A Lateral Theory of Phonology: Volume 1 What is CVCV and why should it be? [Reprint 2012 ed.] 9783110908336, 9783110178715

Table of contents :
Table of contents – detail
1 Editorial note: two volumes
2 Foreword
3 How to use this book
4 Conventions used in this book
Part One: What is CVCV?
1. Introduction
2. Open versus closed syllables in CVCV
3. A unified theory of vowel – zero alternations
4. Alternating vowels are present in the lexicon
5. The beginning of the word: “#” = CV
6. The Coda Mirror
7. Consequences of the Coda Mirror: no confusion between Government and Licensing anymore
8. A syntax of phonology
9. Lateral relations are head-final: length in phonology
10. Syllabic and trapped consonants in CVCV
Part Two: Why CVCV ?
1. Introduction
2. Principles of argumentation I: disjunctive contexts
3. Principles of argumentation II: representations and their function
4. Principles of argumentation III: generality of processes
5. Principles of argumentation IV: a better solution for extrasyllabicity than extrasyllabicity
6. Argument 1. Languages without initial restrictions
7. Argument 2. What you get is NOT what you see: Tina Turner was wrong
8. Argument 3. Description vs. explanation of restrictions on word-initial consonant clusters
9. Argument 4. Lower: empty Nuclei and regressive internuclear relations have been used for over 30 years in the analysis of Slavic vowel-zero alternations
10. Argument 5. The life of “yers” outside of Slavic and in locations where vowels do not alternate with zero
11. Argument 6. Unified representations for the syllable and stress
12. Argument 7. Licensing power of final empty Nuclei parameterised: paired vs. impaired behaviour of internal and final Codas
13. Argument 8. The Coda Mirror
14. Argument 9. News from the yer context: what happens in Codas and before an unpronounced alternating vowel
15. Argument 10. What sonorants do in Codas: a unified theory of melodic reaction on positional plight
General Conclusion
Appendices
1. List of parameters and their translation into CVCV and other theories
2. Closed Syllable Shortening vs. diminutive lengthening in Czech
3. Polish two-membered word-initial consonant clusters
4. A short guide to 1990 Government Phonology
References
Subject Index
Language Index

Citation preview

A Lateral Theory of Phonology

Studies in Generative Grammar 68.1

Editors

Harry van der Hulst Jan Köster Henk van Riemsdijk

Mouton de Gruyter Berlin · New York

A Lateral Theory of Phonology What is CVCV, and why should it be?

by

Tobias Scheer

Mouton de Gruyter Berlin · New York

Mouton de Gruyter (formerly Mouton, The Hague) is a Division of Walter de Gruyter G m b H & Co. KG, Berlin.

The series Studies in Generative Grammar was formerly published by Foris Publications Holland.

® Printed on acid-free paper which falls within the guidelines of the ANSI to ensure permanence and durability.

Library of Congress Cataloging-in-Publication

Data

Scheer, Tobias, 1968A lateral theory of phonology / by Tobias Scheer. p. cm. - (Studies in generative grammar ; 68-) Includes bibliographical references and index. ISBN 3-11-017871-0 (cloth : alk. paper) 1. Grammar, Comparative and general - Phonology. I. Title II. Series P217.S284 2004 414—dc22 2004024594

ISBN 3-11-017871-0 Bibliographic information published by Die Deutsche

Bibliothek

Die Deutsche Bibliothek lists this publication in the Deutsche Nationalbibliografie; detailed bibliographic data is available in the Internet at .

© Copyright 2004 by Walter de Gruyter G m b H & Co. KG, D-10785 Berlin. All rights reserved, including those of translation into foreign languages. No part of this book may be reproduced in any form or by any means, electronic or mechanical, including photocopy, recording, or any information storage and retrieval system, without permission in writing from the publisher. Cover design: Christopher Schneider, Berlin. Happy green gnome by Mogdolina. Printed in Germany.

Table of contents — overview

§ 1 2 3 4

page Table of contents - detail Editorial note: two volumes Foreword How to use this book Conventions used in this book

vii xxxvii xli li liii

Part One: What is CVCV ? 8 14 69 76 83 110 135

1. Introduction 2. Open versus closed syllables in CVCV 3. A unified theory of vowel - zero alternations 4. Alternating vowels are present in the lexicon 5. The beginning of the word: "#" = CV 6. The Coda Mirror 7. Consequences of the Coda Mirror: no confusion between Government and Licensing anymore 165 8. A syntax of phonology 218 9. Lateral relations are head-final: length in phonology 240 1 0. Syllabic and trapped consonants in CVCV

1 7 81 87 95 117 149 181 249 283

Part Two: Why CVCV ? 302 303 304 332 339 381 387

1. Introduction 2. Principles of argumentation I: disjunctive contexts 3. Principles of argumentation II: representations and their function 4. Principles of argumentation III: generality of processes 5. Principles of argumentation IV: a better solution for extrasyllabicity than extrasyllabicity 6. Argument 1 Languages without initial restrictions 7. Argument 2 What you get is NOT what you see: Tina Turner was wrong....

365 369 371 405 415 459 469

vi

Table of contents — overview

§

page

390

8. Argument 3 Description vs. explanation of restrictions on word-initial consonant clusters 411 9. Argument 4 Lower: empty Nuclei and regressive internuclear relations have been used for over 30 years in the analysis of Slavic vowel-zero alternations 426 1 0. Argument5 The life of "yers" outside of Slavic and in locations where vowels do not alternate with zero 501

618 General Conclusion

621 622 623

519 597

625 665

691

15. A r g u m e n t 10

What sonorants do in Codas: a unified theory of melodic reaction on positional plight

620

495

11. A r g u m e n t 6

Unified representations for the syllable and stress 524 1 2. Argument 7 Licensing power of final empty Nuclei parameterised: paired vs. impaired behaviour of internal and final Codas 556 13. Argument8 The Coda Mirror 579 1 4. Argument 9 News from the yer context: what happens in Codas and before an unpronounced alternating vowel 591

473

Appendices 1. List of parameters and their translation into CVCV and other theories 2. Closed Syllable Shortening vs. diminutive lengthening in Czech 3. Polish two-membered word-initial consonant clusters 4. A short guide to 1990 Government Phonology

632 References 633 Subject Index 634 Language Index

707 745

749 753 759 765 779 825 841

Table of contents - detail

§ 1 2 3 4 5 6 7

page Editorial note: two volumes Foreword How to use this book Conventions used in this book 1. General conventions: symbols, languages, cross-references 2. Phonological lingua franca and the skeleton 3. Czech and Polish spelling conventions

xxxvii xli li liii liii lv lvii

Part One: What is CVCV ? 8 Chapter 1 Introduction 9 1. CVCV in a nutshell 10 2. Some core properties of Standard Government Phonology 11 3. Syllabic arborescence is not primitive: it derives from lateral relations among segments 12 4. If lateral relations are primary, syllabic arborescence has to go 13 5. Roadmap of Part One

1 1 3 3 4

14 Chapter 2 Open versus closed syllables in CVCV 15 1. The Empty Category Principle, Proper Government and vowel-zero alternations in Standard Government Phonology 16 1.1. Vowel-zero alternations and the Empty Category Principle.. 17 1.2. The Projection Principle, empty Nuclei and Government 18 1.3. Final empty Nuclei and Coda Licensing 19 1.4. The analysis of vowel-zero alternations in Standard Government Phonology 20 1.4.1. The basic pattern: vocalisation occurs before empty Nuclei

7 8 10 11 15 15

viii

Table of contents - detail

§ 1.4.2. Intervening governing domains block Proper Government 22 1.4.3. Moroccan Arabic vocalisation before geminates induces disjunctivity 23 1.5. Proper Government is not recursive 24 1.6. Domains and domain-final empty Nuclei 25 1.7. Summary: the ECP and Proper Government in Standard Government Phonology 26 2. Open vs. closed syllables in CVCV: the problem and two indications 27 3. Internuclear communication over consonant clusters 28 3.1. The vocalisation of Czech prefixes 29 3.2. Consequences of this préfixai alternation 30 3.2.1. Vowel-zero alternations in Standard Government Phonology 31 3.2.2. Falsification of the statement "intervening governing domains block Proper Government" 32 3.2.3. Unvocalised prefixes occur only before branching Onsets 33 3.2.4. Heteromorphemic TR vs. monomorphemic TR: variable intimacy of adjacent consonants 34 3.2.5. Consonantal interaction 35 3.3. The representation of open vs. closed syllables in CVCV - first definition 36 4. Substantial condition on Infrasegmental Government: the internal structure of consonants 37 4.1. John Harris derives sonority from a non-observational property: complexity 38 4.2. The sonorant is the head of TR clusters 39 4.3. Complexity and the internal structure of consonants 40 4.3.1. Segmental alternations are a sovereign and unquestionable referee 41 4.3.2. Different approaches to the representation of consonants 42 4.3.3. The same set of primes for consonants and vowels 43 4.3.4. The undissociability of velarity and roundness makes wrong predictions

Page

21

17 19 21 22 24 24 26 26 28 28 29 31 33 35 38 40 40 42 44 44 45 46 47

Table of contents - detail

ix

§

Page

44

4.3.5. Dentals, derived sonority and the incompatibility of A and? 51 4.3.5.1. Dentals: [t,d] are empty 51 4.3.5.2. Sonority is a derived category 51 4.3.5.3. A and ? hate each other: they cannot combine 52 4.3.6. Sonorants are heavy: abundant evidence 53 4.3.6.1. [r], [1] and nasals contain A 53 4.3.6.2. [r], [1] and [n] contain I, [1] contains U 55 4.3.6.3. Conclusion: sonorants are too complex to be governees 58 4.3.7. Complexity counts only Place definers 59 4.3.7.1. All stops have the same degree of stopness 59 4.3.7.2. Vocalic sonority disregards manner definers.. 60 4.3.7.3. Counting manner definers is not fair play: sonorants will be demoted before being able to score 61 4.3.7.4. The manner-place distinction is commonplace elsewhere 62 4.4. Infrasegmental Government: how complexity conditions interconso-nantal relations 63 Phonotactic condition on Infrasegmental Government: Government Licensing 65 Consequences for the ECP and final empty Nuclei 67 6.1. The Empty Category Principle - final version 67 6.2. The special status of final empty Nuclei in Standard Government Phonology 67 6.3. The special status of final empty Nuclei in CVCV 70 6.4. Government Phonology predicts that the right edge of words is special 71 Domains of Infrasegmental Government and branching Onsets are not the same 72 7.1. How could the empty Nucleus enclosed by TR clusters betray its existence ? 72 7.2. When TR clusters are bare of Infrasegmental Government: French 73 Open vs. closed syllables in CVCV - second definition 76 The identity of Codas in CVCV - first definition 78

45 46 47 48 49 50 51 52 53 54 55

56 57 58 5. 59 6. 60 61 62 63 64 7. 65 66 67 8. 68 9.

Table of contents — detail

page

§ 69 Chapter 3

A unified theory of vowel - zero alternations 70 71 72 73 74 75

1. Transparency of monomorphemic clusters for Proper Government 2. Vocalisation of alternation sites: Standard Government Phonology is disjunctive, CVCV is not 2.1. Why Standard Government Phonology must invoke two different causes 2.2. Why all vocalisations have the same cause in CVCV 3. Observation vs. explanation 4. Proper Government always applies

81 82 82 84 85 86

76 Chapter 4

Alternating vowels are present in the lexicon 77 78 79 80 81

82

1. Standard Government Phonology: alternating vowels are inserted 2. Alternating vowels may not always be identified phonetically 3. Deletion or insertion? When more than one vowel alternates with zero 4. Deletion is also enforced for theory-internal reasons 5. Alternating vowels are floating chunks of melody 6. The three-way distinction: associated vs. floating vs. no melody at all

87 87 88 89 90 91

83 Chapter 5

The beginning of the word: "#" = CV 84 85 86 87 88 89

1. The beginning of the word is a phonological object 1.1. Phonologists do not appear to be eager to discover the real identity of the unknown 1.2. Nobody was happy to get rid of pink panthers when the Coda was (re)introduced 2. Why do pink panthers always do the same things ? 2.1. There can be no causal relation between things that belong to different worlds 2.2. What left pink panthers do I: co-occurrence restrictions on initial clusters

95 95 97 97 97 98

Table of contents - detail

§ 2.3. What left pink panthers do Π: stability of the first vowel of the word 91 2.4. What left pink panthers do ΓΠ: strength of word-initial consonants 92 2.5. Disjunctivity leads us to # = CV 93 2.6. Initial vowels are stable because they are preceded by an empty Nucleus 94 2.7. Czech prefixes again 95 2.8. Summary: phonology operates only over phonological objects 96 3. Word-initial empty Nuclei prior to CVCV: Magic Licensing 97 3.1. The notorious bad guy: s+C sequences 98 3.2. Initial Codas 99 3.3. Magic Licensing 100 3.4. Initial s+C sequences in CVCV 101 3.5. The mystery of s+C sequences is also melodic: [s] is the bad guy 102 4. Restrictions on initial consonant clusters 103 4.1. Why are there no initial #RT clusters? The classical explanation is circular 104 4.2. CVCV and initial restrictions 105 4.3. What about languages without initial restrictions ? 106 5. Interonset Government 107 5.1. Comparison of Interonset and Infrasegmental Government 108 5.2. Interonset Government violates the most fundamental principles of Standard Government Phonology 109 5.3. Interonset Government in CVCV has no handle on initial clusters

xi

page

90

98 99 100 100 101 103 103 103 105 105 106 107 108 108 109 111 111 111 112 115

110 Chapter 6 The Coda Mirror 111 1. Scope of the Coda Mirror 112 1.1. Positional effects vs. the transmission of primes between (adjacent) segments 113 1.2. The three players and their eventual coalition: position, shared melody and stress 114 2. The basic factor: five positions

117 117 117 119

xii

Table of contents - detail

§

Page

115 3. The Strong Position: empirical evidence 116 3.1. Why the Coda Mirror is the Mirror of the Coda 117 3.2. French obstruents 118 3.3. Somali stops 119 3.4. Sievers' Law: the vocalic face of the Coda Mirror 120 3.4.1. Vocalic effects of the Coda: closed syllables 121 3.4.2. Sievers' Law - the facts 122 3.4.3. Alternation sites are vocalised before Codas and after Coda Mirrors 123 4. Consequences for syllable structure 124 5. The Coda Mirror: descriptive adequacy 125 6. Government inhibits, Licensing enhances the segmental expression of the target 126 6.1. Government Licensing: French 127 6.2. Government Licensing: Czech 128 6.3. Charette (1990): consonant clusters need vocalic support.... 129 6.4. A good guy and a bad guy 130 7. The Coda Mirror: explanatory adequacy, or why strong positions are strong 131 8. Two ways of being weak 132 9. Positional Faithfulness and associated psycho-linguistic explanations miss the disjunction 133 9.1. Positional Faithfulness has overlooked half of the Strong Position 134 9.2. Psycho-linguistic "grounding" is no ground for the Strong Position

121 121 122 124 127 127 128 130 131 132 134 135 136 137 138 139 142 145 145 146

135 Chapter 7 Consequences of the Coda Mirror: no confusion between Government and Licensing anymore 136 1. The confusion of Government and Licensing in Standard Government Phonology 137 1.1. Licensing: one word for two different realities 138 1.1.1. Type I Licensing: Licence in order to be absent

149 149 149

139

1.1.2. Type II Licensing: Licence in order to be present

151

140

1.1.3. Summary

152

Table of contents - detail

§ 141 142 143 144 145 146 147 2. 148 149 150 151 152 3. 153 154 155 156 157 158 159 160 161 4. 162 5. 163 164

xiii

Page 1.2. Government: a different word for similar phonological realities 1.2.1. Type I Government: the segment gluer 1.2.2. Type II Government: the segment-destroyer 1.2.3. Summary 1.3. Did the conception of Harris (1994a) govern the model as it stood in 1990? 1.4. Conclusion: lateral relations need to be identified according to their properties Government and Licensing in CVCV 2.1. There is only one kind of Government 2.2. The special status of Infrasegmental Government 2.3. The identity of Codas in CVCV - second definition 2.4. Government and Licensing are the translation of two antagonistic situations that are found in nature Internuclear Licensing 3.1. Internuclear Licensing was practised before it was named: Kaye (1990a) and Yoshida (1993) 3.1.1. Vowel shortening before an empty Nucleus 3.1.2. The unnamed internuclear relation at hand is Licensing 3.2. Vowel length alternations in CVCV 3.2.1. Italian Tonic Lengthening 3.2.2. The target of spreading must be licensed 3.2.3. Government or Licensing ? 3.2.4. Internuclear Government and Licensing in complementary distribution ? Typology of lateral relations Open vs. closed syllables in CVCV 5.1. Third definition 5.2. The two typical syllable-related vocalic events: alternations in length and vowel-zero alternations

153 153 154 155 156 157 160 160 162 163 164 165 166 166 169 170 170 171 173 174 175 177 177 179

xiv

Table of contents - detail

§

page

165 Chapter 8 A syntax of phonology 166 1. The core identity of Government Phonology: latéralisation of structure and causality 167 1.1. Vowel-zero alternations and Proper Government 168 1.2. Coda Licensing 169 1.3. Government Licensing 170 1.4. Syllable structure and its function 171 1.5. Conclusion 172 2. Remaining vertical structure and causality in Standard Government Phonology 173

2.1. Syllable structure

174 175 176

2.2. Effects of the Coda on consonants 2.3. Effects of the Coda on preceding vowels 2.4. The Binary Theorem: its interpretation and its consequences 177 2.4.1. Ternary constituents are ill-formed, intermediate structure does not matter 178 2.4.2. The price to pay: an unwarranted prediction 179 2.5. Prosodie Government 180 2.6. Coda-Onset vs. Bogus clusters 181 2.6.1. Bogus: a three-way distinction 182 2.6.2. Typical bogus clusters: tl, dl 183 2.6.3. Cam(e)ra: bogus clusters created by post-tonic syncope in English 184 2.6.4. Bogus clusters with a melodic effect: t-lenition in English 185 2.6.5. We never see a true three-way distinction 186 3. Harris (1994a): latéralisation of causality but not of structure, and a new meaning for Government and Licensing 187 3.1. Government curtails distribution, Licensing is a condition on syllable structure 188 3.1.1. New Licensing relations that parallel Government.... 189 3.1.2. Licensing vs. Government I: different function, different effect

190

3.1.3. A new parameter: Rhymal Adjuncts may be licensed once or twice

181 182 185 186 188 190 191 191

193 194 196 196 197 198 199 199 199 200 201 203 204 204 204 206

208

Table of contents - detail § 191

xv Page

3.1.4. Why are there no Coda clusters nor word-final Codas? 210 192 3.1.5. Closed Rhyme Shortness lateralised 211 193 3.1.6. Licensing vs. Government Π: Rhymal Adjuncts are constituent-licensed, but not constituent-governed 211 194 3.1.7. Government and Licensing viewed by John Harris and the Coda Mirror 212 195 3.2. Causality is only lateral: Licensing Inheritance 215 196 3.2.1. The general picture 216 197 3.2.2. Alas, foot-initial Φ word-initial, foot-internal φ intervocalic 218 198 3.2.3. The status of Codas and word-final consonants 220 199 3.2.4. Licensing Inheritance: how it works 222 200 3.2.5. The strength of post-Coda consonants 223 201 3.2.6. Licensing Inheritance: three major problems 225 202 3.2.6.1. The Coda Mirror context remains disjunctive 225 203 3.2.6.2. Stress is a secondary, not a primary factor.... 226 204 3.2.6.3. In many systems, post-Coda consonants are insensitive to the content of the preceding Coda 227 205 3.2.7. Summary 229 206 4. CVCV: latéralisation of both structure and causality 230 207 4.1. Harris' advances establish a hybrid system: lateral causality but vertical structure 230 208 4.2. Arborescence is redundant, the null hypothesis for syllable structure is lateral 231 209 4.2.1. Takahashi (1993): only lateral causality, a farewell to constituency, but no additional empty Nuclei 231 210 4.2.2. The existence of lateral relations makes arboreal structure redundant 235 211 4.2.3. The null hypothesis for syllable structure is lateral: why taking two steps if one is enough ? 237 212 4.2.3.1. The central tacit assumption: co-occurrence restrictions are due to arboreal structure 237 213 4.2.3.2. Co-occurrence restrictions: the parallel with syntax is phoney 237 214 4.2.3.3. The only critical factor is relative sonority, hence a lateral relation 240

xvi

Table of contents - detail

§ 215

216 217

Page 4.3. Bogus clusters again: transforming melodic contrasts among adjacent objects into vertical structure is a bad idea 4.4. HIGH vs. LOW 4.5. Summary

241 243 246

218 Chapter 9 Lateral relations are head-final: length in phonology 219 1. Lateral relations in CVCV are automatically regressive 249 220 2. Vocalic length 251 221 2.1. Systems with inalterable vowel length 252 222 2.2. Syllable-sensitive vowel length 253 223 2.2.1. Closed Syllable Shortening 253 224 2.2.2. Tonic Lengthening 255 225 2.2.3. Three labels, one phonological reality 256 226 2.3. Compensatory Lengthening 259 227 2.4. Compensatory Lengthening is different: spreading may occur in either direction 262 228 2.5. All long vowels obey the same requirement: their complement is licensed 264 229 2.6. LOWER spreading may be in both directions, but UPPER Licensing is only regressive 266 230 2.7. Alternating long vowels are head-initial, their nonalternating peers are head-final 267 231 2.7.1. Only CVCV can build on the contrast head-initial vs. head-final 267 232 2.7.2. Head-final vs. head-initial also expresses the parameter on the existence of super-heavy Rhymes... 268 233 2.8. Progressive Compensatory Lengthening: an argument in favour of CVCV 271 234 3. Consonantal length 275 235 3.1. The representation of geminates in conventional models and in CVCV 275 236 3.2. Standard Government Phonology: why geminates cannot be preceded or followed by consonants 276 237 3.3. CVCV: head-initial and head-final geminates 277

Table of contents - detail § 238 239

xvii Page

3.4. CVCV: why geminates cannot be preceded or followed by consonants 3.5. Geminates, RT clusters and homorganic NC clusters are one

279 280

240 Chapter 10 Syllabic and trapped consonants in CVCV 241 1. Setting the scene: phonological hermaphrodites 242 2. Syllabic and trapped consonants tell us about each other we are well advised to listen carefully 243 3. The synchronic situation of trapped consonants: trapped (Polish) vs. syllabic (Czech) 244 3.1. Trapped consonants in Polish: lexically trapped or trapped by a vowel-zero alternation 245 3.2. Syllabic, but not trapped consonants constitute syllabic peaks and can bear stress 246 3.3. Czech syllabic consonants and préfixai vowel-zero alternations 247 3.4. Vocalisation of Polish prefixes before trapped roots 248 3.4.1. Morphology has got a word to say 249 3.4.2. Polish prefixes and phonological domains 250 3.4.3. How Polish prefixes behave before trapped consonants 251 3.5. Summary syllabic vs. trapped consonants 252 4. What kind of animal is a syllabic consonant ? 253 4.1. Classical approaches since SPE: function, not behaviour decides 254 4.2. Why syllabic consonants do not sit in Nuclei 255 5. Do syllabic consonants spread to their right or to their left ? 256 5.1. Some literature 257 5.2. A fundamental argument for left-branchers: the complementary distribution of Ç and aC 258 6. Do syllabic consonants need to satisfy the ECP ? 259 7. Alternations of syllabic and non-syllabic versions of the same consonant

283 284 286 286 288 289 292 292 295 297 298 298 299 300 302 3 02 304 306 307

xviii

Table of contents - detail

§

Page

260 261

8. Syllabic consonants are left-branching structures: arguments... 309 8.1. Synchronic situation in Germanic: complementary distribution of CI and aC 309

262

8.2. Diachronic situation: syllabic consonants come into being because a preceding vowel is lost 309 9. Syllabic consonants are left-branching: they govern in Czech ..311 1 0. The phonological identity of trapped consonants 314 10.1. Trapped consonants are right-branching: they refuse to govern in Polish 314 10.2. Charette (1992): trapped consonants are an extreme case of Government Licensing 315 11. Summary: the identity of syllabic and trapped consonants 316 1 2. Phonetic correlates of syllabic and trapped consonants 319 1 2.1. Trapped consonants are transparent to voicing, syllabic consonants are not 319 1 2.2. Classical interpretation: trapped = extrasyllabic 321 12.3. An alternative solution: trapped = obstruent 323 12.3.1. Real sonorants do not undergo final devoicing Palatalised sonorants devoice word 323 12.3.2. Palatalised sonorants devoice word-finally 323 1 2.3.3. Trapped sonorants are obstruents 325 12.3.4. Romansch trapped sonorants are also transparent and have voiceless allophones 326 1 2.3.5. Summary 329 13. Diachronic confirmation: yers followed trapped, but preceded syllabic consonants 329 13.1. The Western Slavic comparatistic situation: cz CRC = pol CVRC and CRC 330

263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284

13.2. The Common Slavic and Old Church Slavonic sources of Polish vocalised liquids 13.3. CS tbrt and CS trbt have never merged 13.3.1. CS tbrt and CS trbt remain distinct in Polish 13.3.2. CS tbrt and CS trbt remain distinct in Baltic and Eastern Slavic 13.3.3. The yers of OCS trbt < CS tbrt refuse to vocalise 13.3.4. Summary: trapped consonants result from the loss of a following yer, but what about their syllabic peers ?

333 335 335 335 339

340

Table of contents - detail § 285 286 287 288 289

290 291 292

293 294 295 296 297 298 299 300 301

xix page

13.4. What has happened to Czech (and Slovak) trapped consonants? 13.4.1. Old Czech: sonorants from CS tbrt are syllabic against trapped from CS trbt 13.4.2. Secondary vocalisation of syllabic consonants in Czech (and Polish, Slovak) 13.4.3. The Old Czech phonemic opposition trapped < CS trbt vs. syllabic < CS tbrt 13.4.3.1. Syllabic liquids from CS tbrt do, but trapped liquids from CS trbt do not count in Old Czech verse 13.4.3.2. The opposition trapped vs. syllabic was phonemic in Old Czech 13.4.3.3. Prepositions stopped to vocalise when trapped roots became syllabic 1 3.4.3.4. Why did trapped consonants refuse to become syllabic in word-initial position ? 1 3.4.4. Trapped consonants in Modern Czech: type krtit = tft < tr't < CS trbt 13.4.4.1. Inventory of trapped roots, and why they could not become syllabic 13.4.4.2. Do trapped roots provoke the vocalisation of prepositions ? 1 4. Summary and final amendment due to the right periphery of syllabic consonants 1 4.1. Summary 1 4.2. The right periphery of syllabic consonants 1 4.3. Phonological domains are of no rescue 14.4. Syllabic consonants behave like both pre- and postvocalised structures 15. Conclusion

341 341 343 345

345 347 348

350 351 351 353 356 356 358 360 361 363

XX

Table of contents - detail

§

Page

Part Two: Why CVCV ? 302 Chapter 1 Introduction

365

303 Chapter 2 Principles of argumentation I: disjunctive contexts

369

304 Chapter 3 Principles of argumentation II: representations and their function

371

305 1. What representations are good for: they are the answer of the 80s to the fundamental problem of overgeneration that has dominated the debate in the 70s 306 1.1. SPE and overgeneration 307 1.2. One option: fighting against abstractness limits the generative power 308 1.3. Another option: autosegmental representations automatically restrict the generative power 309 1.4. Structure and process: there is no natural science in absence of one or the other 310 1.4.1. OT holds that monsters could exist in nature, but accidentally do not occur 311 1.4.2. Only competition determines (a)grammaticality in OT: nothing can be inherently (a)grammatical 312 1.4.3. Computation is king: representations are only decorative in OT 313 1.4.4. Since they do not matter, representations are arbitrary and interchangeable in OT: they are doomed to disappear 314 1.4.5. Back to where we started: representations were the enemy of overgeneration

371 371 372 375 377 378 379 379

380 382

Table of contents - detail

§ 1.4.6. OT tools for fighting back overgeneration: markedness and grounded constraints 316 1.4.6.1. Markedness and OT 317 1.4.6.2. The fundamental source of arbitrariness in OT: anything and its reverse can be a constraint 318 1.4.6.3. Grounded constraints 319 1.4.6.4. If phonological events have exclusively non-phonological (grounded) causes, why replicate them as constraints ? 320 1.4.7. Summary: structure and process have an independent existence in nature, but not in OT 321 1.4.8. Shifting burden over to GEN: a promising track to follow? 322 1.4.9. Inventing and turning wheels 323 2. Structure and process: the take of Government Phonology 324 2.1. Representations that are (en)rich(ed) enough to support parameterisation 325 2.2. Internal and final Codas: different but still the same 326 2.3. Vowel-zero alternations: enriched representations avoid absolute neutralisation and allow to express the parameter Havlik vs. Lower 327 2.3.1. Slavic vowel-zero alternations: the basic pattern 328 2.3.2. Two kinds of yers 329 2.3.3. The introduction of representations allows to get rid of the absolute neutralisation 330 2.3.4. Further enriched representations: Government Phonology 331 2.3.5. The parameter Havlik vs. Lower supposes enriched representations

xxi

page

315

384 384

385 387

390 391 393 394 395 395 396

398 398 399 400 401 403

332 Chapter 4 Principles of argumentation III: generality of processes 333 1. Introduction: SPE hocus-pocus, the baby and the bath 334 2. Vowel-zero alternations 335 3. Vowel-zero alternations and sonority sequencing in languages without initial restrictions 336 4. The yer context

405 406 408 410

xxii

Table of contents - detail

§

Page

337 5. The Coda Mirror, Closed syllable shortening and 1-vocalisation 338 6. Summary

412 413

339 Chapter 5 Principles of argumentation IV: a better solution for extrasyllabicity than extrasyllabicity

415

340 1. Some elementary and consensual facts about syllabic theory 341 2. The facts that cannot be accommodated by the theory 342 3. Extrasyllabicity was created by people who believed in their theory 343 4. Type I extrasyllabicity: enforced underpaying, an edge consonant fails to be parsed 344 4.1. Interaction of unparsable consonants with other rules: typical serial solutions 345

346 347 348 349 350 351 352 353 354 355 356 357

416 417 418 420 420

4.1.1. German Jagd: devoicing must apply after

adjunction 4.1.2. Stray erasure and the Prosodie Hierarchy 4.1.3. Polish kadra: devoicing must apply before adjunction 4.2. Can there be more than one extrasyllabic consonant at the right edge? 4.2.1. Reduction of extrasyllabic candidates by morphology 4.2.2. You can get a coronal for free in Germanic 4.2.3. Summary 4.3. Word-initial extrasyllabicity 4.3.1. Typology of initial extrasyllabicity 4.3.2. Word-initial extrasyllabic consonants are different I: they are not transparent to voicing 4.3.3. Word-initial extrasyllabic consonants are different II: they do not degeminate 4.3.4. The contrast is automatic if word-final extrasyllabicity is due to final empty Nuclei 4.4. Peripherally

420 421 422 423 423 424 426 426 426 427 428 429 430

Table of contents - detail

§ 358 5. Type II extrasyllabicity: deliberate underpaying, a word-final consonant does not behave like a Coda 359 5.1. If certain word-final consonants must not be Codas, what could they be ? 360 5.2. Extrasyllabicity effects on both word-final consonants and the preceding vowel 361 5.3. Summary 362 6. Doubts on extrasyllabicity and an alternative view 363 6.1. There is no extrasyllabicity without serialism and a syllabification algorithm 364 6.2. Extrasyllabic once, extrasyllabic forever 365 6.3. Extrasyllabic consonants do not behave like nothing they often behave like Onsets 366 6.4. Can we afford to allow for constituents that do not express any co-occurrence ? 367 6.5. Extrasyllabic consonants adjoined to the phonological word 368 6.6. There are initial and internal s+C effects, but there are only initial extrasyllabic consonants 369 6.6.1. The regular extrasyllabic analysis of initial s+C clusters 370 6.6.2. Word-internal s+C effects 371 6.6.3. All s+C clusters cannot be contour segments either 372 6.6.4. Conclusion 373 6.7. Why are there no words with two, nine or twenty extrasyllabic consonants ? 374 6.7.1. The only definition of extrasyllabicity is negative 375 6.7.2. Wild Polish initial clusters are less wild than their reputation 376 6.7.3. Conclusion: there are no extrasyllabic clusters in Polish, and probably not in Salish or any other language either 377 6.8. Confusion of causalities: there are three, not two phonologies 378 6.9. Edge consonants are special because they occur at edges: all phonology-internal solutions (such as extrasyllabicity) must be wrong

xxiii

page 431 431 433 435 435 436 437 437 438 440 441 441 443 443 444 445 445 446

450 452

454

xxiv

Table of contents - detail

§ 379 380

page 6.10. We need a theory of margins: final empty Nuclei and the initial CV 6.11. Conclusion

455 456

381 Chapter 6 Argument One Languages without initial restrictions 382 1. Classical syllabic theory is made of sonority sequencing, Onset Maximisation and the word-initial anchor 383 2. No co-occurrence restrictions, hence no branching constituents. 384 3. Could a consonant of whatever sonority close an internal syllable ? 385 4. How to have your cake and eat it: vowel-zero alternations have got nothing to do with closed syllables 386 5. Only CVCV offers a unified theory of syllable structure and vowel-zero alternations

459 460 462 465 467

387 Chapter 7 Argument Two What you get is NOT what you see: Tina Turner was wrong 388 1. Phonologists explain that Tina Turner was wrong when teaching phonemics and Onsets 389 2. Phonologists explain that Tina Turner was right when teaching Nuclei

469 471

390 Chapter 8 Argument Three Description vs. explanation of restrictions on word-initial consonant clusters 391 1. What circularity is: the non-independence of prediction and observation 392 1.1. The only reality that is real is the one that humans perceive: Kant, Saussure, Popper 393 1.2. How a sound prediction emerges: back and forth between data and theory 394 1.3. Circularity: overt and hidden varieties 395 1.4. Summary

473 473 474 475 476

Table of contents - detail

xxv

§

page

396

2. The regular account of initial restrictions: "within a branching Onset sonority must increase" 476 3. Constraints are inherently circular and may inhibit research 478 4. Standard Government Phonology: constituency is redundant, it restates the lateral relations among segments a second time .... 479 5. Lateral relations are self-sufficient: CVCV builds on complexity and lateral relations, but does away with redundant arboreal structure 481 6. Standard Government Phonology: the left-headedness of constituents does not follow from anything 482 7. CVCV: why #RT cannot exist in typical Indo-European languages 483 8. The initial CV parameterised: morphology does (#TR-only languages) or does not (anything-goes languages) send down the initial CV 485 8.1. How can anything-goes languages exist, and why are there no #RT-only languages ? 485 8.2. The distribution of the initial CV is parameterised: it is present in #TR-only languages, but absent when anything goes 485 8.3. Diacritics are arbitrary in number, nature and effect, but their effect in natural language is not 487 8.4. Anticipating on Volume 2: phonologically relevant morpho-syntactic information is translated into phonological categories 489 8.5. Why #RT-only languages cannot exist 490 8.6. Word-initial extrasyllabicity in CVCV: why there can be one extrasyllabic consonant at most 490 9. Independence of theoretical devices and the empirical target 491 10. Conclusion 492

397 398 399

400 401 402

403 404

405 406

407 408 409 410

411

Chapter 9 Argument Four Lower: empty Nuclei and regressive internuclear relations have been used for over 30 years in the analysis of Slavic vowel-zero alternations

412

1. Introduction 2. Slavic vowel-zero alternations and their analysis

413

495 497

xxvi Table of contents - detail §

Page

414 3. Distributional facts and the challenge they raise 498 415 4. Lower - how it works and what it implies 501 416 5. The difference between Lower and Havlik 505 417 5.1. Lightner makes modern Slavic look like Common Slavic underlyingly, yet Lower is not exactly like Havlik.... 505 418 5.2. The difference: "secondary vocalisation", that is "every other" (Havlik) vs. "all but the last" (Lower) 506 419 6. The difference between final and alternating yers 509 420 6.1. Synchronic evidence 509 421 6.2. Diachronic evidence: only alternating yers may originate in epenthesis 509 422 6.3. Conclusion: Common Slavic yers and modern abstract vowels 513 423 7. Autosegmentalised Lower 513 424 8. Slavic vowel-zero alternations are caused by a lateral relation.... 516 425 9. Conclusion 517 426 Chapter 10 Argument Five The life of "yers" outside of Slavic and in locations where vowels do not alternate with zero 427 1. Introduction 428 2. Yers all over the place in Slavic? 429 2.1. The general picture 430 2.2. Czech alternations in vowel length 431 2.3. Czech and Polish [o] - [u(u)], Polish ^-ç 432 2.3.1. The synchronic situation of [d] - [u(u)] 433 2.3.2. The only solution is diachronic: the alternation in fact concerns vowel length 434 2.3.3. Polish 3-ç is but the nasal version of o > oo 435 2.4. Summary: the ambition of Lower is too narrow 436 3. French: yers all over the world ? 437 3.1. ATRness of mid vowels in Southern varieties 438 3.2. French couldn't have yers, but it can have abstract vowels 439 3.3. Schwa - [ε] alternations 440 4. The big puzzle: nature produces two antagonistic patterns

519 520 520 521 522 522 523 525 526 527 527 528 530 531

Table of contents - detail

§ 441 5. Three implementations of the insight that yers are empty Nuclei 442 5.1. Empty Nuclei prior to Government Phonology I: Stephen Anderson on French 443 5.1.1. On the structural side: how to get something for nothing 444 5.1.2. On the computational side I: Coda capture 445 5.1.3. On the computational side II: Coda capture misses the basic insight of Lower 446 5.2. Tracy Hall on German: consonants preceding schwa end up as Codas, but for a good reason 447 5.2.1. On the structural side: schwa is a floating x-slot 448 5.2.2. On the computational side: the consonant preceding schwa has a good reason to end up as a Coda 449 5.3. Empty Nuclei prior to Government Phonology II: Andrew Spencer on Polish 450 5.3.1. On the structural side: a less abstract Lower 451 5.3.2. On the computational side: fill-in without causality vs. intervocalic relation 452 5.3.3. Kenstowicz & Rubach's (1987) arguments against empty Nuclei 453 5.3.4. Spencer's analysis cannot be extended to other alternations 454 5.4. Szpyra (1992a): insertion into unsyllabifiable clusters without intervocalic causality 455 5.5. We are looking for a theory of intervocalic relations 456 5.5.1. Synopsis of all analyses regarding the yer context 457 5.5.2. Why does everybody delete material without phonetic existence at the end of the derivation ? 458 5.5.3. The lateral relation described by Lower is (Proper) Government 459 6. Government Phonology and yers 460 6.1. Abstract vowels can be nothing but empty Nuclei 461 6.2. Empty Nuclei are not really empty: their melody is present underlyingly

xxvii

Page 533 533 533 534 536 539 539

541 543 543 545 546 546 547 548 548 550 550 551 551 552

xxviii Table of contents - detail

§ 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487

Page 7. Empty Nuclei after internal Codas 7.1. Slavic vowel-zero alternations and French schwa-[e]: no additional empty Nuclei needed 7.2. More empty Nuclei required 7.2.1. Czech vowel length 7.2.2. French ATRness 7.3. General summary thus far 8. There are two patterns of vowel-zero alternations in nature: Havlik and Lower 8.1. Government derives Havlik, not Lower 8.2. Havlik and Lower: how they are distributed and how they are parameterised 8.3. A terminological clarification 9. Contradictory effects of yers: there are two antagonistic lateral forces in nature 10. Who is who (Government vs. Licensing) and the parameterised lateral ability of schwa 1 0.1. Identification of the lateral relation through its effect 1 0.2. Licensing abilities of schwa are also parameterised (and independent from the parameter on government).... 1 0.3. Western Slavic [oo] is not a regular long vowel: it was born through the voicing of the following consonant 1 0.4. French ATRness is an instance of Licensing 10.5. The tricky French schwa - [ε] is driven by Licensing as well 1 0.6. Summary: lateral abilities and the distribution of Government and Licensing 11. More evidence from German: the velar nasal 11.1. Introduction 11.2. The distributional situation of [η] and [qg] in German 11.3. /g/ does not appear on the surface when it fails to be licensed 11.4. Additional evidence for the governing ability of schwa in Germanic: Dutch 12. Gussmann & Kaye (1993): cyclicity, domains and Reduction.. 1 2.1. The selection of Havlik or Lower has got nothing to do with the particular vowel that alternates 1 2.2. Domains and Reduction

554 555 556 556 559 559 560 560 562 564 565 567 567 568 570 572 574 575 576 576 577 579 582 583 583 583

Table of contents - detail § 1 2.3. As cyclic as Rubach (1984), but Reduction on top of that 489 1 2.4. Boundary abuse: selling phonology for morphology 490 13. Havlik vs. Lower: how Rubach (1984), Standard Government Phonology and CVCV encode the parameter 491 13.1. Standard Government Phonology and CVCV 492 13.2. Classical Lower as exposed in Rubach (1984) 493 13.3. Comparison of the three approaches: procedural vs. lateral solutions 494 13.4. Four different nuclear categories 495 14. Summary and conclusion of chapters 11,9 and 11,10 496 14.1. The study of the yer context in the past 30 years: three separate traditions 497 1 4.2. The West was wrong: the closed, not the open syllable is an optical illusion 498 1 4.3. Government Phonology has given a theoretical status to things that people have been using for a long time: empty Nuclei and lateral relations 499 1 4.4. Parameterisation of the lateral capacities of schwa 500 1 4.5. What vowel-headed Government and Licensing do to vowels and consonants

xxix Page

488

585 586 588 588 588 589 590 591 591 592

594 594 595

501 Chapter 11 Argument Six Unified representations for the syllable and stress 502 1. Introduction 503 2. Classical interpretation of stress: grids and morae 504 2.1. Empirical generalisation I: stress is distributed according to vowels and vocalic quantity (always) and Codas (sometimes) 505 2.1.1. The major parametric division across languages: CVC syllables are either light or heavy 506 2.1.2. A possible third parametric situation: the sonority of Coda consonants decides whether the syllable is light or heavy 507 2.2. Grid-based and moraic accounts of generalisation 1 508 2.2.1. Moraic analysis of the Latin pattern: third but last mora

597 599

599 599

601 603 604

XXX

Table of contents - detail

page

§

2.2.2. Extrametricality (dominus, fáció) and fórmula 605 2.2.3. Latin viewed through the grid: syllable 510 marking rules 608 2.2.4. How grid-based and moraic systems express 511 the three-way typology 609 2.3. Empirical generalisation II: Onsets are invisible to stress .... 610 512 513 2.4. Moraic theory and grids encode, but do not explain generalisations I and II 611 514 3. Stress assignment and CVCV 613 515 3.1. Why Onsets never count: their Nucleus is not empty 613 516 3.1.1. Stress falls on the third but last Nucleus 613 517 3.1.2. Viewed from ABOVE: ungoverned empty Nuclei are invisible because they are silenced by the action of LOWER melody 615 518 3.1.3. CVCV: the metrical irrelevance of Onsets is in-built 618 3.2. Syllabic and prosodie generalisations are expressed 519 by the same structure 619 520 3.3. Consonants are never counted, the parameter known as "Weight by Position" concerns the visibility of governed empty Nuclei 619 521 3.4. How CVCV expresses the third pattern: when weight is sensitive to the sonority of Coda consonants 620 522 3.5. CVCV explains where others observe 623 523 4. Conclusion 623 509

524 Chapter 12

Argument Seven Licensing power of final empty Nuclei parameterised: paired vs. impaired behaviour of internal and final Codas 525 526 527 528 529 530

1. Setting the scene: Standard Government Phonology moved back to SPE 2. Effects on Codas 2.1. Internal Φ final Coda 2.2. Internal = final Coda 2.2.1. L-vocalisation in Brazilian Portuguese 2.2.2. Boundary abuse: selling phonology for morphology..

625 627 627 629 629 631

Table of contents - detail

xxxi

§

Page

531

2.2.3. Boundary abuse produces absurd distributional and diachronic consequences 632 Effects on the vowel preceding Codas 633 3.1. Internal Φ final Coda 63 3 3.2. Internal = final Coda 635 The life of final empty Nuclei 637 4.1. How can you have your cake and eat it ? 637 4.2. How you can have your cake and eat it without turning back the wheel 638 4.2.1. Coda Licensing is too rigid, and it cannot be parameterised 638 4.2.2. Lateral relations can be parameterised: a genuine tradition in Government Phonology 640 4.2.3. Word-final RT clusters exist when final empty Nuclei can govern 642 4.2.4. Final empty Nuclei can only govern Nuclei that are bare of any underlying melody (floating or attached) 643 4.2.5. All the cake: the (im)pairment of final and internal Codas depends on whether word-final consonants are licensed or not 645 4.2.6. The parametric situation of word-final consonants .... 648 4.2.7. When final empty Nuclei license preceding Nuclei: the variable behaviour of vowels in final closed syllables 650 4.2.8. The final parametric table 651 Summary 652 Comparison of extrasyllabicity and the parameterised lateral actorship of final empty Nuclei 652 6.1. Introduction 652 6.2. Why are there no extrasyllabic vowels? Because there are final empty Nuclei, but no final empty Onsets 653 6.3. Deliberate underpaying is Licensing, enforced underpaying is Government 654 6.4. Extrasyllabic once, extrasyllabic forever: why extrasyllabicity is not selective 655 6.5. Why there are no extrasyllabic clusters at the right edge of words 657

532 3. 533 534 535 4. 536 537 538 539 540 541

542

543 544

545 546 5. 547 6. 548 549 550 551 552

xxxii Table of contents - detail

§

Page

553 7. The fourth object: schwa 554 7.1. What a schwa can do 555 7.2. Summary: the four primary nuclear objects and their lateral activity in some languages

658 658 661

556 Chapter 13 Argument Eight The Coda Mirror 557

1. Classical syllable structure is unable to characterise the

558

Coda Mirror as a natural class 1.1. The Coda Mirror: summary

559

1.2. Disjunctions and their consequences in phonology

665 665 667

560 2. Is the Coda Mirror a phonological object ?

669

561

669

2.1. Tiberian Hebrew spirantisation

562

2.2. Non-disjunctive contexts win - but who wins when everybody is non-disjunctive ? 563 2.3. The choice of the context determines the causality of the process 564 3. Why the Coda Mirror is as real as Onsets and Codas 565 3.1. Fortitions 566 3.2. The absence of an event is an event 567 3.3. Lenition is a positional effect that owes nothing to adjacency 568 3.3.1. Positional vs. adjacency effects 569 3.3.2. Obviously, Tiberian Hebrew spirantisation has got nothing to do with adjacency 570 3.4. How could any lenition be caused by adjacency ? 571 3.5. The Mirror effect 572 3.6. When the Strong Position is prone to lenition 573 3.6.1. The Coda Mirror makes only relative predictions 574 3.6.2. The High German Consonant Shift 575 3.6.3. Did fricatives in weak positions go through an intermediate affricate stage ? 576 3.6.4. There is not a shred of evidence for post-vocalic affricates in any old or modern dialect 577 3.6.5. Negative evidence in diachronics: "X does not exist" 578 4. Summary

671 671 672 672 675 675 675 677 679 681 683 683 683 684 686 688 689

Table of contents - detail xxxiii §

Page

579 Chapter 14 Argument Nine News from the yer context: what happens in Codas and before an unpronounced alternating vowel 580 1. A particularly nasty variant of the yer context: when the schwa is unpronounced 581 2. How to escape both resyllabification and disjunctivity 582 3. The Polish palatal nasal and its implosion in internal Codas 583 3.1. Alternation of [n] and β] 584 3.2. Alternation of [ji] and Q] 585 3.3. Rhymal Adjuncts are followed by an empty Nucleus 586 3.4. Schwa is a good licensor in Polish 587 4. The German velar nasal again 588 4.1. Identical behaviour in internal and final Codas: /Ng/ —> [q] 589 4.2. [q] also occurs before unpronounced alternating vowels 590 5. Conclusion

691 692 694 694 695 697 699 700 700 702 705

591 Chapter 15 Argument Ten What sonorants do in Codas: a unified theory of melodic reaction on positional plight 592 1. The logic of the argument 593 2. Homorganic NC clusters: how the mirage of assimilation is created 594 3. The obstruent cannot impose anything on the nasal in CVCV, but the nasal may be active: there is neither master nor servant 595 4. Usually unrelated evidence I: the behaviour of nasals in final Codas (Somali, Southern French, Polish) 596 4.1. Somali: loss of place produces a dental 597 4.2. Southern French: loss of place produces a velar 598 4.3. Polish: loss of place produces a glide 599 4.4. Summary 600 5. Usually unrelated evidence II: the birth of nasal vowels (French, Portuguese, Slavic)

707 708

709 710 711 712 715 716 717

xxxiv Table of contents - detail § 601 6. Something that should not happen: German homorganic CN clusters 602 6.1. Illustration and their incompatibility with Standard Government Phonology 603 6.2. Schwa in /CaN#/ is not dropped under Government 604 6.3. Schwa is killed by the stabilising action of the nasal 605 6.4. Homorganicity has got nothing to do with adjacency 606 6.4.1. When the word-final nasal is promoted to a Coda Mirror consonant 607 6.4.2. Eignung: the nasal must not be homorganic because it stands in strong position 608 6.4.3. Eignung: the absence of schwa is mandatory because it is governed 609 6.4.4. Eignung: devoicing shows who is who 610 7. Usually unrelated evidence III: consequences for the genesis and identity of syllabic consonants 611 7.1. The analysis of syllabic and trapped consonants recalled 612 7.2. Germanic: why syllabic consonants are created only after schwa 613 7.3. Devoicing again: right-branching syllabic consonants do not qualify 614 7.4. Summary 615 8. General summary regarding nasals 616 9. Not only nasals react on positional plight: extension of the analysis to all sonorants 617 10. Conclusion: a unified theory for the behaviour of sonorants in Coda position 618 General Conclusion

Page 719 719 722 723 726 726 730 732 733 735 735 735 736 738 738 739 742 745

619 Appendices 620 1. List of parameters and their translation into CVCV and other theories 621 2. Closed Syllable Shortening vs. diminutive lengthening in Czech 622 3. Polish two-membered word-initial consonant clusters

749 753 759

Table of contents - detail

§ 623 4. A short guide to 1990 Government Phonology 624 4.1. Melodic representations 625 4.2. Constituent structure 626 4.2.1. The basic architecture 627 4.2.2. A depleted version of the familiar syllabic tree 628 4.2.3. Indirect lateral conditions on the existence of skeletal slots 629 4.2.4. No resyllabification 630 4.2.5. Proper Government, the Empty Category Principle and domains 631 4.2.6. A trademark of Government Phonology: empty Nuclei 632 References 633 Subject Index 634 Language Index

xxxv

Page 765 765 768 768 770 771 773 774 777 779 825 841

1

Editorial note: two volumes

When I set out to write this book in fall 2001,1 merely intended to make my habilitation thesis available to the English speaking audience (all academic work in France must be written in French). The initial division included three chapters that were designed to explain what CVCV is (chapter one), why it should be (chapter two) and which place is assigned to locality, morphology and phonology in this kind of theory (chapter three). A draft version of the first two chapters circulated since late summer 2002. While work on the last chapter progressed in spring 2003, it appeared with increasing clarity that the original project would not fit into one volume. The most natural seam was after chapter two: at this stage of the discussion, the reader has been fully introduced to CVCV. The remaining text, then, refines this basic model, explores its potential and positions it in regard of other modules of the grammar and its general architecture. For one thing, the system is made strictly local in the syntactic sense (Relativised Minimality): two constituents can contract a lateral relation only if there is no other constituent of the same kind intervening (locality in phonology). A consequence thereof is the "dephonetisation" of phonology or, in positive terms, the phonologisation of phonology: being a good governor or a good licensor does not depend on any phonetic condition anymore ("only phonetically expressed Nuclei can govern"). Rather, phonology alone decides: all and only those Nuclei which are ungoverned possess lateral actorship. It is also attempted to draw a red line between the area that is properly and exclusively phonological, and other domains such as phonetics and morphology, or eventual blends thereof with phonology (phonology in phonology). In a nutshell, everything that is located above the skeleton (and only this) belongs to the "immaculate" phonology and qualifies for Universal Grammar. The concept of UG must include natural language that uses non-vocal interfaces, i.e. sign language. Hence, "phon" in phonology is a misunderstanding. "Phon"ology is the study of how neuronal linguistic structure is translated back and forth to the extraneuronal world. The particular interface used is a secondary parameter that must not condition any property of the universal human capacity to translate neuronal into extra-neuronal structure (and vice versa). In this context, a number of recent neo-behaviourist raids on phonology are

xxxviii Editorial Note

examined, and it is shown why phonology, rather than syntax or semantics, is singled out for behaviourist attack. Finally, the incidence of CVCV on the representation of morphosyntactic and semantic information in phonology is examined (morphology in phonology). After a look at how higher level information has been implemented into phonology since American Structuralism, I argue for an interface which is privative, representational and translational: morphological, syntactic and semantic information must be translated into the phonological language since phonology is only able to interpret truly phonological objects. Privativity holds that only phonologically relevant information is shipped off to phonology: higher levels do not communicate with the phonological module at all in case it has been decided that a given higher level division will have no phonological effect. Whether some morpho-syntactic property is phonologically relevant or not is a sovereign decision made by the higher modules; in the minimalist perspective, Chomsky's (2000,2001 a,b) phase theory manages the mailing of postcards to the semantic and phonological interfaces. Assuming privativity, thus, morpho-syntactic information is either shipped off to phonology or not; unlike in SPE, phonology is necessarily underfed with higher level divisions. Also, there are no negative messages: a non-intervention of higher levels simply makes phonology follow its regular domestic rule. For example, specific domains across which phonological units do not "see" each other are only created upon an explicit morpho-syntactic order. In absence thereof, heteromorphemic strings are one phonologically speaking. Higher level information materialises as a modification of the phonological representation. The set of possible modifications reduces to two four (see §406): either a CV unit is parachuted, or the properties of final empty Nuclei are modified. In the latter case, final empty Nuclei can either be authorised to remain empty (which on autochtone phonological grounds they would not; they are then governed). In addition, they may be granted lateral actorship to which they do not have access according to domestic phonological rule (they then can govern and/ or license). Finally, the only portion of the representation that is accessible for morpho-syntactic modification is immediately adjacent to the morphological boundary at hand (hence spanning from the last Nucleus of the preceding morpheme to the first Onset of the following morpheme). Higher levels have no power over the morpheme-internal area, nor can they access the melody below the skeleton.

Editorial Note

xxxix

The general architecture of the grammar that this approach requires is parallel, rather than continuous: phonology is not simply the terminal structure of a big tree that begins with syntax and "hands down" information to phonology (via morphology). Rather, the different modules of grammar have a parallel organisation. According to work by Ray Jackendoff (1992,1997,2002) and Michal Starke, syntax, semantics and morphology on one hand and phonology on the other constitute two (or even three) separate worlds whose communication is not top-down. The different modules can talk to each other only via a lexical access (correspondence rules in Jackendoffs vocabulary): they send and receive postcards. This is required because phonology and the other modules do not speak the same language: while syntax, morphology and semantics all know what "plural", "case" or "gender" is, phonology is unable to interpret these concepts. By contrast, things like "labial" or "Coda" make no sense to higher level modules. This supposes that higher level information is translated from the language that is common to morphology, syntax and semantics into the phonological idiom. Such a translational process can only be achieved by a lexical access in the sense that higher levels send a signal to some dictionary, whose entry is associated with a specific phonological property on the other end. As mentioned earlier, I argue that the outlet of this translator's office on the phonological side is made of exactly four slots (further discussion is provided in §402, which actually anticipates on Volume 2). CVCV interprets syllable-based generalisations as the consequence of lateral relations that hold among segments. The resulting structure is entirely flat: there is no syllabic arborescence left at all. In a parallel perspective, nothing withstands a flat structure in phonology since different worlds may implement different architectures: the existence of a treebuilding device in syntax, semantics and morphology does not imply its replication in phonology. If arboreal structure is indeed absent from phonology, as suggested by CVCV, a long-noticed contrast between phonology and syntax falls out automatically: there is no recursion in phonology because recursion supposes arborescence (a structure is recursive iff a given node dominates a node of the same type). The foreword §2 offers further discussion of this issue. The preceding paragraphs describe the volume to come. The first volume is now in the hands of the public. When this editorial note was written (August 2004), the second volume was almost completed in draft.

xl

Editorial Note

The constant reference that is made to it here should therefore reflect its divisions quite closely.

2

Foreword

This book presents a development of Jean Lowenstamm's idea that phonological constituent structure can be reduced to a strict sequence of non-branching Onsets and non-branching Nuclei. This approach is known as "CVCV", and emerged from Government Phonology. The book is divided into two parts, which expose what CVCV is (Part One), and why it is worth considering this idea a valuable and viable approach to phonology (Part Two). The primary goal is not to locate Government Phonology in general and CVCV in particular within the contemporary or foregoing phonological scene. Before general comparisons in the popperian sense can be made, the properties of each competitor need to be known. Therefore, the present book aims at establishing a player in the game: it exposes the characteristics of CVCV as explicitly as possible. In the current OT-dominated phonological scene, then, CVCV appears as a true theory of the 80s insofar as it is representational at core: representations contribute a sovereign and unoutrankable arbitral award that is not subjected to any further computation. Structure and process are related but independent; a theory that dispenses with the autonomy of one of these poles of the natural world must fail. In other words, there is something like ill-formedness (and not just more or less well-formedness). Grammar may assess an arbitral award regarding the grammaticality of a form in complete absence of any competition with other forms. Representations exist. They are primitive, autonomous and contribute a sovereign arbitral award that owes nothing to the computational component of the grammar. Therefore, they are not the result of any competition (such as constraint interaction). Representations do not emerge; they are (see §309). The genuine research programme of Government Phonology is to build "a syntax of phonological expressions" (first page of Kaye et al. 1990).

xlii

Foreword "What is at stake here goes well beyond a mere search for interesting or suggestive similarities. Rather, if (some of) the same principles can be shown to underlie phonological as well as syntactic organisation, the idea that such principles truly express special, idiosyncratic properties of the mind (such as the kind of asymmetries typical of natural language) will be correspondingly strengthened." Kaye et al. (1990:194)

This programme was implemented by lateralising structure and causality. That is, syllable-related processes do not root in contrasting arboreal structure. Rather, they are due to lateral forces that hold among constituents. The present book further develops this line of thought: it shows that Standard Government Phonology ran out of breath half way when pursuing the lateral idea. As a result, a permanent in-between was installed: some arboreal structure and causality was lateralised, but other chunks of the traditional syllabic tree were left in place (see chapter 1,8 §165). Therefore, Standard Government Phonology is a hybrid animal. Quite some problems, many of them long-noticed (such as its inability to handle word-final consonants that behave like Codas, see chapter 11,12 §524), originate in this hybridity. For example, an important consequence of the arboreal-lateral hybridity is the redundancy of arboreal structure, something that was made crystal-clear in a largely underquoted article by Takahashi (1993) called "A farewell to constituency" (see also Takahashi 2004:141 ss). If co-occurrence restrictions are expressed in lateral terms (e.g. a branching Onset: the obstruent governs the sonorant), rather than in regular arboreal fashion (the obstruent and the sonorant are sisters of the same node), the former should take over the function of the latter. This, however, is not the picture that Standard Government Phonology draws: lateral relations cohabitate with the old arboreal structure. The latter is thus redundant. It is obvious that no theory can afford encoding the same information twice, on top of that by two devices that ought to concurrence each other. Takahashi (1993) demonstrates the redundancy of arboreal structure by simply showing what would happen if it were not there: nothing. Since it can be entirely deduced from lateral relations, Standard Government Phonology would have exactly the same face without any mention of arboreal constituency (see §209s). If one were to choose, then, between the classical arboreal expression of syllable structure and the lateral alternative, it appears that the latter is certainly to be considered the null hypothesis. For, unlike in syntax, cooccurrence restrictions in phonology are defined by the relative sonority of adjacent consonants (segments). Hence by a lateral, not an arboreal, relation between neighbours. Arborescence only enters the picture when the

Foreword

xliii

analyst translates this primary lateral reality into a secondary kind of structure. I argue that of course there is no counter-indication to do so. However, the null hypothesis ought to be lateral. The burden of proof should lie on the side of secondary approaches such as the one embodied by arboreal syllable structure (§211 details this line of reasoning). Therefore, there is good reason to complete the missing steps on the way towards a complete latéralisation of structure and causality. This is what CVCV sets out to do: it takes the lateral idea of Kaye et al. (1990) to its logical end. In the light of the preceding discussion, it appears that this is actually a condition on the survival of the lateral enterprise: hybrid models are doomed to failure. Completing the lateral programme, then, produces a result that is entirely flat: no arboreal syllable structure is left at all. Its functional load has been shifted onto lateral relations. Lateral relations are thus the central tool of CVCV: they define both syllable structure and cause phonological processes. When compared to Standard Government Phonology, their number has been shrunk quite radically (see §§136,147). On the other hand, they have been endowed with a clear functional identity. The architect of this evolution is the Coda Mirror (Ségéral & Scheer 2001a): Government and Licensing alone define syllable structure and a good deal of phonological computation.1 Also, they have a stable effect on both consonants and vowels: Government inhibits, Licensing backs up the melodic expression of the target. It this sense, CVCV may well be interpreted as a minimalist enterprise in the Chomskyan sense: too many devices that have been added over the years without any clear definition of their function in the overall structure are cut down to a minimal number: two. A programme that aims at unifying phonology and syntax and at the same time comes up with a flat structure may strike as self-contradictory. Quite obviously, no syntactician is prepared to buy anything that is flat: hierarchically ordered structure seems to be a genuine property of language. Therefore, I try to show two things in this book. First, a flat structure associated with lateral relations is not just a notational variant of the familiar arborescence. It is different in a way that produces theoretical and empirical advantages. Second, nobody has ever claimed that syntax and phonology are identical. Obviously, unifying phonology and syntax supposes that one is aware of what is similar and what is not in the first place. Nobody has ever 1

See §149 and Vol.2,1.8 on the peculiar status of Infrasegmental Government.

xliv

Foreword

suggested that every single bit of one area is replicated on the other side. Rather, the research programme at hand seeks to identify cases where phonological structure and processes, eventually against intuition and the surface mirage, have syntactic peers and hence could be unified with them. This implies the existence of areas where phonology and syntax are different in kind. One such case, co-occurrence restrictions, is discussed in §211. Kaye et al. (1990) also express a balanced view on similarities and differences between syntactic and phonological structure: "Allowing for fundamental distinctions between the objects under study in sentence syntax and in phonology, such as the recursive nature of syntactic, but not phonological categories, it is conceivable that some of the same principles at work in syntax will be seen to be operative in phonology, and •vice versa." (emphasis in original) Kaye et al. (1990:193) When comparing both areas of grammar, then, the most prominent difference that springs to one's eye is certainly the one mentioned by Kaye et al. (1990): there is no recursion in phonology. It is interesting to note that this hard fact, which is a long-standing observation, actually follows from flat structure: if, as I argue, phonology lacks a tree-building device (i.e. Merge in the minimalist vocabulary), there could not be any recursion. For recursion is defined as a node that dominates a node of the same type. A phonological landscape along these lines is also consistent with the general picture that is drawn by Chomsky et al. (2002), who make a difference between the faculty of language in the broad (FLB) and in the naorrow sense (FLN). The latter is the abstract linguistic computational system; it "comprises only the core computational mechanisms of recursion as they appear in narrow syntax and the mappings to the interfaces" (p. 1573). More technically speaking, thus, FLN is made of Merge and Phase. FLB, on the other hand, includes FLN and the two interfaces themselves: the phonological and the semantic module (which the authors call "sensory-motor" and "conceptual-intentional" systems, respectively). Chomsky et al. (2002) argue that FLB is shared with animals, while FLN is uniquely and specifically human. Or rather, to be precise, their only claim is of diachronic nature: the present-day human FLB may have evolved through a series of gradual modifications on the basis of animal pre-homo FLB. The FLN may not. It is a device which specifically and uniquely characterises human communication, and therefore has no biological basis in any non-human ancestor. It must have emerged during the times when

Foreword

xlv

the homo species was alone in its evolutionary branch, i.e. in the past six million years or so. FLN is a human invention, FLB is not.2

2

Chomsky et al. (2002) is often condensed into "the only thing that UG contains is recursion (i.e. Merge)", suggesting that no phonological property could be part of UG and, worse, that there is nothing to be shared by syntax and phonology. This would then be the precise expression of neo-behaviourist stances such as Carr's (2000): phonology lies outside of U G It is important to understand that this interpretation is incorrect. Chomsky et al. (2002) are explicitly agnostic with respect to this issue: "Liberman and his associates [...] have argued that the sensory-motor systems were specifically adapted for language, and hence should be considered part of FLN. There is also a long tradition holding that the conceptual-intentional systems are an intrinsic part of language in a narrow sense. In this article, we leave these questions open, restricting attention to FLN as just defined but leaving the possibility of a more inclusive definition open to further empirical research" (p. 1571). On another occasion, they even include FLB into those properties that make human communication specifically human: "we take as uncontroversial the existence of some biological capacity of humans that allows us (and not, for example, chimpanzees) to readily master any human language without explicit instruction. FLB includes this capacity, but excludes other organism-internal systems that are necessary but not sufficient for language (e.g., memory, respiration, digestion, circulation, etc)" (p. 1571). On page 1573, however, they say that the strongest form of their hypothesis holds that "all peripheral components of FLB are shared with other animals, in more or less the same form as they exist in humans, with differences of quantity rather than kind". This is all consistent with the diachronic focus which, recall from the main text, is the only purpose of their article: it may well be true that FLN is the only part of human language that has been "invented" by humans alone; that FLB is some kind of evolved version of the old animal FLB. The synchronic characteristics of UG as implemented in the human genome, however, do not care for how they have come into being: they may well possess properties that have been invented by the homo family and others that have an animal ancestor - both sets may be genetically encoded and conjointly produce the effect that mankind speaks, while animals do not. In other words, the old animal FLB that humans have inherited may have evolved in such a way that its present version is quite different from the shared animal-human ancestor. Different enough to include specifically linguistic features to which animals, through their unevolved FLB, do not have access. The present-day human genome, then, contains a set of specifically linguistic properties of two different evolutionary origins: some are based on the common animal ancestor, others are "human inventions". In any event, at least some features of FLB are part of the present-day UG, hence of the human genome.

xlvi

Foreword

Hence, the critical difference that Chomsky et al. (2002) establish between syntax on one hand and phonology/ semantics on the other precisely concerns recursion: following their logic, any adequate phonological theory must be unable to produce recursive structure. One way to do that - a radical way - is actuated by CVCV: there is no recursion in phonology because there is no tree-building mechanism in this module. If the minimalist philosophy regarding Merge is taken seriously, this is actually a necessary consequence: since Merge is responsible for treebuilding, hence for recursion, eliminating the latter means to eliminate the former. Or, in other words, there could be no non-recursive tree structure on minimalist assumptions. Either Merge is active and results in both arboreal structure and recursion, or it is not, and none is produced. CVCV arrives at this result "from the other end", and for entirely independent reasons that root in the original research programme of Government Phonology, i.e. the latéralisation of structure and causality. This is to say that the flat result of CVCV is at the same time the consequence of the programme that attempts at unifying syntax and phonology, and the grounds on which the most fundamental difference between both modules may be understood. It goes without saying that Government Phonology is by no means the only or the first theory that attempts at accounting for phonological and syntactic phenomena with the same set of principles. Dependency Phonology, a sister theory not only in this respect, has a genuine tradition and an longer experience in this area. John Anderson (1985,1986,1987,1992, Anderson & Ewen 1987:283ss among others) has always promoted what he calls structural analogy·, in the formulation of Hulst (2000:209), "grammar recapitulates, rather than proliferates, structures and principles". Further work that brings together syntax and phonology includes Riemsdijk (1982), Sauzet (1996,1999) and Michaels (1991,1992). Before moving on, a disclaimer is in order: this book sets out only to flatten syllable structure. Other types of supra-skeletal units that are known from Prosodie Phonology since Selkirk (1984a) and Nespor & Vogel (1986) such as feet, prosodie or phonological words, phonological phrases and the like are not its primary focus. Only chapter 11,11 (§501) argues that nothing else than flat CVCV is needed in order to implement the parameter known

The issue of whether there is some phonology in UG will be addressed at length in Vol.2,II where neo-behaviourist raids on phonology are discussed.

Foreword

xlvii

as Weight by Position, and to explain why Onsets are weightless. Whether higher areas of phonology need to be represented by arboreal structure or not is a question open to further debate. My intuition is that they do not. In any event, this question is discussed at length in Vol.2,ΙΠ where the relation between phonology and other modules of the grammar is addressed. Let us now turn to some features regarding the internal organisation of the book (see also §3). For the sake of better legibility, a stenographic overview of the melodic and syllabic properties of Standard Government Phonology is provided in appendix 4 (§623) (and also in a nutshell at the outset of Part One in §§10s,15). These pages are designed to serve as a shortcut to relevant information for the reader who is less familiar with certain aspects of the theory as it stood in 1990. Moreover, the relation of CVCV with Standard Government Phonology is established in some detail in chapters 1,7 (§135) and 1,8 (§165). The reader who is accustomed with earlier versions of Government Phonology will be put in a position to judge the differences. People coming from other horizons can gain an impression of the genuine properties of Standard Government Phonology, of its tools (Government and Licensing) as well as of their evolution. In any event, historical information is not central in scope and does not represent a condition on the understanding of the book. Therefore, its core is not located at the outset of the text. While chapters 1,7 (§135) and 1,8 (§165) concentrate relevant discussion, they do not exhaust the matter. Rather, reference to earlier versions of Government Phonology is made throughout the entire book whenever this suits the demonstration. It was mentioned earlier that the same holds true for the position of CVCV with respect to other phonological theories. This not withstanding, Part Two is the natural location for some comparative discussion since it is designed to explain why CVCV is worth to be considered a serious competitor in the field. Arguments are drawn from the comparison of solutions for particular phenomena that are proposed by CVCV and other approaches. These range from traditional Kahnian syllabification algorithms over Lexical Phonology and Standard Government Phonology to Optimality Theory. I have tried to focus the discussion on theory-neutral tools that were developed by phonologists over the past 30 years and have become common theoretical background. These include concepts such as branching Onsets, Coda-Onset sequences, disjunctive contexts, bogus clusters, sonority, extrasyllabicity, syllabification algorithms, resyllabification, morae, autosegmental representations and the issues related to the debate on abstractness vs. concreteness.

xlviii

Foreword

It is also worth mentioning that the representational orientation of Government Phonology in general and of CVCV in particular leads to the development of precise identities for basic phonological objects and processes. These include Codas, closed syllables, long vowels, geminates, syllabic and trapped consonants, the beginning and the end of the word, vowel-zero alternations, Closed Syllable Shortening, Tonic Lengthening, compensatory lengthening, lenition and the like. Throughout the book, each issue is examined with particular attention to diachronic evidence in its relation to the synchronic state of affaires. Needless to say, as well: the phenomena and languages discussed represent but an arbitrary choice that reflects my personal interests and the empirical field that I am best familiar with. As far as languages are concerned, this translates as a focus on (Western) Slavic, French, German and some Semitic. Finally, attention needs to be drawn on the fact that the view expressed in this book represents only one possible implementation of Jean Lowenstamm's idea. Other interpretations of CVCV that may or may not be compatible with the present approach in whole or in part include the following. Szigetvári (1999a,2001) (also Dienes & Szigetvári 1999) advocates strings that obey strict CVCV, but begin with a Nucleus and end in an Onset (VC skeletons). Rennison (1999b) and Rennison & Neubarth (2003) develop an x-bar theory that roots in CVCV, and Brandäo de Carvalho (2002a) operates with a "double CVCV helix". Cyran (2001,2003) abandons Proper Government or any other lateral device for the description of vowel-zero alternations. Finally, Rowicka ( 1999a,b,2001) holds that lateral relations are head-initial rather than head-final, while Polgárdi (1998,1999, 2002,2003) rejects final empty Nuclei ("loose CV"). Also, the latter two implementations of CVCV share the view that Government Phonology and Optimality Theory are not incompatible. This is certainly true since OT, in its own understanding, is a metatheory that can work with any input (linguistic or not: "theory X with an OT-top"). Encouraged by the possible marriage of both theories, the two approaches mentioned operate with various OT-type constraints that apply to Government Phonology representations. While writing this book, I was lucky enough to be able to spend some time in Warsaw (repeatedly, thanks to Jerzy Rubach), Leipzig (also repeatedly, thanks to Gerhild Zybatow), Lublin (thanks to Gienek Cyran) and Brno (thanks to Petr Karlik). These stimulating environments have

Foreword

xlix

greatly contributed to the venture, and actually quite some text was produced there. Another important source of inspiration have been the EGG Summer Schools (Central European Summer School in Generative Grammar, coolschool.auf.net), particularly the editions in Plovdiv/ Bulgaria (1999), Nis/ Serbia (2001), Novi Sad/ Serbia (2002), Lublin/ Poland (2003) and Cluj/ Romania (2004). Various parts of the book have been "tried out" in classes that I have taught at EGG, and other people's classes as well as further discussion at the school have greatly contributed to the result that now appears in print. The following people have generously spent time and energy in order to discuss various drafts with me: Klaus Abels, Petr Biskup, Sylvia Blaho, Katalin Balogné Bérces, Joaquim Brandäo de Carvalho, Monik Charette, Jean-Philippe Dalbera, Pierre Encrevé, Edmund Gussmann, Tracy Hall, Daniel Huber, Harry van der Hulst, Jonathan Kaye, Artur Kijak, Ursula Kleinhenz, Ondra Kockour, Kristina Krchñava, Ivona Kuóerová, Laszlo Krísto, Jaromír Nohavica, Gábor Oláh, Karel Plíhal, Stefan Ploch, Curt Rice, Jerzy Rubach, Jaroslav Samson Lenk, Eirini Sanoudaki, Philippe Ségéral, Michal Starke, Péter Szigetvári, Marianna Tóth. The text owes much improvement to them. Thanks a lot for the fruitful exchange. Le Boulou, August 2004

3

How to use this book

This book is not a textbook. It does not aim at being pedagogical: there is no linear progression (in the sense that you have to have read through chapter n-1 in order to understand chapter n), issues are not presented by a "neutral" or "impartial" observer who does not personally support either of the views discussed, and of course there are neither exercises nor learn-byheart summaries. On the contrary, this book is written from a partial point of view: the one of Government Phonology in general and of CVCV in particular. The goal is to demonstrate that CVCV is worthwhile; all the rest follows from this premise. The book is thematically organised. Most probably, il will therefore be best used like a dictionary: you want to know what CVCV says about X, so you look it up. Provisions have been made to facilitate this look-up function. A fairly detailed subject index (§633) is available at the end of the book, and a language index (§634) refers to the languages mentioned and offers a list, language by language, of all individual alternations discussed. Two appendices also enhance the look-up function: appendix 1 (§620) lists and references all parameters that have been discussed, and a short guide to 1990 Government Phonology is offered in appendix 4 (§623) (cf. the foreword §2). All sections and sub-sections are identified by a running number in the page margin, the paragraphs § (see the general conventions §5 for details). All cross-reference in this book (from the main text, from footnotes, from indexes) targets this running number. Within §s, the reference system may point towards relevant thematic units. These are identified by alphabetic characters after the § number. For example, §476c refers to the third paragraph of §476. When following a cross-reference, thus, the reader does not need to go back to the table of contents in order to identify the page number and finally look up the page: he can jump directly to the running number. Each chapter has a thematic unity and may thus be accessed independently. This, I admit, is less true for the very beginning of the book, i.e. chapters 1,2 (§14) to 1,6 (§110), where the bare skeleton of CVCV is exposed. Once this system is understood, however, the order in which the remaining chapters are approached does not matter a lot. Given this dictionary-like organisation, I do not really expect anybody to read through the entire book from the first to the last page.

Iii

How to use this book

Also, the rather scary size that the text has grown into should not be dissuasive for that reason. Finally, the relative thematic independence of the chapters supposes a particular effort for making information from other chapters available. As was mentioned before, I have tried to meet this challenge by using constant cross-reference and offering fairly detailed (and thematically organised) indexes and appendices. Another means of prompting relevant information are short thematic summaries. These appear whenever I found that they may enhance the reader's task, at the risk of some repetition and redundancy here and there. To round off this practical description, it is useful to be aware of the following fact: the book which you hold in hands has changed quite a bit with respect to the various drafts that have circulated over the past two years. The comments that I was lucky enough to receive, as well as a thorough final revision of the text have eliminated a number of errors (though I apprehend in advance all those that have managed to seep through) and prompted minor, but also some major changes in all areas: organisation, presentation, style and content.

4

Conventions used in this book

5

1. General conventions: symbols, languages, cross-references The following general conventions are used in this book.

in text and tables V C Τ R L [Χ] Iiyjl

any vowel any consonant any obstruent any sonorant any liquid X is a phonetic object rendered in IPA transcription. X is a lexical object (as opposed to an intermediate representation /X/ which is the result of a derivation, but not a phonetic object). /Χ/ X is a phonological and/ or underlying object (in case it is used in relation with //...//, it is an underlying non-lexical form).

X is a graphic object that appears as such in spelling/ script §79 the running number in the page margins are paragraphs. Crossreferences in this book are only to these paragraphs, number 79 in this example. Beware·, reference to a § that identifies the beginning of a chapter or a section refers to this chapter or section and to all of its sub-sections. §§45-48 reference to paragraphs 45, 46, 47 and 48. §§45,48 reference to paragraphs 45 and 48. §§45s reference to paragraph 45 and the following paragraph. §476c an alphabetic character after a § number indicates the number of the paragraph within the § at hand. For example, §476c refers to the third paragraph of §476. chapter 1,7 (§135) reference to an entire chapter; in the example, to chapter seven of Part One that begins at §135. Vol.2,II.3.5 reference to volume two, here to section 11,3.5. 18s, 167ss reference to a particular location in a book or article: page 18 and the following page, page 167 and the forthfollowing pages FEN Final Empty Nucleus (Nuclei) ECP Empty Category Principle CSS Closed Syllable Shortening

liv

Conventions used in this book

in text and tables SGP Standard Government Phonology HIGH, (BE)LOW occurring in the area above/ below the skeleton (see §§215s,229,517) UG Universal Grammar NGP Natural Generative Phonology in representations PG Gvt Lie Gvt-Lic IG 10 Gvt CG, ICG CL, ICL y í,y

u V

comment

V

J, 3 ç, 3

tç, d? Ç>J_ ts, dz

? Í.3

ch h

\

ή ó

IPA

Czech doesn't have this series Czech doesn't have this series Polish doesn't have this series there is no voiced version in Czech there is no voiced version see explanations above (also for Czech )

X

fi w J1 u i ii uu VV

h is voiced in Czech

nasal vowels, see explanations above see explanations above see explanations above see explanations above see explanations above

PART ONE: WHAT IS C V C V ?

Chapter 1 Introduction

1. CVCV in a nutshell CVCV holds that syllabic constituency boils down to a strict consecution of non-branching Onsets and non-branching Nuclei in all languages. There are no Codas and no Rhymes, and the minimal syllabic unit that may be manipulated is an Onset followed by a Nucleus: the existence of the former implies the latter and vice versa. For the sake of clarity, the constituent structure of some basic phonological objects appears under (1) below. (1)

closed syllable Ο Ν Ο Ν I c

! ν

c

I l 0

geminate Ο Ν Ο Ν

long vowel Ο Ν Ο Ν

[,..C#] ... O Ν C

0

"branching Onset" Ο Ν Ο Ν Τ

0

R V

This is the core of the research programme laid out in Lowenstamm (1996).3

2. Some core properties of Standard Government Phonology The structures under (1) are the ultimate consequence of the line of thought that was initiated by Kaye et al. (1990) and Kaye (1990a), which is known 3

Work that has followed this idea includes Barillot (1997,2002), Bendjaballah (1998,1999), Brandäo de Carvalho (2002a), Creissels (1989), Cyran (2001,2003a,b), Dienes & Szigetvári (1999), Guerssel & Lowenstamm (ms), Heo (1994), Hérault (1989), Lahrouchi (2001), Larsen (1994,1998), Lowenstamm (1988,1999,2002,2003), Nikiema (1989), Pagliano (1999,2003), Polgárdi (1999,2002,2003), Rizzolo (2002), Rennison (1999b), Rennison & Neubarth (2003), Rowicka ( 1999a,b,2001), Scheer (1996,1997,1998a,b,1999a,b,c, 2000a,b), Ségéral (1995,1996,in press), Scheer & Szigetvári (2002), Ségéral & Scheer (2001a), Seigneur-Froli (2001,2003,in press), Szigetvári ( 1999a,b,2001).

2

1,1

Introduction

as Government Phonology. Kaye et al. (1990) have severely restricted syllabic arborescence in claiming that all constituents are maximally binary, except the Coda that can only dominate one skeletal slot. 4 At the same time, they have given a theoretical status to empty Nuclei (which were formally introduced by Anderson 1982), whose existence is systematically supposed in word-final position and in locations where vowels alternate with zero. Moreover, phonological processing is supposed to be structure-preserving, which means that nothing can ever be resyllabified. Strings are fully syllabified in the lexicon, and hence there is no such thing as a syllabification algorithm. In order to be pronounced, a segment must pertain to a syllabic constituent, on which the usual restrictions regarding sonority rely. 5 Ordered rules (but not serialism as such) are declared undesirable: phonological processes apply whenever (and only when) their environment is met (Kaye 1992b: 141, 1995:291). 6 The two latter statements exclude all versions of extrasyllabicity: nothing may have a phonetic existence without belonging to a syllabic constituent, a segment may not be attached to anything else (such as the phonological word and the like) than a skeletal slot, all skeletal slots are dominated by a syllabic constituent, nothing can be underparsed because there is no such thing as a syllabification algorithm, and "late" incorporation into syllable structure is outlawed by the absence of rule ordering. Constituents are Onsets, Rhymes and Nuclei, but not "Appendices" or "Terminations". In a further step, Kaye (1992a) has introduced the idea that a word may also begin with an empty Nucleus. Although this kind of configuration was restricted to a very peculiar subset of words (those beginning with s+C clusters), it is the first expression of a structure that will later be generalised by Lowenstamm (1999) for very different reasons.

4

5

6

See appendix 4 (§627) on the real status of the lingua franca term "Coda" in Standard Government Phonology: Rhymal Adjuncts do not enjoy constituent status at all. Sonority has no autonomous status as a phonological prime in Government Phonology. In early versions of the theory, it has been expressed by a notion called Charm, which was gradually replaced by complexity in later analyses. This issue does not bear on the present discussion. It is addressed at some greater length in §37,54s. Vol.2,III. 1.2.4.2, Vol.2,III.3.5.2 and note 166 develop the questions related to serialism at greater length.

Syllabic arborescence derives from lateral relations 11

3

3. Syllabic arborescence is not primitive: it derives from lateral relations among segments A crucial property of the syllabic model initiated by Kaye et al. (1990) is the fact that co-occurrence restrictions among neighbouring segments do not stem from the arboreal structure itself. Rather, the syllabic arborescence is demoted to a derived status. It follows from lateral relations that hold among segments. In traditional syllabification, the fact that a sequence such as, say, [pr], but not [φ], qualifies as a branching Onset is ascribed to a constraint on this constituent: "within branching Onsets, sonority must increase". As was mentioned in note 5, sonority is not a primary phonological category (there is no [±son] feature or the like) but follows from the number of primes that participate in the definition of a segment (complexity). Based on complexity, segments contract disbalanced lateral relations where one member governs the other. Governing abilities depend on the complexity (or Charm) value of each item. In this view, [pr], but not [rp], qualifies as a branching Onset because [p] is more complex than [r] and may therefore govern it; the reverse governing relation is excluded. Arboreal syllable structure is only a secondary effect of this kind of governing relation: the head of a branching constituent must be able to govern its dependent, and constituents are head-initial. This disbalanced head-dependent interaction holds also true for relations among segments that belong to different constituents: the Coda is naturally weak and therefore may never be the head of a relation. Its being dependent requires the presence of a head, that is the existence of a following Onset. This Onset must be able to govern its complement, the Coda. If there is no governing Onset available on the righthand side of the Coda, there can be no Coda. This is the essence of Kaye's (1990a) Coda Licensing principle. Its most straightforward effect is the absolute ban on word-final Codas. The technical detail of this system is not important here. It only needs to be understood that already Kaye et al. (1990) did not view syllabic arborescence as a primary object: it is derived from the lateral relations that hold among segments. These relations instantiate the dependency pattern: one member of the lateral domain is the head, the other is the complement.

12

4. If lateral relations are primary, syllabic arborescence has to go The philosophy of CVCV is to take this conception to its logical end: if syllabic arborescence is not primary, if it can be read off the lateral rela-

4

1,1 Introduction

tions that hold among segments, and if all effects that are usually understood as a consequence of arboreal syllable structure (Closed Syllable Shortening etc.) may be described in terms of lateral relations, then it must be concluded that there is no syllabic arborescence at all: what would it be good for? It is self-evident that no theory can afford to host two parallel structures with identical scope and function. In case one is primary and the other derived, the consequence is clear: the derived one is to be done away with (§§209s enlarge on this issue). But even when considering the competition between non-hybrid approaches where syllabic regularities are exclusively encoded by either arboreal or lateral structure, I argue that the null hypothesis must be lateral (§211). This follows from the fact that the observational reality on which all syllabic generalisations build is lateral in kind: the relative sonority of two neighbouring consonants is the angle stone of the causal chain which classically leads to arboreal structure. Arboreal structure thus is a mere projection of lateral information. While, of course, such a transformation of the primary lateral reality into some other kind of structure is perfectly legitimate, it must support the burden of proof: encoding the lateral observation by lateral structure is the first thing to be thought of. Any other move needs special argumentation. The resulting picture is also a fairly accurate description of the research programme that is laid out here: CVCV attempts at expressing all syllabic functions and syllable-related processes in terms of lateral relations, rather than with appeal to any kind of syllabic arborescence. As was mentioned in the foreword §2, this research programme is actually identified on the first page of Kay e et al. (1990), but Standard Government Phonology ran out of breath half way when implementing it: various nests of arboreal structure are maintained (see §172). The critical question regarding the redundancy of arboreal syllable structure was raised by Takahashi (1993) as early as in 1993: his article is entitled "A farewell to constituency" (see also Takahashi 2004:141 ss). Unfortunately, his ideas did not have any sensible echo. This issue is further discussed in §209.

13

5. Roadmap of Part One The present part of the book sets out to introduce the representation of some basic phonological objects and processes in CVCV. Among those are closed vs. open syllables, long vowels, geminates, the beginning and the end of a word, lenition and fortition, Closed Syllable Shortening, Open

Roadmap of Part One

5

Syllable (Tonic) Lengthening, compensatory lengthening and the like. It does not aim at writing a history of Government Phonology since the mid80s. Historical excursuses are made only when a CVCV structure may not be understood without its non-CVCV predecessor. For instance, this is the case for the status of lateral relations that have been discussed above. Some historical elements are thus weaved into the text on various such occasions, with a concentration in chapters 1,7 (§135) and 1,8 (§165). The only goal of this part of the book is to introduce the properties of CVCV: "have a look at how this works". I do not intend to compare CVCV to any concurrent approach here. As was stated in the foreword, the overall goal of this book is not to locate Government Phonology in general and CVCV in particular within the contemporary or foregoing phonological scene. Before being able to operate general comparisons of phonological theories in the popperian sense, the properties of each competitor need to be known. This does not exclude, however, some scanty reference to other phonological approaches when relevant to the discussion. In this part and throughout the entire book, I often refer to "Standard Government Phonology" (SGP). Since CVCV is an outgrowth of Government Phonology, I add the qualification "standard" in order to address the model that was briefly described earlier in this section: Kaye's (1989) foundation-laying book, Kaye et al. (1985,1990) and the following work carried out in the 90s within this frame such as, among others, Kaye ( 1990a,b, 1992a,b, 1995), Harris (1990,1994a), Charette (1990,1991), Gussmann & Kaye (1993), Gussmann (1997,2002). More recently, a tendency can be observed among representatives of Standard Government Phonology whereby the branching structure is more and more depleted by the introduction of additional empty Nuclei (or empty Onsets). Ritter (1995,1998) is a case in point: she argues that Hungarian does not possess any branching Onsets. All TR clusters of the language enclose an empty Nucleus (Ritter 1995:50ss). She also contends that long vowels in Hungarian are made of two independent Nuclei which enclose an empty Onset (Ritter 1995:12Iss, 1998). Gussmann (2003) probably goes as far as one can in this direction. For the particular case of Icelandic, he concludes that yes, there are some branching Onsets left, but only just. The basic concepts of Standard Government Phonology are summarised in appendix 4 (§623). In order to introduce CVCV, I have chosen the syllabic concept that is certainly the most fundamental of all: open vs. closed syllables. How can a theory that denies the existence of the arboreally defined object "Coda"

6

1,1 Introduction

encode this opposition? The following section develops an answer to this question.

14

15

Chapter 2 Open versus closed syllables in CVCV

1. The Empty Category Principle, Proper Government and vowel-zero alternations in Standard Government Phonology In comparison to standard syllable structure, CVCV multiplies the number of empty constituents, and namely that of empty Nuclei. This situation raises the more general question regarding the status of empty categories in linguistic theory. It is commonly agreed that "you cannot get an empty category for free". This idea is expressed in the Empty Category Principle, which states that an empty category may remain unexpressed if and only if precise conditions are met. These conditions are defined in terms of the relation that the empty category contracts with a filled position at some lateral distance. In earlier versions of generative syntactic theory, it was proposed that movement could only take place if the trace of the moved object in the empty base-position is properly governed by the object in its new position. Proper Government was defined as a structural relation between the filled and the empty position, subject to certain locality conditions (c-command, barriers). This example from syntax provides the kind of motivation that is typical for the existence of empty categories. If there were no structure preservation, i.e. if the category the object was moved from were deleted or not even lexically present, no explanation of agrammaticality in terms of movement along these lines would be available. Neither would it be if there were no empty categories.7 Empty categories burden the grammar because they require special care (defined e.g. as Proper Government). Nevertheless, their existence is a necessary condition for explanation. Hence, the burdening of the grammar by more empty categories should not be viewed as an undesirable overload, but on the contrary as a welcome source of explanation. If grammar is not free in its moves because it must create or maintain the conditions required for the existence of empty categories, a step towards a more constrained theory is made. As for all other scientific theories, the challenge, then, is to

7

Rennison (1994) offers more discussion regarding the parallel between empty Nuclei in phonology and empty positions in syntax.

8

1,2 Open vs. closed syllables in CVCV

propose a model that is as constrained as possible while covering all relevant data.

16

1.1. Vowel-zero alternations and the Empty Category Principle The same reasoning holds for phonology. Kaye et al. (1990:219) have proposed phonological Proper Government which is also based on lateral longdistance phenomena involving an empty and a filled category.8 In their view, empty categories are subject to the ECP in phonology as well as in syntax. An adapted version of their phonological ECP is given under (2). (2) Empty Category Principle (ECP) - version one An empty Nucleus may remain unexpressed iff it is properly governed.

The long-distance phenomena at hand are vowel-zero alternations. These are typically sensitive to the object(s) which are found between the zero (empty Nucleus) and the vowel (filled Nucleus) that occurs to its right. Vowel-zero alternations show a remarkably stable pattern in various genetically unrelated languages. A cross-linguistic sample is provided under (3).9

8

9

Another important concern that led to an explicit and restrictive formulation of the ECP was to limit the proliferation of empty categories, whose number and location could otherwise grow out of control. See Dresher (1985) for early discussion of this problem that inevitably arose with the advent of autosegmental structures. Ewen & Hulst (2001:188s) also discuss this motivation for the ECP. Vowel-zero alternations are analysed in greater detail in chapters 1,3 (§69) and 1,4 (§76). The data shown may be controlled for example in Scheer (1996,1997) for Czech, Gussmann (1980a), Rubach (1984) for Polish, Rubach (1993) for Slovak, Nikiema (1989) for Tangale, Kaye (1990b), Barkaoui (2000) for Moroccan Arabic, Barillot (1997) for Somali, Törkenczy (1992), Siptár & Törkenczy (2000:214ss) for Hungarian, Kenstowicz & Kisseberth (1979:175sq) for Hindi, Clements (1993:13 lss) for Kolami and Wiese (1996), Noske (1993) for German.

The ECP, PG and vowel-zero alternations in Standard GP (3)

Moroccan Arabic German Tangale (Chadic) Somali (Cushitic) Turkish Slavic (e.g. Czech) Hungarian Hindi

vowel C_C-CV kittib

gloss

inner dobe

write pf act 3pl, 3sg, 3sg causative inner-lich inner+infl, inner, internal dobu-n-go called, call, called me

nirag-o

nirig

nirig-ta

dev0r-i lok0t-e

devir loket

devir-den loket-ni

young female camel pi, sg indef, sg def transfer ACC, NOM, ABL elbow GENsg, NOMsg, adj.

zero C_C-V kit0b-u

vowel C_C-0 k0tib

inner-e dob0-go

maj0m-on majom kaarek-öö kaarak

Kolami (Dra- kin0k-atun kinik vidian)

9

majom-ra

monkey Superessive, NOM, Sublative kaarak-nee "case" oblique pi, NOMsg, agentive kinik-tan "break" present, imperative, past

Languages vary as to whether the elision is optional (e.g. German) or obligatory (other languages quoted), and with respect to the quality of the vowel(s) concerned. However, the phonotactics of vowel-zero alternations are stable in all systems mentioned. The correct descriptive generalisation is as under (4). (4) alternation sites show a. zero / CV # b. vowel / C Hence, vowels appear before Codas since {#,C} describes the Coda context. In other words, zero surfaces in open syllables, while a vowel is observed in closed syllables. This is the classical interpretation of the facts. It supposes resyllabification: the root-final consonant in a Czech word such as loket "elbow NOMsg" is a Coda in the NOMsg, but pertains to an Onset in GENsg lokot-e "id. GENsg". As was mentioned earlier, resyllabification is outlawed in Government Phonology. Before considering how vowel-zero alternations can be captured in a system that does not allow for resyllabification, the reasons that have led to this syllabic rigidity are discussed in the following two sections.

10 17

1,2 Open vs. closed syllables in CVCV

1.2. The Projection Principle, empty Nuclei and Government Empty Nuclei are a central and genuine feature of Government Phonology. They have been proposed outside of and prior to this particular theory (Anderson 1982, Spencer 1986), but only Government Phonology has given them a theoretical status. Empty Nuclei have been developed in Government Phonology for reasons that are of theoretical, rather than empirical nature. It was mentioned earlier that the research programme of Government Phonology since its early days is to build a syntax of phonology, and to use central concepts form syntactic theory.10 This attitude follows Dependency Phonology, where the Structural Analogy Hypothesis is a landmark since Anderson (1985,1986,1987,1992) and Anderson & Ewen (1987:283ss). In Hulst's (2000:209) words, "grammar recapitulates, rather than proliferates, structures and principles" (see also Durand 1995). By this token, the notion of c-command was used in order to explain the ill-formedness of super-heavy Rhymes (so-called Prosodie Government, see Lowenstamm & Kaye 1986:115, Lowenstamm 1989 and §179). Ritter (1995:36s) suggests to extend the notion of head-operator to the syntactic relation between the predicate and the subject (see also Ritter 1995:152s,237,1997, Hulst & Ritter 1999b). Phonological versions of the Minimality Condition (Charette 1989, Harris 1994a: 169s, Kaye et al. 1990:224s) and the Projection Principle were implemented. The latter requests that "governing relations are defined at the level of lexical representation and remain constant throughout a phonological derivation" (Kaye et al. 1990:221). According to Kaye et al. (1990), syllable structure is a consequence of the governing relations that hold among consonants (see appendix 4 §626 for greater detail). Therefore, ^syllabification is prohibited: a melodic object that is "born" in a Coda cannot surface in an Onset (and vice versa) because this would alter the corresponding governing relations. Another consequence of the Projection Principle is Structure Preservation (also known from syntax): a syllabic constituent is not deleted in case the associated melody, for one reason or another, should happen to be phonetically absent. Facing a typical Slavic alternation such as the Czech case mentioned under (3) loket vs. lokt-e "elbow NOMsg, GENsg", [k] and [t] belong to two independent Onsets in both forms. Did they not, i.e. were 10

Harris (1994a:187ss) and Rennison (1994) provide a more detailed discussion of syntactic concepts and their translation into Government Phonology. Note, however, that nobody has ever claimed that there is, that there should be, or that there could be, total superposition of phonology and syntax: see the discussion of recursion in the foreword §2.

The ECP, PG and vowel-zero alternations in Standard GΡ

11

the [k] resyllabified into the Coda of the preceding vowel in the GENsg lokt-e, the governing relations holding among both consonants would be altered in the course of the derivation. Therefore, all alternations of a segment with zero exclusively concern the melodic part of the autosegmental representation: constituent structure remains untouched. This means that in Czech lokt-e "elbow GENsg", [k] and [t] belong to two independent Onsets which enclose an empty Nucleus. The distribution of empty Nuclei is regulated by the Empty Category Principle that was introduced in the previous section. The lateral relation it appeals to, Proper Government, relates two Nuclei in a disbalanced fashion: one acts as the governor (e in loket-e), the other is the governee (0 in lokot-e), whereby governors need to possess a phonetic existence. (5)

Proper Government relation between a filled and an empty Nucleus Czech "elbow" a. lokt-e GENsg b. loket NOMsg PG

Ο I X I I

Ν I X I I 0

Ο Ν Ο Ν I I I I X X X X I I k t e

Ο I X I 1

Ν I X I I 0

Ο Ν Ο Ν I I I I X X X X I k t

This typical dependency relation has given its name to the theory: Government Phonology. If syllabic representations are marshalled by Structure Preservation, how can Government Phonology handle the fundamental pattern of vowelzero alternations discussed under (3)? In order to be able to introduce the account that relies on Proper Government, another consequence of Structure Preservation needs to be discussed in the following section: final empty Nuclei.

18

1.3. Final empty Nuclei and Coda Licensing Kaye (1990a) has examined so-called Closed Syllable Shortening: a long vowel shortens if a Coda consonant is present in its Rhyme. Consider the

12

1,2 Open vs. closed syllables in CVCV

following pattern from Yawelmani (which is discussed at greater length in § 153).11 (6)

Yawelmani vowel-zero alternations I non-future gloss dubitative to help RvR ?aamil-hin ?aml-al doolul-hun to climb doll-al saalik-hin to wake up RvT salk-al mooxil-hin TvR moxl-ol to take the scent hotn-ol hootin-hin to fall asleep woo?uj-hun wo?j-al

As may be seen, CWCvC-CV sequences such as [?aamil-hin] alternate with CVCeC-VC [?amel-al] in this language. If it is true that vowelzero alternations imply the existence of a non-deleted empty Nucleus even in case zero surfaces, the shortening process may not be ascribed to the presence of a Coda in the Rhyme of the shortened vowel since the alleged Coda ([m] in the example used) is the Onset of an empty Nucleus. In addition, the two consonants that flank the empty Nucleus, and which constitute a Coda-Onset sequence on the regular account, do not show any distributional restrictions. This is unexpected since Coda-Onset interludes usually exhibit a falling sonority profile RT or a sonority plateau RR, TT, but certainly not a rising sonority slope TR (§154offers further discussion of this issue) Therefore, Kaye (1990a) concludes that "Closed Syllable Shortening" has got nothing to do with closed syllables in Yawelmani. Rather, long vowels are shortened iff the following Nucleus is empty. This analysis has interesting consequences since vowels are also shortened before word-final consonants. This is demonstrated by the examples under (7) below. (7)

11

Yawelmani vowel-zero alternations II dubitative non-future predicative pana-1 panaa-hin panaa-xa ?ile-l ?ilee-hin ?ilee-xa

gloss arrive fan

I use lower case "v" in order to identify the location of vowel-zero alternations. The data have been collected from Kaye (1990a) and Kenstowicz & Kisseberth (1977:9-12,34ss,1979:77ss). Note that the suffixes expressing dubitative and non-future value are harmonic. The literature on Yawelmani that I had a chance to access does not identify the meaning of the oft-quoted root moxl-.

The ECP, PG and vowel-zero alternations in Standard GP

13

If shortening is triggered by the existence of a following empty Nucleus, the final consonant of a word where shortening is observed must sit in the Onset of an empty Nucleus as well. The ensuing structures of [?amel-al] and [panal] appear under (8). (8) fake "Closed Syllable Shortening" in Yawelmani: long vowels shorten before empty Nuclei O N O N O N O N

I

I I

I

I

I

I

O N O N O N

I

x x x x x x x x

I I I

I I I

? a m

l a i

I

I

I

I

I I

x x x x x x

I I

I

I I

p a n a l

But there is yet another type of languages where Closed Syllable Shortening is reputed to be active. While shortening occurs both wordinternally and word-finally in Yawelmani, languages such as English operate Closed Syllable Shortening word-internally, but not word-finally: keep, weep [kiip,wiip] vs. kept, wept [kept, wept]. 12 In this case, Kaye (1990a) maintains a true closed syllable analysis: the vowel in kept is short because the [p] is the Coda of its Rhyme. Hence, the final consonant of keep, weep cannot close the syllable that contains the long vowel: long vowels are banned from true closed syllables in English (as much as in all other languages). The only solution is to interpret the [p] of keep, weep as the Onset of an empty Nucleus. Therefore, keep and kept do not posses the same syllable structure: the [p] is an Onset in the former, but a Coda in the latter word. This state of affairs is depicted under (9) below. (9)

true Closed Syllable Shortness in English: long vowels cannot be long in presence of a Coda R

Ο I χ k

12

Ν χ

χ

I χ Ρ

O N I χ

o

keep

χ

χ

χ

k

ε

ρ

Ο

Ν

χ

χ

See note 123 for a more detailed discussion of the English situation.

kept

14

1,2 Open vs. closed syllables in CVCV

A consequence of this analysis is that keep and kept must be two separate lexical entries: Structure Preservation excludes wavering syllabic identities for [p]. In other words, Kaye must pretend that present and past tense forms of English irregular verbs are not derived from a common underlying source. This view is supported by independent evidence such as the fact that natives can always form the past tense of regular verbs, but need to rely on their memory when producing past tense forms of irregular verbs. Depending on their education and other social factors, most natives will have a hard time producing the "correct" past tense or past participle of verbs such as to bear or to lie (in bed). Also, no shortening is observed with regular verbs: to peep [piip] - peeped [piipt] and so forth. Here, the past tense is indeed derived from the present tense, and hence the [pt] cluster in peeped (but not in kept) represents two independent Onsets that enclose an empty Nucleus. Further implications of this analysis are discussed in Kaye (1990a:323ss) and in Vol.2,III.5. The overall result of Kaye's (1990a) analysis of Closed Syllable Shortening is that word-final consonants belong to Onsets in all languages, whether of the Yawelmani or the English type. Moreover, they are always followed by an empty Nucleus. The latter fact requires an extension of the Empty Category Principle (2): final empty Nuclei do not participate in vowel-zero alternations, and they cannot be covered by Proper Government from any following vowel. Therefore, a special provision is made: final empty Nuclei are said to be subject to parametric Licensing. That is, some languages do, while others do not license them. The former group corresponds to systems such as French or English where they are inaudible; hence consonant-final words occur. The latter group is illustrated for example by Italian where all words must end in a vowel: final empty Nuclei are not licensed and hence need to enjoy a melodic existence. See Kaye (1990a:323ss) for further discussion of this typology. The revised Empty Category Principle that takes into account the special status of final empty Nuclei is thus as under (10) below (see Kaye 1990a:314). (10) Empty Category Principle - version two an empty Nucleus may remain unexpressed iff it is a. properly governed or b. word-final

The ECP, PG and vowel-zero alternations in Standard GΡ

15

A remarkable consequence of Kaye's (1990a) analysis of "Closed Syllable Shortening" is the fact that word-final Codas do not exist at all. All word-final consonants in all languages belong to Onsets and are followed by an empty Nucleus. Kaye (1990a) derives the absence of word-final Codas from a principle that he calls Coda Licensing: Codas may exist only if they receive support (i.e. are licensed) from a following Onset. Since wordfinal consonants are not followed by any Onset, they cannot instantiate Codas.

19

1.4. The analysis of vowel-zero alternations in Standard Government Phonology

20

1.4.1. The basic pattern: vocalisation occurs before empty Nuclei We are now in a position to address the question that was raised in §16: if Standard Government Phonology is unable to express the Coda context {C,#} in a non-disjunctive fashion because word-final consonants enjoy the status of Codas in no circumstance, how can the basic pattern of vowelzero alternations (3) be captured? Structure Preservation and final empty Nuclei provide the answer: alternating vowels never occur before a Coda. We already know that word-final consonants can be Codas in no event. But the reader may also verify under (3) that all consonant clusters which provoke the vocalisation of preceding alternation sites are heteromorphemic (with one apparent exception, Moroccan Arabic, on which more shortly in §22). They are achieved by the suffixation of a consonant-initial suffix to a consonant-final stem, i.e. CVC-CV such as loket-ηί "elbow adj." in the familiar Czech example. In these cases, Structure Preservation enforces a syllabic interpretation whereby the [t] pertains to an Onset: the stem also exists without suffix, i.e. loket "elbow NOMsg". If both forms are related by a derivation, the obligatory interpretation of the [t] as an Onset in the latter word implies the same syllabic status in the former. Hence, all alternations shown under (3) reduce to the pattern under (11) below.

16

1,2 Open vs. closed syllables in CVCV

(11) representation of basic vowel-zero alternations in Standard Government Phonology Czech "elbow" c. loket-ni adjective b. loket NOMsg "a. lokt-e GENsg PG, PG, PG

f O N O N O N

O N O N O N

O N O N O N O N

I X X X X X X

X X X X X X

X

1

1

1

o

k

t

e

o

k

epenthesis

I X

o

I X

I I X

k

I

X

X

t

I I X

X

η ί

epenthesis

Since the stem-final consonant always belongs to an Onset, vowelzero alternations may be described by the non-disjunctive statement "empty Nuclei are subject to epenthesis iff they are followed by another empty Nucleus". Hence, it appears that empty Nuclei play a central role in the description of two major alternations: they provoke 1) the shortening of preceding long vowels (§18) and 2) the vocalisation of preceding empty Nuclei as under (11). However, their effects seem to be contradictory: in the former case, their presence weakens a strong object (a long vowel), whereas in the latter they provoke the strengthening of empty Nuclei, which acquire melody under their influence. This intriguing property of empty Nuclei is further discussed in §§151,440,472. In any event, (11) also shows that the Standard Government Phonology analysis relies on insertion: the melody of alternating vowels is absent from the lexicon. The location of alternation sites is identified by the presence of a non-final empty Nucleus. In case these are followed by another empty Nucleus, epenthesis occurs. In other words, two empty Nuclei in a row are ill-formed. This description introduces a causality that contrasts with the one that is usually taken for granted: vowel-zero alternations do not occur because of the presence of a peculiar syllabic constituent, the Coda, in a syllabic arborescence. Rather, a lateral relation that does make no reference to arborescence at all controls the pattern: the lexical identity of alternation sites are empty Nuclei, which do or do not vocalise according to the presence or the absence of another empty Nucleus to their right.

The ECP, PG and vowel-zero alternations in Standard GP

17

The lateral relation that assures the internuclear communication has already been identified; it is implemented under (11): Proper Government (PG). Recall that the distribution of empty Nuclei is regulated by the Empty Category Principle: empty Nuclei may exist only if they are properly governed or word-final. Since only contentful Nuclei can dispense Proper Government, empty Nuclei must acquire a melodic identity in case they are followed by another empty Nucleus. This is what happens under (llb,c) where the following Nucleus is unable to dispense Proper Government.

21

1.4.2. Intervening governing domains block Proper Government But there is yet another configuration where an underlyingly empty Nucleus undergoes vocalisation. This is when more than one consonant separates it from the following contentful Nucleus, which would be ready to provide Proper Government. In this case, the alternation site is vocalised even if the intervening consonant cluster is monomorphemic and hence does not enclose an empty Nucleus as under (11c). Illustration comes for example from Quebec French (Charette 1990,1991). In this language, schwa alternates with zero. Consider the data under (12) below.13 (12) a. no cluster adjacent to schwa [...VCaCV...] or [...VCCV...]

13

b. cluster adjacent to schwa [...VCaTRV...] vs. •[...VCTRV...]

davaniK

devenir

la fanet/

la fenêtre

la tavKet

la levrette

ta K3pa

le repas

nu s9vk5

nous sevrons

la samen

la semaine

Is ctagKe

le degré

la sakxe

le secret

kosavwa/

concevoir

la javKoej

le chevreuil

b simatjex

le cimetière

dapqi

depuis

There are no schwas in closed syllables in French, hence the environment RT is not represented. More detailed discussion of the data is available in Charette (1990,1991). Glosses: (12a) "to become, the window, the meal, the week"; (12b) "the roebuck, the greyhound, we wean, the degree, the secret, to conceive, the cemetery, since".

18

1,2 Open vs. closed syllables in CVCV

Under (12a) where schwa is separated from the following vowel by one single consonant, its realisation is optional (hence [davaniK] and [daveniK] for devenir are both grammatical). If on the other hand a monomorphemic cluster separates schwa from the following full vowel as under (12b), it cannot be left out (hence [ b sakxe] vs. * [ b sekxe]). The reason for the stability of schwa cannot be the one that was invoked in the previous section: the Nucleus that follows the alternation site enjoys full phonetic expression. Distributionally speaking, the solution is obvious: the presence of the cluster enforces the stability of the potentially alternating vowel. The interpretation that Standard Government Phonology (Kaye et al. 1990:219s, Kaye 1990b, Charette 1990:236s, 1991:99,113ss) has given to this fact appears under (13). (13)

Intervening governing domains block Proper Government.

In other words, there may be no embedded governing domains: Proper Government is suspended if it experiences another governing domain on its way from the governing to the governed Nucleus. In Standard Government Phonology, branching Onsets and Coda-Onset sequences are governing domains: Constituent Government holds among the members of the former cluster, while Interconstituent Government relates the two consonants in the latter configuration (see appendix 4 §626). Hence, the representation of a word such as secret "secret" is as under (14). (14) Proper Government cannot reach schwa because it cannot apply over another governing domain PG

O N X

X

s

epenthesis

O

N

X

X

X

k

r

ε

/\

Constituent Gvt

In all cases reviewed, epenthesis occurs when the empty Nucleus escapes Proper Government. However, the grammar recognises two distinct reasons for which empty Nuclei may remain uncovered by Proper Govern-

The ECP, PG and vowel-zero alternations in Standard GΡ

19

ment: either the following Nucleus is empty as under (11 b,c), or a governing domain intervenes between governor and governee. Note that this disjunctivity is not odd. At least is it not any worse than the ordinary closed syllable analysis that ascribes the vocalisation of alternation sites as under (1 lb,c) to the presence of a Coda in their syllable. On this account, the stability of schwa under (14) has an unrelated cause as well: the [k] of secret "secret" certainly does not belong to a Coda. It will be shown in §64 how CVCV derives both instances of vocalisation from one single cause.

22

1.4.3. Moroccan Arabic vocalisation before geminates induces disjunctivity There is one case among all alternations shown under (3) where Standard Government Phonology must give up on the unified solution of (11). This is when Moroccan Arabic requires a stable vowel in presence of a following geminate. Recall from (3) that the surface pattern of this language is strictly identical to all vowel-zero alternations that are found elsewhere: zero occurs before a single consonant which is followed by a vowel ([kiteb-u] "write pf act 3pl", [ketib] "write pf act 3sg"), while schwa surfaces in case it precedes two consonants ([kitb-u] "write pf act 3pl") or one consonant which is word-final ([ktib] "write pf act 3sg"). Alternation sites are also vocalised when followed by a geminate: [kittib] "write pf act 3sg causative". This is expected under the regular closed syllable analysis: the single statement "alternation sites are vocalised in closed syllables" covers all Moroccan Arabic data mentioned. By contrast, Standard Government Phonology cannot account for the vocalisation before geminates in the same way as for those cases where epenthesis occurs before two non-geminate consonants ([kitb-u] "write pf act 3pl") and one single word-final consonant ([ktib] "write pf act 3sg"). Recall that the two latter configurations provoke epenthesis because the alternation site is followed by an empty Nucleus. This is shown under (15) below.

20

1,2 Open vs. closed syllables in CVCV

(15) Moroccan Arabic vowel-zero alternations I: vocalisation because of a following empty Nucleus a. vocalisation before two nonb. vocalisation before a wordfinal consonant geminate consonants PG PG PG PG

Ο Ν Ο Ν Ο Ν

Ο Ν Ο Ν Ο Ν

χ χ χ χ χ χ I I I I I k i t b u

χ χ χ χ χ χ I I I I k

t

i

b

The existence of an empty Nucleus enclosed by the cluster in [kitb-u] is guaranteed by its overt phonetic presence in [ktib]. On the other hand, the word-final [b] must be followed by an empty Nucleus anyway because all word-final consonants sit in Onsets. However, there can be no empty Nucleus enclosed within the two members of the geminate. The only structure that geminates can instantiate in Standard Government Phonology is a Coda-Onset sequence. This is depicted under (16) below. (16) Moroccan Arabic vowel-zero alternations II: vocalisation because of a following geminate

PG.

Ο

Ν

Ο

Ν

[kittib]

tJ epenthesis

"write pf act 3 sg causative"

Interconstituent Gvt

Therefore, Standard Government Phonology produces a disjunctive analysis: the reason why epenthesis occurs under (16) is the fact that the geminate constitutes a governing domain which does not allow the following expressed Nucleus to contract a governing relation with the Nucleus to

The ECP, PG and vowel-zero alternations in Standard GP

21

its left. The cause for the vocalisation of [kitb-u] /k0t0b-u/ and [ktib] /kt0Ò0/, however, is as under (llb,c), respectively: there is an empty Nucleus following the alternation site, and this empty Nucleus is unable to issue Proper Government. Hence, there are two different reasons for the same effect. Chapter 1,3 (§69) demonstrates how CVCV does away with this disjunctivity: the only cause for the vocalisation of alternation sites is the emptiness of the following Nucleus. That is, a geminate encloses an empty Nucleus just like any other consonant cluster. In the following two sections, some additional properties of empty Nuclei and Proper Government are identified.

23

1.5. Proper Government is not recursive Another characteristic of Proper Government is that the governing ability of vowels is exhausted by one single governing relation. That is, a vowel may govern at most one single constituent. This is obvious from chain reactions that involve vowel-zero alternations. Consider the data from Moroccan Arabic (Kaye 1990b) and Czech (Scheer 1997:68) given under (17). (17)

Moroccan Arabic Czech a. kcrtib Sevec b. kitab-u Sev0c-e

gloss "he has written", "shoemaker NOMsg" "they have written", "shoemaker GENsg"

Both the Moroccan Arabic and Czech examples show two vowels that alternate with zero in a row. In sufflxless forms as under (17a), the rightmost Nucleus is phonetically expressed, whereas its lefthand neighbour does not appear on the surface. In other words, the former properly governs the latter. What should be the effect, then, of adding a vowelinitial suffix? Since both stem-internal vowels alternate with zero, they could both disappear under the influence of the suffix. This is, however, not what happens: under (17b), the suffixal vowel properly governs its direct neighbour, but is unable to reach the first vowel of the stem, which therefore surfaces. This kind of observation, which seems to be cross-

22

1,2 Open vs. closed syllables in CVCV

linguistically stable, indicates that vowels are unable to simultaneously govern more than one constituent (see e.g. Kaye et al. 1990:219s).14

24

1.6. Domains and domain-final empty Nuclei The special status of word-final empty Nuclei has been recorded in the Empty Category Principle under (10): they may remain empty even though they escape Proper Government because of a language-specific parameter that defines the possibility for consonant-final words to exist. In other words, final empty Nuclei may remain empty because of a morphological reason: they occur at the right edge of the word. In subsequent work, the initial peculiarity of word-final empty Nuclei has turned out to identify a larger class of empty Nuclei, of which those that occur word-finally are but a subset: empty Nuclei that may remain phonetically unexpressed even though they escape Proper Government in fact occur domain-finally, rather than word-finally. Word-final empty Nuclei are a subset of domain-final empty Nuclei. But what is a phonological domain? Every phonological theory must recognise the existence of domains in one way or another. A phonological domain is the portion of a morphologically complex string on which phonology operates without interference from other modules of the grammar. For example, a string made of a root and a suffix involves a boundary: Depending on the idiosyncratic properties of the suffix, this boundary may or may not be visible for the phonology. That is, phonological rules may either operate on the entire string [ V - sufi] or on the root alone. In English, suffixes divide into so-called level 1 and level 2 suffixes, which show contrasting behaviour. For instance, the suffix -al as in parént-al is invisible for the phonology since it does not modify regular stress assignment, which produces the penultimate pattern. The root in isolation follows the same penultimate scheme: párent. The suffix -hood, however, disturbs the regular course of phonology since it produces antepenultimate stress: parent-hood. The classical interpretation of these facts is procedural in the sense of Lexical Phonology: different phonological rules apply at various moments of the word-formation process (here stress is assigned before -hood is concatenated). The implementation of the same observation in Government 14

Gussmann (1997) advocates Recursive Government (where a vowel governs more than one target at a time) on the grounds of préfixai vowel-zero alternations in Polish. He has abandoned this analysis since then.

The ECP, PG and vowel-zero alternations in Standard GΡ

23

Phonology is based on domains: parént-al identifies as [parent-al] because the boundary is invisible to the phonology (angled brackets represent domain boundaries). By contrast, párent-hood follows the pattern [[parent] hood]: the morphological boundary is phonologically relevant. §§249,349,630 offer some more discussion of domain structure, and Vol.2,III.3.5 examines the issue at length. Relevant literature in Government Phonology includes Gussmann & Kaye (1993), Kaye (1995), Cyran & Gussmann (1998,1999) and Gussmann (2002:45ss). The choice of domains for the representation of morphological information in phonology has consequences namely for the status of empty Nuclei, an issue that does not arise in Lexical Phonology. The move from "word-final" to "domain-final" has direct bearing on the situation of empty Nuclei in heavy word-final clusters of the kind that occur in Germanic languages. For instance, the English word sixths that is always quoted in this context identifies as six-th-s [siks-0-s], This sort of word-final monster clusters is always heteromorphemic in English and German.15 Monomorphemic sequences such as [-ks0s]# do not occur in English, nor does [-ks0s-] exist word-internally. Therefore, the existence of monster clusters must be viewed as a consequence of their heteromorphematicity. In other words, [-ks0s]# may exist because it is composed of several morphemes, and only on this explicit condition. Hence, morphological structure has direct bearing on phonology: the more complex you are on the morphological side, the more complex you are allowed to be phonologically. If domains represent those morphological divisions that are phonologically relevant, six-th-s must be made of three domains. Hence, there are three domain-final empty Nuclei: [[[siks0]00]s0]. If parametric Licensing has scope over domain-final, rather than word-final empty Nuclei, all wordfinal monster clusters in English lose their exotic and cumbersome flavour: they instantiate a perfectly regular phonological structure. Or, in other words, sixths [[[siks0]00]s0] is no more exceptional than, say, pit [pit0]: both words possess domain-final empty Nuclei whose absence from the surface is due to parametric Licensing. The only difference between sixths and pit is the number of domain-final empty Nuclei, that is three in the former, against only one in the latter case.

15

See for example Goldsmith (1990:140ss) and Harris (1994a:66ss) for a detailed discussion of the English data. Hall (1992:110) provides exhaustive facts for German.

24

1,2 Open vs. closed syllables in CVCV

1.7. Summary: the ECP and Proper Government in Standard Government Phonology At the end of this review, the two major tools for the description of vowelzero alternations are summarised below. The Empty Category Principle is repeated under (18), and Proper Government identifies as under (19). (18) Empty Category Principle - version two an empty Nucleus may remain unexpressed if it is a. properly governed or b. domain-final (19) Proper Government (PG) Proper Government is an internuclear relation whereby a. the governor must be phonetically expressed b. the governee is phonetically absent c. each governor can govern at most one target d. governing domains that intervene between the governor and the governee act as a barrier: Proper Government is blocked. In the following sections, the position of CVCV in regard of the fundamental opposition between open and closed syllable is examined.

2. Open vs. closed syllables in CVCV: the problem and two indications How could CVCV represent the difference between clusters of rising and falling sonority? In many (Southern) varieties of French for example, the former make preceding mid vowels +ATR (métro [met/o] "metro"), while the latter provoke -ATR mid vowels (perdu [psKdy] "lost").16 Under the assumption that all consonants belong to non-branching Onsets, these words are represented as under (20).

16

Since Dell (1973:209ss), this pattern is called Closed Syllable Adjustment or "Loi de Position". Relevant literature includes Selkirk (1972:367) and Tranel (1987,1988). The facts are exposed at greater length in §437.

Open vs. closed syllables in CVCV: the problem and two indications (20) a. French métro "metro1II C

V

m

e

C

V

25

b. French perdu "lost'.11

C

V

C

r

o

p

V e

C r

V

C

V

d

y

As my be seen, there is no structural difference of any kind. If no structural property produces the contrast that is observed on the preceding vowel, in which way could the different sonority slope cause this effect? This is not really a trivial question since phonological processes that are conditioned by the difference between TR and RT clusters abound. Almost every language that allows for consonantal clustering reacts on this contrast. Its effects may be visible on the preceding vowel or on the first consonant of the RT cluster (i.e. the Coda). In short, no phonological theory can afford to dismiss this most fundamental opposition. There are two indications that hint at how this contrast could be expressed in CVCV. For one thing, the different syllabifications of TR and RT that are commonly practised are not primary. Rather, they are derived from the sonority value of both consonants: whether Ci in a sequence VC,C 2 V "belongs" to the preceding or to the following vowel is determined by its situation in regard of C 2 . Typically, if Q is more sonorous than C2, it will close the preceding syllable. If it is less sonorous than C2, both consonants are said to share an Onset. Hence, all theories must necessarily derive the effects observed from the relation that both consonants contract, that is from the consonantal property called sonority. The second indication comes from a particular property of Government Phonology. Since syllable structure has been (re)introduced into generative phonology by Kahn (1976), a growing number of restrictions on branching structures have been introduced. For instance, early syllabification algorithms often operated according to the principle "syllabify any number of word-initial consonants into the Onset until you hit the first vowel". The right edge of the word was treated in the same way. Rather soon, however, it was judged advantageous to respect strict patterns of sonority for Onsets and Codas. Supernumerary consonants were therefore viewed as extrasyllabic (this notion discussed at length in chapter 11,5 §339). In regard of this debate, Standard Government Phonology holds that all constituents are maximally binary. 17 CVCV is the ultimate stage of this 17

The so-called Binary Theorem (Kaye et al. 1990:199, Kaye 1990a:306) is discussed in §176.

26

1,2 Open vs. closed syllables in CVCV

evolution: syllabic arborescence has been impoverished in all models since the late 70s; CVCV reduces it to nothing. The reduction of arboreal structure, however, has not gone without developing a surrogate means of expressing the regularities at hand. It was shown in the previous section how the presence of empty Nuclei has led to the introduction of a regulating principle for empty positions, i.e. the phonological Empty Category Principle. One way of satisfying the ECP is a lateral relation among segments that is called Proper Government. Hence, the importance of lateral relations and arboreal syllable structure follows reverse proportionality: the less arboreal structure there is, the more lateral relations need to be developed in order to pick up the functional load. It is thus reasonable to believe that the demotion of the Coda to a constituent that is not hierarchised in any arboreal fashion must go hand in hand with a lateral relation that has not been previously considered. These two indications being borne in mind, let us consider a peculiar property of the vowel-zero alternations that are found in Czech prefixes.

3. Internuclear communication over consonant clusters 3.1. The vocalisation of Czech prefixes In Czech, consonant-final prefixes such as pod- "under" and od- "away" are sometimes vocalised. The vowel appears only if the following root begins with at least two consonants: compare ode-brat "take away p f ' with od-birat "id. ipf'. However, all consonant clusters do not provoke vocalisation of the prefix: pod-bradek "double chin". The nature of the root-initial cluster does not play any role since identical groups ([br-] in the words quoted) produce contrasting results. Whether a prefix is vocalised or not depends on the structure of the root. Given a root that begins with two consonants on the surface [VCiC2Vi...], the prefix will bear a vowel iff Ci is root-initial, and C2 root-final, i.e. in case the root occurs in zero grade /VC10C2-V,/ and V, is thematic or suffixal. If on the other hand Ci and C2 are both root-initial and the following vowel part of the root /VC1C2V/, prefixes will remain unvocalised. This pattern controls all consonant-final prefixes in the language. A more detailed description is available in Scheer (1996,1997,1999a). Some further illustration is offered under (21).

Internuclear communication over consonant clusters

27

vocalised

unvocalised

gloss

VBR

pode-brat

pod0-bradek

to seize from below, double chin

VDR

roze-drat

roz0-drobit

to tear up, to crumble

VSL

vze-Sly

roz0-§lapat

to open (flower), to crush

VDM

roze-dmout

roz0-dmychat

to blow up, to poke (fire)

VPR

roze-pfit

roz0-pfahat

to straddle, to stretch

T h e c l a i m that the vocalisation o f p r e f i x e s d e p e n d s on root structure m a y b e substantiated b y different means, t w o o f w h i c h are d i s c u s s e d here. For o n e thing, a root m a y betray its z e r o grade in other g r a m m a t i c a l forms. T h i s is, a root such as Vb r "to take" appears in z e r o grade in the infinitive br-át "to t a k e " , but s h o w s a v o w e l b e t w e e n the [b] and the [r] in inflected f o r m s ber-u, ber-es

etc. "I, y o u . . . take" and in i m p e r f e c t i v e s p o d - b i r - a t "to

s e i z e f r o m b e l o w i p f A s a matter o f fact, all and o n l y those cluster-initial roots

.] that p o s s e s s f o r m s w h e r e a v o w e l m a y appear b e t w e e n C j

and C 2 a l s o p r o v o k e v o c a l i s a t i o n o f the prefix. T h o s e roots w h o s e Q

and

C2 are n e v e r broken up by any v o w e l possess u n v o c a l i s e d p r e f i x e s . T a b l e (22) o f f e r s s o m e more illustration o f alternating and non-alternating roots. 1 8 (22)

a. root provoking vocalised prefixes

zero grade

two forms of the same root full grade

Vc,c2- /Vc,0c2/

18

root provoking non-vocalised prefixes no occurrence of Vc,vc2

W c , v c2/

VBR-

pode-brat

Pf

pod-birat 'Pf

VDR-

inf

roz-deru

VHR-

roze-drat pfede-hra

noun NOMsg

her

noun

VHN-

ode-hnat

pf

od-hànët

¡pf

VPRVSNVSL-

ode-prat

inf

od-peru

beze-sny vze-Sly

adj

sen

adj

gel

past

VZDVDN-

pode-zdit beze-dny

inf

zed' den

noun

adj

b.

pod-bradek

'sg

!sg noun

noun

roz-drobit GENpl

od-hrabat roz-hnëvat vz-pruha

NOMsg pod-snëinik act part. roz-Slapat NOMsg od-zdola GENpl —

Glosses, line by line: "to seize from below pf, id. ipf, double chin, to tear up inf, id. lsg, to crumble, prelude, game, to sweep away, to expel pf, id. ipf, to enrage, to prewash inf, id. lsg, incentive, sleepless, dream, snowdrop, open adj. (flower), to go past act part., to crush, to underpin, wall, from below, bottomless, day."

28

1,2 Open vs. closed syllables in CVCV

The status of the two consonants may also be controlled when looking at the morphological structure of the root. In a surface cluster .], C 2 may either be root-initial as inpod-brad-ek "double chin", or it closes the root as in pode-ber-at "to seize from below pf'. Whether C 2 is root-initial or root-final is decided by the morphological status of the vowel and the consonants that follow. Slavic and more generally Indo-European roots may possess one root vowel at most. Given this fact, there can be no zero hidden between Ci and C 2 in [VC,C2VC3] if V and C 3 belong to the root (as in pod-brad-ek). If on the other hand V and C 3 belong to a different morpheme as in pode-br-a-t (-a- is thematic, -t marks the infinitive), it must be concluded that the root occurs in zero grade. Examination of the cases under (22) shows that all and only those items where V and C3 do not belong to the root provoke vocalised prefixes. (23)

C2 closes the root, hence the root is in zero grade [VC,C2-] VBR VDR VHR VHN VPR

VSN VSL A/ZD

VDN

=/C,_C 2 / pode- B D rozepfede- H H odeΡ odebeze- S vze§ pode- Ζ beze- D

R R R Ν R Ν L D Ν

-at -at -a -at -at -y -y -ít -y

vs.

C2 is part of the initial cluster of the root, hence the root is not in zero grade =/C,C 2 _/ pod- BRaD -ek roz- DRoB -it od- HRaB -at HNëV -at ΓΟΖPRuH -a vzpod- SNËZ -ník roz- ËLaP -at od- ZDoL -a —

The following section discusses the consequences of this préfixai vowel-zero alternation.

3.2. Consequences of this préfixai alternation 3.2.1. Vowel-zero alternations in Standard Government Phonology Recall from §21 that vowel-zero alternations are blocked in case a consonant cluster intervenes between the potential governor and the potential governee. This is the surface description of what Kay e et al. (1990) take to be a conflict of two governing domains. In their model of syllable structure,

Internuclear communication over consonant clusters

29

all monomorphemic consonant clusters that are marshalled by cooccurrence restrictions constitute a governing domain 19 : Constituent Government holds among TR clusters, while Interconstituent Government relates RT sequences. 20 Both types of Government are illustrated under (24). (24) French patrie [patri] "fatherland"

O N O N I I / \ I X X X X X I I I I I p a t r i l _ t Constituent Gvt

French partie [parti] "part" R 0 1 χ I p

N \ I \ χ χ I I a r t

Ο I χ I

Ν I χ I i

1 Interconstituent Gvt

Given this situation, it was tempting to identify the existence of a governing domain with the blocking effect of consonant clusters that intervene between governor and governee. The observation that "v" in monomorphemic vCCV could never be zero even though it is an alternating vowel was thus recast in terms of government: intervening governing domains block Proper Government. Or, in other words, there may be no embedded governing domains. It is obvious, however, that this statement does not enjoy any theoretical status: it remains a pure observational fact that has been translated into the formal vocabulary of a theory. No property of the theory prohibits the existence of embedded governing domains.

31

3.2.2. Falsification of the statement "intervening governing domains block Proper Government" The generalisation under (13) "intervening governing domains block Proper Government" is falsified by the vowel-zero alternations in Czech prefixes. Recall that prefixes remain unvocalised in spite of the presence of a conso-

19 20

So-called bogus clusters do not. They are further discussed in §§180,215. See appendix 4 (§625) on the general organisation of syllable structure in Standard Government Phonology.

30

1,2 Open vs. closed syllables in CVCV

nant cluster on their righthand side if this cluster does not represent a root in zero grade. A relevant example is podo-bradek "double chin". This pattern is by no means exceptional or marked in the language. The corpus analysed in Scheer (1996) contains all entries of the dictionary Ulbrich (1978) that begin with one of the seven consonant-final prefixes bez-, vz-, pred-, roz-, nad-, pod-, od- "without; inchoative, upwards; before, in front of; inchoative, divide into several parts; on, over; under, underneath; away" and are followed by at least two consonants. Out of a total of 957 items, 195 possess vocalised prefixes, against 762 that occur without préfixai vowel. The former, but not the latter, instantiate roots in zero grade. Hence, the numerically unmarked case is the one that offends the observation "intervening consonant clusters block Proper Government ". One such item is represented under (25). (25) Czech pode-bradek

O

N

I

O

I

N

I

x

x

x

I

I

I

p

o

d

i I

"double chin" in Standard Government Phonology PG

-

I

O

/ \

x

I

x

I

x

I b

u

N

O

I x

I r

x

I a

d

N

O

N

I

I

I

x

x

x

I

I

e

k

Constituent Gvt

There can be no doubt that the préfixai alternation at hand is an orthodox instance of vowel-zero alternations. This is evidenced by the fact that all prefixes appear unvocalised in isolation21, i.e. when no proper governor is available: podo, *pode. The same result is produced in case one single consonant separates the alternation site and the following vowel. VcV-roots always provoke unvocalised prefixes such as for example rozedat "to distribute". Finally, it is noteworthy that Czech, as all other Slavic languages, also possesses vowel-zero alternations that follow the regular cross-

21

Or as prepositions for that matter: most prefixes also exist as prepositions with identical or related meaning: pod "under", nad "over", pred "before", bez "without", od "from".

Internuclear communication over consonant clusters

31

linguistic pattern which was introduced under (3). Table (26) below provides some illustration. (26)

-C0C-V hudeb-a kaváran-a lokert-u

-CeC-0 N O M sg N O M sg GEN sg

hudeb kaváren loket

GEN pi GEN pi N O M sg

-CeC-CV

gloss

hudeb-ni adj kaváren-sky adj loket-ní adj

music café elbow

If it is true that a Nucleus is phonetically unexpressed because it is struck by Proper Government which comes from a following contentful Nucleus, there is no way not to conclude that the zero in pode-bradek is properly governed by the following [a].22 However, the intervening [br] constitutes Constituent Government domain and therefore should act as a barrier with respect to the governing relation between the prefix and the [a]. Hence, if the alternations under (3) and those that occur in Czech prefixes are controlled by the same phonological mechanism, the observation (13) "intervening governing domains block Proper Government" is disqualified. The consequences of this falsification for the theory of vowel-zero alternations are evaluated in a while. Before doing so, another fact needs to be mentioned.

32

3.2.3. Unvocalised prefixes occur only before branching Onsets The vocalisation of Czech prefixes is also interesting for the sake of the primary goal pursued in chapter 1,2 (§14), i.e. the expression of the contrast between open and closed syllables in CVCV. Consider the representation of pode-bradek in CVCV under (27) below.

22

Domain structure (cf. §§24,249,349,630) is of no help here since the relevant parameter is o f phonological, not of morphological nature: the only information that needs to be retrieved in order to compute the préfixai alternation is whether or not the root occurs in zero grade. All prefixes show exactly identical behaviour in regard of this conditioning. Were the préfixai boundary visible to phonology, the process would be sensitive to morphological parameters. The fact that it is not but only looks at the presence or the absence o f the root vowel enforces the view that w e are facing one single phonological domain.

32

1,2 Open vs. closed syllables in CVCV

(27) poda-bradek "double chin" in CVCV PG

i C

V

C

V

I -

C

I

I

I

I

p

o

d

b

V

o

C

I

V

I r

C

I a

d

V

C

I

I

e

k

V

The préfixai alternation site is phonetically unexpressed as in the non-CVCV environment. This necessarily means that it is hit by Proper Government. The only governor available is the radical [a] since the Nucleus that is located between the [b] and the [r] is empty and therefore does not qualify as a governor. Hence, the question arises for which reason this Nucleus V Q is phonetically absent. Whatever this reason, it must be different from Proper Government because the only governor available, [a], already targets another Nucleus. Even though this question cannot be answered at the present stage of the discussion, the resulting situation contributes to our understanding of how the difference between TR and RT could be represented in CVCV. Indeed, a closer examination of the 762 cases where a consonantal cluster encloses a Nucleus that seems to be orphan in regard of the ECP reveals that the two consonants at hand invariably constitute a branching Onset. On one hand, there are 195 items of the aforementioned corpus in Scheer (1996) that provoke vocalised prefixes, and their root-initial consonants are of both kinds: [VTR] as e.g. in pode-brat "to seize from below p f ' , and [VRT] as for example in ode-rvat "to tear off'. On the other hand, the 762 items that produce unvocalised prefixes never begin with an [VRT]-cluster. Only items that bear an initial [VTR] sequence such as poda-bradek "double chin" occur. This fact clearly indicates that the relation between the two consonants of TR- and RT sequences is not only responsible for the fundamental opposition open vs. closed syllable, but may also control the phonetic expression of a preceding Nucleus. The crucial parameter that allows to understand why intervening clusters sometimes do, but at other times do not block Proper Government is introduced in the following section.

Internuclear communication over consonant clusters 33

33

3.2.4. Heteromorphemic TR vs. monomorphemic TR: variable intimacy of adjacent consonants How come that clusters, in the same language, sometimes do but at other times do not block intervocalic communication? In Czech, consonant clusters that are created by suffixation invariably disqualify a preceding zero, while prefix-final Nuclei may happily remain unexpressed in presence of a cluster to their right. Table (28) gathers the relevant examples from the foregoing discussion. (28)

-CeCV hud0b-a kaváran-a lok0t-u

NOM sg NOM sg GEN sg

-CeC0 hudeb kaváren loket

-CeCCV GEN pi hudeb-ni GEN pi kaváren-sky NOM sg loket-ní vs. pode-bradek

adj adj adj etc.

gloss music café elbow double chin

This contrast could be captured by exploiting the different status of the empty Nuclei involved. The stability of alternating vowels before suffixes stems from the fact that the following empty Nucleus is morphemefinal: loket /lokete/, loket-ní /lokete-ni/. By contrast, the empty Nucleus that tolerates the non-vocalisation of its preceding peer is morpheme-internal: pod-bradek /pod0-beradek/. On this account, the general prohibition of two consecutive empty Nuclei must be released in case the second one is morpheme-internal. Or, in other words, morpheme-internal empty Nuclei are able to dispense Proper Government. This generalisation is certainly correct on the descriptive side, but it is entirely unwarranted since it weakens the theory and only transcribes the observation: there is no reason why internal, rather than final empty Nuclei should be able to govern. If anything, the contrary should be true: the lateral actorship of final empty Nuclei is not inferior, but superior to the one that their internal peers enjoy. §59 and chapter 11,12 (§524) are more explicit on this fundamental distinction. The alternative solution builds on the crucial influence of the contrast between suffixation and prefixation: two consonants categorically prohibit the syncope of preceding vowels upon suffixation, but do allow, under certain conditions, for the muteness of alternation sites when prefixation is involved. This statement may be refined when looking at the behaviour of vowel-zero alternations in French. Recall from §21 that alternating vowels

34

1,2 Open vs. closed syllables in CVCV

are obligatorily realised in Quebec French when they are followed by a consonant cluster. Now the behaviour of Metropolitan French is different: certain speakers regularly omit schwa in the same words, which are repeated under (29) for convenience (more data are provided in §66). 23 (29)

-C0CCV le secret le degré le chevreuil la levrette

[ta S0kxe] [ta degKe] [ta 30VKcej] [la Ιβνκεί]

gloss the secret the degree the roebuck the greyhound

Both the Czech and French cases where Nuclei can communicate over a consonant cluster share the following characteristics. (30)

properties of transparent clusters a. the consonants that intervene between the governor and the governee belong to the same morpheme b. they are always of rising sonority

Significantly, these conditions are never met in the languages that were introduced in the cross-linguistic survey under (3). The clusters that act as barriers for Proper Government are always heteromorphemic. It may thus be concluded on a complementary distribution of those clusters that do and those that do not prohibit the communication of their flanking Nuclei. In case the cluster is of falling sonority RT, Proper Government is blocked in any event. If the cluster qualifies as a branching Onset TR, the flanking Nuclei may or may not be allowed to communicate. TR clusters that are monomorphemic are transparent for Proper Government, but those that belong to two distinct morphemes interrupt the communication. The reason why the governor often does not reach its target when more than one consonant intervenes is not the existence of an embedded governing domain, neither does the sheer number of consonants block transparency. Rather, there are two configurations that disqualify internuclear communication independently of one another: 1) the intervening clus23

This pronunciation is conditioned by various geographic, social and agedependent parameters. It is much more likely to be found in the Northern half of France, but there are many speakers on both sides of the Loire that are unable to leave out schwa in this context. Note that schwa more generally alternates with zero in French. See e.g. Dell (1973:219ss), Scheer (1999b) for greater detail.

Internuclear communication over consonant clusters

35

ter is of falling sonority RT, and 2) it is heteromorphemic. If the sequence is of rising sonority TR and monomorphemic, a third condition needs to be satisfied in order for the flanking Nuclei to be able to communicate: the cluster must not enclose a "hidden" zero (as is the case in pode-brat, but not in pode-bradek). These indications are precious in regard of the initial question "how could CVCV represent the opposition between open and closed syllables?". It has been shown that the quality of RT clusters, but not that of branching Onsets TR, produces the same irrecoverable barrier for internuclear communication as does the heteromorphemic character of a cluster of whatever sonority slope. And it is certainly reasonable to conceive that monomorphemic consonants contract a more intimate relationship than the two members of a heteromorphemic cluster. I contend that this contrasting intimacy of two adjacent consonants is the key to the question raised. That is, internuclear communication fails because the two intervening consonants do not contract a relationship that is sufficiently close. This lack of intimacy may be due to either their heteromorphemic status, or to their falling sonority slope RT. By contrast, their cohesion is assured in case they are monomorphemic and possess a rising sonority slope. The following section translates this idea into the formal vocabulary of CVCV.

34

3.2.5. Consonantal interaction In CVCV, there is an in-built reason why Nuclei that are separated by two consonants may not communicate: the existence of an empty Nucleus enclosed within the cluster. This Nucleus absorbs the government coming from the following contentful Nucleus. The alternation site to the left of the cluster can thus never be reached by Proper Government. The relevant situation is illustrated under (31) where the heteromorphemic status of the intervening cluster is responsible for the absence of communication between the [t] and the [n].

36

1,2 Open vs. closed syllables in CVCV

(31 ) *0TRV where TR is heteromorphemic PG

C

V

C

V,

c

I 1

I o

I k

I e

I t

i

v2

I

c

v3

I I n i

Czech loket-ni "elbow adj."

In this situation, the alternating vowel in V) cannot be reached by Proper Government that originates in V 3 because the E C P of V 2 calls for satisfaction. This explanation is shared by C V C V and Standard Government Phonology because Structure Preservation enforces the existence of an empty N u c l e u s in the latter approach as well (see §20). T h e critical cases where C V C V and Standard Government Phonology contrast are those discussed in §22, i.e. Moroccan Arabic [kittib] "to write pf act 3sg causative" where a m o n o m o r p h e m i c cluster follows the alternation site. It is not the purpose of the present section to compare C V C V and Standard Government Phonology. This will be done at length in chapter 1,3 (§69). Rather, I intend to show h o w a theory of vowel-zero alternations has to look like when C V C V is assumed. In this sense, the next thing to be implemented is the observation that w a s made on the foregoing pages: Proper Government does reach the alternation site provided the intervening cluster is m o n o m o r p h e m i c and of rising sonority. The relevant representation in C V C V is shown under (32). (32) ok 0TRV where TR is monomorphemic Czech podfl-bradek "double chin" PG

C

V

I p

I o

C I d

V

i

-

C I b

V

o

I C

I

V

I r

C I

a

d

V I e

C

V

I k

In this configuration, the E C P of V @ is necessarily satisfied by a device other than Proper Government. It was contended in the previous section that the phonetic absence of V 0 is due to the intimate relationship that its flanking consonants contract. A definition of the consonantal interaction at hand that I call Infrasegmental Government is given under (33) (Scheer 1996,1997,1999a).

Internuclear communication over consonant clusters

37

(33) Infrasegmental Government (IG) - version one a. a consonant C x may govern another consonant C y iff C x is more sonorous than Cy. b. the existence of an infrasegmental relation between two consonants circumscribes the enclosed Nucleus, to the effect that its ECP is satisfied.

Infrasegmental Government raises a number of questions. For instance, it may appear as a notational variant of what is known as a true branching Onset. Also, it supposes that the headship within consonantal relations is reversed with respect to regular practice in Standard Government Phonology (Kaye et al. 1990): it is generally believed that obstruents are the prominent part of branching Onsets. In Standard Government Phonology, obstruents govern sonorants. The reverse relation is contended under (33a). The discussion of these issues is postponed until §36 when it will have been demonstrated how CVCV represents the opposition between open and closed syllables. Table (34) below shows two non-final empty Nuclei that are inaudible for two different reasons: Proper Government and Infrasegmental Government (IGt, noted " rom noapte, lupta, piept "night, fight, breast". Hence, theory is called to be able to describe all transitions within the triangle [u,w] - labials - velars. This requirement is neither original nor novel: the triangular relation mentioned has been identified for a long time. It originally comes from phonetic observation, where all members are known to share a property that has been called "grave". Since Jakobson, this fact was carried into phonological representation as the feature [grave], which fed a continuous debate in the 70s (Hyman 1973, Vago 1976, Odden 1978). In later autosegmental

50

1,2 Open vs. closed syllables in CVCV

times, the obvious connection between velars and labials was discussed in feature geometrical models under the heading of the node "peripheral" (see Clements & Hume 1995, Avery & Rice 1989, Rice 1994,1996,1999a,b for a survey of the relevant literature). Harris' (1990) classical system dismisses the velar angle of the triangle and thus needs to be amended. Following Lass (1984:278s) and Rennison (1990:187)30, one solution is to represent roundness and velarity by two separate primes, i.e. Β for labiality/ roundness, and U for velarity. The cornerstone of the triangle is [u,w]. Since these articulations are both rounded and velar, they contain Β and U. The interaction of [u,w] with labials, then, is assured by B, while the alternations with velars stem from U.31 Table (41) shows how Β and U are distributed across segments in this kind of system, anticipating also on the discussion of dentals (heads are boldfaced). Note that the identities shown only concern obstruents. Sonorants are discussed in a separate section below. (41) bilabials Β

labials labio-dentals Β, Α

dentals [t,d] [s,z] β A, I

velars β, U

labio-velars [u,w] U, Β

It goes without saying that the particular place definitions and additional splits within a given place ([t,d] vs. [s,z]) need due argumentation. Space limitations preclude any further detail here, and the reader must be referred to Scheer (1996,1998c, 1999a,2003a).

30

Lass (1984:278s) writes "I propose introducing |ra| 'velarity' and |ω| 'labiality'/'roundness', and dispensing with |u| completely. |u| seems to conflate too many properties anyhow. [...] Overall I think it's a good idea for ALL markedness considerations to be excluded from phonological characterizations. [...] Segments ought to code only their own properties, not statistics of crosslanguage distribution". Rennison (1990:187) defines "U — pull towards high back tongue body position (but with no implication of lip rounding!)". The split of the familiar U into two distinct primes gives birth to B. The very existence of this prime may appear awkward, and I would myself prefer a system where one single prime covers labiality, roundness and velarity. Ritter (1997:346, note 7) offers some interesting discussion of this issue. Broadbent (1996) promotes an alternative solution that could do this job (and which Daniel Huber has also suggested to me) whereby the difference between labials and velars is expressed in terms of headship (for example, the familiar U heads labial consonants, but is only operator in velars). The consequences of this option remain to be worked out; I leave this an open question here.

Substantial condition on IG: the internal structure of consonants

51

Note, however, that the overall goal of the present discussion is not so much to convince anybody of the particular consonantal identities presented. Rather, it focuses on their consequences: "if this representational system is assumed, the following theory of consonantal interaction may be installed". Hence, the presentation concentrates on the properties of the consonantal identities at hand, rather than on their motivation.

4.3.5. Dentals, derived sonority and the incompatibility of A and ? 4.3.5.1. Dentals: [t,d] are empty Three more properties of the consonantal representations that underpin Infrasegmental Government deserve explicit mention. Dental stops [t,d], but not dental fricatives [s,z], are melodically empty. That is, no placedefining prime contributes to their articulation. The proposal that coronals are underspecified for place is not original at all, see for instance Paradis & Prunet (1989,1991), Avery & Rice (1989), Steriade (1995a), Szigetvári (1994) and may others. However, the opposition between [t,d] and [s,z] within coronals is less common. Among other things, its motivation stems from the cross-linguistic epenthetic status of dental stops (see Pagliano 2003 for a typological survey), which the corresponding fricatives do not share. It is also intriguing to observe that [t,d] are very often and very easily affected by phonological processes such as palatalisation etc., but never constitute their output (except epenthesis of course). If [t,d] lack melodic content, this observation receives a natural explanation: phonological processes transport, replace or delink primes. Hence, the adjunction of a prime to a given structure cannot possibly produce an empty object. On the other hand, nothing can be replaced within or eliminated from an empty object. The phonological inertness of coronality has also grounded Backley's (1993) proposal according to which (certain) coronals are empty-headed.

4.3.5.2. Sonority is a derived category It has been mentioned earlier that sonority cannot be a phonological prime in its own right. This stance is also taken by Rice (1992), Ritter (1997:353ss) and Hermans (2003). In the model presented here, sonority is a function of three parameters: 1) the role played by the aperture prime A

52

1,2 Open vs. closed syllables in CVCV

within a segmental expression (head, operator, absent), 2) the constituent that dominates the segmental expression and 3) the presence vs. the absence of those primes that define the manner of articulation (? and h). Sonority may be calculated in regard of these factors in the way shown under (42) below. (42)

segment a e,o i,u liquids nasals glides s,z gutturals fricatives stops

Nucleus or Onset Ν Ν Ν 0 0 0 0 0 0 0

h/? — — — — — —

h h h hand?

role of A head operator absent head head/ operator absent head head / operator operator/ absent absent

sonority

Due to space limitations, the questions that this table may raise will have to remain unanswered here. Foremost the distribution of the prime A in the various segments mentioned would require much more ink. The individual internal structures of all segments shown would help elucidating this synoptic table. They are duly presented and motivated in Scheer (1996, 1998c, 1999a,2003a). For want of a better solution, additional information must be retrieved from these sources.

47

4.3.5.3. A and ? hate each other: they cannot combine However, let me introduce one particular point that concerns the aperture prime A and the occlusion prime ?. I argue that they are in strict complementary distribution within segments (Scheer 1996,1999a,2003a). That is, A and ? cannot cohabitate within a given phonological expression. Their antipodal status has an articulatory reflex since of course occlusion and aperture represent both extremes of the vertical status of the vocal tract. The mutually exclusive character of these primes helps understanding why there are no related simplex (i.e. non-affricate) stops of the same place of articulation for some fricatives, while others do possess a regular correspondence. For example, [p,b] are the stops related to the fricatives [φ,β] (both the stops and the fricatives are bilabial), but there are no simplex

Substantial condition on IG: the internal structure of consonants

53

labio-dental or palato-alveolar stops on record, which would echo the labiodental [f,v] and palato-alveolar fricatives [J",3], respectively. The prime A co-defines the consonantal identities of the labio-dental and the palatoalveolar locus, but is absent from bilabial segments. Hence, the only kind of stop that may exist for the labio-dental and the palato-alveolar places of articulation must not bear the occlusion element ?. Significantly, the stops associated with these places are affricates: [pf, tj", d3]. The incompatibility of A and ? may also shed light on the mysterious relation between aspiration and spirantisation. For example, aspiration triggers the spirantisation known as Grimm's Law, and Classical Greek aspirated stops also become fricatives in Modern Greek under the influence of aspiration. 32

4.3.6. Sonorants are heavy: abundant evidence 4.3.6.1. [r], [1] and nasals contain A The view that [r] contains the aperture prime A is fairly consensual nowadays. It is enforced by recurrent evidence from various languages that [r] has a lowering effect on adjacent vowels. For instance, lowering activity of [r] is reported for English where only mid and low vowels occur before etymological [r] (cf. Harris 1994a:244). Further evidence along these lines is reported by, among others, Lindau (1985), Hume (1994:128), Broadbent (1991,1996,1999), Brockhaus (1995a:115s,228ss) and Cyran (1994: 210). Another case in point is German where etymological [r] "vocalises" in Codas in non-rhotic varieties. As in English, German is represented by dialects where the pronunciation of former [r] is unaltered, while other varieties show uvular realisations in place of [r] across the board. The former are called rhotic; the latter (of which "Standard" German is a representative) are under focus here since the uvular representatives of etymological [r], [χ] and [κ], appear in vocalic clothes in Codas. This is demonstrated under (43) below. 33

32

33

The relationship of aspiration and spirantisation is further developed in Scheer (1996:216ss,2003a). Seigneur-Froli (2003, in press) provides a detailed analysis of the Greek case. Among many others, Drosdowski (1984:35s,57), Brockhaus (1995a:115s,228s), Hall (1992:56ss) and Wiese (1996:252ss) provide more detailed evidence.

54

1,2 Open VÍ. closed syllables in CVCV

(43) r-vocalisation in German a. [B] / V _ #

jveee foOB nuB hOKOB

máuE beu bis leeE foje

b. [B] / V _ C

c. [ a ] / a _ { C , # }

IUBÇ

beek bet baat baaj faat

d. M /

C_

gaa diraj, *dßaj gKajs, *gBajs

e. [κ] / V _ V

p/ajs, *peajs JveeKB piKaat pairada feliiifsn ?OKaan

spelling schwer vor nur Horror Mauer Bär Bier leer Feuer Lurch Berg Lord Bart Barsch Fahrt gar drei Greis Preis schwerer Pirat Parade verlieren Oran

gloss heavy before only horror wall bear beer empty fire amphibian mountain Lord beard perch trip done, cooked three old man price harder pirate parade lose Algerian city

In word-final position after a vowel (other than [a]) (vor) and after a vowel (other than [a]) followed by a consonant (Berg), former [r] has become a low schwa-vowel different from [a] that Drosdowski (1984:57) transcribes central [β]. In the same context, if the vowel preceding etymological [r] is [a], the result is a long [aa] (Bart, gar). Elsewhere, that is intervocalically (Pirat) and after a consonant (drei, Preis), the former [r] is realised [χ] or [κ] according to the voice value of the preceding segment. It must therefore be concluded that there is a melodic relationship between [r]/ [κ/χ] on one hand and low vowels on the other. In the same way, laterals may also give rise to schwa. Broadbent (1999:13) for example discusses a case from Orkney English where root vowels are subject to breaking before [r], [1], [ç] and [x]. The former two

Substantial condition on IG: the internal structure of consonants

55

consonants produce schwa ("bairn" [beam], "hail" [heal]), while [ç] and [x] release [i] ("eight" [eiçt]) and [u] ("through" [traux]), respectively. The ability to lower adjacent vowels is also reported for nasals. Consider the evolution of Middle High German root vowels before nasal consonants as shown under (44) (e.g. Paul et al. 1989:79). (44) nasals lower preceding vowels in the evolution of German Middle High German New High German gloss sunne Sonne sun sumer Sommer summer kumen kommen come münech Mönch monk sun Sohn son kllnec König king gewunnen gewonnen won geswummen geschwommen swum This lowering ability supposes the presence of A in nasals. Again, space restrictions preclude a more elaborate presentation of related processes. The literature that was mentioned earlier provides more ample illustration.

50

4.3.6.2. [r], [1] and [n] contain I, [1] contains U Several processes show that coronal sonorants are co-defined by I. Let us first look at German where they act as I-providers. In German, the distribution of [χ] and [ç] is well known and has been thoroughly studied since American Structuralism (e.g. Moulton 1947). Both segments are indeed in complementary distribution: (45a) shows that [ç] occurs after front vowels, while [χ] appears after [a,o,u].

56

1,2 Open vs. closed syllables in CVCV

(45) alternation of German [χ] and [ç] [χ] after [ç] after [u,o,a,aw] absence of I a. u buux

b.

o

kox

a

bax

aw

bawx

[y,0,i,e,aj,oj] presence of I

spelling Buch, Bücher

gloss book sg, pi

Koch, Köchin

cook mase, fem

Bach, Bäche

creek sg, pi

y_ ce

byyçe kœçin

ε ι

bεç^ viçtik

wichtig

important

bojça

Bauch, Bäuche

bailey sg, pi

a j _ wajç 1_ dolç

weich Dolch

soft dagger

η

mança

manche

some

r

duBç

durch

through

The fricative [ç] may be considered as the palatalised version of underlying /χ/ 3 4 : in [kox] vs. [kœçin] "cook mase, fem" for instance, the I contained in the feminine suffix umlauts the root vowel which, in turn, palatalises / χ / into [ç]. However, (45b) demonstrates that in all cases where / χ / occurs after [η], [1] and (vocalised etymological) [r], its palatal version [ç] surfaces. Since the postvocalic distribution of / χ / unambiguously shows that its palatalisation is triggered by all and only those segments that possess I, it must be concluded that the coronal sonorants [n,l,r] are I-providers as well. Parallel evidence is reported from Austrian German (for example Salzburg, Rennison 1978). Another type of evidence comes from lenition processes. It is quite common to observe that [r] and [1] implode under the pressure of a weak position (see chapter 11,15 §591 for the more general behaviour of sonorants in Coda position). Harris (1992,1994a, 1997) interprets lenition as melodic decomposition: the Coda is unable to license all melodic primes of a given segment. Therefore, a depleted version of its original structure appears on the surface. This process is highly instructive because it allows to 34

This analysis is actually subject to debate, as evidenced for example in Hall ( 1989,1992:220ss), Macfarland & Pierrehumbert (1991), Noske (1997), Wiese (1996:209ss). The alternative view holds that /ç/ is underlying in present-day German. In case this turns out to be correct, the evidence presented is not any less significant, it simply becomes diachronic: the origin of the NHG alternation at hand is the High German Consonant Shift (see §572), which effected the spirantisation of [k]. The result was a velar (uvular) fricative, which subsequently experienced palatalisation.

Substantial condition on IG: the internal structure of consonants

57

isolate individual primes that co-define the segment at hand but are usually merged with other melodic specifications. Table (46) provides relevant evidence from Dutch and Caribbean Spanish (data from Harris 1983, Harris 1997). (46)

[r,l] implode in Codas: Dutch and Caribbean Spanish a. Dutch spelling standard Southern (Rotterdam, Leiden) daar daar daaj kaart kaart kaajt stoort stoort stoojt karnan verpt b. Spanish standard revolver karta papel algo

gloss over there to card to disturb 2sg près

kajnan

kamen

to make buttermilk

vejpt

werpt

to throw 3sg près

Caribbean (e.g. Haiti) revojvej kajta papej ajgo

gloss revolver card paper something

If the lenited version of a segment represents its original identity minus some prime(s) that have been lost, it follows that I, which is the uncontroversial identity of the output [j], must be present in [r] and [1]. Another such process is notorious in Romance diachronics, only is the triggering position different. In all Romance languages except French (and Rheto-Romance), [1] is somehow affected if it occurs after an obstruent, i.e. as the second member of a branching Onset (e.g. Bourciez 1910:176s). In a common initial stage, the [1] was palatalised, and then experienced various subsequent lenitions in the individual Romance languages. In Italian for example, [1] in this position appears as [j]. Table (47) provides some illustration. 35 Words in the righthand column show that positions other than after obstruents do not provoke the appearance of [j].

35

The Italian situation is described in greater detail for example by Bee (1970 I:53s). Italian forms are spelt; after consonants represents [j], hence for example fiore [fjore] etc.

58

1,2 Open vs. closed syllables in CVCV

(47)

Italian: 1 > j in branching Onsets Italian Latin piazza platea P_ piovere vulg piovere bianco b germ * blank biasimare *blastimare florem f fiore fiamma fiamma k chiudere claudere chiodo clavus ghiaccio vulg glacia g_ ghianda glandem V_C multu molto calzare calceare v_v tavolo tabula dolore dolorem #_ luna luna lavare lavare

gloss place to rain white to blame flower flame to close nail ice glans much to shoe table pain moon to wash

The reasoning is the same as before: if the lenited version of [1] is [j] whose identity is unquestionably I, this prime must also be present in the undepleted original lateral. Finally, it does not take a lot of argument in order to be convinced that the velar lateral [1] contains U. In many languages, the result of [i]vocalisation in Codas is [w], Two such cases (French and Brazilian Portuguese) are discussed at length in chapter 11,12 (§524). Polish and SerboCroatian also fall into the category of (diachronically) [l]-vocalising languages.

51

4.3.6.3. Conclusion: sonorants are too complex to be governees The evidence reviewed hints at the presence of all place definers in sonorants: the coronal sonorants [r,l,n] bear I and A, while the velar lateral [1] possesses U and A. On the grounds of evidence that has not been reviewed above, Torre (2003) and Sebregts (2004) also argue for the "richness" of sonorants in general, and of [r] in particular. This result raises the complexity of sonorants to a level where they cannot be governed by obstruents anymore: following Harris (1990), [r] would consist of two elements (A,I), [1] would be defined by three primes (?,A,I) and [n] even by four (?,N,A,I). Were there such a thing as the cor-

Substantial condition on IG: the internal structure of consonants

59

onal prime R, the complexity of these segments would even reach three, four and five elements, respectively. Hence, the examination of sonorants challenges the distribution of governors and governees that is assumed in Standard Government Phonology. If segmental complexity determines the headship in consonantal domains, sonorants cannot be systematically reduced to governees for they are too complex. As the discussion regarding the opposition between closed and open syllables unfolds, it will appear that the only relevant communication among consonants in CVCV is the one that holds within clusters of rising sonority TR (i.e. branching Onsets). The heaviness of sonorants purports that the lateral relation which structures this kind of sequence is head-final, rather than head-initial. Recall from §38 that this is precisely what CVCV predicts: since there are no branching constituents, all interconsonantal (and all other) relations are necessarily of the interconstituent type and hence head-final. The following section questions another fundamental property of Harris' (1990) model.

52

4.3.7. Complexity counts only Place definers

53

4.3.7.1. All stops have the same degree of stopness It was mentioned earlier (§42b) that one cumbersome feature of Harris' (1990) consonantal representations is the fact that they allow for both place and manner definers to be the head of a segmental expression. If heads contribute more to the phonetic output than operators, ?-headed stops (such as bilabials) are expected to be more "stoppish" than those stops where the occlusion element ? is only an operator (such as dental and velars). This prediction is not echoed by any kind of phonetic or phonological evidence. Tolerating primes that define the manner of articulation in head position also contrasts with a firmly established practice from vocalic representations. That is, only place definers can be the head of a segmental expression. As far as I can see, this statement has never been made explicit, but it governs the representation of vocalic sounds in all privative theories. The place definers I, U and A are thus the heads of high, mid (I,U) and low (A) vowels. The angle stone of privative systems is that all primes combine freely, and that all combinations should produce attested segments. In spite of this ground rule, vocalic expressions where the nasal element N, the low

60

1,2 Open vs. closed syllables in CVCV

tone L or the high tone H are the head have never been proposed. It is tacitly assumed that only I,U and A (and of course the null set), i.e. the place definers (sometimes called resonance elements), can endorse headship. Therefore, it is reasonable to assume that this restriction also controls the representation of consonants.36 The problem related to the unwarranted over-occlusiveness of bilabial stops would be solved under these conditions since ? could not be the head of any consonant anymore. But of course, this move produces shockwaves through Harris' (1990) entire system of consonantal representation, the damage of which would need to be evaluated in detail.

54

4.3.7.2. Vocalic sonority disregards manner definers The difference in kind between place and manner definers is also evidenced by another fundamental property of privative representations. Recall that Standard Government Phonology has set out to build a system of segmental representation where the traditional and fundamental notion of sonority is no primary phonological object. Rather than recurring to a conventional prime mirroring [±son] and the like, the sonority value of each segment was supposed to be a function of Charm at first. Since Harris (1990), this notion has been progressively replaced by complexity. The difference between TR- and RT clusters that exists on the surface is traditionally referred to in terms of sonority. Complexity captures this contrast in counting the number of primes that contribute to the articulation of a segment. Hence, complexity endorses the function of sonority. In this situation, it is interesting to consider how sonority is defined in the vocalic area. It is commonplace that sonority increases with aperture: low vowels are more sonorous than mid vowels, which in turn outrank the sonority of high vowels. The three degrees of aperture are defined by I, U and A alone: A, the most sonorant prime, is left alone to define low vowels. 36

For independent reasons, Ritter (1997) proposes a system where manner definers are almost banned from the head position of segments: headedness alone decides on stricture. That is, stops and laterals are headed (but never by a manner-defining prime, except the glottal stop which is H-headed), while fricatives and rhotics are headless. Only nasals are L-headed. Even though Ritter's model of consonantal representation does not completely dispense with manner specifications in head position, the idea of expressing stricture in terms of headedness obviously converges with the claim that only place definers (and, of course, the null set) can head segmental expressions.

Substantial condition on IG: the internal structure of consonants

61

I and U, in isolation or in combination, identify as high vowels, while mid vowels are a merger of I, U on one hand and A on the other. Whether a vowel is nasal or not, whether it does or does not bear a low or a high tone has no influence whatsoever on its sonority. Hence, vocalic sonority explicitly disregards manner defmers such as N, L and H. It is therefore reasonable to assume that the same holds true for consonants. It will turn out shortly (in the following section) that Infrasegmental Government, just as Constituent and Interconstituent Government in Standard Government Phonology, relies on complexity: a consonant A may govern another consonant Β provided that A is more complex than B. Only is complexity calculated according to different principles. Sonorants are more complex than obstruents at the level of place definition. In order to define the headship within a domain of consonantal interaction, the complexity of place counts alone. Therefore, sonorants are governors and obstruents are governees, which implies that TR clusters form a head-final domain.

55

4.3.7.3. Counting manner definers is not fair play: sonorants will be demoted before being able to score Before turning to the implementation of this insight, it is useful to mention another argument in favour of the exclusion of manner definers from the calculus of complexity and hence of lateral relations. One major reason why Charm has been progressively abandoned is the fact that this notion amounted to a hidden sonority prime in the spirit of [±son]. Negative Charm was supposed to be carried by the low and the high tone L" and H", which control non-spontaneous voicing on the consonantal side. Hence, L" or H" were present in all obstruents, but absent from all sonorants. Therefore, obstruents were negatively charmed, whereas sonorants bore a neutral Charm value. However, the question why the low and the high tone, rather than any other prime, were selected as the vector of negative Charm was never raised, let alone answered. Or rather, the only reason that could be found was the reproduction of the surface-opposition between sonorants and obstruents via spontaneous (absence of L" and H") vs. non-spontaneous (presence of L" and H') voicing. As a matter of fact, there was no means at all that would have allowed to determine headship independently from the surface: the overt opposition between spontaneous and non-spontaneous voicing was simply recast as Charm, which in turn determined governing abilities and hence headship.

62

1,2 Open vs. closed syllables in CVCV

The same objection applies to models such as Harris' (1990), which count manner and place deflners alike. In Harris' (1990) and most other systems, the two primes that contribute typical obstruental properties, i.e. the occlusion and the noise elements ? and h, are present in all stops. The noise element h alone defines a segment as a fricative. Nasals and laterals bear ?, but not h. However, L and H are only present in obstruents. These elements never contribute to the articulation of sonorants. The overall picture thus shows a clear disbalance between the complexity of sonorants and obstruents: the latter are more complex than the former. But this contrast in complexity is just as artificial as was the one established by Charm, and for identical reasons. Sonorants will never be able to compete with obstruents because they lack the tone elements L and H as well as the noise element h by definition. The complexity slope that allows to crown obstruents as heads of interconsonantal domains is largely created by the institutionalised advantage that L, H and h confer to obstruents. In other words, Charm-driven headship was fixed once and forever before any argument involving segmental alternations could be produced. The same holds true for complexity-driven headship if manner and place definers are counted on a par: no argument based on a segmental alternation can raise sonorants to the complexity level of obstruents, which benefit from the in-wired presence of L, H and h. This surface-anchored disbalance between sonorants and obstruents ought to be done away with: only comparable properties of both types of articulations should compete. That is, place is the only feature that is shared by all sounds: vowels, sonorants and obstruents possess a place of articulation and the relevant primes. On the other hand, only consonants can be occlusive, fricative or lateral. And within consonants, only obstruents can be voiced or voiceless. The only common ground on which competition for complexity that is subjected to surface-independent arguments can be based is the level of place definition. This is another reason why Infrasegmental Government only considers place definers in order to determine segmental complexity, and hence lateral headship.

56

4.3.7.4. The manner-place distinction is commonplace elsewhere In other privative models such as Dependency Phonology, this clear distinction between Place and Manner has always been taken for granted. The most fundamental split of melodic representation in Dependency Phonology relies on the opposition of the so-called Categorial (Manner) and Ar-

Substantial condition on IG: the internal structure of consonants

63

ticulatory (Place) Gesture (Anderson & Ewen 1987). This approach has been further developed by Hulst (1994b, 1995:92s, 1999:94ss,2000) into Radical CV Phonology, where Place is defined in the Locational or the Place Gesture, as opposed to the Categorial or Manner Gesture. The same holds true for all versions of Feature Geometry, where the place of articulation is defined under a specific node of the arborescence that does not admit any manner specifications. The following section shows how the contrasting complexity at the level of place definition can be recast in terms of a lateral relation.

57

4.4. Infrasegmental Government: how complexity conditions interconsonantal relations It could be asked why complexity rather than any other consonantal property should control interconsonantal relations. And also, why a more complex consonant should govern a less complex peer, rather than the reverse. The present section answers these questions. Let us first look at the organisation of place definers when two consonants come to stand next to each other. The general properties of consonantal representations have been introduced on the foregoing pages. They recognise four primes that define the place of articulation, consonantal as well as vocalic: I (palatality), U (velarity), A (aperture) and Β (labiality/ roundness). Since I and U never cohabitate within a given segmental expression (front rounded vowels are a combination of I and B), they always share an autosegmental line. Under these provisions, two adjacent consonants may oppose, on a given autosegmental line, 1) one prime each, 2) no prime on either side or 3) a prime on one side, facing no prime on the other. Table (48) provides illustration for some pairs of adjacent consonants that involves the place-defining lines I/U and A (heads of segmental expressions are underscored). 37

37

The element Β must be granted a special status since it is a place definer in the consonantal area (labiality), but contributes only manner on the vocalic side (roundness). This issue is further discussed in Scheer (1996,1999a).

64

1,2 Open vs. closed syllables in CVCV

(48) consonantal interaction a. consonantal interaction possible t r Ρ r _0— 1 I/U 0— I A

0—A

0—A

b. consonantal interaction impossible η r s r I/UA

k r 0—1

f r 0—1

U—A

A—A

I r

i—ι-

i— i

U— I —

A—A

A—A

A—A

t Ρ - 0— 0 -

0—0

r 3 I— I

A—A

Further detail regarding the internal structure of the consonants shown cannot be addressed here (see the aforementioned references). The only information that feeds the purpose of the present discussion is the observation that under (48a), a prime belonging to one consonant is opposed to an empty slot ("0") at least on one line. By contrast under (48b), either two primes face each other, or the slots are empty on both sides. Based on these representations, the definition of Infrasegmental Government is as under (49). (49) Infrasegmental Government (IG) - final version a. a consonant A may contract a governing relation with its neighbour Β iff there is a place-defining autosegmental line where A possesses a prime, while the corresponding slot in the internal structure of Β is empty. In this situation, the prime belonging to A governs the empty position of B. b. the empty Nucleus that is enclosed within such a domain of Infrasegmental Government is circumscribed. Its Empty Category Principle is satisfied.

Infrasegmental Government is the equivalent of Proper Government at the level of the internal structure of segments. At the syllabic level, Proper Government describes a lateral relation whereby a contentful position establishes Government over an empty position. Infrasegmental Government does the same thing below the skeleton (and is therefore called "infrasegmental"). Also, the effects of both operations are identical: an empty Nucleus is circumscribed and must not appear on the surface. This is the case of the target at the syllabic level, and also of the Nucleus that is enclosed within a domain of Infrasegmental Government.

Substantial condition on IG: the internal structure of consonants

65

Infrasegmental Government also shows that relations of the dependency type are not restricted to inter-constituent relations (Proper Government, Licensing, i.e. "lateral") and inner-segmental structure (headoperator, i.e. "vertical"). They also exist among phonological primes that belong to different segments. Infrasegmental Government may be seen as the lateral expression of the infrasegmental dependency relation that is known to hold "vertically" among the primes of a given segment. In sum, the idiosyncratic property of each consonant which is known as complexity produces a situation where sonorants are governors, while obstruents are governees. Now recall from §35 that a governing relation holds among TR clusters, but not among RT sequences. If sonorants are governors and obstruents governees, this is rather unexpected: why should the R be able to interact with Τ in a TR, but not in an RT cluster? The following section addresses this question.

58

5. Phonotactic condition on Infrasegmental Government: Government Licensing Apart from complexity, consonantal interaction is subject to a condition of phonotactic nature, which is identified under (50) below. (50) Government Licensing a domain of Infrasegmental Government may only be established if the head of the domain is licensed to act as a governor by its own Nucleus. Only contentful Nuclei are able to license.

The notion of Government Licensing has been introduced by Charette (1990,1991) in a different syllabic environment, and using different vocabulary.38 The very idea expressed by Government Licensing, however, is independent from any particular syllabic theory. It expresses a fundamental insight into the relationship between consonants and vowels in phonology: consonant clusters need a vocalic support in order to exist. In CVCV, the licensing condition on Infrasegmental Government has the effect of excluding consonantal clusters before empty Nuclei. This is shown under (51) below.

38

The original data and analysis are further discussed in § 125.

66

1,2 Open vs. closed syllables in CVCV

(51)

a. well-formed structure Lie

b. ill-formed structure Lie

C I

V, I

C I

V2

C

V, I

Τ

[ha kkalaßim] "the dogs" vs. /ha raqahim/ —• [haa raqahim] (gutturals and [r] cannot geminate in this language; lengthening of [a] occurs before nouns that begin with a consonant of this kind).

93

2.6. Initial vowels are stable because they are preceded by an empty Nucleus The interpretation of the left pink panther as an empty CV unit also makes sense in regard of the obligatory presence of the first vowel of the word in many languages that tolerate clusters word-internally. The empty Nucleus of the initial CV is subject to the Empty Category Principle just as any other empty Nucleus. Since it is neither word-final nor enclosed within two consonants that can interact, it requires Proper Government from a follow-

Why do pink panthers always do the same things ? 101

ing Nucleus. This Nucleus may not be empty. In languages of the Semitic type where two consonants are always separated by a vowel that enjoys a morphological value, this situation leads to a ban against #CC sequences, as in the aforementioned Tiberian Hebrew and Akkadian case. The French configuration that allows for #TRaC, but does not tolerate #TR0C, has been discussed at greater length in §90.

94

2.7. Czech prefixes again Finally, another fact indicates that word-initial consonants occur after an empty Nucleus. Recall from §§32s that Proper Government may apply over a sequence of Czech word-initial consonants only if this cluster is 1) monomorphemic and 2) of rising sonority TR. Moreover, the cluster must not be broken up by any vowel in some other form of the word. Thus, a root like /ber/ that appears in both zero grade bor-àt "to take" and as vocalised ber-u "I take" will provoke vocalised prefixes: ode-bor-at "to take away". The final Nucleus of the prefix is vocalised because the radical [a] is called to govern the empty Nucleus that is enclosed within the [b] and the [r]. Therefore, the préfixai Nucleus remains ungoverned and appears on the surface. This situation is depicted under (74). (74)

C

V I o

C I d

V A

C

V

C

V

C

V

b r a e ode-br-at "to take away"

Roots that begin with RT clusters produce the same effect regardless of whether they show a vowel within the cluster in other forms of the word. Some illustration appears under (75).

102

1,5 The beginning of the word: "# " = CV

(75) Czech prefixes are always vocalised when they precede an RT-initial root cluster

VRT-

>/RVT

a. RvT-forms of the same root exist mk ode-mkn-out od-myk-at jm pode-jm-out pod-jim-at 1st beze-lst-ny lest b. RvT-forms of the same root do not exist rv ode-rv-at fv roze-Fv-at

to unlock pf, ipf to seize from below pf, ipf guileless, ruse to tear off to begin to shout

Hence, roots with initial TR clusters that are disrupted by a vowel in some other form (ode-brat) line up with RT-initial roots (ode-rvat): both provoke the vocalisation of prefixes. They are opposed to TR-initial roots whose initial cluster is never broken up by any vowel, e.g. podo-bradek "double chin". In this case and only in this case, the radical vowel may reach the prefix, where it silences the final Nucleus. Illustrated is provided under (76) below. (76)

PG

i C

V

C

I

I

I

p

o

d

V

I -

C

V

O

C

I

I

I

V

I

C

I

V

I

C

b mauvais. The evolution of intervocalic velars is a good deal more complex because they may also be subject to palatalisation. Detailed evidence and discussion are provided in Vol.2,1.7.3.3. In any event, velars weaken in intervocalic position in one way or another, just as other consonants do. See Bourciez (1926:130ss), Pope (1934:294,302,333,341), Chaussée (1974:46ss,54ss) on this issue. Velars disappear in Codas, but leave a trace in the form of a palatal element (or a labio-velar in the case of [g] before [m]), which then combines with the preceding vowel, cf.facta > faite, rig(i)da > raide, sagma > (bête de) somme. Throughout Gallo-Romance, stops are affected by various palatalisations. Latin [j] palatalises all stops, and in addition [k,g] palatalise when followed by [i,e,a]. The case of Latin [pj], [bj] is further discussed in Scheer & Ségéral (2001) and §565. For all stops that undergo palatalisation, the contrast between strong (i.e. Coda Mirror) and weak position (i.e. V V, Coda) is preserved. In palatalising contexts, Modern French shows fricatives ([sj,3], from Old French affricates [ts.?>d3]) in strong, but a palatal glide (or zero) in weak positions. Compare the development of [k] in caru, area > cher [$εκ], arche [axJ] vs. pacare > payer [peje], and of [d] in diurnu > jour [3UK], ordiu > orge [0K3] vs. modiolu > moyeu [mwaje].

124

I,6 The Coda Mirror

is, the Coda Mirror produces maximal contrast when compared to the intervocalic position, and to Codas. Damage or preservation of Latin obstruents crucially depends on the position in which they occur. For this reason, the environments that are identified as the Coda Mirror here have been traditionally described as the Strong Position in the Romance literature. 70 The French data discussed illustrate the diachronic reality of the Coda Mirror. In the following section, its relevance in the synchronic distribution of stops is demonstrated. The language under focus is Somali (Cushitic), which guarantees that the Coda Mirror is not an areal or a genetic feature of Romance or Indo-European.

3.3. Somali stops Somali, a Cushitic language spoken in Somalia, Djibuti, Ethiopia and Kenia, also illustrates the Coda Mirror pattern. The distribution of Somali stops is such that plosives can only be observed word-initially and after heterosyllabic consonants, while lenited allophones thereof occur in other positions. First note that Somali lacks branching Onsets altogether: words may not begin with more than one consonant, and word-internal clusters are maximally twofold. Hence, the syllabic inventory is limited to CV(V) and CV(V)C. It may thus be concluded that all word-internal clusters are heterosyllabic.

70

In short, palatalisations affect stops in all environments, but the results respect the difference between strong and weak positions. This fact is striking enough to lead Pope (1934:96) to call both contexts "initial": "consonants are said to be initial : (i) when they stand at the beginning of a word, (ii) when they stand at the beginning of a syllable, if preceded immediately by a consonant, e.g. in the word portare both ρ and t are termed initial" (emphasis in original). Bourciez (1926:147) writes: "Pour une consonne, la position la plus forte consiste à se trouver soit à l'initiale du mot, soit à l'intérieur derrière une autre consonne". On the other hand, Pope (1934:97s) states that "final consonants were in a weak position", and "single consonants in intervocalic position [...] were [...] in a weak position".

The Strong Position: empirical evidence 125 The stops under concern are /b,t,d,k,g/.71 Let us first consider the dis72 tribution of the voiced subset, as under (89).72 (89)

a. # _ sg indef b beer

c. Coda

b. Coda pi/ 3 sg mase pf garb-o

d. V _ V

_C sg def

_# sg indef

pi

garab'-ta dab'-ka

garab 1 dab 1

daß-atf

heßed 7 -ka

heßed 1

geed'-ka

geed 1

nirig'-ta

nirig"

d dile hebM-aj

geeö-ad 1

g gaf nirg-o

n

deg -ta

deg 1

φγ-ο

As may be seen, the only context in which /b,d,g/ appear as such on the surface is the Coda Mirror, i.e. word-initially (89a) and after a Coda (89b). In all other environments, allophones thereof occur, that is the spirantised versions [β,δ,γ] intervocalically as under (89d), and unreleased plosives in Codas as under (89c).73 The alternations shown are based on suffixation that triggers a zero in the place of the second vowel in bisyllabic stems CViCViC in case the suffix is vowel-initial. For example, /nirig/ "young female camel" appears as nireg- when the plural morpheme -o is added, but surfaces as nirig- with the singular definite suffixes -ka (masc.) and -ta (fem.), as well as in the singular indefinite where no suffix is attached. Therefore, the stem-final consonant comes to stand in a position adjacent to its root-medial neighbour, i.e. after a Coda in nirg-, but not in nirig-.

71

[?] and [33] are not mentioned since they are only partially involved in the regularity at hand. More information on the retroflex [dJ and the uvular [q] is

72

73

offered in note 76. There is no [p] in Somali. Glosses (line by line, -o in column b is the plural marker, while -aj denotes 3 sg masc pf): "filed, shoulder, fire, killer, tame animal (heb^d-aj "he became tame"), tree, error, young female camel, ear". Unless otherwise specified, the gloss provides the lexical meaning of the various grammatical forms for every line. See Orwin (1993), Armstrong (1934) for a detailed phonetic characterisation of the Coda allophones.

126

1,6 The Coda Mirror

The distribution of voiceless stops, while following the same pattern, is slightly more complex. Table (90) provides relevant illustration. 74 (90)

a. # _

b. Coda

c. Coda _C

sg indef t

1 sg

l_pl

_# imp. 2sg

gunt-aa

gunucT-naa

gunud" !

sunt-aa

sumad'-naa

sumad1 !

ark-aa

arag'-naa

arag' !

durk-aa

durug'-naa

durug1 !

d. V _ V sg def

tuug"1 /mindi-ta/ = [mindi-öa]

k kal /kursi-ka/ = [kursi-ya]

Again, the only opportunity to observe /t,k/ on the surface is when they appear in a Coda Mirror position, i.e. under (90a,b). 75 In all other contexts, the same allophones occur as before, i.e. unreleased stops in Codas, and fricatives intervocalically. In addition, allophones of /t,k/ are always voiced. The underlying identity of the root-final consonant of the verbs gunt-, sunt-, ark- and durk- is evidenced in post-Coda position (90b). The

same consonant surfaces as an unreleased voiced stop in Codas, cf. (90c). Underlying voiceless plosives spirantise and voice when occurring intervocalically. This process is shown under (90d), where the singular definite markers /-ta/ (feminine) and /-ka/ (masculine) appear as [V-öa], [V-ya], respectively. The allophonic variation of Somali stops may thus be summarised as follows. Stops occur in Coda Mirror positions, spirants intervocalically and

74

75

Glosses (line by line): "thief, knife, to tie a knot, to brand, pestle, chair, to see, to move". Somali does produce intervocalic [t] and [k]. However, on the basis of (i) morphological evidence, (ii) inhibition of regular vowel-zero alternations and (iii) resistance to intervocalic voicing, it can be shown that all of these cases in fact represent underlying geminates. Compare for instance [joog'sadaj] "I ceased" = /joog-sat-0-aj/ = /root + affix + personal marker + preterite/ with [joog'sataj] "you (sg) ceased" = /joog-sat-t-aj/. Ségéral & Scheer (2001b), Barillot & Ségéral (forth) and Ségéral (in press) discuss this issue at length. Significantly, [t] and [k] never appear in Codas (Orwin 1993:253) where, of course, geminates cannot exist.

The Strong Position: empirical evidence

127

unreleased stops in Codas. 7 6 Unless the Coda Mirror {C,#} is granted a non-disjunctive phonological identity, this distributional situation remains mysterious.

119

3.4. Sievers' Law: the vocalic face o f the Coda Mirror

120

3.4.1. Vocalic effects o f the Coda: closed syllables Like the Coda, the Coda Mirror also has a vocalic manifestation. Codas trigger so-called closed syllable effects, among which vowel-zero alternations. Relevant data have been introduced in §16. They are repeated under (91) for convenience. (91) Moroccan Arabic German

zero C_C-V kit0b-u

vowel C_C-0 k0tib

vowel C_C-CV kittib

gloss

inner-e

inner

inner-lich

write pf act 3pl, 3sg, 3sg causative inner+infl, inner, internal

Tangale (Chadic) Somali (Cushitic) Turkish

dob0-go

dobe

dobu-n-go

called, call, called me

nireg-o

nirig

nirig-ta

dev0r-i

devir

devir-den

young female camel pi, sg indef, sg def transfer ACC, NOM, ABL

Slavic (e.g. Czech) Hungarian

lok0t-e

loket

loket-ni

elbow GENsg, NOMsg, adj.

maj0m-on

majom

majom-ra

Hindi

kaarak-ôô

kaarak

Kolami (Dra- kin0k-atun kinik vidian)

76

monkey Superessive, NOM, Sublative kaarek-nee "case" oblique pi, NOMsg, agentive kinik-tan "break" present, imperative, past

This generalisation also holds true for [ j in branching Latin platea P_ vulg piovere b germ * blank *blastimare f florem fiamma k claudere clavus vulg glacia g_ glandem multu V_C calceare v_v tabula dolorem luna lavare

Onsets Italian piazza piovere bianco biasimare fiore fiamma chiudere chiodo ghiaccio ghianda molto calzare tavolo dolore luna lavare

gloss place to rain white to blame flower flame to close nail ice glans much to shoe table pain moon to wash

As may be seen, lat [CI] becomes [Cj], whereas lat [1] is left undamaged in Codas (multu > molto "much"), intervocalically (tabula > tavolo "table") and in word-initial position (luna > luna "moon"). However, the lenition of branching Onsets is a rather difficult matter since there are also cases where the head of a branching Onset undergoes lenition. The detail of one such instance is discussed in Vol.2,1.7: in the evolution from Latin to French, heads of branching Onsets are either lost altogether (with subsequent compensatory lengthening of the following sonorant, e.g. petra > pierre "stone") or spirantised (e.g. labra > lèvre "lip"). But here again, the behaviour of branching Onsets is entirely unrelated to stress, to the effect that Licensing Inheritance makes a prediction that does not meet fact: branching Onsets is promised the same destiny regardless of whether they occur before the dominant or the recessive Nucleus of a foot. However, Licensing Inheritance predicts that they should be stronger in the former case since the licensing path is one step longer from the prosodie head via the recessive Nucleus.

198

3.2.3. The status of Codas and word-final consonants Harris (1994a:207) calculates the status of Rhymal Adjuncts only according to the licensing path that is run since the last nuclear licensor. Also, he disregards Constituent Licensing that is provided by the preceding Nucleus

Harris (1994a): latéralisation

of causality, but not of structure

221

when determining to which extent Rhymal Adjuncts are incline to lenite.131 Consider the diagram under (164) that shows the local licensing configuration of Rhymal Adjuncts. (164) Rhymal Adjuncts are indirectly licensed R

X

X

X

X

tJtJ ICL

ICL

It may be seen that Rhymal Adjuncts occur at two removes from the local head. It is not clear, however, why it should not matter whether this is a dominant or a recessive Nucleus. In the latter case, the Rhymal Adjunct is predicted to inherit less licensing potential than in the former. In other words, stress is predicted to condition a contrast in the behaviour of two kinds of Codas. Internal and word-final Codas may exhibit such a differential (cf. chapter 11,12 §524), but this is again a purely positional event. Cases where two different lenition patterns are encountered in Codas before stressed vs. unstressed vowels do not seem to be on record. Finally, let us turn to word-final consonants. Harris (1994a:209) recurs to parametric Licensing of final empty Nuclei in order to establish that they are only indirectly licensed. Consider their status under (165). (165) word-final consonants are indirectly licensed Licensing of final empty Nuclei

ICL^

Π Ο I χ I C

131

Ν I χ I 0

#

The only function that Harris grants to Constituent Licensing within Rhymes has been exposed in §190: Rhymal Adjuncts in languages from which superheavy Rhymes are banned need to be licensed twice.

222

1,8 A syntax of phonology

Word-final consonants belong to an Onset whose Nucleus is empty and licensed to remain without phonetic expression according to a language-specific parameter. Therefore, their Licensing potential is filtered by the final empty Nucleus, which is itself a licensed position.

199

3.2.4. Licensing Inheritance: how it works On the bottom line, Harris is able to define all locations that are notorious for lenition across languages at the same rate: they occur at two removes from the source of Licensing. This is true for foot-internal Onsets (provided the foot is head-initial), Rhymal Adjuncts as well as word-final consonants. In contrast to the resulting impoverished Licensing potential of these positions, foot-initial Onsets are directly licensed by their Nucleus and therefore strong. Harris thereby construes a three-to-one opposition, which is identified under (166) below. 132 ( 166)

classification of positions according to Licensing Inheritance weak: 1. consonants before unstressed vowels (i.e. foot-internal) 2. internal Codas 3. final Codas b. strong consonants before stressed vowels (i.e. foot-initial for left-

distance from the head

a.

headed feet)

"Ί I Γ J "Ί λ

two removes

one remove

J

Also, Harris has developed a surrogate way of referring to the familiar Coda context {#,C}: he is able to address this typical lenition site in a non-disjunctive fashion. Consonants are not weak, he argues, because they belong to the same constituent. Rather, they are prone to lenition since their Licensing potential is only indirectly inherited. Harris thus carries out the latéralisation of a vertical causality. However, internal and final Codas are not alone in this situation: post-tonic intervocalic consonants are also indirectly licensed. Therefore, it can hardly be claimed that the Coda context

132

Recall that this result hinges on the restriction of Licensing paths to some local level in the case of internal Codas. The Licensing potential of word-initial and word-internal Onsets is determined at the foot-level.

Harris (1994a): latéralisation of causality, but not of structure

223

has found a lateral definition: all, but not only the consonants known as Codas occur in indirectly licensed positions. The overall goal of Licensing Inheritance is to uniformly characterise the three locations where lenition is notorious: (166a). Now we have seen in §131 that consonants are often weak in two different ways according to whether they occur in Codas or in intervocalic position. This contrast is unexpressed by Licensing Inheritance. It was mentioned earlier that the soul of Harris' system is stress. This is why feet come into play, and this is why stable patterns that are defined by positional parameters alone cannot be expressed. For one thing, many lenition systems do not care about stress, whereas all take positional strength into account. Also, the lenition of Codas does not appear to ever depend on stress. Hence the implicational relationship between positional and tonic influence that has been discussed in §113. In contrast, the Coda Mirror is a theory that predicts lenition only according to the position of a consonant. It neglects all prosodie factors. An ideal theory would have to stand on positional grounds while allowing for the inclination of the positional pattern by prosodie factors.

3.2.5. The strength of post-Coda consonants In has been mentioned earlier that Licensing Inheritance does not assign any contrasting strength to Onsets according to whether they are preceded by a Coda or not. The particular strength of post-Coda consonants, however, is a hard fact that enjoys a stable record in many languages. Harris (1994a:217ss) implements post-consonantal strength in a way that is entirely independent from Licensing Inheritance. He contends that consonants which occur after Codas are less prone to lenition because they must observe a governing duty. In Government Phonology (since Harris 1990), lenition is viewed as a process whereby a segment loses phonological primes, i.e. becomes less complex. For instance, in Harris' system of consonantal representations, [t] is made of three elements, R (coronality), ? (stopness) and h (noise). When [t] lenites to [?], it loses R and h, hence [?] = ?. Unreleased [t] is lacking h, a flap [r] is deprived of both ? and h, while the corresponding fricative [s] does not license ?. That is, the various outputs of lenition that affects [t] = R,?,h are [ f ] = R,?, [r] = R, [?] = ? and [s] = R,h. In each case, the damage effected by lenition translates as a loss of segmental complexity: all lenited outputs are less complex than [t].

224

1,8 A syntax of phonology

However, recall that complexity also controls Constituent- and Interconstituent Government. Namely, governors must not be less complex than governees. In Harris' view, lenition is present in a given language, and puts pressure on potential targets. In the English case, this predisposition to lenition affects all [t]s regardless of the position in which they occur. This propensity to lenite may be actuated or not, and it can lead to various results ([t\s,?,r]). In case a possible target of lenition has to observe governing duties, i.e. follows a Coda, the reduction of melodic complexity has consequences that are not encountered when it occurs in Codas or in intervocalic position. In order to maintain the complexity slope required for the existence of Interconstituent Government, post-Coda consonants cannot lose primes in such a way that they become less complex than their governee. This is how Harris explains the relative protection of [t] when it comes to stand after a Coda. According to this reasoning, the complexity of [t] should be able to decline as long as the number of primes does not fall below the complexity of the Coda. In other words, if the Coda may be occupied by sounds that show contrasting complexity, the result of lenition on following [t]s should vary according to the melodic identity of the Coda. Harris demonstrates that this prediction is indeed borne out in English: the four lenition systems that were introduced in §184 react in different ways according to whether the preceding Coda is an obstruent, a lateral or a nasal. Table (167) shows the various effects recorded. 133 (167) English t-lenition depending on the preceding consonant (Harris 1994a:217s) conetymological [t] text appears as example A Β C D a. after obstruents Τ V t t chapter, doctor, after, mister t t b. after laterals L__V t t ? t filter, shelter 1 L _# ? ? t fault, belt t c. after nasals t ? 0 t winter, twenty, plenty Ν V N__ # ? ? t hint, sent f That is, [t] cannot give way to lenition when it is called to govern a preceding obstruent, while it is able to decompose in presence of laterals 133

In both rhotic and non-rhotic varieties, preceding etymological [r] behaves like a vowel, i.e. produces the same result on [t] as the intervocalic position. See Harris (1994a:217s).

Harris (1994a): latéralisation of causality, but not of structure

225

and nasals. Harris argues that this is a direct consequence of the contrasting complexity levels: obstruents are made of more primes than laterals and nasals. The latter two possess comparable complexity, and hence produce almost identical results. In every case, the admissible minimal complexity of the governor [t] is a function of the number of primes that is associated with the Coda: governors may not be less complex than their governees.

201

3.2.6. Licensing Inheritance: three major problems The overall picture of Harris' delinearisation of strong and weak positions is objectionable on three grounds: the undisputed role of stress, the obligatory reference to the melody of Codas and the remaining disjunctivity.

202

3.2.6.1. The Coda Mirror context remains disjunctive Let us first address the latter. The major observation that founds the Coda Mirror as a theory of lenition and as a phonologically significant context is the fact that on a cross-linguistic record, consonants seem to be protected against lenition in word-initial position and after Codas. This disjunctive context identifies as {#,C} , and hence turns out to be the exact mirror image of the Coda context {#,C}. Not only is the structural description the reverse of the one that characterises the Coda, but the effects encountered in both locations are also opposite, i.e. strength in the former, against weakness in the latter. The challenge raised by this situation is twofold: for one thing, in regard of obvious reasons that are identical to the treatment of the Coda context in the 70s, the disjunction {#,C} needs to be reduced to a non-disjunctive statement. Moreover, the non-disjunctive identity proposed should somehow possess mirror-properties with respect to the object that characterises the Coda context. The Coda Mirror complies with both requests in identifying the Coda as " 0", while its mirror corresponds to Under Harris' analysis, however, the identity of the Coda Mirror context remains disjunctive. That is, there are two entirely independent reasons why consonants are strong in word-initial position and after Codas. Protection against lenition in the former environment follows from Licensing Inheritance, that is from the fact that word-initial consonants are directly licensed by their Nucleus, which is tonic and thus the head of the foot (in trochaic systems). On the other hand, consonants that occur after Codas are

226

1,8 A syntax of phonology

strong for a reason that is entirely independent from Licensing Inheritance: they may not lenite (or rather, they may not lenite too much) because they have to meet governing obligations in regard of the preceding Coda. Hence, the Coda Mirror {#,C} does not enjoy a unified identity in Harris' system. It remains a disjunctive object whose two members are guaranteed against lenition for reasons that have nothing in common.

203

3.2.6.2. Stress is a secondary, not a primary factor The second issue concerns the primary and obligatory role that Harris grants to stress (something that has already been discussed on various occasions on the foregoing pages). It may be illustrated by the interspreading of positional and accentual criteria within Harris' model. One case in point is precisely the Coda Mirror-disjunction. The strength of word-initial consonants is claimed to stem from the vicinity of the stressed vowel, while postCoda consonants are protected against lenition for a reason that is purely positional. Governing duties within Coda-Onset clusters must be respected regardless of the stress pattern: foot-initial consonants (i.e. preceding a stressed vowel in trochaic feet) of an internal foot have to govern preceding Codas just as much as their foot-internal peers (which occur before an unstressed vowel). The ensuing prediction falls foul of fact: there is no reason why Licensing Inheritance should influence foot-internal consonants if there is no preceding Coda (they are weak), but have no bearing on them when they follow a Coda. In other words, there should be a differential in lenition between post-Coda consonants that occur before stressed (footinitial) vs. unstressed (foot-internal) vowels. This claim is neither supported by fact nor by Harris' system itself, where foot-internal and foot-initial consonants are assigned the same status in regard of lenition in case they occur after a Coda. The only parameter that determines their damage is the preservation of the complexity slope within the cluster. Empirically speaking, thus, the strength of post-Coda consonants does not seem to depend on stress in any way. Their status in regard of lenition is exclusively determined by their position in the string. This is true even for languages like English where stress plays a major role in lenition. Yet, there is a more significant reason why the angle stone of a theory of lenition should be position, not stress. It seems that the behaviour of consonants in all lenition systems on record depends on their position in the string. In some languages, stress comes in as an additional player. But as far as I can see, there is no language where lenition depends exclusively on

Harris (1994a): latéralisation of causality, but not of structure

227

stress and disregards positional factors altogether. Given this including relationship, every theory of lenition should implement an optional module that manages stress, but also be able to describe systems where stress plays no role (such as for example the French paradigm discussed in §117). Also, it should be borne in mind that stress-based systems such as Harris' are forced into a tacit prediction that may appear rather uncomfortable: they have to claim that there are no iambic feet in natural language. This consequence was pointed out in §197b,c: the strength of word-initial consonants can only be derived when the first vowel of the word is stressed, and hence the associated foot trochaic. If iambic feet exist at all, the cross-linguistic co-occurrence of those languages where word-initial consonants are strong with those where feet are trochaic would thus have to be regarded as purely accidental. Since hardly anybody is prepared to believe in such a coincidence, a tacit prediction is made to the effect that iambic feet do not exist at all: they would produce unattested effects across the board. On the bottom line, it appears, again, that the fundamental factor for the calculation of lenition is not stress (and associated categories such as feet), but position.

204

3.2.6.3. In many systems, post-Coda consonants are insensitive to the content of the preceding Coda Finally, the way Harris explains the strength of consonants in post-Coda position leads to another prediction that sits uneasily with cross-linguistic observation. That is, the lenition of post-Coda consonants in all systems should depend on the melodic content of the preceding Coda. This follows from the fact that according to Harris, consonants in this position are strong for one single reason: they have governing duties and therefore must maintain a sonority incline. Hence, they should be free to lenite as long as their governing ability is not affected, which in turn depends on the complexity of the segments that occupy the preceding Coda. In other words, consonants in post-Coda position should always "look at" the melody of the preceding Coda before deciding whether or not they can tolerate depletion. It appears that some languages indeed lenite post-Coda consonants gradually according to the melodic identity of the preceding Coda. But this condition is by no means universal. There are many systems where consonants in post-Coda position are strong regardless of the kind of consonant that occupies the preceding Coda. One such system has been discussed in

228

1,8 A syntax of phonology

§117. Relevant data from the evolution of Latin obstruents in French are reported under (168) below. 134 (168)

cluster T_ -

N__ r 1_

Ρ s Ρ t k s m η r r 1 1

P P t t t t P t P t P t

a. # porta porte tela

toile

b. Coda suppa soupe crispare crêper route rupia goutte [t] gutta voiture vectura festa fête lampe * lampa chanter cantare serpente serpent virtute vertu taupe talpa alteru autre d. V.

c. Coda C route Ρ rupta t plat(a)nu plane

lup(u) marit(u)

loup [lu] mari

ripa vita

This table demonstrates that Latin [p] and [t] in post-Coda position survive into French without any damage regardless of the kind of consonant that occupies the preceding Coda. If it is true that lenition puts equal pressure on all obstruents regardless of the position in which they occur (a scenario that is supposed by both Licensing Inheritance and the Coda Mirror), post-Coda consonants should lenite as much as they can as long as their depletion does not endanger their governing status. The fact that in systems such as under (168) their behaviour does not take into account the melodic identity of the preceding Coda indicates that we are dealing with a parameter: the status of post-Coda consonants in regard of lenition may or may not depend on the melodic properties of the preceding Coda. We will see in Vol.2,III.9.1 that those systems which effect gradual lenition according to the melodic properties of the preceding Coda always produce the same pattern: consonants are relatively stronger if they follow an obstruent than if they are preceded by a sonorant (i.e. the behaviour evidenced for Eng-

134

For expository reasons, only [p] and [t] are considered here: various palatalisations that affect [k] would complicate the picture inordinately. See §117 for further detail.

Harris (1994a): latéralisation of causality, but not of structure

229

lish). In no case will lenition effect more damage in the vicinity of an obstruent than in presence of a sonorant. Note that this discussion directly relates to §180 where the phonological reality of bogus clusters and their translation into CVCV was examined. Also, the parameter at hand reflects the third player that is active in lenition systems (see §113): consonants are relatively stronger if they share some (melodic) property with a neighbour (Honeybone 2001,2002, see also chapter 11,15 §591).

205

3.2.7. Summary Licensing Inheritance has introduced a new function of Licensing: while this lateral force only defined the conditions under which skeletal slots may exist in Harris' (1994a) general system (§193), it now appears as a mechanism that also has bearing on the melody of the associated segments. Hence, the labour on this side is actually shared by Government and Licensing: the former defines the melodic structure in static relations in the lexicon that will not be modified by any phonological event. By contrast, the latter is responsible for melodic alternations which are due to phonological processes that operate on underlying representations. For the sake of completeness, the final version of the picture that Harris draws for Government and Licensing is shown under (169) below. (169) Properties of Government and Licensing according to Harris (1994a) final version Licensing Government a. scope constituent structure melodic structure (dynamic) melodic structure (static) b. condition on the existence of x-slots the existence of melodic contrast c. effect provides support for its target inhibits the melodic expression of its target The overlap of labour that Government and Licensing do is actually objectionable. Indeed, Harris (1994a) has set out to disentangle the confusion that has reigned in Standard Government Phonology: the two lateral forces sometimes concerned the same empirical object, and the same lateral relation covered phenomena that were different in kind (see §136). Harris achieves to make Licensing responsible for the existence of x-slots (which cannot survive in its absence), while Government is at the origin of static

230

1,8 A syntax of phonology

melodic effects (i.e. the distributional asymmetries of neighbouring sounds in regard of sonority). Calling a force that bears on the melodic properties of segments Licensing is thus a step back into former times when the foundations of Government and Licensing were blurred. Therefore, Licensing Inheritance would actually be better off as Government Inheritance. But this would not suit the purpose either since Harris' basic idea is that support is great at the source of strength, i.e. stress, and gradually fades away when it is transmitted to the periphery. Unfortunately, Government does not dispense support. On the contrary, it curtails the segmental expression of its target. It therefore seems that Licensing Inheritance is necessarily at odds either with itself or with the independence of two distinct lateral forces.

206

4. CVCV: latéralisation of both structure and causality

207

4.1. Harris' advances establish a hybrid system: lateral causality but vertical structure The overall result of the foregoing discussion is the following: Harris (1994a) achieves full latéralisation of phonological causality and thereby stands in direct filiation of Kaye et al. (1990) where a syntax of phonology was seeded. However, the structure that supports Harris' lateral system remains vertical in essence: there are branching Onsets, branching Nuclei and branching Rhymes. This in itself indicates that we have not reached the end of the road towards laterality. Now it is interesting to observe that the latéralisation of causality on the grounds of a vertical syllabic support leads to the problems that were mentioned in §191. Recall that Harris seeks to eliminate the residue of vertical causality which controls phonological events (see §§174s,190). The new lateral relation he introduces in order to overcome vertical causality is a Licensing relation which holds between the Nucleus and the Rhymal Adjunct. He is then able to describe Closed Syllable Shortness in terms of this relation: Nuclei may not branch in presence of a Rhymal Adjunct because the latter would fail to be licensed by the head of the domain, that is the first skeletal slot of the Nucleus. In languages where all closed syllables are obligatorily short, Rhymal Adjuncts are said to need double-licensing from both their Nucleus and a following Onset. In languages that tolerate superheavy Rhymes, Rhymal Adjuncts need to be licensed only once (by the following Onset).

CVCV: latéralisation of both structure and causality

231

The existence of this lateral force, however, overgenerates structures that ought not to exist. That is, there does not appear to be any reason why two Rhymal Adjuncts should not be able to cohabitate within a Rhyme in languages that allow for super-heavy Rhymes (where Rhymal Adjuncts need to be licensed only once), as under (156). Also, the same kind of languages should tolerate word-final Rhymal Adjuncts as under (157). I believe that these difficulties arise from the attempt to marry two conceptions that repel each other, i.e. vertical structure and lateral causality. On the assumption of CVCV where both phonological structure and causality are non-vertical, problems of that kind are unheard of. It will be shown in §232 how CVCV tells languages that possess super-heavy Rhymes from those that put a ban on this structure. Yet another problem related to the hybrid character of models with lateral causality and vertical structure has been mentioned earlier. There is no hope that arboreal constituent structure will ever be able to do justice to the word-final situation on non-extrasyllabic grounds. If there are no wordfinal Codas in any language, no parameter can be set that accounts for the fact that word-final consonants sometimes do and at other times do not behave as internal Codas. Chapter 11,12 (§524) discusses this issue at length. These points can be made in order to provoke the last step on the way towards a syntax of phonology: the latéralisation of constituent structure, hence CVCV.

208

4.2. Arborescence is redundant, the null hypothesis for syllable structure is lateral

209

4.2.1. Takahashi (1993): only lateral causality, a farewell to constituency, but no additional empty Nuclei Perhaps Takahashi's ( 1993,2004:141 ss) consequent exploration of the lateral idea in a Standard Government Phonology environment may be considered as the second but last step on the way towards the latéralisation of syllable structure. Takahashi wants to do the same thing as Harris (and even a little more), but without constituent structure at all. He fully accepts Harris' Licensing Inheritance and the additional lateral relations that Harris (1994a) has introduced (Takahashi had access to Harris' 1994a book manuscript in 1993). That is, Coda consonants are not only licensed by the following Onset, but also by the preceding Nucleus. Without discussion, Takahashi does away with Government altogether, to the effect that the

232

1,8 A syntax of phonology

kahashi does away with Government altogether, to the effect that the only lateral force known in phonology is Licensing, as conceived of by Harris (1994a). This is because Government has no function anymore in the model developed by Takahashi: distributional generalisations do not depend on melodic properties of segments in any way. The basic observation that Takahashi makes is the logical consequence of the presence of lateral relations in phonology: vertical constituent structure and lateral relations do the same job. Saying that the two members of the distributional class TR are tied together because they pertain to the same constituent, or because they contract a governing/ licensing relationship, is equivalent. Grammar needs only one device in order to encode distributional generalisations. And theoretically speaking, of course, several devices that do the same j o b twice are a nightmare: no theory can afford redundant statements. I argue in §§398-400 that the implementation of lateral devices into phonology, i.e. the very research programme of Government Phonology, necessarily leads to the kind of hybrid vertical-lateral model that is represented by Standard Government Phonology and Harris (1994a). It can hardly be expected, indeed, that lateral relations, in one single move, throw constituent structure over board and take over its function. An intermediate stage that works on the basis of a compromise between the familiar vertical structure and the new lateral devices is probably inevitable. Facing the hybrid situation, then, vertical constituent structure must be done away with if lateral relations deserve any credit at all. However, the flattening of constituent structure can be done in two different ways: by increasing the number of empty Nuclei, or by maintaining the number of x-slots without any modification. The former solution is the one favoured in CVCV, while the latter is explored by Takahashi (1993,2004:141ss). To cut a long story short, the structures proposed by Takahashi are identical to the ones of Standard Government Phonology as far as the number of x-slots and their consonantal/ vocalic labelling is concerned. The lateral network is the one of Harris (1994a), but there are no constituents left at all. Table (170) shows the familiar representations for two English words that illustrate basic syllabic objects, as well as their equivalents in Takahashi's constituent-less approach.

CVCV: latéralisation of both structure and causality

233

(170) constituents are redundant: Takahashi (1993) a. classical: with constituents b. Takahashi (1993): without constituents tree

entry

Ο

Ν

x'< I X > X

x' I X

t

e

r

>X

R 0 N\ \ 1 X X X X 1 1 1 1 e η t r

Ν χ y

χ

x' x'< I I χ — — χ — > χ I I I I e n t r y

x' I χ—> χ

Indeed, no piece of information is lost. And even graphically, the equivalence of vertical and lateral devices appears since vertical lines under (170a) simply correspond to horizontal arrows under (170b). Headship is indicated by slanting lines in the former, against arrowed lines in the latter representations. Note that "x1 " in Takahashi's conventions stands for the projection of heads. Hence, a dependency relation on the x' level indicates that the leftmost x' is consonantal (an Onset) and receives Licensing from the neighbour to its right (a Nucleus). Two x' units that do not contract any lateral relation identify a Rhyme-Onset sequence. A Coda is normally licensed twice. In case it is only licensed once (something that may occur according to Harris 1994a, see §190), it contracts a relation with its righthand neighbour. This is also how the difference between a Coda and an Onset complement is achieved: the latter never contracts any lateral relation with its righthand neighbour. On the bottom line, there are three and only three different kinds of x-slots: 1) nuclear heads (i.e. heads of the phonological domain), 2) preheads (which is how Takahashi calls Onsets: they occur before nuclear heads) and 3) complements. The latter fall into two categories: those which are licensed only once, and from their left (Nucleus dependent and Onset dependent) on one hand, against those which may be licensed twice, but minimally receive Licensing from their right (Codas). Surprisingly, Takahashi (1993:402ss) eliminates any kind of melodic control over the clustering of segments, even though, as far as I can see, the mechanisms he proposes do not enforce that move. In all models of syllable

234

1,8 A syntax of phonology

structure, the higher organisation into syllabic constituents, morae, lateral relations or whatever phonotactic device that embodies distributional generalisations is a function of some melodic property of segments. Classically, syllabification algorithms build on sonority. In place of this merely observational category, Government Phonology has developed the notion of complexity. At any rate, the fact that TR ends up as a cluster that is united by some force (a constituent, a lateral relation), whereas RT shows disunited behaviour, is a fundamental fact in phonology which necessarily roots in the contrasting melodic substance of Τ and R (Vol.2,11.4.1 enlarges on this issue). Contrary to that, Takahashi (1993:403) "place[s] the burden of indicating information concerning the types of positions onto the positions themselves." In other words, melody does not contribute to the definition of clusters at all since "positions in phonological representations themselves should be responsible for the information regarding licensing relations" (404), and these relations alone decide on consonantal clustering. In my opinion, Takahashi's approach is severely handicapped by putting melody off-side. Nonetheless, it is the most consequent implementation of the lateral idea into an originally arboreal system. Subtracting redundant arboreal information from phonological representation is probably as far as one can get within a hybrid model. For it would be erroneous to believe that wiping out arboreal relations makes them undone: the structure, the clustering and the number of x-slots in Takahashi's (1993,2004) model are exactly the ones of the initial arboreal constituent structure. This is because lateral relations were injected into a vertically defined model in the first place. They were then made to parallel the function of vertical constituent structure. Finally, constituent structure is abandoned in Takahashi's (1993,2004) move. The difference with CVCV, then, is twofold. For one thing, CVCV is not just a lateralised version of Standard Government Phonology from which constituents have been removed. In CVCV, the network of lateral relations that define clusters of segments are not the same as in Standard Government Phonology. This may be seen when comparing for example the definition of Coda consonants: "a consonant belongs to a Coda iff it is licensed by a following consonant" (Standard Government Phonology and Takahashi 1993,2004) vs. "a consonant belongs to a Coda iff it is followed by a governed empty Nucleus" (CVCV). The other difference is the existence of additional empty Nuclei in CVCV, as compared to Standard Government Phonology and Takahashi (1993,2004). Arguments in favour of these are discussed in the second part of the book.

CVCV: latéralisation of both structure and causality

210

235

4.2.2. The existence of lateral relations makes arboreal structure redundant In linguistics, arboreal structure is the formal translation of co-occurrence restrictions that hold among objects which are adjacent in the linear string: X and Y cohabitate within some constituent if some properties or the occurrence of X depend on some properties or the occurrence of Y, and vice versa. In case no such distributional restrictions are encountered, X and Y are said not to be sisters of the same node (§211 below enlarges on this issue). In addition to the vertical structure that expresses co-occurrence restrictions, Government Phonology has produced a parallel device that is concerned with the same empirical record: lateral relations. In Standard Government Phonology as well as in Harris' (1994a) system, constituent structure is doubled by Government and Licensing, to the effect that cooccurrence restrictions are expressed twice: once by the arboreal structure, and another time by Government. An obvious objection against this kind of hybrid interpretation is its redundant character: if a vertical property (constituent structure) explains why the distribution of adjacent objects is not free, why do we need a second reason for the facts observed (Government)? In Standard Government Phonology, arboreal and lateral structure is hierarchised in the sense that the former results from the latter. Constituent structure exists because X can govern Y, but Y cannot govern X. And in turn, governing abilities are defined in terms of a complexity differential among adjacent objects. In other words, a lateral contrast is transformed into arboreal structure. Therefore, the relevant syllabic generalisations are expressed twice, whereby one of the resulting structures (the vertical arborescence) may be completely derived from the other (lateral relations). The reverse is not true. It does not make sense for any theory to declare an entirely predictable structure the real cause of phenomena that can as well be expressed by the primary lateral network. §§398-400 further examines the consequences of its hybridity for Standard Government Phonology: its mixed arboreal-lateral essence enforces unnecessary axiomatic statements such as "constituents are leftheaded". Mainstream models of syllable structure that rely on some syllabification algorithm actually show identical practice. A difference in sonority among adjacent consonants that may not be reversed is transformed into vertical syllable structure via descriptive statements such as "within a branching Onset, sonority must increase". These approaches, however, lack

236

1,8 A syntax of phonology

lateral relations and hence escape the objection of hybridity (§211 below develops this line of reasoning). Of course, CVCV does not deny the existence of co-occurrence restrictions among objects that are adjacent on the surface. Only does it not grant any vertical interpretation to the lateral differential observed. The functional load of vertical syllable structure is taken over by the additional empty Nuclei which CVCV supposes. That is, the consonants of a TR cluster can contract a lateral relation, but no such interaction is possible among the members of an RT sequence. The consequence thereof is not the construction of contrasting arboreal structures, but the different status of the empty Nucleus that both types of clusters enclose: it is circumscribed in the former, but not in the latter case. This difference, in turn, enforces the existence of another lateral relation in the RT configuration: the following vowel is called to govern the empty Nucleus that is enclosed within an RT cluster, while no such request is issued by TR clusters. And again, this fact has consequences on the vowel that precedes the cluster, as well as on its second member: the vowel preceding an RT sequence will be the target of no internuclear relation (it therefore stands in a closed syllable) since the following vowel exhausts its lateral potential in satisfying the ECP of the intervening unpronounced Nucleus. On the other hand, if this empty Nucleus is taken good care of as is the case for TR clusters, the following vowel is free to contract a relation with its preceding peer, which therefore stands in an open syllable. For the same reason, the second member of the cluster escapes Government in RT configurations, but not in TR clusters. Hence its strength in the former, but not in the latter case. This discussion is designed to show the fundamental difference between Standard Government Phonology and mainstream models on one hand, and CVCV on the other. On the analysis of the latter system, the initial difference in an idiosyncratic property of adjacent objects (complexity) gives rise to a lateral contrast (presence or absence of a lateral relation among them), which then produces various secondary effects through a domino reaction. In the entire picture, all causalities are strictly lateral, and in no event does arboreal structure emerge. In other words, there is no phonological effect that stems from the fact that two or more segments enjoy a contrasting vertical status (i.e. belong to different constituents). Therefore, the functional load of arboreal syllable structure is nil.

CVCV: latéralisation of both structure and causality

237

211

4.2.3. The null hypothesis for syllable structure is lateral: why taking two steps if one is enough ?

212

4.2.3.1. The central tacit assumption: co-occurrence restrictions are due to arboreal structure We are thus left with a picture where syllabic regularities may be encoded either by traditional arboreal or by lateral structure. It should not be quite difficult to agree on the fact that hybrid models do not qualify: they are redundant since they express the same regularity twice. Having the same job done twice, however, is a nightmare for any scientific theory: one of the two devices must go. The question to be asked, then, is whether it can be decided which is the one to be done away with: the arboreal or the lateral candidate for the representation of syllable structure. Part Two exposes ten arguments in favour of the lateral solution. However, besides regular phonological argumentation there is a strikingly simple reason: the lateral option encodes the relevant observations directly, while they are only indirectly reflected by arboreal structure. In other words, the null hypothesis of syllable structure is lateral. This is what I set out to show in the present section. In the generative tradition since Kahn (1976) (but not only in there), syllabic generalisations are looked at through the arboreal prism. In line with syntax, co-occurrence restrictions of two adjacent segments are interpreted as the consequence of arboreal structure: the segments in question are dominated by the same node. This approach seems self-evident: it is tacitly transposed to phonology without any explicit motivation.

213

4.2.3.2. Co-occurrence restrictions: the parallel with syntax is phoney Let us thus have a closer look at the alleged parallel with syntax. The fundamental motor for syllabification is the status of consonant clusters: in a V1C1C2V2 sequence, it must be found out whether Q "belongs" to the preceding or to the following vowel. All other syllabic matters follow from this decision: Vi will either stand in an open or in a closed syllable, and Ci will either belong to a Coda or to an Onset. Various melodic effects are then conditioned by the syllabic status of Vi and Ci. However, co-occurrence restrictions do not help us in any way when it comes to make the decision regarding Ci: in an average Indo-European language, the distribution of obstruents and sonorants in intervocalic clus-

238

1,8 A syntax of phonology

ters is free. All combinations are found: TT, RT, TR, RR. Hence, there can be no argument made on these grounds in order to erect a common arboreal node over either of the clusters mentioned, rather than over another. The syntactic picture is quite different: given two major categories of lexical items (say, prepositions, nouns, verbs, adjectives), their order is not usually free. Therefore, phonologists rely on the "initial anchor" (Kahn 1976, §§382,396 expose classical syllabification algorithms at greater length). In word-initial position, indeed, only TR clusters are found in average IndoEuropean languages. Since the first consonant in this position cannot be attached to anything to its left, both consonants must cohabitate within the same constituent, which will be called the Onset. It is then understood that the sonority slope of the initial TR clusters is a condition on sisterhood that holds for the entire language: "within a branching Onset, sonority must increase". Hence, intervocalic TR clusters are branching Onsets as well, while RT, TT and RR clusters are not. Since they do not qualify for sisterhood, they must belong to distinct constituents, which means that they instantiate Coda-Onset sequences. But here again, this way of determining sisterhood is unprecedented in syntax. It is certainly true that syntactic co-occurrence restrictions are brought to light by applying various substitution and movement tests. However, there is nothing like a unique and recurrent anchor that can be applied mechanically: a computer can perfectly well determine the phonological constituency of an entire language according to the algorithm "the members of a cluster are sisters iff the cluster exists word-initially"; no machine can be designed that is able to achieve the same result in syntax. Finally, let us examine the notion of sonority. All phonological major classes contract a clear relationship since they are ordered on the sonority scale: stops < fricatives < liquids < nasals < glides < vowels. This is absolutely critical when it comes to express syllabic generalisations: it is not mere class membership that induces co-occurrence restrictions. That is, sisterhood or non-sisterhood is never determined by making reference to major classes alone: "class X can co-occur with class Y". Rather, generalisations always make reference to the relative sonority value of adjacent consonants: "an obstruent is the sister of the following consonant iff this consonant is more sonorous". Only the relative sonority in regard of the neighbour, i.e. the lateral relation among the two items, counts. Again, there is nothing remotely comparable in syntax. For one thing, major classes of items are not arranged on any scale: there is nothing like sonority in syntax. Major classes are; they do not contract any particu-

CVCV: latéralisation of both structure and causality

239

lar relation with other major classes in terms of inferiority or superiority on some common scale. 135 And of course, syntactic decisions on sisterhood are never taken according to any "slope" that adjacent segments show according to this inexistent common scale. Finally, and perhaps most importantly, attention needs to be drawn on the fact that arboreal structure in syntax enjoys a kind of motivation that has got nothing to do with co-occurrence restrictions. That is, many syntactic phenomena such as Binding or Locality (relativised minimality, Rizzi 1990) make sense only when certain chunks of the linear string, hierarchically speaking, do not stand on a par with others. This kind of independent motivation is hard to come by in phonology - in any event, phonologists do not call on any evidence different from co-occurrence restrictions in order to determine arboreal syllabic structure. The overall picture, then, is consistent if it is true that the major difference between syntax and phonology is the absence of a tree-building mechanism in the latter module. This has been argued for in the foreword §2, where the notorious absence of recursion in phonology was shown to follow from the inability of phonology to create arboreal structure: a structure is recursive iff a given node dominates a node of the same type. We now see that the absence of arborescence in phonology has still another correlate: syllable structure is motivated by co-occurrence restrictions alone in phonology. The arguments that can be made on the grounds of independent hierarchical phenomena in syntax such as Binding and Locality are absent from phonology: phenomena of this kind, i.e. which suppose arboreal structure, do not occur.

135

The subcategorisation of major syntactic classes according to whether they are [±noun] and [±verb] (Chomsky 1970) does not give rise to any scale on which all basic lexical syntactic categories are hierarchically ordered. Also, this subcategorisation is entirely irrelevant for establishing co-occurrence restrictions.

240 214

1,8 A syntax of phonology

4.2.3.3. The only critical factor is relative sonority, hence a lateral relation If we thus remove the arboreal glasses through which everybody contemplates syllable structure, the facts appear in a different light: with lateral contours. We have seen that the only information which is needed in order to discover syllable structure is the relative sonority of two adjacent consonants. Relative means lateral, for it is the specific relation of a consonant with its neighbour that is at stake. In CVCV, this lateral relation is encoded directly as Infrasegmental Government (see §§57s): sonorants can establish a lateral relation over obstruents if they are the second member of the cluster. Hence TR is a possible domain of consonantal interaction. By contrast, no lateral relation can be established among RT, TT and RR clusters. Therefore, the opposition is binary, lateral and non-arboreal: RT, TT and RR are "naked", that is unmarked: they occur without any special provision. TR on the other hand stands out because its members are bound by a lateral relation. All syllabic matters, then, follow from this initial contrast "lateral relation vs. no lateral relation", without any arboreal structure being needed (see the last part of §210). Since relative sonority of neighbouring segments is the basic observational fact, the classical arboreal approach also builds on this lateral relation. Only is its ultimate goal not to build a linguistic structure that encodes this primary information. The tacit aim is to arrive at something that looks like a (syntactic) tree. Therefore, the primary lateral information is transformed into arboreal structure. The critical word here is "transformed": it means that the arboreal structure, which is supposed to represent the phonological and cognitive reality, is artefactual: it is manufactured by the wish to see it in place. This is not to say, of course, that primary information is always the correct one and reflects most closely linguistic reality. That would be grossly misunderstanding theoretical endeavour and the generative enterprise. It means, however, that a theory which encodes the lateral facts directly without projecting them onto a second level deserves some attention. Before any transformation is operated, the primary facts could be explored for what they are: lateral. A good example of the direct lateral, against the indirect arboreal approach, is discussed in §446. What is the reaction in a situation where a particular vowel, schwa, is responsible for an effect on the preceding consonant (which is lost)? The lateral solution is to simply record this lateral relation: schwa, being weak by nature, cannot license the preceding conso-

CVCV: latéralisation of both structure and causality

241

nant, which therefore is lost. On the arboreal side, the consonant is said to be resyllabified into the Coda of the preceding syllable, although its regular place is in the Onset of schwa. Its loss is then interpreted as a consequence of its Coda status: Codas are weak. The former solution directly encodes the observation, while the latter takes the long way round via arboreal structure. In sum, what I want to say is that expressing syllable structure in lateral terms should be the null hypothesis. The stance taken by CVCV may thus appear somewhat less exotic than it does at first sight: classical Kahnian algorithmic models as well as Standard Government Phonology (which on top of that is hybrid and hence redundant) transform the primary lateral information into arboreal structure. CVCV encodes this information directly. Unless there are good reasons not to, phonological theory should express the lateral reality in terms of lateral relations and hence give syllable structure a lateral face. This is precisely what CVCV has set out to do. Non-lateral approaches must accept the burden of proof: explaining why the lateral solution falls short of theory or fact.

215

4.3. Bogus clusters again: transforming melodic contrasts among adjacent objects into vertical structure is a bad idea Bearing in mind the redundancy of arboreal structure within Government Phonology, as well as the primary character of lateral relations, let us now return to the challenge raised by bogus clusters (cf. §180). Various distributional situations suggest that in some instances, consonant clusters which do not qualify as branching Onsets may not be accommodated as CodaOnset sequences either. Therefore, it is contended, phonology must be able to express this three-way distinction in theoretical terms. Since Standard Government Phonology (and other theories) interpret the contrast between branching Onsets and Coda-Onset sequences in terms of contrasting arborescence, the third situation is given a representation that relies on a vertical distinction as well: bogus clusters belong to two independent Onsets, and hence their members are not subject to any co-occurrence restrictions. A bogus structure is assumed by default when a cluster does not qualify as either a branching Onset or an interlude. Table (171) recalls the three relevant representations.

242

1,8 A syntax of phonology

(171) Standard Government Phonology: three-way distinction of surface clusters a. branching Onset b. Coda-Onset

c.

Onset-Onset (bogus)

R Ν

0

n \

1

A X ι

X I

X I

1 X ι

Τ

R

V

V

1

1

1

1

0

Ν

Ν

1

I

I

I

1

1

1

X ι

X I

X I

X I

X

X

X ι

X I

R

Τ

V

V

\ \ 1

1

1

0

0

Ν

ι 1 t

1

Ν

1 1

1

V

The reason why [tl] clusters cannot be interpreted as Coda-Onset sequences is the sonority of their first members: the R under (171b) is a Rhymal Adjunct because it is a sonorant, and thus may be governed by the following obstruent. By contrast, the [t] under (171c) is an obstruent and as such creates a sonority slope with the following [1] that does not allow for the same syllabic interpretation. Finally, [tl] does not qualify as a branching Onset either since it does not occur word-initially. Therefore, both [t] and [1] must belong to two independent Onsets. As may be seen from this description, bogus clusters are just another instance of the conventional practice that transforms a lateral melodic difference into a vertical contrast. As was indicated earlier, CVCV in its present coat is unable to express a three-way distinction in the spirit of (171) since there is no vertical structure left at all. This is evident from table (172) below. (172)

CVCV: three-way distinction of surface clusters a. branching Onset b. Coda-Onset Gvt/ Lie Gvt

f V C I I V

Τ

V

C

V I I

/ka-koma/ —> /ma-ino/ —> /ka-eleka/ —>

147

matama kakoma miino keeleka

cheeks one who kills eyes one who cooks

The Latin case is probably the single most frequently quoted one in the literature on Compensatory Lengthening. The data illustrating Chilungu are taken from Bickmore (1995). The Old French situation is exposed in, among many others, Bourciez (1926:207ss) and Pope (1934:151). Gess (1998a,b) provides an very detailed review of the relevant data and literature.

Vocalic length

261

Under (181a,b), the diachronic loss of preconsonantal [s] provokes the lengthening of the preceding short vowel. The Old French case under (181b) is illustrated by Latin and Modern French forms. Preconsonantal [s] was lost in the course of Old French and provoked lengthening of the preceding vowel. The resulting length is still indicated in Modern French spelling by the circumflex accent. 148 In Chilungu as under (181c), the prefix-final vowel is elided before a root-initial vowel, which then lengthens. If Compensatory Lengthening is triggered by the non-realisation of a consonant as in Latin and Old French, the preceding vowel lengthens. Cases where short vowels lengthen that follow the unrealised consonant do not seem to be on record, and it is reasonable to assume that this is a universal property of Compensatory Lengthening. In Chilungu on the other hand, a hiatus is resolved by the elision of its first member with ensuing colonisation of the vacant position by the second member. The three cases shown under (181) thus illustrate Compensatory Lengthening in both directions: the short vowel spreads to its right in Latin and Old French (missing consonant), while it expands to its left in Chilungu (missing vowel). 149 However, it is interesting to note that the target of spreading is in internuclear communication with the following vowel in all cases. Since long vowels that participate in the paradigm of Closed Syllable Shortening can only be long if their complement is licensed by the following vowel, we may safely assume that the internuclear relation at hand is Licensing. The directional contrast is depicted under (182) below.

148

149

The last three forms bear a [t] in their modern versions that is absent from the Latin source. The [t] in question is of epenthetic origin and arose regularly through the contact of [s] and [r], which in turn was provoked by the loss of post-tonic vowels (bracketed in the examples shown). Bourciez (1926:256s) for example provides more general information regarding syncope-triggered epenthesis. The directionality of spreading in absence of a consonant seems to be universal. By contrast, Compensatory Lengthening which is consecutive to the elision of a vowel appears to occur both ways.

262

1,9 Lateral relations are head-final: length in phonology

(182) Compensatory Lengthening a. progressive spreading: absence of a consonant Lie

b. regressive spreading: absence of a vowel Lie

c v c v c v c v I k

I a

-

^

I

s

I n

u

C I m

I s

V C

C

C I C

V I V

b. Tonic Lengthening (Open Syllable Lengthening) Lie

C I C

V I V

[C

V]stress >

C I C

0 Latin *fideslia > fedeelia

Italian /fáto/ ->· [fáato]

V I V

Vocalic length 263 All other things being equal, the only difference is the origin of the Nucleus on which spreading operates: it is provided by stress in Tonic Lengthening systems, while it stems from the loss of a consonant in progressive Compensatory Lengthening. This superficial contrast is accompanied by a more fundamental one: there is no way to control how progressive lengthening of the compensatory type reacts on a following RT cluster. Since progressive Compensatory Lengthening supposes the loss of a consonant, the structure at stake in its original form would need to bear two Codas in a row, that is RRTV. Of course, double Coda sequences do not occur. Therefore, we can only suppose that the preceding vowel would refrain from lengthening in such a case: the freed Nucleus on which it could spread would not be licensed. On the other hand, progressive Compensatory Lengthening allows to observe that following TR clusters do not block the vocalic expansion. This is indicated by the behaviour of Old French under (181b) where Latin nostru comes out as nôtre. We are thus left with regressive Compensatory Lengthening of the Chilungu kind. As was indicated in the previous section, the orphan Nucleus that is subject to spreading has been liberated on a reaction against hiatus. I am not aware of other sources that could provide potential targets (such as stress). Also, I do not know whether regressive Compensatory Lengthening is sensitive to the right context of the alternating vowel. According to the analysis developed, it should not. The kind of vowel elision with ensuing lengthening of the following vowel that is illustrated by Chilungu under (181c) is a very common feature of black African languages, among which the Bantu group. 150 Unfortunately, most of these languages do not cluster at all, so that the influence of consonantal sequences on preceding vowels that are ready to spread seems to elude intelligence. In any event, I could not come by critical data in the aforementioned literature. The same can be said about hypothetical cases where progressive Compensatory Lengthening is triggered by the loss of a vowel, i.e. the mirror image of the Chilungu situation (181c). It is hard to quote any relevant evidence, and the literature quoted does not help. According to the analysis developed, progressive Compensatory Lengthening whereby V 2 of a CViV 2 CC sequence is lost should be inhibited in case the following CC cluster is a Coda-Onset sequence RT; by contrast, it should be able to go into effect if the cluster identifies as a branching Onset TR. Again, critical 150

See for instance Clements (1986) on LuGanda, or Creissels (1994) for a broad overview.

264

1,9 Lateral relations are head-final: length in phonology

evidence not only requires a language where the first part of a hiatus eats up the second member. Such a language must also allow for consonant clusters that follow the hiatus. For want of reliable evidence, I leave this an open question. Be that as it may, the analysis of regressive Compensatory Lengthening is as under (182), which I reproduce below for convenience. ( 184) regressive Compensatory Lengthening Lie

¿ I

C V I ^ m a

C

V I

C V I I i n o

The interesting property that sets Compensatory Lengthening apart from Closed Syllable Shortening and Tonic Lengthening is the directionality of spreading: while the latter processes are exclusively progressive, Compensatory Lengthening comes along in both progressive and regressive flavours. The following section centralises all information on stable long vowels and non-arbitrary alternations in vowel length that has been collected so far.

228

2.5. All long vowels obey the same requirement: their complement is licensed We are now in a position to compare the three types of non-arbitrary alternations in vowel length. Table (185) below is an augmented version of (180).

Vocalic length 265 (185)

typology of non-arbitrary alternations in vowel length Tonic Length- Closed Sylening lable Short(Op. Syll. Le.) ening yes a. conditioned by the right yes environment b. distribution of length in only short - closed syllables only short

Compensatory Lengthening presumably yes

only long

only long

c. long vowels are

head-initial

head-initial

d. spreading is

progressive

progressive

e. control over RT f. control over TR g· origin of the target of spreading h. alternating vowels are underlyingly i. process

yes yes stress

yes no lexicon

short

long

prediction: only short prediction: only long head-initial head-final progressive regressive no no loss of a segment short

lengthening

shortening

lengthening

- open syllables

Hence, it appears that the complement of long vowels is always licensed, whatever its status and origin. This is the single most important result. There are no alternating long vowels that spread on a target which is unlicensed. In spite of the apparent variety of processes that involve alternations of long and short vowels, it thus appears that vowel length in phonology obeys one single and simple mechanism, which is identified under (186). (186)

the complement of alternating long vowels must be licensed. Vocalic melody cannot spread onto unlicensed targets.

There do not appear to be cases on record where vowel length is conditioned by the left environment. The different processes that may be identified on the surface, known as Closed Syllable Shortening, Tonic Lengthening and Compensatory Lengthening, are but different illustrations of the general principle (186). Their diversity is due to different origins of the syllabic material that is targeted by spreading. Spreading is an active phonological process in the two latter cases that is triggered by an independent phonological event, i.e. stress placement or the loss of a segment.

266

1,9 Lateral relations are head-final: length in phonology

On the other hand, the association of the melody to both the head and the complement Nucleus is present at the lexical level in Closed Syllable Shortening systems. A morphological event may put the structure in a position where the complement fails to be licensed, in which case its association with the head breaks down.

229

2.6. LOWER spreading may be in both directions, but UPPER Licensing is only regressive The initial question that was raised earlier in this chapter can now be answered. Recall that CVCV makes a prediction to the effect that all governing and licensing domains are head-final. It now appears that progressive spreading of vocalic melodies in (tonic or compensatory) lengthening processes does not challenge this generalisation at all. Again, the distinction HIGH vs. LOW comes in handy here: Government and Licensing belong to the UPPER area. It was mentioned earlier (§§215s) that only this part of the grammar is subjected to strict regressive directionality. On the other hand, spreading of melody takes place BELOW the skeleton. Hence, the LOWER area is active in both directions, right-toleft and lefìt-to-right, a property which has already been identified (palatalisations, voice assimilations etc.). By contrast, whatever the directionality of the melodic activity, the segmental result is conditioned by a restriction that is of exclusively HIGH, hence regressive character: the condition under (186).

In this sense, vowel length is parallel to lenition and fortition: the control over segmental alternations of this kind is only positional. It makes no reference to adjacency or melodic properties of either the agent or the patient. In short, this is what is commonly referred to as syllable-related processes in phonology. This notion thus receives the following expression in CVCV. (187)

syllable-related processes are segmental effects which are a. independent of adjacency b. independent of the melodic properties of the agent and the patient c. conditioned by the actors in the UPER area, i.e. Government and Licensing

In the following section, we turn to the macro-distinction between alternating and non-alternating vowels.

Vocalic length 267 230

2.7. Alternating long vowels are head-initial, their non-alternating peers are head-final

231

2.7.1. Only CVCV can build on the contrast head-initial vs. head-final This leads us to the obvious question regarding systems where long vowels do not shorten in whatever syllabic or other circumstances. Two such cases, German and Somali, have been presented in §221. It looks like long vowels in these languages are immune against external influence of any kind: they mock at the number and quality of consonants that occur to their right. This is certainly unexpected if it is true that long vowels are made of a head and a complement whereby the melody of the former spreads onto the latter, and the target needs to be licensed. Regressive Compensatory Lengthening is the key to this question. It shows how the existence of inalterable long vowels may be reconciled with (186). In case a vowel expands onto the position that was abandoned by a peer on its lefthand side, the resulting long vowel is head-final as under (184). As was mentioned in the discussion of regressive Compensatory Lengthening, long vowels of this kind are "self-licensors" in the sense that the head licenses its own complement. It does not need any external help to do so since it is phonetically expressed and therefore possesses full lateral actorship. Hence, the parameter that somehow must distinguish alternating and non-alternating long vowels falls into place when the contrast headinitial vs. head-final is fertilised. Table (188) shows this difference. (188) a. alternating long vowels are head-initial Lie

C V C V I I > C V

b. non-alternating long vowels are head-final Lie

I C

C V C V ^ I V

Notice that this kind of contrast is predicted by CVCV: in contrast to other theories, long vowels involve two independent Nuclei that are related to one single chunk of melody. This stance does not imply or exclude any particular headship a priori. Therefore, it is expected that some long vowels will be head-initial, while others turn out to be head-final. Note that this is not the case in Standard Government Phonology where long vowels are instantiations of branching Nuclei and as such necessarily head-initial. On

268

1,9 Lateral relations are head-final: length in phonology

these assumptions, headship and contrasting directionality of spreading cannot be used in order to express the parameter at hand.

232

2.7.2. Head-final vs. head-initial also expresses the parameter on the existence of super-heavy Rhymes The necessity for any theory to be able to assign two different underlying identities to alternating and stable long vowels is demonstrated by languages where both varieties cohabitate. A case in point is Czech. Recall that this language illustrates Closed Syllable Shortening in a certain morphologically defined paradigm, i.e. feminine nouns in -a and neuter nouns in -o (see appendix 2 §621). Elsewhere, however, long vowels occur freely without any control of the following environment. Consider the data under (189) below.151 (189)

cohabitation of alternating and stable long vowels in Czech C CV C _# C CCV Vkra(a)vkraava kraf kraf-ka vs. Vflaamflaam-ονεε flaam flaam-ka flaam-skii flaam-Jc-ina

There must be a reason why the long vowel in flámka, which belongs to the same paradigm as brava, refuses to shorten in the appropriate environment. 152 In CVCV, this behaviour is attributed to the difference in headship: flámka bears an autonomous head-final long vowel, while kráva has a head-initial long vowel. The contrast shown under (188) is interesting in yet another respect. As a matter of fact, it expresses the difference between languages that allow for super-heavy Rhymes and those which do not tolerate this kind of structure. Recall the discussion in §190 where Harris' (1994a) way of encoding 151

152

Glosses of table (189): line one kráva "cowNOMsg", krav "id. GENpl", kravka "id. diminutive NOMsg"; line two flámové "flame (person) NOMpl", flám "id. NOMsg", flámka "flamish woman"; line three flámsky "flamish (adj.)"; line four flámStina "flamish (language)". The fact that flámka has probably entered the language more recently than kráva may be relevant here. But whatever the origin of the contrast at hand, it must somehow be represented in the lexicon.

Vocalic length

269

this parameter was exposed. On his account, Closed Rhyme Shortness is a consequence of the fact that the Rhymal Adjunct fails to be licensed twice in languages that require double Licensing. In languages where the Rhymal Complement needs to be licensed only once, however, nothing withstands the existence of super-heavy Rhymes. If vocalic length is controlled along the lines that have been exposed in this section, head-final long vowels do not object to the existence of a governed empty Nucleus to their right. This means that they may be followed by an RT cluster. By contrast, head-initial long vowels cannot precede a governed empty Nucleus: their complement would fail to be licensed. Both situations are depicted under (190) below. (190)

languages prohibiting vs. allowing for the existence of super-heavy Rhymes a. head-initial long vowels cannot be b. head-final long vowels can be followed by an RT cluster: followed by an RT cluster: supersuper-heavy Rhymes are heavy Rhymes are ill-formed well-formed Lie Gvt Lie Gvt

f

c v c v c v c v I

I

C

V

\

>

I R

"[CVVRTV]

I

C

I T V

I

f

V

^

C

V

C

V

C

V

I I V

R

τ

I I V

[CVVRTV]

The familiar notion of super-heavy Rhyme refers to a rhymal sequence that contains a long vowel and a consonant. More descriptively speaking, this amounts to a long vowel that is followed by a consonant cluster of falling sonority. The head-final structure under (190b) is the translation of a super-heavy Rhyme in CVCV: a long vowel is followed by an RT cluster. Put another way, a long vowel exists on the lefthand side of a governed empty Nucleus. The vowel to the left of the RT cluster may only be long because it is a self-licensor and therefore does not depend on external support from the following Nucleus, which, being empty and governed, cannot act as a licensor. On the other hand, head-initial long vowels such as under (190a) cannot exist before RT clusters because the only candidate for the Licensing of their complement is the following governed empty Nucleus, which is laterally disabled.

270

1,9 Lateral relations are head-final: length in phonology

Hence, the parameter according to which languages do or do not tolerate super-heavy Rhymes translates into CVCV as under (191). (191) parameter on the existence of super-heavy Rhymes a. long vowels in languages that allow for super-heavy Rhymes are headfinal. b. long vowels in languages that do not tolerate super-heavy Rhymes are head-initial. In other words, long vowels that undergo Closed Syllable Shortening are always head-initial. As was demonstrated under (189) for Czech, alternating and non-alternating long vowels may co-exist within the same language. This simply means that there are languages where some long vowels are head-initial, while others are head-final. This parameter thus is (or at least may be) decided on a case-by-case basis in the lexicon. Another benefit of this view on long vowels is the following. Recall from §191 that Harris' (1994a) system encounters two problems which are related to super-heavy Rhymes. That is, languages which allow for superheavy Rhymes, i.e. where the Rhymal Adjunct must be licensed only once, should tolerate word-final Rhymal Adjuncts and rhymal sequences that are made of a short vowel plus two consonants. In the former case, the wordfinal consonant can be interpreted as a Rhymal Adjunct since it will be licensed by its Nucleus (see (156)). The latter structure should be wellformed because both putative Rhymal Adjuncts will be licensed once: the first one by its Nucleus, the second one by the following Onset (see (157)). These problems disappear when the existence of super-heavy Rhymes is controlled by the lexical parameter on headedness under (191). Harris1 (1994a) proposal was the first attempt to lateralise the causality of Closed Syllable Shortening (Shortness). Recall that Standard Government Phonology could only account for this process by virtue of a vertical analysis (the Binary Theorem that restricts the number of rhymal x-slots or Prosodie Government, cf. §§176,179). The research programme of Government Phonology is to build a syntax of phonology. This means that vertical structure and causality should be replaced by lateral analyses (see chapter 1,8 §165). Regarding the issue of super-heavy Rhymes, CVCV now provides a solution along these lines that does not suffer from the problems that are germane to Harris1 (1994a) lateral account.

Vocalic length 271 2.8. Progressive Compensatory Lengthening: an argument in favour of CVCV Progressive Compensatory Lengthening as illustrated under (181a,b) grounds an argument in favour of CVCV that relies on the existence of an empty Nucleus which is absent in conventional analyses. In a case like lat *kasnus > kaanus, the short [a] expands on the position that is liberated by the loss of the following consonant. Table (192) shows that under traditional syllabic analysis, the skeletal slot which has hosted the consonant is called to receive vocalic material. (192)

relabelling of skeletal slots in Compensatory Lengthening a. the skeletal slot hosts [s] b. the skeletal slot hosts [a] R R κ I 1 0 0 Ν 0 Ν 0 Ν 0 Ν 0 I I 1 \ 1 I I 1 I 1 I X Χ Χ X χ X X —> X χ χ χ χ χ I I ι I I ι I1 11 II ~7' 1 1 I 1 1 I k a s η u k η U s s a s

Ν I χ

The Coda (or Rhymal Adjunct) under (192a) dominates the consonant [s]. In the course of the derivation (which is of diachronic or synchronic type, depending on the kind of Compensatory Lengthening), this Coda is reinterpreted as the second member of a branching Nucleus since it has been colonised by the melody of the formerly short vowel. This is odd at any standards, both in conventional syllabic analyses 153 and in Standard Government Phonology. In the latter model, resyllabification is ruled out altogether, and hence there is no way for a skeletal slot which is "born" as a Coda to end up as a Nucleus. In contrast to other theories, CVCV holds that the cluster [sn] in *kasnus encloses an empty Nucleus, i.e. /kasenuso/. Accordingly, the short [a] does not expand on the position left behind by the [s]. Rather, it spreads on the empty Nucleus which has become available for melodic identification through the loss of [s] since association lines do not cross anymore. Reconsider the representation of progressive Compensatory Lengthening in CVCV as under (193).

153

Where the syllabification algorithm would have to reapply. See Bickmore (1995:130ss) on this issue.

272

1,9 Lateral relations are head-final: length in phonology

(193) CVCV: nothing is relabelled or resyllabified Lie

C I k

V I a

Ψ V

C > s

c I η

V I u

C I s

V

Hence, CVCV accounts for progressive Compensatory Lengthening without having to transform consonantal into vocalic x-slots. This advantage is even more salient in systems where progressive Compensatory Lengthening is triggered by the loss of a consonant which disappears across the board without contextual condition. Indo-European Laryngeals are a case in point. Saussure's (1879) finding, which has founded the Laryngeal Theory, describes the following diachronic event. The Indo-European mother tongue possessed a class of consonants that are called Laryngeals. Their phonetic identity is commonly assumed to be post-velar, i.e. guttural, hence the name. Due to a spontaneous sound change, Laryngeals were lost in all individual Indo-European dialects (except Anatolian). If they occurred word-finally or in preconsonantal position {#,C}, the preceding vowel lengthened. Accordingly, (late) Indo-European and its descendants have no Laryngeals anymore. However, the existence of these consonants can be recovered through the distribution of long vowels (and partly by Anatolian script, which has been discovered after Saussure died). 1 4 Some evidence regarding Laryngeals and their fate in various contexts is provided under (194) below. Note that Laryngeals come along in three (according to some versions of the theory, four) flavours depending on the "colouring" that they have produced on the preceding vowel. The conventional symbol for Laryngeals is "H", and the sub-categorisation mentioned is indicated in subscript ( H b H2, H 3 ).

154

The Laryngeal Theory is exposed for example in Lindeman (1987). Also see classical handbooks such as Mayrhofer (1986), Szemerényi (1990), Gamkrelidze & Ivanov (1995).

Vocalic length

( 194) reflexes of Laryngeals in various environments IE root V_C C_V *VpeH3 athematic inf skr pâ-ti •peHj-ti redupl. thematic skr pi-b-a-ti perf *pi-pH3-e-ti WpleH, *VpleH,-isto-

V_V

273

gloss to drink to drink

grpleistos most

The Indo-European root *VpeH3 ends in a Laryngeal. In Sanskrit, the Laryngeal comes to stand in preconsonantal position in the infinitive because this verb happens to be athematic and thus provokes the contact of the last root-consonant with the infinitive morpheme -ti. By contrast, the perfective is achieved through reduplication, which recurs to thematic formation (in addition, the root appears in zero grade). Therefore, the Laryngeal is located before a vowel. Sanskrit has conserved a trace of the Laryngeal in the former grammatical form where the preceding vowel has been lengthened. In the latter case, however, the Laryngeal has been lost without any trace. Finally, the Greek example shows that Laryngeals in intervocalic position also disappear without leaving any trace. The point is that models which allow for relabelling of skeletal slots or resyllabification predict that Compensatory Lengthening should always occur when a skeletal slot becomes available through the loss of a consonant. If consonants disappear only in preconsonantal position as in Latin and Old French under (181a,b), orphan skeletal slots that are ready for colonisation exist only in this specific context. Accordingly, vowels are expected to lengthen only if they precede such a vacant skeletal slot, which is then resyllabified as a Nucleus. In the Indo-European case, however, the consonant in question disappears everywhere without any contextual condition. It may thus be expected that all vowels which precede an orphan skeletal slot expand. This, however, is not what happens: Compensatory Lengthening is observed only if the deleted Laryngeal occurred in the Coda context {#,C}, to the explicit exclusion of intervocalic environments. Hence, IE ...VH X C, ...VH X # > VVC, VV#, as opposed to IE V,H X V 2 > ViV 2 . Table (195) below depicts this situation.

274

1,9 Lateral relations are head-final: length in phonology

(195) a. Laryngeal in a Coda R Ο

Ν

C

Ρ

e

Η;

-

Ο

R >

Ν

t

b. Laryngeal in V

Ν C I ^ e H3

-

Ο I

Ν I t i

V

N O N o

O I ρ

Ha

>

N O N o

e

H2

e

If relabelling and/ or ^syllabification are permitted in order to turn a Coda into a Nucleus, there is no reason why an Onset as under (195b) should not be able to receive vocalic melody and thereby become a Nucleus. Contrasting with this analysis, the absence of Compensatory Lengthening on the orphan position of intervocalic consonants is predicted by CVCV. This follows from the fact that vowels do not spread on the position that is left behind by the deleted consonant. Rather, the following empty Nucleus is colonised. Consider the representations under (196) that evidence the different result of the deletion in Codas and intervocalic position. (196) a. Laryngeal in a Coda C V C V - C V > C V C V - C V I I I I I I I ^ I I ρ

e H3

t i

b. Laryngeal in V V - Ν Ο Ν > ο Η2 e

ρ

e Ha

t i

- Ν Ο Ν ο Hi e

Under (196a), an empty Nucleus becomes available for segmental identification because the Laryngeal has been deleted. By contrast under (196b), there is no empty Nucleus that has been liberated by the loss of the Laryngeal since only consonant clusters enclose "additional" empty Nuclei. Therefore, an appropriate target for spreading is available under (196a), but not under (196b).

Consonantal length

275

234

3. Consonantal length

235

3.1. The representation of geminates in conventional models and in CVCV The analysis of vocalic length that has been exposed on the foregoing pages also suits the representation of consonantal length. It is cross-linguistically notorious that geminates occur only in intervocalic position. Degemination is generally observed in pre- or post-consonantal environments (e.g. McCarthy 1986, one case of initial non-degemination is discussed in §355).155 In CVCV, this condition on the existence of geminates is a straightforward consequence of the existence of ("additional") empty Nuclei, which are all subject to the ECP. In order to demonstrate that, let us first compare the treatment of geminates in CVCV and elsewhere. In conventional representations, geminates are interpreted as a single chunk of melody that is associated to a Coda and its following Onset. By contrast, CVCV holds that the two parts of a geminate are separated by an empty Nucleus. This is nothing but the straightforward translation of "...associated to a Coda and its following Onset" since, recall, a Coda is a consonant that occurs before an empty Nucleus (cf. §150). Hence, the underlying identity of an intervocalic geminate [VCXCXV] is /VCx0CxV/. Its first part is associated to an Onset which occurs before an empty Nucleus, while its second part belongs to an Onset that is followed by a Nucleus with phonetic content. However, the attentive reader will have noticed that the definition of the Coda in CVCV requires an additional property: only those consonants that occur before a governed empty Nucleus stand in a Coda (as opposed to those whose Nucleus is empty but ungoverned, which are the first member of a branching Onset). Therefore, the representation of a geminate in CVCV implies a single chunk of melody that is associated to two Onsets, of which the first occurs before a governed empty Nucleus, while the second is followed by a Nucleus with phonetic content. Table

155

This generalisation is sometimes contradicted by heteromorphemic structures. That is, a geminate may follow or precede another consonant in case this consonant belongs to a different morpheme, or if a morpheme boundary separates the two members of the geminate. Berber (e.g. Dell & Elmedlaoui 1985 Elmedlaoui 1993) is a notorious case in point. The responsibility of the morpheme boundary is obvious; hence the phonological generalisation as such (i.e. in absence of all non-phonological forces) remains untouched.

276

1,9 Lateral relations are head-final: length in phonology

(197) below opposes the representation of geminates in traditional syllabic environments and in CVCV.156 ( 197) representation of geminates a. in conventional syllabic envi ronments

b. in CVCV Gvt

R Ο I c

236

Ν I ν

C ^

Ο c

Ν I ν

3.2. Standard Government Phonology: why geminates cannot be preceded or followed by consonants In Standard Government Phonology, the ban on geminates that are adjacent to another consonant is explained by Government Licensing and the rhymal status of Codas. The head of branching Onsets and Coda-Onset sequences must be licensed in order to be able to govern its complement. If geminates are represented as under (197a), they are subject to this condition. Hence, the head of the geminate, which is located in the Onset, needs to be government-licensed by its Nucleus. Therefore, this Nucleus must not be empty, which means that surface sequences where a consonant follows a geminate *[VCxCxCyV] are ruled out: they would necessarily imply the existence of an empty Nucleus after the geminate /VCxCx0CyV/, which therefore would fail to be government-licensed. On the other hand, sequences where a consonant precedes a geminate are ill-formed because such a structure can only be achieved if the Nucleus that precedes the geminate is empty. This putative configuration is shown under (198).

156

Under (197b), the headship of the geminate is not indicated on purpose. This issue will be addressed as the discussion unfolds.

Consonantal length 277 (198)

Standard Government Phonology: why geminates cannot be preceded by consonants

Ο

Ν

*[CyCXCxV]

The only way to silence the empty Nucleus is to make it the target of Proper Government. This, however, is not a possible option since the heart of vowel-zero alternations as conceived of in Standard Government Phonology would be violated: "Proper Government is blocked by intervening governing domains" (cf. §21). Since the geminate instantiates a governing domain, the structure under (198) is ill-formed.

237

3.3. CVCV: head-initial and head-final geminates What does the situation in CVCV look like? Recall that one major difference between Standard Government Phonology and CVCV is the status of Coda-Onset sequences. The former model conceives of a lateral relation between the two consonants of both branching Onsets (Constituent Government) and Coda-Onset sequences (Interconstituent Government). By contrast, CVCV holds that the opposition between both types of clusters is privative: there is a lateral relation between both members of a branching Onset (Infrasegmental Government), while Coda-Onset sequences do not represent any lateral domain (see § 180). A direct consequence of this stance is that the head of a geminate does not need to be government-licensed: the only domain of consonantal interaction are branching Onsets. This not withstanding, geminates are headed. Hence, there should be both head-initial and head-final geminates in nature, just as much as there are head-initial and head-final long vowels. By contrast, geminates can only be head-final in Standard Government Phonology since they instantiate a Coda-Onset sequence.

278

1,9 Lateral relations are head-final: length in phonology

Table (199) shows how CVCV represents head-initial and head-final geminates in intervocalic position. ( 199) geminates in CVCV a. head-final Gvt

C I C

V I V

C ^

f V

C I C

b. head-initial Gvt

V I V

C I C

V I V

C I C

f V

C ^

V I V

In case geminates are recorded in the lexicon and hence do not result from a phonological process, it is difficult to determine their headship. The same holds true for "morphological" geminates that often arise when two identical consonants occur on both sides of a morphological boundary. On the other hand, when geminates arise through spreading on a position that has become available because of a phonological process, headedness may be controlled. For instance, the Italian gemination known as Radoppiamento Sintattico describes a process whereby the first consonant of a word geminates if the preceding word ends in a stressed vowel. As was the case in Italian Tonic Lengthening, the target of this spreading identifies as a syllabic unit that is provided by stress. Hence, the initial [p] of the second word in paltò pulito "clean coat" where the accent on the "o" indicates stress is subject to gemination. In this case, it is reasonable to assume that we are in presence of a head-final structure. By contrast, other circumstances suppose head-initial geminates. For instance, the original morpheme that represents the infinitive in prehistorical Latin is *-se, which surfaces in most Latin verbs as -re because of rhotacism s > (z >) r. However, rhotacism only transforms [s] into [r] in intervocalic position, e.g. *am-a-se > am-a-re "to love". Since nearly all verbs are thematic, i.e. possess a thematic vowel that appears between the root and the suffix (-a- in amaré), the regular representative of *-se in Latin is -re. In the small class of athematic verbs, however, the absence of a thematic vowel blocks rhotacism. Hence, athematic verbs such as *vel-se, *fer-se "to want, to carry" should appear as such in Latin. Instead, velie and ferre are recorded. 157 The gemination which has occurred is progressive and therefore supposes a head-initial structure. 157

Ernout (1989:171) and Niedermann (1991:140ss), among many others, expose the Latin facts at greater length.

Consonantal

238

length

279

3.4. CVCV: why geminates cannot be preceded or followed by consonants Given the representations under (199), for which reason are geminates unable to occur in the vicinity of other consonants? The answer is simple and involves no special reference to geminates at all: the ECP of each and every empty Nucleus must be satisfied. One difference between Standard Government Phonology and CVCV is the fact that geminates are free from lateral conditions like Government Licensing in CVCV: a consonantal melody is autonomous and may spread on a preceding or following Onset without restriction. Another difference, of course, is the very existence of an empty Nucleus in the middle of the geminate. It is this Nucleus that carries the functional burden of explaining why geminates cannot be preceded or followed by another consonant. In other words, the existence of geminates is not subjected to a lateral relation that concerns its own body. Rather, it depends on the status of the enclosed empty Nucleus. Table (200) shows the kind of structure that would be produced by the presence of a consonant before or after a geminate. 158 (200)

CVCV: why geminates cannot be preceded or followed by consonants a. preceding consonant

b. following consonant GvJ.

GvJ, C

V

I C

I V

C

I C

WM

V

C

V

Gvt

Gvt C

V

I C

I I V

C V

I C

I V

C V ^

C J C

V

C I C

V I V

Both hypothetical structures are ill-formed for the same reason: there are two empty Nuclei in a row, the leftmost of which will not satisfy the ECP. This way of accounting for the obligatory presence of expressed Nuclei on both sides of geminates is only available if geminates enclose an empty Nucleus. It has the advantage of unifying the reasons why geminates may not exist in presence of a preceding or a following consonant. Recall that two independent reasons have to be invoked in Standard Government Phonology: geminates plus a following consonant are ruled out because the head of the geminate cannot be government-licensed. On the other hand,

158

Only head-final geminates are represented, the situation is identical for their head-initial peers.

280

1,9 Lateral relations are head-final: length in

phonology

there can be no consonant preceding geminates because this would imply that Proper Government applies over a governing domain. On the analysis presented here, there is only one reason for the agrammaticality of the structures under (200), that is the violation of the ECP.

239

3.5. Geminates, RT clusters and homorganic NC clusters are one But CVCV unifies the grammar even further: geminates and the distributional restrictions that they are subjected to are not consequences of any special provision made for geminates. CVCV treats geminates as regular RT clusters, i.e. as two independent Onsets that enclose a governed empty Nucleus. The only difference with respect to true RT sequences is of melodic, i.e. LOWER nature: RT clusters are made of two independent chunks of melody, whereas geminates possess only one melody that is shared by two syllabic positions. An intermediate case is represented by what is sometimes called a partial (or nasal) geminate, i.e. homorganic nasalobstruent clusters (e.g. Harris 1994a:69). These are ordinary geminates whose first member has an additional nasal identity.159 Table (201) provides illustration of the three structures at hand. (201)

three objects, one structure in CVCV: interludes, geminates and partial geminates. All clusters enclose an empty Nucleus that requires Government. a. interlude (RT seb. geminate c. partial geminate, quence) e.g. homorganic nasal-obstruent cluster Gvt Gvt Gvt

i

I

c v c v c v i m C V R

i

l

T V

i I

i I

c v c v c v m C V

ι C V

c v c v c v ι

ι

r - ^ j

C V N

C V

In other words, RT clusters, true and partial geminates obey identical phonotactic restrictions and represent the same object at the UPER level. The difference between the three types of clusters is a consequence of the organisation at the melodic (i.e. LOWER) level.

159

Homorganic N C clusters are further discussed in chapter 11,15 (§591).

ι

Consonantal length

281

Finally, the account of geminates in CVCV also allows for a rather unified theory of length in phonology: vocalic as well as consonantal length is conditioned by a lateral internuclear relation between Nuclei. In the former case, the lateral relation at hand is Licensing, and it concerns the second part of the body of the long object. Geminates on the other hand are controlled by Government, which applies to the empty Nucleus that separates both parts of their body.

Chapter 10 240 Syllabic and trapped consonants in CVCV

241

1. Setting the scene: phonological hermaphrodites This chapter sets out to inquire on the identity of a basic phonological object: syllabic consonants. Its close cognate, trapped consonants, will be considered in parallel. But let us first have a look at syllabic consonants because they are a great deal more common and have been studied for a long time. Syllabic consonants are objects of wonder: they are phonological hermaphrodites. Neogrammarians used to call them "consonants in vocalic function" (Saussure's 1879 laryngeal theory is entirely based on this insight), and this is probably as close as one can get to reality. That is, syllabic consonants are consonants physiologically speaking - yet they behave as if they were vowels. There is no way of mixing these roles: as a sound, their body is only consonantal - not a bit of vocalicity is in sight. As a phonological object, however, they are only vocalic - showing no consonantal behaviour at all. In other words, they are males physiologically, but whose behaviour is exclusively female. Except when they are followed by another female (i.e. a vowel), in which case their behaviour patterns with their physiological identity. In fact they are the male counterpart of glides, which represent the opposite distribution of physiological properties and phonological behaviour: glides are female hermaphrodites. They belong to the female vocalic world, but endorse male consonantal function. In this chapter, I develop a specific version of the well-known branching analysis for all hermaphrodites. That is, physiological properties determine the syllabic home of all individuals, while spreading defines their function: glides are vowels that have spread onto a consonantal position, while syllabic consonants are consonants which have colonised a vocalic slot.

284 242

1,10 Syllabic and trapped consonants in CVCV

2. Syllabic and trapped consonants tell us about each other - we are well advised to listen carefully Consonants in syllabic function do not usually stand in the focal spot of phonological theory. They are typically treated as a step-child whose strange properties are notorious and embarrassing. Therefore, the common way to address them is to declare that their identity follows from their function: syllabic consonants must sit in a Nucleus because they function as a syllabic peak. This simplistic view not only confuses function and representation, it also muddies autosegmental water: since the advent of non-linear structure, the consonantal or vocalic identity of a segment is not recorded in its idiosyncratic makeup anymore. A given set of phonological primes does not bear any inherent information that defines whether it will be realised as a consonant or as a vowel. Rather, this contrast is decided by the syllabic constituent in which it is interpreted. For example, the Element I (or its equivalent bundle if binary features are used) will come out as [i] if it pertains to a Nucleus, while its pronunciation will be consonantal [j] in case it comes to stand in an Onset (or in a Coda). If this indirect definition of vowel- and consonanthood is correct, there is absolutely no way for syllabic consonants to exclusively belong to Nuclei: did they, a vowel would be heard. There are also theory-internal reasons which lead to doubt that syllabic consonants have an exclusively nuclear identity. That is, the very frequent alternations between syllabic and non-syllabic realisations of the same consonant (e.g. English bottle [boti] vs. bottling [bDtlirj]) would imply continuous resyllabification if syllabicity were equated with nuclearhood. This is why the study of syllabic consonants has gained growing interest in recent work within Government Phonology (e.g. Szigetvári 1999:117ss, Rowicka 1999a:261ss, 2003, Rennison 1999b:333ss, Afuta 2002, Blaho 2001,2002,2004, Kijak 2003, Toft forth). Independently of this fact, the survey of the representation of basic phonological objects that has been undertaken in this part of the book would be incomplete without a statement regarding syllabic consonants. The way in which syllabic consonants are approached here bears a number of peculiarities. For one thing, a catalogue of the specific synchronic and diachronic behaviour that syllabic consonants constantly display in Slavic and Germanic is established. This aims at giving as much, as

Syllabic and trapped consonants tell us about each other - let's listen

285

varied and as detailed flesh as possible to the common-sense statement "syllabic consonants behave like vowels". But the most important piece of evidence comes from the comparison with a closely related relative of syllabic consonants (both genetically and phonologically): so-called trapped consonants. T h e s e occur most 160 prominently in Polish, and have been extensively discussed in the literature (foremost in the work by Jerzy Rubach) under the banner of extrasyllabicity. O n Rubach's analysis (e.g. Rubach & Booij 1990a, R u b a c h 1997a), the [r] in Polish words such as rdza, trwac, Piotr "rust, to last, Peter" is extrasyllabic. However, word-internal extrasyllabicity is problematic: it is c o m m o n l y assumed that extra-prosodicity, extra-metrical ity, extrasyllabicity and the like are exclusive properties of domain-edges (cf. the socalled Peripherality Condition discussed in §357). The Polish data are perfectly clear, but their interpretation may be subject to discussion. The overall picture can be accurately described by a very simple statement: all Polish candidates for word-internal extrasyllabicity ought to be syllabic, but are not. They should be syllabic because they are trapped between t w o consonants, and they enjoy a syllabic status in neighbouring languages in the same words. They are not syllabic for some reason, to be discovered. It will turn out, moreover, that to all extents and purposes, trapped consonants show the exact opposite behaviour with respect to their syllabic relatives: every time the latter act as if they were a vowel, the former show regular consonantal behaviour. Finally, the very existence of trapped consonants as an independent phonological category that is found in nature needs to be established before any further discussion can unfold. In the face of the above characterisation "trapped consonants ought to be syllabic, but they are not", some authors simply declare them syllabic and carry on as if English or Serbo-Croatian syllabic consonants were at stake (e.g. Rowicka 1999a:263ss, 2003). The non-recognition of the specific characteristics of trapped consonants and their subsequent classification as a mere sub-variety of syllabic consonants is dramatic theoretically speaking, but foremost in descriptive work where

160

But also in Czech (§§245,293) and Romansch (§275). Another candidate is Georgian (Kartvelian): I strongly suspect the massive clusters that this language is famous for to be created to a large extent by trapped consonants. For example, what is usually called "syllabic" consonants in this language is transparent to voicing (i.e. the voice value of the adjacent consonants must agree, a typical feature betraying trapped consonants, see §269). Relevant material is discussed for example in Butskhrikidze (2002) and Ritter (ms).

286

1,10 Syllabic and trapped consonants in CVCV

the reader is left guessing whether the object described is not, perhaps, a trapped, rather than a syllabic consonant as the description suggests. Given this general picture, the ambition of this chapter is threefold. For one thing, I wish to establish trapped consonants as an autonomous player in the phonological scene, i.e. one that is on a par with, and thus distinct from, other fundamental categories such as syllabic consonants, vowels and non-syllabic consonants. Second, I intend to cross the information that is conceded by syllabic and trapped consonants so that a consistent picture emerges. Any analysis of these peculiar consonants must come up with representations that somehow encode their absolute antagonicity. Finally, this endeavour should clarify the alleged extrasyllabic status of trapped consonants in Polish. In short, the analysis developed is original because it confronts both types of consonants: the behaviour of one will tell us something about the identity of the other. Building a theory of syllabic consonants without looking at their trapped mates must lead to partial and inaccurate results. And so does the isolated analysis of trapped consonants. There are two closely related, yet different and indeed even antagonistic cluster-building consonants in nature. Only a contrastive analysis may establish all relevant pieces of the puzzle. I first attempt at identifying these pieces on strictly empirical grounds. That is, I work out a check-list of properties that characterise both syllabic and trapped consonants. In a second step, I try to join the pieces of the puzzle so to create a general picture that makes sense. The following section begins with a descriptive survey of how syllabic (Czech) and trapped (Polish) consonants behave.

243

3. The synchronic situation of trapped consonants: trapped (Polish) vs. syllabic (Czech)

244

3.1. Trapped consonants in Polish: lexically trapped or trapped by a vowelzero alternation Synchronically speaking, two patterns of trapped consonants must be distinguished in Polish: those that are lexically trapped and those that end up trapped because of a vowel-zero alternation. Table (202) illustrates the former situation for word-internal and word-final contexts. 161 Relevant Polish 161

Word-initial trapped consonants are discussed at length in §375 (Polish) and Vol.2,II.6.5 (cross-Slavic). They do not bear on the present demonstration.

The synchronic situation of trapped consonants: trapped vs. syllabic

287

words and their Czech cognates are lined up in order to show that the consonants at stake occur in identical structures in two neighbouring languages and yet are syllabic only in Czech. Among many others, Biedrzycki (1978), Rowicka (1999a:231ss,2001) and Pawelec (1989) offer more data and further discussion. 1 6 2 (202)

lexically trapped consonants in Polish a. word-internally Czech Common Polish Slavic CrC travati trvat trwac CrzC dvbri drzwi dvefe grbmëti grzmiec hïmët brbnëti brzmieé brnët chrbbbtt grzbiet hFbet trbstina trzcina trstina klnç C1C klbnklnout plbvplwocina arch plvat > plivat b. word-finally Common Polish Czech Slavic Cr bebn. bóbr bobr vëtrb wiatr vitr pbpbrb Crz pieprz pep? vbnjçtrb wewn^trz vnitf

gloss (Polish) to last door to thunder to sound back reed (plant) I curse sputum

gloss (Czech) to last door to thunder to tickle back reed (plant) to curse to spit

gloss (Polish) beaver wind pepper inside

CI

thought

gloss (Czech) beaver wind pepper inner, inside sense

myslb

mysl

mysl

Cases where Polish trapped consonants arise through a vowel-zero alternation are identified under (203). The list aims at exhaustivity.

162

There are no syllabic consonants in Polish. In their place, the trapped consonants under focus appear. Rubach (1974,1977:68ss) reports on the existence of syllabic consonants in fast speech, but this does not bear on the present discussion. All examples hereafter are spelt. Polish and Czech spelling conventions are explained at the outset of this book (§7).

288

1,10 Syllabic and trapped consonants in CVCV

(203) Polish trapped consonants that are created by a vowel-zero alternation Polish Common gloss GENsg NOMsg Slavic blood krwi krew Cr C krbvb eyebrow brwi bnvb brew baptism chrztu chrzest CrzC krbstb sex pici pteó C1C plttb tear lez GENpl slbza Iza < slza czosnku garlic czosnek CnC Cesm>kb primroses pierwiosnek pierwiosnka piosnek GENpl song pë-snb piosnka piosenka The following section compares the behaviour of trapped and syllabic consonants in poetry and with respect to stress.

245

3.2. Syllabic, but not trapped consonants constitute syllabic peaks and can bear stress Consonants are called syllabic when they assume a vocalic function: for instance, they bear stress and count in poetry. This is the case for Czech syllabic consonants such as in krk, trvat, vlk, slza "throat, to last, wolf, tear". When asked, natives invariably identify two peaks in these words, which also count as two units in poetry. In addition, syllabic consonants bear stress in Czech: stress is wordinitial in this language. It thus regularly falls on the [r] of krk, trvat, and on the [1] of vlk, slza. We also know that syllabic consonants count just as much as vowels in Czech because of a bimoraic constraint that controls infinitives: a wellformed infinitive must either bear a long vowel, e.g. znát [znaat] "to know" (vs. po-znat [poznat] "to recognise"), two short vowels, e.g. topit [topit] "to heat", or one short vowel and one syllabic consonant, e.g. trvat [trvat] "to last" (see Kastler 1995:26, Scheer 1996,2001b). By contrast, Polish trapped consonants never count in poetry, and natives recognise only one peak in words such as trwac, krwi, bóbr, wiatr "to last, blood GENsg, beaver, wind". Furthermore, trapped consonants are unable to bear stress. Polish has invariable penultimate stress (e.g. malin vs. malina vs. malinami "raspberry GENpl, NOMsg, INSTpl"). However, words such as trwác "to last" and krwi "blood GENsg" are stressed on the vowel, not on the trapped [r]. Were trapped consonants able to bear stress,

The synchronic situation of trapped consonants: trapped vs. syllabic

289

they surely would in these examples. Stress also ignores word-final trapped consonants. The word jesiotr "sturgeon" for example is stressed on the first vowel. Were the final trapped consonant counted, the would be tonic. This picture is confirmed by Czech trapped consonants. In this language, [r] and [1] are trapped if and only if they occur word-initially before another consonant. Thus in words such as rdít se, rzi, rty, Ihát, Izíce "go red, rust GENsg, lips NOMpl, to lie, spoon" (see Vol.2,111.6.5 for the exhaustive list of these words), the initial sonorant is trapped. As in Polish, it is not counted as a syllabic peak by either poetry or natives, and it may not be stressed. It was mentioned before that stress falls on the initial syllable in Czech. Hence, were the sonorant in the above words a stress-bearing unit, it would be tonic. As a matter of fact, it is not: in all cases mentioned, stress falls on the first vowel. Finally, there is yet another category of trapped consonants in Czech: palatalised rhotics that are flanked by two consonants or occur in wordfinal position after a consonant. These environments, which make nonpalatalised rhotics syllabic (cf. krk, bratr "throat, brother" etc.), produce trapped results with : hrbitov, krtit, trpytit, hrbet, pepr, vnitr "cemetery, to baptise, to glance, back (human), pepper, interior" (see §§288,293 for full synchronic and diachronic detail). As before in Czech and Polish, in these words does not count in poetry, natives do not identify it as a syllabic peak (hrbitov for example has two peaks), and it is unable to bear stress (stress always falls on the first vowel). On the bottom line, thus, we can record a consistently opposite behaviour: syllabic consonants are visible for stress and in poetry, whereas trapped consonants are not. We will see on the following pages that this antipodal behaviour is also observed in regard of another test, i.e. the vocalisation of prefixes.

246

3.3. Czech syllabic consonants and préfixai vowel-zero alternations Syllabic consonants also line up with full vowels in regard of another property: in case there is a vowel-zero alternation to their left, zero surfaces. In other words, syllabic consonants are governors in the same way as ordinary vowels. Recall the behaviour of Czech consonant-final prefixes that has been discussed at greater length in §§28,94. They are vocalised iff the following root 1) begins with at least two consonants 2) that are root-initial and root-final, respectively. In less technical terms, prefixes are vocalised if the following root occurs in zero grade. There are various ways of identify-

290

1,10 Syllabic and trapped consonants in CVCV

ing the fact that the root occurs in zero grade. One is the existence of another form of the same root where a vowel overtly separates what appears to be the initial cluster. Such cases are shown under (204a). 163 (204)

a. root provoking vocalised prefixes two forms of the same root zero grade full grade /Vc,vc2/ Vc,c 2 - Wc,0 c 2 / pod-birat 'Pf VBR- pode-brat Pf inf roz-deru ls e VDR- roze-drat noun noun NOMsg her GENpl VHR- pfede-hra Pf od-hànët ¡pf VHN- ode-hnat inf ode-prat Λ/PRod-peru LSE noun adj sen NOMsg Λ/SN- beze-sny VSL-

vze-Sly

adj

§el

past act part.

VZDVDN-

pode-zdit beze-dny

inf

zed'

noun

adj

DEN

noun GENpl

NOMsg

b. root provoking non-vocalised prefixes no occurrence of Vc,vc2 pod-bradek roz-drobit od-hrabat roz-hnëvat vz-pruha pod-snëânik

roz-Slapat od-zdola —

The difference between the [br] of pode-brat "to seize from below' and the [br] of pod-bradek "double chin" is that the former, but not the latter, represents a root which occurs in zero grade. This first vowel of the root thus governs the zero of the root in /pode-borat/, but reaches the prefix in /pod0-bradek/. This is shown under (205) below. (205)

vocalisation of Czech prefixes a. unvocalised prefix: root in full grade pode-bradek "double chin" Gvt

I

163

C

V

I p

I o

C I d

V

-

C

Vo

c

V

C

V

C

I b central vowel > zero.

262

8.2. Diachronic situation: syllabic consonants come into being because a preceding vowel is lost The Germanic situation actually describes a scenario for a great many, if not for all languages: syllabic consonants are never diachronically primitive. They come into being because of an evolution that makes the melodic content of a neighbouring Nucleus fade away. 173 In case this emptied Nucleus occurs before a word-final consonant C# or in a closed syllable RTV, no governor is available that could secure its phonetic absence. One way of resolving this awkward situation is to provide new melodic 171

172

173

On the grounds of this complementary distribution in English, Gussmann (1991a) argues for the existence of empty Nuclei that are subject to a late schwa-insertion rule. Bell (1978:166) reports on cases where syllabic consonants have come into being because a following vowel was lost. However, he does not make any difference between syllabic and trapped consonants, to the effect that this statement needs to be verified for each language quoted. Be that as it may, the only source for syllabic consonants in English and German is the syncope of a preceding vowel. Bell (1978:165ss) confirms this statement on the grounds of a cross-linguistic record of 85 languages that bear what he takes to be syllabic consonants, which actually may well include trapped items as well.

310

1,10 Syllabic and trapped consonants in CVCV

content to the orphan empty Nucleus via spreading from a neighbouring 174

consonant. The question which kind of consonant may be the source of spreading is orthogonal to the present discussion. Some discussion is offered in §§253,617 and notes 195,386: sonorants are commonly syllabic, while the syllabicity of obstruents is subject to debate. In any event, English and German belong to those languages where only sonorants can be syllabic. If syllabic consonants exist in order to deliver melodic content to an adjacent orphan Nucleus, in principle this could be done by preceding as well as by following consonantal melody-providers. Now the hard observational fact is that syllabic consonants always seem to be born through the syncope of a preceding, not of a following vowel. This is the fundamental argument for representing syllabic consonants as left-branching structures, rather than as right-branching items. In any event, this is the motivation for proposals along the lines of (213a) that have been made by Harris (1994a:224s), Wiese (1986,1996) and others (even if this fact fails to be made explicit on many occasions). Another instance of this pattern is Slavic. Its modern representatives Czech, Slovak and Serbo-Croatian display syllabic consonants. Only the liquids [r] and [1] can assume this function in these languages. 175 It is a well-known fact that Slavic syllabic consonants continue former vowelliquid sequences. 176 The vowels at hand are so-called yers, which faded away since late Common Slavic. Yers come along in two flavours, one front "b", the other back ' V . They continue Indo-European short [i] and [u], respectively. Table (219) provides some illustration of the regular correspondences and the diachronic origin mentioned: a Common Slavic CyerRC sequence is continued by a syllabic consonant in Czech, Slovak and Serbo-Croatian, while Russian and Polish vocalise the yer (in a predictable way in the former, but in a rather complicated fashion in the latter language, see §278).

174

175

176

This reasoning will be refined in §§604,610, where I argue that the driving force of the consonantal expansion is the consonant's own positional precariousness, rather than the loss of the preceding vowel. With the exception of two words in Czech, i.e. sedm "seven" and osm "eight", which may be pronounced [sedm], [osm] in high-style speech, but most commonly appear as [sedum], [osum]. In Serbo-Croatian, only [r] can be syllabic since the lateral has vocalised in Codas, where it appears as [o]. In syllabic position, however, the vocalisation produces [u] (see table (219). Full detail of this rather complicated pattern will be provided in §277.

Syllabic consonants are left-branching structures: arguments (219)

311

Common Slavic VCbRC-/ VCtRC= Czech, Slovak, Serbo-Croatian VCRC= Polish, Russian VCVRCCbRC> syllabic Common Czech Slovak SerboCroatian Slavic gbrdlo hrdlo hrdlo grlo mirk-ivb mrkev mrkva stmbrtb smrt smrt smrt' pbrvb prvy prvy prvi vblna vina vina vuna vblkb vlk vlk vuk

> vocalised Polish Russian

gloss

garello marchew ámieré pierwszy weina wilk

throat carrot death first wool wolf

gorlo morkov' smert' pervyi volna volk

The genesis of syllabic consonants in Slavic is described in greater detail for example by Stieber (1979:33ss,54ss), Rospond (1979:94ss), Dtugosz-Kurczabowa & Dubisz (1993:84ss), Nahtigal (1961:11 lss), Panzer (1991:296ss), Carlton (1991:15lss), Vondrák (1924:180ss), Vaillant (1950:173ss), Meillet (1934:73ss), Mikkola (1913-50 II:200ss), Mann (1957:54). All authors take up the traditional 19th century description according to which "liquids took over the syllabic function from preceding vowels (yers) as they faded away". It thus appears that the Slavic evidence is strictly parallel to the Germanic case which was discussed earlier: in all instances, syllabic consonants come into being because a preceding vowel has been lost.

263

9. Syllabic consonants are left-branching: they govern in Czech At the present stage of the discussion, we are thus left with one single argument that has bearing on the competition between left- and rightbranching representations. It is of distributional nature and governs the synchronic as well as the diachronic pattern: consonantal syllabicity and preceding (not following) vowels are in complementary distribution. Let us now return to one of the three criteria that have demonstrated the true vocalic behaviour of Czech syllabic consonants, as opposed to the non-vocalic status of their trapped cognates in Polish. Recall from §§246s that prefixes remain unvocalised before syllabic consonants in Czech, whilst they produce a vowel before trapped consonants in Polish. The absence of préfixai vocalisation in Czech implies that Government strikes the prefix-final Nucleus. Hence, this Government relation must have a source.

312

1,10 Syllabic and trapped consonants in CVCV

Who could be the governor? This question was already asked under (212); the diagram is reproduced under (220) for convenience. (220)

who governs the préfixai Nucleus? a. option 1 : the Nucleus preceding the syllabic consonant Gvt

f C V C V - C V C V C V C V h a t

r ο ζ

roz0-trhat "to tear up"

b. option 2: the Nucleus following the syllabic consonant Gvt

f C V C V - c v c v c v c v h a t

r ο ζ

roze-trhat "to tear up"

By the time the question was raised, we did not have any clue as to what a syllabic consonant may look like phonologically. Therefore, the excursus into the discussion of syllabic consonants on the foregoing pages was necessary. We are now in a position to decide which option is correct. Nuclei that are involved in syllabic consonants are fully enabled lateral actors. If this is true, the putative governors under (220) must be linked to the melody of the syllabic consonant. In other words, the opposition "the governor is the Nucleus that precedes (option 1) vs. follows (option 2) the syllabic consonant" turns out to be a contrast between the two possible representations of syllabic consonants that we are keen to run against each other: left-branching (option 1) vs. right-branching (option 2). Hence, the real contrast of (220) is as under (221). (221 )

who governs the préfixai Nucleus? a. option 1 : the Nucleus of a leftbranching syllabic consonant Gvt

f C V C V - C V C V C V C V I N I I I r ο ζ

t

r

h a t

roze-trhat "to tear up"

b. option 2: the Nucleus of a rightbranching syllabic consonant Gvt

f C V C V - C V C V C V C V I 1/ I I I r ο ζ

t

r

h a t

roze-trhat "to tear up"

(221a) is well-formed provided that the empty Nucleus on the righthand side of the syllabic consonant is taken care of by the following

Syllabic consonants are left-branching: they govern in Czech 313 [a]. Nothing prevents the [a] from doing so. Cases where syllabic consonants precede clusters are discussed in §298. Let us now turn to option 2 under (221b) that represents syllabic consonants as right-branching items. This structure is ill-formed because the empty Nucleus enclosed by the [t] and the [r] remains orphan. Put another way, the Nucleus to the right of the syllabic consonant acquires lateral actorship through the spreading of the consonantal melody. However, it is preceded by two empty Nuclei, i.e. the one located in the prefix and the one that precedes the syllabic consonant. Neither of these empty Nuclei appears on the surface, to the effect that whichever one is targeted, the other will remain orphan. Now it could be argued that the [tr] cluster which involves the syllabic consonant and the preceding obstruent form a domain of Infrasegmental Government and thereby circumscribe the enclosed empty Nucleus. This option is depicted under (222). (222) who governs the préfixai Nucleus? right-branching structure plus Infrasegmental Government Gvt

v c I I ο

ζ

f v

-

c

v

c

I

v L ^

t [frekt] vs. lat frigida > [fregda]; in the latter example, the Latin [d] is protected by a following vowel. The words under (235b) demonstrate that word-final sonorants which follow the hardened velar stop behave exactly like the obstruents under (235a): they are 1) devoiced themselves and 2) hand over their voiceless character to the hardened velar stop, which surfaces as [k]. Finally, (235c) allows to control that sonorants, in identical position, are not devoiced if they are protected by a following vowel. Consequently, the preceding hardened velar stop does not undergo devoicing either and appears as [g]. Note that the presence of a word-final schwa is a direct function of the Latin origin: elsewhere in Romance, all final vowels disappear without any trace except -a, whose reflex is a schwa. The relevant contrast can be best appreciated when comparing the masculine and feminine forms of the word "hard": lat duru > [dikr] vs. lat dura > [digra].

328

1,10 Syllabic and trapped consonants in CVCV

The picture described is entirely parallel to the Polish pattern: wordfinal sonorants cause the devoicing of preceding voiced obstruents (kadr "staff GENpl" = [katr]). Only is the voiceless character of the sonorant in this position not commonly transcribed in the Polish tradition. In both Romansch and Polish, the devoicing of the sonorant is due to the conjunction of two parameters: the fact of 1) occurring word-finally and 2) after an obstruent. It is the latter characteristic that makes the sonorant trapped. This may be controlled by looking at its behaviour in word-final position when preceded by a vowel. It has already been demonstrated under (231) that sonorants do not devoice in this position (Polish czar [tjar] "magic (noun)"), and the same holds true for Bravuogn: [gatar, paadsr, flajval] "grid, father, feeble" etc. (Montreuil 1999:531). Hence, there is an overt causal relation between the occurrence in word-final position after a consonant and the fact of being trapped. Trappedness, in turn, exposes the sonorant to final devoicing. In other words, it transforms the sonorant into an obstruent. Finally, Montreuil (1999:542) also demonstrates that the mechanism at hand is still active in the synchronic grammar of Bravuogn: the [-T-voice r#] vs. [-T+voice ra] alternation operates in various paradigms such as masculine and feminine forms of adjectives (236a), collective (236b) and diminutive (236c) formation. (236) synchronic devoicing of trapped sonorants and preceding obstruents in Bravuogn a. masc. fem. gloss frekt fregda cold dikro digra hard pokr pogra farmer b. singular collective gloss pear la pegra iX pekr c. noun diminutive gloss pokr pogret farmer Let us now summarise the foregoing discussion.

Phonetic correlates of syllabic and trapped consonants 329 276

12.3.5. Summary Trapped sonorants are sensitive to voice assimilation in a way that is untypical for sonorants: they devoice word-finally and hand over voice values in CrC clusters. An alternative to the classical analysis that assigns an extrasyllabic status to trapped consonants is to acknowledge the fact that trappedness implies a demotion from a sonorant to an obstruent. Arguments in favour of this option are provided by the Romansch alternations discussed. The fact that the trapped status of a sonorant causes its transformation into an obstruent is rather unexpected. I do not know why sonorants that branch on the following Nucleus should lose their status as a sonorant, while the parallel left-branching structures remain fully sonorant. On the other hand, this very fact fits the general picture since syllabic consonants certainly deserve to be described as "strong" when compared to their trapped cognates. It seems natural to expect that the weak, rather than the strong way of colonising an adjacent Nucleus is demoted. In any event, it is interesting to observe that spreading on the following Nucleus has the same effect on a sonorant as the modification of its place of articulation: in the same way as trappedness, both palatalisation (Polish , Czech ) and uvularisation (French and German = [χ,κ]) cause demotion. In the following section, the diachronic situation of Polish trapped consonants is examined.

277

13. Diachronic confirmation: yers followed trapped, but preceded syllabic consonants The purpose of this section is to introduce the upshot of the analysis that has been presented so far: the structure that was derived for trapped and syllabic consonants on purely synchronic grounds is a diachronic reality. In short, trapped consonants have come into being iff a following yer was lost, whereas their syllabic cognates were born when a preceding yer fell out. In order to establish this fact which is commonplace in all diachronic grammars, we need to retrace the philological reasoning in some detail.

330

278

1,10 Syllabic and trapped consonants in CVCV

1 3.1. The Western Slavic comparatistic situation: cz CRC = pol CVRC and CRC In §244, Polish trapped consonants were introduced with reference to Czech, where they are syllabic in the same words. Table (202), which provides relevant information, is repeated below for convenience. (237)

Czech VCRC- = Polish VCRCCommon Slavic CrC trbvati CrzC dvbri grbmëti brtnëti chrbbbtb trtstina C1C slbza klbnplbvblicha

Polish

Czech

gloss ish)

trwaé drzwi grzmieé brzmieé grzbiet trzcina Iza < slza klnç plwocina

trvat dvefe hfmët brnêt hfbet trstina slza klnout arch plvat > plivat old cz bicha > blecha

to last door to thunder to sound back reed (plant) tear I curse sputum

to last door to thunder to tickle back reed (plant) tear to curse to spit

flea

flea

ρ chía

(Pol- gloss (Czech)

Unfortunately, the relation "pol trapped = cz syllabic" is not an equivalence: it is true that all consonants which are trapped in Polish appear as syllabic in Czech, but the reverse is false. Table (238) shows that the Polish response of a Czech syllabic consonant may also be a sonorant preceded by a vowel, which therefore is neither trapped nor syllabic. For reasons that will be clear as the discussion unfolds, I refer to this kind of Polish reflex as "vocalised syllabic consonants".

Diachronic confirmation: yers followed trapped, but preceded syllabic Cs (238)

331

Czech VCRC- = Polish VCVRCPolish reaction CaRC: 34

Common Slavic gtr-dlo g-brt-tb

dbr-m> sbr-na CieRC: 16 pbrsi sbrpi> CiRC: 4 vblki CeRC: 6 vblna sbrdb-ce pblni> Total: 60

Czech

Polish

Czech gloss

Polish gloss

hrdlo hrst

gardlo garsé

dm srna prsa srp vlk vina srdce piny

dará sarna pierá sierp wilk welna serce pelny

throat (cupped) hand lawn roe breast sickle wolf wool heart full

throat (cupped) hand lawn roe breast sickle wolf wool heart full

This table shows Polish and Czech cognates, whose Common Slavic origin is identified in column two. Polish reflexes of these words always show a vowel before the sonorant that is syllabic on the Czech side, but the quality of this vowel is greatly variable: vocalisations in [a,ie,i,e] are found, and even some marginal cases in (onomatopoeia) and . In order to be able to get a handle on this situation, the table also attempts at giving a flavour of the numeric proportions. The first column shows the number for each vocalisation, on the account of a total of 60 instances of the equation "pol C V R C = cz CRC". 1 7 9 It is a classical issue in Polish diachronic grammar to predict the quality of the vowel from contextual parameters. Even though it can be clearly established that the place of articulation of the consonants preceding and following the sonorant condition the quality of the vowel that appears, some vocalisations resist explanation. The discussion of that matter is led in, among others, Stieber (1973:23s,42ss,1979:54ss), Dlugosz-Kurczabowa & Dubisz (1993:84ss), Rospond (1979:94ss), Nahtigal (1961:11 lss), Carl179

This 60-item list is a compilation of various etymological dictionaries and historical grammars that I have established. The sources include Havlová & Erhart (1989-2002), Bankowski (2000), Brückner (1927), Machek (1957), Holub & Kopeòny (1952), Rejzek (2001), Rospond (1979:95ss), Stieber (1979:33ss,54ss), Nahtigal (1961:11 lss). The philological situation is rather complex and cannot be discussed at length here. The total number of roots that can be come by probably exceeds 60 a bit (yet not much). This not withstanding, table (238) certainly provides an accurate overview of the proportions.

332

1,10 Syllabic and trapped consonants in CVCV

ton (1991:249s), Vondrák (1924:183ss), Mikkola (1913-50 II:201s), Wijk (1949-50:44s). For the purpose of the present demonstration, the distribution of vowels before vocalised syllabic consonants is not relevant. The only thing that needs to be borne in mind is that vowels always appear be1 Riì

fore, never after the sonorant. Of course, the challenge is to sort out this Polish double-reflex for Czech syllabic consonants: can it be predicted whether the Polish response to a Czech syllabic consonant is a vocalised or a trapped sonorant? The answer is yes. All diachronic grammars inform about the fact that Polish has vocalised reflexes if a yer preceded the sonorant in Common Slavic, while trapped consonants are produced in case a yer followed the sonorant in Common Slavic. This distribution can actually be read off the tables (237) (trapped sonorants) and (238) (vocalised sonorants): the Common Slavic origin of all words of the former is CRiAC, while all instances of the latter go back to C i A R C . In the Slavic philological tradition, the relevant cases are referred to under the banner of two formulas, tbrt vs. trbt where "t" is any consonant. 181 In diachronic grammars, the evolution of sonorants that neighbour a yer is always strictly separated according to this pattern, which, among other distinctions, allows to tell trapped from vocalised sonorants in Polish. Typical presentations of the facts can be found in, among others, Stieber (1979:54ss), Nahtigal (1961:111s), Carlton (1991:151ss,249s), Vaillant (1950:173ss), Panzer (1991:296ss), Vondrák (1924:181). This diachronic reality may thus be booked as an outstanding confirmation of the analysis that has been developed for syllabic and trapped consonants. However, several questions remain. For one thing, it is not clear why Czech does not reproduce the Common Slavic opposition tbrt vs. trbt in the way Polish does: both origins are merged and appear as syllabic consonants. Also, how is the Common Slavic opposition between tbrt and trbt established? These questions are addressed on the following pages. 180

181

With one apparent exception, that is CluC- vocalisations such as in CS tblstb > pol thisty = cz tlusty = slk tlsty "thick". This post-vocalisation is secondary and hence does not bear on the present discussion. It is addressed at greater length in §287. In actual fact, both formulas are further subdivided according to whether the yer is front or back, and the sonorant an [r] or an [1], Hence, Common Slavic formula that diachronic grammars operate with extend to tbrt, tbrt, tblt, tblt for cases with a preceding yer, and trbt, trbt, tlbt, th>t for those where a yer follows the sonorant. For the time being, this detail does not suit the demonstration; it will only be mentioned when necessary.

Diachronic confirmation: yers followed trapped, but preceded syllabic Cs 333 279

13.2. The Common Slavic and Old Church Slavonic sources of Polish vocalised liquids The ultimate origin of the words whose sonorants are prevocalised in Polish but syllabic in Czech is undisputed: the sonorants in question were syllabic in Indo-European. The words in which they are found instantiate the IndoEuropean equation that identifies Indo-European syllabic sonorants (i.e. the zero grade of roots). The following table provides some illustration for Indo-European syllabic r (see for example Meillet 1937:118ss, Szemerényi 1990:47ss, Panzer 1991:296ss). (239)

equation establishing IE r skr gr lat Germanic r ar, ra or, ur ur mrtam mors got maúrpr kardia cordis got hairto

lit ir, ur mirtis Sirdis

CS br, i r simbrtb sbrdbce

pol Vr smierc serce

cz r smrt srdce

As may be seen, Baltic stands out in that its vocalisation of the IndoEuropean syllabic liquid is twofold: either -ir- or -ur- is found (the same holds true for the lateral). The Slavic response to the Baltic pattern is -br-, -ir-, whereby the correspondence bait ir = si br vs. bait ur = si t r is entirely regular (e.g. Vondrák 1924:420). Therefore, the question arises in which circumstances i-vocalisation has occurred, and which words have been subject to u-vocalisation. This debate is reflected for example in Vondrák (1924:420s), Arumaa (1964:151ss), Mann (1957:51ss); Meillet (1937:119) is explicit on the mystery of this distribution. Another problem is the fact that the Common Slavic state of affairs is not witnessed by direct recordings. Only Old Church Slavonic (OCS) provides written testimony. And as a matter of fact, OCS texts consistently show the yer after the liquid. Hence, bait ir, il, ur, ul = OCS rb, lb, n>, I t as for example in lit mirtis, sirdis, gurklys, burzdus = OCS stmrbtb, srbdbce, grblo, b r b z t "death, heart, throat, moving". At first sight, it seems that the presence of the yer "on the wrong side" of the liquid has a direct explanation in terms of the Common Slavic liquid metathesis: the Slavic response of CVRC sequences that are found in other Indo-European languages is CRVC (e.g. Panzer 1991:291 ss, Nahtigal 1961:108, Carlton 1991:144ss). Compare for example germ Berg, Milch, lat hortus, lit galva with OCS b r ë g i , mlëko, gradi>, glava. Hence, on the grounds of the more general Slavic metathesis that makes concord Baltic and Slavic reflexes of Indo-European syllabic conso-

334

1,10 Syllabic and trapped consonants in CVCV

nants, a Common Slavic stage sr, i r , bl, τ,Ι is reconstructed. Of course, the yers suppose former ir, ur, il, ul since they continue IE i,u (e.g. lat vidua = CS vbdova, lat duo = CS d t va). Given these observations, the general chronology is as under (240).182 (240)

evolution of Indo-European syllabic liquids in Slavic IE r, J > balt-slav ir, ur, il, ul > CS br, ir, bl, τ>1 > OCS rb, rb, lb, It

However, there are strong indications that , lb, h>> in OCS script do not actually represent a CV sequence. For instance, it is not infrequent to observe that the yer misses altogether (Vondrák 1924:181) in the texts. Also, the scribes consistently mismatch both yers: CS b regularly appears in OCS texts as >, and vice versa (Wijk 1949-50). Therefore, it is generally assumed that OCS is just a way to transcribe syllabic consonants: [r', r,}', J] (where r' and J' are palatalised versions of r, 1) were the actual sounds present in OCS (Rospond 1979:94, Vondrák 1924:181, Carlton 1991:152, Wijk 1949-50). On this analysis, there was no metathesis of yer-liquid clusters at all. Common Slavic CbRC, C t R C sequences simply lost their yer and gave birth to syllabic consonants that kept the memory of the original front vs. back opposition of the yers: CbRC > CR'C with a palatalised syllabic liquid, against C t R C > CRC where the syllabic consonant is not palatalised. Hence, (240) needs to be recast as (241). (241)

evolution of Indo-European syllabic liquids in Slavic IE r, 1 > balt-slav ir, ur, il, ul > CS br, tr, bl, t l > OCS r', r, Γ, |

The following section inquires on the reality of the reconstructed Common Slavic opposition between tbrt and trbt.

182

This is the picture that is most widely accepted for the reasons mentioned. For instance, Stieber (1973:17,1979:35), Wijk (1931), Arumaa (1964:151ss), Vondrák (1924:180s,420s), Carlton (1991:151ss), Vaillant (1950:173 ss), Schenker (1995:94) adhere. See Peciar (1941) for a peculiar view on the relation of syllabic consonants and the Common Slavic metathesis. However, another scenario is argued for by Pedersen (1905:340), Rospond (1979:95) and DhigoszKurczabowa & Dubisz (1993:84s) who hold that syllabic consonants did not vocalise in Common Slavic. Instead, the Indo-European syllabic consonants were inherited as such by CS, and only later developments led to pre- or postvocalised liquids.

Diachronic confirmation: yers followed trapped, but preceded syllabic Cs 335 280

13.3. CS tbrt and CS trbt have never merged

281

13.3.1. CS tbrt and CS trbt remain distinct in Polish But there is yet another reason why OCS , lb, 1τ>> from CS br, t r , bl, •bl could not possibly represent a CV sequence. If they did, they would have merged with the original CS rb, rb, lb, It, which appear in OCS script as > as well. We are sure, however, that such a merger has never occurred, for CS tbrt and trbt show contrasting reflexes all over Western and Eastern Slavic languages, e.g. Wijk (1931:59), Vondrák (1924:181), Stieber (1979:56s), Vaillant (1950:173ss), Panzer (1991:297). We already know one example where both origins remain distinct down to the modern pattern, that is the Polish opposition between vocalised CVRC (238) vs. trapped CRC (237) sonorants. Both have syllabic cognates CRC in Czech, but the former continues tbrt, while the latter is the reflex of trbt. Hence, it all comes down to the crucial distinction between tbrt and trbt in Common Slavic. Both patterns are graphically merged into trbt in OCS, which therefore does not help to ascertain the CS contrast. Is there an independent way of establishing that tbrt and trbt were really different in Common Slavic? Also recall that the entire diachronic argument in favour of the representations that have been derived earlier hinges on the opposition tbrt vs. trbt: trapped consonants are right-branching structures because they continue a trbt sequence where the liquid has taken over the syllabic function of the following yer by branching on its Nucleus. On the other hand, syllabic consonants of the Czech type branch on the Nucleus to their left because they are the reflexes of tbrt sequences. Hence, there is no argument unless the Common Slavic distinction between tbrt and trbt can be safely established.

282

13.3.2. CS tbrt and CS trbt remain distinct in Baltic and Eastern Slavic Fortunately, both inner-Slavic and Baltic comparatism allows to tell CS tbrt from CS trbt without ambiguity. Reconsider the comparatistic situation of Polish trapped consonants. The information contained in table (237) has been augmented by Baltic, Eastern Slavic and other Indo-European reflexes under (242) below.

336

1,10 Syllabic and trapped consonants in CVCV

(242) trapped in Polish: Baltic CRi/uC = ESI CRe/oC = Czech CRC I = Polish CRC other IE Baltic Common Estern Polish (lit) Slavic Slavic (ms) travati ukr tryva- trwac skr dhruva, lat durua ty kn.vb krew, krwi kraujas krov', krovi skr dvaaras dvbri dver' drzwi dvaras germ Gram, grumenti grbmëti gremet' grzmieó gr khromos lat fremo, brbnëti ukr brzmieó germ brenity Bremse, skr bhramaras chrbbbth grzbiet chrebet truSis trbstina trostina trzcina < germ krist krbstb chrzest, krest, kresta chrztu slbza germ schluiliukti sleza Iza < stza cken klbnkljanu klnç lat glutire gl-btati glotat' οροί kltaé plbvplwaé plevat'

skr plutas, gr plytos germ Floh

latv pluts ph>tb blusa

bltcha

plot', pioti blocha

pteé, ptci pchla

Czech

trvat krev, krve dvefe hfmët brnët

hrbet trstina kïest, kïtu s Iza klnout hltat arch plvat > plivat pit', piti ocz bicha > blecha

The evidence is conclusive: in all cases where Polish shows a trapped consonant CRC, both Eastern Slavic and Baltic respond with postvocalised sonorants CRVC. Moreover, the quality o f the vowel in Eastern Slavic and Baltic concords: Baltic i, whose Slavic version is b, appears as in Eastern Slavic, while Baltic u, which corresponds to Slavic t , is vocalised as on Eastern Slavic grounds. This simply means that yers in C R h / i C are regularly vocalised in Eastern Slavic as far as their quality is

Diachronic confirmation: yers followed trapped, but preceded syllabic Cs

337

concerned. 183 Other Indo-European languages also confirm that these words originally possessed a vowel on the righthand side of the sonorant. In any event, it is for sure that the pan-Indo-European equation under (239), which identifies original Indo-European syllabic consonants, is certainly not reproduced by the trapped pattern under (242). We are thus in a position to expect that Baltic and Eastern Slavic also provide concordant evidence for those sonorants which are vocalised in Polish. These should go back to CS tbrt and hence bear a vowel on the lefthand side the sonorant. Table (243) reproduces the equation pol CVRC = cz CRC (238), which is augmented by Baltic, Eastern Slavic and other IndoEuropean information. 184

183

184

However, their very presence is unexpected. The general pattern in Slavic languages distinguishes between so-called strong and weak yers: the former occur in _ C # , _ C C V and _Cb/tCV, the latter appear in _ # , __CV. Strong yers are vocalised, while weak yers fall out, hence for example CS dbn-b > e.g. cz den "day NOMsg", but CS dbn-a > cz dne "day GENsg". Of course, this is also true for Russian: den', dn'a, respectively, e.g. Kiparsky (1963 I:93ss); the general evolution of yers is described for example in Panzer (1991:303ss), Vondrák (1924:160ss). Therefore, yers in CRb/tCV should be lost. Contrary to expectation, the above table consistently shows that yers in this position are vocalised, e.g. CS glitati > rus glotat'. Carlton (1991:153) is puzzled by this fact, and a solution does not seem to be in sight. The consistent Eastern Slavic reflex Ce/oRC which, recall, corresponds to OCS CRB/TJC also allows to firmly discard the view that the OCS situation is the result of regular Slavic metathesis (cf. §279c). Under the metathesis analysis indeed, CS tbrt would have produced OCS trbt where the sequence was really pronounced [CV], Were OCS trbt the result of metathesis, Eastern Slavic would have to show so-called pleophonia. This term refers to the regular Eastern Slavic output of the Slavic metathesis, which bears a vowel on both sides of the sonorant. Compare for example the Russian reflex of the words quoted in §279c: germ Berg, Milch, lat hortus, lit galva = OCS brégb, mlëko, gradi, glava = rus bereg, moloko, gorod, golova. If words such as OCS srbna were the result of metathesis, Russian should produce **serena, which it does not: only serna is attested. MareS (1956:457, 1965:23) makes the same point, and Wijk (1949-50:42) also provides a consistent scenario of the evolution of CS tbrt in Russian. This is further support in favour of the assumption that OCS trbt < CS tbrt is but a way of transcribing syllabic sonorants.

338 (243)

1,10 Syllabic and trapped consonants in CVCV vocalised in Polish: Baltic Ci/uRC = ESI Ce/oRC = Czech CRC = Polish CVRC other IE Baltic (lit) Common Estern Sia- Polish Slavic vie (rus) gardlo gir-dlo gorlo lat gurgulio, germ gurklis Gurgel garáó gurste gbrt-tt gorst' gr a-gortos ptrstb arch perst parst piritas skr prStiS, oiran pariti, germ Fürst serna sarna lat cervus, gr keras, latv simas, sbr-na skr áiras oprus sirvis, lit stima pbrsi skr parsu piráis persi piers sbrpi serp sierp lat sarpio, gr harpee, latv sirpe vblkb wilk skr vrkas, got wulfs, vilkas volk alb ulk vblna vilna, volna welna oiran varna, got wulla oprus vilna arm sirt, lat cordis, Sirdis sbrdb-ce serdce serce got herto, gr kardia got fills, skr purnas, pilnas pblnb polnyi pelny but lat plenus, gr pleios

Czech hrdlo hrst prst sma

prsa srp vlk vina srdce piny

Our expectations are satisfied: the sonorant is always preceded by a vowel in both Baltic and Eastern Slavic. In a great majority o f words, this is also the case in other Indo-European languages, where various languagespecific treatments or the inheritance of the full degree may disturb the picture. In any event, Polish vocalised and trapped sonorants may now be safely retraced to a Common Slavic opposition tbrt vs. trbt. The following table embodies both relevant equations, which are established on the grounds o f the comparatistic evidence reviewed.

Diachronic confirmation: yers followed trapped, but preceded syllabic Cs (244)

339

Polish vocalised vs. trapped consonants continue CS tbrt vs. trbt185 hence: CS Baltic ESI Pol a. Polish trapped sonoCRt/tC CRi/uC CRe/oC CRC rants, cf. (242) b. Polish vocalised sonoCiART Ci/uRC Ce/oRC CVRC rants, cf. (243)

This result shows that tbrt and trbt were perfectly distinct in Common Slavic and moreover have never merged in Eastern or Western Slavic (more on Czech and Slovak shortly). Hence, OCS script, which writes trbt for both CS tbrt and trbt, does not represent the actual pronunciation.

283

13.3.3. The yers of OCS trbt < CS tbrt refuse to vocalise Yet another argument comes from yer-vocalisation, which cannot be understood if OCS , lb, l i > represent anything else than a mere graphic merger of CS tbrt and trbt. As a matter of fact, "metathesised" CS tbrt > OCS refuses to undergo yer-vocalisation (e.g. Komárek 1962:24). Words that instantiate the opposite pattern OCS trbt < CS trbt regularly vocalise the yer in strong position: CS krbvb, krbstb, ph>tb = OCS krbvb, krbstb, p l i t b > cz krev, pol krew, chrzest, plec. By contrast, "metathesised" CS tbrt > OCS appears only in prevocalised (Polish) or syllabic (Czech) dress: e.g. CS dbrm>, sbrna, vblkb = OCS drbnt, srbna, vlbkb > pol dará, sarna, wilk, cz dm, srna, vlk. Had CS tbrt and CS trbt really merged into trbt in OCS, the yer in OCS trbt would have been strong. Therefore, it should have been vocalised in all instances of OCS trbt, including those that come from CS tbrt. The fact that OCS only undergoes yervocalisation in case it continues CS tbrt proves that 1) CS tbrt and CS trbt were not merged in OCS and 2) OCS script that represents CS tbrt does not translate a [CRVC] sequence.

185

I could only identify two irregular representatives for Polish, both of which instantiate the same root: CS gbrdlo = lit gurklis, > rus gorlo, cz hrdlo that gives the regular pol gardto "throat" also produces pol grdyka "Adam's apple" and pol krtañ "larynx" = ru gortan' with a trapped consonant instead of a prevocalised solution. Etymological dictionaries such as for example Machek (1957:185), Rejzek (2001:215) invoke an irregular "expressive" derivation.

340

284

1,10 Syllabic and trapped consonants in CVCV

1 3.3.4. Summary: trapped consonants result from the loss of a following yer, but what about their syllabic peers ? For all these reasons, most grammars interpret OCS trbt < CS ti>rt as [trt] anyway (cf. §279 and note 184). Therefore, the scenario that is presented for example by Carlton (1991:152s) certainly matches reality: tbrt and trbt were distinct in Common Slavic. The loss of the yers caused the syllabicity of the sonorant in CS tbrt > OCS [trt]186, but not in CS trbt > OCS [trbt], both being written in OCS script. We may now come back to the initial question regarding the representation of syllabic and trapped consonants. The analysis according to which the latter colonise the following Nucleus, while the former are leftbranching structures receives solid diachronic support: the Polish opposition vocalised vs. trapped is the true expression of the Common Slavic contrast tbrt vs. trbt. Or rather, we now are in presence of positive evidence for the view that trapped consonants can only be born if the Nucleus to their right is empty (or emptied). For the time being, however, there is no diachronic support in favour of the left-branching analysis of syllabic consonants that has been derived earlier from synchronic alternations. This is due to the simple fact that Polish does not possess any syllabic consonants. In order to be able to evaluate the diachronic evidence regarding syllabic consonants, we must examine their situation in Czech and Slovak.

186

At least in those dialects that possess modern syllabic reflexes, i.e. South Slavic (save Bulgarian), Czech and Slovak. According to Carlton (1991:151s), the modern prevocalised results of Polish and Eastern Slavic are the direct continuation of the yers and did not go through a syllabic stage. The alleged evolution CS tbrt > trt in South Slavic, Czech and Slovak is also consistent with the interpretation of syllabic consonants that is developed here. On the other hand, the fate of CS trbt is less clear: these sequences may have directly resulted in trapped sonorants in dialects where the modern situation allows for a trapped interpretation, i.e. South Slavic, Czech and Slovak (on which more shortly), but certainly not in Eastern Slavic where post-vocalisation CS trbt > ESI tret witnesses the faithful contrast with CS tbrt > ESI tert, but leaves us with the puzzling and irregular vocalisation of yers in weak position that was discussed in note 183. Unconvincing solutions for the problem regarding the consistent maintain of the CS contrast between tbrt and trbt in the evolution of particular dialects are proposed by Vondrák (1924:181) and Wijk (1931:59s). As usual in this kind of situation when apparent exceptions need to be treated, analogy is invoked.

Diachronic confirmation: yers followed trapped, but preceded syllabic Cs 341 At first sight, though, the Common Slavic contrast tbrt vs. trbt does not seem to have any effect in these languages since both configurations are represented by syllabic consonants (which, recall, correspond to both vocalised and trapped consonants on the Polish side). However, this will turn out to be but a modern mirage.

285

13.4. What has happened to Czech (and Slovak) trapped consonants?

286

1 3.4.1. Old Czech: sonorants from CS tbrt are syllabic against trapped from CS trbt Tables (242) and (243) show that the Czech reflex of both Common Slavic tbrt and trbt is a syllabic consonant. The expected pattern, however, is an independent continuation for both origins. Since Czech allows for syllabic consonants, there is no reason why CS tbrt > OCS trt should be vocalised as in Polish. However, CS trbt should produce trapped consonants as much as it does in Polish. Czech indeed shows the expected syllabic consonants for CS tbrt > OCS trt > cz trt (= pol tVrt), but the same result (instead of the trapped solution) is also encountered for CS trbt > cz trt (= pol trt). In other words, the evidence for viewing syllabic vs. trapped consonants as left- vs. right-branching structures only relies on the Polish opposition CS tbrt > pol tVrt vs. CS trbt > pol trt. Both origins remain distinct and identify trapped consonants as right-branching structures, but there does not seem to be any evidence to the effect that syllabic consonants are leftbranching: the supposed syllabic OCS trt < CS trbt comes out as pol tVrt. On the other hand, Czech indeed produces the syllabic consonants requested, but is mute on the trapped side. An ideal language in support of the analysis presented would have a syllabic reflex for CS tbrt, as opposed to a trapped continuation of CS trbt. This language exists: Old Czech. The following table summarises the comparatistic situation regarding the reflexes of CS tbrt and trbt for the various languages discussed.

342 (245)

1,10 Syllabic and trapped consonants in CVCV Western Slavic reflexes of Common Slavic tbrt and trbt Common Slavic tbrt OCS trt Old Czech trt Modern Czech, trt Slovak tVrt Polish

syllabic syllabic syllabic vocalised

«

trapped (?) trapped syllabic

example sbrna - trbvati srna - trvati snia - trvati srna - trvat

trt

trapped

sarna - trwac

trbt trt tit

It is notorious in all diachronic grammars of Czech that the modern situation is the result of an evolution which has taken place in historical times and can be followed step by step in written testimony. That is, Modern Czech syllabic consonants, when identified according to their origin, fall into so-called "primary" and "secondary", or "old" and "new" items. Old syllabic consonants go back to CS tbrt, while their younger peers that have emerged in historical times continue CS trbt. The historical change that is transparent in Old Czech script transforms trapped into syllabic consonants. Hence, CS trbt > ocz trapped trt > mcz syllabic tit. Table (246) summarises this evolution and, for the sake of completeness, also provides additional information on parallel events (CS trbt > ocz tr't > mcz trapped tft, ocz Cray, tlsty > mcz öerny, tlusty), which will be discussed in the following section. Relevant material on the distinction old vs. new syllabic consonants, as well as on the evolution ocz > mcz syllabic consonants is available in Tràvniôek (1935:57s, 111 ss, 226ss), LehrSpíawmski & Stieber (1957:97ss), Komárek (1962:60s, 82, 97ss, 127ss), Liewehr (1933:93s, 162s).

Diachronic confirmation: yers followed trapped, but preceded syllabic Cs 343 (246)

evolution of Common Slavic tbrt and trbt in Czech

This table reads as follows. The r in the formulas tbrt, tbrt and trbt, trbt represents both liquids [r] and [1] in this position. The distinction between front b and back τ> yers is also relevant: the former has produced palatalised versions of syllabic and trapped consonants, which are traditionally noted r' and Γ. The opposition trapped palatalised tr't < trbt vs. trapped non-palatalised trt < trbt needs to be mentioned because it is preserved in Modern Czech: the latter has become syllabic, while the former is still trapped (more on this shortly).

287

13.4.2. Secondary vocalisation of syllabic consonants in Czech (and Polish, Slovak) Before discussing the crucial development ocz trapped slza > mcz syllabic sjza and the existence of trapped consonants in Modern Czech (type kftit "to baptise") at greater length, a few words on the pattern mcz cerny, tlusty are in order. Old Czech syllabic consonants from CS tbrt have been vocalised in two different environments. An [ε] has appeared before the syllabic consonant in case it was preceded by a palatal obstruent, i.e. type ocz crny, crt,

344

1,10 Syllabic and trapped consonants in CVCV

örpati, zrt > mcz öerny, fiert, ¿erpat, íert "black, devil, to scoop, joke" 187 , against ocz = mcz trh, trpëti, smrt, srp, krk, hrdlo "market, to suffer, death, sickle, neck, throat" (see for example Trávníóek 1935:122, Stieber 1979:54, Lehr-Splawmski & Stieber 1957:78, Carlton 1991:233,240s, Mann 1957:55, Wijk 1949-50:44s). The Old Czech syllabicity of the sonorant in these words is guaranteed by the script itself, but can also be read off Slovak where the old state of affairs has been preserved in most words, compare slk zrd', érpat' = mcz zerd', öerpat "flagpole, to scoop". 188 On the other hand, ocz syllabic non-palatal 1 and palatal 1' have developed an [u] (or ocz long [uu] > mcz [ow]) on their righthand side, except for palatal Γ that follows labials. This is illustrated by the table under (247) below. 189 (247)

type tlusty: u-postvocalisation of syllabic } in Modern Czech CS ti.lt > ocz tit mcz tlut CS tblt > ocz tit silnbce after dentals dblgb dlouhy stblpi and palatals tblkti tlouci tblpmb tlum tblstb frblnt Turkish tilmai tlumoöit 2blva 21uva 2blna 21una 21uty 2bltb CS tblt > ocz tjt mcz tlut CS tblt > ocz tit after velars chblmb kblki> gblkb CS tbit > ocz tit after labials vblna pbim> vbikb iDblketi

187

188

189

mcz tit vina piny vlk mlöet

CS tblt > ocz tit rrrh 1 vit i

mcz tlut slunce sloup tlusty Clun

mcz tlut chlum kluk hluk mcz tlut mluvit

Other members of this paradigm are mcz 2erd', öerv, öerstvy, óermák, ôerven, öerveny, öervenec, Stërbina, Stërbâk, Stèrk, iernov "flagpole, worm, fresh, kind of bird, June, red, July, crack/ gap, endive, gravel, millstone". But slk öierny, Cert, öerv, öerveny = mcz Cerny, Cert, öerv, öerveny "black, devil, worm, red". Glosses for table (247), line by line: after dentals and palatals "long, sun, to hit, pillar, group (people), fat, to translate, (small) boat, golden oriole, kind of bird, yellow"; after velars "hill, boy, noise"; after labials "wave/ wool, full, wolf, to shut up".

Diachronic confirmation: yers followed trapped, but preceded syllabic Cs

345

The same movement has affected Polish, although in slightly different contextual conditions. 190 And again, Slovak has not participated in the abandon of syllabic consonants, this time without exceptions: mcz tlusty, slunce, dlouhy, zluty, zluc, clun, kluk = slk tlsty, slnko, dlhy, zlty, zlc, cln, klk "fat, sun, long, yellow, gall bladder, (small) boat, boy" etc. The postvocalisation of syllabic } is described for example in Lamprecht (1986:41), Trávnícek (1935:112), Lehr-Splawmski & Stieber (1957:78), DhigoszKurczabowa & Dubisz (1993:88s), Vondrák (1924:183 ss), Stieber (1979:55), Carlton (1991:233 Czech, 240s Slovak, 248s Polish).

288

13.4.3. The Old Czech phonemic opposition trapped < CS tri>t vs. syllabic < CS tbrt We can now turn to the major interest of table (246), that is the synchronic situation that was prevailing in Old Czech. Recall that this language is supposed to provide direct evidence for the opposition "syllabic consonants < tbrt vs. trapped consonants < trbt".

289

13.4.3.1. Syllabic liquids from CS tbrt do, but trapped liquids from CS trbt do not count in Old Czech verse Of course, Old Czech is only known by written testimony. Therefore, the question arises how the contrast syllabic vs. trapped could be read off a text. The major argument is versification, i.e. one of the criteria that allow to distinguish syllabic and trapped consonants: the former do, the latter do not count in poetry. Consequently, as soon as there is a versified text with a stable number of syllables per line, it may be established without ambiguity whether a sonorant that is surrounded by other consonants is trapped or not. Written testimony from Old Czech has been handed down since the second half of the 13th century A.D. For about a hundred years, clusters from CS trbt did not count in poetry and thereby identify as trapped, whereas the reflexes of CS tbrt > ocz weigh in the versification of the same texts. By the end of the 14th century, however, trapped from CS trbt start to count as well. Therefore, the evolution CS trbt

190

Compare mcz tlusty, slunce, mluvit = pol tlusty, stoñce, mówié (< motwié) "fat, sun, to talk" with mcz iluty, ölun, hluk, chlum = poi zóhy, czólno, gieik, Chetai "yellow, (small) boat, noise, hill".

346

1,10 Syllabic and trapped consonants in CVCV

> trapped ocz trt > syllabic ocz, mcz trt, which is shown under (247), can be almost followed in real time. Consider some examples of older sources under (248). In all cases, the poetry obeys typical Old Czech Alexandrine verse, i.e. counting eight syllables. 1 9 1 The change from trapped to syllabic consonants in Old Czech is studied in greater detail by Smetánka (1940), w h o provides much raw material, dating and counts for individual texts. The following examples have been collected by Lehr-Sptawiñski & Stieber (1957:97) and Komárek (1962:128s). (248)

older sources of Old Czech: r in trt < CS trht does not count context CrC < trbt text dating a. C C within a root 1 23 4 5 6 78 we krwi jakito vodë kalé krwi < krbve AlxB. verse 3,18, late 13th, early 14th 1 2 3 4 567 8 a ζ jich srdce krwe utoöie krwe < krbve AlxV. verse 1517, srdce late 13 th, early q / p. Argumentation is hardly needed in order to understand that nature does not work like that. Certain processes are recurrent (e.g. dental nasals become velar before a velar obstruent, dental nasals become labial before a labial obstruent), others are rare, and some do never occur in natural language (e.g. [s] is turned into [b], dental nasals become velar before a labial obstruent, cf. Ewen & Hulst 2001:3ss on this). Hence, it is wrong for sure to leave both variables A and Β without any restriction. It is also obvious that any context may not provoke any effect: there are precise causalities in nature, and it is not reasonable to assume that a given object may trigger a process X and its reverse. Hence, the relation between Κ and A in A —> Β / Κ is not free but obeys precise causal patterns: given A, Κ may only be chosen among a very restricted subset of all possible objects. And for a given trigger K, only a restricted subset of objects that instantiate A will react.

1.2. One option: fighting against abstractness limits the generative power Both problems were already identified in the ninth chapter of SPE. Chomsky & Halle (1968) believed that the solution would come from markedness. All through the 70s, these issues were rehearsed over and over again (e.g. Foley 1977, Kenstowicz & Kisseberth 1977, 1979). Namely, they underpin the most important debate that was led in early generative phonology: the issue of abstractness. The question "How abstract is phonology ?" was raised in 1968 by Kiparsky (1968) and has provoked reactions that led to the constitution of specific schools of thought both within and outside generative phonology. The latter is represented by Natural Phonology (e.g. Stampe 1972, Dressler 1974,1984, Donegan 1978, Donegan & Stampe 1979), the former by Natural Generative Phonology (e.g. Hooper 1976, Vennemann 1974a,b) and Lexical Phonology (Kiparsky 1982a,1985, Hargus & Kaisse 1993). Anderson (1985:331 ss) provides an informative historical survey of this debate.

What representations are goodfor: the answer of the 80s to overgeneration

373

The discussion of abstractness originally questioned the dissimilarity that phonological theory should allow for between phonetic reality and phonological representations. It led to the view that abstract structures should be dispensed with in the presence of solutions that are bound to the surface in a more direct fashion. This position is expressed in subsequent work by Kiparsky (Kiparsky 1982b contains a number of relevant papers) and for example in Leben & Robinson (1977), Tranel (1981). Natural Generative Phonology aimed at limiting the expressive power of the grammar by restrictively defining what a possible phonological rule is: the True Generalization Condition advocated in Hooper (1976:13ss) states that rules may only make reference to objects that are contained in the phonetic signal (including syllable and pause boundaries). Hence, "natural" generalisations relate one surface form to another, rather than an underlying to a surface form. All phonological rules are thus fully transparent, and any kind of opaque alternation is relegated to suppletive activity in the lexicon. Another outgrowth of the debate on abstractness was Lexical Phonology. In contrast to Natural Generative Phonology, Paul Kiparsky was always concerned with the articulation between underlying and surface representations. A first step that defined what a possible underlying representation is was taken with the Alternation Condition (Kiparsky 1968,1973): in case a morpheme shows no alternation on the surface, it must not be any different in its underlying form. This resulted in a ban against absolute neutralisation. In a further step, Kiparsky (1973) restricts the application of rules in such a way that a certain rule-class may only target derived environments. This was called the Revised Alternation Condition: "obligatory neutralization rules apply only in derived environments". An environment is derived iff it has been produced by either the concatenation of two morphemes or the application of a phonological rule. Hence, the abstract SPE analysis of trisyllabic shortening ( d i v f a j j n e div[i]nity, op[ej]que - op[ae]city etc.) for example is ruled out. In order to account for the fact that trisyllabic shortening fails to apply to monomorphemic items such as nfajjghtingale, [ajjvory and [awjmaha {Omaha), SPE recurs to abstract underlying structures such as /nixt.../ for nightingale: since there is no long vowel in the underlying representation of this word, it cannot be subject to shortening (and ultimately acquires the [aj] by independent rules). When assuming the Revised Alternation Condition, this kind of abstract underlying representations, can be dispensed with because there is a good reason why nightingale and the like remain untouched by trisyllabic shortening: they are underived.

374

11,3 Principles of argumentation II: representations and their function

This move, then, is an indirect way of constraining the abstractness of underlying representations: something that cannot be managed by a rule must be encoded in the lexicon (Rubach 1984:7ss, Cole 1995:73ss and McMahon 2000a:35ss offer more discussion of this issue). Further restrictions on the rule component came from the Strict Cycle Condition which was imported into phonology from syntax (Chomsky 1973) by Kean (1974), Mascaró (1976) and others. Also, the computational part of phonology was modified by allowing word-formation rules to be interspread with phonological rules. In the early 80s, the result of this redefinition of the rule component and its scope has emerged as Lexical Phonology (e.g. Kiparsky 1982a,c,1985, Mohanan 1982,1986, Rubach & Booij 1984, Rubach 1984,1993, Hargus & Kaisse 1993). A more radical response to the abstractness debate was developed by Natural Generative Phonology (Vennemann 1974a,b, Hooper 1974,1975, 1976). In the name of "phonological realism" (Vennemann), alternations that are not one hundred percent automatic (i.e. postlexical in the terminology of Lexical Phonology) were eliminated from the computational component of the grammar. The True Generalization Condition (Hooper 1976:13ss) assured that rules may only make reference to objects that are contained in the phonetic signal. On the other hand, the No-Ordering Condition (Hooper 1976:18ss) prohibited any rule ordering: either underlying representations were related to the surface by only one rule; else, all rules had to apply simultaneously. As a result, everything that could not be described with reference to the surface was excluded from phonological computation (P rules). It was managed by a morpho-phonemic module (MP rules) where processes necessarily make reference to morpho-syntactic categories. In addition, suppletive activity and lexical marking were provided for (via-rules). In sum, thus, Natural Generative Phonology reinstalled the old structuralist principle of level independence: it is strictly forbidden to appeal to any morpho-syntactic kind of information while doing phonology (see Vol.2,3.3.2.7.2 on this issue). In spite of this effort, the debate on abstractness never came to a conclusive end. In particular, no satisfactory definition of what is natural, simple, elegant, phonetically plausible, psychologically real or typologically invariant could be achieved (Hellberg 1978, Koutsoudas 1980, Dinnsen 1980). Neither could an evaluation metric be derived that would be able to measure different degrees of abstractness (Kiparsky 1974, Campbell 1981, Goyvaerts 1981). Proponents of abstract analyses with no other limitation than the learnability of rules (cf. Skousen 1981) are, among others, Schane

What representations

are good for: the answer of the 80s to overgeneration

375

(1968,1974), Hyman (1970), Brame (1972), Dell (1973), Selkirk & Vergnaud (1973), Vago (1973), Anderson (1974), Halle (1977), Dinnsen (1980), Gussmann (1980a), Dresher (1981) and, taking abstractness to an extreme point, Lightner (1978, 1981). Kenstowicz & Kisseberth (1977:1-62) offer extensive discussion of the matter. They argue in Kenstowicz & Kisseberth (1979:204ss) that in some cases, there is no alternative to abstract analyses which are motivated by compelling internal evidence.

308

1.3. Another option: autosegmental representations automatically restrict the generative power Abstractness as a topic of phonological debate has played little or no role in the 80s and further on. The focus was moved to the internal structure of segments, as well as to issues regarding the autosegmental organisation. All approaches that have been quoted so far operated in the 70s and early 80s where multi-dimensional representations did not exist or were in an embryotic state. It became clear in the 80s that they restrict the generative power of the grammar in rather spectacular fashion for the sake of properties that are in-built. A very early and entirely universal feature of autosegmental representations for example is that association lines may not cross. This rules out a fairly remarkable number of phonological processes that would have been well-formed in linear paradigms. Hence, multi-levelled representations had come into being for a given set of reasons (foremost the Coda context and tone), but they turned out to also serve the purpose of fighting back overgeneration. This independence of their raison d'être and the effects they produce were a guarantee against circularity. They could be viewed as a key to the old problem of abstractness: the reason why anything could not be a phonological event is not the restrictive definition of what a well-formed underlying representation is, nor the definition of the formal properties (Strict Cycle) and the scope (derived environments) of the rule component. The crucial property of representations is that in contrast to SPE-type matrices, they may be well- or ill-formed. Simple and entirely universal principles such as the non-crossing constraint on association lines rule out a wide range of virtually possible phonological events. This potential was well understood, and hence early spartan representations were continuously enriched throughout the 80s, leading to rather complex structures. This evolution was parallel to the expansion of arboreal structures in syntax during the same period. Feature Geometry (e.g. Clements 1985,1993, Sagey 1986, Clements & Hume 1995) for

376

11,3 Principles of argumentation II: representations and their function

Clements 1985,1993, Sagey 1986, Clements & Hume 1995) for instance has built autosegmental structures that are self-restrictive in the sense that their arboreal properties exclude quite a number of processes. A more radical way of developing representations was taken by Government Phonology. Kaye et al. (1990:194) ambition at doing away with the rule component altogether: phonological processes apply whenever their triggering environment is met, cf. Kaye (1992b, 1995:291) (more on this in Vol.2,III.3.5.2). As was the case in other attempts at restricting the generative power of the grammar, this approach supposes an important shift from the procedural part of the system towards the lexicon. A typical example is so-called velar softening: the words electri[k] and electri[s]ity contract no derivational relationship of any kind. They represent two distinct lexical entries. In the same way, trisyllabic shortening is banned from the phonological component: opacity [-aes-] is not derived from opaque [-ejk]. The putative structural changes [k] —• [s] and [ej] —• [as] do not qualify as possible phonological events. More generally speaking, much (but not all) of the heart of Lexical Phonology, i.e. phonological processes that apply only under certain morphological conditions, are removed from the synchronically active phonology and placed in the lexicon. The alternations at hand are viewed as the result of suppletive activity, much in the spirit of Natural Generative Phonology's via-rules (Hooper 1976:14ss). Also, derived environment effects are not the result of a synchronically computed phonological process. All these questions are discussed at greater length in Vol.2,III.3.5. Whether this rather radical program can achieve competitive empirical coverage while maintaining relevant phonological generalisations remains to be seen. The only thing that I want to show here is that the enrichment of representations was the specific answer to the problem of abstractness and overgeneration that united phonological research throughout the 80s. The fact that representations cannot be "a little ill- or a little wellformed" was entirely consensual: ill-formedness results in agrammaticality, and cannot be "salvaged" by any operation on the computational side. For example, a representation where association lines cross can result in a grammatical object under no circumstance: there is no such thing as "ok, line-crossing is bad, but X is worse, so line-crossing 'wins'". In short, representations have an intrinsic power on grammaticality. Their rule absolute (it cannot be "outranked") and independent of any computational event.

What representations are goodfor: the answer of the 80s to overgeneration 309

377

1.4. Structure and process: there is no natural science in absence of one or the other202 The following pages examine the role of representations in Optimality Theory, and their relation with the computational part of the grammar. I adopt Anderson's (1985) prism for looking at the evolution of phonological theories: Anderson has observed a movement back and forth between representationally-oriented and process-oriented approaches. This large-scale analysis that spans over the 20 th century (and arguably over more than that), I submit, also provides interesting insights into the more recent evolution of generative phonology, of which Anderson had a strikingly precise premonition: SPE offers much process, but little structure if any; autosegmental representations rebalance the situation in the 80s, but OT is moving back to the "computation is king" attitude of the 60s and early 70s. Anderson (1985) identifies the scope of his book on its first page: "Our intent is to study this history [the history of linguistics] in relation to a particular issue: the balance between rules and representations as components of a theory of language and, more particularly, as components of a theory of sound structure." (emphasis in original) Anderson (1985:1) Throughout the book, he argues for a position which I believe is correct and actually extends to all natural systems (physical, chemical, biological etc.): nature is made of structure and process. Ignoring either one is misrepresenting nature. Anderson writes: "In this work, the history of the balance between the study of rules and the study of representations [...] will be of primary importance. [...] It is not our intention to argue that one sort of consideration is right and the other wrong in a linguistic theory. In fact, theories of rules and theories of representations deal with intimately interrelated and indissoluble aspects of the same linguistic structure. In order to understand the structure, however, both aspects must be appreciated, and this has certainly not always been the basis on which inquiry into sound structure has proceeded." (emphasis in original) Anderson (1985:9s)

202

This section is a written version of Scheer (2003 c). Hyman (2004) has very similar ideas to the ones that are expressed here.

378

11,3 Principles of argumentation II: representations and their function

Writing in the mid-80s, Anderson observes a trend that grants more and more importance to representations, while impoverishing the computational part of the grammar: This, he contends, is a cumbersome evolution. "If current attention to the possibilities of novel sorts of representations leads to a climate in which the importance of explicit formulation of rulegoverned regularities disappears from view, the depth of our knowledge of phonology will in all likelihood be poorer for it." Anderson (1985:350) The last sentence of his book admonishes phonologists, who were then going down the representational road, not to forget about computation: "We hope that this book has demonstrated that neither a theory of rules nor a theory of representations constitutes a theory of phonology by itself." Anderson (1985:350) I argue here that his word has been heard - actually heard too much. Or rather, that it has not been heard at all: by depleting phonology of sovereign representations altogether, OT has arrived at the opposite extreme twenty years later, to the effect that the latter quotation has never been more relevant than today.

310

1.4.1. OT holds that monsters could exist in nature, but accidentally do not occur In OT, the role and the function of representations has been set back in a rather radical fashion. Different and mutually exclusive representations may be taken as the input into the constraint chamber, and will be able to be modified in such a way that both produce the correct surface form. For instance, Optimality Theory has no claim to make regarding the question whether Codas exist, whether they are able to branch one, two, three, ten or twenty times. Any theory of syllabic representation will do, even the one which claims that Codas can branch twenty times. In practice, monsterCodas o f t h a t kind will not survive the application of the constraints including NOCODA, but nothing prevents them from existing in nature. It will be said that a language with Codas that branch twenty times is a possible human language, but that the probability of its actual occurrence is very small: unfortunately, NOCODA is never ranked low enough not to filter out this kind of monster-Coda. Monster-Codas are objects that occur in nature in

What representations

are good for: the answer of the 80s to overgeneration

379

the same right as non-branching Codas or non-branching Onsets; they just never happen to be optimal, and this is a pure accident.

311

1.4.2. Only competition determines (a)grammaticality in OT: nothing can be inherently (a)grammatical One fundamental and inalterable property of OT (as it stands, and as far as I can see) is that whatever structure may exist, it does not exist in absence of competition with other candidates. That is, OT is unable to evaluate the (a)grammaticality of an object or a candidate for its own sake, i.e. in absence of comparison with other objects or other candidates. The logical consequence is the prediction that objects do not possess any inherent degree of grammaticality: they may have a life in nature in absence of competition (e.g. in the lexicon), but they do not possess any inherent property that makes them more or less grammatical (and in any event, nothing is ever ruled out per se, see the preceding section). An OT grammar does not perform any judgement on a form X out of context. It is only when form X is in competition with form Y that one will be worse than the other. It seems hard to believe that this could reflect the nature of language: a Coda that branches seven times is inherently ill-formed. There is no need to compare it to any other candidate: nature does not admit this kind of object, not even a little bit, and not even in an extreme or a highly marked situation. There is no place for Codas that branch seven times in nature. Nature does not need to run this kind of object against anything to understand that it is ill-formed and banned from any human brain. (A)grammaticality is a notion that exists in complete absence of any competition. It goes without saying that this does not mean that there is no competition in grammar. It means that competition is not the only referee who decides on (a)grammaticality. This, however, is precisely one of the central claims of OT.

312

1.4.3. Computation is king: representations are only decorative in OT Another fundamental claim of OT is that constraints and their interaction are the only devices that determine (a)grammaticality. There is no arbitral award from any other source. A direct consequence thereof is that representations are demoted to some kind of decoration which reminds former times, but has lost its function. The very essence of representations is the

380

11,3 Principles of argumentation II: representations and their function

fact that they are a structure, not a process. And that they may be illformed. In OT, computation is king: representations contribute to the arbitral award only indirectly insofar as they are the input into the constraint chamber. Nothing is decided on the sole grounds of representations: it does not make sense anymore to talk about well- and ill-formedness, since nothing is ever ill-formed and thus ruled out. The only arbitral award is relative: a structure is more or less ill- or well-formed in comparison to another structure. There is no property of a representation that could disqualify it once and forever because there is nothing that cannot be violated. For example, "this representation features line-crossing. However, it is the winner because line-crossing is less fatal than X in this particular grammar" is a statement which is not counter-indicated in OT. Representations cannot possibly play an autonomous role in the refereeing process. They may exist, but are necessarily subordinated to constraint interaction (actually they are the result of constraint interaction). Therefore, computation does not operate on representations in OT, but with representations. The irrelevance of representations in OT is also well documented by the fact that they are freely exchangeable: did they matter, opposite and mutually exclusive representations could not possibly produce the same result.

313

1.4.4. Since they do not matter, representations are arbitrary and interchangeable in OT: they are doomed to disappear Since representations are disqualified from contributing a sovereign and unoutrankable award to the refereeing process, research on their exact properties has not been on the menu of mainstream phonological endeavour in the past ten years. In the introduction to a volume on Distinctive Feature Theory, Hall (2001) for example writes: "A consequence of the shift away from representational questions [...] is that there is at present much uncertainty concerning certain fundamental questions pertaining to [...] phonological representations [...]. With respect to features, the most obvious question [...] is: What featural representations (e.g. feature geometry, underspecification) are necessary in a phonological theory?" Hall (2001:1)

What representations are goodfor: the answer of the 80s to overgeneration

381

In the same volume, Clements (2001) draws a similar picture. "One less desirable consequence [of the rise of constraint-based phonological theories] has been an increasing uncertainty regarding such fundamental questions as: What is a lexical representation? What is a phonological representation? Of what features or feature specifications do they consist? How do these features combine? What is the trade-off between constraints and representations in understanding phonological regularities?" Clements (2001:71) The kind of representations that are known from the times before their function was taken over by constraints are still practised, but they are hardly questioned or renewed. Had they been different in the 80s, these different representations would have received an OT-top in the same way syllables, feet, features etc. did, and the result would have been the same. O T does not argue in favour of any particular representation, nor can it admit that a representation of any kind is incompatible with the theory: any representation and its reverse may be accommodated within OT, and produce the "correct" result. The arbitrariness of everything that has to do with the content of phonological representations is even open-heartedly admitted. Linda Lombardi writes in her introduction to a volume that is concerned with the status of melody in OT: "The tenets of OT, regarding constraint violability and ranking, make no particular claims about phonological representations. We could, for example, do OT with any kind of feature theory: SPE feature bundles or feature geometric representations, privative or binary features, and so on." Lombardi (2001:3) It is hard to see how any serious theory could happily rely on interchangeable representational assumptions. Addressing this issue, Lombardi writes on the same page that "much early work in OT paid little attention to representational questions, simply taking over assumptions from previous work in derivational autosegmental phonology. But representational arguments are theory-internal and need to be reexamined in light of fundamental theoretical changes; the choice of correct representations to use in OT analyses must be based on arguments couched in OT terms." (emphasis mine) Lombardi (2001:3) Hence Lombardi contends that there is such a thing as correct representations in OT, i.e. which are not interchangeable. It is certainly true that

382

11,3 Principles of argumentation II: representations

and their function

representational arguments are theory-specific. This statement, however, supposes that the theories in question admit the sovereign arbitral award of representations. How could a theory like OT, which does not, tell correct from incorrect representations? Also, the book at the outset of which Lombardi makes this statement ranges over 300 pages, but representations are fairly hard to detect indeed. Instead of autosegmental or any other kind of structure, constraints such as * DORSAL, *LABIAL, *[+cont] etc. are used in order to describe and manipulate melodic properties. It has been pointed out to me that objects of the kind *DORSAL, or the feature [dorsal] itself, could be said to be representations. This depends on the semantics that one is prepared to grant to the word "representation". I cannot see any meaningful way of talking about objects as representations if they cannot be ill-formed: something is a representation iff if can be ill-formed. Obviously, the feature [dorsal] cannot. By contrast, a feature geometric or a syllabic tree can. In short, only autosegmental representations or equivalent structures can be deemed representations. Representations of former times are used in OT not because they have any function in the theory, but because they are there, and people have been brought up with them. Their selection is arbitrary, and they are interchangeable. The movement towards * DORSAL and the like eliminates parts of the representational heritage of the 80s. This is only consequent and logical: if a tool has lost its function, it must be replaced. Living with an old emptied shell that duplicates the new carrier of the function makes little sense. This is the direction in which OT is evolving, and in which it should evolve: representations will disappear altogether.

314

1.4.5. Back to where we started: representations were the enemy of overgeneration One obvious consequence when representations are dysfunctional ised is that their job is undone. It was pointed out earlier that representations were the specific answer of the 80s to the problem of overgeneration. If nothing is ruled out anymore because of ill-formed representations (there is no such thing as ill- or well-formedness in OT, things are only more or less illformed), phonology is set back to where it stood in post-SPE times. That is, just as SPE, OT can accommodate all linguistic data that are found on this planet without exception, and all the others as well. There is not a thing that could not be described by OT, or that would be declared not

What representations are goodfor: the answer of the 80s to overgeneration

383

to be able to exist in natural language. Given that everything is only relative in OT, one ought to add: "able to exist not even a little bit, not even potentially, not even with the slightest probability". This clarification is in order because the very essence of OT is to never make any discrete statement: candidates are not ill- or well-formed by themselves; they are only more or less well-formed in regard of other candidates (§311). Given a set of 50 universal constraints (which is an extremely conservative number: McCarthy 2002:305s lists 102 constraints in his index, which for sure does not exhaust all constraints that are found in the literature), and supposing free ranking, factorial 50 grammars are generated. Even if rankings that produce the same grammar are counted out, 203 it may be doubted that a computer screen is able to represent this number without scrolling down. 204 This is what McMahon (2003) calls "an embarrassment of the riches". Strangely enough, it really seems necessary to rehearse the same arguments related to overgeneration that were made in post-SPE times, i.e. the days when most of the major OT voices were students. Therefore, McMahon (2003) recalls that "power corrupts", and provides illustration with a quote from Lass (1984) on absolute neutralisation. "The more powerful a theory is, the less responsive to empirical checks. [...] As for 'power' itself, in the extreme case, if anything can be anything, then you end up with nothing." Lass (1984:211)

203

204

Whose exact number of course is impossible to calculate since it depends on the formulation of the individual constraints, which is subject to fluctuation. Simulations on a typical constraint set with a restricted number of constraints seem to suggest that around 70% of the grammars that the factorial mechanism produces are identical. OT is sometimes said to particularly suit dialectal variation: "there are 50 dialects, and you just need to rerank these X constraints in order to get them all." No other theory can do that. This may be true. But no other theory can produce the 500 other systems that are not attested in the dialectal space either. Is it any wonder that a theory is able to cover 50 different dialects if it generates hundreds of systems that may or may not exist? There are two approaches to parametric variation: either the theory generates everything and tries to restrict the generative power afterwards (SPE and OT), or the theory is (over-) restrictive in the first place and tries to implement parameters in the face of variation. Throughout this book, I pursue the latter strategy (e.g. in chapter 11,12 §524, see appendix 1 §620).

384

11,3 Principles of argumentation II: representations and their function

The same observation is commented for example in M c M a h o n (2000b) and Hulst & Ritter (2000:282s).

315

1.4.6. OT tools for fighting back overgeneration: markedness and grounded constraints A s far as I can see, the question of overgeneration is addressed under two headings in OT. Or rather, two devices in O T are responsive to the issue of overgeneration, even though this is probably not their raison d'être in the mind of OT-practitioners. The devices at hand are markedness and so-called grounded constraints.

316

1.4.6.1. Markedness and O T The idea of markedness is that those properties of language which are cross-linguistically frequent are theoretically more relevant and should be encoded by the theory (in a more prominent way). This idea, of course, is as old as the study of language. Its specific impact on overgeneration was recognised in the ninth chapter of SPE, and developed in various generative 90S theories thereafter. O T is the most faithful modern heir of the direction that was given by SPE (and not only with respect to markedness, see Hulst & Ritter 2000) in the sense that it grants a central role to markedness probably the central role in the theory. 206 The specific way of implementing markedness in O T is some kind of tacit understanding a m o n g practitioners to always formulate constraints according to cross-linguistic statistics. For instance, the most important constraints regarding syllable structure are ONSET and NOCODA. Exactly the same result with the same winning candidates could be produced with the reverse constraints NOONSET and CODA. There is no property of Optimality Theory that requires the former formulation, rather than the latter, 205

206

E.g. in Natural (Generative) Phonology, Government Phonology as far as melody is concerned ("what is a possible vocalic system?"), and in the work of Brandäo de Carvalho ( 1994,1997,2002a,b). Contrary to this tradition, there have always been voices which argue that markedness is entirely irrelevant to linguistic theory: grammar defines what a possible language looks like, not whether a possible object is more or less probable. Arguments along this line are put forth for example by Lass (1984:278s), Hale & Reiss (2000a,b) and Newmeyer (1998).

What representations are goodfor: the answer of the 80s to overgeneration

385

and both can be made empirically equivalent. The only reason why ONSET and NOCODA are favoured is cross-linguistic observation: most languages "want" to have Onsets, while they "are reluctant" to accept Codas. If the reality observed were the reverse, i.e. if many languages showed Codas but no Onsets, O T would work with the reverse set of constraints. 207 This way of recording the observational situation cuts down overgeneration insofar as it makes non-occurring or impossible monster systems improbable (though not impossible, see §310). A related issue is the so-called "emergence of the unmarked" (abbreviated T E T U , McCarthy & Prince 1994), which is advertised as an achievement of OT. In short, it is argued that a theory which automatically produces unmarked patterns probably reflects the linguistic reality quite closely. Or, in other words, O T is able to predict what is marked and what is unmarked, especially when higher ranked constraints are unable to determine the optimal candidate and lower ranked markedness constraints, although dominated, make the unmarked candidate win. I have never been able to understand this argument, which is used at a large scale. Its complete circularity is overt: if markedness is fed into the constraints in the first place, how could these constraints produce anything else than the markedness pattern that was programmed? The only thing that "emerges" in T E T U is what the phonologist who has formulated the constraints has fed into the system beforehand: the cross-linguistic observation. A system that works with NOONSET and CODA will declare languages with Codas but without Onsets unmarked. Is that any surprising? H o w can an observation that is built into a system be interpreted as a prediction?

317

1.4.6.2. The fundamental source of arbitrariness in OT: anything and its reverse can be a constraint The other device that helps fighting back overgeneration in O T are socalled grounded constraints (e.g. Hayes 1996, Beckman 1997,1998, Boersma 1998, Kirchner 2000, Bermúdez-Otero in press). Before looking at grounded constraints themselves, this section describes a fundamental crux of OT, which is closely related to grounded constraints. That is, any-

207

Hence, nothing in the theory prevents from declaring human a system which is non-human in every respect. Linguistic UFOs may be happily accommodated. Or rather, there are no UFOs at all: any logically possible system could be human.

386

11,3 Principles of argumentation II: representations and their function

thing and its reverse can be a constraint: the formulation of constraints is unconstrained. In my opinion, this is the fundamental source of arbitrariness in OT (see for example Angoujard 1997 on this issue). It is a fact that anything which can be written down in prose may be a constraint in OT. There is no formal or semantic property of any utterance that would be ruled out by some provision of OT. OT is a theory of constraint interaction, and these two words truthfully characterise the whole enterprise: given a set of constraints, their various ways to interact produce phonology (or syntax, or chemistry...). OT is not a theory of constraints. It is obvious that the way nature functions, the effects of this functioning that are visible and the human perception thereof are three things that may be light-years away from one another. For example, the description of physical, chemical or biological events with simple vocabulary may produce something that is entirely satisfactory for the purpose of observation. In all likelihood, however, it will have nothing to do with the natural mechanics that are at work. Therefore, natural scientists have always tried to establish the specific categories of their field: elements, electrons, quarks and so forth. Their theories, then, refuse to use anything different from these categories. The universe of discourse of physicists is thus settled: only physical objects qualify. Since OT is a theory of constraint interaction, but not of constraints, there is no universe of discourse: nothing prohibits reference to nonlinguistic items. Practitioners of OT actually make a large use of this license: constraints are regularly formulated according to very general human properties that are not specifically linguistic: LAZY is a typical representative of this class. While it may seem plausible that humans refrain from wasting energy, the linguistic effects that are allegedly caused by laziness may have very different mechanics: linguistic mechanics. In its present-day shape, thus, OT imposes no constraint on how a constraint may look like. For the reasons exposed, this is not a desirable state of affairs. If on the other hand some auxiliary theory defines what a possible and what an impossible constraint is, this theory is necessarily different from OT. It necessarily stands outside of OT, hence it must not be subjected to the refereeing process that is carried out within the OT module. This is the essence of a discussion that I had with Marc van Oostendorp in July 2003. In other words, OT needs to be couched into a universe of discourse that defines its properties from the outside before any OT-internal mecha-

What representations are goodfor: the answer of the 80s to overgeneration

387

nism is launched. 208 Representations are one possible universe of discourse. As was required earlier, their status would then be autonomous with respect to the constraint chamber, and their arbitral award would be absolute. Markedness is another possible universe of discourse. It is a fact that OT constraints are formulated according to markedness - or at least, they are not formulated against markedness. But as far as I can see, this has never been an explicit requirement, and certainly does not follow from any property of the theory. Practitioners of OT follow markedness tacitly. This is actually true for all other properties of OT constraints. Hale & Reiss (2000a: 166s, 2000b: 179ss) point out the same fact in regard of the OT tendency towards functionalist constraints: "In fact, there is no necessary connection between OT as a theory of computation and functionalist reasoning, and an OT proponent might invoke what we call the National Rifle Association defence ('Guns don't kill people; people kill people'): computational theories are not inherently functionalist, people are functionalist." Hale & Reiss (2000b: 179) That is, formulating constraints according to markedness or not, implementing functionalist ideas or not, is a private decision of the analyst. The theory itself is fine with constraints following markedness and functional ism as much as with those that encode anti-markedness and dysfunctionalism. Hale & Reiss (2000b: 179ss) for example show that dysfunctional constraints such as " O B F U S C A T E : merge contrasts, use a small inventory of distinctive sounds" and "No P A I N - N O G A I N : maintain contrasts, use a large inventory, generate allomorphy", if appropriately ranked, produce the same empirical result as their functional counterparts. Therefore, the absence of constraints on constraints is also the source for the interchangeability of statements and representations that was discussed earlier (§312s). It is surely a motivation for the (more or less recent) popularity of grounded constraints as well.

1.4.6.3. Grounded constraints Given the situation described in the foregoing section, many OT practitioners feel the need to be able to tell good from bad constraints. Typically, overarching constraints that encode some very general human property and 208

Oostendorp & Weijer (forth) make specific proposals how OT could be inserted into a universe of discourse.

388

11,3 Principles of argumentation II: representations

and their function

are void o f specifically linguistic content are the "bad" ones: LAZY, •EFFORT, *STRUCTURE and the like.

This reasoning replicates the situation of ordered rules in the late 70s: any existing and all non-existing processes could be described because anything and its reverse could be a rule. That is, Χ, Y and Κ in X —» Υ / Κ could be instantiated by anything irrespectively of the values of the other variables (see §306). For this reason, people were eager to be able to tell "good" from "bad" (or so-called "crazy") rules. The idea of grounded constraints is that their "grounding" is a guarantee against craziness, i.e. against constraints such as "velar nasals must be followed by a labial" or "delete a vowel every time you blink". Now what does "grounded" mean? What is the difference between a grounded and an ungrounded constraint? The best answer that I can come up with is the following: grounded constraints enjoy an "independent" motivation, while ungrounded constraints do not. And what does "independent" mean? Independent means "extra-phonological". Following this line of reasoning, grounded constraints make reference to phonetic, psycho-linguistic, perceptual, effort-based or functional findings. In short, they replicate a non-phonological pattern. Proponents of grounded constraints explicitly argue that it is a good thing to be grounded in any event better than not to be grounded. What they do not claim explicitly is that grammar should contain no ungrounded statement at all - but this stance is implicit. If it is true that grounding is desirable, it follows that for any competition between a grounded and an ungrounded alternative, the former is selected. This amounts to nothing less than the complete evacuation of phonological causes from phonology. If we deplete phonology of phonology, what are we left with? We are precisely left with "grounding": phonetics, psycho-linguistics, perception, functionalism, the least effort and the like. In other words, the grounding philosophy leads to reject phonology as a cause of phonological events. Extra-phonological explanation is superior to phonological explanation. In this world-view, phonology is nothing more than an appendix of phonetics, anatomy, functionalism etc. Hayes' (1996) "inductive grounding" is an incarnation of this position. "It is reasonable to suppose [...] that virtually all of segmental phonology [...] is driven by considerations of articulatory ease and perceptual distinctness." Hayes (1996:14)

Phonology as such, then, ceases to exist.

What representations

are good for: the answer of the 80s to overgeneration

389

While nobody can deny the influence of non-phonological factors on phonological phenomena, it is hard to see how phonology could be done, or why it should be done, without appeal to phonology. "Grounded" competitors are not superior to ungrounded alternatives per se. Evaluating which part of the pool of phonological events is due to extra-phonological causes is a difficult and painstaking task that has been undertaken for more than a hundred years. Most probably, it has not reached a final conclusion with the advent of OT. Phenomena in physics are also conditioned by all kinds of extraphysical parameters, which may be biological, chemical etc. Would physicists try to explain physical phenomena by extra-physical causes before having tried physical explanations? Would anybody believe that extraphysical explanations are superior to physical ones per sel One case in point where the run on grounded constraints eludes relevant data has been discussed in §132. For grounded psycho-linguistic and functional reasons, Beckman (1997,1998) holds that the beginning of X is strong, while the middle of X is not, where X can be the word, the syllable, the morpheme, the root, the stem, the foot, the prosodie word or any other relevant phonological unit. This exclusively extra-phonological perspective dismisses the existence of phenomena whereby objects are strong not only at the beginning of X, but also word-internally in a specific syllabic position, i.e. after a Coda. The Coda Mirror context {C,#} therefore goes unnoticed on grounded grounds. Representations are incompatible with the grounded philosophy precisely because they are phonological, and only phonological. The only way to argue for or against a representation is to do phonology. Phonetics, psycho-linguistics, perception, functionalism, the least effort and so forth have no bearing on representations at all. Given the overall picture, then, it is a fair assumption that there is a causal relation between the demotion of representations to a decorative ghost-ship of the 80s and the development of the grounded philosophy: extra-phonological alternatives are sought because the possibility of theoryinternal and truly phonological explanation is lost. The title of Brandäo de Carvalho's (2002b) article is rather instructive in this light: "Formallygrounded phonology: from constraint-based theories to theory-based constraints". That is, Brandäo de Carvalho suggests the opposite of what has happened in OT: constraints need to be constrained by some theory, rather than by extra-phonological patterns. Also note that his statement implies that constraints and theory are two notions which must not be co-extensive.

390

11,3 Principles

of argumentation

II: representations

and their function

Finally, it is worthwhile pointing out that people who follow the research strategy of grounded constraints do not stand on Saussurian grounds anymore. Grounded causes are necessarily extra-phonological, which means that they recur to properties of the Parole. Not only does Saussure (1915) claim that there is a linguistic (phonological, syntactic etc.) core which is entirely independent and uninfluenced by the Parole. He also asserts that the only possible research strategy is to study the Langue with no regard to the Parole. Trying to get a handle on the Langue by looking at the Parole first is necessarily doomed to failure. "La Langue, distincte de la Parole, est un objet qu'on peut étudier séparément. [...] Non seulement la science de la Langue peut se passer des autres éléments du langage, mais elle n'est possible que si ces autres éléments n'y sont pas mêlés." Saussure (1915:31) [The Langue is distinct from the Parole. It is an object that can be studied in its own right. [...] Not only can the science of Langue ignore the other elements that are constitutive of speech; the study of the Langue is indeed impossible if these other elements are taken into account.]

319

1.4.6.4. If phonological events have exclusively non-phonological (grounded) causes, why replicate them as constraints ? Let us now consider another consequence of inductive grounding. We have seen that the grounded philosophy looks out for extra-phonological causes when explaining phonological patterns. Since a competition between a grounded and an ungrounded explanation will always prompt the former as the winner, there is a natural drift towards the absence of phonological explanation for phonological phenomena. If it is true that extra-phonological mechanisms are responsible for the facts observed, however, the following general question arises: why should they be replicated in the form of constraints in the first place? If phonetics, functionalism, psycho-linguistics and the like produce the phonological patterns "for free", what is the job of the constraints? Why should a set of phonological constraints retranscribe the activity of non-linguistic mechanisms of the brain and the body? Shouldn't we simply face the ultimate consequence: on the assumption of inductive grounding, there is no phonology left at all? I can see only reason to maintain any phonological apparatus at all in this landscape: cross-linguistic variation. The various kinds of extra-

What representations are goodfor: the answer of the 80s to overgeneration

391

phonological pressure are the same for all languages and for all speakers, but grammar decides how much each individual conditioning weighs in a particular language. For example, phonetic pressure may be ranked high here but low there, while lexical access may be important in this, but not in that language. What we are left with, then, is a system where 1) there are universal properties of language that produce phonological patterns, but 2) they are not phonological; 3) the only phonological apparatus that remains is constraint interaction, which is responsible for cross-linguistic variation. In other words, OT would boil down to the parameter function of a classical Principles & Parameter framework with the peculiarity that the principles lie outside of phonology. It may be asked whether the bearing of grammar on sound may be, or should be, reduced to this narrow action.

320

1.4.7. Summary: structure and process have an independent existence in nature, but not in OT The foregoing pages have examined the role of representations in Optimality Theory. It was shown that they are still present physically, but do not carry any function anymore. In OT, representations do not contribute any sovereign arbitral award to the refereeing process that decides on (a)grammaticality. Different and incompatible representations are interchangeable, rather than standing in a competing relationship. No matter what their properties, representations X and Y are suitable as long as they produce the "correct winner". This is parallel to the status of grammatical tools in SPE and the early 70s (see Hulst & Ritter 2000). On the bottom line, representations in OT are only decorative. OT has eliminated structure in favour of computation, and it seems to me that this is a straightforward consequence of its very core. Anderson (1985) has shown at a larger scale that phonological theories of the 20th century (and arguably in continuity of the 19th century) may be classified in a meaningful way according to the balance between structure and process. It appears that the micro-evolution inside generative phonology reproduces this pattern: autosegmentalism has installed a counterweight to the computation-only approach of SPE. Among other things, autosegmental representations served to fight back overgeneration. OT undoes autosegmentalism (de facto, even if not (yet) in print), a move that has a number of consequences: 1) we are back to SPE-type over-

392

11,3 Principles of argumentation II: representations and their function

generation; 2) the judgements that grammar performs on linguistic objects are not absolute and discrete, but only relative (in regard of other competitors) and gradual (no object is banned from nature per se, including monsters); 3) predictions are only probabilistic, never on a hit-or-miss basis; 4) phonology is more and more depleted of phonology: the ultimate causes of phonological events lie outside of phonology (drift towards grounded constraints). These consequences should be given a second thought: do we really want phonology to have this face? An alternative (actually the classical) view holds that (a)grammaticality is the result of the tension between two modules that are necessarily different and autonomous: structure (i.e. representations) and process (i.e. computation of whatever kind: ordered rules, constraint interaction etc.; the nature of computation is an independent debate). 209 Representations are primary and autonomous phonological objects. They are not the result of the computational system: they do not emerge, they are. Their arbitral award cannot be undone or outranked by the computational system. Grammaticality is not exclusively relative: natural objects do experience a grammatical judgement in complete absence of any alternative candidate. Among other things, grammar is about to define what possible natural objects and events are, and what are not. Monsters do not exist, not even a little bit, and they could never be produced by any human being, not even in whatever exceptionally marked situation. Note that this does not mean that there are no constraints, that there is no constraint ranking, that there is no competition in grammar or that there are no extra-phonological causes for phonological events. It just means that there are representations, and that they are central, rather than decorative. Also note that OT has probably overstated the role of computation in grammar, neglecting its structural properties. It is certainly true, however, that Government Phonology, on the other hand, does not possess a balanced record either: quite some things have been done on the structural side, but the theory is almost virgin of any reflection regarding phonological processes. As for a theory of computation indeed, Government Phonology has not much to offer other than Kaye's (1992b: 141) statement "processes apply whenever the conditions that trigger them are satisfied". The arguments that have been put forth here are directed towards OT as it stands. I believe that the absence of sovereign representations is noth209

This is also the basic distinction made in the work of Hale & Reiss (2000a,b).

What representations are goodfor: the answer of the 80s to overgeneration

393

ing that follows from the basic claims made by OT (although it is a natural consequence thereof). The fundamental crux of OT is the fact that nothing constrains constraints. I argue that representations could play this role, as they did in the late 70s when they cut back the computational power of SPE. That is, they could be the universe of discourse in which constraints evolve, and at the same time acquire the status of a device that issues an unoutrankable arbitral award.

1.4.8. Shifting burden over to GEN: a promising track to follow ? The architecture of OT, however, offers a "natural" alternative to the rehabilitation of representations. That is, all universal properties of sound structure that I am talking about (for example the non-existence of monsterCodas and other UFOs) could be built into GEN. They probably could indeed, but this does not appear to be a priority of OT practitioners. OT is spending much ink, time and effort to invent opacity-killers (with results that are not exactly encouraging), but shows far less concern for the problems that are raised by its core architecture and scope. The functionalisation of GEN in the sense described would require the following attitude: "we have to deplete the constraint chamber of many things over which we thought it rules. Here is what we take out, here are the criteria that define what is taken out and what remains; and here is how GEN will have to look like after the recalibration of its function." We then can expect a revival of the "old" debates that have been called for on the foregoing pages: building this or that representation into GEN will be absolutely crucial since GEN will produce either X or Y, not both. X and Y will not be interchangeable, and hence require an independent debate as in the old days: proponents of X will have to argue with the proponents of Y without any reference to constraints and their interaction. But even if there is a general redefinition of GEN, the division of labour between GEN and the constraint chamber will, in all likelihood, be difficult to operate. In the original version of the OT architecture, constraints and their interaction were supposed to take care of everything that is cross-linguistically variable, while universal properties of sound structure were directly built into GEN (Prince & Smolensky 1993).210

210

For example, Prince & Smolensky (1993:5) write "Gen contains information about the representational primitives and their universally irrevocable relations", but there was some early debate on the distribution of labour between

394

11,3 Principles of argumentation II: representations and their function

Also, a functionalisation of GEN along these lines still leaves us with a sizable amount of overgeneration: many properties of language do not occur in factorial variation. Rather, they only appear in two, three or four different coats. For example, there are only two patterns of vowel-zero alternations, Havlik and Lower (see §468). Disregarding the many factorial and logical possibilities, it appears that nature has not produced any third pattern. However, free ranking of all constraints, including the ones that are not designed for the specific purpose at hand, will produce much more variation. I therefore doubt that the original criterion "variable properties for constraints, universal properties for GEN" can work out in the general landscape of free ranking and factorial typology. If this is true, however, it is hard to see in which way OT and classical Principles & Parameters are different: GEN generates candidates according to universal principles and language-specific parameters (the ones that produce the limited two- or three way distinctions). The resulting candidate set is then run through a constraint chamber that handles those parameters which describe a high number of cross-linguistic alternatives according to factorial typology. In any event, all this is only speculation done by somebody who does not have a competent look at the OT-scene from the inside. Perhaps some OT practitioners take up the issue and pose conditions on GEN. The new trade-off between GEN and the constraint chamber will then show whether OT is really so radically different from all preceding theories as is suggested by its self-understanding.

322

1.4.9. Inventing and turning wheels

If the foregoing pages make any sense at all, phonology over the past decade will have turned back the wheel just in order to go through the historical evolution since SPE again. Anderson's (1985) prism thus has not only proven to be a meaningful descriptive tool of the past at the time when Anderson wrote. It was also premonitory for the 90s and 00s. In line with Durand & Laks (1996) and Encrevé (1997), a fair case for cumulativity in phonology and elsewhere can probably be made: reinventing the wheel every now and then consumes quite some energy, ink and time. When John Anderson received a Honorary Doctorate from the university of Toulouse in

GEN and the constraint chamber: McCarthy & Prince (1994:4, note 5) also allow for the presence of non-universal information in GEN.

What representations

are goodfor: the answer of the 80s to overgeneration

395

July 2003, he put it this way: "we'll be a lot longer discovering the future if we don't recover the past".

323

2. Structure and process: the take of Government Phonology

324

2.1. Representations that are (en)rich(ed) enough to support parameterisation The foregoing historical excursion serves the purpose of locating the approach that is defended in this book. CVCV is a development within Government Phonology, and as such privileges the explanatory and restrictive virtue of representations. Government Phonology in general and CVCV in particular are sometimes said to be representationally oriented (e.g. Hulst & Ritter 1999a). This characterisation is appropriate. A number of arguments that are made in this part of the book crucially rely on 1) the existence of representations and 2) specific properties thereof that are not interchangeable. Government Phonology in general and CVCV in particular increase the representational device: there are more constituents than in classical syllabic approaches (final empty Nuclei and empty Nuclei in alternation sites in Standard Government Phonology, empty Onsets in the middle of long vowels, empty Nuclei enclosed within geminates, empty Nuclei after internal Codas, empty Nuclei in the middle of branching Onsets in CVCV), and there are lateral relations (Government and Licensing) that take over the functional load of traditional arboreal structure. The increased richness of representations in Standard Government Phonology and CVCV is actually a major feature that crafts the identity of these approaches and sets them apart from other models (another genuine property is the latéralisation of phonological structure and causality, cf. chapter 1,8 §165). Description and explanation in Standard Government Phonology and CVCV crucially rely on representational devices such as empty Nuclei that other theories do not endorse. CVCV goes one step further on this path than Standard Government Phonology: there are more empty Nuclei and more empty Onsets, the lateral devices Government and Licensing are distinct and clearly defined according to their effect. In this part and throughout the entire book, arguments in favour of CVCV are made on the grounds of these representations, which are rich enough to allow for parts of their body to be parameterised.

396

11,3 Principles of argumentation II: representations and their function

In order to illustrate what this means in actual practice, two examples below briefly anticipate on arguments that will be made in this second part of the book.

325

2.2. Internal and final Codas: different but still the same In chapter 11,12 (§524), it will be shown in greater detail that internal and final Codas sometimes provoke identical effects, but at other times do not. The same holds true for internal and final closed syllables. In systems where laterals are vocalised for example, both word-final and preconsonantal laterals may be affected (Brazilian Portuguese), or only those that occur in internal Codas (French). The reverse case where final Codas show an effect, but their internal peers do not, seems to be absent from natural language. 211 Hence, theory is called to be able to both account for the uniform and the contrasting behaviour of the two kinds of Codas. In traditional terms, the Coda is referred to as a syllabic constituent that is defined by its position in the arboreal structure of the syllable. In case both Codas show identical behaviour, the effects observed are ascribed to the "weakness" of this specific constituent, which is not explained but merely noted (there is no answer to the question "why are Codas weaker than Onsets?"). This descriptive adequacy is lost when the effects produced by both Codas are not in line. In this case, final Codas are usually declared extrasyllabic, to the effect that 1-vocalisation in French for example can be described as "occurring in the Coda": there are no word-final Codas since word-final consonants are not syllabified. The same holds true for final closed syllables that do not participate in the phenomenology observed for their internal peers: if word-final consonants are extra syllabic, there are no word-final closed syllables. For the reasons exposed in chapter 11,5 (§339), extrasyllabicity is not considered an option here. In its absence, Standard Government Phonology is in the reverse situation with respect to classical syllable structure: all word-final consonants in all languages are interpreted as Onsets of an empty Nucleus (Kaye 1990a), while preconsonantal consonants are true Codas (provided they obey the usual sonority pattern and there is no vowelzero alternation to their right). Hence, a prediction is made to the effect that both word-final and preconsonantal positions never provoke the same ef-

211

Except of course eternal final devoicing. See note 340 on this issue.

Structure and process: the take of Government Phonology

397

feet: they do not share the same syllabic identity. This prediction is obviously wrong. CVCV accepts the fact that word-final consonants sit in an Onset of an empty Nucleus. But in addition, it holds that there is an empty Nucleus after preconsonantal consonants as well, and hence takes a step that enriches representations. Therefore, we are back to the dilemma of classical syllable structure: in CVCV, both Codas are defined as "consonants that occur before a (governed) empty Nucleus". They are thus identical with respect to constituency. How come they can behave in different fashion, then? There seems no way to have your cake and eat it. Unless the effects that are observed in Codas are no longer ascribed to constituency, but to the lateral relations weaved among constituents. This is precisely the research programme of CVCV: the latéralisation of structure and causality (chapter 1,8 §165). According to the Coda Mirror, a consonant is in a weak position called Coda iff it is neither governed nor licensed. Obviously, the identical or contrasting behaviour of final Codas is a parameter that differentiates phonological systems. The dilemma mentioned comes from the fact that constituent structure can hardly be parameterised: word-final consonants cannot be said to be syllabified into Codas in one language, while ending up in Onsets in another. By contrast, nothing prevents lateral relations from being parameterised: different nuclear categories may have different lateral actorship. This is known to be true for final empty Nuclei for a long time (they can do more than their internal peers, see §§61,266 and chapter 11,12 §524). It is also true for schwa (i.e. vowels that alternate with zero, see §§266,553 and chapters 11,9 §411,11,10 §426). Hence, we can maintain that both Codas are Onsets of an empty Nucleus. Only do final and internal empty Nuclei not necessarily possess the same ability to govern and to license. Internal empty Nuclei can act as governors and licensors under no circumstances, while final empty Nuclei may or may not govern and license. This kind of analysis cannot be developed unless it is understood that phonological phenomena such as lenition and fortition are not a consequence of constituency itself. Rather, they are due to the lateral relations that constituents contract. And it supposes enriched representations: no parameters concerning the lateral actorship of empty Nuclei could be set if these empty Nuclei, both after internal and final Codas, did not exist. Hence, representations need to be rich enough in order to be able to support their own parameterisation.

398

11,3 Principles of argumentation II: representations and their function

326

2.3. Vowel-zero alternations: enriched representations avoid absolute neutralisation and allow to express the parameter Havlik vs. Lower

327

2.3.1. Slavic vowel-zero alternations: the basic pattern In chapters 11,9 (§411) and 11,10 (§426), Slavic vowel-zero alternations are discussed at some length. For reasons that are exposed there, the classical generative analysis ascribes the alternating pattern to the presence or the absence of a particular kind of vowel in the syllable that follows the alternation site. Consider the fragment of data displayed under (264).212 (264) typical Slavic vowel-zero alternations open syllable zero C_C-V C C-yer C-0 dom-eö-ek Czech dom-0k-u Slovak kríd-01-ο kríd-el-iec Polish bul-0k-a bul-ecz-ek

1

closed syllable vowel C_C-CV e /

399

Co i

It is assumed that vowels which alternate with zero are so-called yers at the underlying level. Yers are transcribed as Iii. Lower thus reads "a yer is turned into if and only if it is followed by another yer". Hence, both alternation sites of dom-ec-ek are vocalised because they are followed by a yer: the one enclosed by and precedes another alternation site whose underlying identity is a yer. This second yer is itself followed by another yer, which is word-final. We are thus faced with a chain of three successive yers at the underlying level: /dom-iô-ik-i/. Cyclic application of Lower produces the correct result dom-ec-ek, whereby the word-final yer (which enjoys a morphological identity: it is the NOMsg marker) triggers the vocalisation of the alternation site located between and . In order to achieve this surface form, however, this word-final yer must disappear later on, since it is inaudible.

328

2.3.2. Two kinds of yers

We are therefore in presence of an absolute neutralisation: the object "yer" Iii that is supposed to exist underlyingly never appears on the surface as such. Its phonetic manifestations are either or nothing. Therefore, the classical account appeals to two rules: Lower and a context-free rule of yerdeletion that removes any "surviving" yer, whereby the former is ordered before the latter. The surface contrast "phonetic absence vs. presence of yers" is thus due to the rule component, while both values amount to a single object at the underlying level: Iii. However, there appear to be two kinds of lil\ one whose derivational fate depends on contextual parameters (followed by another III or not), and another one that reaches the surface in no event: its phonetic manifestation is always zero regardless of the context. The former are yers in alternation sites, the latter are word-final yers (or yers with a morphological value). In chapters 11,9 (§411) and 11,10 (§426), I call the former "alternating", and the latter "final" yers. Terminology is not important here, but these labels stand for the salient properties of both species of lif: alternating and final yers both trigger the vocalisation of a preceding alternation site, but vowel-zero alternations appear only in place of alternating yers. It is also shown in chapters 11,9 (§411) and 11,10 (§426) that there

400

11,3 Principles of argumentation II: representations and their function

is much more empirical evidence that calls for a differentiation between both kinds of yers. On the classical analysis, this contrast is expressed by the rule component in such a way that two different rules apply: Lower turns III into under certain contextual conditions, and yer deletion then removes all yers that have not been transformed by Lower.

2.3.3. The introduction of representations allows to get rid of the absolute neutralisation The picture is further complicated by the fact that the alternating quality of an cannot be told from its phonetic properties: there are alternating [e]s such as in domek vs. domak-u, and non-alternating [e]s such as in les vs. lesa "forest NOMsg, GENsg". In linear accounts of Slavic vowel-zero alternations, this contrast is achieved by mapping the difference in behaviour onto the underlying melody: ill alternates and is subject to Lower, while /ε/ is not. Hence, the underlying phonemic inventory contains objects that never appear on the surface under their lexical melodic identity, that is Iii. Therefore, theory is called to express the following three-way opposition on the surface. (266)

a. vowels that do not alternate but are phonetically identical to vocalised yers. b. yers that may or may not be vocalised on the surface (alternating yers). c. yers that are never vocalised (final yers).

The classical linear analysis operates the distinction between (266a) and (266b,c) by recording two different melodies in the lexicon: /ε/ vs. /i/. The opposition (266b) vs. (266c) on the other hand is relegated to the rule component. The introduction of autosegmental representations has changed the picture insofar as the opposition (266a) vs. (266b,c) is still made in the lexicon, but does not concern melody anymore: alternating and nonalternating [e]s are melodically identical. However, the latter is attached to a skeletal slot in the lexicon, while the former is not. This is a typical example of how enriched representations (or rather the introduction of representations from scratch) allow to view a contrast for what it is: a difference in kind, not in melody. Table (267) shows Rubach's (1986) implementation of this idea. Note the lexical difference between the

Structure and process: the take of Government Phonology

401

of this idea. Note the lexical difference between the alternating boldfaced and the non-alternating word-final case marker under (267a). (267)

representation of vowel-zero alternations according to Rubach (1986) Czech "elbow" a. lokt-e GENsg b. loket NOMsg c. loket-ni adjective X X X

X X

I I I I I l o k e t e

X X X

Χ

I I I I l o k e t e

X X X

X

X X

I I I I I I l o k e t e n i

However, the opposition (266b) vs. (266c), i.e. between alternating and final yers, is still managed by the rule component. Only has Lower now an autosegmental formulation: floating vowels acquire a skeletal slot iff they are followed by another floating vowel.

330

2.3.4. Further enriched representations: Government Phonology In Standard Government Phonology, the representations under (267) were further enriched so that alternating vowels are not floating matrices but empty Nuclei. To be precise, they were enriched in this respect, but impoverished elsewhere: the underlying identity of alternation sites was thought of as an empty Nucleus alone without any lexically specified melody. Alternating vowels were born through epenthesis into empty Nuclei in case they escaped Proper Government. The ensuing representations appear under (268).

402 (268)

11,3 Principles of argumentation II: representations and their function representation of vowel-zero alternations in Standard Government Phonology Czech "elbow" a. lokt-e GENsg b. loket NOMsg c. loket-ni adjective Gvt Gvt Gvt Γ Ί Γ " O N O N O N O N

f O N O N O N

f O N O N O N

X X X X X X I I I I I 1 o k t e

χ χ χ χ χ χ

x x x x x x x x

1 o k

1 o k

epenthesis

epenthesis

η ί

However, the melody of alternating vowels may not be predicted in all cases and interacts with surrounding segments in various ways. It must therefore be recorded in the lexicon (see §§450,461 and chapter 1,4 §76 for greater detail). The following representations are the result of this move (Scheer 1997:80s, 1998b:264s, 1999a:229s). (269)

representation of vowel-zero alternations in CVCV Czech "elbow" a. lokt-e GENsg b. loket NOMsg c. loket-ni adjective Gvt Gvt Gvt Gvt f C V C V C V I I I I I 1 o k e t e

f C V C V C V I I I φ I l o k e t

C V C V C V C V N i f i I I 1 o k e t e η ί

This option combines the insights of (267) and (268): alternating vowels are empty Nuclei in the sense that they are not associated to any melody lexically. The melody in question, however, is lexically present. It remains floating if its Nucleus ends up being governed, but associates in case it escapes Government. Government acts as an association inhibitor. Also, (269) represents the opposition between alternating (266b) and final (266c) yers: the former are empty Nuclei that possess a lexically nonassociated melody, while the latter are "real" empty Nuclei which are deprived of any melody (more on this in §461).

Structure and process: the take of Government Phonology 403 In sum, the three-way opposition of (266) is translated into the representations under (269) by three distinct underlying objects: vowels that are homophonous with alternating vowels but do not alternate (266a) (cz les lesa "forest NOMsg, GENsg") consist of a Nucleus and a melody which are associated (the second in GENsg /lokete/ under (269a)). Vowels that appear in alternation sites (266b) (alternating yers) are Nuclei that possess a melody underlyingly, but which is not associated (the first in GENsg /lokete/ under (269a)). In case they escape Government, this melody will attach to the Nucleus. Government acts as an association inhibitor. Finally, final yers (which exist only in word-final position) are Nuclei that do not possess any melody in the lexicon (the final empty Nucleus under (269b)). As a consequence, the different behaviour of alternating and final yers (266b) vs. (266c) is not a matter of the ordered application of rules anymore, but stems from the different representations that both objects enjoy. In regard of this specific issue, thus, the programme of shifting the burden from rule ordering to representations has been successfully realised.

2.3.5. The parameter Havlik vs. Lower supposes enriched representations The modern Slavic alternations reviewed follow the Lower pattern, whose peculiar characteristic is to provoke the vocalisation of alternation sites both before word-final empty Nuclei (final yers) and other alternating vowels (alternating yers). Cross-linguistically, however, this pattern cohabitates with another implementation of vowel-zero alternations: Havlik's Law. The only difference between Lower and Havlik is that alternation sites are vocalised when occurring before a phonetically expressed alternating vowel in the former, but not in the latter case. Most languages actually follow Havlik: modern occidental Arabic varieties, German, French, but also Old Czech and Old Polish. Hence opol pies - ps-ek (pies-k-a) vs. mpol pies pies-ek (pies-k-a) "dog NOMsg, dim. (GENsg dim)". Chapter 11,9 (§411) and §468 offer further discussion of the difference between Lower and Havlik. The only thing that needs to be borne in mind here is the existence of a parameter that controls both patterns (and as far as I can see these exhaust the cross-linguistic variation). Following the logic of vowel-zero alternations that is practised in Government Phonology in general and in CVCV in particular (chapters 1,3 §69 and 1,4 §76), the implementation of this parameter must concern the existence of a lateral relation. Since the phonetic absence of an alternation site is due to Government, its presence necessarily

404

11,3 Principles of argumentation II: representations and their function

means that it escapes this lateral force. In cases such as cz domecek "house double dim. NOMsg" where both s are alternation sites, thus, the second must not govern the first . This is how the parameter "alternating vowels may (Havlik) or may not (Lower) govern" is derived in §470 and further developed in §553. This kind of analysis supposes enriched representations: not only are additional (empty) constituents needed, but the parameterisation must be able to address a level of representation that is different from both syllabic constituency and melody: lateral relations among constituents. It is true, of course, that there may be other ways in other theories to encode the parameter Havlik vs. Lower. The modern Slavic pattern has always been considered a typical Slavic event that has no parallel in other languages (yers), to the effect that existing theories of vowel-zero alternations have not attempted to accommodate both Havlik and Lower under the roof of a unified theory. They may therefore come up with something that does not appeal to the parameterisation of lateral forces. In any event, the analysis that is proposed here could not exist if there were no lateral relations such as Government and Licensing. Vowel-zero alternations provide good illustration of how CVCV uses representations and in which way they are crucial for both description and explanation of phonological phenomena. A number of arguments that are developed in this part of the book stand on the grounds of this kind of enriched representations.

Chapter 4 332 Principles of argumentation III: generality of processes

333

1. Introduction: SPE hocus-pocus, the baby and the bath Some arguments that are made in this part of the book suppose that a typical generative reasoning is accepted: the best theory is the one that offers the most general formulation of a regularity. That is, linguists try to place as much of the phenomena observed as possible into Universal Grammar. In case a process is cross-linguistically stable, it is not reasonable to assume that it is driven by mechanism X in one language, but by mechanism Y in another language, where X and Y do not share any property and may be irreconcilable. Rather, such a situation requires a general and unique principle that is located in UG or as close to UG as possible. Its different expression in various particular languages is viewed as a consequence of language-specific parameters. On identical empirical coverage, the more general theory is to be preferred. This approach is obviously located on the "abstract" side of the debate that was reported in chapter 11,3 (§304). It has been advocated over the years by, among many others, Chomsky (1965:35s), Schane (1968,1974), Dell (1973:163ss), Gussmann (1980a), Dinnsen (1980). It goes without saying that I do not mean to escort the reader into a revival of the wild times of SPE where it was taken for granted that all items which are semantically, etymologically or paradigmatically related are the result of a synchronic derivation over one single underlying form.214 214

For English, the wildest picture was drawn by Lightner (1978:18s,1981,1985), who applied this principle in all its consequences without any inhibition. As a result, Lightner holds that eye and ocular, thunder and detonation, dental and tooth, rebel and bellicose, cardiac and heart, three and third, gynaecology and queen, sweet and hedonism and so on are synchronically derived from one single underlying form. Since the alternations h-k (heart - cardiac), d-θ (third fourth) and s-h (sweet - hedonism) suppose Grimm's Law, Verner's Law and the Ancient Greek s > h shift, respectively, Lightner concludes that all these processes are performed by the grammar of present day English natives. Sometimes, however, Common Germanic and Ancient Greek cannot account for all synchronic alternations. One case in point is the word tooth, which Lightner

406

11,4 Principles of argumentation III: generality

ofprocesses

Generative Phonology has (hopefully: some doubts are permitted) left behind these early teething troubles (see Vol.2,ΠΙ.3.3.2.7). We must be careful, however, not to throw out the baby with the bath. The general intellectual attitude that runs under the heading "generality of processes" should remain unchallenged: a theory that is able to cover different data sets (possibly in different languages) with a single mechanism is to be preferred over a theory that uses two distinct mechanisms for the same empirical coverage. Both attitudes - rejection of SPE hocus-pocus and abstract phonological principles - are not contradictory at all. Government Phonology has always practised both, and especially Jonathan Kaye has insisted that the former - rejection of SPE hocus-pocus such as velar softening (ele[k]tric electri[s]ity) and Trisyllabic Shortening (op[ej]que - op[as]city) is actually a condition on the latter, i.e. abstract phonological mechanisms. No consistent and general phonological theory may be constructed if it is sentenced to account for velar softening, Trisyllabic Shortening and the like. This issue will be discussed at length in Vol.2,III.3.3.2.7.

334

2. Vowel-zero alternations One telling example where the argument of generality is decisive are vowel-zero alternations. Cross-linguistic observation reveals that languages vary as to whether the elision is optional (e.g. German, French) or obligatory (e.g. Slavic), and with respect to the quality of the vowel(s) concerned. However, the phonotactics of vowel-zero alternations are remarkably stable over a wide range of genetically unrelated languages: a sample of vowelzero alternations in different languages has been introduced in §16 and

(1978:26) seriously proposes to be synchronically derived from the underlying representation /H3d/, where H3 is the Proto-Indo-European o-colouring laryngeal. Of course, Lightner is an extreme case, which is certainly not representative for the global phonological scene in the post-SPE period. However, he simply took the morpho-phonological idea to its logical end. And the idea that etymologically, paradigmatically or semantically related items are the result of a synchronic derivation was perfectly consensual (except of course in the Natural Phonologies, generative and non-generative: Vennemann 1974a, Hooper 1976, Stampe 1972, Dressler 1984).

Vowel-zero alternations 407 discussed subsequently. The correct descriptive generalisation that controls all cases on record is as under (270).215 (270)

alternation sites show a. zero / CV b. v o w e l /

C

#

e.g. Czech lokert-e "elbow GENsg" loket-0 "id. NOMsg" loket-ni "id. adjective"

Hence, vowels appear in closed syllables, while they are absent from the surface in open syllables. Conflicting systems where vowels would occur in open syllables and zero before Codas are not on record, and it is reasonable to assume that they do not exist, i.e. that they are not a possible instantiation of human speech. If the best theory is the one that proposes the most general formulation, vowel-zero alternations in all languages must be handled by a single mechanism. In other words, there is only one vowel-zero alternation in nature. Two different theories of vowel-zero alternations are necessarily mutually exclusive: a scenario whereby one theory accounts for vowel-zero alternations in language X, while another one covers language Y is out of the question. This is a welcome effect on the epistemological side since conflicting theories may not coexist and thus can be compared to each other on large empirical grounds. A theory of vowel-zero alternations that has been developed on the grounds of data from language X can be checked against the data of language Y. If the latter may not be accommodated in the theory, this theory is falsified and cannot be maintained for language X either. Along these lines, I have argued in favour of a theory of vowel-zero alternations that is based on empty Nuclei and Proper Government as in Standard Government Phonology, but couched within CVCV (Scheer 1998a). My goal was to show that Czech falsifies the statement "Proper Government may not apply over intervening governing domains", which is central for the standard account (Kaye et al. 1990, Kaye 1990a, Charette 1990).216 In this language, Proper Government freely applies over branching Onsets, and hence the theory practised in Standard Government Phonology is not a possible theory for Czech. I then demonstrated that the Czech data may be covered by CVCV without contradiction, and also that 215

216

Not withstanding the fact that the modern Slavic pattern obeys an additional complication: see the discussion of the yer context (Lower) in regard of the Havlik pattern in §§331,416,468,472s. §27 offers a more detailed discussion of this issue.

408

11,4 Principles of argumentation HI: generality

ofprocesses

CVCV is able to accommodate the French and Moroccan Arabic facts on which the standard theory was initially built. It must be concluded, then, that the theory which relies on the statement "Proper Government may not apply over intervening governing domains" is falsified not only for Czech, but also for all other languages. The theory based on CVCV, on the other hand, continues to be a serious competitor.

335

3. Vowel-zero alternations and sonority sequencing in languages without initial restrictions Another case in point is the argument made in chapter 11,6 (§381), which is based on two cross-linguistic regularities: vowel-zero alternations and the sonority sequencing principle. On one hand, vowel-zero alternations crucially refer to the contrast open vs. closed syllables. Hence, if there is just one theory of vowel-zero alternations in the world, all languages must possess open and closed syllables. On the other hand, however, the existence of any regularity regarding the sonority slope of consonant clusters is called into question by languages such as Moroccan Arabic. Here, no word-initial restrictions of the usual Indo-European kind ("sonority must increase") can be detected at all: any combination of any two consonants of the inventory occurs word-initially. Hence, word-initial consonants that are followed by another consonant may be treated in four different ways: 1) they are Codas in case of #RT clusters, and part of a branching Onset in #TR sequences; 2) they are always part of a branching Onset: Moroccan Arabic imposes no sonority-restrictions on branching Onsets; 3) they are extrasyllabic; 4) they always belong to an Onset whose Nucleus is empty, and the following consonant pertains to an independent Onset: [#CC] = IttCaCI. Extrasyllabicity is disregarded for independent reasons that are exposed in chapter 11,5 (§339). Options one and two force to release one of the cornerstones of syllable structure: under the assumption of 1), languages are divided into two sets according to whether they allow for the existence of word-initial Codas or not. Typical Indo-European languages do not, while Moroccan Arabic does. Hence, there is no universal syllable structure anymore, and the account of the absence of initial #RT clusters in typical Indo-European is demoted to pure observation: in an #RT cluster, R could belong to a Coda, but happens not to because of a language-specific parameter that has got nothing to do with syllabification.

Vowel-zero alternations: sonority sequencing in anything-goes languages

409

If 2) is assumed, the sonority sequencing principle is demoted to a language-specific existence: there is no ban against branching Onsets with a free sonority slope in natural language. Some languages have branching Onsets with a falling sonority slope (Moroccan Arabic), while others do not (typical Indo-European). By contrast, option 4) does not renegotiate the universality of either syllable structure or sonority sequencing: there are no word-initial Codas in any language, and all branching Onsets that exist in nature have a rising sonority slope. If, for this reason, we conclude that initial clusters are two independent Onsets in Moroccan Arabic, how do we syllabify word-internal consonant sequences? The usual word-initial anchor is not available. For example, [tl] qualifies as a branching Onset with respect to sonority, but will be denied this status because of its absence in word-initial position in a language like English (antler, butler etc.). It is demonstrated under (286) in chapter 11,6 (§381) that vowel-zero alternations in Moroccan Arabic treat any first member of an internal consonant cluster as a Coda: the preceding Nucleus will always be vocalised before two consonants, no matter whether they are of rising or of falling sonority. Since there are no branching Onsets in this language, we must conclude on the absence of Coda-Onset interludes as well: if any first member of a cluster behaves like a Coda, this will be true even for TR clusters. But declaring Τ a Coda in this case is in overt violation of sonority sequencing. Therefore, it must be concluded that just like initial clusters, all internal clusters are instances of two independent Onsets which enclose an empty Nucleus. As a consequence, the "closedsyllable" effect that we have observed in relation with vowel-zero alternations must be recast as an "empty Nucleus" effect: vowels occur before an empty Nucleus, while zero surfaces before expressed Nuclei. We are thus left with the following paradox: either there are two kinds of vowel-zero alternations. That is, the Moroccan Arabic one where alternations are controlled by the presence or absence of an empty Nucleus that follows the alternation site. On the other hand, languages such as German for example where empty Nuclei do not condition the alternation at all: a vowel occurs in a syllable that possesses a Coda, while zero appears in those syllables that lack Codas. If this picture is judged undesirable because it misses the crosslinguistic generality of vowel-zero alternations, an alternative unified theory necessarily supposes the abandon of universal sonority sequencing. That is, if the presence of a Coda in a syllable is the only cross-linguistic cause for the vocalisation of its Nucleus, any first member of an internal

410

11,4 Principles of argumentation III: generality ofprocesses

consonant sequence in Moroccan Arabic must be a Coda. This violates sonority sequencing in case of TR clusters. Both of these options are cumbersome and disregard the generality of processes. If there is only one sonority sequencing principle in all languages, and if there is only one single reason why alternation sites vocalise across languages, the only solution to this puzzle is to reinterpret vowelzero alternations as "empty Nucleus effects" in all languages. The argument for CVCV is thus obvious if it is accepted that the best theory is the one that makes the most general statements while offering the same empirical coverage as alternatives that must ascribe invariable crosslinguistic behaviour to language-specific parameter settings.

336

4. The yer context Chapters 11,9 (§411) and 11,10 (§426) discuss processes that occur in closed syllables and in open syllables iff the following vowel alternates with zero. Modern Slavic vowel-zero alternations are known to instantiate this pattern, and I therefore refer to it as the "yer context" (or the Lower pattern). However, they are not the only alternation that is governed by this context. I show that both within and outside Slavic segmental effects are conditioned by this disjunction. Among those are alternations in vowel length in Czech, Czech and Polish alternations between [o] and [u(u)], the Polish alternation involving the two nasal vowels of the language, French schwa - [ε] alternations, Southern French ATR-alternations of mid vowels and the realisation of the German velar nasal. Anticipating on the actual demonstration, table (271) illustrates these processes (ftill detail and glosses appear in chapters Π,9 §411 and 11,10 §426).217

217

Vowels that alternate with zero are referred to as "schwa" in this table.

The yer context 411 (271)

alternation a. mod. Slavic vowel-zero b. French schwa - [ε] c. Czech vowel length d. Southern French ATR

object occurring in closed syllables CV and _ C V if ifw schwa V=schwa zero vowel dom-ek-u [apale]

example dom-ek, dom-eô-ek, dom-eô-k-u [apel] j'appelle,

schwa

[ε]

appeler

[apelsKa] il appellera

VV

V

2áb-a

¡tab, ¿ak-ek, ¿ab-k-a

+ATR

-ATR

[fete] fêter [fet] fête, [alexte] alerter, [selsifi] céleri

e. German velar nasal

fog]

f. Polish ο-ό

[o]

g· Czech o-û h. Polish ή-ç

M

to] [u] [uu]

^ (>?) ¡Ά (>«0

[?iqgoo] Ingo krow-a ηοέ-e zfb-a

[ ? n p ] Inge, [zig] sing !, [zigt] er singt krów, krów-ek, krów-k-a ηΰέ, nûï-k-y, ηΰέ-ek z^b, z^b-ek, z^b-k-a

The contextual conditions of these alternations are highly specific and rather unexpected: "in closed syllables and in open syllables iff the following vowel alternates with zero". Could its recurrence in a number of different languages (which are all Indo-European, but belong to different families) be accidental? Could this behaviour in a language like Czech be due to a mechanism that is different from the one that provokes identical results in French or Polish? Do we observe the disjunction at hand by accident, and will processes occurring "in closed syllables and in open syllables iff the following vowel does NOT alternate with zero" be uncovered, or is the reverse context unobservable because it stands outside the possibilities of natural language? Will there be any prediction made to the effect that the reverse context will never be found to govern a phonological process in any language, or will we refrain from making falsifiable predictions of that kind? In other words, are there different yer contexts, possibly as many as there are languages, and is the converging behaviour under identical surface conditions pure accident? Or should the same effects in identical circumstances - again: which are highly specific - be ascribed to a single cause? In case it is judged that a multicausal interpretation misses a relevant generalisation that helps us gaining insight into syllable structure, particular analyses for any of the individual phenomena mentioned under (271) may be

412

11,4 Principles of argumentation III: generality

ofprocesses

falsified on the grounds of any other process of that pool. In chapters 11,9 (§411) and 11,10 (§426), I take for granted that the different instances of the yer context are not unrelated pieces which are produced at random by different language-specific phonologies. Rather, they are variations of one single phonological mechanism. Only on this assumption regarding the generality of processes does the argument in favour of CVCV that I develop make sense.

337

5. The Coda Mirror, Closed syllable shortening and 1-vocalisation It has been shown in chapter 1,6 (§110) and in Ségéral & Scheer (2001a) that consonants are invariably stronger in word-initial position and after Codas {#,C} , i.e. in the Coda Mirror, than in Codas and in intervocalic position. They are relatively more protected against lenition in the Coda Mirror than they are in Codas and intervocalically. On the other hand, fortition affects consonants in the Coda Mirror when weak positions remain unchanged. This generalisation has been made on the grounds of IndoEuropean, Semitic and Cushitic languages. The reverse phenomenology where consonants would be relatively weaker in the Coda Mirror than they are in Codas and intervocalic position does not appear to exist. On these grounds, I argue in chapter 11,13 (§556) that phonological theory ought to be able to assign a single, unique and positive identity to the Coda Mirror. Classical syllable structure is unable to achieve this goal, while CVCV can conceive of the Coda, its mirror and the intervocalic position as three phonological objects that enjoy a single, unique and positive characterisation each. The argument is strictly parallel to the one that was made on the grounds of the Coda context {#,C} in the late 70s. Its impact on phonological theory is known, i.e. the (re)introduction of syllable structure into generative phonological theory. However, this conclusion could only be reached on the assumption that processes which occur in highly specific disjunctive contexts are not accidental manifestations of arbitrary properties that are found in individual grammars. Rather, there is only one reason across languages for which consonants behave alike in word-final and preconsonantal position. The absence of other kinds of disjunctions such as, say, {#,V} in the structural description of phonological processes strengthens this view. Hence, if the recurrence of the Coda context {#,C} in a wide range of genetically unrelated languages allows to establish universal properties

The Coda Mirror, Closed syllable shortening and l-vocalisation

413

of syllable structure, the existence of its mirror image {#,C} in the same conditions must licence analogous conclusions. This line of reasoning is a straightforward expression of the assumption regarding the generality of processes: linguistic structure ought to express relevant generalisations, and if none is to be missed, the most general formulation of a process is the one that will be taken home. Coda effects such as Closed Syllable Shortening and l-vocalisation, which are discussed in chapter 11,12 (§524), are other instances of segmental reaction on a specific position in the string. They are approached along the same lines. For instance, the comparison of systems where internal and final Codas do vs. do not produce the same effect (on the consonant which they host or on the preceding vowel) may only be compared if it is asserted that there is only one reason for which we observe a specific phenomenology in and before Codas. Somebody in whose opinion [t] vocalises in Brazilian Portuguese in both Codas for some reason which does not coincide with the reason for which French [1] undergoes the same process only in internal Codas will be entirely immune against the argument. Indeed, there is no point in trying to achieve a unified theory that can accommodate both systems by virtue of a parameter setting if we are dealing with different causes that produce identical segmental results by accident. Here again, the argument crucially hinges on the generality of processes.

6. Summary Some of the statements that have been made in this section may appear to be trivial. For example, it would probably not cross any phonologist's mind to ascribe positional effects such as those provoked by Codas to different causes. This is a fundamental consensus of post-SPE phonology: structural descriptions that are recurrent in a wide range of genetically unrelated languages and whose reverse pattern is not recorded must share a common cause. On this assumption, and only on this assumption, may crosslinguistic data be used in order to argue for or against a specific take on syllable structure. A side-effect of this insight was discussed in chapter 11,2 (§303), that is disjunctive contexts. It is only when cross-linguistic data may be compared that the issue raised by disjunctions appears as a clear challenge. Most phonologists are probably willing to apply this argument to the

414

11,4 Principles of argumentation III: generality

ofprocesses

Coda Mirror without too much hesitation: just as the Coda, the Coda Mirror describes positional effects, and these are commonly viewed as the result of a specific syllabic configuration. However, people will probably be less easily convinced that vowel-zero alternations (§334), eventually combined with sonority sequencing and syllable structure (§335), as well as the yer context (§336), fall into the same category. The goal of the present section was to clearly identify the logical principle that grounds many of the forthcoming arguments. Generality of processes is applied to all cases discussed. Either all of them are under its spell, or none is. The issue at stake is genuinely generative: be as general as possible and as language-specific as necessary.

Chapter 5 339 Principles of argumentation IV: a better solution for extrasyllabicity than extrasyllabicity

Extrasyllabicity is mentioned on various occasions throughout this book. In all cases, it appears as an alternative solution to the one that is exposed. Indeed, the literature systematically appeals to extrasyllabicity in the analysis of various phenomena that are of interest here. I will try to explain in this section why the concept of extrasyllabicity is discarded from consideration in this book. The point that I make proceeds in two steps: one is taken here, the other will only be made in chapter 11,12 (§524) (right-edge) and Vol.2,III.6 (left edge). On the following pages, I set out to explain why extrasyllabicity is incompatible with Government Phonology in general and CVCV in particular. The exposition of these theoretical reasons is then followed by theory-neutral arguments against extrasyllabicity and a case study of one of the wildest instances of word-initial extrasyllabicity that is on record: Polish. I show that even here one consonant at most can be extrasyllabic at a time, something that is entirely unexpected when extrasyllabicity is take literally: a word should be able to begin with two seven, ten or twenty-five extrasyllabic consonants. As is obvious from the title of this chapter, I use the word "extrasyllabicity" in two different ways: as a particular phonological pattern that is known under this label, and as a syllabic analysis. I contend that the former is perfectly real and needs to be accounted for by all phonological theories. The latter however, I argue, should be abandoned in favour of a different analysis. What this surrogate analysis looks like will not be shown in the present chapter. This second step is only taken in chapter 11,12 (§524) for word-final extrasyllabicity, and the reader will even have to wait until Vol.2,III.6 to see how word-initial extrasyllabicity is dealt with in CVCV. But before actually evaluating the conceptual tool of extrasyllabicity, the following discussion up to §358 determines what its exact empirical basis is, why it was created, and in which theoretical environment it was conceived. As far as I can see, there are two (and only two) raisons d'être for extrasyllabicity: a consonant is said to be extrasyllabic if it wither cannot be parsed by the syllabification algorithm, or does not behave like a Coda. The detail of these two motivations is introduced first.

416

11,5 Principles IV: a better solution for extrasyllabicity

than

extrasyllabicity

Finally, a disclaimer is in order. As for most other concepts of the 80s, there is an OT-version of extrasyllabicity that does not tell its name. The same reality is analysed in Prince & Smolensky (1993: chapter 4), McCarthy & Prince (1993) and much subsequent work (e.g. Wiltshire 2003) under the heading of misalignment. Extrasyllabic consonants are said to be misaligned, which means that syllable- and word margins do not coincide: ...VC1.C2] is misaligned since the syllable boundary "." does not match the word boundary "]". C2, then, is the extrasyllabic consonant on the classical account. On the other hand, extrasyllabicity has also been analysed in Optimality theory along the classical representational lines where stray consonants need to link up to some constituent (syllable, foot, prosodie word) in order to acquire a phonetic existence. This stance is promoted for example by Green (2003) and Féry (2003). Whether the classical analysis is directly implemented or mediated by translation into the alignment vocabulary, virtually everything that will be said about extrasyllabicity also applies to its OTed version.

340

1. Some elementary and consensual facts about syllabic theory Phonologists have established a theory of the syllable. Since the advent of autosegmental structure, one of its non-controversial cornerstones is the fact that a phonetic action is the result of the association of three distinct phonological objects: 1) a skeletal slot; 2) a melody; 3) an association line. A skeletal slot without melody does not produce any phonetic trace (e.g. empty Onsets). In case a piece of melody has no skeletal anchor, nothing is heard (e.g. floating consonants of the French kind). If a skeletal slot faces a chunk of melody, but both items remain unassociated, there is no phonetic result either (e.g. vowel-zero alternations in CVCV, cf. §§81,461). It is also understood that the existence of a skeletal slot implies the existence of a syllabic constituent that dominates this slot. In short, phonetic output is only produced if a piece of melody is connected to a syllabic constituent. In the standard view, there are four syllabic constituents: Onset, Rhyme, Nucleus and Coda (and perhaps the syllable node itself). The co-occurrence of the pieces of melody that are eligible for syllabification into each one of these constituents is governed by strict rules: sonority must increase within Onsets, but decrease within Codas (provided a model is assumed that allows for branching Codas). The Nucleus and the Coda rather than, say, the Onset and the Nucleus, are

Some elementary and consensual facts about syllabic theory

417

grouped together under a higher constituent, the Rhyme, because they experience co-occurrence restrictions of the kind embodied by Closed Syllable Shortening ("the maximal space allocated to the couple Nucleus and Coda stretches over two skeletal slots"). Just like everywhere else in linguistics, phonological constituents are the formal expression of cooccurrence restrictions that language imposes on objects which are visible on the surface. If there were no co-occurrence restrictions, there would be no constituent structure.

2. The facts that cannot be accommodated by the theory Now suppose you explore unknown languages with this theory in mind. Sooner or later, you will come across some facts that are incompatible with your theory. Among those are the following observations. (272)

situations that give rise to extrasyllabic interpretations a. internal Codas react, but final Codas do not example: 1-vocalisation in French b. vowels in internal closed syllables react, but they show no effect in final closed syllables example: Icelandic Closed Syllable Shortening c. word-initial #RT sequences example: Czech rty "lips", lhát "to lie" etc. d. heavy word-final clusters example: English sixths [siks0s], German Herbst [hexpst] "autumn" e. co-called "trapped" consonants (chapter 1,10 §240) example: the [r] in Polish trvaó "to last", the [n] in Polish czosnku "garlic GENsg"

Let us first consider (272a,b), for which detailed illustration is provided in chapter 11,12 (§524). In some languages, Coda phenomena are visible in both internal and final position. For instance, [I] vocalises in internal and final Codas alike in Brazilian Portuguese (§529). In the same way, Czech Closed Syllable Shortening affects both vowels that occur in internal and final closed syllables (§534). On the other hand, there are languages like French where 1-vocalisation is observed in internal, but not in final Codas (hence modern alternations such as cheval [Javal] vs. chevaux [Javo] "horse, horses", see §527 on that). Analogously, Closed Syllable Shortening produces effects only on internal closed syllables in Icelandic, while leaving long vowels unchanged in final closed syllables (§533).

418

11,5 Principles IV: a better solution for extrasyllabicity

than

extrasyllabicity

An important additional information is the fact that the only impairment between internal and final Codas that seems to exist in natural language is the one that has been described. Languages where a phenomenon affects final Codas or final closed syllables, but not their internal peers, are not on record.218 This typological situation is summarised under (273). (273)

non-arbitrary impairment of final and internal Codas / closed syllables internal final a. do consonants in Codas react ? Brazilian Portuguese + + French + — trivial — —

does not exist b. do vowels in closed syllables react ? Czech Icelandic, Palestinian Arabic trivial

does not exist

—

+

+ + —

+ — —

—

*·

We are certainly entitled to conclude that "there is something wrong" with the right edge of the word: it repeatedly behaves in a peculiar way. Our theory is rigid enough to predict that syllabification is invariable: preconsonantal and word-final consonants pertain to Codas. A consonant-final word always ends in a Coda, never in an Onset. Hence, the regular case where nothing needs to be explained is the one where both internal and final contexts behave alike. The offending facts are those where word-final consonants ought to sit in a Coda, but do not behave accordingly: either they do not react themselves in the way their word-internal peers do, or the vowel preceding them behaves as if it stood in an open syllable.

342

3. Extrasyllabicity was created by people who believed in their theory Facing this situation, the reaction of phonologists has been sound. Instead of abandoning or modifying syllable theory because it could not accommodate an important and recurrent set of facts, they believed in their theory. Theories exist because they hit reality in an interesting way. Hitting reality means "not being compatible". The incompatible facts will then be looked 218

Except, of course, the notorious case of final devoicing. See §536b (note 340) on this issue.

Extrasyllabicity was created by people who believed in their theory 419 at through the prism of the theory, and appear in a new light, which tells us something about them. The only way to reconcile syllabic theory with the facts at hand was to allow for a parameter on word-final consonants: the general syllabification algorithm continues to make them Codas and hence remains unchanged. In case word-final consonants do not behave like Codas, though, they are no Codas. If they are not, what could they be? Since there are only two consonantal constituents (the Onset and the Coda) and syllabic theory at that time could not imagine the existence of word-final Onsets, there was only one solution: they are neither Onsets nor Codas. This means that, syllabically speaking, word-final consonants are nothing in these cases. They remain unsyllabified, and hence are extrasyllabic. It is worthwhile pointing out one consequence of this reasoning: the right edge, which we know is special, loses its magic. That is, extrasyllabicity is the faithful implementation of the reality observed. It does not ask any further questions, such as "why is it that the right edge of the word is special since (273), rather than the middle of the word?" For there is a way to accommodate the reality by opening a parameter, the peculiar properties of the right edge are not expected anymore to follow from some property of the theory: the facts are recorded; no invitation is issued for the search of explanation. I will come back to this issue below (§378s). For the time being, let us look at the implementation of the extrasyllabic strategy. The literature contains various outgrowths thereof. The idea that a structure-building mechanism may elude an object which occurs at a margin in order to make it stand astray roots in the treatment of stress. Since Liberman & Prince (1977), syllables that are left unconsidered by stress assignment are called extrametrical. Clements & Keyser (1983:104ss) have extended this concept to syllable structure on the grounds of French floating consonants (petit café [pati kafe] "small coffee" vs. petit effort [patit είοχ] "small effort"). All models of syllable structure that were available at the time when extraprosodicity was extended to the syllabic level were serial at heart: a syllabification algorithm gathers lexically unsyllabified segments into syllabified clusters. In some versions, extrasyllabic consonants simply stand astray (e.g. Steriade 1982, Itô 1986, Giegerich 1992, Hall 1992, Wiese 1996), in others they are dominated by a constituent called Appendix (Kiparsky 1979, Halle & Vergnaud 1980, Ewen & Hulst 2001:136ss distinguish between word-initial Prependices and word-final Appendices), Termination (Fudge 1969) or Specifier (Cairns & Feinstein 1982, Lapointe & Feinstein 1982). Hulst & Ritter (1999a) provide an informed overview of

420

11,5 Principles IV: a better solution for extrasyllabicity

than

extrasyllabicity

the various approaches to supernumerary consonants at word edges that have been developed over the years. Rubach (1999:292ss) also provides a broad introduction to the concept of extrasyllabicity. In the following sections, I identify the two reasons that motivate an extrasyllabic analysis. The one that was already mentioned (word-final consonants do not behave like Codas) will be exposed last. Let us first look at the other: consonants are extrasyllabic when they occur in a cluster that fails to be parsed by the syllabification algorithm.

343

4. Type I extrasyllabicity: enforced underparsing, an edge consonant fails to be parsed

344

4.1. Interaction of unparsable consonants with other rules: typical serial solutions Extrasyllabicity may be detected by the simple fact that the syllabification algorithm is unable to parse a given sequence. For instance, the initial cluster of Czech rty "lips" does not qualify as a branching Onset, and since a word may not begin with a Coda, the [t] will end up as an Onset, while the initial [r] is left unsyllabified. However, classical rule ordering allows for the amendment of this situation: at a later stage of the derivation, the extrasyllabic consonant may be reintegrated into syllable structure. This is done by so-called "adjunction-rules" that link extrasyllabic consonants to some constituent which is nearby.

345

4.1.1. German Jagd: devoicing must apply after adjunction Consider the following case borrowed from Hall (1992:122ss). In the German word Jagd [jaakt] "hunt (noun)", the final [-t] will end up being extrasyllabic because the syllabification algorithm requires a falling sonority slope for Codas. However, the underlying identity of this [-t] is /-d/, as is evidenced by the plural Jagd-en [jaakdan] "hunts". The /-d/ must thus have been subject to syllable-final devoicing which is otherwise general in German. Note that obstruents do not only devoice word-finally, but also in internal Codas: the /g/ of the plural Jagd-en [jaakdsn] is devoiced as well (its underlying identity may be determined by looking at the related verb jag-en [jaagan] "to hunt"). Hence, the extrasyllabic /-d/ must pertain to a Coda by the time syllable-final devoicing takes place.

Type I extrasyllabicity:

enforced underpaying

421

Core syllabification thus produces Jag (where square brackets indicate extrasyllabicity). At a later derivational stage, the is adjoined to the Coda, which by that time hosts /-gd/. This is the input to the "late" rule of syllable-final devoicing, which devoices all voiced obstruents that are found in Codas. Hence on this analysis, generalisations about sonority in constituents hold only at the very moment when the syllabification algorithm applies, but not any later in the derivation. As Hall (2000:248) puts it, sonority sequencing governs "deeper", but not phonetic representations. Serial rule ordering thus allows to interspread the derivational stages where consonants are extrasyllabic and those where they are reintegrated into syllable structure with various phonological rules. In the German example, adjunction to the Coda needs to be effected before the rule of syllable-final devoicing applies. In other cases that will be reviewed below, rules must apply while extrasyllabic consonants are still invisible for tree structure, i.e. before they are reintegrated.

346

4.1.2. Stray erasure and the Prosodie Hierarchy Note that even when there is no extra motivation of the German kind, it is held that extrasyllabic consonants must somehow be integrated into the arboreal structure. This is a consequence of Stray Erasure that was proposed by Steriade (1982), which deletes melodic material that does not belong to any constituent. On these grounds, McCarthy (1979), Itô (1986), Myers (1987) and others have derived a general principle according to which no segment may enjoy a phonetic existence if it is not integrated into the prosodie hierarchy. This term is understood as the tree structure in a broader sense which includes all kinds of constituents such as Onsets, Rhymes, Codas, morae, the syllable node, feet, the phonological word etc. Under this assumption, extrasyllabicity may only be temporal: there is a derivational stage where a consonant is extrasyllabic, and another where it is not anymore. Rules may apply at any time before or after the integration of the extrasyllabic consonant into the prosodie hierarchy. Another way of putting this is to say that all audible segments must be prosodically licensed. As a matter of fact, this is but a different formulation of the autosegmental ground-rules that are mentioned in §340: phonetic action is the result of the association of a melody and a (syllabic) constituent.

422 347

11,5 Principles IV: a better solution for extrasyllabicity than extrasyllabicity

4.1.3. Polish kadra: devoicing must apply before adjunction Adjunction rules that operate the reintegration of extrasyllabic consonants into the prosodie hierarchy may target Codas, Onsets, phonological words or other constituents. Sometimes as in the German example, the final destination of the extrasyllabic consonant can be determined through the interaction with other rules. In German, the word-final extrasyllabic consonant must be linked to the Coda since it is affected by a rule whose domain is this constituent. In other cases such as those discussed by Rubach & Booij (1990a), it may be established independently that final devoicing has scope over the phonological word, rather than over the syllable. Consonants that cannot be parsed by the syllabification algorithm such as the final [r] of kadr [katr] "staff GENpl" appear to be transparent for final devoicing: the dental is underlyingly voiced (as is evidenced by kadra [kadra] "staff NOMsg"), but devoices in GENpl although it is not word-final. In other words, the final [-r] behaves as if it were not there. 219 Hence, the Polish case is the reverse of the German process as far as rule ordering is concerned: in the latter language, the extrasyllabic consonant must be integrated into the prosodie hierarchy before final devoicing applies, while the situation in the former language requires that final devoicing affects the dental while the extrasyllabic consonant is still invisible for constituent structure. That is, Rubach «fe Booij (1990a) argue that the /d/ in GENpl /kad/ undergoes final devoicing because it is the last segment of the phonological word by the time the is still extrasyllabic. It is only once final devoicing has been effected that the stray is adjoined to the phonological word. In short, the fact that there are no "phonetic" extrasyllabic consonants is consensual in the literature on extrasyllabicity. 220 By contrast, the final host that picks up the stray material varies according to the language and the analysis (Goldsmith 1990:147). Roca (1992) provides a survey of the early literature on extrasyllabicity.

219

220

In fact the [r] is trapped. See §268 where an alternative to the extrasyllabic analysis is developed. Since Rubach (1997a), the extrasyllabic analysis of the data at hand has served to establish a derivational version of Optimality Theory, DOT. Only Piggott (1991,1999) argues against the need for a stray consonant to be adjoined to some constituent. According to him, being extraprosodic alone qualifies for phonetic existence.

Type I extrasyllabicity: enforced underpaying 348

423

4.2. Can there be more than one extrasyllabic consonant at the right edge? I will try to show in this section that the answer to the question raised in its title is no: there are no cases on record where extrasyllabic clusters occur at the end of words. This statement seems counter-intuitive at first sight. It hinges on two demonstrations: 1) only monomorphemic word-final clusters count, and 2) the supernumerary coronals that are typical for Germanic have got nothing to do with extrasyllabicity. The following section demonstrates the consolidating influence of morphological boundaries on final monster clusters. Also, the analysis that Government Phonology has developed in regard of these heteromorphemic clusters is exposed. §350 then inquires on the notorious "extra" coronals that may appear at the end of words in Germanic.

349

4.2.1. Reduction of extrasyllabic candidates by morphology Alongside with the relatively light clusters that were mentioned in the previous section, the word-final context is particularly incline to produce much heavier groups. However, it appears that all such instances may be reduced to a maximum of one extrasyllabic consonant when morphology is taken into account. Heavy word-final clusters are always heteromorphemic in English and German. For instance, the English example sixths that is always quoted in this context identifies as six-th-s [siks-6-s], Monomorphemic clusters such as [-ks0s]# do not occur in English, nor does [-ksBs-] exist wordinternally. Hence, the role of morphology moderates the issue of extrasyllabicity.221 Government Phonology has given a particular interpretation to this insight. Final empty Nuclei are licensed by a device whose source is not identified. This situation is usually referred to as "Licensing of final empty Nuclei" (see §61). It is obvious that the final character of these Nuclei is responsible for their extraordinary behaviour. The observation that the right edge of the word hosts monster clusters and thus deserves a special status in phonological theory is consensual. Its classical interpretation grants a special identity to the offending word-final consonants themselves (extrasyllabicity), whereas Government Phonology shifts the focus on the 221

See for example Goldsmith (1990:140ss) and Harris (1994a:66ss) for detailed discussion of the English data. Hall (1992:110) provides exhaustive facts for German.

424

11,5 Principles IV: a better solution for extrasyllabicity

than

extrasyllabicity

labicity), whereas Government Phonology shifts the focus on the following word-final empty Nuclei. Heavy word-final clustering stems from the extraordinary properties of these Nuclei. That is, Gussmann & Kaye (1993) and Kaye (1995) give a broader interpretation to the word "final" in "final empty Nucleus": it is not the fact of being word-final that licenses empty Nuclei in a special way, but rather the fact of being domain-final. Phonological domains are produced by morphology.222 Roughly speaking, level 2 affixes that are familiar from Lexical Phonology create two distinct domains (analytic morphology), while level 1 affixes form a single domain with their host (non-analytic morphology). Hence, six-th-s includes three domains and three domain-final empty Nuclei (rendered by "0" hereafter): [[[siks0]00]s0]. Word-final empty Nuclei were not created in order to account for heavy word-final clusters at all (see §§18,628). However, they now allow for an interpretation of the morphological complexity of word-final monster clusters that does away with their exceptional status. Or rather, sixths is no more exceptional than, say, pit = /pita/: both words possess domain-final empty Nuclei whose absence from the surface is due to a special property of the final site that remains to be understood. The only difference between both words is the number of those domain-final empty Nuclei, that is three in the former, against one in the latter case. In sum, thus, big word-final heteromorphemic clusters have got nothing to do with extrasyllabicity. They exist because of their morphological complexity. One could say that they are licensed by morphology. This statement is entirely theory-neutral: everybody must admit the fact that morphological complexity is the determining factor here. I have shown how Government Phonology interprets this situation.

350

4.2.2. You can get a coronal for free in Germanic Determining which consonants are extrasyllabic of course depends on the syllabification algorithm that is used. Typically, different algorithms have been proposed for different languages. And anyway, right-edge clusters may vary across languages. Since all reported cases of right-edge extrasyllabicity cannot be rehearsed here, I will proceed by the strongest case strategy. That is, I look at 222

§§24,249 have introduced the notion of phonological domains at some greater length (see also §630). Vol.2,III.3.5 offers full discussion.

Type I extrasyllabicity: enforced underpaying

425

two families where clusters reportedly flower at the end of words: Slavic and Germanic. In the former family, Polish is certainly the best studied language with respect to extrasyllabicity. Rubach & Booij (1990a,b) assume a syllabification algorithm that is able to accommodate clusters of falling sonority RT# into final Codas. Only the second consonant of clusters with rising sonority TR#, or of those that present a sonority tableau TT#, RR#, fail to be parsed. Examples are bóbr "beaver", szept "whisper", cierñ "thorn". As far as I can see, the Polish lexicon does not contain any monomorphemic word that, according to Rubach & Booij's criteria, presents more than one extrasyllabic consonant at the right edge. Let us now look at Germanic. Here again, the decision on the extrasyllabicity of a consonant depends on the syllabification algorithm used. Regarding German, Hall (1992) works with the same assumptions as Booij & Rubach (1990a,b). When the German lexicon is scanned according to these criteria, the result is much the same as before: there are words in TT# (e.g. nackt [nakt], Haupt [hawpt]) and RR# (e.g. Halm "blade (of grass)"), only TR# is missing since the sonorant in this position is syllabic (e.g. Nagel [naagj]). There is just one feature that sets German apart from Polish: "regular" word-final clusters that include extrasyllabic consonants may still be followed by additional consonants, sometimes even by clusters. Consider words such as Herbst [herbst] "autumn", Markt [markt] "market", Arzt [?artst] "doctor". Now it does not take long to see that these supernumerary consonants are coronal obstruents in all cases: Herb, Mark, Arz. In fact, we face a stable pan-Germanic pattern. It is notorious that whatever the "regular" phonotactics of a Germanic language is, "you can get a coronal for free" on top of that at the end of words. This distributional fact is well studied: Hall (1992:1 lOss) and Wiese (1991) report on German, Goldsmith (1990:140ss), Myers (1987) and many others on English, Ewen & Hulst (2001:136ss) on Dutch. What should be the phonological attitude towards these extra coronals? Obviously, we face but one particular effect of the well-known peculiarity of coronality, which appears on many other occasions. There is a large literature on the special status of coronals: Paradis & Prunet (1991) collect some of the evidence. Hence, it seems that the existence of extra coronals at the right edge of words has got nothing to do with the phonotactics of Germanic. Restrictions on clusters disregard additional coronals. If this is true, we are left with little evidence for right edge extrasyllabicity. It does not matter

426

11,5 Principles IV: a better solution for extrasyllabicity than extrasyllabicity

whether we wish to count sonorants into the class of supernumerary coronals or not (in words such as Harn [harn] "urine"). In any event, the maximal number of extrasyllabic items that may occur at the end of words is one.

351

4.2.3. Summary All in all, it seems that clusters which appear as real monsters at first sight lose their teeth when 1) morphological structure and 2) the special status of coronals is taken into account. Again, the cases studied are but a window on what is going on out there. But they are reportedly wild. In any event, I have not come across a convincing case where more than one consonant is extrasyllabic at the right edge of words. I therefore take this working hypothesis for granted: word-finally, natural language produces one extrasyllabic consonant at most.

352

4.3. Word-initial extrasyllabicity Extrasyllabic consonants are also found word-initially. They are identified along the same lines as their word-final peers: the syllabification algorithm cannot parse them. That is, they are the first member of initial clusters that show either a decreasing sonority slope #RT or a sonority plateau #TT, #RR.

353

4.3.1. Typology of initial extrasyllabicity Word-final extrasyllabicity is common in many Indo-European languages. Word-initial extrasyllabic consonants, however, are exotic in IndoEuropean. The very constraint on branching Onsets that requires sonority to increase stems from this observation which is typical for most IndoEuropean languages: there are no initial #RT clusters. One family massively misbehaves in this respect: Slavic. Another language, Greek, does possess offending initial clusters, but these are limited to #pt-, #kt- and #mn-. The Slavic case will be studied at length in Vol.2,III.6.5, and some attention will also be given to Greek (Vol.2,III.6.3). Modern occidental Arabic (e.g. Moroccan Arabic) and Berber are another group of genetically and geographically defined languages where initial

Type I extrasyllabicity: enforced underpaying

427

#RT clusters are commonplace. Their pattern is discussed in chapter 11,6 (§381). Finally, the Salish family (native American Northwest languages) is reputed for monster clusters. Relevant evidence is discussed in §376 (see also note 195). Other languages with initial #RT clusters exist, but their distribution over the globe and according to genetic kinship appears to be erratic. At least, this is what Clements' (1990:288s) list of #RT languages suggests. 223

354

4.3.2. Word-initial extrasyllabic consonants are different I: they are not transparent to voicing Hence, word-initial extrasyllabic consonants are set apart from their peers at the right edge of the word in regard of their restricted occurrence in IndoEuropean and other languages. Interestingly, this distinction is reproduced by their behaviour. One of the major results of Rubach & Booij (1990a) is the impaired reaction of initial and non-initial extrasyllabic consonants with respect to voice assimilation and degemination. Recall that in Polish, the underlying /d/ of kadr [katr] "staff GENpl" devoices by virtue of final devoicing as if the final extrasyllabic were not there. This is what Rubach & Booij call the transparency of word-final extrasyllabic consonants. In contrast to this behaviour, word-initial extrasyllabic consonants are not transparent to voice assimilation. Consider first that voice assimilation applies across word boundaries in Polish: the underlying Iti in teatr [teatr] "theatre", which is followed by an extrasyllabic , is voiced in case the next word begins with a voiced obstruent such as in teatr wojenny [teadr vojsnni] "war theatre" (Rubach & Booij 1990a:445). Hence, voice assimilation affects the /t/ across 1) a word-boundary and 2) a word-final extrasyllabic consonant. However, no such assimilation is observed in identical conditions word-initially. For instance, the underlying l-dl of pod "under" is not devoiced in case the following word begins with an extrasyllabic consonant plus a voiceless obstruent: pod mchem [pod mxem] "under the nose", od mszy [od mji] 223

Greenberg's (1978) list can hardly be used for the purpose of identifying #RT languages since he counts a word as #RT-initial even if there is a morphological boundary between both initial consonants or, worse, if both initial consonants belong to two different words. For instance, Czech ν domu [v domu] "in the house" where "v" is a preposition meaning "in" and "domu" the noun meaning "house", is counted as one word that begins with an offending [#vd]-cluster. This way of counting is worse than not counting at all.

428

11,5 Principles IV: a better solution for extrasyllabicity

than

extrasyllabicity

mxem] "under the nose", od mszy [od mji] "since the mass" illustrate this pattern. Assimilation does not operate in the opposite sense either: the final /-k/ of brak "lack" is not voiced when it comes to stand before rdza [rdza] "rust" where the initial is extrasyllabic, i.e. brak rdzy [brak rdzi] (Rubach & Booij 1990a:447). It must thus be concluded that word-initial extrasyllabic consonants, unlike their word-final peers, are not transparent for the purpose of voice assimilation.

355

4.3.3. Word-initial extrasyllabic consonants are different II: they do not degeminate The behaviour of geminates reveals the same contrast. In words such as flotylla [flotilla] "fleet NOMsg" and Sybilla [sibilla] "Sybilla", the geminate occurs as a simplex consonant in word-final position flotyll [flotil] "fleet GENpl" and before consonants Sybilski [sibilski] "Sybilla, adjective". Rubach & Booij (1990a) relate this degemination to the extrasyllabic character of the second part of the geminate in degeminating contexts: Sybil-ski, flotyl. The degeminating rule they invoke makes direct reference to extrasyllabic consonants in its structural description. However, word-initial geminates freely occur in Polish. They do not show any inclination to degeminate: ssac [ssatç] "suck", na czczo [tjtfo] "on empty stomach", dzdzysty [d3d3isti] "rainy" etc. Rubach & Booij (1990a) treat this contrast in terms of rule ordering: word-initial and non-word-initial extrasyllabic consonants are incorporated into the prosodie hierarchy by two different rules that apply at two different derivational stages. The one concerning word-initial consonants is called Initial Adjunction and is "early", i.e. applies before degemination and voice assimilation rules. Hence, there are no word-initial extrasyllabic consonants anymore that could be the target of degemination by the time this rule is effective. In the same way, word-initial extrasyllabic consonants are not stray anymore by the time voice assimilation occurs, to the effect that the last consonant of the preceding word and the one that follows the formerly extrasyllabic item are not adjacent. Consequently, voice assimilation is blocked. On the other hand, word-final and word-medial extrasyllabic consonants are adjoined by a "late" rule called Housekeeping Adjunction that saves its targets from stray erasure after degemination and voice assimila-

Type I extrasyllabicity: enforced underparsing

429

tion have operated. Hence, word-final and word-medial extrasyllabic consonants are still extrasyllabic when degemination applies. They are also transparent for the purpose of voice assimilation since their extrasyllabic status makes the flanking consonants adjacent at the syllabic level.

356

4.3.4. The contrast is automatic if word-final extrasyllabicity is due to final empty Nuclei The contrast between word-initial and non-word-initial extrasyllabicity could be a phonologically relevant generalisation. We have two indications to the effect that these sets of consonants are different in kind. For one thing, they enjoy contrasting distribution across languages. Word-initial extrasyllabic consonants are much more "exotic" than their word-final peers (e.g. Goldsmith 1990:107). To be precise, it appears that both sets contract an implicational relationship: if a language has word-initial extrasyllabic consonants, it will also possess word-final extrasyllabicity. The reverse is not true. Second, they show contrastive behaviour, as demonstrated by Rubach & Booj (1990a). If this empirical contrast is an invitation to consider both types of consonants distinct theoretical objects, the approach that was introduced in §348 seems to be on the right track: word-final extrasyllabic consonants are special because they occur before a final empty Nucleus. Under these provisions, it is obvious that the peculiar status of word-initial extrasyllabic consonants cannot be ascribed to the same cause: they are not followed by a final empty Nucleus. In the view of Rubach & Booij (1990a), all supernumerary consonants enjoy identical representation as extrasyllabic items. The difference between the two kinds of extrasyllabicity is then achieved by the rule component: two distinct rules that apply at two different derivational stages target word-initial and word-final extrasyllabic consonants. The representational alternative of Government Phonology holds that the contrasting behaviour is not due to different rules with different serial properties. Rather, the two classes of consonants are different since the lexicon. Their contrasting behaviour stems from this representational difference: they are either followed by a final empty Nucleus or not. Note that for the time being, the distribution of final empty Nuclei does not offer any analysis of (word-initial or word-final) extrasyllabicity. Recall that such an analysis is postponed until chapter 11,12 (§524), especially §547 (word-final), and §408 (word-initial, Vol.2,III.6.5 treats this

430

11,5 Principles IV: a better solution for extrasyllabicity than extrasyllabicity

issue at length). The only thing that I wish to point out at this stage of the discussion is that the contrast between both edges is in-built anyway if it is true that word-final extrasyllabicity is a consequence of the special properties of final empty Nuclei.

357

4.4. Peripherally As is obvious from the foregoing discussion, extrasyllabic consonants occur either in word-initial or in word-final position. On the preceding pages, word-medial extrasyllabicity was briefly mentioned in relation with degemination in Polish, but these cases are so extremely rare and unclear that the literature does not usually take them into account. Even though Polish seems to be the only case on record, Rubach & Booij (1990a), Rubach ( 1996,1997a,b) attribute extrasyllabic status to [n] and [r] in Polish words such as piosnka "song" and trwac "to last", respectively. In the more recent references, Rubach (1997a,b) actually makes crucial use of the behaviour of [r] in the latter word in order to argue that Optimality Theory should allow for serialism (Derivational Optimality Theory DOT). These special Polish consonants are discussed at length in chapter 1,10 (§240), where they are referred to as trapped consonants. Since the uniform identity of word-initial, word-final and word-medial unsyllabifiable consonants is called into question in this book, an alternative interpretation of apparent word-medial extrasyllabicity is developed. On this account, the alleged unity of the three kinds of extrasyllabic consonants (initial, medial and final) is but a mirage created by syllabification algorithms: the only reason why the three kinds of consonants are understood as a homogeneous set is the fact that algorithms cannot parse them (more on that shortly). If initial and final supernumerary consonants are special for reasons that have nothing to do with extrasyllabicity, the place of the special Polish (trapped) consonants in the picture must be reconsidered. This has been undertaken in chapter 1,10 (§240). In any event, there is reason to doubt the existence of word-internal extrasyllabic consonants that is entirely independent of the orientation of the present book. That is, the literature on extrasyllabicity unanimously takes the peripherality of extrasyllabic consonants as an empirical fact and/ or a theoretical requirement. This has been formulated as the so-called Peripherality Condition, of which Roca's (1994:213) version appears under

Type I extrasyllabicity: enforced underpaying

431

(274) (see also Harris 1983, Milliken 1988, Clements 1990:290,1997, Hayes 1995:57s). (274)

Peripherality Condition Extrametrical elements must be peripheral in their domain.

"What this means is that we can only mark the left or rightmost element of a syllable, foot, word, or whatever extrametrical. It would be impossible to mark, say, the second syllable of a three syllable word extrametrical", as Spencer (1996:246) puts it. This rigid prohibition is but the statement of an observational fact. It does not derive from any property of the theory. According to (274), word-final extrasyllabic consonants for instance are "predicted" to lose their extrasyllabic status if a suffix is added that dispenses with their word-final status. This is indeed what happens: Old French 1-vocalisation for example does not apply to word-final [I] in a word such as vassal [vasal] "vassal", but is observed as soon as the plural-marker [-s] is suffixed, vassau-s [vasaws] "vassals". If word-final [I] is extrasyllabic, the generalisation "[I] vocalises in Codas" may be maintained (see §527). It is fairly obvious that the peripherality of extrametrical elements is not a random property of language. It denotes the special status of edges. Regular treatments of extrametricality merely encode this observation into prose, but refrain from inquiring on the cause for the extraordinary properties of domain edges. Identical practice is encountered when arbitrary diacritics are used in order to refer to domain boundaries, that is "#", "+", "-" and the like. A label is given to the linguistic reality at hand. Its properties, however, remain uninspected and unquestioned (see chapters 1,5 §83, 11,8 §390 and Vol.2,III for further discussion).

358

5. Type II extrasyllabicity: deliberate underparsing, a word-final consonant does not behave like a Coda

359

5.1. If certain word-final consonants must not be Codas, what could they be? Except the failure of being parsed by the syllabification algorithm, there is yet another reason why consonants commonly receive extrasyllabic status. In many instances where they are perfectly parsable, they behave as if they were not parsed. Or, in other words, they should behave as Codas because

432

11,5 Principles IV: a better solution for extrasyllabicity

than

extrasyllabicity

regular syllabification would assign them Coda status. However, they refuse to do so. As far as I can see, such motivation is only found word-finally. Word-initial (and word-medial) extrasyllabicity never hinges on this kind of reasoning. The evidence that prompts word-final extrasyllabic consonants in absence of any parsing urge is exposed at length in chapter 11,12 (§524). In some languages, Coda phenomena are visible in both internal and final position. For instance, [I] vocalises in internal and final Codas alike in Brazilian Portuguese (§528). In the same way, Czech Closed Syllable Shortening affects vowels that occur both in internal and final closed syllables (§534). On the other hand, there are languages like Old French where 1-vocalisation is observed in internal, but not in final Codas (§527). Analogously, Closed Syllable Shortening produces effects only on internal closed syllables in Icelandic, but leaves long vowels unchanged in final closed syllables (§533). Also recall from (273) that the impairment of internal and final Coda effects is not random: languages where a phenomenon only affects final Codas or final closed syllables, but not their internal peers, do not seem to be on record. It thus appears that word-final consonants whose syllabic status should be a Coda sometimes refuse to behave as a Coda: they do not show the effects observed in internal Codas, and the vowel preceding them does not react in the way it does when a Coda closes its syllable in word-internal position. Therefore, it is reasonable to conclude that word-final consonants do not pertain to Codas in these cases. However, their syllabification into a Coda is the only option available if syllable structure is built by a regular syllabification algorithm. Therefore, the only way of making word-final consonants something else than Codas while preserving the universality of syllabification is to assume that they are left unconsidered by the algorithm. They are invisible for the purpose of syllabification. Note that this is the only motivation for extrasyllabicity in these cases: extrasyllabicity is the consequence of the impaired behaviour of word-final consonants and syllables with respect to their internal peers. Extrasyllabicity can never be predicted on the grounds of any parameter that is independent from the behaviour of word-final consonants and the vowels that precede them. Standard Government Phonology shares the conclusion that wordfinal consonants must not pertain to Codas. However, this insight has been implemented in a different way: since strings are fully syllabified in the lexicon, there is no algorithm that builds syllable structure, and hence no universality of such a mechanism has to be taken care of. Moreover, the

Type II extrasyllabicity: deliberate underparsing

433

theory recognises the existence of empty Nuclei as regular phonological objects. Therefore, there is no reason why word-final consonants should not belong to a syllabic constituent that is not a Coda: they may well be dominated by an Onset whose Nucleus is empty. This is the solution favoured by Kaye (1990a). In contrast to regular extrasyllabic accounts, however, Kaye's (1990a) C o d a Licensing grants universal status to the Onset analysis of word-final consonants: they are supposed to be the Onset of a final empty N u c l e u s in all languages. I come back to the position of Government Phonology shortly (§363). Also, chapter 11,12 (§524) further examines Kaye's maximal claim according to which no word-final consonant in no language can be a Coda. B e f o r e summarising the issues related to extrasyllabicity, let us look at a typical case of word-final extrasyllabicity that is particularly interesting because it combines effects on both the final consonant and the preceding vowel.

360

5.2. Extrasyllabicity effects on both word-final consonants and the preceding vowel Kenstowicz (1994:274s) discusses stress assignment in Levantine varieties of Arabic such as Palestinian. In these languages, stress falls on the first vowel of a word unless it is attracted by a closed syllable that is closer to the right margin of the word. Hence, stress assignment can be formulated in the following way: "coming in f r o m the right edge of the word, stress the first closed syllable that you meet. If there is no, stress the last vowel you meet (i.e. the first vowel of the word)". This mechanism is illustrated in conjugation: /katab-u/ "write pf 3 rd pi" (where /-u/ is the personal marker) is stressed on the first vowel because it does not bear any closed syllable: [ká.ta.bu]. However, in case a consonant-initial suffix is added as in pf 1 st pi I-nal, stress falls on the second [a]: /katab-na/ comes out as [ka.táb.na]. In case of zero suffixes, /katab-o/ should bear stress on the final syllable since it is closed by the word-final /-b/. This, however, does not happen: pf 3 rd sg /katab-0/ receives stress on the first vowel [kátab] j u s t like /katab-u/ and unlike /katab-na/. It must thus be concluded that the final syllable is not closed in /katab/. Or rather, that its final consonant is invisible for the mechanism that assigns stress. Kenstowicz (1994:274) therefore proposes a rule that makes any word-final consonant extrasyllabic: [+cons] —• [+extrasyllabic] / #.

434

11,5 Principles IV: a better solution for extrasyllabicity

than

extrasyllabicity

As usual, extrasyllabicity occurs only at margins, an observation that has been encoded in the Peripherality Condition (§357). That is, the /-b/ is extrasyllabic only when it occurs at the right edge of the word. In case a suffix consisting of a sole consonant such as pf 1st sg /-t/ is added, /-b/ does not behave anymore is if it did not close the preceding syllable. The actual surface form of /katab-t/ is [katábit] with an [i] between the Ibi and the /t/, but this vowel can be shown to be epenthetic on independent grounds. The crucial observation here is that stress falls on the second, not on the first [a], which thus necessarily belongs to a closed syllable. It must therefore be concluded that Ibi closes the preceding syllable: in /katab-t/, it is not extrasyllabic "anymore" because it does not occur at a word-margin. In a serial approach, then, the events are ordered as follows: 1) the rule that makes word-final consonants extrasyllabic, 2) syllabification, 3) stress assignment, 4) revocation of extrasyllabicity (as Kenstowicz calls it, corresponding to the adjunction to some syllabic or prosodie constituent in other analyses), 5) epenthesis of [i]. Epenthesis needs to occur after stress has been placed since it makes the Ibi an Onset. Finally, Kenstowicz reports another observation, which is consistent with what we already know: long vowels are prohibited in internal, but not in final closed syllables. Hence, a root that possesses a long vowel in the lexicon such as /stajaar/ "to consult" appears unchanged in case a vowelinitial suffix is added [stajaar-u] "to consult pf 3pl", while shortening is observed with consonant-initial suffixes: [stajar-na] "to consult pf lpl". In pf 3sg where there is no phonetically expressed suffix, however, the vowel is long: [stajaar]. Hence, if final consonants are invisible for the rule that effects vowel shortening, Closed Syllable Shortening, which is attested in a wide range of languages, may be applied without amendment: "long vowels shorten in closed syllables". This is a very typical case of the application of the extrasyllabic idea: there is a cross-linguistically well attested process, and a language-specific situation that looks very much like an instance of this process. However, the word-final pattern seems to prohibit its application. In such a situation, an amendment is made within the grammar of this language (the presence of the rule that creates extrasyllabicity), which allows not to give up on the universality of the process itself.

Type II extrasyllabicity: deliberate underpaying 361

435

5.3. Summary Unlike the cases of extrasyllabicity that were discussed in §343, consonants are not extrasyllabic here because they fail to be parsed. Rather, the extrasyllabic status is assigned by a sovereign decision of the analyst, whose goal is to preserve the generality of some process. 224 However, both kinds of extrasyllabicity are actually motivated by the principle that was exposed in chapter 11,4 (§332): not only processes, also theories should be as general as possible. That is, either consonants are made extrasyllabic as soon as they cannot be parsed by the syllabification algorithm in order not to bleed the generality of this central device of grammar. Or consonants are declared invisible for the algorithm if this move allows a phonological process to enjoy a more general description that is also found in other languages.

362

6. Doubts on extrasyllabicity and an alternative view After having exposed the way extrasyllabicity is classically conceived of, I now set out to pinpoint some concerns that are raised by this device. The critical review will be closed by the sketch of an alternative.

224

Therefore, formulations of the kind "let us view X as extrasyllabic in order to achieve the generalisation Y" are typical: the "diacritic feature [of extrasyllabicity] allows segments to be made invisible to the syllabification rules or algorithm" (Kenstowicz 1994:274); "we can say that the Warao system is identical to that of Weri, provided we ignore the final syllable of every word" (Spencer 1996:246); "in order to extend the procedure [...] to Winnebago, we first need to account for the stresslessness of its initial syllable. This can readily be done by appeal to extraprosodicity, a device which enforces invisibility on a peripheral element [...]. The extraprosodicity of the initial syllable [...] yields the desired formal identity between the Winnebago and Yidiny stress algorithms" (Roca 1994:213). Gussenhoven & Jacobs (1998:217) write: "[...] in order to maintain the basic bounded foot types (maximally two syllables in a foot), Hayes (1981) uses the concept of extrametricality. At the periphery of a word that is, at the right or left edge - a phonological constituent (syllable, segment, consonant, vowel, rhyme, mora etc.) may be declared extrametrical, that is, be made invisible to the metrical tree construction rules."

436

11,5 Principles IV: a better solution for extrasyllabicity

than

extrasyllabicity

6.1. There is no extrasyllabicity without serialism and a syllabification algorithm Strings are fully syllabified in the lexicon in Government Phonology, and syllable structure remains stable under phonological processing. This stance was originally encoded by the Projection Principle (Kaye et al. 1990:221). It follows that there is no resyllabification, and that phonological processes are structure preserving (cf. §§10,17). Another trivial consequence is that there is no such thing as an algorithm that assigns syllable structure. Hence, consonants that are called extrasyllabic cannot fail to be parsed by any algorithm, nor can they be made invisible to any syllable-building mechanism. All segments belong to a syllabic constituent since the lexicon. Therefore, consonants cannot be extrasyllabic in the sense of the foregoing pages. Since there is no syllabification algorithm, all segments belong to a syllabic constituent at any derivational stage.225 Hence, Government Phonology must come up with an alternative analysis for the facts that are usually covered by extrasyllabicity. Such an alternative has already been mentioned in the foregoing discussion, and will be developed at length in chapter 11,12 (§524). Another property of Government Phonology is its explicit rejection of rule ordering (Kaye et al. 1990:194, Kaye 1992b). However, the serial application of rules is central for the conception of extrasyllabicity. In order to account for various phenomena, some rules must crucially apply before or after the reintegration of extrasyllabic consonants into the prosodie hierarchy. In the same way, word-final consonants that are declared extrasyllabic because they do not behave like Codas must not be adjoined to any element of the arboreal structure before relevant rules have gone into effect. What is usually taken to be caused by extrasyllabicity may not be achieved if processes do not apply serially. This is the second reason why there can be no such thing as extrasyllabicity in Government Phonology. In the following sections, individual arguments against extrasyllabicity are developed from inside the extrasyllabic perspective.

225

Except floating consonants of the French kind and other floating autosegmental material of course.

Doubts on extrasyllabicity and an alterative view 437 364

6.2. Extrasyllabic once, extrasyllabic forever Recall from the discussion of Palestinian Arabic in §360 that word-final consonants behave as if they were not there for the purpose of two distinct phonological processes: stress assignment and Closed Syllable Shortening. This uniform behaviour of consonants seems to be cross-linguistically stable: either a segment is extrasyllabic, in which case it enjoys this status in regard of all processes, or it is not, which means that it will be extrasyllabic with respect to no event. A language where word-final consonants are extrasyllabic in regard of, say, Closed Syllable Shortening, while counting for the purpose of stress assignment, does not appear to be on record. The ever uniform behaviour of word-final consonants in regard of all processes that are active in a language, however, is entirely unexpected in a serial approach: rules relative to stress, vowel length or any other kind of segmental phenomenon may be interspread without any restriction related to their content. This follows from the very principle of extrinsic rule ordering: in a given language, the order of rules may not be predicted from any of their properties; the only way to establish their ordering is to look at their effect. Hence, there is no reason why, say, stress assignment should not be able to apply before extrasyllabic consonants are reintegrated into the prosodie hierarchy, while Closed Syllable Shortening occurs only once this adjunction is effected. In other words, serialism predicts that within the same grammar, final consonants may be invisible for a given process, but taken into account by another event. In absence of any empirical response of that kind, extrasyllabicity appears in a dubious light. It will be shown in §551 that CVCV predicts the invariable typological situation: since both the relative strength of wordfinal consonants and the length of vowels in final closed syllables depend on the same parameter (the Licensing ability of final empty Nuclei), their fate is bound together.

365

6.3. Extrasyllabic consonants do not behave like nothing - they often behave like Onsets It was mentioned earlier that everybody agrees on what word-final consonants which do not provoke Coda effects are not: Codas. If they are no Codas, what could they be? There are two and, as far as I can see, only two logical possibilities: either they are Onsets, or they are nothing, i.e. extrasyllabic. This contrast thus provides a fair measure of the two alternatives,

438

11,5 Principles IV: a better solution for extrasyllabicity

than

extrasyllabicity

and it makes a prediction: the word-final consonants in question should pattern with word-internal Onsets if they are Onsets - but with neither word-internal Onsets nor Codas if they are nothing. Gussmann & Harris (2002:22ss) and Harris (1994a) present various kinds of evidence showing that word-final consonants which for sure are not Codas pattern with Onsets. They quote for example English word-final consonants, whose distribution is fairly free and certainly does not meet the severe restrictions to which word-internal (monomorphemic) Coda-Onset sequences are subjected. Rather, they match the distributional latitude of word-initial and intervocalic consonants. Also, Gussmann & Harris bring neutralisation patterns from Khosian languages to bear, which are fairly radical. In Zu|'hoasi for example, no less than 88 (sic, including 47 different clicks) consonants can appear word-initially. Intervocalically and in wordfinal position, however, this flowering plurality shrinks radically: only four different consonants occur in the former location, and two in the latter. Gussmann & Harris conclude that these positions must share some property which induces this radical kind of melodic depletion: onsethood. Obviously, word-initial consonants are also Onsets, and hence Gussmann & Harris will have to argue for a special protective status of this position. Research along this line which tries to tackle the identical behaviour or distribution of word-final and intervocalic consonants is not very developed, hence there may be more evidence out there. Whatever the empirical situation, I wish to point out here that this is an interesting area because it bears exactly on the difference between the literally extrasyllabic approach and the one where extrasyllabic consonants are Onsets.

366

6.4. Can we afford to allow for constituents that do not express any cooccurrence ? In linguistics, tree structure expresses co-occurrence restrictions. For example, a Determiner Phrase cannot host a verb and a noun. Syllabic constituents exist for the same reason: two stops are never observed within an Onset. There are co-occurrence restrictions between the Nucleus and the Coda such as those evidenced by Closed Syllable Shortening where both constituents must not cover more than two skeletal slots. Therefore, the Nucleus and the Coda are dominated by a higher node, the Rhyme, whereas the Onset and the Nucleus are not. Some versions of extrasyllabicity differentiate between "phonological" and "phonetic" extrasyllabicity. Recall from §345 that a German word

Doubts on extrasyllabicity and an alterative view 439 like Jagd "hunt (noun)" is supposed to be parsed as Jag. Subsequently, the extrasyllabic is adjoined to the Coda because it must pertain to this constituent by the time final devoicing applies, which has scope over Codas (Hall 2000:248). Sonority sequencing, which prohibits the syllabification of two stops into a Coda, is therefore held to apply at a certain "deeper" level, while "phonetic" representations may include monster Codas and any other object of miracle. In the presence of this kind of proposal, it js legitimate to ask what constituent structure is there for: if it can be violated in any way that suits the linguist, what kind of generalisation does it express? Constituent structure and sonority sequencing are derived from the observation of possible clustering on the surface. It is jwr/òce-clusters at the beginning of words that motivate the generalisation "within an Onset, sonority must increase". No consideration regarding underlying structure of any kind is appealed to in order to establish the properties of a well-formed branching Onset. In those theories that admit branching Codas, the same holds true for these constituents: "within Codas, sonority must decrease" is the expression of a surface-generalisation. So why should the surface tolerate heavy violation of the principles that were built on its own grounds? And why should these principles be unviolable at some earlier derivational stage? Another objection may be raised if a German Coda can host [kt] at the phonetic level as in Jagd [jaakt]: why is it that clusters such as [ktp] or [ktfpslr] may not be covered by a Coda? In other words, why are there no monster clusters of unbounded size at German word edges? Any arbitrary number of unparsable consonants would be made extrasyllabic by the syllabification algorithm, and adjunction rules would link them to Onsets or Codas at some later derivational stage where sonority sequencing does not constrain syllable structure anymore. The question of the number of possible extrasyllabic consonants that are predicted by different theories is further discussed in §373. The point of this section is to suggest that either constituents express regularities regarding the co-occurrence of consonants that are observed on the surface, or they do not. In case they do, it does not make sense to suspend their effect in a certain area of the grammar. They exist in order to exclude non-occurring patterns in a principled way.

440 367

11,5 Principles IV: a better solution for extrasyllabicity than extrasyllabicity

6.5. Extrasyllabic consonants adjoined to the phonological word One way of eluding monster constituents is to adjoin extrasyllabic consonants to some higher node of the tree structure for which no co-occurrence restrictions are defined. A popular candidate for this function is the phonological word. Rubach & Booij (1990a) for example argue in favour of this option: extrasyllabic consonants that cannot be parsed by the algorithm are ultimately linked to this constituent in the way shown under (275). (275)

adjunction of extrasyllabic consonants to the phonological word according to Rubach & Booij (1990a) a. initial b. initial

Polish ssaé [ssatç] "to suck"

Polsih rdza [rdza] "rust"

German Herbst [hexpst] "autumn"

The phonological word is held to dominate syllables and/ or feet. The organisation of these constituents is governed by certain rules, which are discussed for example by Spencer (1996:240ss) and Gussenhoven & Jacobs (1998:247ss). The phonological word is not designed for the accommodation of consonants. In all versions that I can think of, it exists in order to dominate feet and/ or syllables. Therefore, putting up consonants in the phonological word does not fall foul of any generalisation regarding cooccurrence restrictions: this constituent is simply not competent for consonants. However, this way of eluding the abandon of syllabic generalisations has two cumbersome consequences. For one thing, syllable structure exists in order to represent the grouping of segments: only of segments, and of all segments. It is difficult to conceive that all segments are organised by syllable structure, except those that are problematic. Under extrasyllabic analyses that adjoin to higher non-syllabic constituents, the objects that are problematic for syllable theory are kept out-

Doubts on extrasyllabicity and an alterative view

441

side of syllable structure at all derivational stages: when parsed, the supernumerary consonants stand outside of syllabic constituents because they are extrasyllabic. After their integration into the prosodie hierarchy, they still escape the control of syllabic generalisations since they belong to higher units. That is, "misbehaving" items are treated in a way so that they never meet the structure that should govern their existence. The other worrisome consequence is the unwarranted representation of geminates as under (275a). One fundamental insight of autosegmentalism is that long objects are not a consecution of two distinct melodic items. Rather, long vowels consist of a single feature bundle that is associated to two skeletal slots. In the same way, a geminate consonant is one single melodic item that belongs to two different constituents: one leg is dominated by a Coda, the other by an Onset. Monosyllabic geminates where both legs would pertain to a branching Onset or to a branching Coda are alien in natural language. The representation of the initial geminate under (275a) is contrary to both of these principles. Instead of being a single melodic entity that is linked to two skeletal slots, there are two distinct consecutive /s/. Also, they do not instantiate a Coda-Onset sequence. It will be shown in Vol.2,111.6.5.9 how Polish initial geminates may receive a regular representation that stands in no conflict with autosegmental principles.

368

6.6. There are initial and internal s+C effects, but there are only initial extrasyllabic consonants

369

6.6.1. The regular extrasyllabic analysis of initial s+C clusters Word-initial s+C clusters fall into the category of clusters that are unparsable for syllabification algorithms: #spr, #spl, #str, #skr, #skl and their voiced counterparts do not fit within a branching Onset because the fricative [s] is more sonorous than the following stop. Therefore, initial #s is commonly interpreted as an extrasyllabic element in the way shown under (276). This view is expressed in, among others, Steriade (1982,1988), Halle & Vergnaud (1980), Levin (1985), Hall (1992:74ss, 2000:247), Giegerich (1992), Wiese (1988), Ewen & Hulst (2001:136ss).

442 (276)

11,5 Principles IV: a better solution for extrasyllabicity than extrasyllabicity extrasyllabicity of [s] in initial s+C clusters σ

#

Ο

Ν

C

I

I

I

t

ο

ρ

s

It is a well-known fact that [s] is a phonological alien in every possible sense: when phonologists have discovered a regularity, they may be fairly sure that s+C clusters will not obey. Put another way, a generalisation that s+C sequences conform to has good chances to be wrong. The peculiar properties of s+C clusters appear in a wide range of phenomena, which have been rehearsed again and again in the literature. Relevant material has been gathered for example by Kaye (1992a) and in the references quoted. It is obvious that the key to the special status of s+C sequences must be sought in the idiosyncratic properties of [s] : clusters that begin with another fricative such as f+C, Θ+C, χ+C etc. do not show any of the peculiar behaviour that makes the reputation of s+C sequences. For example, there are no words beginning with #fC or #yC in typical Indo-European languages where #sC freely occurs. Another important property of the phenomenology related to s+C is that [s] alone does not behave in any special way. Only the adjacency of [s] and a following stop produces peculiar effects. Among the crowd of individual events, Italian may illustrate this fact. The Italian masculine definite article has three allomorphs, i.e. /', il and lo (e.g. Marotta 1993, Davis 1990). Elision produces /' in case the following noun begins with a vowel: l'amico "the friend". The selection of il or lo is operated according to the initial properties of consonant-initial nouns. The former is chosen iff the noun begins with a single consonant or a cluster of rising sonority, thus il parco, il sole, il treno "the park, the sun, the train". The latter occurs in case the noun begins with a geminate or an s+C sequence: lo zio [lo ttsio] "the uncle", lo gnomo [lo jipomo] "the gnome", lo zero [d3zero] "the zero", lo sporto, lo sbaglio,

lo studio, lo sporco "the sport, the error, the study, the

dirty (one)". Hence, regular il is produced by nouns that begin with an [s] followed by a vowel, while lo is selected in case a stop follows this [s]. The same pattern is produced by another well-known process of Italian phonology, Raddoppiamento Sintattico (e.g. Chierchia 1982, Repetti 1991).

Doubts on extrasyllabicity

370

and an alterative view

443

6.6.2. Word-internal s+C effects It is true that many s+C effects occur at the left edge of the word and may thus be captured in one way or another by appealing to extrasyllabicity. However, s+C effects are also found word-internally. Czech vowel-zero alternations are one case in point. The general pattern is as under (277) below (see §31). (277) Czech vowel-zero alternations C_C-V C_C-0 C_C-CV lokot-e

loket-0

loket-ni

gloss "elbow" GENsg, NOMsg, adj.

That is, zero occurs in open syllables, while [ε] is observed in closed syllables.226 As may be seen, the presence of two heteromorphemic consonants to the right of the alternation site enforces the presence of a vowel. This is also true in case both consonants of a cluster that is preceded by an alternation site belong to the same morpheme. Indeed, there is no word in Czech where the radical vowel of a root that ends in two consonants alternates with zero. Roots of the shape VCeCC have always stable vowels. The only exceptions to this rule are roots that end in s+C sequences such as under (278).227 (278)

371

NOMsg lest kfest öest

GENsg l0st-i ki0t-u C0t-i

gloss "ruse" "baptism" "honour"

6.6.3. All s+C clusters cannot be contour segments either That is, s+C sequences seem to behave as if they were just one single consonant. Ana analysis along these lines would also solve the problem of s+C clusters at the left margin of the word: if they were some kind of contour segment in the spirit of affricates, #str would instantiate only two consonants, [st] and [r], which then could happily be accommodated within a branching Onset. The interpretation of s+C sequences as contour segments 226

227

The detail of this alternation is more intricate, further discussion is provided in chapters 11,9 (§411) and 11,10 (§426). Note that two of them, cest and kfest, lose the [s] when the root vowel is absent.

444

11,5 Principles IV: a better solution for extrasyllabicity than extrasyllabicity

has indeed been proposed for example by Selkirk (1982:346ss), Carr (1993:212), Wiese (1996:42s), Weijer (1994:165ss). However, Kaye (1992a) has gathered a number of s+C effects in word-initial position that lead to the reverse conclusion: the two members of s+C clusters do not cohabitate within the same constituent. Kaye (1992a) therefore concludes that the [s] of initial s+C-sequences belongs to a Coda, while the following stop is dominated by an Onset. This interpretation supposes a word-initial empty Onset followed by an empty Nucleus (cf. §96). Actually, evidence in support of a heterosyllabic analysis of initial s+C clusters has already been mentioned in §369. In Italian, word-initial s+C sequences pattern with geminates: both clusters select lo as the definite article. Hence, it is reasonable to conclude that both geminates and s+C sequences share the same syllable structure, i.e. a Coda-Onset interlude.

372

6.6.4. Conclusion The purpose of this section is not to review evidence regarding s+C effects, nor to evaluate the various proposals that have been advocated. Rather, I wish to make the following point: an extrasyllabic treatment of initial s+Csequences is inaccurate since the same clusters also show peculiar behaviour word-internally. If it is true that only peripheral consonants may be extrasyllabic (§357), there is no extrasyllabic interpretation available for word-internal s+C effects such as under (278). It is clear, however, that there is only one kind of s+C effect. Whatever the key to this phonological Nobel-prize problem, it should be able to apply to all of its instances, initial and internal alike. Another issue is the role that needs to be attributed to the idiosyncratic properties of [s]. There must be something in the internal structure of [s] that makes this sound so special. Hence, there is good reason to believe that an exclusively syllabic interpretation of this segmental peculiarity is on the wrong track: why should a special property of the melodic structure of a segment be encoded as a peculiar syllabic property? Anybody who works on s+C effects needs to explain how the special melodic properties of [s] induce an extraordinary syllabic status, which in turn is held responsible for the effects observed.

Doubts on extrasyllabicity and an alterative view

445

Extrasyllabic analyses do not help in this respect since they do not make any reference to the internal structure of [s].22 Neither are they able to derive both initial and internal s+C effects from a single cause.

373

6.7. Why are there no words with two, nine or twenty extrasyllabic consonants ?

374

6.7.1. The only definition of extrasyllabicity is negative One important formal property of all theories that appeal to extrasyllabicity is the fact that extrasyllabic consonants enjoy a purely negative definition: "is extrasyllabic what cannot be parsed" for reason one (§343), "is extrasyllabic what does not behave like a Coda" for reason two (§358). This stands in sharp contrast with everything that we know from regular syllable structure: all syllabic constituents enjoy a positive definition. A branching Onset is a constituent that accommodates clusters of rising sonority. The opposite sonority slope is required for Coda clusters (in those theories that allow for branching Codas). No such statement can be made for Appendices or, in approaches that do not recognise this kind of constituent, for extrasyllabic clusters that do not pertain to any specific constituent. What is the maximal size of extrasyllabic clusters? Is there any restriction on the sonority of their members? The purely negative definition of what an extrasyllabic consonant is inevitably leads to the conclusion that any number of consonants could be extrasyllabic if the lexicon contained a string of three, eleven or thirty unpayable consonants at a word margin. One major goal of syllable theory is to explain why the clustering at word margins is not random but follows precise regularities. If extrasyllabicity is allowed for by phonological theory on exclusively negative grounds, the set of existing vs. non-existing clusters at word margins is taken back to arbitrary definitions. A theory that contains an extrasyllabic device which roots in a negative definition is unable to tell the occurring clusters from the non-occurring ones in a principled way. If scientific theory is called to be able to explain why the things that surround us do exist, but also why those that never occur are banned from nature, extrasyllabicity ought to beg the question. This is even more so in the generative tradition where the fundamental first-year sentence to memorise says that "the theory must be able to generate all and only the

228

The same comment applies to Kaye's (1992a) Coda analysis.

446

11,5 Principles IV: a better solution for extrasyllabicity than extrasyllabicity

sentences that are well-formed". Extrasyllabic devices can accommodate all clusters that occur at word margins for sure, but it is perfectly unable to exclude any of the non-existing ones. Hence, there is no such thing as a "possible initial or final cluster" under extrasyllabic auspices. Notice that the purely negative and restrictionless definition of extrasyllabic clusters holds true for all derivational stages: the syllabification algorithm cannot parse the five first consonants of an underlying string such as /#fthkrbat/, and thus builds just one syllable that accommodates /bat/, while /#fthkr/ remains extrasyllabic. Phonology operates, and at some derivational stage, the five-membered extrasyllabic cluster is incorporated into the prosodie hierarchy. Depending on the particular theory, this is done by adjunction to either the Onset or some higher prosodie unit such as the phonological word. In neither case is there any restriction on the result of this adjunction: the phonological word may dominate as many consonants of any sonority slope as are submitted, and in case adjunction targets the Onset, sonority sequencing is not applicable at the "phonetic" level.

375

6.7.2. Wild Polish initial clusters are less wild than their reputation Let us consider the detail of one of the "wildest" cases of initial extrasyllabicity that is reported in the literature: Polish word-initial clusters. Among all word-initial situations within Slavic and outside of this language family, Polish initial clusters have probably received most careful attention. The challenge has always been to characterise the occurring sequences as a natural class, and to accommodate them within a more general syllabic theory. This project was undertaken for example by Kuryiowicz (1952), Gussmann (1991b), Cyran & Gussmann (1998,1999), Rowicka (1999a: 257ss), Rubach & Booij (1990a), Rubach (1996, 1997a,b). Let us start the discussion with two-membered initial clusters. A record which ambitions to be exhaustive, reproduced from Rowicka (1999a:309ss), is spelt out in appendix 3 (§622). The way roots and clusters were counted is explained there. Table (279) represents the condensed result of appendix 3 (§622). The existing initial sequences are only a subset of the logically possible combinations. Table (279) demonstrates in which way the existing clusters exactly relate to those that are unattested. Grey-shaded columns recall that s-initial clusters are absent from this table (because of the notorious s+C effect that was mentioned earlier, cf. appendix 3 §622). Those clusters that conform to sonority sequencing (in the large interpretation "C2

Doubts on extrasyllabicity and an alterative view

447

must be more sonorous than Ci") are marked as "+", while sequences that do not qualify as a branching Onset are represented by "—". Empty cells indicate that the corresponding cluster does not occur word-initially in Polish. Hence, the first member of all clusters that are represented by " — a n d only of those, is extrasyllabic. (279) #C!C2: existing vs.jion-existing initial two-membered clusters in Polish Ci ρ I k b d g Is tf tç d î d î d j f v s z J j ç M n i n j i r

Ci ρ

—

k

—

ρ

—

b d

— —

—

i.

—

ts

1tç - —

—

k

—

b d

— —

-

—

dz

—

— JS

ts

1 te

—

—

—

¿3

dz

—

«ζ

d;

—

f + +

+ + +

ν s

l w j

—

++ +

ζ J ++ 3 Ç + * χ ++ m η+ + JJ+ + Γ+ + 1+ + W+ +

d»

—

+

f

+

ν

+

s

+

+

+

+++

+ + + + + + +

—

+

+

+

+ + ++ ++ ++

+ ++++ ++ ++ + ++ +++

— — — — ++ ++ +

—

— : — i

— ,+ . — ; — ! +— +— +—

ζ J 3 e — * χ m — η _ j j r 1 w

j

j ρ t k b d

g ts TJ tç dz d3 dï f ν s ζ J 3 ç ? χ m η ji r

I w j

Looking at the table, it is probably safe to bet quite some money that nobody will manage to characterise either the set of occurring or the set of non-occurring initial clusters as a natural class. Out of 616 logically possible clusters (22 possibilities for Q times 28 possibilities for C2), only 130 are attested (of which 56 respect sonority sequencing, against 74 violating it). The reader may verify that for an occurring sequence, its mirror image may or may not be attested, and that this distribution is perfectly random. For example, why are there words in #mr, but not in *#rm? Why are #tr and its mirror image #rt possible initial clusters, while the mirror image of #pr

448

11,5 Principles IV: a better solution for extrasyllabicity than

extrasyllabicity

does not exist, *#rp? Also, many clusters that are expected to occur do not exist. If a word may begin with a cluster whose first member is a nasal, as is witnessed by #mg, #mj, #mr, #ml and the like, why is there no initial cluster with the unmarked dental: *#nC? All these questions remain unanswered. Only among the classical branching Onsets can some regularity be found: all possible stop-liquid sequences are attested. Before continuing the interpretation of these results, let us consider three-membered initial clusters. All initial sequences of three consonants that are represented in the language appear in table (280) (still following Rowicka 1999a:309ss). They are arranged according to the sonority slope of their elements: if two adjacent members constitute a well-formed branching Onset, they are quoted as a block. Those clusters whose second member is an s-sound (i.e. [s,z,ç,2,^,3]) are mentioned separately.229 (280) three-membered monomorphemic word-initial consonant clusters in stressable Polish roots final branching Onset the second member is an s-sound cluster example cluster example b3d3

bzdienie

tkn^c

b3

bul-ek-a

The challenge raised by this distribution is its disjunctivity. General properties of disjunctions and their consequences for phonological theory are discussed in chapter 11,2 (§303). Theory is called to be able to attribute identical effects to one single cause, rather than to different and unrelated mechanisms. In our case, the question thus arises in which way closed syllables and yers constitute a natural class. This question may also sound "why do yers behave like a consonant, rather than like a vowel?", since CCV patterns with Cb,t, and both are opposed to CV. Yet in other words, one could ask why yers behave as if they were not there: if Cb,iCV reduced to CCV, the closed syllable generalisation would be correct in all instances. In the following section, it is shown how this challenge was approached since the 60s.

415

4. Lower - how it works and what it implies If there is any chance to capture the distribution of vocalised and unvocalised alternation sites in terms of a non-disjunctive statement, the formulation must not include any reference to closed vs. open syllables: the closed syllable analysis is contrary to fact. Hence, the only alternative is to explore

502

11,9 Arg. 4 Lower: empty Nuclei and internuclear relations in Slavic

the opportunity of generalising the yer context: a vowel appears in an alternation site if and only if a yer is present in the following syllable. This is the essence of Lightner's (1965) proposal, which is known as "Lower". The rule appears under (302), where yers are transcribed according to classical ν w 251 generative practice, b=i, ί>=ι. (302)

Lower ï,ï^e,o/_C0{tf}

In Lightner's view and in all subsequent analyses that use a version of Lower, yers are vowels that are present in underlying representations and appear in various colours throughout Slavic languages. In some idioms, the front-back opposition is carried onto the surface, while in others both yers merge into one single phonetic object. For instance, Western Slavic merges both yers into [ε], Eastern Slavic transforms /ϊ,ϊ/ into [ε,ο], respectively, while both yers appear as [a] in South Slavic252: compare /din/ "day" and /sin/ "dream" in Polish (dzien, sen), Russian (den1, son) and Serbo-Croatian (dan, san). For theory-internal reasons that are of no interest for the present demonstration25 , as well as for the sake of their diachronic identity (Common Slavic b,T> < IE i,u), yers were located in the central upper part of the vocalic triangle, enjoying synchronic descriptions such as "high front centralised lax vowel" for lïl and "high back centralised lax vowel" for III. Throughout derivation, yers could have two fates: either they were lowered to some mid or low vowel (according to the language at hand), or they were deleted.254 The latter event was said to take place after the former. Thus, a 251

252

253 254

The historical object "yer = vowel in Common Slavic" had appeared in synchronic analysis since Halle (1959). These are only general indications. The detail language by language is much more complicated. In Slovak for example, various vocalisations can be found, cf. Rubach( 1993:134ss). See e.g. Gussmann (1980a:39ss,63ss), Rubach (1984:28s) on that issue. The debate on the kind of deletion that occurs (stray erasure vs. erasure by rule) is not relevant here; see Rubach (1984:184ss, 1993:140s) for discussion. Note that even in case both yers are merged into the same phonetic result, it may be necessary to maintain two distinct underlying identities. In Polish for example (see §§450,452), the front yer leaves a trace on the preceding consonant /din/ - * [diep], whereas the back yer does not /sin/ —» [sen] (but see Gussmann 1992a for a solution that favours underlying palatalised consonants).

Lower - how it works and what it implies

503

yer could never be surprised in its underlying cloths on the surface: the only phonetic trace of yers were said to be non-high vowels. This is where the name of Lightner's rule comes from: it lowers high vowels. Another important property of yers is their unpredictable distribution across the lexicon: they appear in roots and affixes just as any other vowel does. Their occurrence may not be predicted by reference to any other parameter. For instance, the surface representatives of yers may be phonetically identical with regular vowels whose quality has not been acquired through Lower, and which do not alternate with zero: compare e.g. Czech pes [pes] - psa [psa] "dog NOMsg, GENsg" with les [les] - lesa [lesa] "forest NOMsg, GENsg". Hence, yers must be present in the lexicon, and insertion strategies that epenthesise a vowel in the appropriate syllabic environment do not qualify.255 Also, yers must be underlyingly different from those vowels which are phonetically identical but do not alternate, i.e. /pis/ for "dog" vs. /les/ for "forest" in the above example. Let us now return to the issue that was raised at the end of the previous section. We have seen why Lower was proposed, but we have not yet discussed its consequences. The chief incidence of Lower on underlying structure is evident from simple cases such as /pis/, which reaches the surface as [pes] "dog NOMsg". If it is true that yers are vocalised in case they are followed by another yer, the word-final consonant must be followed by a yer: only /pisi/ can be turned into [pes] via Lower. Therefore, all consonant-final words were assumed to end in yers underlyingly: the yers in question was attributed the morphological value of a case marker (i.e. NOMsg in (303b), GENpl in (303c)).256 In other words, Lower enforces the existence of underlying (wordfinal) yers that never appear on the surface. The derivation under (303) shows how a sequence of two alternation sites is treated by Lower on these

255

256

Treatments relying on insertion have been proposed by, among others, Laskowski (1975), Czaykowska-Higgins (1988), Piotrowski (1992a). They are convincingly refuted in Gussmann (1980a:26ss), Rubach (1984:28s, 1993: 134ss), Szpyra (1992a:280ss, 1995:94ss). Bethin (1992) advocates a compromise that combines both epenthesis (in borrowings) and underlying specification. See Gussmann (1980a:36ss), Rubach (1984:41ss) for discussion. The choice of the front or back version of the yer is usually determined by its palatalising effects.

504

11,9 Arg. 4 Lower: empty Nuclei and internuclear relations in Slavic

assumptions. The example used is the Polish word for bread roll bulka (see (299). 257 (303)

two alternation sites in a row: derivation with Lower yerunderlying —» Lower —• deletion —> surface bul-ecz-Ik-a bui-ecz-k-a buleczk-a a. buMcz-Tk-a butek bul-ek b. bul-ïk-ï buJ-ek-i bulecz-ek c. buMcz-ïk-ï but-ecz-ek-ί but-ecz-ek but-k-a butk-a d. buMk-a buMk-a

relevant yer occurs in C yer C V _ C yer # C yer C yer CV

Note that under (303c), Lower must apply two times and from left to right (or from the root towards the periphery) in order to transform //but-icz-ïk-ï// into /bul-ecz-ek-ï/. This effect is commonly viewed as a consequence of the cyclic application of Lower, which was introduced by Rubach (1984:184ss) (see also Rubach 1993:139s). 258 According to Lower, thus, there are two ways of identifying the presence of an underlying yer: yers exist 1) in locations where a vowel alternates with zero and 2) after word-final consonants. The former may appear on the surface under certain conditions, while the latter never enjoy a phonetic existence. We can already note at this early stage of the discussion that this pattern exactly coincides with the distribution of empty Nuclei in Standard Government Phonology (see (297)). The following section examines the diachronic reality of yers.

257

258

Certain aspects of the derivations that are not relevant here, cf. Gussmann (1980a:41ss), Rubach (1984:184ss). They are left unmentioned. Contrary to that, Gussmann (1980a:30) and Anderson (1974) advocate a noncyclic interpretation: "the string is first scanned for the [alternating] segments; once these are identified, the change is implemented simultaneously" (Gussmann 1980a:30).

The difference between Lower and Havlik

505

416

5. The difference between Lower and Havlik

417

5.1. Lightner makes modern Slavic look like Common Slavic underlyingly, yet Lower is not exactly like Havlik In 1889, Antonin Havlik discovered the following diachronic regularity when comparing Common Slavic (CS) to Old Czech (ocz).259 (304)

Havlik's Law a. given a sequence of consecutive yers in Common Slavic, every other yer survives into Old Czech, counting from the right edge of the sequence. b. illustration thereof 4 3 2 1 4 J 2 Y CS s t pbs-imb > ocz se p0s-em0 se psem "with the dog" 5 4 3 2 1 S1 4 3f 2 Y CS si. äbv-bc-bmb > ocz se äev-ec-ema s Sevcem "with the shoemaker"

It was soon understood that Havlik's Law, allowing for some variation, in fact underlies all Slavic languages. It covers the third line of the synchronic generalisation (301) which states that a yer is vocalised if it occurs before another yer. While both Havlik's Law and (301) define the vocalisation of yers according to the presence or the absence of a following yer, they do not implement this condition in the same way. The former qualifies a yer for vocalisation if and only if it is 1) followed by another yer and 2) even-numbered in a yer chain. By contrast, while the latter does not inform on any condition regarding even- or odd-numberedness. This is the first remarkable difference between Havlik on one hand and the synchronic generalisation (302) on the other. The second one concerns directionality: Havlik attributes odd and even numbers of a yer chain from right to left, whereas Lower (302) must apply cyclically in case it encounters several yers in a row. In terms of directionality, however, cyclic means from left to right: //bul-ïcz-ïk-ï// comes out as /bul-ecz-ek-i/ only after recursive application of Lower first to the leftmost, then to the intermediate yer (cf. §415).

259

See general descriptions of the evolution of Common Slavic yers for example in Trávníóek (1935:46ss), Lamprecht et al. (1986:46ss), Liewehr (1933:91ss).

506

11,9 Arg. 4 Lower: empty Nuclei and internuclear relations in Slavic

In sum, Havlik's generalisation amounts to saying that evennumbered yers are "strong", while odd-numbered yers are "weak". The former are vocalised, the latter fall out.260 The two first lines of the synchronic generalisation (301) are irrelevant for Common Slavic and Havlik since Common Slavic lacked closed syllables altogether: all words end in a vowel, and sequences of consonants are only of rising sonority.261 The move that is taken by Lower thus makes the underlying structure of modern Slavic languages look like Common Slavic: all words end in a vowel (words that are consonant-final on the surface possess an underlying word-final yer with morphological value), and Codas do not exist in alternation environments (Coda clusters that host a vowel-zero alternation are separated by a yer in underlying representations). Therefore, it is only consequent to assimilate the grammar of modern Slavic languages to the Common Slavic state of affairs in yet another respect: when compared to the distributional generalisation (301), Lower (302) does away with any reference to closed syllables. The only parameter that drives yer-vocalisation is the presence or the absence of a yer in the following syllable.

418

5.2. The difference: "secondary vocalisation", that is "every other" (Havlik) vs. "all but the last" (Lower) Given all these similarities, why is it, then, that Lower and Havlik's Law are different with respect to directionality (cyclicity) and the mention of odd and even numbers? The answer is actually contained in table (299), of which relevant parts are repeated under (305). Also, diachronic information regarding Old Czech and Old Polish is added.262 260

261 262

These are the common terms in the philological tradition as exposed in, among others, Koschmieder (1958), Panzer (1991:277ss), Vaillant (1950:124ss), Trávniöek (1935:46ss), Lamprecht (1987:137ss), Lamprecht et al. (1986:46ss), Nahtigal (1961:96s), Liewehr (1933:91ss), Carlton (1991:165ss). See for example Carlton (1991 :lOOss), Vaillant (1950:284ss), Martinet (1952). Unless otherwise indicated, the forms of the modern and the old languages coincide. The Old Czech pattern is described in the literature that is discussed after (305); Rospond (1979:74) for example informs about the situation in Old Polish pattern. Glosses: Czech "house" dim.GENsg, double dim. NOMsg, dim. NOMsg, double dim. GENsg; Polish: "dog" GENsg, dim.NOMsg, NOMsg, dim.GENsg.

The difference between Lower and Havlik

507

(305) vocalisation in open syllables: diachronic situation closed syllable vowel C c-cv iC C-0 dom-eò-0k-u ¡dom-ek

open syllable

Czech

zero C C-V dom-0k-u

Polish

pes-a

C C-ver C0 ocz dom-eC-ek mcz dom-eC-ek οροί p0S~ek mpol pies-ekg

sài ÜSl

¡pies

pies-0k-a

»

»

All diachronic grammars report that Havlik's Law correctly describes the state of affairs in the old languages (Old Czech, Old Polish etc.) which continue Common Slavic, but that "secondary vocalisations" have disturbed the picture since then. The forms where "secondary vocalisation" occurs are precisely those of the offending grey-shaded column (recall (299)) where a Common Slavic yer chain has been reduced according to Havlik's Law: CS dom-bò-bk-b > ocz dom-eí-ek-e > mcz dom-eC-ek-e. The third yer (counted from the right margin) is odd-numbered and thus weak. It should fall out, which it does indeed in Old Czech. However, it is "restored" in Modern Czech. Philologists always invoke analogy in order to explain this "secondary vocalisation" of weak yers. Examples of this attitude are Trávniöek (1935:47), Komárek (1962:48), Liewehr (1933:98) and Vondrák (1924:180). However, the statement (300) "alternation sites are vocalised in open syllables iff the following vowel alternates with zero itself' is absolutely true for the entire language. This fact points to phonological, rather than analogical activity. The analogisation of the entire language according to a perfectly regular phonological parameter can hardly be granted credit.263 Supposing analogical activity in this case is even more bewildering when considering the fact that several modern languages have operated exactly the same "secondary vocalisation" of yers in identical conditions, cf. (299). One of the challenges of this chapter is to seek an explanation which is not lexical, analogical or accidental, but phonological.

263

To the best of my knowledge, there is only one root in Czech that disregards the generalisation expressed under (300), i.e. where yers still alternate according to Havlik in the modern language: Sev "seam", which derives Sv-ec and not *ievec "shoemaker", the genitive of which is Sev-ec-e. Strangely enough, this hapax is always used in diachronic textbooks in order to show the modern reality of Havlik's Law, cf. for instance Lamprecht et al. (1986:47), Trávníóek (1935:46), Liewehr (1933:95).

508

11,9 Arg. 4 Lower: empty Nuclei and internuclear relations in Slavic

The "secondary vocalisation" discussed is the only real difference between Lower and Havlik: two consecutive yers are never both vocalised in the latter regularity, but do appear on the surface in the former pattern. Both statements describe realities which are slightly different but share the feature of solely depending on the presence vs. absence of a yer in the following syllable. Neither Havlik's Law nor Lower are influenced by any other parameter. That is, directionality (cyclicity) and odd-/ even-numberedness also define the relation with the following yer. In fact, as will be argued in §468, Havlik and Lower describe the two patterns of vowel-zero alternations that occur in natural language. They may be informally stated as under (306) below. (306) the two patterns of vowel-zero alternations a. Havlik given a chain of alternation sites, vocalise every other one, counting from the right margin. b. Lower given a chain of alternation sites, vocalise all of them save the last one. It thus appears that Lightner has simply made Common Slavic the underlying structure of modern Slavic languages, and adjusted Havlik's Law according to the "secondary vocalisations" discussed. This was current practice in early generative endeavour, where synchronically underlying representations strangely but precisely coincided with earlier stages of the language. 264 Before we can concentrate on the crucial importance of Lower for the purpose of the argument to be made, some additional diachronic information needs to be introduced in the next section.

264

Nobody could beat Theodore Lightner in this sport: he was not joking when he proposed that the English word tooth is synchronically derived from /H3d/ by modern English natives, where H3 is the Indo-European o-colouring laryngeal (Lightner 1978:26). See §333 for more engineering of that kind.

The difference between final and alternating yers

419

6. The difference between final and alternating yers

420

6.1. Synchronic evidence

509

We have seen in §415i that there is good reason to distinguish two kinds of yers: those that sometimes appear on the surface, and those that may never be observed in phonetic cloths. This contrast actually allows to detect yers at the underlying level. (307) yers exist underlyingly a. alternating yers in locations where a vowel alternates with zero. b. final yers after word-final consonants

The former category may be called alternating, the latter final yers. We will see in the following section that this distinction also has a diachronic reality. More synchronic evidence in support of it will be introduced in §428.

421

6.2. Diachronic evidence: only alternating yers may originate in epenthesis A good deal of the philological literature holds that alternating vowels in modern Slavic languages always originate in yers: if a vowel alternates with zero, it was a yer in Common Slavic. This is a legend.265 An alternating vowel in a modern Slavic language may have two Common Slavic sources: either a yer or nothing. In other words, some modern alternating vowels were born through epenthesis. The yer-origin does not need any specific illustration. Relevant evidence regarding the epenthetic origin of

265

Which is provoked by the antipathy against non-etymological objects that is widespread among philologists. See Scheer (1996:92ss) for discussion. Even though diachronic grammars such as Gebauer (1894-98 I:154ss), Trávníóek (1935:230), Trávníóek (1948-49 I:41ss), Havránek & Jedliöka (1988:31) clearly expose the fact that alternating vowels may originate either in a yer or in epenthesis, the same authors sometimes properly invent yers or recur to analogical explanations in order to be able not recognise an object that is bare of any etymological existence. For GENpl forms, instead of admitting epenthesis, Gebauer (1894-98 1:160), Gebauer (1894-98 II,139ss), Trávníóek (1935:230), Lamprecht (1987:138) and Komárek (1962:150) for example invoke analogy with yer-bearing items.

510

11,9 Arg. 4 Lower: empty Nuclei and internuclear relations in Slavic

alternating vowels in Czech is provided under (308). The group of words concerned is identified according to the traditional classification of stems that relies on the thematic vowel in Indo-European. (308) epenthesis of "yers" a. feminine -i stems in NOMsg266 CS case-suffix -b. Cause: loss of a yer in the following syllable Modern Czech Common Slavic píseñ - pisen-ë "song NOMsg, GENsg" NOMsg pë-sn-b báseñ - bàsen-ë "poem NOMsg, GENsg" NOMsg ba-sn-b b. neuter o-stems and feminine a-stems, both in GENpl267 CS case-suffix -τ>. Cause: loss of a yer in the following syllable Modern Czech Common Slavic óísel-o - ôisel η. "number NOMsg, GENpl" GENpl Cit-sl-i. sester-a - sester f. "sister NOMsg, GENsg" GENpl sestr-T> c. some masculine o-stems in NOMsg268 CS case-suffix -i. Cause: loss of a yer in the following syllable Modern Czech Common Slavic mozek - mozek-u "brain NOMsg, GENsg" NOMsg mozg-i

266

267

268

Material on feminine i-stems is exposed for example in Váíny (1963:73ss), Vondrák (1924:478ss), Trávníóek (1948-49 1:43), Trávníóek (1935:230), Gebauer (1894-98 I:160ss), Gebauer (1894-98 II:343ss), Arumaa (1985:49ss,120ss), Havránek & Jedlióka (1988:149ss), Vaillant (1958:142ss). On epenthetic vowels in genitive plural forms, see e.g. Gebauer (1894-98 1:160,165), Gebauer (1894-98 II:139ss), Panzer (1991:324), Válny (1963:45,61ss), Trávníóek (1948-49 1:44), Trávníóek (1935:230), Arumaa (1985:68ss,141s), Lamprecht (1987:138), Komárek (1962:128s,150), Vaillant (1958:35s). On these stems, see for example Vondrák (1924:344), Trávníóek (1948-49 1:43), Gebauer (1894-98 1:160).

The difference between final and alternating yers (308) epenthesis of "yers" d. prepositions and prefixes269 Modern Czech vze-pnout se - vze-pínat se "to straighten up pf, ipf ' roze-psat - roze-pisovat "to begin to write pf, ipf' beze-dny - beze-brady "without bottom, without beard" ode-mknout - ode-mykat "to open (key) pf, ipf'

511

Common Slavic *ντ>ζ*orz*bez*od-

It is not difficult to see that epenthesis occurred when a yer fell out in the following syllable. All cases quoted under (308) follow this pattern.270 This behaviour of course hints at Havlik's Law: in the period of its activity (CS > particular Slavic languages), vowels appear out of nothing iff a yer in the following syllable was lost. Both in case of yer chains and when an epenthetic vowel emerges, the vocalisation is identical, e.g. [ε] in Czech and Polish. In other words, the loss of a yer causes the vocalisation of either a yer, or of nothing in the preceding syllable. If identical causes produce identical effects, the "nothing" at hand and the yer must share some property. What could that be? Can nothing be a yer? No. This is what some philologists have tried to do: inventing yers (see notes 265 and 269). Can a yer be nothing? Yes, of course. We know that yers were "fading away". That is, they were first centralised, and then fell out. Hence, the objects that were vocalised are not yers, as is commonly believed. Rather, the items that undergo vocalisation are "nothing", or zeros. Zeros of vocalic nature, to be precise. In autosegmental representations, a "vocalic zero" of course can be nothing else than an empty Nucleus.

269

270

The effects of the antipathy against non-etymological objects appear most strikingly in the treatment of prefixes and prepositions. In answer to the question "which prepositions/ prefixes were terminated by a yer?", almost anything and its reverse can be found in the literature. The item roz(e) is identified as *orzi in etymological dictionaries (Machek 1957:424, Holub & Lyer 1978:391), although no yer can be established on the basis of either comparatism or Old Church Slavonic texts. The same holds true for *otb > od(e) (Lamprecht et al. 1986:332ss). Machek (1957:579) invents yers when a vowel-zero alternation without etymological yer-source has to be brought back to yer regularity: for example, he derives ocz vzezvati from CS *vbz-i-zbvati, identifying the yer between the prefix and the stem as "added" ("pfidáváno ι , dávající e"). Including the prepositions and prefixes mentioned under (308d), see Scheer (1996).

512

11,9 Arg. 4 Lower: empty Nuclei and internuclear relations in Slavic

Hence, it appears that CVCV makes a correct prediction: the existence of empty Nuclei is assumed exactly where epenthesis occurs. No special provision needs to be made in order to insert vowels: they simply fill in empty Nuclei that have always existed. These empty Nuclei have either always been empty (cases of "epenthesis"), or they contained a yer and were emptied as the yers faded away (cases of "yer vocalisation"). Empty Nuclei of both origins then were vocalised iff the vowel in the following syllable fell out (because it was a yer). Hence, Havlik's Law does not only concern yer chains, its scope is larger: objects that are vocalised are former yers and former nothings. This means that a slight modification of Havlik's Law is in order, at least for the Czech situation. This adjustment, as well as a summary of the preceding discussion, is provided under (309). (309) the story of real and "epenthetic" yers a. in late CS, yers were centralised and fell out. b. they were not vocalised but dropped, leaving behind an empty Nucleus. c. empty Nuclei were vocalised iff the following Nucleus was empty (because it contained a yer that fell out). d. hence, b,t> > 0 > vowel (and not: b,t > vowel) e. Havlik's Law conventional formulation "given a sequence of consecutive yers in Common Slavic, every other yer is vocalised, counting from the right edge." reformulation "given a sequence of consecutive empty Nuclei in Common Slavic, every other empty Nucleus is vocalised, counting from the right edge."

This reformulation of Havlik's Law describes the Czech state of affairs, i.e. the language on which the Law was originally built. In other Slavic languages, epenthesis has been less regular, and hence (309e) cannot be applied.271 I will come back to the phonological interpretation of the version of Havlik's Law that suits Czech in §468.

271

For example in Polish: pol NOMsg siostra - GENpl sióstr = cz sestra - sester; pol basñ = cz báseñ etc.

The difference between final and alternating yers 513 422

6.3. Conclusion: Common Slavic yers and modern abstract vowels This being clarified, we may now return to the main purpose of this section: alternating vs. final yers. It was shown that alternating yers are not always etymologically real; they may originate in a vowel, or in nothing. By contrast, final yers have only one single source: yers. In all instances that have been reviewed above, the case marker whose loss triggers the vocalisation of the preceding empty Nucleus is a yer (NOMsg of fem i-stems and mase o-stems, GENpl of fem a-stems and neuter o-stems). The synchronic assumption that word-final yers are actual case markers (Gussmann 1980a:36ss, Rubach 1984:41ss) is thus a diachronic fact. Hence, it appears that the contrast between alternating and final yers which has been established on purely synchronic grounds is paralleled in diachrony. Also, it follows that it is necessarily erroneous to view the modern alternations as a simple reflex of the diachronic situation: the object called "yer" that we deal with in the formulation of Lower is different from the historical vowels b,T> that occurred in Common Slavic. Lower includes etymological "nothings", whereas Common Slavic yers do not. The yers that are supposed to have a modern existence in underlying representations are abstract theoretical vowels (Gussmann 1980a, Rubach 1984:28), rather than the modern version of a diachronic reality.272 The following section shows how Lower was adapted to the autosegmental environment in the mid-80s.

423

7. Autosegmentalised Lower Lower and its presuppositions have been subject to substantial modifications in the mid-80s when autosegmental representations became a settled part of phonological theory. The autosegmentalisation of Lower has been operated by Spencer (1986), Rubach (1986) and Kenstowicz & Rubach 272

Further clarification of the use of the word "yer" is in order here. Yers are two Common Slavic vowels that possess a certain phonetic identity. They have always been called "yers", and so are they here. For obvious reasons, the same word is widely used in the synchronic analysis of Slavic languages in order to refer to the vowels that alternate with zero and those that are assumed in wordfinal position. Using the word "yer" in synchronic analysis is handy and does not mean that the synchronic and the historical objects are mixed up. It does, however, entertain a terminological confusion. It should therefore be borne in mind that "yer" means only "abstract vowel" in a synchronic context.

514

11,9 Arg. 4 Lower: empty Nuclei and internuclear relations in Slavic

(1987). 273 Since the first article mentioned is autosegmental, but not a faithful instantiation of Lower, it is not discussed in the present section. Spencer (1986) will be introduced below in §449. Recall from §415d that the distribution of alternating yers is unpredictable. Their occurrence in a root or a morpheme is a lexical property of every single item. In non-autosegmental terms, the only way to express that two vowels are different in underlying representations is to make them contrast in quality. Hence, a six-vowel system such as the one encountered in Polish ([i,u,i,e,o,a]) will have to be augmented by two yers, whose melodic identity must not coincide with any of the six existing vowels. The traditional solution was to make yers high vowels for historical reasons, but which were attributed a [-tense] feature that isolated them from the other three high vowels. Polish was thus believed to possess no less than five high vowels: /i,u,i,T,'i,8,o,a/.274 In autosegmental representations, a vowel that enjoys phonetic expression is defined as the association of a melodic unit and an x-slot, which in turn is dominated by a syllabic constituent. If there is an x-slot but no melody, nothing is heard (empty Onset or Nucleus); if there is a melody available but no x-slot, no phonetic trace of this melody will appear (e.g. floating consonants of the French kind); finally, if both melody and x-slot are present but remain unassociated, nothing is heard either. These basic principles of autosegmentalism do not call for any special motivation here. Autosegmental representations thus offer an alternative way of making yers different from other vowels: their peculiar properties may be encoded structurally, rather than melodically. It was not a very lucky idea anyway to attribute the peculiar phonotactic behaviour of yers to some melodic property of theirs. This is ignoring and mismatching causalities: the reason why yers alternate with zero is a lateral relation that they do or do not contract with the following vowel (cf. the next section); none of their supposedly peculiar melodic properties plays any role at all. The alternative explored by Rubach (1986) and Kenstowicz & Rubach (1987) grants a melodic, but no skeletal identity to yers: yers are floating melodies that do not possess any skeletal anchor. The corresponding underlying representations are shown under (310) below for the three

273

274

Further exploration of the analysis that is discussed in the two latter articles may be found in Rubach & Booij (1990a,b). The exact situation is still more complicated than that, see Gussmann (1980a: 63ss), Rubach (1984:27ss,139ss).

A utosegmentalised Lower

515

relevant distributional situations, which are illustrated by familiar Czech data. (310)

yers are floating matrices: Rubach (1986) Czech "elbow" a. lokt-e GENsg b. loket NOMsg

c. loket-ni adjective

X X X

X X

X X X

Χ

X X X

Χ

X X

I I I

II

I I I

I

I I I

I

II

l o k e t e

l o k e t e

l o k e t e n i

This option offers several advantages. First, there is no need anymore to locate the melodic identity of yers in the high area. Yers and nonalternating [ε,ο] are identical as far as their melody is concerned. The only difference comes from the fact that the latter are associated to a skeletal slot, while the former are not (observe the difference between the two s under (310a)). Hence, the acrobacy related to the feature [±tense] can be dispensed with, and the Polish system of underlying melodies is driven back to the regular six items which appear on the surface. Another advantage is that the rule of yer deletion can be dispensed with: the phonetic absence of unassociated melodic material is automatic under autosegmental auspices (stray erasure). Finally, vowels that alternate with zero are predicted not to be subject to any melodic restriction anymore: in the earlier system, at most two vowels that were distinct on the surface could alternate with zero: this is the case in Eastern Slavic. But what about languages like Slovak where more than two vowels alternate with zero? Relevant facts are reported by Rubach (1993:139ss): o-zero bahor - bahra "belly NOMsg, GENsg", e-zero sev svu "seam NOMsg, GENsg", a-zero jedlo - jedàl "food NOMsg, GENpl". In autosegmental terms, the associated representations are straightforward: whatever vowel alternates with zero, its underlying representation is simply its floating melodic identity without a skeletal slot. Any alternating vowel will thereby be different from its melodically identical peer that does not alternate because it is associated to a skeletal slot. By contrast, every additional vowel that alternates with zero augments the inventory of underlying melodies in the pre-autosegmental system. In cases such as Slovak, then, the historical motivation for two different yers breaks down. The autosegmental expression of Lower is shown under (311) below.

516 (311)

11,9 Arg. 4 Lower: empty Nuclei and internuclear relations in Slavic autosegmentalised Lower (Rubach 1986, Kenstowicz & Rubach 1987)

χ 0

—

V

/ _C

©

Yers are circled. A skeletal slot is associated to a yer iff this yer is followed by a consonant and another yer. The cyclic application of the autosegmentalised version of Lower transforms the underlying floating melodies of (310) into the attested surface forms. That is, it promotes all members of a chain of yers except the final one to a phonetic existence. The following section evidences the lateral essence of Lower.

424

8. Slavic vowel-zero alternations are caused by a lateral relation There is a crucial difference between Lower, which recurs to "invisible" yers, and the distribution (301) that is read off the surface. This fundamental difference has gone completely unnoticed in the literature on Slavic vowel-zero alternations. Under (301), indeed, the reason why vowels alternate with zero is of paradigmatic nature: syllable structure determines whether an alternation site is vocalised or not; in a closed syllable, vowels occur, while nothing is visible in open syllables (provided the following vowel does not alternate with zero itself). By contrast, Lower denies any causal relation between syllable structure and the vocalisation of alternation sites: the only information that is needed in order to calculate the phonetic value of alternation sites is of lateral nature. Either the following vowel is a yer, or it is not. If it is, the alternation site is vocalised, if it is not, the site remains phonetically unexpressed. This situation is summarised under (312) below.

Slavic vowel-zero alternations are caused by a lateral relation 517 (312) the causality of Slavic vowel-zero alternations is lateral a. vowel-zero alternations are not triggered by the presence or the absence of a consonant in a given syllable (Coda analysis). Rather, it is controlled by an intervocalic communication. b. this intervocalic communication involves two yers whereby the rightmost yer determines the phonetic status of the leftmost yer: e.g. Czech pes "dog NOMsg"

Í ρ

ï

S

I

}

^ vocalisation ε The arrow under (312b) materialises the intervocalic causality that is the very essence of Lower, but which curiously has always remained unmentioned in the literature on yers. As far as I can see, the enormous body of writings on the matter never makes explicit that we are in presence of a communication between two neighbouring vowels, and that this lateral relation is regressive: the rightmost vowel performs an action on its preceding peer. An arrow such as under (312b) never appears in the literature, which means that the most important property of Lower is neither discussed nor represented. Saussure's (1915:23) "le point de vue crée l'objet" [the point of view creates the object] may thus be an accurate description of the situation at hand: the lateral and regressive identity of the object Lower was not perceived by those linguists who created it because it had no function in their world-view of phonology. By contrast, it springs immediately to the eye of somebody who is looking at Lower through glasses which are biased by Government Phonology and a lateral world-view.

425

9. Conclusion In this chapter, the analysis of vowel-zero alternations in Slavic languages until the beginning of the 90s has been retraced. With minor revisions, the core of the reasoning still stands up to present-day standards. The synchronic analysis of Slavic vowel-zero alternations, developed prior to Government Phonology and for entirely independent reasons, relies on the existence of "abstract vowels" that are called yers. The goal of this

518

11,9 Arg. 4 Lower: empty Nuclei and internuclear relations in Slavic

section was to show that these yers have exactly the same distribution as empty Nuclei in Standard Government Phonology. They occur 1) in locations where a vowel alternates with zero and 2) after word-final consonants. Since it is not very difficult to equate "abstract vowel" with "empty Nucleus" when autosegmental structures are assumed, it may be hoped that empty Nuclei in the end appear to be a good deal less outlandish than they usually do in the phonological scene. Autosegmental structures call for their existence anyway (chapter 11,7 §387), and the present chapter shows that there is good empirical reason to believe that they are real since they allow for the unified expression of otherwise disjunctive causalities (the yer context (301)). The analysis that Lower embodies of course is not carried out in the framework of Government Phonology, nor in its spirit. The empirical problem at stake simply calls for the abstract vowel analysis. Government Phonology on the other hand provides the tools that allow to take the Slavic facts to a level where they do not need to be considered as an idiosyncratic pattern that is specific to this group of languages: they could well exist in languages whose ancestor is not Common Slavic. That is, yers are exclusively Slavic, but empty Nuclei are universal. The following chapter develops this line of reasoning. The second genuine property of Government Phonology, lateral relations (cf. chapter 1,8 §165), is also constitutive of Lower, even though this fact has gone entirely unnoticed in the literature: it had no function in the analysis (see the previous section). The point that is made in this chapter thus is clear: two central devices of Government Phonology, empty Nuclei and regressive intervocalic relations, have been used for over thirty years in order to analyse Slavic vowel-zero alterations. The motivation for their use was strictly empirical: the triple-disjunctive yer context (301) cannot be otherwise reduced to a non-disjunctive statement. Hence, it appears that Government Phonology has proposed theoretical tools that had been used for a long time without being aware of this fact. On the other hand, analysts of Slavic have been using them without knowing that they have received a theoretical status in the meantime. Could the phonological essence of these tools be so very mistaken if they have been invented twice, for different reasons, on different empirical grounds and with no awareness of the sister development?

Chapter 10 426 Argument Five The life of "yers" outside of Slavic and in locations where vowels do not alternate with zero

Audience: people who do not believe that all Coda-Onset clusters are separated by an empty Nucleus Conclusion regarding syllable structure: all RT sequences are separated by an empty Nucleus.

427

1. Introduction In this chapter, phenomena from Slavic and other languages that do not engage alternations between a vowel and zero are discussed. They all share the fact of being conditioned by the familiar yer context (301). I argue that a unified analysis which is able to cover all events that are controlled by the yer context requires the move from Standard Government Phonology to CVCV. Slavic vowel-zero alternations are not the only phonological processes that are conditioned by the aforementioned context (301) "in closed syllables and in open syllables if the following vowel alternates with zero". A number of other Slavic alternations are governed by this structural description as well. The same context is also active in French where it drives the ATRness of mid vowels and the alternation between schwa and [ε]. If the recurrence of this highly specific structural description is not to be regarded as accidental, a unified analysis of all alternations mentioned is called for. Such an analysis, I argue, may only be achieved if all groups that are traditionally viewed as Coda-Onset sequences are interpreted as distinct Onsets which enclose an empty Nucleus.

520

11,10 Arg. 5 The life of'yers" outside of Slavic

428

2. Yers all over the place in Slavic?

429

2.1. The general picture The evidence that has led Lightner, Gussmann, Rubach and others to seek a non-disjunctive analysis of Slavic vowel-zero alternations is convincing: there is only one single cause for the vocalisation of a yer, that is the existence of another yer in the following syllable. The cornerstone of the entire venture that allows for this excellent insight has been pinpointed in §424: Slavic vowel-zero alternations have got nothing to do with syllable structure; they are the consequence of an intervocalic relation between two yers. This causality being borne in mind, let us review some additional evidence from Slavic. Under (313), consider four well-known segmental events from Czech and Polish where different melodic effects are conditioned by the yer context (3Ol).275 (313) the distributional pattern "in closed syllables and in open syllables if the following vowel is ayer" (301) extends beyond vowel-zero alternations open syllable closed syllable C_C-V C C-yer C_C-0 C_C-CV a. Czech VV-V b. Czech û-o c. Polish ó-o d. Polish q-ç

¿áb-a jmén-o ηοέ-e krov-a zfb-a

iab-ek nûi-ek króv-ek z^b-ek

¿ab jmen ηύέ króv zqb

2ab-0k-a jmen-ny nû2-0k-y króv-0k-a zçb-ak-a

The following sections discuss these alternations one by one.

275

Polish and Czech spelling conventions are explained in §7 at the outset of this book. In Polish spelling, a hook under a vowel notes its nasal character, whereby ^ is a nasal [o], and ç a nasal [ε]. The phonetic value of nasal vowels in Polish depends on the following segment. Roughly speaking, [εΝ], [oN] appear before a stop ("N" is a homorganic nasal consonant that shares place with the stop), while [ε\ν], [5w] surface elsewhere. §598 discusses this pattern at length (descriptions are also available for example in Ostaszewska & Tambor 2000:53ss and Nagórko 1998:35ss). This allophony is of no relevance for the present discussion. Glosses: a) "frog" NOMsg, dim. GENpl, GENpl, dim. NOMsg; "name" NOMsg, GENpl, adj.; b) "knife" GENsg, "scissors" (i.e. dim. of "knife") GENpl, "knife" NOMsg, "scissors" NOMpl; c) "cow" NOMsg, dim. GENpl, GENpl, dim. NOMsg; d) "tooth" GENsg, dim. NOMsg, NOMsg, dim. GENsg.

Yers all over the place in Slavic ? 521 430

2.2. Czech alternations in vowel length The alternation in vowel length shown under (313a) is typical for Czech. 276 The distribution of long and short vowels is constant in the paradigm mentioned, i.e. feminine a-stems and neuter o-stems to which yer-initial suffixes such as the diminutive -ek and the adjectival -ny are attached. The presence of a yer is guaranteed by the vowel-zero alternation observed in zab-ek-a - zab-ek "frog dim. NOMsg, GENpl". We also know from socalled short adjectives that occur in predicative function where the casemarker -y is absent that the adjectival -ny in fact is yer-initial: nemoc-en-y nemoc-en "ill; long, short form". Also recall that all consonant-final words end in a yer. Hence, abstracting away from vowel length, the underlying representations of zába, zabek, zab, zabka, jméno, jmen-ny, jmen are /zab-a, zab-ïk-ί, zab-í, zab-ïk-a, jmen-o, jmen-ïn-y, jmen-i/, respectively. If the distribution of long and short vowels that occur in the root were to be characterised with no reference to yers, the correct statement would be "long vowels occur in open syllables if the following vowel does not alternate with zero, whereas short vowels occur in closed syllables and in open syllables if the following vowel alternates with zero". In other words, the distribution of vowel length and the pattern of vowel-zero alternations are controlled by the same contextual conditions (301): long vowels appear in the same environment as zeros, short vowels occur in the same context as vowels. Could this identical behaviour be accidental? We have good reason to believe that it is not. Therefore, the usual Closed Syllable Shortening analysis that phonologists are tempted to apply in these circumstances will not do for the Czech alternation shown. This follows from the same reason that made the closed syllable analysis inapplicable to Slavic vowel-zero alternations: saying that short vowels occur in closed syllables is contrary to fact because they appear in open syllables if the following vowel alternates with zero. 276

Vowel length is a century-old problem in Czech, see e.g. Òerny (1897-1900), Trávníóek (1921,1948-49 1:20Iss). There is no general distributional pattern for the entire language, regularities can only be found in certain morphologically defined paradigms. It is shown in Scheer (200lb,c,2003b, in press b) that the lion part of the length alternations is templatic in the Semitic sense. The alternations discussed here must be kept apart from those that are produced by another category of diminutives, which yield opposite results (vlak - vlác-ek "train, dim"). More illustration and discussion of both paradigms is provided in appendix 2 (§621).

522

II, 10 Arg. 5 The life of "yers " outside of Slavic

By contrast, the analysis is simple and non-disjunctive if we call on yers: short vowels occur before yers, while their long versions appear before other vowels. In spite of this striking identity of the triggering environment, we cannot apply Lower to these data, be it in its linear or autosegmental flavour. Indeed, the triggering conditions are identical, but the effects are very different: vocalisation in one case, shortening in the other. Let us leave this issue open for the time being. Before we can pursue the discussion, the other alternations that are mentioned under (313) need to be examined.

431

2.3. Czech and Polish [o] - [u(u)], Polish q-ç

432

2.3.1. The synchronic situation of [o] - [u(u)] In both Czech and Polish, [o] alternates with [u(u)]. This phenomenon is sometimes referred to as "raising". In Czech, the alternation in quality is accompanied by length: [o] is short, while [uu] is long; the former is spelt , the latter . In Polish where length is not contrastive, [o] (spelt ) alternates with [u] (spelt ). Polish spelling makes a difference between [u] that alternates with [o] and [u] that does not: the former is written , while the latter spells . Relevant data are shown under (313b) and (313c). Both Czech and Polish alternations are governed by the yer context (301) which does not need to be commented anymore: [u(u)] occurs in closed syllables and in open syllables if the following vowel alternates with zero. On the other hand, [o] appears in open syllables if the following vowel does not alternate with zero.277 This statement can be recast in non-

277

The majority of Czech roots that show the ü-o alternation does in fact react on yers: dum - domek "house, id. diminutive", stul - stolek "table, id. diminutive" etc. The cases where yers are treated as regular vowels seem to be those where the form that contains the vocalised version of the yer is the Nominative singular, i.e. domek etc. By contrast, feminine nouns that show the vocalised yer in GENpl forms bear the unaltered : nûz-ek-y - nùz-ek "scissors NOMpl, GENpl", hxd-ok-a - hùl-ek "little stick NOMsg, GENpl", pul-ok-a - pul-ek "half NOMsg, GENpl". There is no hope to tell both sets from each other on phonological grounds. Whatever the solution, if any, it does not bear on the demonstration. Note 279 provides the exhaustive record of alternating stems.

Yers all over the place in Slavic ? 523 disjunctive fashion as before if yers are assumed. Their motivation can be read off the data in the way that is familiar by now.278 We do certainly not deal with two separate phenomena here. Rather, there is one single event that has a Polish and a Czech manifestation (and is also present in Slovak and Sorbían). Spelling in both languages hints to the common origin: the [u(u)] is fake, its real identity is [o], The Polish symbol betrays this source overtly, and so does the Czech : the little circle on the "u" (krouzek) is a reminder that the [uu] in question comes from an [o] (long [uu] that has never been [oo] is written ). Another important factor is that this alternation occurs only if the following consonant is voiced. In Czech, this is true for the entire set of alternating roots. 279 In Polish, many more alternating roots are found. The majority shows a voiced root-final consonant. However, some alternating roots possess a voiceless consonant following the , e.g. powrót - powrotu "return NOMsg, GENsg".

433

2.3.2. The only solution is diachronic: the alternation in fact concerns vowel length In no event can the synchronic situation be called regular, neither in Czech nor in Polish. When looked at from the diachronic perspective, the initial process turns out to involve vowel quantity, rather than vowel quality. Short [o] is lengthened iff it occurs 1 ) before a voiced consonant 2) that is wordfinal: [d] > [oo] / _C+voice#·280 This statement is true for Old Polish (where 278

279

280

The Czech alternation is discussed in, among others, Lamprecht et ai. (1986:113), TrávníCek (1935:82ss,268ss), Komárek (1962:169s). On the Polish side, see Gussmann (1980a:53s,113ss) who is explicit on the triggering yer context (301), Szpyra (1989:160ss,1992a:288ss). Grzegorczykowa et al. (1999:114ss) provide a descriptive survey. The exhaustive list of stems that present an ü-o alternation features 20 items, that is bûh - boha, hnùj - hnoje, lùj - /o/e, mùj - moje, tvùj - tvoje, svùj - svoje, stùj! - stojím, stùl - stole, sùl - soli, dui - dolu, hùl - hole, kùl - kolu, pùl polovina, vìil - vola, dum - domu, kùh - konë, dvùr - dvoru, -ùv - -ova, vùz vozu, nuz - noze "God, liquid manure, tallow, my, your, his, stand, table, salt, mine, stick, stake, half, ox, house, horse, yard, possessive adjective marker, carriage, knife" (for each word, NOMsg is contrasted with GENsg). The activity of the second condition may be seen in adjectives: cz kiiñ - kon-ë koñ-sky "horse NOMsg, GENsg, adj". In the latter form, the consonant following the alternating vowel stands before another consonant and is thus not word-

524

II, 10 Arg. 5 The life of "yers" outside of Slavic

vowel length was contrastive) and Old Czech.281 Later on, long [oo] was raised to ü/ ó = [uu] in both languages, and Polish lost distinctive length. In any event, this diachronic scenario does not explain why lengthening has not taken place before regular vowels (NOMsg krow-a), but did occur before yers (GENpl krów-b). Another diachronic interpretation of the facts at hand does not either. As was mentioned earlier (§417, note 261), Common Slavic lacked closed syllables altogether. Therefore, all consonant-final words in modern Slavic languages have been born through the loss of a word-final yer. Hence, "before a word-final consonant" actually means "before a word-final Common Slavic yer that has been lost". In the example mentioned, the yer at hand was the GENpl case marker -i>. Building on this fact, Vondrák (1924:324s), Trávnícek (1935:268ss) and Rospond (1979:65ss) for example try to interpret o > oo as a case of Compensatory Lengthening. This analysis is also argued for by Kavitskaya (2002:117ss). That is, the vocalic unit (mora) that drowns when the final yer falls out is transferred to the preceding root vowel, which therefore lengthens. While this analysis may work for GENpl krów < krow-τ, and dim. NOMsg krów-ka < krow-tk-a, it fails to explain the lengthening of dim. GENpl krów-ek < krow-ik-t. Here the CS o is followed by a yer, but this yer does not fall out. On the contrary, it is vocalised and hence preserves its mora. Nevertheless, lengthening occurs.

281

final. Therefore, lengthening does not occur, and the modern language shows [a], not [uu]. This pattern also disqualifies putative analyses that are based on syllable structure: [a] occurs in closed syllables if they are non-final, while [uu] is observed in final closed syllables. Pedersen (1905:305) for example discusses the diachronic identity of alternations involving [O] (nasal or not) in both languages: [O] > [OD] / C+voice#. The relation between the lengthening of CS o and the voicing of the following consonant was first discovered by Vondrák (1924:324s, first edition: 1906). Trávniôek (1935:82ss,268ss) describes the Czech situation, whose interpretation according to the rule mentioned was still not undisputed by the time he wrote. The activity of the rule on the Polish side is described for example by Stieber (1958:17s, 1973:55ss), Klemensiewicz et al. (1964:52s), Rospond (1979:65ss), Dtugosz-Kurczabowa & Dubisz (1993:79ss). The existence of many nonalternating stems that do meet the contextual conditions in both languages and for both alternations [o] - [u(u)] and ç-^ also shows that the phonological process at hand is not synchronically active anymore, but occurred in a former (common) stage of Polish and Czech (grammars are reluctant to advance a dating).

Yers all over the place in Slavic ? 525 Hence, the action of the yer context on o > oo remains as mysterious as before. All cases of lengthening are not a compensation of the loss of a yer. Yers are responsible for o > oo, but not in the compensatory sense.

434

2.3.3. Polish q-ç is but the nasal version of o > oo Finally, the alternation shown under (313d) that concerns the two nasal vowels of Polish (sometimes called "Nasal Backing") belongs to the same family: it may be traced back to a diachronic reality (that is Old Polish) where short nasal was lengthened iff it occurred 1) before a voiced consonant 2) that was word-final. In actual fact, Polish is but the nasal version of the aforementioned lengthening process [o] > [oo] / C+V0,ce#. All back mid vowels were concerned, nasal and non-nasal alike (note that transcribes a nasal [o]). Later on, an independent diachronic event turned short nasal o into nasal e, i.e. . Finally, contrastive length was abandoned. 2 8 2 The same remarks apply as before: this diachronic scenario does not explain why yers do not behave like regular vowels: lengthening was regu-

282

See for example Stieber (1973), Carlton (1991:128s) for a summary of the diachronic situation of Polish nasal vowels. Gussmann (1980a:54,84ss), Rubach (1984:130ss,229s), Szpyra (1989:163ss, 1995:104) and Bethin (1992) provide exhaustive synchronic data. The alternation is highly irregular and does by no means cover the entire lexicon. This may be best seen when considering the from rzqd, whose GENsg is rzqdu when it means "row", but rzqdu when it means "government". It is curious to observe that most authors hint at the close relationship between the alternations involving ç-^ and o-ó because of the identical triggering context (301), but do not compare them to vowel-zero alternations even though these are governed by the same contextual conditions. Since Rubach (1986) and Kenstowicz & Rubach (1987), the fact that yers "are invisible" for syllablebased processes (such as comparative and imperative allomorphy as well as Iotation) has been a standard argument for their representation as a floating matrix (lacking any x-slot, rather than as an empty Nucleus). Szpyra (1992a:288ss, 1995:105ss) argues that both alternations concerning ç-^ and o-ó fall into this category, but rejects a causal relation between their distribution and the existence of a yer in the following syllable. However, she convincingly moderates the orthodox statement: yers fail to be syllabified, but they are not entirely invisible for syllable structure since they block the tautosyllabicity of flanking consonants that qualify for branching Onset status.

526

II, 10 Arg. 5 The life of "yers " outside of Slavic

larly inhibited before ordinary vowels in GENsg zçb-a (< zqb-a), but did occur before yers in dim. NOMsg z^b-ek (< z^qb-ek < z^b-ik-t).

435

2.4. Summary: the ambition of Lower is too narrow The table under (314) shows a summary of the four phenomena examined. It supposes that Polish/ Czech [o] - [u(u)] and Polish ^-ç are traced back correctly to their historical identity: we face alternations in vowel length that have been coloured by secondary processes such as raising ([oo] > [uu]), and the loss of contrastive length in Polish. Please note that this diachronic interpretation makes no statement concerning the cognitive reality of the processes in the synchronic grammar of present-day speakers. Given their great irregularity, it may be safely assumed that they are "dead" in the modern languages. There were times, however, when they were alive. And of course, the fact that these alternations were active some time ago does not make them less real. (314) summary of the alternations discussed object occurring in example alternation _ C V _Cyer _CV _Cyer dom-0k-u dom-ek, dom-eò-ek, a. vowel-zero zero vowel dom-eé-0k-u ¿áb-a b. Czech vo- VV V ¿ab, áak-ek, ¿ab-0k-a wel length c. Polish o-ó 0 00 krow-a krów, krów-ek, krów-ek-a d. Czech o-û 0 00 no2-e ηΰέ, nü2-0k-y, nüi-ek e. Polish 3-ç zçb-a zqb, zqb-ek, zqb-ek-a Looking at this table immediately raises a question: how can long vowels be provoked by the absence of a yer in the following syllable in one case (Czech vowel length), but by its presence in the others (Polish/ Czech [o]-[u(u)], Polish 3-ç)? This important issue will be discussed at length in §§472s. For the time being, we can close this section on the following observation, which is its actual goal: yers condition a number of segmental variations in the preceding syllable, among which vowel-zero alternations and vowel length. However, Lower in its initial or autosegmental formulation aims only at describing the influence of yers on preceding vowel-zero al-

Yers all over the place in Slavic ? 527 ternations, i.e. a relation that involves two yers. The fact that vowel length obeys exactly the same pattern (301) as vowel-zero alternations suggests that this ambition is too narrow: a more general formulation of what yers can do to vowels in the preceding syllable is called for. Also, the theory-internal and diachronic motivation for the fundamental difference between final and alternating yers is now enriched by an overt contrast. That is, all alternations discussed are controlled by a lateral relation among vowels. We now know that this intervocalic relation does not necessarily target a y er. However, it necessarily originates in a yer. Final and alternating yers must thus be different. In the following section, I show that the familiar yer context extends beyond Slavic.

436

3. French: yers all over the world ? In this section, two more occurrences of the yer context (301) are reviewed: in French, the ATR-value of mid vowels and alternations involving schwa [a] and [ε] are controlled by this environment.

437

3.1. ATRness of mid vowels in Southern varieties In French, the only vowel that alternates with zero is schwa [a].283 Unlike in Slavic, the alternation is optional, not obligatory, thus la semaine "the week" may be pronounced as either [la samen] or [la semen], depending on dialectal, sociological and idiolectal parameters.284 French possesses six mid vowels that subdivide into two sets: +ATR [e,o,0] and -ATR [ε,o,ce],285 ATRness is distributed according to the famil283

284

285

The phonetic value of schwa is greatly variable, ranging from central rounded or unrounded to front rounded (see note 42). This fact has no bearing on the discussion; the only symbol used below is [a], Dell (1973:219ss) and many others expose the general facts about schwa-zero alternations in French. In Northern French (Paris), realisations in open syllables are floating und inconsistent (both [fete] and [fete] may be heard for fêter "to party" by the same speaker in free variation). In the South on the other hand (sometimes called Midi French; any label of course is but a rough generalisation of complicated geographic and sociological variation), ATRness is constant und subject to no

528

11,10 Arg. 5 The life of "yers " outside of Slavic

iar yer context: +ATR versions occur in open syllables if the following vowel does not alternate with zero, while mid vowels are -ATR in closed syllables and in open syllables if the following vowel alternates with zero. Since Dell (1973), this pattern is referred to as "Closed Syllable Adjustment" or "Loi de Position" in the relevant literature. 2 8 6 Table (315) illustrates this description. 2 8 7 (315) French ATR alternations of mid vowels closed syllable open syllable _CV _c# C.CV _Cs e

seismi bet9Kav

fete metxik

ssKensmä

seirenite

poXte normal

moksKi

kode

mètre, perdu, betterave, métrique sereine, sereinement, sérénité le code, porter, moquerie, coder

rozare

rozje

rose, normal, roseraie, rosier

sabKsmä

sobKÍjete sobre, sobrement, sobriété

0KOEZ

œ/te

0Koezamä

ap0Ke

œvK

sœxfe

bœvsKi

0VKe

vœtairi

30nes

fet

alexte

metx

peirdy

s3Ken o

kod roz sabir

0

3œn

438

spelling je fête, alerter, céleri, fêter

heureuse, heurter, heureusement, apeuré œuvre, surfer, beuverie, œuvrer jeune, veulerie, jeunesse

3.2. French couldn't have yers, but it can have abstract vowels The yer context is not j u s t an ordinary context in the same way as, say, the C o d a context {#,C}, which we expect to meet every n o w and then in

286

287

kind of free variation (only [fete] is possible for fêter). In closed syllables such as [fet] je fête "I party", mid vowels are always -ATR in both Northern and Southern varieties. The following discussion addresses only the Southern pronunciation. See for example Dell (1973:209ss), Selkirk (1972:367ss), Schane (1968:30ss), Valdman (1972), Morin (1986,1988), Tranel (1987,1988). Glosses (left to right, top down): e-ε "I party, to alert, celery, to party, meter, lost, turnip, metrical, serene, serenely, serenity"; o-o "code, to carry, mockery, to code, rose, normal, rosary, rose tree, sober, soberly, sobriety"; 0-ce "happy fem, to punch, happily, frightened, work (opus), to surf, boozing, to work, young, weakness, youth".

French: yers all over the world ?

529

various languages. The yer context is highly specific. Is it reasonable, then, to suppose that the distribution of French ATRness, Slavic vowel-zero alternations and Czech alternations in vowel length is exactly identical by accident? If not, we should recast the statement "-ATR vowels occur in closed syllables and in open syllables if the following vowel is schwa" as "mid vowels are -ATR iff a yer occurs in the following syllable". Hence, should we conclude that there are yers in French? The answer to that question probably depends on the definition of the word "yer": if a "yer" is a high centralised lax vowel with either a palatal or a velar flavour that underlies vowel-zero alternations and corresponds to Indo-European short [i] and short [u], respectively, there is no doubt that we conclude on the absence of yers from French. No such objects are known in either French diachronics or synchronic analysis. If on the other hand yers are abstract vowels, as is suggested in the classical Slavic literature (Gussmann 1980a, Rubach 1984:28) and at the end of §422, the answer may be positive. If a Slavic word that is consonant-final on the surface can end in an abstract vowel underlyingly, there is no reason why a French word could not be analysed in the same way, and for identical reasons.288 Also, if an abstract vowel may occur after Codas in Slavic, so it can in French. Table (316) below shows what kind of underlying structure this reasoning produces for French. Abstract vowels are transcribed as the traditional Slavic yer "b" for reasons of exposition (glosses are as before). (316) abstract vowels in French closed syllable open syllable _C# _CV _C.CV Ca Cb -ATR

CbCV Cb -ATR -ATR

e

ssKenb peKbdy

O

KOZb

0

0Kcezb

CV +ATR

ssKenbmä

seKenite

noKtmal

KOZbKE

BOzje

œ/bte

0Kcezbmä

apene

spelling sereine, perdu, sereinement, sérénité rose, normal, roseraie, rosier heureuse, heurter, heureusement, apeurer

On the assumption of (316), then, +ATR mid vowels become -ATR iff they occur before an abstract vowel. 288

Interestingly, Dell (1995:18ss) makes a parallel suggestion for the sake of entirely independent evidence.

530

II, 10 Arg. 5 The life of "yers " outside of Slavic

However, there is also a major difference between Slavic vowel-zero alternations and French ATR: in the former case, yers after Codas always enjoy additional motivation because they appear on the surface in the appropriate context (pol /but-ïcz-îk-a/ bul-ecz-ek-a "bread roll NOMsg" vs. /bul-ïcz-ïk-!/ bul-ecz-ek "id., GENpl"). This is not the case in French: the inferred abstract vowel that separates [r] and [d] in [peirdy] perdu "lost" never appears on the surface. This is an issue of utmost importance for it enforces the existence of additional empty Nuclei which are not motivated by vowel-zero alternations. It will be further discussed in §462. The price to pay if the perspective of abstract vowels in French were to be rejected is high: 1) the analysis of French ATRness would have to live with an unwarranted disjunction; 2) a generalisation regarding this disjunction and its identical occurrence in Slavic would be dismissed; 3) the same effects would be ascribed to different causes.

439

3.3. Schwa - [ε] alternations Before evaluating the consequences of the presence of "yers" in French, the other piece of evidence in support of the reality of French "yers" needs to be mentioned. It concerns the well-known French alternation between schwa and [ε] that is commonly referred to as "Schwa-Adjustment (a-AJ)".289 This alternation and the aforementioned variation in ATRness are classically assumed to be two effects of a single phonological mechanism called Closed Syllable Adjustment.290 The distribution of schwa and [ε] obeys the familiar pattern: schwa appears in open syllables if the following vowel is not a schwa itself, while [ε] is observed in closed syllables and in open syllables if the following vowel is a schwa. Table (317) offers some illustration.291

289

290

291

See for example Dell (1973:198ss), Schane (1968:30ss), Tranel (1987,1988), Charette (1991:172ss). Tranel (1987,1988) and Morin (1988) provide a detailed comparison of both alternations and a survey of the relevant literature. Glosses (left to right, top down): "to cut into pieces" lsg, id. noun, id. lpl, id. infinitive; "to call" lsg, id. future 3sg, id. infinitive; "to bewitch" lsg, id. noun, id. infinitive; "to harass" lsg, id. noun, id. infinitive, "to achieve" lsg, id. noun, id. infinitive; "to wean" lsg, id. future 3sg, id. infinitive, id. noun.

French: yers all over the world ? 531 (317) French schwa - [ε] alternation closed syllable open syllable eCa aCV eC# ηιοχβεΐ mD/sebmä moxsalS, mo/sale apel

apebKa

apsle

äso/sel

äso/setamä

äso/sale

axsel

axsebmä

axssle

ajev

ajevamä

ajave

sevK

sevKSKa

ssvKe sevKa3

spelling je morcèle, morcèlement, nous morcelons, morceler j'appelle, appellera, appellation j'ensorcèle, ensorcèlement, ensorceler je harcèle, harcèlement, harceler j'achève, achèvement, achever je sèvre, sèvrera, sevrer, sevrage

Internal closed syllables ( C.CV) are not represented because French morphology does not allow to produce relevant C.CV clusters where both consonants are heteromorphemic. On the other hand, schwa does not exist in closed syllables at all in French. Hence, it will never appear in roots that end in two consonants of falling sonority V RT such as alert-er, perd-re "to alert, to lose" etc. Everything that has been said in the previous section concerning the introduction of abstract vowels into the phonology of French also applies to the schwa - [ε] alternation under (317). In sum, thus, the two French alternations shown require yers (abstract vowels) in order to be understood as a non-disjunctive phonological reality.

440

4. The big puzzle: nature produces two antagonistic patterns Let us now gather the pieces of the puzzle that have been collected in the previous sections. The environment (301) that has been referred to as the yer context conditions the following alternations ("schwa" is shorthand for "vowel that alternates with zero").

532

(318)

11,10 Arg. 5 The life of "yers" outside of Slavic

the big puzzle: alternations controlled by the yer context object occurring in example _CV Cyer alternation _CV Cyer strong alternant before yers dom-ek-u dom-ek, dom-eô-ek, vowel Slavic vowel- zero dom-eö-0k-u zero [apa le] [apel] appelle French schwa [ε] schwa - [ε] appeler krow-a krów, krów-ek, Polish ο-ό VV V krów-ak-a VV ηοέ-e nui, nu2-0k-y, Czech o-û V ηύέ-ek zçb-a zrçb, zçb-ek, Polish 3-ç V VV zqb-ek-a weak alternant before yers ¿áb-a V ¿ab, ìak-ek, Czech vowel VV 2ab-0k-a length -ATR [fete] fêter [fet] fête French ATR +ATR

The individual processes are explicitly grouped: Slavic vowel-zero alternations go along with French schwa - [ε], Polish/ Czech V - VV and Polish under (318a), while Czech alternations in vowel length are paired with French ATRness under (318b). This classification is made on purpose. It reflects the relative strength and weakness of the particular alternating items: in the former group, the objects that are relatively weaker with respect to their alternating partners (i.e. zero, schwa and short vowels) appear before real vowels, while the strong alternants (i.e. vowels, [ε] and long vowels) occur before yers. By contrast, strong and weak alternants have the reverse distribution in the second group: the strong long and +ATR vowels occur before real vowels, whereas their weak alternants (i.e. short and -ATR vowels) are found before yers. This situation is overtly contradictory: why should an identical context sometimes create the conditions for the appearance of strong allophones, but at other times promote weak objects? Further discussion of this issue must be postponed until §472. The following section discusses three analyses (Anderson 1982, Hall 1992 and Spencer 1986) that arrive at the same conclusion as chapter 11,9 (§411): yers are empty Nuclei.

Three implementations of the insight that yers are empty Nuclei

441

533

5. Three implementations of the insight that yers are empty Nuclei A unified solution for the challenge that is raised by the yer context was proposed by Anderson (1982) and Spencer (1986): yers are empty Nuclei. Both articles come to this conclusion in the atmosphere that was created by the advent of autosegmental structure, and they are motivated by the two major sets of data that have been discussed on the foregoing pages: French alternations involving schwa (Anderson) and Slavic vowel-zero alternations (Spencer). A third approach to the same problem is exposed by Hall (1992:199ss). He considers the same empirical object as Spencer and Anderson, i.e. the yer context, but in a rather different incarnation which is neither Slavic nor French nor vocalic. Hall's interest for the yer context is raised by the German velar nasal. On the bottom line, his analysis also follows the insight that schwa is an empty Nucleus. On the following pages, the three analyses mentioned are discussed.

442

5.1. Empty Nuclei prior to Government Phonology I: Stephen Anderson on French

443

5.1.1. On the structural side: how to get something for nothing Let us first look at Anderson's (1982) problem. He observes that the two French alternations which have been exposed under (315) and (317) are conditioned by the same context and therefore wishes to conflate the corresponding rules into one. Considering the schwa - [ε] alterantion (317), what could be the underlying identity of the object whose surface manifestations are [ε], [a] and zero? Recall that schwa may be optionally realised as zero, including instances where it alternates with [ε] as in appelle Ι^ρεί] "to call lsg" (only possible realisation) vs. appeler "to call inf', which may be pronounced either [apale] or [aple]. However, the phonetic realisation of schwa most frequently is not [a], but [ce] (see note 283). Since both [ε] and [œ] are also surface realisations of underlying objects that do not participate in the alternation schwa - [ε] - zero (e.g. allaite [atet] "to breastfeed lsg" vs. allaiter [alete], * [alate] "id., inf'; abreuve [abrœv], *[abKev] "to give water (animal) lsg" vs. abreuver [abK0ve] "id., inf."), neither /ε/ nor /schwa/ (realised as [a] or [ce]) qualify for underlying representations (Anderson 1982:544ss,550). The third candidate, zero, does in principle.

534

11,10 Arg. 5 The life of'yers" outside of Slavic

But this option was excluded by Dell (1973:187s) for the same reason that disqualifies epenthesis-strategies in Slavic (cf. §415d): in both Slavic and French, the occurrence of vowels that alternate with zero is a lexical property of each word and morpheme. It cannot be predicted from any environmental parameter. If zero were underlying in French, the two words pelage [pala3] "coat (animal)" and plage [pla3] "beach" would possess an identical underlying structure: /plaj/. Accordingly, either both forms are expected to surface with schwa, or none is. As a matter of fact, the former may be realised as either [psla3] or [pla3], while the only possible pronunciation of the latter is [pla3]. This is where the title of Anderson's (1982) article comes from: How to get something for nothing. In linear terms, the two insights that Anderson had made could not be married: melodically (or paradigmatically) speaking, the only possible underlying identity for the object at stake is zero: "nothing"; but structurally (or laterally) speaking, the identity of the same object has to be different from zero: "something". Autosegmental theory provides a natural solution to this equation: since the structural level is separated from the definition of melody, nothing prohibits the structural presence of the vocalic object at hand, i.e. a Nucleus, but which is melodically zero. On this account, schwas are different from any other vowel since they are the simple spell-out of an empty Nucleus (whereas stable [ce] represents an underlying /œ/ that is already associated with its Nucleus in the lexicon). Their location in the linear string is anchored in the lexicon. 444

5.1.2. On the computational side I: Coda capture Once this important step away from the unwarranted syllabic philosophy "What you get is what you see" is made, the question arises how these empty Nuclei come into segmental being in case they appear on the surface. Anderson evolves in a rule ordering type of grammar that may be described as a version of SPE augmented with autosegmental representations. In this spirit, he operates with various linking- and delinking rules. The cornerstone of his analysis is so-called Coda capture. Anderson's version of this device resyllabifies the Onset of an empty Nucleus as the Coda of the preceding syllable if and only if the empty Nucleus is not itself followed by a Coda (Anderson 1982:553ss). Table (319) illustrates how Coda capture works: its application ensures that consonants which are followed by schwa (i.e. an underlying empty Nucleus) become Codas: (319a). The preceding vowels therefore

Three implementations of the insight that yers are empty Nuclei

535

stand in closed syllables just as much as those that have enjoyed this status since the lexicon: (319b). Only those intervocalic consonants remain Onsets that occur before a regular vowel and hence escape Coda capture: (319c). Table (319) represents Coda capture for schwas that alternate with [ε], where Anderson assumes the common underlying identity to be schwa, i.e. an empty Nucleus. Alternations in ATRness work in the same way, only that here the +ATR version is supposed to be underlying. (319) Anderson (1982): Coda capture before schwa a. appellera [apetala] b. appelle [apel]

c. appeler [apale]

R lexicon

O N O N O N O N

O N O N O N

II

I

II

II

I

II

II

a p

1

κ a

a p

1

a p

l e

R Coda capture

Κ O N O N C

Ν O N O N C N O N | | | | | a p 1 κ a R

Κ Closed O N O N C N O N Syllable I M I II Adjusta p e l κ a ment R Γ\ fill-in of O N O N C N O N schwa ι ι ι I M (optioa ρ e 1 * κ a

.J

R Κ O N O N C | | | a p 1

O N O N O N | | | | a p l e

R l \ O N O N C I II I a p e l

O N O N O N II II a p l e

R κ O N O N C I I I I a p e l

O N O N O N II II a ρ A 1 e

.J

This is how Anderson gets rid of the yer disjunction (301) "in closed syllables and in open syllables if the following syllable contains schwa": something that behaves like a Coda (i.e. the Onset of schwa syllables) will be made a Coda, even if it is not.

536

11,10 Arg. 5 The life of'yers" outside of Slavic

Once Coda capture has operated, subsequent rules epenthesise an [ε] into empty Nuclei that occur in closed syllables, hence under (319a) and (319b) (Closed Syllable Adjustment). For the other alternation, mid +ATR vowels are turned into their -ATR partners in closed syllables. Finally, the remaining empty Nuclei that have not been concerned by [e]-insertion optionally receive schwa292: il appellera [apel(a)Ka] "he will call" may be pronounced with or without schwa. In case schwa is left out, the entire syllable (which then would not possess either Onset material nor a Coda) is deleted.

445

5.1.3. On the computational side II: Coda capture misses the basic insight of Lower The solution favoured by Anderson is typical for the early period of generative endeavour: anything could be a rule and change any object into any other object, the only goal being to produce the attested result on the surface.293 Foley (1977:3ss) for example lays out a substantial critique of this practice, which is nowadays standard textbook argumentation (e.g. Ewen & Hulst 2001:3ss). Harris (1999) and Gussmann & Harris (2002:22) pinpoint the real identity of Coda capture: this device is an attempt to force an unpleasant reality into a theory. Some open syllables behave as if they were closed, so let us write a rule that makes them closed.

292

293

The optional character of the rule that deletes syllables with empty Nuclei, as well as the exact ordering of rules is inferred by me, Anderson (1982:555s) is evasive on this issue. It is true, however, that this has not been Anderson's exclusive motivation. He also connects his analysis with other facts from French such as the ordering of other rules and h-aspiré. Another attempt at reducing the yer context (301) was made by Selkirk (1978). She proposes that the relevant domain of application of the rule which governs the French alternations is not the syllable, but the foot (see also Basball 1981 along these lines). Tranel (1987:312s) has shown that this approach does not resist the French evidence. Moreover, it is transparent that the definition of a French foot as "containing one syllable, except if the following vowel is a schwa, in which case it contains two syllables", serves no other purpose than making the yer context (301) non-disjunctive. Schwa being forced into the foot, the yer context can be recast as "everywhere but at the end of a foot". This mechanism is overtly circular.

Three implementations of the insight that yers are empty Nuclei

537

Doing this makes sure to miss the basic insight of Lower: the presence or absence of a Coda in a given Rhyme is completely irrelevant for the alternations that are governed by the yer context (301). Rather than in the arboreal relation of a Nucleus and a Coda, the solution is to be sought in the lateral communication between two vowels. There is actually good reason to be surprised that the "French" reaction on the yer context (301) went the paradigmatic, rather than the syntagmatic way. Indeed, since French was carried into the spotlight of generative investigation by Schane (1968), all analyses used a kind of final yer which occurs after word-final consonants. These abstract vowels were called "protective schwas". They are almost identical to word-final yers in function and behaviour: 1) they never reach the surface, 2) they exist in order to prevent (Slavic: in order to trigger) the application of rules, 3) they are assigned morphological value. For instance, first, second and third person singular of verbs that belong to the first group (infinitives in -er like porter "to carry") were said to be marked by /-a/. In the same way, the feminine marker of adjectives was supposed to be /-a/, and the thematic vowel of first group verbs was /-a-/ as well. On the account of alternations between consonants and zero, consonant deletion rules were proposed. For instance, adjectives typically lose their final consonant in masculine forms, but maintain it in feminine occurrences: fort [fox] "strong masc." vs. forte [ίοχί] "strong fem." and the like are characteristic examples. Both were said to derive from the common and unique underlying root /fort/. The masculine morpheme was zero, while the identity of the feminine suffix was schwa /-a/. If the rule of consonant deletion applies before the one that evacuates final schwas, the correct surface forms are derived. The same kind of argumentation was made for the other morphemes mentioned that involve protective schwas. Their discussion exceeds the scope of the present purpose. Relevant information is available in, among many others, Dell (1973:177ss), Tranel (1981:277ss). Hence, half of the technology that is needed in order to run a lateral implementation of the yer context (301) in the spirit of Lower was already established in the analysis of French: word-final "protective yers" and the rules deleting them were needed anyway (though not in all instances of words that end in a consonant on the surface, that is true). Perhaps was the ongoing debate on abstractness (e.g. Kiparsky 1968,1982b, Kenstowicz & Kisseberth 1977:1-62, Hooper 1976) with firm supporters of "concreteness" (i.e. anti-abstractness) on the French side (Tranel 1981) a brake on general-

538

II, 10 Arg. 5 The life of "yers " outside of Slavic

ising "abstract schwas" to all triggering contexts, including after Codas. Or perhaps did the label "protective" intuitively preclude from considering this object as a trigger of phonological processes. Be that as it may, let us now return to the discussion of Anderson's Coda capture. Decreeing that the nasty open syllables are closed also misses another important insight that Lower expresses: it is not just any arbitrary kind of open syllable that behaves as if it were closed. Only those syllables are concerned that are followed by a very peculiar vowel, i.e. schwa. Why is this vowel so special? Of course, Anderson and the other authors that have identified the relevance of the yer context in French were aware of the list of special properties of French schwa, among which the fact that it alternates with zero. But they did not establish any causal relation between this property and the fact that only this vowel has a special influence on the preceding syllable. If any other vowel, say non-alternating [u], had displayed this behaviour, the Coda capture rule would simply have stated that Onsets of syllables whose Nucleus is /u/ are resyllabified as the Coda of the preceding syllable if their Rhyme does not branch. In yet another scenario, the Coda capture rule would have sounded exactly as proposed by Anderson if French schwa did not alternate with zero. This is in fact another typical property of early generative analysis, consubstantial with the one that was mentioned earlier: rules transform X into Y in the vicinity of a context A. We are not interested in the causal relation involved: why does A turn X into Y rather than into Z? Why is X turned into Y by A rather than by B? These questions had no answers, and they would not even be asked (Foley 1977:4, see §333).294 By contrast, the very essence of Lower precisely makes this causality explicit: a yer is vocalised iff followed by another yer. There would be no way to replace the triggering yer by some other vowel since only the fact that yers alternate with zero allows to posit underlying triggering yers that never reach the surface.295 294

However, it must be said in defence of the "French" branch that the identification of the crucial property of schwa in regard of its relation to the preceding syllable, i.e. its alternating status, was not self-evident since the Slavic data and analysis were obviously set in no relation with the French facts. This is probably due to the fact that there were no slavicists on the "French" side, and the literature dealing with Slavic data was not reviewed (at least is it completely absent from the texts that discuss the French alternations).

295

Gussmann explicitly identifies the crucial property of triggering vowels: "the deleting vowel disappears if followed directly by a nondeleting vowel" (Gussmann 1980a:30).

Three implementations of the insight that yers are empty Nuclei

539

It is only when both the Slavic and the French pieces of the puzzle (318) are identified that the crucial question can be raised: what is the common feature of yers (which appear as [ε], [o], [a] etc. in various particular Slavic languages) and French schwa? The answer is straightforward: the fact that both alternate with zero. Hence, any analysis that does not encode this causality between the alternating character of the triggering vowel and its bearing on the preceding syllable is on the wrong track. Anderson was wrong, Lightner and his followers were right. On the other hand, the following question may be asked: what made the Slavic side refrain from acknowledging empty Nuclei in the French fashion, and for the same reasons? In order to understand the motivation for autosegmentalising Lower in the way that was shown in §423, i.e. without using empty Nuclei, we need to discuss the missing link in the study of Slavic vowel-zero alternations: the article that Spencer (1986) has written on Polish.

446

5.2. Tracy Hall on German: consonants preceding schwa end up as Codas, but for a good reason

447

5.2.1. On the structural side: schwa is a floating x-slot As was mentioned in the introduction of §441, Hall (1992:199ss) comes across the yer context in a language that is neither Romance nor Slavic, and the particular incarnation he considers bears on a consonant, rather than on a vowel: the German velar nasal. 296 For expository reasons, relevant data and literature are introduced only in §480. The cocktail version of the facts, however, is as follows: in German, [η] and [qg] are in complementary distribution. Underlying /ng/ is reduced to [η] word-finally (e.g. lang [lag] "long"), in preconsonantal position (e.g. Angst [?aqst] "fear") and before schwa (e.g. Inge [?iqa] "female first name"). That is, [η] surfaces in the yer

296

Interestingly, Hall (1992) and Schenker (1995:97) (on which more in note 306) are the only references that I have come across where the facts from either language (family) where the yer context is active, i.e. French, Slavic, German, are related to another set of facts in another of these languages. Hall (1992:219, note 9) mentions the French alternations in ATRness. Finding the coincidence strange and anyway too precise in order to be accidental, he calls for further study of the pattern.

540

II, 10 Arg. 5 The life of "yers " outside of Slavic

context. Before a full vowel, however, [rjg] appears (e.g. Ingo [?iggoo] "male first name corresponding to Inge"). The solution that Hall (1992) proposes follows the same philosophy as Anderson's (1982) analysis: schwa behaves as if it were not there, hence it must be somehow made absent in the relevant representations. As a consequence, then, the preceding consonant will not be interpreted as an Onset, but as a Coda. In Hall's system, the notorious weakness of schwa is expressed by its lexical representation: schwa is an orphan skeletal slot without melodic content that acquires phonetic expression through a late epenthesis. Hall (1992) evolves in a classical environment where strings are unsyllabified in the lexicon and acquire syllable structure through the action of a syllabification algorithm. However, x-slots are recorded in the lexicon in his system. On a first pass, then, the syllabification algorithm deliberately ignores x-slots that are melodically empty. Assimilation and g-deletion apply at the resulting derivational stage. Finally, schwa is inserted by default into empty slots, which are then syllabified on a second pass of the algorithm. Table (320) below shows the derivation of the word Inge according to this analysis. (320) Hall (1992:210s): schwa is an unsyllabified skeletal slot derivation of Inge "female first name" a. lexical representation b. first pass c. assimilation σ σ

χ

χ

χ

χ

?

ι

Ν

g

χ

Ο I χ

Ν I χ

C χ

χ

?

ι

Ν

g

χ

Ο I χ ?

Ν I χ

C χ

Γ \ χ

q

g

χ

Three implementations of the insight thatyers are empty Nuclei

541

(320) Hall (1992:210s): schwa is an unsyllabified skeletal slot derivation of Inge "female first name" d. g-deletion: e. schwa insertion f. second pass and resyllabification g -» 0 / [+nasal] _ ]„ of η σ σ σ σ M

Ο

Ν

C

χ

χ

χ

?

I

r,

χ

Ο

Ν

C

χ

χ

χ

?

ι

i)

Ο

Ν

Ο

Ν

χ

χ

χ

χ

χ

a

?

ι

Γ)

3

According to Hall (1992), the fact that the x-slot which ends up hosting schwa is unsyllabified has a consequence: the preceding consonant cannot be interpreted as its Onset. The only alternative is a syllabification into the Coda of the preceding syllable. Therefore, the /g/ under (320b-e) belongs to a Coda, rather than to an Onset. It is then obvious why objects before schwa (/ng/ here, vowels in French and Slavic) show the same behaviour as in true closed syllables: they occur in a Coda (German /Ng/) or in a closed syllable (French and Slavic vowels) as well. Therefore, a rule may simply target all objects in Codas (or vowels in closed syllables) once it has been made sure that the consonant preceding schwa belongs to a Coda. In the German case, the environment of g-deletion can thus be formulated as follows: /g/ is lost "after a nasal in syllable-final position". This also holds true for word-internal and word-final Codas: Angst "fear" will be syllabified as /?aNg.st/ 297 , while Ding "thing" identifies as /diNg/. Let us now compare this analysis with Anderson's (1982) approach.

448

5.2.2. On the computational side: the consonant preceding schwa has a good reason to end up as a Coda Hall's (1992) scenario is functionally and representationally equivalent to Anderson's (1982) solution: the critical move is to achieve codahood of the 297

The final /t/ remains extrasyllabic, while /s/ joins the Coda of [η] after g-deletion, see Hall (1992:205).

542

II, 10 Arg. 5 The life of "yers" outside of Slavic

consonant that precedes schwa. In other words, an intervocalic consonant (or RT cluster) will be a regular Onset (a regular Coda-Onset sequence) if the following vowel is a non-schwa, i.e. V.CV (VR.TV), while it is syllabified as a Coda in case the following vowel is a schwa, i.e. VC.a (CRTs). Also, the interpretation of schwa itself as an empty Nucleus is common to both analyses. This is overt in Anderson (1982), while schwa is not explicitly rated as an empty Nucleus in Hall's (1992) scenario. However, the syllabic identity of schwa has all properties of an empty Nucleus except, precisely, the fact of being syllabified at the critical derivational stage. That is, the phonetic object schwa comes into being by filling in an empty x-slot by default, as shown under (320e). There is, however, an important difference between Anderson's and Hall's analysis. It was shown in §445 that Anderson's Coda capture is arbitrary: there is no reason for decreeing that the relevant consonant becomes a Coda other than the wish to achieve this very result, i.e. to be able to write a rule which uniformly targets Codas (or vowels in closed syllables). There is no causal relation between the syllabic or representational identity of schwa and the fact that consonants preceding this particular vowel are captured, rather than, say, consonants before [a], [i] or mid vowels. Hall (1992) is explicit on this argument: "These rules [i.e. rules making explicit mention of the phonetic object schwa] cannot explain why the only vowel in the German language that triggers g-Deletion is schwa, the only unspecified vowel in the language. Were one to posit a g-Deletion rule with a schwa environment the question of why schwa and not some other vowel such as /oe/ triggers the rules would remain a mystery." Hall (1992:212)

On Anderson's (1982) Coda capture, the chief insight of Lower, intervocalic causality, is completely (dis)missed. By contrast on Hall's (1992) analysis, there is a very clear causal chain that leads form the "weakness" of schwa over its representational properties to the fact that the preceding consonant ends up in a Coda. The initial choice of representing schwa as a floating empty x-slot can be argued to express its "weakness": floating and empty objects are weaker than those that are integrated into syllable structure. The fact that the preceding consonant cannot be syllabified as the Onset of this floating empty x-slot, then, is obvious since it needs to be integrated into syllable structure in order to be pronounced. Therefore, the only alternative is its interpretation as the Coda of the preceding syllable.

Three implementations of the insight that yers are empty Nuclei 543 In sum, thus, Hall (1992) achieves the same effect as Anderson (1982) without recurring to Coda capture. He invokes a plausible causal relation between the weakness of schwa and the codahood of the preceding consonant. This codahood, in turn, induces the segmental eifect that is observed on the surface. The causality is thus ultimately intervocalic, true, but only indirectly so. Recall that this is one of the basic arguments for CVCV (see §211): why are phenomena that are overtly lateral forced into an arboreal picture? The lateral causality (/g/ is lost before a schwa, but not before a regular vowel) is first converted into an arboreal distinction (Coda capture before schwa), which is then said to be the reason of the segmental effect observed (/g/ is lost in Coda position). Why do phonologists always make a detour via arborescence when the phenomenon is lateral in nature and can directly be expressed as such? Therefore on the bottom line, Hall's (1992) solution fares better than Anderson's (1982) because it expresses the intervocalic causality. However, this causality is only indirect, and there is no reason to have a mediating arboreal artefact that serves no purpose.

449

5.3. Empty Nuclei prior to Government Phonology II: Andrew Spencer on Polish

450

5.3.1. On the structural side: a less abstract Lower Unlike Anderson (1982) and Hall (1992), Spencer (1986) did not have to struggle with the data and their analysis in terms of yers since he could build on Gussmann (1980a) and Rubach (1984). Spencer fully accepts the fundamental insights of Lower that have already been discussed: yers exist, that is Polish does not possess any Codas in alternating environments nor word-finally in underlying representations. Spencer's purpose is not to build a new analysis. Rather, he attempts at making Lower and the formal apparatus used by Gussmann and Rubach less abstract. His motivation stems from the long lasting debate on abstractness that was initiated by Kiparsky (1968) (see §307). Spencer is firmly engaged against abstract analyses in case a more concrete alternative is available. Therefore, he aims at proposing a version of Lower that does not need to appeal to underlyingly present yers which are subject to absolute neutralisation, the extensive use of extrinsic rule ordering and the cyclic application of rules. In short, Spencer's goal is to achieve the same

544

II, 10 Arg. 5 The life of "yers " outside of Slavic

descriptive adequacy as Gussmann and Rubach while cutting down their formal machinery to a minimum, and expanding the role of both the lexicon and allomorphy. He contends that autosegmental representations are the adequate tool that allow to successfully implement this program. One direct result of his approach is the equation "yer = empty Nucleus" (Spencer 1986:255). Hence, the alternating pairs of the word for "day" [Sieji] dzien NOMsg - [dji-a] donia GENsg and "dream" [sen] sen NOMsg - [sn-u] S0n-u GENsg enjoy the following lexical representation. (321) dzien NOMsg C V C V

I dj

C V C V

I I η

d0nia GENsg

C V C V

t i l l *

dj

*

n

S0n-u GENsg

sen NOMsg

a

I

I I

s

η

C V C V *

I

I

I I

s

*

η

u

The difference between both words is that the former, in addition to the vowel-zero alternation, shows a [d$]-[d] alternation, while the initial consonant of the latter is not affected. This contrast classically motivates the existence of two distinct yers, even though the phonetic value of alternating vowels is always the same: the front yer ill has a palatalising effect, while the back yer Γ\! does not provoke palatalisation. On the account of Gussmann (1980a) and Rubach (1984), the underlying identity of the alternating vowel in dzien is a front yer, while the [ε] in sen originates in a back yer. Since Spencer wishes to do away with abstract vowels that are part of the phonemic inventory of the language, he equates the vowel yer with an empty Nucleus. He also conflates both yers into one single object: there is just one variety of empty Nuclei. As shown under (321), the palatality effect is achieved by the lexical presence of a /j/ in the root of dzien vs. its absence in the root of sen (on the fate of this /j/ see below). On this account, the vocalisation of the lexically present empty Nuclei proceeds as follows. Spencer (1986:252) correctly translates the action of Lower into prose: "its effect is to lower all but the rightmost of a consecutive string of yers. The remaining yer is then deleted". His computation of empty Nuclei is a simple recast of this formulation: first, he gives a special status to the last member of a yer chain: this yer is called "extrametrical" (even though it may occur word-internally). In the representations under (321), these empty Nuclei are identified by an asterisk. Spencer further stipulates that extrametrical empty Nuclei may not receive melodic identification. All other empty Nuclei are then subject to epenthesis: [ε] is inserted. Finally, a late postcyclic rule deletes unassociated empty Nuclei. This rule

Three implementations

of the insight that yers are empty Nuclei

545

feeds another rule that deletes the 1)1 mentioned in case it occurs between two C-slots (like in GENsg dnia). If /j/ is not deleted, it causes palatalisation of the preceding consonant (like in NOMsg dzieñ). Further implications of Spencer's account regarding the organisation of the lexicon and the exact nature of allomorphy are not relevant here. In the following section, his contribution to the analysis of vowel-zero alternations is evaluated in the light of the previous discussion.

5.3.2. On the computational side: fill-in without causality vs. intervocalic relation As was mentioned earlier, Spencer is not concerned with the reduction of the nasty yer context. He takes for granted that underlying representations bear vocalic items in exactly the locations specified by Lower: word-finally (final yers) and in alternation sites (alternating yers). Only does he "desegmentalise" the objects in question, giving them a purely syllabic, nonmelodic identity. As far as I can see, a much simpler account is available under these provisions (and Spencer was seeking concreteness ~ simplicity). Recall that the effect of Lower "is to lower all but the rightmost of a consecutive string of yers. The remaining yer is then deleted". A good question to be asked, then, is why the "remaining yer", i.e. the last yer of a yer chain, should exist in the first place: it is deleted anyway and needs the very suspicious stipulation of (word-internal) "extrametricality". But most importantly and unlike yers in the classical Lower account, extrametrical empty Nuclei do not trigger any process. Spencer's mechanism would produce identical results if the last yer of a yer chain were not embodied by an empty Nucleus. This follows from the fact that there is no causal relation between the existence of an empty Nucleus (i.e. a yer) and the vocalisation of its preceding peer. Spencer's rule that inserts melodic content into empty Nuclei makes no reference to the melodic status of the following syllable at all. It simply says "insert [ε] into empty Nuclei".298

298

The absence of an intervocalic causal relation in his analysis exists even though Spencer (1986:255) explicitly mentions that one of his motivations is an insight from the analysis of Finnish: "The C slot, though empty, can condition syllablesensitive rules, but if left unassociated at the end of a derivation is deleted". See Piotrowski (1992a,b) and Piotrowski et al. (1992) for more discussion of Spencer's (1986) analysis.

546

II, 10 Arg. 5 The life of "yers " outside of Slavic

In this respect, thus, Spencer (1986), Anderson (1982) and Hall (1992) behave alike: they posit empty Nuclei, but neither incorporates the fundamental insight from Lower according to which the effects of the yer context do not originate in the paradigmatic organisation of syllable structure, but are caused by a lateral relation between two vowels.

452

5.3.3. Kenstowicz & Rubach's (1987) arguments against empty Nuclei While Rubach (1986) and Kenstowicz & Rubach (1987) welcome the idea of eliminating yers from the underlying inventory of melodies, they argue against empty Nuclei. Several points are made, among which namely palatalisation. This process is triggered for example by the adjective marker [-n-] in words such as pol glos [gwos] "voice" vs. glos-n-y [gwoç-n-i] "loud".299 We know that the adjective marker in question is yer-initial since a vowel occurs in predicative forms where the final -y is absent: pol win-ien [vip-εη] - win-on-y [vin-n-i] "guilty", cz nemoc-en - nemoc-en-y "ill". Hence, if the underlying identity of the adjectival [-n-] were an empty Nucleus followed by an [n] /-Vn-/, how should palatalisation be achieved? Since the empty Nucleus is not subject to epenthesis, it will be deleted on Spencer's account. Consequently, no palatal agent is available that could trigger palatalisation.300 On the grounds of this argument and a few others, Rubach (1986:257s) concludes that 1) the autosegmental idea is welcome, 2) the elimination of yers from the underlying inventory of melodies is called for, 3) yers must enjoy a melodic identity, and 4) representations must allow for the existence of two different yers, one palatalising, the other not. The consequences of this scenario have been exposed in §423 (autosegmentalised Lower).

453

5.3.4. Spencer's analysis cannot be extended to other alternations When looking at the other Slavic alternations that have been discussed under (313) (Czech VV-V, Polish ^-ç, Polish and Czech [o]-[u(u)]), Spencer's 299

300

But see the absence of palatalisation in -ny formations such as miçso [mjëwso] miqsny [mjewsni] "meat, adj.", zelazo Relazo] -zelazny fóelazni] "iron, adj.". But see the solution proposed by Gussmann (1992a) and Bethin (1992) in defence of Spencer that records palatalised consonants in the lexicon.

Three implementations of the insight that yers are empty Nuclei

547

analysis runs into even more serious trouble. Since there is no causal relation between the site where the segmental effect is observed and the following vowel, what could be the trigger of these alternations? The underlying identity of the target vowels in question is not an empty Nucleus for sure. The same holds true for French alternations in ATRness. In sum, the discussion of (313) concluded that not only must there be an intervocalic relation, but that theory is also called to be able to distinguish final from alternating yers. On Spencer's account, vocalisation is not triggered by any object but occurs by default. Therefore, no such difference can be made, and the associated processes where vowels alternate with something different from zero remain unexplained. Yet another fact constitutes a serious hurdle for Spencer's analysis: there are languages where more than one vowel alternates with zero. Among these are Eastern Slavic (typically [e]-zero from Common Slavic soft yers vs. [o]-zero from Common Slavic hard yers) and Slovak. This fact is incompatible with insertion strategies (see §415d).

454

5.4. Szpyra (1992a): insertion into unsyllabifiable clusters without intervocalic causality Along the same lines as Spencer (1986), Szpyra's (1992a) contribution to the debate is engaged against abstract underlying structures. She aims at doing away with yers that never appear on the surface, that is word-final yers with morphological value. This is achieved in denying the intervocalic causality of Polish vowel-zero alternations altogether. Only alternating, but not final yers, are present in underlying representations. They are not vocalised because of the presence of a yer in the following syllable. Rather, they come into phonetic being in order to salvage unsyllabifiable consonants that could not otherwise be accommodated. Szpyra's argumentation crucially hinges on the analysis of the alternations mentioned under (313c,d), that is Polish and ο-ό. For the sake of her demonstration, the yer context (301) that controls them must not be deprived of its disjunctivity (see also note 282). She thus explicitly welcomes the disjunctive statement "in closed syllables and before a yer". Disjunctions are suspect per se. But Szpyra also draws a red line in the middle of three processes that are governed by identical contextual conditions: ç-3. and ο-ό on one hand obey a disjunction, whereas vowel-zero alternations on the other are due to a single cause, that is the unsyllabifiable character of certain consonants.

548

II, 10 Arg. 5 The life of "yers" outside of Slavic

Finally, it is not clear how the French data could be accommodated in Szpyra's analysis without abandoning the relevant cross-linguistic generalisations.

455

5.5. We are looking for a theory of intervocalic relations

456

5.5.1. Synopsis of all analyses regarding the yer context We are now in a position to compare the different views that have been discussed in the preceding sections. Table (1) provides a synopsis (the term "abstract vowel" subsumes yers in the traditional generative sense, empty Nuclei in Spencer's 1986 analysis and German alternating vowels, which are phonetically schwa).301

301

Szpyra (1995:107ss) offers a more elaborate discussion of the various representations of yers that have been proposed.

Three implementations of the insight thatyers are empty Nuclei

υ O •β υ β

V

ω

o c

o β

o

1/3

α> >>

>

Λ S

o Β

o c

o c

o c

o β

o c

Ol« c

>>

o Β

ac2

« >>

c

c

(O υ>>

o

C fi

o

O Β

υ

ω >>

ω

Ο Β

Ο Β

Ο Ε

Ο Β

Ό ij h OO OO » Ν 9ι Ο

"S C g J3 u 2 u Λ co Λ Λ C Χ) 3 ω 3

JS

VI ^c VI o 2 >>

o c

»

ω

β

OÍ i¿ OÍ Η «Ν >>Ον

CO

w

ω U «

vo 00

È5

•e

/—V Tf 00 os

cd

Τ! α> s 3

S. •S o s ¡3 • 3 κ» ι-· "ÖJ υ>> ε cd

(Ν (Ν m

2 β

Λ

VI

ω

S 0>

Β O

60 •S

Τ3

Ε

CO

_çd "ω t

C C *Τ3 ϊλ |

S

1 C/3

i

i

O cd OD - b •S £ ed

co cd ° u ö u -ι (ν:— (si — (Ni— (N > ^ £ « 2 05 ™ ä O Q. υ ε oD. c «« % M o 2 " 0¿• oΙΛO-S * c2: o•Λ ce η ce (U o cO ευ ce Ε . T 3 •SÍ3o O .5 οΌ 11 o δ > * =tfc υ * . U U _J> S ·· 5 g ü-fC í -Λ bp > > e a fc C/î u 3: υ C 5/ βojO o •υo ot ocί ϊ •β £o > > ου S 3 -β S -5 o. O (Λ .2 5Λ *

2

Zi

•S0 Λ υ

U

¿>

0. α> ε § 2 t-i - Ä b «>4> λ ·- ° c Ε·§ * ϊ 3Ü ffj

•S 3

Ζ ë 00

ία ) > υ β

tiυ > υ β

£

υ > υ > < eL> υa

< >U (eU

ζ

υ 3 s* 73s C

>C Sy 6ot> + +

se εS> 8~ o60

The fourth object: schwa

663

The typology shown is arrived at according to the following criteria: 1) Closed Syllable Shortening does or does not occur before word-final consonants, 2) word-final consonants do or do not behave like internal Codas (in case they do not, they may be in intervocalic, post-Coda or worsethan-Coda position), 3) word-final RT# clusters do or do not occur, 4) Closed Syllable Shortening does or does not occur before schwa, 5) alternation sites show a vowel or zero before schwa and 6) before schwa, consonants behave like if they were in intervocalic, Coda-, post-Coda or worsethan-Coda position. The former three criteria define the properties of final empty Nuclei, while the latter three position schwa. Table (388) below shows the distribution of the few languages that have been addressed in this book according to the six criteria mentioned. It goes without saying that there are surely more empirical tests that allow to detect the lateral actorship of the two parameterised nuclear categories. And it is also clear that all processes which allow to identify the lateral properties of final empty Nuclei and schwa do not necessarily co-occur in a given language. Therefore, empty cells do not come as a surprise. (388) lateral actorship of final empty Nuclei and schwa in some languages FEN schwa conclusion C# = CSS CSS V-0 in c in Ca internal ¡ n in —Ca behaves FEN schwa Coda c # __Ca like gvt lie gvt lie + + yes no ?

RT#

Old French Mod. French Icelandic German Braz. Port. Old Cz., Old Pol Dutch mod Czech 345

yes

?

yes 3 ' 5

yes

β

?

yes yes no

? yes yes

no ? ?

?

0

a Coda an Onset

yes yes yes

yes

yes

0 V

-

+

+ +

+ +

+

+

0

?

+

?

+ +

+

+

-

+ -

-

The value of this cell follows the analysis of Rizzolo (2002) who interprets the ATR alternations of French mid vowels (§§437s,477) as a contrast in vowel length: long vowels project +ATR on the surface, while short vowels come out as -ATR. If this analysis is correct, Closed Syllable Shortening is active in Modern French.

664

II, 12 Arg. 7 Different but still the same: internal and final Codas

The prediction implied by this kind of typology is that no language will produce a self-conflicting parameter setting. Whether this is true or not, whether all logically possible combinations are found in nature and whether the setting for final empty Nuclei stands in any causal relation with the one for schwa needs to be investigated by further analysis of individual languages.

Chapter 13 556 Argument Eight The Coda Mirror346

Audience: 1) people who believe in classical constituency including Rhymes and Codas. 2) people who do not believe that all Codas (not just some) are in fact Onsets of an empty Nucleus. Conclusion regarding syllable structure: all RT sequences are separated by an empty Nucleus.

557

1. Classical syllable structure is unable to characterise the Coda Mirror as a natural class

558

1.1. The Coda Mirror: summary The Coda Mirror has been established on empirical grounds in Ségéral & Scheer (2001a), and some evidence has been reported in §115. Recall the three-way partition of syllabic environments that need to be distinguished by any theory of lenition and fortition. Table (389) below repeats the relevant diagram.

346

The content of this section is based on joint work with Philippe Ségéral (Ségéral & Scheer 1999,2001a).

666

II, 13 Arg. 8 The Coda Mirror

(389)

Positional influence on segmental health

Strong Position

Weak Position

Intervocalic

#_

Coda

_C

_#

V_V

A particular position in the string may bear on the segmental health of consonants such that they are either prone to lenition or preservation/ fortition. The latter phenomenology is found word-initially and after Codas (i.e. in the Coda Mirror), while the former occurs in Codas and in intervocalic position. Even though it is true that consonants are weak in both latter contexts, they are weak in two rather different ways. Consider the following list of phonological processes that are typically encountered in one, but not in the other position (repeated from §131). (390)

process affecting a segment because of its position in the string

Coda

V_V

devoicing deaspiration (Ch—>C)

typical typical

highly improbable highly improbable

velarisation (l,n—4,q)

typical

highly improbable

s-debuccalisation (s—>h) liquid gliding (r,l—»j)

typical typical

highly improbable highly improbable

depalatalisation (ji—*n)

typical

highly improbable

1-vocalisation (1—>w/o) r-vocalisation/ loss ([kaad] "card") [NC] hom : homorganisation of nasals

typical

highly improbable

typical

highly improbable

typical only if also in V V highly improbable highly improbable

highly improbable typical

spirantisation (b,d,g-*ß,ÖY) voicing (t—>d) rhotacism (s,z-*r)

typical typical

On top ofthat, both Codas may or may not show different behaviour. This issue was discussed at length in chapter 11,12 (§524). We will see in

Classical constituency fails to describe the Coda Mirror as a natural class 667 Vol.2,III.6.3.2 that the same variation is also observed for the Coda Mirror: consonants in post-Coda position are strong in all languages, while the status of word-initial consonants is parameterised. Languages vary as to whether initial consonants are strong or weak (Seigneur-Froli 2003, in press). Note that the Coda Mirror, then, is the exact symmetric object when compared to the Coda in this respect as well: internal Codas and Coda Mirrors are always weak and strong, respectively, while the weakness or strength of Codas and Coda Mirrors that occur at the margins of the word is subject to parametric variation.

559

1.2. Disjunctions and their consequences in phonology The argument to be made in this chapter concerns the three-way opposition Coda Mirror vs. Coda vs. intervocalic position. The reasoning that has led to the syllabic object "Coda" was the recurrent identical behaviour of wordfinal and preconsonantal consonants in a large number of genetically unrelated languages. Theory is called to be able to characterise these two environments as a natural class, which means that the following characteristics must be met. (391) challenge raised by disjunctions theory needs to characterise disjunctions as a natural class that is a. uniform (i.e. non-disjunctive) b. unique (i.e. different from any other) c. positive (i.e. not just the complement set of an object that is defined in positive terms) These requirements are self-evident when talking about the Coda context {#,C}. They are therefore not usually spelt out one by one. However, according to my experience with the Coda Mirror, the behaviour of phonologists may be a little different in regard of other disjunctions that possess exactly identical formal properties. While (391a,b) are commonly accepted, it is worthwhile to make sure that (391c) be clear. Positive in this context means that it would not have crossed anybody's mind to refer to Coda consonants as "all and only those consonants that do not occur before a vowel". This would have been a negative statement: given the total set of consonants, the uniform and unique identity of the subset that occurs in {#,C} will be characterised as the complement set of those consonants that are followed by a vowel. In other words, the Coda would enjoy only an

668

II, 13 Arg. 8 The Coda Mirror

identity by default: "you are a Coda iff you do not stand before a vowel". This would also mean that the Coda is nothing that is cognitively real: it is not an object that grammar refers to; only "after a vowel" is an address that the cognitive system uses. This, of course, is absurd. The Coda is a phonological object in its own right that is addressed as such by the cognitive system. I leave it up to the reader to imagine what our understanding of syllable structure would look like if phonologists had denied a positive identity to the Coda, treating it only as a default. Since the Coda Mirror context is recurrent in a number of genetically unrelated languages, the formal requirements of (391) must also the Coda Mirror disjunction "word-initially or after a heterosyllabic consonant". Theory is called express this phonological reality in terms of a natural class. As before, the phonological object in question must not be of disjunctive nature (391a), and it ought to be different from any other syllabically determined item (391b). Clearly, classical syllable structure cannot comply with these desiderata since the syllabic identity of consonants in wordinitial, post-Coda and intervocalic position is defined by the constituent they pertain to: an Onset. Therefore, there is no hope that the Coda Mirror {#,C} will ever be distinguished from the intervocalic position on the grounds of the familiar arboreal syllable structure that operates with Onsets, Rhymes and Codas (§123). Also, the requirement regarding the non-disjunctive and unique description of the Coda Mirror is accompanied by an explanatory issue. Recall that both the Coda and its Mirror are not only antipodal with respect to their structural description ( {#,C} vs. {#,C} ), but also in regard of the effects they produce: consonants in Codas are damaged, while they are guaranteed against damage in the Coda Mirror. The antagonistic properties of the position and its effects can hardly be regarded as accidental (§124). Any theory of lenition and fortition is thus invited to tell why the Coda provokes weakness and the Coda Mirror strength, rather than the reverse. It has been demonstrated how CVCV characterises the Coda Mirror as a uniform and unique object (§124), and how the antagonistic effects of the Coda and its Mirror follow from the distribution of Government and Licensing (§130). Classical syllable structure may not seek a solution along the lines of the strategy that was used when the Coda context was at issue. When SPE was faced with the disjunctive context {C,#}, a new constituent called "Coda" was added to the syllabic inventory that consisted only of vowels

Classical constituency fails to describe the Coda Mirror as a natural class 669 and consonants. Two kinds of consonants were now distinguished, rather than only one: those pertaining to Onsets, and those associated to Codas. A similar move cannot be made in order to implement the Coda Mirror into syllable structure. If the minimal syllabic unit is considered to be CV, the Coda could be introduced because the right edge of the syllable was virgin. Its left edge, however, is already occupied by a constituent, the Onset. Subdividing Onsets into "real" Onsets in intervocalic position and Coda Mirrors would lead to the absurd situation of encountering syllables that sometimes begin with an Onset, and sometimes with a Coda Mirror. Unlike in the case of Codas, syllable typology would not bear the alternative parameter "presence vs. absence of the one object (Coda)", but the mutually exclusive presence of two objects, i.e. an Onset or the Coda Mirror. This kind of structure would fail to express any relevant syllabic generalisation, and merely grant a notational existence to a phonological reality that cannot be accommodated.

560

2. Is the Coda Mirror a phonological object ? Another solution that could be thought of in order to salvage classical constituency is a purely negative definition of the Coda Mirror: a consonant sits in the Coda Mirror if and only if it is not a Coda nor stands in intervocalic position. In other words, a consonant belongs to the Coda Mirror iff it does not occur after a vowel. Consequently, Coda Mirror effects are no phonological events, they are non-events. For example, lenition affects positions that are prone to lenition, i.e. those after a vowel, while consonants in the Coda Mirror are left untouched simply because they do not meet the structural description of the phonological process at hand.

561

2.1. Tiberian Hebrew spirantisation

A typical example along this line of reasoning is the well-known spirantisation found in Tiberian Hebrew whereby each underlying stop /p,b,t,d,k,g/ may appear as either plosive [p,b,t,d,k,g] or fricative [φ,β,θ,δ,χ,γ] (e.g. Joiion 1923, Kenstowicz 1994:410ss, Lambdin 1973:XIX). The distribution of both allophonic variants is commonly referred to as a function of vocalic contexts: fricatives appear post-vocalically, stops elsewhere. This statement is correct. However, the description under (392b) that builds on the reverse context is strictly equivalent.

670 (392)

II, 13 Arg. 8 The Coda Mirror two empirically equivalent descriptions of Tiberian Hebrew spirantisation a. fricatives appear after vowels, stops occur elsewhere. b. stops appear word-initially and after consonants, fricatives occur elsewhere.

For instance, consider the behaviour of Ibi under (393), where it occurs as first, second and third radical, respectively. 347 (393)

Tiberian Hebrew spirantisation

root

pf3msg pattern CiaaC2aC3

qal = simple ipf3mpl yi-CiC23C3-uu

imperative 2f C1ÍC2C3-Ü

Vbq?

baaqa?

yi-ßqai-uu

biqî-ii

to split

Vjbr Vktb

Jaaßar kaa0aß

yi-jbar-uu yi-xtaß-uu

Jißr-ii ki0b-ii

to break to write

If Ibi is the first radical of a verb like in VbqT "to split", it appears as [b] in initial position [baaqa?, biqîii], while the spirant allophone [β] is observed when it comes to stand in an internal Coda [yißqaTuu] or in intervocalic position in intensive forms [yaßaqqa?uu] (cf. note 347). The roots Vfbr and Vktb demonstrate the same behaviour. In addition, they show that Ibi surfaces as [b] in post-Coda position [yijbaruu, kiBbii], while it appears as [β] word-finally [kaa0aß]. The general picture that may be drawn from these alternations thus features strong plosive allophones in the Coda Mirror context {#,C} , while weak fricative versions are found in Codas and intervocalically. The behaviour of those plosives that have not been examined is identical to the one observed for Ibi. 347

Assuming either description, a special provision must be made for geminates. Indeed, the first part of a geminate, although occurring in a Coda, is not affected by spirantisation, as can be seen in intensive forms: the relevant binyan pi?el is [y3-C|aC2C23C3-uu] in ipf 3m pi, which produces [ya-ßaqqaT-uu, ya-jabbar-uu, ya-xattaß-uu], respectively, for the three verbs mentioned under (393). This behaviour illustrates the well-known phenomenon of geminate integrity (e.g. McCarthy 1986:226ss, Kenstowicz 1994:41 Iss, Perlmutter 1995:309s), which does not interfere with the present purpose. It has also been noticed that in some cases, fricatives do occur after Codas, e.g. [malxee] construct state pi. from [melex] "king". These instances are due to a vowel syncope, cf. McCarthy (1986:234).

Is the Coda Mirror a phonological

562

object?

671

2.2. Non-disjunctive contexts win - but who wins when everybody is nondisjunctive ? As was mentioned earlier, it is commonly assumed that Tiberian Hebrew spirantisation is triggered by the action of a vowel on the following consonant. This interpretation is due to the distributional situation which seems to suggest that fricatives occur in a uniform context (postvocal ically), while the environment of stops ({#,C} ) is heterogeneous. It follows the principle that out of two contexts which describe a complementary distribution, the one which allows for a non-disjunctive formulation is mentioned in the rule. Since it is true indeed that classical constituency cannot reduce the stop environment {#,C} to a non-disjunctive object, the relevant context must be "post-vocalic". CVCV, however, offers a non-disjunctive formulation for both complementary contexts: as before, "post-vocalic" characterises the environment in which fricatives occur; but "after an empty Nucleus", i.e. in the Coda Mirror, can now refer to the contexts where stops appear without any disjunction. In this situation, then, the motivation for declaring that the post-vocalic environment is the active phonological agent disappears.

563

2.3. The choice of the context determines the causality of the process The difference between both formulations (392a) and (392b) is important because it determines the causality at hand: if the former description is correct, there is a process of spirantisation which selectively chooses postvocalic consonants as targets. On this analysis, the phonologically active agent are vowels: there is a transmission of some vocalic property from the vowel into the following consonant. If on the other hand the Coda Mirror is the phonologically relevant environment, there is a general lenition process in the language that potentially affects all stops. Only are some stops sheltered against the sharp lenition wind because they stand in a position where this wind is moderated, i.e. the Coda Mirror. In this case, the causality at hand has got nothing to do with the transmission of any melodic property from one segment to another: there is no communication between adjacent segments at all. Consonants are prone to lenition according to the position in which they occur. Therefore, the two formulations (392a) and (392b) are not just a matter of vocabulary; they imply two contrasting causalities, hence two very different phonological realities: what we see is either due to a positional

672

11,13 Arg. 8 The Coda Mirror

parameter (stops are protected in the Coda Mirror), or to the contamination of a segment by a neighbour (transmission of a melodic prime in postvocalic contexts). 348 If the post-vocalic solution is correct, the Coda Mirror is no phonological object because it merely constitutes the complementary set of the environments in which a phonological process takes place. Consequently, there is no need to address it positively as such. Classical constituency could then comply with the challenge that is raised by the phenomenology of the Strong Position: in phonology, consonants that are strong are not strong inherently; they just happen not to be weak. This would contrast with the opposition of the other two kinds of consonants: Codas are not just "consonants that happen not to be intervocalic" or, alternatively, intervocalic consonants are not just "consonants that happen not to occur in a Coda". Here, both kinds of consonants enjoy a positive definition in terms of syllabic constituency (Codas are post-vocalic but not pre-vocalic, while intervocalic consonants are both post- and pre-vocalic). The three relevant contexts for lenition that are identified under (389) would thus be heterogeneous in kind: two of them, the Coda and the intervocalic position, are defined positively as phonological objects, while the Coda Mirror has no positive phonological or cognitive existence: it is only the complementary set of all other environments. It follows that a strong argument regarding constituency is made if it can be shown that the Coda Mirror not only represents a phonological object that is uniform (391a) and a unique (391b), but also has a positive existence in phonology and in the cognitive system (391c). Classical constituency would then be in a desperate position with no way out even on the descriptive side, let alone explanation (i.e. the mirror effect). In the remainder of this chapter, I develop six arguments in order to show that the Coda Mirror must be regarded as a positive object whose phonological identity is as real as the one of Onsets and Codas.

564

3. Why the Coda Mirror is as real as Onsets and Codas

565

3.1. Fortifions A fairly obvious argument against the view that the Coda Mirror is only the residue of other environments is the existence of fortifions (also called 348

This issue will be further discussed in Vol.2,II.3.6.

Why the Coda Mirror is as real as Onsets and Codas 673 strengthening or hardening). The relevant literature (e.g. Foley 1977:90ss, Lass 1984:177ss, Collinge 1985:93ss,243ss, Harris 1990, 1994a:132s, 1997, Hock 1991:162ss, Clements 1993:122ss, Kenstowicz 1994:35, Trask 1996:55ss) quotes various instances in a number of genetically unrelated languages. The focus is generally oriented on the segmental variations that occur, rather than on the position in the string that conditions fortition. However, Kenstowicz (1994) summarises the relevant issue as follows: "postvocalic context is the most typical environment for the change from stop to fricative [...] Many systems restrict weakening to contexts in which a vowel follows as well as precedes [...] Fortitions from fricative to stop tend to occur in the complementary set of contexts: postconsonantal and initial." Kenstowicz (1994:35) If there is a process that turns weaker into stronger segments in a given environment, this environment must be explicitly addressed by the phonology of the language. Hence, in case fortition occurs in both wordinitial and post-consonantal position, the disjunctivity begs the question in positive fashion: this disjunction must enjoy a cognitive reality. Note that the Coda Mirror makes no prediction to the effect that fortitions are always observed both in word-initial position and after Codas. Fortition is expected to occur either in all of the Coda Mirror, or in one of its sub-parts only. The fact of being observed in half of the Coda Mirror is not at odds with the existence of the Strong Position. Just as is the case for Codas, relevant processes may well affect only half of the disjunction. This issue was briefly discussed in §558c where I suggested that the behaviour of the Coda Mirror is symmetric with what we know from the Coda. If there is a process that affects only half of the Coda Mirror, consonants in post-Coda position will be concerned. Events related to fortition that occur word-initially but not after a Coda do not appear to exist. Vol.2,III.6.3.2 offers more discussion of this issue. On the other hand, the Coda Mirror does make a prediction to the effect that no language shows a fortition intervocalically and/ or in Codas, while no fortition of initial and/ or post-Coda consonants is observed. Conversely, if lenition occurs in (part of) the Coda Mirror, it must also be observed intervocalically and in Codas. A number of fortitions that occur in one or both parts of the Coda Mirror are reviewed in Ségéral & Scheer (2001a). The focus of the present section is on the phonological reality of the Coda Mirror as a disjunctive object. Therefore, the fortition discussed below represents a case where both word-initial and post-Coda consonants are strengthened.

674

11,13 Arg. 8 The Coda Mirror

The process which is most commonly quoted in the literature in order to illustrate strengthening concerns the evolution of Latin [j] in French (e.g. Lass 1984:177ss, Trask 1996:55ss, Clements 1993:124).349 Table (394) provides relevant illustration. (394)

fortition of Latin [j] in French a. # b. Coda

c. Coda

C jocu jurare

0

3 3yire

sapjam saj rubju ku33

d. V

V

# maj(u) me

raja

jejunu 3 ι d α. υ υ

C> O a Χ

? ? ρ e A i •δ ο IΡ fl« ω ω ff Μ Λ

S

c

•à Ο (J Ρΐ S¿ if· ω

± β Ss 'S C

cm

C7 D ι

s ta § oo 8 ;§· cm X« ΰ (ν 1 ίV Βω Λ S O e ωn I NU > C M

X Î

C O ω s Ν

C rt So ·«? CM

C S m S; c

C β ti ce S >

C η tt cd «>

C η u S ta

^

c c Β () (t Οι S c β to tu ISP s I I I ρ XI υ

5 ω S u ~ o 3 .«í -e £ fe (Λ " u o s δ -S ¡> o o Cfl — ¡> ? Pi D ..û Ν . - ΙΛ

SO

Tj-

fx


ì§

>I

"" Ό J3

60

|

C O I

VI

C ω

741

742

617

11,15 Arg 10 Sonorants in Codas: melodic reaction on positional

plight

10. Conclusion: a unified theory for the behaviour of sonorants in Coda position The analysis that was originally made for nasals has been progressively extended to other sonorants. In this light, a whole range of phonological processes that are well known but generally not assumed to share any particular feature may be interpreted as the result of one single causality: the weakness of sonorants in Coda position. This positional discomfort may result in three major types of reaction: 1) the sonorant seeks to share melodic (place) primes with a neighbour and thereby achieves a branching status; 2) the sonorant spreads onto a syllabic position of a neighbouring segment and thereby achieves a branching structure; 3) the sonorant does not succeed any of those performances and falls prey to lenition: it must release manner and/ or place primes. For the sake of clarity, the table under (438) summarises the set of apparently unrelated melodic phenomena that now receive a unified account. (438) processes that fall under the scope of the theory type of reaction result on another only nasals segment: 1. homorganic NC and CN clusters spreading shared place 2. genesis of nasal vowels (successful nasals and liquids stabilisation) on another 1. genesis of syllabic consonants position 2. genesis of trapped consonants nasals and liquids 1. liquids: depletion of manner and place primes lenition l,r — Ü1, * — [w], r — [β], r — [r], (unsuccessful stabilisation) [i] - ra 2. nasals: depletion of place primes Imi — [η], /η/ — [η] In sum, this theory has bearing on (almost) all phonological processes that are known to affect sonorants. Hence, an interesting question to be asked is certainly why obstruents do not enter this picture. And the answer, alas, cannot be any different than "because they are obstruents". That is, we are confronted with the intrinsic and inalterable natural properties of so-called major classes of segments. Obstruents are never observed to conform their place of articulation to a neighbouring segment because they are

Conclusion: a unified theory of sonorants in positional plight

743

obstruents. The same holds true for liquids. Only nasals enjoy this privilege. This is but another way of expressing the well-known "volatile" character of nasals. Let us now return to the statement according to which obstruents do not enter the picture under (438). This is not entirely true since obstruents do share one possible reaction on positional plight with sonorants: lenition. Also, it is not fully uncontroversial to say that only sonorants may be syllabic (or trapped). Recall from §294 (note 195) that some languages are reported to tolerate syllabic obstruents. However, there is no doubt that sonorants are the "preferred" syllabic (and trapped) consonants, and that there is an implicational relationship between sonorants and obstruents: all systems that allow for syllabic obstruents also possess syllabic sonorants, while the reverse is not true.386 Finally, there is no doubt at all that only nasals can become homorganic. This means that the behaviour of consonants can also be used as a definitorial property of major classes. Table (439) summarises the foregoing discussion. (439)

definition of major classes according to their behaviour under positional pressure can spread onto another can become syllabic position (i.e. becan experience homorganic come syllabic or trapped) lenition nasals yes yes yes liquids no yes yes obstruents no no (?) yes

Finally, it is worthwhile observing that the action of sonorants under positional pressure can be either progressive or regressive. Below the skeleton, the directionality of processes is not restricted (see Vol.2,II.3). The positional pressure, however, is created by Government and Licensing. 386

Saussure's (1879) laryngeal theory, i.e. the Mémoire sur le système primitif des voyelles dans les langues indo-européennes, is a famous case where the critical difference between sonornats and obstruents is their ability to be(come) syllabic: the zero grade of CejC and CewC roots is CiC and CuC, respectively. It can exist because the glides are able to take over syllabic function. The same holds true for liquids and nasals: the zero grade of CeRC, CeNC is CRC, CNC, respectively. CeTC roots, however, are unable to appear in zero grade. Also, the entire classificational system of roots in traditional Indian grammar is built on this distinction.

744

11,15 Arg 10 Sonorants in Codas: melodic reaction on positional

plight

These lateral forces are located ABOVE the skeleton and therefore must be regressive. Hence, the same pattern that was observed on the occasion of the study of vocalic and consonantal length (§229) is reproduced here. The argument in favour of CVCV should be clear from the preceding discussion. The rather general picture of (437) results from an initial reasoning that was a good deal more narrow in scope: the question was simply how the homorganicity of NC clusters could be represented in CVCV. The well-established and largely consensual way of analysing this phenomenon is not available in CVCV, something that looks like a serious setback on the face of it. It is only when the mechanisms of CVCV are explored that a new scenario emerges that reverses roles: nasals are not passive, they are active. There are neither servants nor masters. In this light, new pieces of data that are all well known but generally unrelated to homorganic NC clusters enter the picture. The final unified theory for the behaviour of sonorants in Codas could not have emerged in absence of the prism that CVCV requires to use when looking at the data.

General Conclusion

In conclusion of this book, there is not much to say. I hope that the programme which was defined in the foreword has been fleshed out to a certain extent. The major issues are introduced there. Perhaps it is worthwhile mentioning the two most important goals of the book, one concerning the evolution of Government Phonology from a theory-internal perspective, the other reaching beyond theory-specific debate. That is, CVCV takes the research programme of Government Phonology, i.e. the latéralisation of structure and causality, to its logical end. Thereby, it dispenses with hybrid in-betweens that inevitably run into hybridity-caused trouble (§§208,398400). Second, the flatness of CVCV and the claim that phonology lacks a tree-building device is a straightforward consequence of the purely phonology-internal and theory-specific approach developed. 387 However, it matches, actually predicts, the absence of recursion in phonology and meets the general picture that Chomsky et al. (2002) draw of the human faculty of language (see the foreword §2). Also, there is one point of relative importance that has not been addressed yet. At the outset of Part Two (chapter 11,1 §302), a roadmap gathers the ten arguments that are produced, as well as the syllable-relevant conclusions which are drawn on their grounds. When looking at the latter, it appears that seven times out of ten, the demonstration aims at establishing the existence of an empty Nucleus in the middle of what is classically held to be a Coda-Onset sequence (arguments one §381 and five §426 to ten §591). One chapter shows that empty Nuclei and lateral relations have existed for a long time in the analysis of Slavic (argument four §411), and another argues that there is no way of accepting autosegmentalism without acknowledging the existence of empty Nuclei (argument two §387). Only one single argument is made to the effect that what is classically held to be a branching Onset in fact straddles an empty Nucleus (argument three §390).

387

However, recall the disclaimer that was mentioned in the foreword: the flatness claim is restricted to syllable-related processes for the time being. Whether higher prosodie phenomena require arboreal structure is a matter of further study, which will be engaged in Vol.2,III.

746

General conclusion

When looking at this record, it appears that the reader has been provided with quite some material that could make him incline to break up Coda-Onset sequences. However, CVCV requires more than that: branching Onsets and branching Nuclei are supposed to enclose an unpronounced empty slot as well. The record on this side is poor or inexistent. To which extent, then, does this book actually do justice to its subtitle, which promises that the reader will discover why CVCV is a interesting alternative? The answer is twofold. For one thing, it was mentioned in the introduction of Part Two (chapter 11,1 §302) that the collection of arguments put forth in this book represents but a personal choice which reflects the empirical field and the languages that I happen to be familiar with. It neither aims at exhaustivity nor at providing a balanced picture of what could be said. But also, if this book succeeded to convince the reader of just one thing, the existence of an empty Nucleus separating Coda-Onset sequences, it is hard to see how the missing steps towards CVCV could not be taken. That is, an overall picture where Codas do not exist and Rhymes cannot branch, but Onsets and Nuclei can, is nothing that makes sense. Therefore, I believe that breaking away the pillar of Coda-Onset sequences will be fatal for the entire branching edifice. Writing this conclusion about two and a half years after the first line was put on paper, there is not much left of the original conception, and also the content is not exactly what I thought it would be. For one thing, the book has evolved in a direction that is probably more historical than it was intended to be: at various occasions (with a concentration in chapters 1,7 §135 and 1,8 §165), explanation is provided why and how CVCV is different from Standard Government Phonology, and how the transition was made. It remains true, however, that readers without any prior knowledge of Government Phonology should be able to take advantage of the book. Its primary function is announced in the subtitle and in the foreword: the reader is introduced to the properties of CVCV, which means first of all that the representation of basic phonological objects is discussed; in a second step, a number of arguments is provided that may incline the reader to consider CVCV a serious competitor in the present-day phonological scene. A second feature, namely of Part Two, which to a certain extent enlarges the original conception, is the increased number of locations where direct discussion with concurrent theories is engaged. The most prominent case in point is §309 on the necessary presence of both sovereign structure and process in theory. This issue addresses OT directly; my thoughts on the

General conclusion

747

matter have been prompted by a conference presentation in July 2003, which I have decided to incorporate into the book. Finally, these last pages of the book should be considered as a transition, rather than as a conclusion: a transition towards volume two, which is referred to all throughout the present book, and of which a draft version is almost ready for circulation. A flavour of what it will look like is given in the editorial note at the outset of this book (§1).

619

Appendices

620

Appendix 1 List of parameters and their translation into CVCV and other theories On various occasions in this book, contrasting empirical patterns have been interpreted as a language-specific variation of a given principle. It is useful to summarise the different parameters that have been proposed, to show which kind of empirical contrast they are supposed to cover, and to provide their traditional interpretation. The following table offers a global survey.

750

Appendix 1

oo m o Ό 00 Os ^ eoo coi u»

(N m CN

O 5 > .SP o J > Κ o ω C > C* (Ν

List of parameters and their translation into CVCV and other theories

«Ν

00 o

T3 υ Ζ ' s £ •o

CO

-Ô)• Λ g * ^ Ό ST o * 2 O ' ^ r\

& i, ss ^s v «^ Όo „¡ CA ω

* o u o I

¡tí ce

U

S

•a 3

.3 «

υ

« « -S

υ S

-κ ο « CA S .S

·

-s

S« U e 'S ω S I 'v s •ρ "c3

t/f

οο .3 ο υ ti Ο - 133 ν O 3 ce

s s s «

G

Μ (β „ ·Ό CΟ

.2 υ +•» ν—'

ο α, ν. ο ξ ΐ à i υ .. Ο J •S υ

υ sη § « g

-g O3u

ο α Ë α υ S

ο υ -§

260- "S ä •S o -m S & ε .2c y.-β s 5 c3 S s

co

cod :s α ευ 0> en J3

Appendix 2 Closed Syllable Shortening vs. diminutive lengthening in Czech Closed Syllable Shortening affects both internal and final closed syllables in Czech. This is the empirical situation that is made reference to at several occasions in this book, including §§154,222,430,534. Some comments and restrictions are in order here. For one thing, as is mentioned in the main text, Closed Syllable Shortening does not occur throughout the entire language, but operates only in certain morphologically defined paradigms. These are feminine nouns in -a (on the model of zena "woman") and neuter nouns in -o (on the model of mesto "city"). Moreover, the alternations at hand are most probably not performed by the synchronic grammar of Czech speakers: even within the paradigms mentioned, some nouns with long root vowels alternate, while others do not react on the closed syllable environment. Table (440) provides some illustration of this fact for the feminine -a declension.388 (440)

long roots + -ka: Closed Syllable Shortening a. shortening altershortening in #, CCV nation noun -ka GENpl -CV skyva skyvka y-y i-i ¿ila Zilka ¿il ¿ilní á-a skála skalka skal skalni ou-u bud bouda budka

b. no shortening alternation noun dyha y-y i-i bfiza é-e bariéra á-a fasáda ó-o sezóna ou-u roura ü-o puda

388

no shortening in #, CCV -ka GENpl -CV dyh dyika bfizka bfiz bariér bariérka fasádka fasád fasádní sezón sezónní sezónka rourka rour püdní pûdka pûd

nb 1 10 13 7 31

nb 5 12 3 17 1 9 5 52

Glosses of the roots (-ka forms are usually diminutives, -ni produces the related adjective): a) "slice, vein, rock, small house or place", b) "veneer, birch, barrier, façade, season, pipe, loft".

754

Appendix 2

The table provides numeric proportions for alternating (31 items) and non-alternating (52 items) roots. The count is based on the reverse dictionary Slavícková (1975) and further lexical material that I have collected in Scheer (200lb,c). 389 Each timbre and each long vowel are represented by one root (in NOMsg). The following columns provide the diminutive form in -ka, the GENpl and forms with consonant-initial suffixes (e.g. the adjectival -ni). It may be seen that there are more long roots of the -a declension that do not participate in Closed Syllable Shortening than there are that do. This is one good reason to believe that the alternation at hand is not synchronically active. Also note that recent loans such as sezóna "season", bariéra "barrier",fasàda "façade" etc. do not shorten. Of course, the non-synchronic character of Czech Closed Syllable Shortening does not call into question its existence. The process was active at a former stage of the language, and traces thereof have survived into its modern variety. The synchronic inactivity of Closed Syllable Shortening is also evidenced by the interference of another length-modifying process that operates on the same paradigm. In Czech, diminutive suffixes have a lengthening effect on roots that possess a short vowel (e.g. Trávnícek 1948-49:440, Dokulil 1962:170ss, Dolezel 1967, Hausenblas 1967). This holds true across the board for masculines, neuters and feminines. Hence, the nondiminutives vlak (mase), bficho (neuter), kniha (fem) "train, stomach, book" are lengthened when the diminutive suffixes -ek (mase), -ko (neuter), -ka (fem) are added: vlácek, brísko, knízka "little train, little stomach, little book" (an acute accent on vowels indicates length). The same lengthening is provoked by another kind of diminutive morpheme, -ë, which is used in order to derive "joung individuals" from living species. Hence, junior individuals of had (mase), hovado (neuter), sova (fem) "snake, cattle, owl" are hadë, hovâdë and sùvë, respectively. This shows that the conditioning does not take into account the phonological properties of the suffixes at hand, which may be either consonant- or vowel-initial. Rather, it is the semantic property "diminutive" that triggers the action on the root vowel. This kind of phenomenon is classically known as templatic in Semitic languages. Greater detail is available in Scheer (2001b,c,2003b, in press b).

389

Where all numbers are fleshed out by the examples they represent. Detailed data on Closed Syllable Shortening of the -o declension are also available there.

Closed Syllable Shortening vs. diminutive lengthening in Czech

755

Table (441) provides illustration of diminutive lengthening in the feminine -a declension. Numeric proportions are indicated as before. 390 (441 )

short roots + -ka a. lengthening alternoun nation i-i kniha Stika 0-é vëtev e-í stïecha ë-ί vëi dëva e-é konev hlava a-á straka o-ü hora sova u-ou duha husa

-ka kntëka

StíCe vôtévka sth'Ska vfëka divöe konévka hlávka stráde hûrka sûvë dou2ka 24

390

joung individual

house 15

nb 3 5 4 5 2 1 1 2 4 4 3 3 2 39

Glosses of the roots (-ka forms related diminutives): a) "book, pike, branch, roof, tower, virgin, jug, head, magpie, mountain, owl, rainbow, goose", b) "fish, otter, bank, fox, sheep, bee, squirrel, female first name, virgin, heel, duck, foot, swallow, hand, swan". Note that the "0" in the first column identifies vowels that alternate with zero (e.g. vëtev "branch NOMsg" vs. vëtve "id., GENsg")

756

Appendix 2 b. no lengthening alter- noun nation ryba y-y vydra i-i lavice liSka ovce 0-é e-í vëela veverka ë-ί Bèta dëva a-á pata kachna o-û noha vlaStovka u-ou ruka labut'

-ka

joung individual

rybka vydfe laviòka liSöe oveíka vöelka veveròe Bëtka dëvôe patka kachnë noika vlaStovòe ruòka 13

labutë 29

nb 2 1 3 3 1 2 4 1 1 1 10 1 5 2 5 42

As before, there are far too many roots that do not lengthen in the appropriate diminutive context in order to think of this process as a part of the synchronic grammar of Czech. And again, this fact does not make the process less real: it has operated in the synchronic grammar of an earlier stage of the language. Diminutive lengthening interferes with Closed Syllable Shortening because the feminine diminutive suffix -ka creates closed syllables if it is concatenated to consonant-final roots. In cases such as hlava "head", one could therefore expect a short root vowel when -ka is added, but this is not what happens: the diminutive value of the suffix provokes lengthening, i.e. hlávka "little head". There are words in the language that do not allow for long vowels in closed syllables, while others are comfortable with this kind of configuration. The latter may either already possess a long vowel in the lexicon as in the right half of table (440), or lengthen an underlyingly short vowel due to diminutive semantics in presence of a closed syllable that is created by the diminutive suffix. The only thing that needs to be understood for the sake of the various demonstrations that use Czech Closed Syllable Shortening throughout the book is that both length-modifying processes are orthogonal. Diminutive lengthening does not call into question the existence of Closed Syllable

Closed Syllable Shortening vs. diminutive lengthening in Czech

757

Shortening (and vice versa) since it can be established independently that neither process is active in the present-day grammar of Czech.

622

Appendix 3 Polish two-membered word-initial consonant clusters §375 has inquired on existing and possible consonant clusters at the beginning of Polish words. The exhaustive list of three- and more-membered initial clusters was provided, but space restrictions precluded the spell-out of all two-membered groups. They are listed in the present appendix. However, a number of provisions need to be made before going into further detail. For one thing, trapped consonants such as in trwac [trfatç] "to last" call for a comment. Trwac shows an initial [trf-] sequence that contains a trapped [r]. Trapped consonants, their behaviour and representation have been discussed at length in chapter 1,10 (§240). It was concluded there that they are the mirror image of syllabic consonants: the latter branch on the preceding Nucleus, while the former colonise the following vocalic slot. If this analysis is correct, the situation of trapped roots in regard of their initial cluster is parallel to the one of regular roots that begin with a twomembered cluster: in both cases, two initial consonants are followed by a vocalic position that is phonetically expressed (as trappedness in the former, as a vowel in the latter configuration) and therefore endowed with lateral actorship. Table (442) illustrates this situation. (442) trapped roots instantiate a two-, not a three-membered word-initial cluster a. trapped root b. root with a two-membered initial cluster Gvt Gvt

i

I

f

c v c v c v c v I t

U—•— I r w

I a

I é

c v c v c v c v I C

I

I I I I C V C a é

Polish trwaé "to last"

Whether one is inclined to follow the analysis of trapped consonants that appears under (442a) or not, trapped roots require a special provision anyway. Their empirical properties do not allow for a solution that treats them on a par with ordinary consonant clusters. For all these reasons and following (442), thus, trapped roots will be counted as roots with a twomembered initial cluster.

760

Appendix 3

Another provision needs to be made for s+C sequences where "s" represents all s-sounds [s,z,J",3,ç,?]. S+C sequences are not taken into account. It is one of the most striking, cross-linguistically stable and notorious observations that these clusters represent a phenomenology of their own (see §96). In #sC sequences, s- plays a mysterious role since it sometimes behaves as a kind of secondary articulation of the following consonant, but on other occasions patterns with geminates (see for example the discussion in §368, Rowicka 1999a:257, Weijer 1994:165ss). Also, word-initial s+C sequences disobey the regular sonority restrictions on initial clusters in typical Indo-European languages such as English or French. Hence, whatever the correct interpretation of s-sounds (which I do not intend to discover here), the problems they raise are independent from the issue that is at stake in the present discussion of word-initial clusters. Therefore, s+C sequences are counted as one single consonant. Or, in other words, s-sounds are treated as if they were not there. Table (443) does not contain any word with an initial s+C cluster on purpose: roots beginning with #CsC- and #sCC sequences fall into the ordinary #CC-class. Table (443) is reproduced from Rowicka (1999a:309ss), who has collected relevant data in Bargietówna (1950) and Sawicka (1974). It lists all word-initial two-membered consonant clusters that occur in Polish roots. Heteromorphemic sequences are not taken into consideration. 391

391

The table is phonetic in the sense that [J/3], which alternates with [r], is considered as a fricative, not as a liquid.

Polish two-membered word-initial consonant clusters 761

60 Τ3

60

là J3 y o e o CO 60 .S

V

Μ Χ)

ΚΊ Τ3

Ν 60

Ν60

F.

60

$

£

2

tí

> Í>\ -2 -s Β l i¿2aO 'S 1í q=¡ Ο •g « — Iυ Τ3 B O c d Ss m 4υ—i cυo C O υ — 3σ"M > ä> cυo cd •ΉJ* C oa* υ "Λ ï/> S li © ε cο èχa S D -O r« g TO ÑI « C OS υ y CO o O » £ £ » 60 .s υ Χ> — :Χ> Χι IO Ο > >Τ3 δ ÊH? ««StoGS» α ο υ

763

ω α» ω > 13 > te .y ^ O ib .y «ti cd

764

Appendix 3

o "eh eo . 3 o Β ω

*ì co

Κ ω Έ *r

•S" υ Ü

>>

•1 O Β O cfl β c . ί 53 l u ?

öO .S υ jo o

em

.S

CO o co Β O υ "öb 3 σ" ω CO

— C ο α ο

Μ

"Ξ. Η S3 χ α>

ÖD . 3 4> Χ> ο

co

•5

α> «

3 ΰ

g

ε

Urn

623

Appendix 4 A short guide to 1990 Government Phonology This appendix is designed as a glossary that may be consulted by those readers who are less familiar with the initial version of Government Phonology as it stood in 1990 (Phonology Yearbook 7.2, and also Kaye's 1989 foundation-laying book). No effort is made to derive, explain or discuss the structures that are reported, nor does this presentation aim at being exhaustive. Its only function is to make this book more reader-friendly. At various stages of the discussion and namely in Part One, it may be useful to have the simple list of segmental and syllabic representations in the backhand. A more advanced "Users' Guide to Government Phonology", which has similar scope, is available as Kaye (2000). This very condensed and useful introduction reflects Jonathan Kaye's recent views on the matter, where melodic representations have changed a whole lot (there are only six, perhaps five Elements left), but constituent structure has more or less remained the same.

624

4.1. Melodic representations As in all other models of melodic representation, the basic building blocks have been derived from the inspection of vowels. In Government Phonology, this was done in Kaye et al. (1985). Subsequently, the tools gained on vocalic grounds have been used in order to represent consonants. This step was taken by Harris (1990). Government Phonology uses privative (also called unary, monovalent or holistic) melodic primes. This interpretation of phonological substance was first introduced by Anderson & Jones (1974,1977). Various implementations of their idea have then been proposed by Government Phonology, Dependency Phonology (Anderson & Ewen 1987) and Particle Phonology (Schane 1984). Privative systems of melodic representation contrast with more traditional approaches that directly root in SPE where the basic building blocks are binary features. These are represented by Feature Geometry (e.g. Clements 1985, Clements & Hume 1995) and developments thereof such as Underspecification (e.g. Archangeli 1988, Pulleyblank 1988,1995, Steriade 1995a), which partly align themselves on the privative idea (see §41). Privative primes express oppositions by the fact of being either present or absent, while binary features are always present in representations. The two options make contrasting predictions since,

766

Appendix 4: a short guide to 1990 Government Phonology

for example, something that is not there cannot spread or be an active agent in phonological processes. In Government Phonology, the privative primes in question are called Elements (in explicit reference to chemistry). Elements freely combine within the frame of some basic rules. Abstracting away from Charm (see §§37,54s), these boil down to a Dependency relation: a segmental expression that is co-defined by more than one Element contains exactly one head. Those Elements which are non-heads are called Operators (i.e. dependents). The properties of the head are supposed to leave a more salient acoustic trace than those of Operators. For instance, the combination of the Elements I and A produces a mid [ε] or a low [as] vowel according to the distribution o f headship: in case the expression is I-headed, the phonetic result will be closer to I (i.e. [ε]), while [ae] is the interpretation of an A headed structure. An additional important property of Elements that sets them apart from binary features is that they enjoy an independent pronunciation. The Elements that were assumed to define vowels are identified under (444a). Leaving aside secondary articulations such as ATRness 3 9 2 , nasality and tone, their combinations and phonetic result are shown under (444b). (444)

Elements and their occurrence in vowels according to Kaye et al. (1985) a. description of individual Eleb. internal structure of vowels ments (heads first) I i I frontness e I,A υ roundness u U ae A,I A aperture a A 0 U,A i ν ν neutrality, the "cold vowel" i,u y Ν nasality I,U,A e ATR ì

The consonantal representations that have been developed by Harris (1990) on the grounds of this system needed a minor readjustment in the articulatory characterisation of Elements. These are displayed under (445a). The way Harris (1990) defines various places and manners o f articulation is shown under (445b), while (445c) identifies the internal structure of some individual consonants in Harris' (1990) system.

392

Only +ATR vowels are shown, but the ATR Element i is left unmentioned for the sake of exposition.

Melodie representations (445)

767

Elements and their occurrence in consonants according to Harris (1990) a. description of individual Elements place definers manner definers ? constriction I palatality υ labiality h noise A L slack vocal chords absent from consonants R H stiff vocal chords coronality V Ν nasality velarity Ν nasality ϊ absent from consonants b. some places and manners places (heads first) bilabial ?,u labio-dental h,U interdental R,h alveolar R palatal I alveolo-palatal (J,3) h,I velar V

manners (no heads) — glides ? liquids nasals ?,N h fricatives stops h,?

c. some consonants (heads first) [m] ?,U,N [ρ] ?,U,h,H h,U,H [n] R,?,N Μ I,N [θ] R,h,H M R,?,h,H [t] [r] ?,R R,?,h,L [d] [1] R h,R,H [s] [c] I,?,h,H [k] v,?,h,H ? m [h]

h

Harris (1990) introduces the important notion of segmental complexity, which makes only sense in privative approaches to melodic representation. The complexity of a segment is calculated by simply counting the melodic primes that contribute to its articulation. For instance, the sound [p] under (445c) is made of four primes. It is therefore more complex than the sound [r], which is produced by only two Elements. Binary systems

768

Appendix 4: a short guide to 1990 Government Phonology

cannot possibly refer to this notion because all segments are defined by the same number of binary features. 393 Complexity is central for constituent structure because it takes over the function of sonority. Syllable structure relies on some version of sonority in all theories known. However, sonority is not a phonetic property. It can only be determined by the behaviour of the segments, and is independent from their melodic structure. Harris' complexity does away with any autonomous notion of sonority by making it a direct function of the independently motivated melodic structure of segments (see §37 and Vol.2, II.4.1). In a relation contracted by two consonants (a "sonority slope"), the one that is more complex dominates the less complex one. No dependency relation of this kind is possible if the candidate for the dependent position is more complex than the putative head. This principle is expressed as (446c).

625

4.2. Constituent structure

626

4.2.1. The basic architecture In Government Phonology, syllable structure is recorded in the lexicon. Consequently, there is no such thing as a syllabification algorithm. 394 Constituency is a function of two properties: the internal structure of segments and universal principles that govern constituent structure. The former have been introduced in the foregoing section. The latter consist of the axioms identified under (446), which may not be derived from any other property of the theory.

393

394

In underspecification approaches, however, complexity may become a relevant issue. But this of course is only because underspecifying a segment means nothing else than introducing the privative idea into the binary system (this fact is not usually mentioned in the underspecification literature, see §41). Except perhaps in acquisition: see Vol.2,11.4.2.1-4.

Constituent structure (446)

769

axioms controlling syllable structure according to Kaye et al. (1990) a. segments contract lateral relations whereby a head dominates a complement. b. syllable structure is a function of lateral relations. These are called Government. c. governees may not be more complex than governors. d. strict locality all governing relations are strictly local: no skeletal slot may intervene between the governor and the governee. e. strict directionality all governing domains are strictly directional 1. domains within constituents are head-initial Constituent CG Government 2. domains among constituents are head-final Interconstituent ICG Government

These provisions allow for the structures that are shown under (447), and for no others. (447) possible constituents in Standard Government Phonology Constituent Government Interconstituent Government (CG) relates two members of the same (ICG) relates two segments that constituent pertain to different constituents branching b. branching c. Coda-Onset d. OnsetOnset Nucleus Nucleus R R R 1 1 1 1 0 Ν N \ 0 Ο Ν 1 1 1 \ 1 χ χ χ X χ X X X X

/\

/\

1 t

L_t

CG

CG

ICG

t _ l ICG

That is, constituents may either be non-branching as under (447d), or branching as under (447a-c). Ternary constituents are outlawed by strict locality and strict directionality since for a putative ternary structure [xi X2 X3], the former principle requires X2 to be the head. However, X2 could not govern xi in this case because strict directionality allows only for headinitial relations within constituents. This consequence of (446d,e) is known as the Binary Theorem (Kaye et al. 1990:199s, Kaye 1990a:306s): all syllabic constituents are maximally binary. At the expense of making the awkward prediction that Nuclei and Rhymal Adjuncts are bound by co-

770

Appendix 4: a short guide to 1990 Government

Phonology

occurrence restrictions (see §178 on this), a stronger version of the Binary Theorem may be derived: "no constituent may dominate more than two skeletal slots" (Kaye 2000:6). That is, constituents with more than two terminal elements are outlawed, no matter whether intermediate nodes (such as the Nucleus in a super-heavy Rhyme) assure that there is no ternary constituent. Also, the internal structure of consonants derived by Harris (1990) assures that the usual major classes are reproduced: obstruents are governors, sonorants are governees. This follows from the fact that the former are more complex than the latter. Hence, only TR sequences qualify as branching Onsets: Τ can govern R, but R cannot govern T. Strict directionality produces the reverse situation for Coda-Onset interludes: as before, only Τ can govern R, but this time the relation must be head-final. Therefore, only RT sequences may instantiate the structure shown under (447c).395

627

4.2.2. A depleted version of the familiar syllabic tree The system described demotes the Coda to a non-branching existence, and ultimately to a non-constituent (Kaye et al. 1990:201s). This follows from the strong version of the Binary Theorem (see the preceding section): the Rhyme of a branching Coda would dominate three skeletal slots (the one of the Nucleus plus two from the Coda). Also, typical candidates for branching Codas show a falling sonority slope {carp, salt etc). In the terms of Standard Government Phonology, obstruents are typical governors, while sonorants are typical governees. Hence, the only governing relation that could hold within the cluster [φ] of carp is one where the [p] governs the [r]. This, however, would violate strict directionality: government within a constituent, hence within a putative branching Coda, is head-initial. Moreover, the traditional Coda never governs anything. It is therefore denied the status of a constituent (Kaye et al. 1990:201s). Its skeletal slot is directly attached to the Rhyme. Instead of "Coda", it is referred to as "Postnuclear Rhymal Complement" or as "Rhymal Adjunct". So-called super-heavy Rhymes are excluded by the strong version of the Binary Theorem (see the preceding section and §176).

395

This is only an approximation since, pending on their complexity slope, some RR and TT sequences also qualify as interludes.

Constituent structure

771

Finally, the principles introduced do not allow for the existence of a syllabic constituent that dominates the Onset and the Rhyme, i.e. the syllable node itself (see Kaye et al. 1990:200s, Brockhaus 1995b, Harris 1994a:45ss, Roca 1994:159ss on this). Were the Onset and the Rhyme sisters within a constituent, the former would have to govern the latter. This, however, would be in violation of the strong version of the Binary Theorem (see § 176), and moreover falls foul of the fundamental asymmetry between consonants and vowels: the centre of gravitation of the syllable is the vowel; consonants are adjuncts.

4.2.3. Indirect lateral conditions on the existence of skeletal slots This system relies on the lateral relation called Government. It is subjected to two well-formedness conditions that decide on the existence of dependent skeletal slots. One is called Coda Licensing, the other Government Licensing. Coda Licensing (Kaye 1990a) holds that a Rhymal Adjunct cannot exist on its own: it needs support from a following Onset in the form of a licensing relation that Kaye calls Coda Licensing (see §18). This lateral force is provided by the Interconstituent Government relation that a Rhymal Adjunct contracts with the following Onset. 396 As a consequence, there can be no word-final Rhymal Adjunct at all: since there is not following Onset that could provide Coda Licensing, all word-final consonants must belong to Onsets. This is shown under (448) below.

396

The terminological confusion is obvious: Interconsituent Government provides the Licensing that is necessary for the Rhymal Adjunct to exist. §136 elaborates on the confusing use of Government and Licensing in Standard Government Phonology.

772

Appendix 4: a short guide to 1990 Government

Phonology

(448) Coda Licensing: all word-final consonants in all languages belong to an Onset a. word-internal Rhymal b. word-final Rhymal c. word-final Onset: Adjunct: well-formed Adjunct: ill-formed well-formed R R R * R I 1

N \

Ο

Ν

N

tJ Coda-Lie

I 1 0

Ν

I 1

ι

1

I 1

X I 1

χ

χ

V

C

ι 1

Coda-Lie

In turn, this implies the existence of a Nucleus after word-final consonants: the minimal syllabic unit is an Onset-Rhyme pair; one does not occur without the other. Therefore, all words in all languages that are consonant-final on the surface in fact end in an empty Nucleus. These final empty Nuclei are one of the trademarks of Government Phonology (see §631). The other condition on the existence of dependent positions builds on Charette's (1990,1991) idea that governors do not enjoy their governing status per se. They need an external source of energy in order to be able to dominate their complement. This concept is known as Government Licensing: consonants need to be licensed in case they want to act as governors. Hence simplex, but not branching Onsets can be followed by an empty Nucleus: only contentful Nuclei qualify as licensors. For the same reason, a simplex Onset that is called to govern a preceding Rhymal Adjunct must not be followed by an empty Nucleus. In practice, Government Licensing was developed on the grounds of Quebec French where the realisation of schwa is optional unless it is preceded by a consonant cluster, in which case its presence is mandatory. The preceding cluster may either be a Coda-Onset sequence as in forteresse [fortares], *[fort'res] "fortress", or a branching Onset as in autrement [otramä], *[otr'mä] "in another way". In both cases, schwa cannot be dropped since the head of the preceding consonantal governing domain needs to be licensed in order to be able to govern its complement, and of course only contentful Nuclei can license. More detailed discussion of Government Licensing and its empirical basis can be found in §§125,169. Table (449) shows the two relevant configurations.

Constituent structure

773

(449) Government Licensing according to Charette ( 1990,1991) Gov-Lic

R

R

R

Χ

Gov-Lic

ν

I

Ο

Ν

V Ο

Χ

Χ

X X X

Χ

tJ

Li

ICG

CG

I Ν

At first, Government Licensing was only designed for non-nuclear governors. In later work by Yoshida (1993) and Kaye (1995), however, Government Licensing was found to be a condition on nuclear governors as well (see §155 for further detail).

629

4.2.4. No resyllabification Government Phonology outlaws any kind of resyllabification. That is, constituent structure is present in the lexicon, and no phonological operation of any kind can modify the lexical constituency. This principle can also be referred to with appeal to the notion of structure preservation (see §§ 10,17): if you are born in an Onset, you will live your life in this Onset position no matter what phonological processing may do to you. The same holds true for consonants in Rhymal Adjuncts and vowels in Nuclei. Initially, resyllabification was encoded by another notion known from syntax, i.e. the Projection Principle. The formulation used originally by Kaye et al. (1990:221) was "Governing relations are defined at the level of lexical representation and remain constant throughout a phonological derivation". Since it is Government that defines syllable structure, the prohibition of its modification means that resyllabification is outlawed. 397 There are two major consequences. One is that stem-final consonants

397

Note that resyllabification in this context is to be taken literally: an x-slot has been resyllabified iff it was dominated by constituent X at some derivational stage, but then has been reassociated to another constituent. In CVCV, there could not be any resyllabification anyway because all consonants belong to Onsets, and all vowels to Nuclei, in the lexicon as well as under any kind of phonological processing.

774

Appendix 4: a short guide to 1990 Government Phonology

belong to Onsets (cf. the foregoing section) and will not become Rhymal Adjuncts in case a consonant-initial suffix is added. For example, the Turkish root for "transfer" is [devir]. Its last consonant belongs to an Onset that is followed by an empty Nucleus, hence /devira/. In the Ablative case, the suffix -den is added, producing [devir-den]. Although the cluster [rd] meets all requirements for being interpreted as a Coda-Onset sequence, this is not its syllabic identity: the [r] belongs to an Onset and is followed by an empty Nucleus as before. This empty Nucleus separates it from the [d]: /devira-den/. In other words, the prohibition of resyllabification creates wordinternal empty Nuclei, which adds a second item on the list of vacuous vocalic positions that so far featured only word-final empty Nuclei (due to Coda Licensing). But the Projection Principle creates yet another kind of empty Nuclei, which is not only word-, but also morpheme-internal. This is when vowels alternate with zero. In the aforementioned Turkish root devir, the [i] alternates with zero in predictable fashion. The Accusative marker for example is an [-i], and the corresponding form sounds [devr-i]. Even though there is no phonetic trace of any vocalic articulation between the [v] and the [r], there must be an empty Nucleus, hence /dev0r-i/. This is because /v/ and Irl, although qualifying as a branching Onset, lexically pertain to two independent Onsets and therefore may not merge.

630

4.2.5. Proper Government, the Empty Category Principle and domains The foregoing section has mentioned vowel-zero alternations and the fact that the constituent structure of words remains stable even in case the alternating vowel is absent from the surface. That is, the Turkish word [devr-i] "transfer ACC" identifies as /dev0r-i/. The existence of empty Nuclei like the one that separates the [v] and the [r] in this example raises the question of how the proliferation of empty categories can be limited. Following the syntactic insights of the late 80s, a phonological Empty Category Principle was proposed to this effect. Table (450) reproduces Kaye's (1990a:314) wording (see also Kaye et al. 1990:219 and §§15,138).

Constituent structure 775 (450) Empty Category Principle - Kaye (1990a:314) a. a licensed empty Nucleus has no phonetic realisation b. an empty Nucleus is licensed if 1. it is properly governed or 2. if it is domain final in languages which license domain-final empty Nuclei The obvious confusion between Government and Licensing lain aside (see §136), Standard Government Phonology thus holds that empty Nuclei can only exist in two locations: word-internally if they are properly governed, and word-finally in languages that possess word-final consonants (which is the translation of "which license word-final empty Nuclei"). Before introducing Proper Government, the difference between "word-final" and "domain-final" needs to be explained. In this appendix, I use only the former because I do not intend to properly introduce the notion of domains. This is done on several occasions in the book, cf. §§24,249,349. Domains are the specific means by which Government Phonology expresses the influence of morphological information on phonology. This broader issue is discussed at length in Vol.2,III. In short, however, the idea of domains is the following. Word-final empty Nuclei are a special case of domain-final empty Nuclei. Phonological domains delineate those chunks of the linear string that remain fully untouched by higher level structure (morphological, syntactic, perhaps semantic), i.e. where the only rule is phonological. Phonological domains may or may not be morphologically complex: some morphological boundaries are "visible" to the phonology (i.e. they influence the regular course of phonology), while others are not (i.e. they do not influence the regular course of phonology). In case the boundary between two morphemes A and Β impacts phonology, A is a domain of its own, and its final Nucleus, if empty, may remain phonetically unexpressed while being word-internal: it will be domain-final. An typical example is the English word sixths where the final sequence [-ks0s] is obviously unsyllabifiable for any theory (note that nothing of the kind does or could occur word-internally). Everybody must admit that it can exist only because it represents in fact several morphemes: its morphological structure identifies as six-th-s. Since morphology impacts phonology here (the cluster could not exist on phonological grounds alone), three domains need to be assumed: /[[[sikso] Θ 0 ] se]/ (where "[ ]" indicate domain boundaries). Hence, there are three empty Nuclei, which can all happily be absent from the surface because of their domain-final status.

776

Appendix 4: a short guide to 1990 Government

Phonology

Let us now turn to the other clause of the Empty Category Principle (450), Proper Government. This lateral force describes vowel-zero alternations. It relates two Nuclei and hence instantiates an communication between two different constituents. Therefore, it is always head-final (cf. strict directionality (446e)). As in all other cases, the governor must not be empty: only phonetically expressed categories qualify for the headship of a lateral relation (cf. Government Licensing as exposed in §628d). The peculiarity of Proper Government, whose name and function have been borrowed from syntax on purpose, is to silence its target: Nuclei that are subjected to this lateral force may remain phonetically unexpressed. That is, the Turkish word [devr-i] "transfer ACC" that has been discussed earlier not only identifies as /dev0r-i/; the empty Nucleus is actually properly governed. Table (451) below shows the three relevant configurations, where Proper Government relations are made explicit. (451) representation of basic vowel-zero alternations in Standard Government Phonology Turkish "transfer" a. devr-i ACC b. devir NOM c. devir-den ABL PG PG, PG, PG

f

f

Λ

O N O N O N

O N O N O N

O N O N O N O N

x x x x x x

x x x x x x

x x x x x x x x

d e

ν

r

ι

d e ν

epenthesis

d e ν

d e

epenthesis

The Nucleus where [i] sometimes appears is lexically empty. It must acquire phonetic content under (451b-c) (via epenthesis) because the following Nucleus is empty. When the following Nucleus is filled as under (451a), however, it remains phonetically unexpressed. §19 provides further discussion. Proper Government is the only lateral relation in Standard Government Phonology which is not strictly local: it relates two skeletal slots that are not adjacent. Kaye (1990a:313) therefore talks about adjacency "on the projection" of the governor and the governee, i.e. Nuclei. This is but a way to escape strict locality (446d), which was made for governing relations between consonants: these are literally adjacent at the skeletal level. In case

Constituent structure

777

t w o Nuclei communicate, Government will always apply over at least one skeletal slot, the one of the intervening Onset. Hence, the "nuclear projection", where only Nuclei exist, needs to be invoked in order to make the t w o communicating constituents adjacent. Finally, Standard Government Phonology holds that no governing domain may intervene between the governor and the governee. Otherwise Proper Government breaks down. This is the translation of the observation that alternating vowels must be realised if they are followed by a consonant cluster (which is either TR, i.e. representing a domain of Constituent Government, or RT, i.e. representing a domain of Interconstituent Government). For example, the word secret [sakre] "secret" in Quebec French cannot be realised without schwa even though the following N u c l e u s is contentful. §§27,126. discusses this issue and further empirical material at length. M o r e detail regarding the phonological E C P and Proper Government can be retrieved f r o m the main text: both concepts are central all throughout the book, which opens on their introduction and further develops them as the presentation of C V C V unfolds.

4.2.6. A trademark of Government Phonology: empty Nuclei Empty Nuclei are probably the feature of Government Phonology that characterises this theory most saliently in the appreciation of a larger audience. Empty Nuclei are often viewed as a property that discredits the f r a m e w o r k : w h o could believe in empty ghosts? This issue is addressed on various occasions in the book, for example in chapter 11,7 (§387). In order to precisely locate empty Nuclei in the theory, it is useful to establish their typology. This is done under (452) below. (452)

exhaustive typology of empty Nuclei in Standard Government Phonology empty Nuclei occur a. after the final consonant of consonant-final morphemes (this subsumes the word-final location and the position before a consonant-initial suffix). b. in alternation sites (vowel-zero alternations) when the zero surfaces. c. in so-called bogus clusters, that is enclosed by clusters that do not qualify for either branching Onsets nor Coda-Onset status, such as [tl,dl] sequences (atlas, atlantic etc). The latter kind is discussed at greater length in §§180,215.

778

Appendix 4: a short guide to 1990 Government

Phonology

It was shown in the previous section how the existence of empty Nuclei is controlled by the ECP and Proper Government.

632

References

The references below contain Internet locations where electronic versions can be freely accessed. Of course, there may be electronic resources that I ignore, and the reader should be aware of the fact that indications may become outdated after the publication of the book. R e f e r e n c e to four particular web sites is recurrent. They therefore appear in abbreviation at the end of reference entries that are available online: R O A : http://roa.rutgers.edu T O B : http://www.unice.fr/dsl/tobias.htm U C L : http://www.phon.ucl.ac.uk/home/PUB/WPL/uclwpl.html SZI: http://seas3.elte.hu/szigetva/papers.html Afuta, Peggy 2002

Representation of syllabic consonants and statement of a final site in yiddish. Poster at the 10th Manchester Phonology Meeting, 2325 May 2002.

Alber, Birgit ms Maximizing First Positions. University of Marburg. ROA (#516). Allen, Sidney 1973 Accent and Rhythm. Cambridge: Cambridge University Press. Amimi, Abdeljabber, and Georges Bohas 1996 Les formes nominales [CeCC] et [CCeC] en arabe marocain ou la persistance des schèmes. Revue canadienne de linguistique 41: 128. Anderson, John 1985 Structural analogy and Dependency Phonology. Acta linguistica hafniensia 19: 5-44. 1986 Structural analogy and case grammar. Lingua 70: 79-129. 1987 Structural Analogy and Dependency Phonology. In Explorations in Dependency Phonology, John Anderson and Jacques Durand (eds.), 15-48. Dordrecht: Foris. 1992 Linguistic representation: Structural analogy and stratification. Berlin & New York: Mouton de Gruyter. Anderson, John, and Charles Jones 1974 Three theses concerning phonological representations. Journal of Linguistics 10: 1-26. 1977 Phonological Structure and the History of English. Amsterdam: North Holland.

780

References

Anderson, John, and Colin Ewen 1987 Principles of Dependency Phonology. Cambridge: Cambridge University Press. Anderson, Stephen 1974 The Organization of Phonology. New York: Academic Press. 1982 The analysis of French shwa: or, how to get something for nothing. Language 58: 534-573. 1985 Phonology in the Twentieth Century. Chicago: University of Chicago Press. Anglade, Joseph 1949 Grammaire élémentaire de l'ancien français. Paris: Colin. Angoujard, Jean-Pierre 1997 La Phonologie Déclarative. Langages 125: 35-54. Archangeli, Diana 1988 Aspects of underspecification theory. Phonology 5: 183-208. Armstrong, Lilias 1934 The Phonetic Structure of Somali. Mitteilungen des Seminars für Orientalische Sprachen zu Berlin 37.3: 116-161. Árnason, Kristján 1998 Vowel shortness in Icelandic. In Phonology and Morphology of the Germanic Languages, Wolfgang Kehrein and Richard Wiese (eds.), 3-25. Tübingen: Niemeyer. Aronoff, Mark, and Mary-Louise Kean (eds) 1980 Juncture. Saratoga: Anma Libri. Arumaa, Peeter 1964 Urslavische Grammatik. Band I: Einleitung, Lautlehre. Heidelberg: Carl Winter. 1985 Urslavische Grammatik. Band III: Formenlehre. Heidelberg: Carl Winter. Avery, Peter, and Keren Rice 1989 Segment structure and coronal underspecification. Phonology 6: 179-200. Bach, Adolf 1969 Deutsche Mundartforschung. Heidelberg: Carl Winter. 1970 Geschichte der Deutschen Sprache. 9th edition Heidelberg: Quelle & Meyer. Backley, Philip 1993 Coronal: the undesirable element. UCL Working Papers in Linguistics 5: 301-323. Bagemihl, Bruce 1991 Syllable Structure in Bella Coola. Linguistic Inquiry 22: 589-646.

References Bailey, Todd 1995

781

Nonmetrical Constraints on Stress. Ph.D dissertation, University of Minnesota. Searchable database at http://www.cf.ac.uk/psvch/ssd/index.html. Bankowski, Andrzej 2000 Etymologiczny Slownik jçzyka polskiego. Vols 1+2. Warszawa: PWN. Bargietówna, M. 1950 Grupy fonemów spótgtoskowych wspófczesnej polszczyzny kulturalnej. Biuletyn Polskiego Towarzystwa Jqzykoznawczego 10: 125. Barillot, Xavier 1997 Alternances voyelle-zéro et gémination consonantique en somali. DEA thesis. University Paris 7. 2002 Morphophonologie gabaritique et information consonantique latente en somali et dans les langues est-couchitiques. Ph.D dissertation, University Paris 7. Barillot, Xavier, and Philippe Ségéral forth Alternances voyelle-zéro en somali. To appear in Linguistique Africaine. Barkaoui, Ayesha 2000 Contribution à l'étude de la phonologie en arabe marocain : accent, schwa et syllabe. PhD. dissertation, University of Nancy. Basbell, Hans 1981 Metrical Theory and the French Foot. In Phonologica 1980, Wolfgang Dressler, Oskar Pfeiffer and John Rennison (eds.), 35-43. Innsbruck: Institut für Sprachwissenschaft Innsbruck. Bates, Dawn, and Barry Carlson 1992 Simple Syllables in Spokane Salish. Linguistic Inquiry 23: 653659. Bee, Pierre 1970 Manuel Pratique de Philologie Romane. 2 Vols. Paris: Picard. Beckman, Jill 1997 Positional Faithfulness, Positional Neutralisation and Shona Vowel Harmony. Phonology 14: 1-46. 1998 Positional Faithfulness. Ph.D dissertation, University of Massachusetts at Amherst. Published by Garland Press, New York 1999. ROA (#234). Bell, Alan 1978 Syllabic Consonants. In Universals of Human Language, Vol 2, Joseph Greenberg (ed.), 153-201. Stanford: Stanford University Press.

782

References

Bendjaballah, Sabrina 1998 La palatalisation en somali. Linguistique Africaine 21: 5-52. 1999 Trois figures de la structure interne des gabarits. Ph.D dissertation, University Paris 7. Bermúdez-Otero, Ricardo in press Markedness in phonology and in syntax: the problem of grounding. To appear in Lingua. Bertinetto, Pier Marco 1981 Strutture prosodiche dell'italiano. Firenze: Accademia della Crusca. Bethin, Christina 1984 Voicing assimilation in Polish. International Journal of Slavic Linguistics and Poetics 29: 17-32. 1992 Polish Syllables: the Role of Prosody in Phonology and Morphology. Columbus OH: Slavica. 1998 Slavic Prosody. Language change and phonological theory. Cambridge: Cambridge University Press. Bickmore, Lee 1995 Accounting for Compensatory Lengthening in the CV and moraic frameworks. In Frontiers of Phonology, Jacques Durand and Francis Katamba (eds.), 119-148. Londres & New York: Longman. Biedrzycki, Leszek 1978 Fonologia angielskich i polskich rezonantów. Warszawa: Polskie Wydawnictwo Naukowe. Blaho, Sylvia 2001 The representation of Slovak syllabic consonants in strict CV. The Odd Yearbook 6: 3-24. 2002 Representing length in autosegmental frameworks: the status of syllabic consonants. Poster at the 10th Manchester Phonology Meeting, 23-25 May 2002. 2004 Syllabic consonants in strict CV. MA thesis, Pazmany Peter U n i versity. Blevins, Juliette 1995 The Syllable in Phonological Theory. In The Handbook of Phonological Theory, Goldsmith John (ed.), 206-244. Oxford, Cambridge, Mass: Blackwell. Boas, Franz 1947 Kwakiutl grammar with a glossary of the suffixes. Transactions of the American Philosophical Society. New Series 37, Part 3. Boersma, Paul 1998 Functional phonology: formalizing the interactions between articulatory and perceptual drives. The Hague: Holland Academic Graphics.

References

783

Booij, Geert 1995 The Phonology of Dutch. Oxford: Clarendon Press. Bourriez, Edouard 1910 Eléments de Linguistique Romane. Paris: Klincksieck. 1926 Précis Historique de Phonétique Française. 6th edition Paris: Klincksieck. Brame, Michael 1972 On the Abstractness of Phonology: Maltese In Contributions to Generative Phonology, Michael Brame (ed.). Austin, Texas: University of Texas. Brandäo de Carvalho, Joaquim 1989a L'origine de la terminaison -äo du portugais: une approche phonétique nouvelle du problème. Zeitschrift für Romanische Philologie 105: 148-161. 1989b L'évolution des sonantes ibéro-romanes et la chute de -N-, -L- en gallaïco-portugais. Revue de Linguistique Romane 53: 159-188. 1993 De quoi sont faites les voyelles? Phonologie tridimensionnelle des particules et harmonie vocalique. In De natura sonorum. Essais de phonologie, Bernard Laks and Marc Plénat (eds.), 65-100. Paris: Presses Universitaires de Vincennes. TOB. 1994 What are vowels made of? The 'no-rule' approach and Particle Phonology. Studia Linguistica 48: 1-27. 1997 Primitives et naturalité. Langages 125: 14-34. 2002a De la syllabation en termes de contours CV. Habilitation thesis. Ecole des Hautes Etudes en Sciences Sociales, Paris. TOB. 2002b Formally-grounded phonology: from constraint-based theories to theory-based constraints. Studia Linguistica 56: 227-263. Brandäo de Carvalho, Joaquim, and Marc Klein 1996 A subsymbolic approach to phonological primitives. In Current Trends in Phonology. Models and Methods I, Jacques Durand and Bernard Laks (eds.), 97-121. Salford, Manchester: ESRI. Braune, Wilhelm 1874 Zur Kenntnis des Fränkischen und zur hochdeutschen Lautverschiebung. Beiträge zur Geschichte der deutschen Sprache und Literatur 1: 1-56. Braune, Wilhelm, and Ernst Ebbinghaus 1981 Gotische Grammatik. 19th edition Tübingen: Niemeyer. Braune, Wilhelm, and Hans Eggers 1987 Althochdeutsche Grammatik. 14th edition Tübingen: Niemeyer. Bretschneider, Anneliese 1934 Deutsche Mundartenkunde. Marburg: Elwert.

784

References

Broadbent, Judith 1991 Linking and Intrusive r in English. LJCL Working Papers in Linguistics 3: 281-302. 1996 The representation of coronal segments. Ph.D dissertation, University College London. 1999 A new approach to the representation of coronal segments. In Issues in phonological structure, S.J. Hannahs and Mike Davenport (eds.), 1-25. Amsterdam: Benjamins. Brockhaus, Wiebke 1992 Final devoicing: principles and parameters. Ph.D dissertation, University College London. 1994 Segmental representations without [R]? Paper presented at the Government Phonology Workshop, April 1994 Vienna. 1995a Final devoicing in the phonology of German. Tübingen: Niemeyer. 1995b Skeletal and suprasegmental structure within Government Phonology. In Frontiers of Phonology, Jacques Durand and Francis Katamba (eds.), 180-221. London & New York: Longman. Broselow, Ellen, Su-I Chen, and Marie Huffman 1997 Syllable weight: convergence of phonology and phonetics. Phonology 14: 47-82. Brückner, Aleksander 1927 Slownik etymologiczny jçzyka polskiego. Warszawa: Wiedza Powszechna. Burzio, Luigi 1994 Principles of English stress. Cambridge: Cambridge University Press. Butskhrikidze, Marika 2002 The Consonant Phonotactics of Georgian. Ph.D dissertation, University of Leiden. Cairns, Charles, and Mark Feinstein 1982 Markedness and the Theory of Syllable Structure. Linguistic Inquiry 13: 193-225. Calabrese, Andrea 1994 Sievers' Law in Gothic: A synchronic analysis with some notes on its diachronic development. The Linguistic Review 11: 149-194. Campbell, Lyle 1974 Phonological features: problems and proposals. Language 50: 5265. 1981 Generative phonology vs. Finnish phonology: retrospect and prospect. In Phonology in the 80's, Didier Goyvaerts (ed.), 147-182. Ghent: Story-Scientia. Carlton, Terence 1991 Introduction to the phonological history of the Slavic languages. Columbus, Ohio: S lavica.

References Carr, Philip 1993 2000

785

Phonology. London: Macmillan. Scientific Realism, Sociophonetic Variation, and Innate Endowments in Phonology. In Phonological Knowledge. Conceptual and Empirical Issues, Noel Burton-Roberts, Philip Carr and Gerard Docherty (eds.), 67-104. Oxford: Oxford University Press.

Casali, Roderic 1996 Resolving Hiatus. Ph.D dissertation, UCLA. 1997 Vowel elision in hiatus contexts: Which vowel goes? Language 73: 493-533. Cerny, Fr. 1897-1900 Studie o òeské kvantitè. Listy Filologické 24: 343-354, 421-431; 27: 17-22. Charette, Monik 1989 The Minimality Condition in Phonology. Journal of Linguistics 25: 159-189. 1990 Licence to govern. Phonology 7: 233-253. TOB. 1991 Conditions on Phonological Government. Cambridge: Cambridge University Press. 1992 Mongolian and Polish meet Government Licensing. SO AS Working Papers in Linguistics and Phonetics 2: 275-291. 1998 Empty and pseudo-empty categories. SOAS Working Papers in Linguistics and Phonetics 8: 167-176. 2003 Empty and pseudo-empty categories. In Living on the Edge. 28 papers in honour of Jonathan Kaye, Stefan Ploch (ed.), 465-479. Berlin, New York: Mouton de Gruyter. Charette, Monik, and Asli Göksel 1994 Vowel Harmony and Switching in Turkic languages. SOAS Working Papers in Linguistics and Phonetics. 4: 31 -52. Also in Kardela, Henryk, Bogdan Szymanek (eds), A Festschrift for Edmund Gussmann, 29-56. Lublin 1996: University Press of the Catholic University of Lublin. 1996 Licensing constraints and vowel harmony in Turkic languages. SOAS Working Papers in Linguistics and Phonetics 6: 1-25. Also in Cyran, Eugeniusz (ed), Structure and Interpretation. Studies in Phonology, 65-88. Lublin 1998: Folium. Chaussée, François de la 1974 Initiation à la phonétique historique de l'ancien français. Paris: Klincksieck. Chekayri, Abdellah, and Tobias Scheer in press The appearance of glides in Classical Arabic defective verbs. To appear in Folia Orientalia.

786

References

Chierchia, Gennaro 1982 An autosegmental theory of raddoppiamento. In NELS 12, J. Pustejovsky and P. Sells (eds.), 49-62. Amherst: GLSA. 1986 Length, syllabification and the phonological cycle in Italian. Journal of Italian Linguistics 8: 5-34. Chomsky, Noam 1965 Aspects of the Theory of Syntax. Cambridge, Mass.: MIT Press. 1970 Remarks on nominalisation. In Readings in English Transformational Grammar, R.Jacobs and P.Rosenbaum (eds.), 184-221. Walthamm, Mass.: Ginn. 1973 Conditions on Transformations. In A Festschrift for Morris Halle, Stephen Anderson and Paul Kiparsky (eds.), 232-286. New York: Holt, Rinehart & Winston. 2000 Minimalist inquiries: the framework. In Step by Step. Essays on Minimalist Syntax in Honor of Howard Lasnik, Roger Martin, David Michaels and Juan Uriagereka (eds.), 89-155. Cambridge, Mass.: MIT Press. 2001a Derivation by Phase. In Ken Hale: A Life in Language, Michael Kenstowicz (ed.), 1-52. Cambridge, Mass.: MIT Press. 2001b Beyond explanatory adequacy. MIT Occasional Papers in Linguistics 20: 1-28. Chomsky, Noam, and Morris Halle 1968 The Sound Pattern of English. Cambridge, Mass. : MIT Press. Chomsky, Noam, Marc Hauser, and Tecumseh Fitch 2002 The faculty of language: what is it, who has it, and how did it evolve ? Science 298: 1569-1579. Clements, George 1985 The geometry of phonological features. Phonology 2: 225-252. 1986 Compensatory Lengthening and Consonant Gemination in LuGanda. In Studies in Compensatory Lengtheing, Leo Wetzels and Engin Sezer (eds.), 37-77. Dordrecht: Foris. 1990 The role of the sonority cycle in core syllabification. In Papers in Laboratory Phonology I, John Kingston and Mary Beckmann (eds.), 283-333. Cambridge: Cambridge University Press. 1993 Lieu d'articulation des consonnes et des voyelles: une théorie unifiée. In L'architecture des représentations phonologiques, Bernard Laks and Annie Rialland (eds.), 101-145. Paris: CNRS Editions. 1997 Berber syllabification: derivations or constraints? In Derivations and Constraints in Phonology, Iggy Roca (ed.), 289-330. Oxford: Clarendon. 2001 Representational economy in constraint-based phonology. In Distinctive Feature Theory, Tracy Hall (ed.), 71-146. Berlin, New York: Mouton de Gruyter.

References

787

Clements, George, and Samuel Keyser 1983 CV Phonology. A Generative Theory of the Syllable. Cambridge, Mass.: MIT Press. Clements, George, and Elizabeth Hume 1995 The Internal Organization of Speech Sounds. In The Handbook of Phonological Theory, John Goldsmith (ed.), 245-306. Oxford: Blackwell. Cobb, Margaret 1997a Conditions on Nuclear Expressions in Phonology. Ph.D dissertation, SAOS, London. 1997b 'ATRVHeight' Harmony: the Manifestation of Complexity Effects in the Vowel Harmony of Natal Portuguese, Yoruba and Ogori. SOAS Working Papers in Linguistics and Phonetics 7: 169-192. 2003 Government Phonology and the vowel harminies of Natal Portuguese and Yoruba. In Living on the Edge. 28 papers in honour of Jonathan Kaye, Stefan Ploch (ed.), 223-242. Berlin, New York: Mouton de Gruyter. Coetsem, Frans van 1972 The Germanic consonant shift: Compensatory processes in language. Lingua 30: 302-15. Cole, Jennifer 1995 The Cylcle in Phonology. In The Handbook of Phonological Theory, John Goldsmith (ed.), 70-113. Oxford: Blackwell. Coleman, John 2002 Phonetic Representations in the Mental Lexicon. In Phonetics, Phonology and Cognition, Bernard Laks and Jacques Durand (eds.), 96-130. Oxford: Oxford University Press. Collinge, N. E. 1985 The Laws of Indo-European. Amsterdam, Philadelphia: Benjamins. Creissels, Denis 1989 Structures syllabiques du dialonke de Faleya. Mandenkan 18: 1-30. 1994 Aperçu sur les structures phonologiques des langues négroafricaines. Grenoble: Ellug. Csides, Csaba 2000 Government and Licensing: a CV analysis of consonant lenition. SOAS Working Papers in Linguistics and Phonetics 10: 41-80. Cyran, Eugeniusz 1994 Vocalic Elements in Phonology. A study in Munster Irish. Ph.D dissertation, University of Lublin. 2001 Parameters and scales in syllabic markedness: the right edge of the word in Malayalam. In Constraints and Preferences, Katarzyna Dziubalska-Kolaczyk (ed.), 1-42. Berlin & New York: Mouton de Gruyter.

788

References

Cyran, Eugeniusz (continued) 2003 a Complexity Scales and Licensing Strength in Phonology. Lublin: KUL. 2003b Branching onsets in Polish. In Living on the Edge. 28 papers in honour of Jonathan Kaye, Stefan Ploch (ed.), 303-320. Berlin, New York: Mouton de Gruyter. Cyran, Eugeniusz, and Edmund Gussmann 1998 Polish consonantal sequences: a phonological testing ground. In Structure and Interpretation, Eugeniusz Cyran (ed.), 127-138. Lublin: Pase. 1999 Consonant clusters and governing relations: Polish initial consonant sequences. In The syllable, Views and Facts, Harry van der Hulst and Nancy Ritter (eds.), 219-248. Berlin, New York: de Gruyter. Czaykowska-Higgins, Ewa 1988 Investigations into Polish morphology and phonology. Cambridge, Mass.: MIT Press. Czaykowska-Higgins, Ewa, and Marie Louise Willett 1997 Simple syllables in Nxa?amxcin. International Journal of American Linguistics 63: 385-411. Dalbera, Jean-Philippe 1994 Les parlers des Alpes-Maritimes. Etude comparative. Essai de reconstruction. London: AIEO. Davis, Stuart 1988 Syllable onsets as a factor in stress rules. Phonology 5: 1-19. 1990 Italian onset structure and the distribution of il and lo. Linguistics 28:43-55. Davis, Stuart, and Gina Torretta 1998 An Optimality Theoretic Account of Compensatory Lengthening and Geminate Throwback in Turkese. Vol I, Papers from the Main Sessions. In Proceedings of the 28th North East Linguistic Society, Pius Tamanji and Kiyomi Kusumoto (eds.), 111-125. Amherst: GLSA. De Chene, Brent, and Stephen Anderson 1979 Compensatory Lengthening. Language 55: 505-535. De Lacy, Paul 2002 The formal expression of markedness. Ph.D dissertation, University of Massachusetts. Dell, François 1973 Les règles et les sons. Paris: Hermann. 1976 Schwa précédé d'un groupe obstruante-liquide. Recherches Linguistiques de Vincennes 4: 75-111. 1995 Consonant clusters and phonological syllables in French. Lingua 95: 5-26.

References

789

Dell, François, and Mohamed Elmedlaoui 1985 Syllabic consonants and syllabification in Imdlawn Tashlhiyt Berber. Journal of African Languages and Linguistics 7: 105-130. 1988 Syllabic consonants in Berber: some new evidence. Journal of African Languages and Linguistics 10: 1-17. 2002 Syllables in Tashlhiyt Berber and in Moroccan Arabic. Dordrecht: Kluwer. Devine, Andrew, and Laurence Stephens 1976 The Function and Status of Boundaries in Phonology. In Linguistic Studies offered to Joseph Greenberg on the occasion of his sixtieth birthday, Alphonse Juilland (ed.), 285-312. Saratoga, Cal.: Anma Libri. Dienes, Péter, and Péter Szigetvári 1999 Repartitioning the skeleton: VC phonology. Ms, Eötvös Loránd University. SZI. Dinnsen, Daniel 1980 Phonological rules and phonetic explanation. Journal of Linguistics 16: 171-191. Dhigosz-Kurczabowa, Krystyna, and Stanislaw Dubisz 1993 Gramatyka historyczna jçzyka polskiego. Pochodzenie jqzyka polskiego, Fonetyka, Fonologia. Warszawa: Wydawnictwa Uniwersytetu Warszawskiego. Dogil, Grzegorz 1988 Phonological configurations: natural classes, sonority and syllabicity. In Features, Segmental Structure and Harmony Processes. Part 1, Harry van der Hulst and Norval Smith (eds.), 79-103. Dordrecht: Foris. Dokulil, MiloS 1962 Tvorenl slov ν éestinë. Dil 1: Teorie odvozování slov. Praha: Nakladatelství Òeskoslovenské Akademie Vëd. Doleáel, Lubomír 1967 Jména zdrobnélá. In Tvorení slov ν cestinë. Dil 2: Odvozování podstatnych jmen, MiloS Dokulil, FrantiSek DaneS and Jaroslav Kuchaf (eds.), 494-530. Praha: Nakladatelství Òeskoslovenské Akademie Vëd. Donegan, Patricia 1978 The Natural Phonology of Vowels. Ph.D dissertation, Ohio State University. Published by Garland Press, New York 1985. Donegan, Patricia, and David Stampe 1979 The study of natural phonology. In Current approaches to phonological theory, Daniel Dinnsen (ed.), 126-173. Bloomington: Indiana University Press.

790

References

Dresher, Elan 1981

Abstractness and Explanation in Phonology. In Explanation in Linguistics. The logical problem of language acquisition, Norbert Hornstein and David Lightfoot (eds.), 76-115. London, New York: Longman. 1985 Constraints on empty positions in tiered phonology. Cahiers Linguistiques d'Ottawa 14: 1-51. Dresher, Elan, and Aditi Lahiri 1991 The Germanic foot: Metrical coherence in Old English. Linguistic Inquiry 22: 251-286. Dressier, Wolfgang 1974 Diachronic Puzzles for Natural Phonology. In Papers from the parasession on Natural Phonology, A.Bruck, R.Fox and M.La Galy (eds.), 95-102. Chicago: Chicago Linguistic Society. 1981 Outlines of a Model of Morphonology. In Phonologica 1980, Uli Dressler, Oskar Pfeiffer and John Rennison (eds.), 113-122. Innsbruck: Institut fìir Sprachwissenschaft Innsbruck. 1984 Explaining Natural Phonology. Phonology Yearbook 1: 29-51. Drosdowski, Günther (ed.) 1984 Grammatik der deutschen Gegenwartssprache. 4th edition Mannheim: Dudenverlag. Dubina, Andrzej 2001 Lukashenka's native tongue defies Proper Government. Paper presented at Generative Linguistics in Poland 3, April 2001, Warsaw. Durand, Jacques 1988 Les phénomènes de nasalité en français du Midi: phonologie de dépendance et sous-spécification. Recherches Linguistiques de Vincennes 17: 29-54. Durand, Jacques, Catherine Slater, and Hilary Wise 1987 Observations on schwa in Southern French. Linguistics 25: 9831004. Durand, Jacques, and Bernard Laks 1996 Why Phonology is one. In Current trends in Phonology: Models and Methods, Jacques Durand and Bernard Laks (eds.), 3-13. Salford, Manchester: ESRI. Edgerton, Franklin 1934 Siever's Law and IE weak-grade vocalism. Language 10: 235-265. 1943 The Indo-European Semivowels. Language 19: 83-124. Elmedlaoui, Mohammed 1993 Gemination and Spirantization in Hebrew, Berber and Tigrinya: A 'Fortis-Lenis Module'Analysis. Linguistica Communicatio 5: 121176.

References

791

Encrevé, Pierre 1988 La liaison avec et sans enchaînement: phonologie tridimensionnelle et usages du français. Paris: Seuil. 1997 L'ancien et le nouveau. Quelques remarques sur la phonologie et son histoire. Langages 125: 100-123. English version published as Encrevé, Pierre 2000: The Old and the New: Some Remarks on Phonology and its History. Folia Linguistica 34, 57-85. Ernout, Alfred 1989 Morphologie historique du latin. 4th edition Paris: Klinksieck. Everett, Daniel 1988 On Metrical Constituent Structure in Pirahä. Natural Language and Linguistic Theory 6: 207-246. Everett, Daniel, and Keren Everett 1984 Syllable onsets and stress placement in Pirahä. Proceedings of the West Coast Conference on Formal Linguistics 3: 105-116. Ewen, Colin, and Harry van der Hulst 2001 The Phonological Structure of Words. Cambridge: Cambridge University Press. Farina, Donna 1991 Palatalization and jers in modern Russian phonology: an underspecification approach. Ph.D dissertation, University of Illinois. Farnetani, Edda, and S. Kori 1986 Effects of syllable and word structure on segmental durations in spoken Italian. Speech Communication 5: 17-34. Fava, Elisabetta, and Emanuela Magno Caldognetto 1976 Studio sperimentale delle caratteristiche elettroacustiche delle vocali toniche e atone in bisillabi italiani. In Studi di fonetica e fonologia, Raffaele Simone, Ugo Vignuzzi and Giulianella Ruggiero (eds.), 35-79. Roma: Bulzoni. Féry, Caroline 2003 Onsets and nonmoraic syllables in German. In The Syllable in Optimality Theory, Caroline Féry and Ruben van de Vijver (eds.), 213-237. Cambridge: Cambridge University Press. Feyerabend, Paul 1976 Wider den Methodenzwang. Frankfurt am Main 1986: Suhrkamp. 1994 Tuer le temps. Une autobiographie. Paris 1996: Seuil. Foley, James 1970 A systematic phonological interpretation of the Germanic consonants shifts. Language Sciences 9: 11-12. 1977 Foundations of theoretical phonology. Cambridge: CUP. Fouché, Pierre 1966-1973 Phonétique historique du français. Three vols. Paris: Klincksieck.

792

References

Fox, Anthony 2000 Prosodie Features and Prosodie Structure. The Phonology of Suprasegmentals. Oxford: Oxford University Press. Fudge, Erik 1969 Syllables. Journal of Linguistics 5: 253-286. Gamkrelidze, Thomas, and Vjaöeslav Ivanov 1995 Indo-European and Indo-Europeans. Berlin, New York: Mouton de Gruyter. Gebauer, Jan 1894-1898 Historická mluvnice jazyka ceského. 3 vols. Praha. 1970 Slovník starocesky. Dil I [A-J], Dil II [K-NJ. 2nd print Praha: Academia. Gess, Randall 1998a Compensatory lengthening and structure preservation revisited. Phonology 15: 353-366. 1998b Old French NoCoda effects from constraint interaction. Probus 10: 207-218. Giegerich, Heinz 1987 Zur Schwa-Epenthese im Standarddeutschen. Linguistische Berichte 112:449-469. 1992 English Phonology: An Introduction. Cambridge: Cambridge University Press. Goedemans, Rob 1996a Syllabe Weight and Prominence. In Stress Patterns of the World. Part 1: Background, Rob Goedemans, Harry van der Hulst and Ellis Visch (eds.), 115-163. The Hague: Holland Academic Graphics. 1996b An Optimality account of onset-sensitive stress in quantityinsensitive languages. The Linguistic Review 13: 33-47. 1998 Weightless segments. Ph.D dissertation, University of Leiden. Goldsmith, John 1976 Autosegmental Phonology. Ph.D dissertation, MIT. 1990 Autsegmental and Metrical Phonology. Oxford: Blackwell. Goossens, Jan 1974 Historische Phonologie des Niederländischen. Tübingen: Niemeyer. Gordon, Matthew 1999 Syllable Weight: Phonetics, Phonology and Typology. Ph.D dissertation, UCLA. in press Syllable weight. In Phonetic Bases for Phonological Markedness, Bruce Hayes, Robert Kirchner and Donca Steriade (eds.). Cambridge: Cambridge University Press. Downloadable at www.linguistics.ucsb.edu/faculty/gordon/pubs.html.

References

793

Goyvaerts, Didier 1981 Introduction. In Phonology in the 1980's, Didier Goyvaerts (ed.), 1 26. Ghent: Story-Scientia. Grammont, Maurice 1914 Traité Pratique de Prononciation Française. Paris: Delgrave. Green, Antony 2003 Extrasyllabic consonants and onset well-formedness. In The Syllable in Optimally Theory, Caroline Féry and Ruben van de Vijver (eds.), 238-253. Cambrdige: Cambridge University Press. Greenberg, Joseph 1978 Some Generalizations Concerning Initial and Final Clusters. In Universals of Human Language, Joseph Greenberg (ed.), 243-279. Stanford: Stanford University Press. Grzegorczykowa, Renata, Roman Laskowski, and Henryk Wróbel (eds.) 1999 Morfologia. 3rd edition Warszawa: PWN. Guerssel, Mohand, and Lowenstamm Jean ms The derivational morphology of the Classical Arabic Verb. Ms. University Paris 7 & UQAM. Gussenhoven, Carlos, and Haike Jacobs 1998 Understanding Phonology. London: Arnold. Gussmann, Edmund 1980a Studies in Abstract Phonology. Cambridge Mass.: MIT Press. Introduction to phonological analysis. Warszawa: PWN. 1980b Schwa and syllabic sonorants in a non-linear phonology of English. 1991a Anglica Wratislaviensia 17: 25-39. 1991b Polish syllable structure: a hypothesis and its problems. In Words are Physicians for an Ailing Mind, Maciej Grochowski and Daniel Weiss (eds.), 207-213. München: Sagner. 1992a Back to front: non-linear palatalization and vowels in Polish. In Phonnological Investigations, Jacek Fisiak and Stanislaw Puppel (eds.), 5-66. Amsterdam: Benjamins. 1992b Resyllabification and delinking: The case of Polish voicing. Linguistic Inquiry 23: 29-56. Govern or perish: sequences of empty nuclei in Polish. In A Fest1997 schrift for Jacek Fisiak on his 60th birthday, Raymond Hickey and Stanislaw Puppel (eds.), 1291-1300. Berlin: de Gruyter. Domains, relations, and the English agma. In Structure and Inter1998 pretation. Studies in Phonology, Eugeniusz Cyran (ed.), 101-126. Lublin: Folium. 2002 Phonology: Analysis and Theory. Cambridge: Cambridge University Press. 2003 Are there branching onsets in Modern Icelandic? In Living on the Edge. 28 papers in honour of Jonathan Kaye, Stefan Ploch (ed.), 321-337. Berlin, New York: Mouton de Gruyter.

794

References

Gussmann, Edmund, and Jonathan Kaye 1993 Polish notes from a Dubrovnik Café: I. The yers. SOAS Working Papers in Linguistics and Phonetics 3: 427-462. Gussmann, Edmund, and John Harris 1998 Final Codas: why the west was wrong. In Structure and Interpretation. Studies in Phonology, Eugeniusz Cyran (ed.), 139-162. Lublin: Folium. 2002 Word-final onsets. UCL Working Papers in Linguistics 14: 1-42. UCL. Hale, Mark, and Charles Reiss 2000a Substance Abuse and Dysfunctionalism: Current Trends in Phonology. Linguistic Inquiry 31: 157-169. 2000b Phonology as Cognition. In Phonological Knowledge. Conceptual and Empirical Issues, Noel Burton-Roberts, Philip Carr and Gerard Docherty (eds.), 161-184. Oxford: Oxford University Press. Hall, Tracy 1989 Lexical Phonology and the distribution of German [ç] and [x]. Phonology 6: 1-17. 1992 Syllable Structure and Syllable-Related Processes in German. Tübingen: Niemeyer. 2000 Phonologie. Eine Einführung. Berlin & New York: de Gruyter. 2001 Introduction: Phonological representations and phonetic implementation of distinctive features. In Distinctive Feature Theory, Tracy Hall (ed.), 1-40. Berlin, New York: 2001. Halle, Morris 1959 The Sound Pattern of Russian. The Hague: Mouton. 1977 Tenseness, Vowel Shift and the Phonology of the Back Vowels in Modern English. Linguistic Inquiry 8: 611-625. 1998 The Stress of English Words 1968-1998. Linguistic Inquiry 29: 539-568. Halle, Morris, and Jean-Roger Vergnaud 1980 Three-dimensional phonology. Journal of Linguistic Research 1: 83-105. 1987 An Essay on Stress. Cambridge, Mass.: MIT Press. Halle, Morris, and William Isardi 1995 Stress and metrical structure. In The Handbook of Phonological Theory, John Goldsmith (ed.), 403-443. Oxford: Blackwell. Hargus, Sharon, and Ellen Kaisse (eds.) 1993 Studies in Lexical Phonology. New York: Academic Press. Harris, James 1983 Syllable Structure and Stress in Spanish. Cambridge, Mass.: MIT Press.

References Harris, John 1990

795

Segmental complexity and phonological government. Phonology 7: 255-300. TOB. 1992 Licensing Inheritance. UCL Working Papers in Linguistics 4: 1-44. 1994a English sound structure. Oxford: Blackwell. 1994b Nonspecified coronality as weak licensing. Paper presented at the Government Phonology Workshop, April 1994 Vienna. 1997 Licensing Inheritance: an integrated theory of neutralisation. Phonology 14:315-370. 1999 Release the captive coda: the foot as a domain of phonetic interpretation. UCL Working Papers in Linguistics 11: 165-194. UCL. Harris, John, and Geoff Lindsey 1990 Phonetic interpretation in Generative Grammar. UCL Working Papers in Linguistics 2: 355-369. 1995 The elements of phonological representation. In Frontiers of Phonology, J.Durand and F.Katamba (eds.), 34-79. Harlow, Essex: Longman. Hausenblas, Karel 1967 Jména mlád'at. In Tvorení slov ν cestinë. Dil 2: Odvozování podstatnych jmen, MiloS Dokulil, FrantiSek DaneS and Jaroslav Kuchaï (eds.), 552-561. Praha: Nakladatelství íeskoslovenské Akademie Vëd. Havlík, Antonín 1889 Κ otázce jerové ν staré ceStine. Listy Filologické 16: 45-51, 106116, 248-258, 342-353, 436-445. Havlová, Eva, and Adolf Erhart (eds.) 1989-2002 Etymologicky slovník jazyka staroslovënského. 1st - 11th booklet Apoditi. Praha: Nakladatelství Òeskoslovenské Akademie Vëd/ Akademie Vêd Òeské Republiky. Havránek, Bohuslav, and Alois Jedliôka 1988 Ceská mluvnice. Praha: SPN. Havránek, Bohuslav (ed.) 1968 Starocesky Slovník. Uvodní stati, soupis pramenù a zkratek. Praha: Academia. Hayes, Bruce 1981 A Metrical Theory of Stress Rules. Ph.D dissertation, MIT. Published by Garland Press, New York 1985. 1986 Inalterability in CV Phonology. Language 62: 321-351. 1989a Compensatory Lengthening in Moraic Phonology. Linguistic Inquiry 20:253-306. 1989b The Prosodie Hierarchy in Meter. In Rhythm and Meter, Paul Kiparsky and G. Youmans (eds.), 201-260. Orlando, Florida: Academic Press.

796

References

Hayes, Bruce (continued) 1994 Weight of CVC can be Determined by Context. In Perspectives in Phonology, Jennifer Cole and Charles Kisseberth (eds.), 61-79. Stanford: CSLI. 1995 Metrical Stress Theory. Principles and Case Studies. Chicago, London: University of Chicago Press. 1996 Phonetically driven phonology: the role of Optimality Theory and Inductive Grounding. Ms, University of California, Los Angeles. Published in Functionalism and formalism in linguistics. Vol.1: General papers, Michael Darnell, Edith Moravcsik, Frederick Newmeyer, Michael Noonan and Kathleen Wheatley (eds.), 243285. Amsterdam: Benjamins. ROA (#158). Hellberg, Staffan 1978 Unnatural phonology. Journal of Linguistics 14: 157-177. Heo, Yong 1994 Empty categories and Korean Phonology. Ph.D dissertation, School of Oriental & African Studies, London. Hérault, Georges 1989 Les rections syllabiques en soninké. Linguistique Africaine 3: 4390. Hermans, Ben 2003 Russian Vowel Reduction as Diminution of Α-Complexity. Talk presented at Formal Description of Slavic Languages 5, Leipzig, November 1996. Hock, Hans Heinrich 1986 Compensatory lengthening: in defense of the concept 'mora'. Folia Linguistica 20: 431-460. 1991 Principles of Historical Linguistics. 2nd edition Berlin, New York: de Gruyter. Hockett, Charles 1955 A Manual of Phonology. Baltimore: Waverly Press. 1958 A Course in Modem Linguistics. New York: Macmillan. Holub, Josef, and FrantiSek Kopeöny 1952 Etymologicky slovník jazyka Ceského. Praha: Státní nakladatelství uCebnic. Holub, Josef, and Stanislav Lyer 1978 Strucny etymologicky slovník jazyka ceského. Praha: Státní Pedagogické Nakladatalství. Honeybone, Patrick 2001 Lenition inhibition in Liverpool English. English Language and Linguistics 5: 213-249. 2002 Another difference between prosody and melody: patterns in process inhibition. Paper presented at the 9th International Phonology Meeting, Vienna 1-3 November 2002.

References Hooper, Joan 1974

797

Rule morphologization in natural generative phonology. In Papers from the parasession on Natural Phonology, A. Bruck, R. Fox and M. La Galy (eds.), 160-170. Chicago: Chicago Linguistic Society. 1975 The archi-segment in Natural Generative Phonology. Language 51: 536-560. 1976 An Introduction to Natural Generative Phonology. New York: Academic Press. Hume, Elizabeth 1994 Front vowels, coronal consonants and their interaction in nonlinear phonology. New York & London: Garland. 1996 Coronal consonant, front vowel parallels in Maltese. Natural Language and Linguistic Theory 14: 163-203. Hristova, Vanya 1994 Nominal vowel/ zero alternations in Bulgarian and Russian. Ph.D dissertation, University of Delaware. Hualde, José 1990 Compensatory lengthening in Friulian. Probus 2: 31-46. Hubbard, Kathleen 1995 Transphonemic Compensatory Lengthening in Chichewa. In Papers from the 31st Regional Meeting of the Chicago Linguistic Society. Vol I, The Main Session, Audra Dainora, Rachel Hemphill, Barbara Luka, Barbara Need and Sheri Pargman (eds.), 193-204. Chicago: Chicago Linguistic Society. Hujer, Oldfich 1918 Κ nauce o poboCnych slabikách ν ôeStinë. Listy Filologické 45: 8487. 1955 Κ òeskému f . Listy Filologické 78: 255-258. Hülst, Harry van der 1988 The geometry of vocalic features. In Features, Segmental Structure and Harmony Processes. Part 2, Harry van der Hulst and Norval Smith (eds.), 77-126. Dordrecht: Foris. 1989 Atoms of segmental structure: components, gestures and dependency. Phonology 6: 253-284. 1994a An introduction to Radical CV Phonology. In SKY 1994: 1994 Yearbook of the Linguistic Association of Finland, S. Shore and M. Vilkuna (eds.), 23-56. Helsinki. 1994b Radical cv phonology: the locational gesture. UCL Working Papers in Linguistics 6: 439-478. 1995 Radical CV Phonology: the categorial gesture. In Frontiers of Phonology: Atoms, Structures, Derivations, Jacques Durand and Francis Katamba (eds.), 80-116. London & New York: Longman.

798

References

Hulst, Harry van der (continued) 1996 The segment-syllable connection. In Current Trends in Phonology. Models and Methods, Jacques Durand and Bernard Laks (eds.), 333-361. Salford, Manchester: ESRI. 1999 Features, segments and syllables in Radical CV Phonology. In Phonologica 1996, John Rennison and Klaus Kühnhammer (eds.). The Hague: Holland Academic Graphics. 2000 Modularity and Modality in Phonology. In Phonological Knowledge. Conceptual and Empirical Issues, Noel Burton-Roberts, Philip Carr and Gerard Docherty (eds.), 207-243. Oxford: Oxford University Press. Hulst, Harry van der, and Nancy Ritter 1999a Theories of the syllable. In The Syllable, Views and Facts, Harry van der Hulst and Nancy Ritter (eds.), 13-52. Berlin, New York: de Gruyter. 1999b Head-Driven Phonology. In The syllable, Views and Facts, Harry van der Hulst and Nancy Ritter (eds.), 113-167. Berlin, New York: de Gruyter. 2000 The SPE-heritage of Optimality Theory. The Linguistic Review 17: 259-289. Hyman, Larry 1970 How Concrete is Phonology? Language 46: 58-76. 1973 The feature [Grave] in phonological theory. Journal of Phonetics 1:329-337. 1985 A Theory of Phonological Weight. Dordrecht: Foris. 2004 Can Phonetics account for the Phonology of contour tones? Paper presented at the annnual LSA Meeting, Boston January 2004. Hyman, Larry, and Francis Katamba 1999 The syllable in Luganda phonology and morphology. In The syllable, Views and Facts, Harry van der Hulst and Nancy Ritter (eds.), 349-416. Berlin, New York: de Gruyter. Ingria, Robert 1980 Compensatory lengthening as a metrical phenomenon. Linguistic Inquiry 11:465-495. Isardi, William 1992 The computation of prosody. Ph.D dissertation, MIT. Itô, Junko 1986 Syllable Theory in Prosodie Phonology. Ph.D dissertation, University of Massachusetts at Amherst. Itô, Junko, and Armin Mester 1992 Weak Layering and Word Binarity. Ms, University of California at Santa Cruz. Updated version 2003 downloadable at http://people.ucsc.edu/~mester/PAPERS/weaklayering.pdf.

References

799

Itô, Junko, and Armin Mester (continued) 1993 Licensed segments and safe paths. Revue Canadienne de Linguistique 38: 198-213. Jackendoff, Ray 1992 Languages of the mind. Cambridge, Mass.: MIT Press. 1997 The Architecture of the Language Faculty. Cambridge, Massachusetts: MIT Press. 2002 Foundations of Language. Brain, Meaning, Grammar, Evolution. Oxford: Oxford University Press. Jacobs, Haike 1989 Nonlinear Studies in the Historical Phonology of French. Ph.D dissertation, University of Nijmegen. 1993 La palatalisation gallo-romane et la représentation des traits distinctifs. In Architecture des représentations phonologiques, Bernard Laks and Annie Rialland (eds.), 147-171. Paris: CNRS Editions. Jordan, Richard 1932 Handbuch der mittelenglischen Grammatik. Heidelberg: Winter. Joüon, Paul 1923 Grammaire de l'hébreu biblique. Rome: Institut Biblique Pontifical. Kager, René 1989 A Metrical Theory of Stress and Destressing in English and Dutch. Dordrecht: Foris. Kahn, Daniel 1976 Syllable-based generalizations in English phonology. Ph.D dissertation, MIT. Kaisse, Ellen 1985 Connected Speech. The interaction of Syntax and Phonology. London, New York: Academic Press. Kant, Immanuel 1781 Kritik der reinen Vernunft. Hamburg 2003 : Meiner. Kastler, Claude 1995 La langue tchèque. Grammaire tchèque pratique et raisonnée. Paris: Ophrys. Kavitskaya, Darya 2002 Compensatory Lengthening. Phonetics, Phonology, Diachrony. New York & London: Routledge. Kaye, Jonathan 1989 Phonology. A cognitive view. Hillsdale: Erlbaum. 1990a 'Coda' licensing. Phonology Yearbook 7: 301-330. TOB. 1990b Government in Phonology: the case of Moroccan Arabic. The Linguistic Review 6: 131-159. TOB.

800

References

Kaye, Jonathan (continued) 1992a Do you believe in magic? The story of s+C sequences. SO AS Working Papers in Linguistics and Phonetics 2: 293-313. Reprinted in A Festschrift for Edmund Gussmann, Henryk Kardela and Bogdan Szymanek (eds.), 155-176. Lublin 1996: Lublin University Press. TOB. 1992b On the interaction of theories of Lexical Phonology and theories of phonological phenomena. In Phonologica 1988, Uli Dressler, Hans Luschützky, Oskar Pfeiffer and John Rennison (eds.), 141-155. Cambridge: Cambridge University Press. TOB. 1994 Vowel harmony: the left hand of God. GLOW newsletter 32. 1995 Derivations and Interfaces. In Frontiers of Phonology, Jacques Durand and Francis Katamba (eds.), 289-332. London & New York: Longman. Also in SOAS Working Papers in Linguistics and Phonetics 3, 1993, 90-126. TOB. 2000 A Users' Guide to Government Phonology. Ms. University of Ulster. TOB. 2001 Working with licensing constraints. In Constraints and Preferences, Katarzyna Dziubalska-Kolaczyk (ed.), 251-268. Berlin & New York: Mouton de Gruyter. TOB. 2003 EGG Phonology Tool Kit. Handout from the 10th Central European Summer School in Generative Grammar, Lublin. Kaye, Jonathan, and Jean Lowenstamm 1984 De la syllabicité. In Forme Sonore du Langage, François Dell, Daniel Hirst and Jean-Roger Vergnaud (eds.), 123-159. Paris: Hermann. TOB. Kaye, Jonathan, Jean Lowenstamm, and Jean-Roger Vergnaud 1985 The internai structure of phonological representations: a theory of Charm and Government. Phonology Yearbook 2: 305-328. TOB. 1990 Constituent structure and government in phonology. Phonology 7: 193-231. TOB. Kean, Mary-Louise 1974 The Strict Cycle in Phonology. Linguistic Inquiry 5: 179-203. Keller, Rudolf 1961 German Dialects. Manchester: Manchester University Press. Kenstowicz, Michael 1994 Phonology in Generative Grammar. Oxford: Blackwell. Kenstowicz, Michael, and Charles Pyle 1973 On the phonological integrity of geminate clusters. In Issues in phonological theory, Michael Kenstowicz and Charles Kisseberth (eds.), 27-43. The Hague: Mouton. Kenstowicz, Michael, and Charles Kisseberth 1977 Topics in Phonological Theory. New York: Academic Press.

References

801

Kenstowicz, Michael, and Charles Kisseberth (continued) 1979 Generative Phonology. Description and Theory. San Diego: Academic Press. Kenstowicz, Michael, and Jerzy Rubach 1987 The Phonology of Syllabic Nuclei in Slovak. Language 63: 463497. Kijak, Artur 2003 Polish trapped sonorants and strict CV. Poster at the 11th Manchester Phonology Meeting, 22-24 May 2003. Kiparsky, Paul 1968 How abstract is phonology? Published 1973 in: Three Dimensions of Linguistic Theory, O. Fujimura (ed.), 5-56. Tokyo: TEC. 1973 Abstractness, Opacity and Global Rules. In Three Dimensions in Phonological Theory, O. Fujimura (ed.), 57-86. Tokyo: TEC. 1974 On the Evaluation Measure. In Papers from the parasession on Natural Phonology, A.Bruck, R.Fox and M.La Galy (eds.), 328337. Chicago: Chicago Linguistic Society. 1979 Metrical Structure Assignment is Cyclic. Linguistic Inquiry 10: 421-441. 1982a From Cyclic Phonology to Lexical Phonology. In The structure of phonological representations I, Harry van der Hulst and Norval Smith (eds.), 131-175. Dordrecht: Foris. 1982b Explanation in Phonology. Dordrecht: Foris. 1982c Lexical phonology and morphology. In Linguistics in the morning calm, In-Seok Yang (ed.), 3-91. Seoul: Hanshin. 1985 Some Consequences of Lexical Phonology. Phonology Yearbook 2: 85-138. 1991 Catalexis. Ms. Stanford University. Kiparsky, Valentin 1963 Russische Historische Grammatik. Band I: Die Entwicklung des Lautsystems. Heidelberg: Winter. Kirchner, Robert 2000 Geminate inalterability and lenition. Language 76: 509-545. Kleinhenz, Ursula 1991 Die Alternation von Schwa mit silbischen Sonoranten im Deutschen. MA thesis, University of Köln. Klemensiewicz, Zenon, Tadeusz Lehr-Splawmski, and Stanislaw Urbanczyk 1964 Gramatyka historyczna jqzyka polskiego. Warszawa: PWN. Kloeke, Wus van Lessen 1982 Deutsche Phonologie und Morphologie. Merkmale und Markiertheit. Tübingen: Niemeyer. Komárek, Miroslav 1962 Historická mluvnice Ceská. Dil I: Hláskosloví. Praha: SPN.

802

References

Kortlandt, Frederik 1979 On the history of the Slavic nasal vowels. Indogermanische Forschungen 84: 259-272. Koschmieder, Erwin 1958 Schwund und Vokalisation der Halbvokale im Ostslavischen. Die Welt der Sloven A.F.3: 124-137. Koutsoudas, Andreas 1980 The question of rule ordering: some common fallacies. Journal of Linguistics 16: 19-35. Kuhn, Thomas 1970 La structure des révolutions scientifiques. Paris 1983 : Flammarion. Kürti, Anna 1999 Post-tonic vowel-zero alternations in English. The Odd Yearbook 4: 23-70. Kurylowicz, Jerzy 1952 Uwagi o polskich grupach spólgloskowych [Remarks on Polish consonantal groups]. Biuletyn Polskiego Towarzystwa Jçzykoznawczego 11: 54-69. Lahrouchi, Mohamed 2001 Aspects morpho-phonologiques de la dérivation verbale en berbère (parler chleuh d'Agadir). Ph.D dissertation, University Paris 7. Lakatos, Imre 1984 Preuves et réfutations. Essai sur la logique de la découverte mathématique. Paris: Hermann. Lambdin, Thomas 1973 Introduction to Biblical Hebrew. London: Darton, Longman & Todd. Lamprecht, ArnoSt 1987 Praslovanstina. Brno: Univerzita Purkynë. Lamprecht, ArnoSt, DuSan § losar, and Jaroslav Bauer 1986 Historická mluvnice Cestiny. Praha: SPN. Lapointe, Steven, and Mark Feinstein 1982 The Role of Vowel Deletion and Epenthesis in the Assignment of Syllable Structure. In The structure of phonological representations, Part II, Harry van der Hulst and Norval Smith (eds.), 69-120. Dordrecht: Foris. Larsen, Bergeton Uffe 1994 Some aspects of vowel length and st0d in modern Danish. MA thesis, University Paris 7. 1998 Vowel length, Raddoppiamento Sintattico and the selection of the definite article in Italian. In Langues et Grammaire II-III, Phonologie, Patrick Sauzet (ed.), 87-102. Paris: Université Paris 8.

References

803

Laskowski, Roman 1975 Studia nad morfonologiq wspólczesnego jqzyka polskiego. Wroclaw, Warszawa, Kraków, Gdansk: Wydawnictwo Polskiej Akademii Nauk. Lass, Roger 1973 A case for making phonological rules state things that don't happen. Edinburgh Working Papers in Linguistics 3. 1984 Phonology. An introduction to basic concepts. Cambridge: Cambridge University Press. Leben, William, and Orrin Robinson 1977 Upside-down Phonology. Language 53: 1-20. Lehmann, Winfred 1993 Theoretical Bases of Indo-European Linguistics. London & New York: Routledge. Lehr-Splawinski, Tadeusz, and Zdzistaw Stieber 1957 Gramatyka Historyczna jqzyka czeskiego. Warszawa: PWN. Lerchner, Gotthard 1971 Zur II. Lautverschiebung im Rheinisch-Westmitteldeutschen; diachronische und diatopische Untersuchungen. Halle: Niemeyer. Levin, Juliette 1985 A Metrical Theory of Syllabicity. Ph.D dissertation, MIT. Liberman, Mark 1975 The Intonational System of English. Ph.D dissertation, MIT. Liberman, Mark, and Alan Prince 1977 On Stress and Linguistic Rhythm. Linguistic Inquiry 8: 249-336. Liewehr, Ferdinand 1933 Einführung in die historische Grammatik der tschechischen Sprache. 1. Teil: Lautlehre, Erste Lieferung. Brünn: Rohrer. Lightner, Theodore 1965 Segmental Phonology of Contemporary Standard Russian. Ph.D dissertation, MIT. 1972 Problems in the Theory of Phonology, vol.1. Edmonton, Alberta: Linguistic Research. 1978 Generative Phonology. In A Survey of Linguistic Science, William Orr Dingwall (ed.), 1-32. 2nd edition Stamford, Connecticut: Greylock. 1981 New explorations into Derivational Morphology. In Phonology in the 80's, Didier Goyvaerts (ed.), 93-99. Ghent: Story-Scientia. 1985 On universal rules. Lingvisticae Investigationes 9:419-420. Lindau, Mona 1985 The story of /r/. In Phonetic linguistics: essays in honor of Peter Ladefoged, Victoria Fromkin (ed.), 157-168. Orlando: Academic Press.

804

References

Lindeman, Fredrik Otto 1965 La loi de Sievers et le début du mot en Indo-Européen. Norsk Tidsskrift for Sprogvidenskap 20. 1987 Introduction to the 'Laryngeal Theory'. Oslo: Norwegian University Press. Lombardi, Linda 1999 Positional Faithfulness and voicing assimilation in Optimally Theory. Natural Language and Linguistic Theory 17: 267-302. 2001 Introduction. In Segmental phonology in Optimality Theory. Constraints and Representations, Linda Lombardi (ed.), 1-9. Cambridge: Cambridge University Press. Lowenstamm, Jean 1988 Another look at Ngizim syllable structure. Paper presented at the 19th Conference on African Linguistics, Boston April 1988. 1989 Prosodie Government. Langues Orientales Anciennes, Philologie et Linguistique 2: 221 -223. 1996 CV as the only syllable type. In Current trends in Phonology. Models and Methods, Jacques Durand and Bernard Laks (eds.), 419441. Salford, Manchester: ESRI. TOB. 1999 The beginning of the word. In Phonologica 1996, John Rennison and Klaus Kühnhammer (eds.), 153-166. La Hague: Holland Academic Graphics. TOB. 2002 A note on the segmental identification of templatic sites. Ms, University Paris 7. 2003 Remarks on mutae cum liquida and branching onsets. In Living on the Edge. 28 papers in honour of Jonathan Kaye, Stefan Ploch (ed.), 339-363. Berlin, New York: Mouton de Gruyter. Lowenstamm, Jean, and Jonathan Kaye 1986 Compensatory Lengthening in Tiberian Hebrew. In Studies in Compensatory Lengthening, Leo Wetzels and Engin Sezer (eds.), 97-132. Dordrecht: Foris. Macfarland, Talke, and Janet Pierrehumbert 1991 On ich-Laut, ach-Laut and Structure Preservation. Phonology 8: 171-180. Machek, Václav 1957 Etymologicky slovník jazyka ceského a slovenského. Praha: Nakladatelstvf Òeskoslovenské Akademie Vëd. Mann, Stuart 1957 Czech historical grammar. Revised edition Hamburg 1977: Buske. Mareä, FrantiSek Václav 1956 Vznik slovanského fonnologického systému a jeho vyvoj do konce období slovanské jazykové jednoty. Slavia 25: 443-495.

References

805

Mareä, FrantiSek Václav (continued) 1965 The Origin of the Slavic Phonological System and Its Development up to the End of Slavic Language Unity. Ann Arbor: Department of Slavic Languages and Literatures. Marotta, Giovanna 1984 Aspetti della struttura ritmico-temporale in italiano. Studi sulla durata vocalica. Pisa: ETS. 1993 Selezione dell'articolo e sillaba in italiano: un'interazione totale? Studi di grammatica italiana 15: 255-296. Martin, Bernhard 1939 Die Deutschen Mundarten. Leipzig: Quelle & Meyer. Martinet, André 1952 Langues à syllabes ouvertes: le cas du slave commun. Zeitschrift für Phonetik und allgemeine Sprachwissenschaft 6: 145-163. Mascaró, Joan 1976 Catalan Phonology and the Phonological Cycle. Ph.D dissertation, MIT. Mayrhofer, Manfred 1986 Indogermanische Grammatik, Band 1, Zweiter Halbband: Lautlehre. Heidelberg: Winter. McCarthy, John 1979 Formal problems in Semitic phonology and morphology. Ph.D dissertation, MIT. 1986 OCP Effects: Gemination and Antigemination. Linguistic Inquiry 17: 207-263. 2002 A Thematic Guide to Optimality Theory. Cambridge: Cambridge University Press. McCarthy, John, and Alan Prince 1986 Prosodie Morphology. Ms. Various (electronic) versions circulate, the latest (2001) is downloadable at ROA (#482). 1990 Foot and Word in Prosodie Morphology: The Arabic Broken Plural. Natural Language and Linguistic Theory 8: 209-283. 1993 Generalized Alignment. In Yearbook of Morphology 1993, Geert Booij and Jaap van Marie (eds.), 79-153. Dordrecht: Kluwer. Abridged version in McCarthy, John (ed.) 2004. Optimality Theory in Phonology, 451-463. Oxford: Blackwell. ROA (#7). 1994 The Emergence of the Unmarked: Optimality in Prosodie Morphology. In Proceedings of the North-East Linguistic Society 24, Merce Gonzalez (ed.), 333-379. Amherst: Graduate Linguistic Student Association. ROA (#13). McCawley, James 1968 The Phonological Component of a Grammar of Japanese. The Hague: Mouton.

806

References

McMahon, April 2000a Lexical Phonology and the history of English. Cambridge: Cambridge University Press. 2000b Change, Chance and Optimality. Oxford: Oxford University Press. 2003 Heads I win, tails you lose. Paper presented at the conference on Representations and Constraints, Toulouse 7-11 July 2003. Meillet, Antoine 1934 Le Slave Commun. 2nd edition Paris: Champion. 1937 Introduction à l'étude comparative des langues indo-européennes. 8th edition Paris: Hachette. Meinhold, Gottfried 1962 Formen und Bedingungen der Realisation des Endsilben-[schwa]. Biuletyn Fonograficzny 5: 117-144. Melvold, Janis 1989 Structure and stress in the phonology of Russian. Ph.D dissertation, MIT. Mester, Armin 1994 The quantitative trochee in Latin. Natural Language and Linguistic Theory 12: 1-61. Michaels, David 1991 Movement Rules in Phonology. In Certamen Phonologicum II, Pier Marco Bertinetto, Michael Kenstowicz and Michele Loporcaro (eds.), 65-80. Torino: Rosenberg & Sellier. 1992 Natural and unnatural phonology. In Phonologica 1988, Uli Dressler, Hans Luschützky, Oskar Pfeiffer and John Rennison (eds.), 207-214. Cambridge: Cambrige University Press. Mikkola, Jooseppi Julius 1913-50 Urslavische Grammatik. Teil 11913: Lautlehre, Vokalismus, Betonung. Teil II 1942: Konsonantismus. Teil III 1950: Formenlehre. Heidelberg: Winter. Milliken, Stuart 1988 Protosyllables: A theory of underlying syllable structure in nonlinear phonology. Ph.D dissertation, Cornell University. Milner, Jean-Claude 1989 Introduction à une science du langage. Paris: Seuil. Minkova, Donka 1982 The environment for open syllable lengthening in Middle English. Folia Linguistica Histórica 3: 29-58. Mithun, Marianne, and Hasan Basri 1986 The phonology of Selayarese. Oceanic Linguistics 25: 210-254. Mitzka, Walther 1943 Deutsche Mundarten. Heidelberg: Winter. Mohanan, Karuvannur 1982 Lexical Phonology. Ph.D dissertation, MIT.

References

807

Mohanan, Karuvannur (continued) 1986 The Theory of Lexical Phonology. Dordrecht: Reidel. Montreuil, Jean-Pierre 1999 The Romansch syllable. In The syllable, Views and Facts, Harry van der Hulst and Nancy Ritter (eds.), 527-550. Berlin, New York: de Gruyter. Morén, Bruce 2003 The Parallel Structures Model of Feature Geometry. The Working Papers of the Cornell Phonetics Laboratory 15: 194-270. Morin, Yves-Charles 1986 La loi de position ou de l'explication en phonologie historique. Revue Québécoise de Linguistique 15: 199-232. 1988 De l'ajustement du schwa en syllabe fermée dans la phonologie du français. In La phonologie du schwa français, Paul Verluyten (ed.), 133-202. Amsterdam & Philadelphia: Benjamins. 1994 Phonological interpretations of historical lengthening. In Phonologica 1992. Proceedings of the 7th International Phonology Meeting, Uli Dressler, Martin Prinzhorn and John Rennison (eds.), 135-155. Torino: Rosenberg & Sellier. Moulton, William 1947 Juncture in Modern Standard German. Language 23: 212-226. Murray, Robert 1988 Phonological Strength and Early Germanic Syllable Structure. München: Fink. Murray, Robert, and Theo Vennemann 1983 Sound change and syllable structure in Germanic phonology. Language 59: 514-528. Myers, Scott 1987 Vowel shortening in English. Natural Language and Linguistic Theory 5: 485-518. Nagórko, Alicja 1998 Zarys gramatyki polskiej. Warszawa: PWN. Nahtigal, Rajko 1961 Die Slavischen Sprachen. Abriß der vergleichenden Grammatik. Wiesbaden: Harroassowitz. Nespor, Marina, and Irene Vogel 1979 Clash Avoidance in Italian. Linguistic Inquiry 10: 467-482. Nespor, Marina, and Irene Vogel 1986 Prosodie Phonology. Dordrecht: Foris. Newmeyer, Frederick 1998 The irrelevance of typology for linguistic theory. Syntaxis 1: 161197.

808

References

Ni Chiosáin, Márie 1990 Syllabification, Compensatory Lengthening and Epenthesis in Irish. UMASS Occasional Papers 14: 67-90. Niedermann, Max 1991 Précis de phonétique historique du latin. 5e édition Paris: Klincksieck. Nikiema, Emmanuel 1989 Gouvernement Propre et licenciement en phonologie: le cas du Tangale. Langues Orientales Anciennes, Philologie et Linguistique 2:225-251. Noske, Manuela 1997 Feature Spreading as dealignment: the distribution of [ç] and [x] in German. Phonology 14: 221-234. Noske, Roland 1993 A theory of syllabification and segmental alternations. With studies on the phonology of French, German, Tonkawa and Yawelmani. Tübingen: Niemeyer. Nykiel-Herbert 1985 The Vowel-Zero Alternation in Polish Prefixes. In PhonoMorphology. Studies in the Interaction of Phonology and Morphology, Edmund Gussmann (ed.), 113-130. Lublin: Katolicki Universytet Lubelski. Odden, David 1978 Further evidence for the feature [GRAVE], Linguistic Inquiry 9: 141-144. Oostendorp, Marc van, and Jeroen van de Weijer forth Phonological alphabets and the structure of the segment. In The internal organisation of the phonological segment, Marc van Oostendorp and Jeroen van de Weijer (eds.). Berlin: Mouton de Gruyter. Orwin, Martin 1993 Phonation in Somali Phonology. In Anthropologie somalienne: Actes du Uè Colloque des Etudes Somaliennes, Mohamed Cabdi (ed.), 251-257. Besançon: Université de Besançon. Ostaszewska, Danuta, and Jolanta Tambor 2000 Fonetyka i fonologia wspólczesnego jqzyka polskiego. Warszawa: PWN. Pagliano, Claudine 1999 Formation des adverbes en -ment en français. DEA thesis. University of Nice. 2003 L'épenthèse consonantique en français. Ce que la syntaxe, la sémantique et la morphologie peuvent faire à la phonologie. Ph.D dissertation, University of Nice. TOB.

References

809

Palková, Zdena 1997 Fonetika a fonologie Ceätiny. Praha: Univerzita Karlova. Panzer, Baldur 1991 Die Slavischen Sprachen in Gegenwart und Geschichte. Sprachstrukturen und Verwandtschaft. Frankfurt am Main: Peter Lang. Paradis, Carole, and Jean-François Prunet 1989 On coronal transparency. Phonology 6:317-348. Paradis, Carole, and Jean-François Prunet (eds.) 1991 The Special Status of Coronals: Internal and External Evidence. San Diego: Academic Press. Paul, Hermann, Peter Wiehl, and Siegfried Grosse 1989 Mittelhochdeutsche Grammatik. 23th edition Tübingen: Niemeyer. Pawelec, Przemyslaw 1989 Cyclic phonology and the inventory of morpheme-initial consonant clusters in Polish and English. Anglica Wratislaviensia 16: 35-63. Peciar, Stefan 1941 Vznik praslov. slabiònych likvíd 1, r. Linguistica Slovaca 3: 49-53. Pedersen, Holger 1905 Die nasalpräsentia und der slavische akzent. Zeitschrift für vergleichende Sprachforschung 38: 297-425. Penzl, Herbert 1969 Geschichtliche Deutsche Lautlehre. München: Hueber. 1975 Vom Urgermanischen zum Neuhochdeutschen. Eine historische Phonologie. Berlin: Schmidt. 1984 Zur relativen Chronologie der hochdeutschen Konsonantenverschiebungen. In Dialectology, linguistics, literature. Festschrift for Carroll E. Reed, Wolfgang Moelleken (ed.), 208-221. Göppingen: Kümmerle. Perlmutter, David 1995 Phonological Quantity and Multiple Association. In The Handbook of Phonological Theory, John Goldsmith (ed.). Oxford, Cambridge, Mass.: Blackwell. Piggott, Glyne 1991 Apocope and the licensing of empty-headed syllables. The Linguistic Review 8: 287-318. 1998 Foot form and the parsing of weightless syllables. Chicago Linguistics Society 34: 315-332. 1999 At the right edge of words. The Linguistic Review 16: 143-185. 2003 The phonotactics of a "Prince" language: a case study. In Living on the Edge. 28 Papers in Honour of Jonathan Kaye, Stefan Ploch (ed.), 401-425. Berlin, New York: Mouton de Gruyter.

810

References

Piotrowski, Marek 1992a Polish yers in non-linear phonology. In Phonologica 1988, Uli Dressler, Hans Luschützky, Oskar Pfeiffer and John Rennison (eds.), 215-227. Cambridge: Cambridge University Press. 1992b Polish yers and extrasyllabicity: an autosegmental account. In Phonological Investigations, Jacek Fisiak and Stanislaw Puppel (eds.), 67-108. Amsterdam: Benjamins. Piotrowski, Marek, Iggy Roca, and Andrew Spencer 1992 Polish yers and lexical syllabicity. The Linguistic Review 9: 27-67. Ploch, Stefan 1998 Non-switch harmony in Yawelmani (and Turkish and Sakha). SOAS Working Papers in Linguistics and Phonetics 8: 209-238. 1999 Nasals on my mind. Ph.D dissertation, SOAS, London. 2003 Can phonological 'nasality' be derived from phonetic nasality? In The Phonological Spectrum. Vol I: Segmental Structure, Harry van der Hulst, Vincent van Heuven and Jeroen van de Weijer (eds.), 73116. Amsterdam & Philadelphia: Benjamins. Polgárdi, Krisztina 1998 Vowel Harmony. Ph.D dissertation, University of Leiden. 1999 Constraint ranking, Government Licensing and the fate of final empty nuclei. In Phonologica 1996, John Rennison and Klaus Kühnhammer (eds.), 167-182. The Hague: Holland Academic Graphics. 2002 Hungarian Superheavy Syllables and the Strict CV Approach. In Approaches to Hungarian 8, István Kenesei and Péter Siptár (eds.). Budapest: Akadémiai Kiadó. 2003 Hungarian as a strict CV language. In The Phonological Spectrum. Vol II: Suprasegmental Structure, Harry van der Hulst, Vincent van Heuven and Jeroen van de Weijer (eds.), 59-79. Amsterdam & Philadelphia: Benjamins. Pope, Mildred 1934 From Latin to Modern French. Manchester: Manchester University Press. Popper, Karl 1959 The logic of scientific discovery. French edition 1973 Paris: Payot. 1962 La société ouverte et ses ennemis. Tome 2: Hegel et Marx. Paris 1979: Seuil. Prince, Alan 1983 Relating to the Grid. Linguistic Inquiry 14: 19-100. Prince, Alan, and Paul Smolensky 1993 Optimality Theory. Constraint Interaction in Generative Grammar. Ms, Rutgers University, University of Colorado. ROA (#537).

References

811

Prokosch, Eduard 1939 A comparative Germanic Grammar. Philadelphia: Linguistic Society of America. Pulleyblank, Douglas 1988 Underspecification, the feature hierarchy and Tiv vowels. Phonology 5: 299-326. 1995 Feature Geometry and underspecification. In Frontiers of Phonology, Jacques Durand and Francis Katamba (eds.), 3-33. London & New York: Longman. Reis, Marga 1974 Lauttheorie und Lautgeschichte. Untersuchungen am Beispiel der Dehnungs- und Kürzungsvorgänge im Deutschen. München: Fink. Rejzek, Jiff 2001 Ce.sky etymologicky slovník. Voznice: Leda. Rennison, John 1978 Bidialektale Phonologie: Die Kompetenz eines Salzburger Sprechers. Ph.D dissertation, University of Siazburg. 1990 On the elements of phonological representations: the evidence from vowel systems and vowel processes. Folia Linguistica 24: 175-244. 1994 Syllables, variable vowels, and empty categories. In Phonologica 1992. Proceedings of the 7th International Phonology Meeting, Uli Dressler, Martin Prinzhorn and John Rennison (eds.), 205-216. Torino: Rosenberg & Sellien 1998 Contour segments without subsegmentai structures. In Structure and Interpretation. Studies in Phonology, Eugeniusz Cyran (ed.), 227-245. Lublin: Folium. 1999a Can there be empty phonological elements? On empty heads and empty operators. In Phonologica 1996, John Rennison and Klaus Kühnhammer (eds.), 183-193. The Hague: Holland Academic Graphics. 1999b Syllables in Western Koromfe. In The syllable, Views and Facts, Harry van der Hulst and Nancy Ritter (eds.), 311-347. Berlin, New York: de Gruyter. Rennison, John, and Friedrich Neubarth 2003 An x-bar theory of Government Phonology. In Living on the Edge. 28 papers in honour of Jonathan Kaye, Stefan Ploch (ed.), 95-130. Berlin, New York: Mouton de Gruyter. Older version downloadable at www.univie.ac.at/linguistics/gp/papers/index.htm. Repetti, Lori 1991 A Moraic Anaylsis of Raddoppiamento Fonosintattico. Rivista di Linguistica 3: 307-330.

812

References

Rice, Keren 1992

On deriving sonority: a structural account of sonority relationships. Phonology 9: 61-99. 1994 Peripheral in Consonants. Canadian Journal of Linguistics 39: 191-216. 1996 Default variability: the coronal-velar relationship. Natural Language and Linguistic Theory 14: 493-543. 1999a Featural markedness in phonology: Variation. Part 1. Glot International 4.7: 3-6. 1999b Featural markedness in phonology: Variation. Part 2. Glot Internationaux·. 3-7. 2003 On the syllabification of right-edge consonants - evidence from Ahtna (Athapaskan). In Living on the Edge. 28 papers in honour of Jonathan Kaye, Stefan Ploch (ed.), 427-448. Berlin, New York: Mouton de Gruyter. Riemsdijk, Henk van 1982 Locality Principles in Syntax and Phonology. In Linguistics in the morning calm, In-Seok Yang (ed.), 693-708. Seoul: Hanshin. Ritter, Nancy 1995 The role of Universal Grammar in Phonology: a Government Phonology approach to Hungarian. Ph.D. dissertation, New York University. 1997 Headedness as a means of encoding stricture. In Phonology in progress - Progress in phonology, Geert Booij and Jeroen van de Weijer (eds.), 333-365. Leiden: Holland Academic Graphics. 1998 The unifying effects of Proper Government. In Papers from the Amsterdam Conference. Approaches to Hungarian 6, Casper de Groot and István Kenesei (eds.), 43-60. Szeged: JATE. ms Georgian consonant clusters: the complexity is in the structure, not in the melody. Ms, University of Connecticut. Rizzi, Luigi 1990 Relativized Minimality. Linguistic Inquiry Monograph 16. Cambridge, Mass.: MIT Press. Rizzolo, Olivier 2002 Du leurre phonétique des voyelles moyennes en français et du divorce entre Licenciement et Licenciement pour gouverner. Ph.D dissertation, University of Nice. Roca, Iggy 1992 Constraining Extrametricality. In Phonologica 1988, Uli Dressler, Hans Luschützky, John Rennison and Oskar Pfeiffer (eds.), 239248. Cambridge: Cambridge University Press. 1994 Generative Phonology. London: Routledge. Rospond, Stanislaw 1979 Gramatyka historyczna jçzyka polskiego. Warszawa: ΡWN.

References

813

Rowicka, Grazyna 1999a On Ghost vowels. A Strict CV Approach. Ph.D dissertation, Leiden University. 1999b On trochaic Proper Government. In Phonologica 1996, John Rennison and Klaus Klihnhammer (eds.), 273-288. The Hague: Holland Academic Graphics. 1999c Prosodie optimality and prefixation in Polish. In The prosodymorphology interface, René Kager, Harry van der Hulst and W. Zonneveld (eds.), 367-389. Cambridge: Cambridge University Press. 2001 How Far Can You Go? Vowelless Roots and Clusters in Polish. In Current Issues in Formal Slavic Linguistics, Gerhild Zybatow, Uwe Junghans, Grit Mehlhorn and Luka Szucsich (eds.), 15-24. Frankfurt am Main: Lang. 2003 Irl syllabicity: Polish versus Bulgarian and Serbo-Croatian. In Living on the Edge. 28 papers in honour of Jonathan Kaye, Stefan Ploch (ed.), 511-526. Berlin, New York: Mouton de Gruyter. Rubach, Jerzy 1974 Syllabic consonants in Polish. Journal of Phonetics 2: 109-116. 1977 Changes of consonants in English and Polish. A generative account. Wroclaw, Warszawa, Kraków, Gdansk: Polska Akademia Nauk. 1984 Cyclic and Lexical Phonology: The Structure of Polish. Dordrecht: Foris. 1986 Abstract vowels in three dimensional phonology: the yers. The Linguistic Review 5: 247-280. 1990 Final Devoicing and Cyclic Syllabification in German. Linguistic Inquiry 21: 79-94. 1993 The Lexical Phonology of Slovak. Oxford: Clarendon Press. 1996 Nonsyllabic Analysis of Voice Assimilation in Polish. Linguistic Inquiry 27: 69-110. 1997a Extrasyllabic Consonants in Polish: Derivational Optimality Theory. In Derivations and Constraints in Phonology, Iggy Roca (ed.), 551-581. Oxford: Clarendon. 1997b Polish Voice Assimilation in Optimality Theory. Rivista di Linguistica 9: 291-342. 1999 The syllable in phonological analysis. Rivista di Linguistica 11: 273-314. Rubach, Jerzy, and Geert Booij 1984 Morphological and prosodie domains in Lexical Phonology. Phonology 1: 1-27. 1987 Postcyclic versus Postlexical Rules in Lexical Phonology. Linguistic Inquiry 18: 1-44.

814

References

Rubach, Jerzy, and Geert Booij (continued) 1990a Edge of constituent effects in Polish. Natural Language and Linguistic Theory 8: 427-463. 1990b Syllable structure assignment in Polish. Phonology 7: 121-158. Russ, Charles (ed.) 1990 The Dialects of Modern German. London: Routledge. Saeed, John Ibrahim 1998 Somali. Dublin: John Benjamins. Sagey, Elizabeth 1986 The representation of features and relations in nonlinear phonology. Ph.D dissertation, MIT. Saussure, Ferdinand de 1879 Mémoire sur le système primitif des voyelles dans les langues indoeuropéennes. Leipzig: Teubner. 1915 Cours de linguistique générale. Paris 1972: Payot. Sauzet, Patrick 1993 Attenance, gouvernement et mouvement en phonologie. Les constituants dans la phonologie et la morphologie de l'occitan. Ph.D dissertation, University of Montpellier. 1996 Ordre des mots, ordre dans les mots. Langue française 111: 10-37. 1999 Linéarité et consonnes latentes. Recherches Linguistiques de Vincennes 28: 59-86. Sawicka, Irena 1974 Struktura grup spôlgloskowych w jqzykach slowianskich. Wroclaw, Warszawa, Kraków, Gdansk: Wydawnictwo Polskiej Akademii Nauk. Schane, Sanford 1968 French Phonology and Morphology. Cambridge, Mass.: MIT Press. 1974 How abstract is abstract? In Papers from the parasession on Natural Phonology, A.Bruck, R.Fox and M.La Galy (eds.), 297-317. Chicago: Chicago Linguistic Society. 1984 The fundamentals of particle phonology. Phonology Yearbook 1: 129-155. 1995 Diphthongisation in particle phonology. In The Handbook of phonological theory, John Goldsmith (ed.), 586-608. Oxford: Blackwell. Scheer, Tobias 1996 Une théorie de l'interaction directe entre consonnes. Ph.D dissertation, University Paris 7. 1997 Vowel-zero alternations and their support for a theory of consonantal interaction. In Certamen Phonologicum III, P.M.Bertinetto, L. Gaeta, G. Jetchev and D. Michaels (eds.), 67-88. Torino: Rosenberg & Sellier. TOB.

References

815

Scheer, Tobias (continued) 1998a A unified model of Proper Government. The Linguistic Review 15: 41-67. TOB. 1998b Governing domains are head-final. In Structure and Interpretation. Studies in Phonology, Eugeniusz Cyran (ed.), 261-285. Lublin: Folium. TOB. 1998c La structure interne des consonnes. In Langues et Grammaire IIIII, Phonologie, Patrick Sauzet (ed.), 140-172. Paris: Université Paris 8. TOB. 1999a A theory of consonantal interaction. Folia Linguistica 32: 201-237. TOB. 1999b Aspects de l'alternance schwa-zéro à la lumière de "CVCV". Recherches Linguistiques de Vincennes 28: 87-114. TOB. 1999c On constraints vs. non-circular approaches to word-initial clusters. In Phonologica 1996, John Rennison and Klaus Kllhnhammer (eds.), 289-304. The Hague: Holland Academic Graphics. TOB. 2000a L'immunité de schwa en début de mot. Langue Française 126: 113-126. TOB. 2000b De la Localité, de la Morphologie et de la Phonologie en Phonologie. Habilitation thesis, University of Nice. TOB. 2001a The Rhythmic Law in Czech: Vowel-final Prefixes. In Current Issues in Formal Slavic Linguistics, Gerhild Zybatow, Uwe Junghans, Grit Mehlhorn and Luka Szucsich (eds.), 37-48. Frankfurt am Main: Lang. TOB. 2001 b Cestina ve WarSawë. Course handout. University of Warsaw. TOB. 2001c The key to Czech vowel length (Arabic rule in Middle Europe). Paper presented at Formal Description of Slavic Languages 4, 2730 November 2001, Potsdam. Handout TOB. 200Id A propos de la vie des yers en slave et en français. Travaux du Cercle de Linguistique de Nice 20: 143-230. French and English versions at TOB. 2003a On spirantisation and affricates. In Living on the Edge. 28 papers in honour of Jonathan Kaye, Stefan Ploch (ed.), 283-301. Berlin, New York: Mouton de Gruyter. 2003b The Key to Czech Vowel Length: Templates. In Investigations into Formal Slavic Linguistics, Petr Kosta, Joanna Blaszczak, Jens Frasek, Ljudmila Geist and Marzena Zygis (eds.), 97-118. Frankfurt am Main: Lang. TOB. 2003c Structure and Process: Computation is not King. Talk presented at the conference From representations to constraints and from constraints to representations, Toulouse 9-11 July 2003. Handout TOB. forth a What final empty Nuclei are good for. In Phonologica 2002, John Rennison, Friedrich Neubarth and Markus Pöchtrager (eds.).

816

References

Scheer, Tobias (continued) forth b How yers made Lightner, Rubach, Gussmann, Spencer & Co invent CVCV. In Phonology in Poland, Piotr Banski and Beata Lukaszewicz (eds.). München: Lincom. TOB. in press a How minimal is phonological change? To appear in Folia Linguistica Histórica. in press b O samohláskové délce pfi derivaci ν ôeStinë. In Cestina - univerzália a speciflka 5, Zdeftka Hladká and Petr Karlík (eds.). Praha: Lidové noviny. Scheer, Tobias, and Philippe Ségéral 2001 Les séquences consonne + yod en gallo-roman. Recherches Linguistiques de Vincennes 30: 87-120. TOB. Scheer, Tobias, and Péter Szigetvári 2002 Unified representations for the syllable and stress. Paper presented at the 10th Manchester Phonology Meeting, Manchester 23-25 May 2002. Handout at TOB. Schein, Barry, and Donca Steriade 1986 On geminates. Linguistic Inquiry 17: 691-744. Schenker, Alexander 1995 The Dawn of Slavic. An Introduction to Slavic Philology. New Haven & London: Yale University Press. Schindler, Jochem 1977 Notizen zum Sieverschen Gesetz. Die Sprache 23: 56-65. Schwarz, Ernst 1950 Die Deutschen Mundarten. Göttingen: Vandenhoeck & Ruprecht. Sebregts, Koen 2004 The representation of complexity and Dutch /r/ variation. Talk presented at the 12th Manchester Phonology Meeting, May 2004. Seebold, Elmar 1972 Das System der indogermanischen. Halbvokale. Heidelberg: Winter. Ségéral, Philippe 1995 Une théorie généralisée de l'apophonie. Ph.D dissertation, University Paris 7. 1996 L'apophonie en ge'ez. In Studies in Afroasiatic Grammar, Jacqueline Lecarme, Jean Lowenstamm and Ur Shlonsky (eds.), 360-391. La Hague: Holland Academic Graphics. in press Morpho-phonological processes in Somali 3rd conjugation. In Studies in Afro-Asiatic Grammar III, Denise Perrett (ed.). Ségéral, Philippe, and Tobias Scheer 1999 Is the Coda Mirror a phonological object? Paper presented at the First Annual Meeting of the GDR 1954 "Phonologie" on Lenition and Fortition, Nice 24-25 June 1999. Handout at TOB.

References

817

Ségéral, Philippe, and Tobias Scheer (continued) 2001a La Coda-Miroir. Bulletin de la Société de Linguistique de Paris 96: 107-152. TOB (also an older English version). 2001b Abstractness in phonology: the case of virtual geminates. In Constraints and Preferences, Katarzyna Dziubalska-Kolaczyk (ed.), 311-337. Berlin & New York: Mouton de Gruyter. TOB. Seigneur-Froli, Delphine 2001 De la lénition des "Codas" initiales en grec. DEA thesis, University ofNice. 2003 Diachronic consonant lenition & exotic word-initial clusters in Greek: a unified account. In Studies in Greek Linguistics. Proceedings of the 23nd annual meeting of the department of linguistics, M. Stavrou-Sifaki and A. Fliatouras (eds.), 345-357. Thessaloniki: University of Thessaloniki. in press Greek ptero: neither extrasyllabic nor magic. To appear in Rennison (ed.), Phonologica 2002. Selkirk, Elisabeth 1972 The phrase phonology of English and French. Ph.D dissertation MIT, published 1980 by Garland Press, New York. 1978 The French foot: on the status of "mute" e. Studies in French Linguistics 1: 141-150. 1980 The Role of Prosodie Categories in English Word Stress. Linguistic Inquiry 11: 563-605. 1982 The syllable. In The Structure of Phonological Representations, Harry van der Hulst and Norval Smith (eds.), 337-384. Vol.2, Dordrecht: Foris. 1984a Phonology and Syntax: The Relation between Sound and Structure. Cambrdige, Mass.: MIT Press. 1984b On the Major Class Features and Syllable Theory. In Language Sound Structure, Mark Aronoff and Richard Oehrle (eds.), 107136. Cambridge, Mass: MIT Press. 1991 On the inalterability of geminates. In Certamen Phonologicum II, Pier Marco Bertinetto, Michael Kenstowicz and Michele Loporcaro (eds.), 187-209. Torino: Rosenberg & Sellien Selkirk, Elisabeth, and Jean-Roger Vergnaud 1973 How abstract is French phonology? Foundations of Language 10: 249-254. Seynnaeve, Johan 1996 Another Look at Open Syllable Lengthening in Dutch. American Journal of Germanic Linguistics and Literatures 8: 10-32. Shevelov, George 1964 A Prehistory of Slavic. The historical phonology of Common Slavic. Heidelberg: Winter.

818

References

Sievers, Eduard 1878 Zur accent- und lautlehre der gemanischen sprachen. Paul und Braunes Beiträge 5: 63-163. Siptár, Péter, and Miklos Törkenczy 2000 The Phonology of Hungarian. Oxford: Oxford University Press. Skousen, Royal 1981 Analogical sources of abstractness. In Phonology in the 80's, Didier Goyvaerts (ed.), 55-92. Ghent: Story-Scientia. Slavíóková, Eleonora 1975 Retrográdní Morfematicky Slovník Öestiny s pripojenymi inventárními slovníky ceskych morfémû korenovych, prefixálních a sufixálních. Praha: Academia. Sloan, Kelly 1988 Bare-consonant reduplication: implications for a prosodie theory of reduplication. Proceedings of the West Coast Conference on Formal Linguistics 7: 319-330. Smetánka, Emil 1940 Κ vzniku m, r, 1 ζ m, r, 1 ν CeStinS. Listy Filologické 67: 354-357. âmilauer, Vladimir 1971 Novoéeské tvoreníslov. Praha: SPN. Smith, Jennifer 2003 Phonological Augmentation in Prominent Positions. Ph.D dissertation, University of Massachusetts. Smith, Norval 1988 Consonant place features. In Features, segmental structure and harmony processes. Part 1, Harry van der Hulst and Norval Smith (eds.), 209-236. Dordrecht. Spencer, Andrew 1986 A non-linear analysis of vowel-zero alternations in Polish. Journal of Linguistics 22: 249-280. 1996 Phonology. Oxford: Blackwell. Stampe, David 1972 How I Spent my Summer Vacation. Ph.D dissertation, University of Chicago. Stanley, Richard 1969 The Phonology of the Navaho Verb. Ph.D dissertation, MIT. 1973 Boundaries in phonology. In A Festschrift for Morris Halle, Stephen Anderson and Paul Kiparsky (eds.), 185-206. New York. Steriade, Donca 1982 Greek Prosodies and the Nature of Syllabification. Ph.D dissertation, MIT. 1988 Reduplication and Syllable Transfer in Sanskrit and Elsewhere. Phonology 5: 73-155.

References

819

Steriade, Donca (continued) 1993 Positional neutralisation. Paper presented at NELS 23, University of Massachusetts, Amherst. 1995a Underspecification and Markedness. In The Handbook of Phonological Theory, John Goldsmith (ed.), 114-174. Oxford: Blackwell. 1995b Neutralization and the expression of contrast. Ms, University of California, Los Angeles. 1997 Phonetics in Phonology: The case of Laryngeal Neutralization. Ms, University of California, Los Angeles. Downloadable at http://web.mit.edU/afs/athena.mit.edu/org/l/linguistics/www/bibliog raphy/steriade.html. Stieber, Zdzistaw 1958 Rozwój fonologiczny jçzyka polskiego. 2nd edition Warszawa: PWN. 1973 A Historical Phonology of the Polish Language. Heidelberg: Winter. 1979 Zarys gramatyki porôwnawczej jçzykôw slowianskich. Warszawa: PWN. Strauss, Steven 1982 A Lexicalist Phonology of English and German. Dordrecht: Foris. Swan, Oscar 2002 A Grammar of Contemporary Polish. Bloomington: Slavica. Szemerényi, Oswald 1990 Einführung in die vergleichende Sprachwissenschaft. Fourth edition Darmstadt: Wissenschaftliche Buchgesellschaft. Szigetvári, Péter 1994 Coronality, Velarity and Why They are Special. The Even Yearbook 1: 185-224. SZI. 1999a VC Phonology: a theory of consonant lenition and phonotactics. Ph.D dissertation, Eötvös Loránd University, Budapest. SZI. 1999b Why CVCV? The Even Yearbook 4: 117-152. SZI. 2001 Dismantling syllable structure. Acta Linguistica Hungarica 48: 155-181. SZI. Szpyra, Jolanta 1989 The Phonology - Morphology Interface. London & New York: Routledge. 1992a Ghost segments in nonlinear phonology: Polish yers. Language 68: 277-312. 1992b The phonology of Polish prefixation. In Phonlogical Investigations, Jacek Fisiak and Stanislaw Puppel (eds.), 185-218. Amsterdam: Benjamins. 1995 Three Tiers in Polish and English Phonology. Lublin: Wydawnictwo Universytetu Marii Curie-Sklodowskiej.

820

References

Szulc, Alexander 1974 Diachronische Phonologie und Morphologie des Althochdeutschen. Warszawa: PWN. 1987 Historische Phonologie des Deutschen. Tübingen: Niemeyer. Takahashi, Toyomi 1993 A farewell to constituency. UCL Working Papers in Linguistics 5: 375-410. 2004 Syllable Theory without Syllables. Ph.D dissertation, SOAS London. Taki, Mohamed 1995 Structure syllabique de l'Arabe Classique et représentation des voyelles longues. In Recherches en Linguistique Arabe, A. Chouta and A. Jahfa (eds.), 133-158. Casablanca: University Ben M'sick. Teyssier, Paul 1980 Histoire de la langue portugaise. Paris: PUF. Toft, Zoë forth The phonetics and phonology of some syllabic consonants in Southern British English. To appear in the proceedings of the Conference on the Phonetics-Phonology Interface, Berlin ZAS October 2001. Törkenczy, Miklos 1992 Vowel-zero Alternations in Hungarian: a Government Approach. In Approaches to Hungarian 4, István Kenesei and Csaba Pléh (eds.), 157-176. Szeged: Jate. Torre, Erik-Jan van der 2003 Dutch sonorants. The role of place of articulation in phonotactics. Ph.D dissertation, University of Utrecht. Tóth, Marianna 2002 Clustering sounds of English in CV and VC Phonology. MA thesis, Eötvös Loránd University Budapest. Tranel, Bernard 1981 Concreteness in Generative Phonology. Evidence from French. Berkeley: University of California Press. 1987 Floating Schwas and Closed Syllable Adjustment in French. In Phonologica 1984, Wolfgang Dressier, Hans Luschiitzky, Oskar Pfeiffer and John Rennison (eds.), 311-317. London: Cambridge University Press. Tranel, Bernard 1988 A propos de l'ajustement de e en français. In La phonologie du schwa français, Paul Verluyten (ed.), 89-131. Amsterdam & Philadelphia: Benjamins. Trask, Robert Lawrence 1996 Historical Linguistics. London: Arnold.

References

821

Trávnííek, Frantiäek 1921 De la quantité en tchèque. Revue des Etudes Slaves 1: 204-227. 1935 Historická mluvnice Ôeskoslovenskà. Praha: Melantrich. 1948-49 Mluvnice Spisovné Öestiny. 2 vol. Praha: Melantrich. Trubetzkoy 1939 Principes de Phonologie. French translation Paris 1986: Klincksieck. Ulbrich, Rolf 1978 Langenscheidts Taschenwörterbuch der tschechischen und deutschen Sprache. Erster Teil Tschechisch-Deutsch. Berlin, München, Wien, Zürich, New York: Langenscheidt. Vago, Robert 1973 Abstract Vowel Harmony Systems in Uralic and Altaic Languages. Language 49: 597-605. 1976 More evidence for the feature [GRAVE]. Linguistic Inquiry 7: 671674. Vaillant, André 1950 Grammaire comparée des langues slaves. Tome 1: Phonétique. Paris & Lyon: Institut d'Etudes Slaves. 1958 Grammaire comparée des langues slaves. Tome 2: Morphologie. Paris, Lyon: Institut d'Etudes Slaves. Valdman, Albert 1972 The Loi de Position as a pedagogical norm. In Papers in Linguistics and Phonetics to the memory of Pierre Delattre, Albert Valdman (ed.), 473-485. The Hague, Paris: Mouton. Vá2ny, Václav 1963 Historická mluvnice ceská. Dil li,2: Tvarosloví, Skloñování. Praha: SPN. Vennemann, Theo 1968 German Phonology. Ph.D dissertation, University of California at Los Angeles. 1971 The phonology of Gothic vowels. Language 47: 90-132. 1974a Phonological concreteness in natural generative grammar. In Toward tomorrow's linguistics, R. Shuy and C. Bailey (eds.), 202219. Washington, DC: Georgetown University Press. 1974b Words and syllables in natural generative grammar. In Papers from the parasession on Natural Phonology, A.Bruck, R.Fox and M.La Galy (eds.), 346-374. Chicago: Chicago Linguistic Society. Verluyten, Paul (ed.) 1988 La phonologie du schwa français. Amsterdam & Philadelphia: Benjamins. Vogel, Irene 1982 La sillaba come unità fonologica. Bologna: Zanichelli.

822

References

Vondrák, Wenzel 1924 Vergleichende Slavische Grammatik. Band I: Lautlehre und Stammbildungslehre. 2. Auflage Göttingen: Vandenhoeck & Ruprecht. Watkins, Laurel 1984 A grammar of Kiowa. Lincoln, London: University of Nebraska Press. Weijer, Jeroen van de 1994 Segmental structure and complex segments. Ph.D dissertation, University of Leiden. Wiese, Richard 1986 Nichtlineare Phonologie: Eine Fallstudie des Chinesischen. Linguistische Berichte 102: 93-135. 1988 Silbische und Lexikalische Phonologie: Studien zum Chinesischen und Deutschen. Tübingen: Niemeyer. 1991 Was ist extrasilbisch im Deutschen und warum? Zeitschrift für Sprachwissenschaft 10: 112-133. 1996 The Phonology of German. Oxford: Oxford University Press. Wijk, Nicolaas van 1931 Geschichte der altkirchenslavischen Sprache. Berlin, Leipzig: de Gruyter. 1949-50 Les groupes i>r, br, t i , bl en slave commun et en russe. Juznoslovenski Filolog 18: 39-47. Williams, Edwin 1976 Underlying tone in Margi and Igbo. Linguistic Inquiry 7: 463-484. Wilmanns, Wilhelm 1911 Deutsche Grammatik. Gotisch, Alt-, Mittel- und Neuhochdeutsch. Strassburg: Trübner. Wiltshire, Caroline 2003 Word and phrase-final alignment. In The Syllable in Optimality Theory, Caroline Féry and Ruben van de Vijver (eds.), 254-268. Cambridge: Cambridge University Press. Yearly, Jennifer 1995 Jer vowels in Russian. In Papers in Optimality Theory, J. Beckman, S. Urbanczyk and L. Walsh (eds.), 533-571. Amherst, Mass.: GSLA. Yoshida, Shohei 1990 A government-based analysis of the "mora" in Japanese. Phonology7: 331-351. 1993 Licensing of empty Nuclei: The case of Palestinian vowel harmony. The Linguistic Review 10: 127-159. 2003 The syllabic nasal in Japanese. In Living on the Edge. 28 papers in honour of Jonathan Kaye, Stefan Ploch (ed.), 527-542. Berlin, New York: Mouton de Gruyter.

References Zee, Draga 1988 1995 Zink, Gaston 1986

823

Sonority constraints on prosodie structure. Ph.D dissertation, Stanford University. Sonority constraints on syllable structure. Phonology 12: 85-129. Phonétique historique du français. Paris: PUF.

633 Subject Index

-

-

-

-

this index refers to paragraphs §, that is the running number in the page margins. reference to a § that identifies the beginning of a chapter or a section refers to this chapter or section and to all of its sub-sections. §476c: an alphabetic character after a § number indicates the number of the paragraph within the § at hand. For example, §476c refers to the third paragraph of §476. cross-reference to other entries of the subject index are introduced by "—•": look up here. cross-reference to a sub-entry of the subject index uses the structure of computer files: the sub-entry "empirical object" of the entry "Coda" is identified as "—• Coda/ empirical object". the headings of level n+1 entries continue the headings of level η entries. For example, if "co-occurrence restrictions in # " is followed by "absence of 1 at the next level, the actual heading of the lower level entry reads "absence of co-occurrence restrictions in # ". boldfaced § numbers indicate that the subject is most prominently studied there. reference to footnotes is made by indicating their number, but also by mentioning the number of the § in which it occurs: "note 258 (415)" refers to footnote 258 which occurs in §415.

A abstractness (and concreteness) general: old generative debate 307 French: protective schwa 445 drive against in Slavic (Spencer 1986) 450 adjacency effects vs. positional effects 567, 570 algorithm —• syllabification algorithm Alternation Condition (Revised Alternation Condition) 307 anything-goes languages (i.e. any CC is a good word-initial CC) 396 arboreal structure —• branching constituents

beginning of the word 83, 402 diacritics in phonology 84, 88, 405 typology of initial restrictions (anything-goes vs. TR-only) 396 stable effects of 87 co-occurrence restrictions in # 89, —» co-occurrence restrictions stability of the first vowel 90 strength of the first consonant 91 BELOW -> HIGH, LOW Binary Theorem what it is 177, 626c unwarranted prediction: the Nucleus governs the Coda 178

826

Subject Index

bogus clusters 180 typical instance: tl,dl clusters 182 created by post-tonic syncope in English 183 and t-lenition in English 184 inCVCV 185,215 boundary abuse 489, 530s branching constituents (and their translation into CVCV) —• syllable structure

c-command 15,17,179 (Prosodie Gvt) Charm 6, 11,37, 54a, 55c circularity (in general and in phonology) 390 catalexis (Kiparsky 1991): when final empty Nuclei count 509 Closed Syllable Shortening (lingua franca) empirical situation 222 in Standard GP 18, 176-179, 153 in CVCV 156, 164 comparison with Compensatory Lengthening and Open Syllable (Tonic) Lengthening 228 closed syllable (lingua franca) vs. open syllable, in CVCV 35 (first definition), 67 (second definition), 163 (third definition) causality of closed syllable effects 175 Closed Syllable Adjustment (French) —• yer context/ LOWER pattern/ bearing on other alternations outside of Slavic/ French Coda (lingua franca) in CVCV first definition 68 second definition 150 effects on Coda consonants 174 effects on preceding vowels 175 word-initial 96 and elimination of diacritics 86 both Codas (internal and final) may or may not behave alike 325 final Codas (= word-final consonants) ban on word-final Codas in Standard GP Licensing/ Coda Licensing

Coda (lingua franca)/ final Codas (continued) inability of Standard GP to express the Coda context _{#,C} 525 empirical situation: comparison of final and internal Codas final φ internal effect on Codas 527 effect on preceding vowels 533 final = internal effect on Codas 528 effect on preceding vowels 534 behaviour of sonorants in Codas 593600 behaviour of nasals homorganicity of nasals in NC clusters 593-595 contrast before stops vs. before fricatives 597s loss of place of word-final nasals 595 genesis of nasal vowels < VN{#,C} 600 homorganicity of nasals after obstruents CN in German 602 nasals kill a schwa in order to become syllabic 602,604 homorganicity has got nothing to do with adjacency 605 behaviour of laterals laterals kill a schwa in order to become syllabic 604c Coda capture —> vowel-zero alternations/ Lower pattern/ consequences Coda Licensing —» Licensing Coda Mirror empirical object: lenition & fortition definition 112, 567-570 three conditioning factors: position, stress, sharing of melody 113, 203s empirical evidence evolution of Latin obstruents in French 117 Somali stops 118

Subject Index Coda Mirror/ empirical object/ empirical evidence (continued) S ievers1 Law 119 Tiberian Hebrew spirantisation 560 French fortition of Latin [j] 565 High German (Second) Consonant Shift 572 positional influence general 114 vs. adjacency effects 567, 570 there are two distinct weak positions 131 lenition in branching Onsets 50, 197 lenition in the Strong Position: High German Consonant Shift 572 parameter: post-Coda consonants may or may not be sensitive to the content of the preceding Coda 200, 204 theory scope 111 descriptive adequacy: no disjunction 124 explanatory adequacy: why strong positions are strong 130 relative strength of positions 130 relativity: positions are not strong or weak, they are stronger or weaker than other positions 573 vocalic vs. consonantal lenition (Szigetvári) 131 and Licensing Inheritance 131, 195 consequences of the Coda Mirror 148 parameter: word-initial consonants may or may not be strong (but post-Coda consonants are always strong) (Seigneur-Froli) 558c

827

Coda Mirror/ theory (continued) the mirror effect: opposite structural description and effects of the Coda and its Mirror 124, 571 same consequences as those provoked by the Coda 337, —• disjunctions and their consequences the Coda Mirror is a positive phonological object, as real as Onsets and Codas 556 concreteness —* abstractness Compensatory Lengthening general: progressive, regressive 226 conditioned by following clusters? 227 comparison with Closed Syllable Shortening and Open Syllable (Tonic) Lengthening 228 Cz u-o, Pol ó-o, ^-ç instances of Compensatory Lengthening ? 433 complement —» head complexity 11,37, 40, 51 counts only Place definers 52 concreteness —• abstractness consonantal interaction —» Government, Infrasegmental consonantal (vs. vocalic) lenition —• Coda Mirror, theory consonant clusters, restrictions on wordinitially —* co-occurrence restrictions in# constraints —• OT co-occurrence restrictions in # why in initial position? 89 absence of: anything-goes languages where both #TR and #RT occur general 105, 382,402 a case study: Moroccan Arabic 383-385 another case study: Polish 375s, 622 the regular analysis: extrasyllabicity 352

828

Subject Index

co-occurrence restrictions in # ued)

(contin-

#TR-only languages classical account 103, 382s, 396 observation in classical approaches (syll. algorithms) 396s and Standard GP 398-400 vs. explanation in CVCV 401-409 account in CVCV 104, 109, 401409 why are there no #RT-only languages (i,e, where #RT, but not #TR exists)? 402 initial s+C sequences s+C clusters cyclic application of processes Slavic vowel-zero alternations: cyclic application of Lower 327, 415h Lower vs. Havlik pattern (Havlik = non-cyclic Lower) 417s, 492s alternative solution: note 258 (415) (Anderson, Gussmann) against cyclicity (Spencer 1986) 450 in Standard GP: Gussmann & Kaye (1993) 487s

Dependency Phonology structural analogy (John Anderson) 2, 17, note 116(166) the sonorant is the head of branching Onsets 38 privative melodic primes and subsegmental structure 41-43, 56, 624 derived environment effects 307s devoicing (syllable-final) in German 609 diacritics (morphological) —• beginning of the word disjunctive contexts, disjunctivity disjunctions and their consequences for phonological theory 303, 559, 562s the yer context (in closed syllables and in open syllables if the following vowel alternates with zero) —• vowel-zero alternations, LOWER pattern Coda _{#,C} — Coda

disjunctive contexts, disjunctivity (continued) closed syllable C{#,C} —> closed syllable syllable weight {V, _ . C } 501 Coda Mirror {#,C} —• Coda Mirror in Codas and before unpronounced alternation sites {#,C} + _0.CV 579 analysis of vowel-zero alternations in Standard GP —» vowel-zero alternations domains (phonological) and domain-final empty Nuclei, also —• Final Empty Nuclei general 24, 630 from word-final to domain-final 24 in Polish prefixes 249 in word-final monster clusters (sixths) 349 in the Standard GP analysis of Slavic vowel-zero alternations Gussmann & Kaye (1993) 485

edge (of the word) right —• final empty Nuclei left —• beginning of the word Elements —• phonological primes Empty Category Principle (ECP) in Standard GP 16-18, 25, 138, 630 in CVCV 16 (version 1), 18 (version 2), 60 (final version), 258 empty Nuclei in Standard GP 17 penologists' schizophrenia. Tina Turner was wrong: what you get is NOT what you see 387 final —» Final Empty Nuclei extrametricality —• syllable weight extrasyllabicity 10,339 empirical scope general 341 type I: enforced underpaying (too many consonants around at word edges) 343 (in CVCV: parameterised governing abilities of FEN 550)

Subject Index extrasyllabicity/ empirical scope (continued) type II: deliberate underparsing (word-final consonants are parsable but do not behave like Codas) 358 (translation into CVCV: parameterised licensing abilities of FEN 550) word-final general 344, 348 non-arbitrary impairment of the behaviour of internal and final Codas 341 word-final monster clusters are morphologically complex 348 word-initial general 352 Polish initial clusters: one extrasyllabic consonant at most 375s, 622 in CVCV 408 word-medial Polish trapped consonants 269s, 357 Peripherality Condition 357 word-initial and word-final extrasyllabic consonants show different behaviour 352 motivation for extrasyllabicity general philosophy 340-342 a consonant is unparsable 343, 550 classical serial analyses 344, 352 a word-final consonant does not behave like a Coda 358, 550 why there can be no extrasyllabicity in GP 363 arguments against extrasyllabicity extrasyllabic consonants pattern with Onsets 365 word-final consonants are either extrasyllabic for all processes of the language, or for none: extrasyllabic once, extrasyllabic forever (360), 364,551 if constituents do not express cooccurrence, they express nothing at all 366s

829

extrasyllabicity/arguments against extrasyllabicity (continued) s+C effects occur both wordinitially and word-intemally 368 extrasyllabicity predicts the existence of random-sized extrasyllabic clusters 373, 552 extrasyllabicity takes no notice of the obvious cause for the peculiar behaviour of consonants at edges: their location. It must be wrong since it is a purely phonological analysis of effects that are due to the morphological category "edge" 378, 549 extrasyllabicity predicts the existence of extrasyllabic vowels but there is no such thing 379, 549 "extrasyllabicity" in CVCV: how the empirical facts that are classically accounted for by extrasyllabicity are handled right edge extrasyllabicity 379, 547 left edge extrasyllabicity 379, 408

F Final Empty Nuclei (FEN) as such (their existence) 11,18, 24, 628, 631 special properties of in Standard GP parametric Licensing of FEN 18, 24 FEN are able to dispense Government Licensing 61 genuine tradition in GP: soft parameters, i.e. on lateral relations rather than on constituency 266, 539 in the debate on whether Lie was about the existence of x-slots and Gvt about the melodic properties thereof in Standard GP 145

830

Subject Index

Final Empty Nuclei (FEN)/ special properties of (continued) in CVCV parametric Government of FEN 62 parameterisation of the governing and licensing abilities of FEN 535 parameterisation of FEN covers the empirical field of what is classically interpreted as word-final extrasyllabicity 547, detail —• extrasyllabicity RT# exist = schwa can govern 540 FEN cannot govern Nuclei that possess melody underlyingly (floating or attached), i.e. alternation sites 541 Closed Syllable Shortening only wordinternally = FEN can license 544 word-final "extrasyllabic" consonants are licensed by FEN 542, 550 summary parameterisation of FEN and schwa: FEN 545 , schwa 554, both 555 special properties of the right edge of words 63 extrametricality (in Latin): FEN may (= CU is not extrametrical) or may not (= C# is extrametrical) count for the purpose of stressassignment 509 fortition (strengthening) —• Coda Mirror

G geminate in Standard GP vs. CVCV 235 in CVCV 237, 239

geminate (continued) why they cannot be adjacent to another consonant in Standard GP 236 in CVCV 238 geminate integrity 113,594 partial (nasal) geminate 594 GEN (what should it contain?) 321 Government in Standard GP Proper Government general 17,19, 25, 72, 77, 630 blocked by intervening governing domains 21s, 66, 71-74,108 confusion with Licensing 136, 159 Recursive Government (Govern or Perish) 23 Interonset Gvt 106 Constituent/ Interconstituent Gvt 626 Government Licensing —• Licensing in Standard GP Prosodie Government 179 in Harris' (1994a) system Constituent/ Interconstituent Government 189 Proper Government 188 vs. Licensing 187, esp. 193 in CVCV of final empty Nuclei 150 inventory of different forms of Gvt 148 (first version), 161 (final version), (also 500) intemuclear Gvt translates as vowel-zero alternations 164 Infrasegmental Gvt general: its properties 34, 52, 57s special status of 149 vs. branching Onsets 64 TR without Infrasegmental Government 66 Government Licensing —» Licensing in CVCV

Subject Index 831 Government/ in CVCV (continued) Government and Licensing are antagonistic forces introduction of Gvt vs. Lie 125 (esp. 129) empirical: two different patterns in nature 151, 164, 440, 472 yers have opposite effects on preceding vowels 468, 472 complementary distribution of internuclear Gvt and Lie 160 unification of Proper Gvt with other forms of Gvt 148 comparison with Gvt in Harris' (1994a) system 194 grid —• metrical grid grounded constraints —• OT

hardening (fortition) —• Coda Mirror Harris (1994a), his system modifying Standard GP —* Licensing/ in Harris' (1984a) system Havlik's law —* vowel-zero alternations head (vs. complement) in branching Onsets traditional view: the obstruent is the head 40 CVCV: the sonorant is the head, literature 38 sonorants are the big guys 51 derived from Gov-Licensing 58 in long vowels: alternating VVs are head-initial, non-alternating VVs are head-final 230 HIGH, LOW 215s, 229, 517 High German (Second) Consonant Shift 572 homorganic NC clusters classical account and CVCV 593s homorganicity has got nothing to do with adjacency 605 behaviour of nasals in Codas —» Coda/ behaviour of nasals hybrid models: lateral causality, but vertical structure (Harris 1994a, Takahashi 1993)207s

I,J,Κ initial CV —» beginning of the word, cooccurrence restrictions initial Codas —• Coda initial clusters —• co-occurrence restrictions in # internal structure of consonants/ vowels—» phonological primes Interonset Government —• Government intervening governing domains (block PG) —• Government

L Langue vs. Parole, Saussurian principle violated by the grounded philosophy ofOT 318 lateral relations Government —• Government Licensing —• Licensing that replace arboreal syllabic structure — geminate lenition —• Coda Mirror lexical representation of vowels that alternate with zero —» vowel-zero alternations Lexical Phonology reaction on abstractness 307s morphology: level 1 vs. level 2 etc. 24, 248, note 166 (249), 349

832

Subject Index

Licensing in Standard GP general 137 Government Licensing general 58,126-128, 153, 169 vs. other forms of Licensing in SGP 137 internuclear (Yoshida 1993) 155 lateral causality of consonant loss 169 cross-linguistic parameterisation 266 p-Licensing 138 Licensing Principle 138, 189 Magic Licensing 96 Coda Licensing what it is 18, 628 vs. other forms of Licensing in SGP 137 lateral definition of the Coda 168, 538 unable to express the Coda context {#,C}, demotion to a parameter (Piggott, Rice) 525, 538 confusion with Government 136, 159 Licensing Inheritance general 131,195 troublesome predictions 197-205 lenition is often entirely unrelated to stress 197, 203 the status of Codas is calculated only on local grounds 198 the Coda Mirror context {#,C} remains disjunctive 202 the Coda context {#,C} does not enjoy a unique identity 199 Licensing Inheritance does not control the lenition of postCoda consonants 200, 204 Licensing Inheritance blurs the freshly gained separation of Gvt and Lie 205

Licensing (continued) in Harris' (1994a) system Constituent/ Interconstituent Licensing 188 Projection Licensing 188 vs. Gvt 189 (version 1), 193 (version 2), 205 (final version) Closed Rhyme Shortness lateralised 190-192 trouble: Coda clusters and wordfinal Codas 191 in CVCV definition 129 inventory of different forms of Licensing 148 (first version), 161 (final version), (also 500) internuclear Licensing translates as vowel length alternations 164 overview 148 Government Licensing 58 internuclear Licensing (vowel length) 158s complementary distribution of internuclear Gvt and Lie 160 Licensing and Government are antagonistic forces 125 esp. 129), 151,164 yers have opposite effects on preceding vowels 468,472 comparison with Licensing in Harris' (1994a) system 194 Loi de Position (French) —• yer context/ LOWER pattern/ bearing on other alternations outside of Slavic/ French long vowel empirical situation stable (i.e. non-alternating) 221 alternating comparison of all kinds of vowel length alternations 222 —> Closed Syllable Shortening —» Open (Tonic) Syllable Lengthening —• Compensatory Lengthening

Subject Index

833

long vowel (continued) theory: representation of non-alternating in Standard GP 626 in CVCV 231 alternating in Standard GP 18,153 in CVCV 158-160 opposition alternating (headinitial) vs. non-alternating (head-final) long vowels 230 CVCV: the complement of all long vowels must be licensed 228 LOW HIGH, LOW LOWER (pattern and rule) —* vowel-zero alternations

Magic Licensing —• Licensing margin (of the word) right —» Final Empty Nuclei left —* beginning of the word markedness 307, 315-317, (597) metrical grid 502s manner of articulation —» phonological primes mirror effect: opposite structural description and effects of the Coda and its Mirror 124, 571 morae 501

nasals, nasal vowels —» Coda/ behaviour of sonorants in Codas/ behaviour of nasals Natural Phonology, Natural Generative Phonology 307s, note 214 (333) (antiabstractness) negative evidence in diachronics: "X does not exist", i.e. absence of traces of an alleged intermediate stage in dialects 576s null hypothesis for syllable structure is lateral 211

obstruents in branching Onsets —• head internal structure —• phonological primes open syllable (vs. closed syllable) —> closed syllable Open (Tonic) Syllable Lengthening general 157, 164, 222 comparison with Compensatory Lengthening and Closed Syllable Shortening 228 Optimality Theory (OT) structure and process: only process in OT, no structure 309 grounded constraints 134,317s TETU 316 factorial typology 321 and extrasyllabicity: introduction to 339 positional faithfulness (Beckman 1997) 132 visibility of melody by prosodie structure (de Lacy 2002) 517 overgeneration in SPE 306 reactions against option 1 : reducing abstractness (Lexical Phonology, Natural (Generative) Phonology) 307 option 2: (autosegmental) representations 308 in OT 309, especially 314s

P,Q parameters list of parameters treated in this book 620 genuine tradition in GP: soft parameters, i.e. on lateral relations rather than on constituency 266,539 four nuclear categories: full vowels, schwa, final empty Nuclei, internal empty Nuclei; parameterisation of their lateral actorship 494, 555

834

Subject Index

parameters

(continued)

schwa (=vowel that alternates with zero): parameterisation of its gov. and lie. abilities 473, summary 479 in modern Slavic: Gvt 474 in Western Slavic: Gvt 474, 476 in modern Czech: Lie 475 in modern Polish: Lie 581s in Old Czech and Old Polish: Gvt 469s in Moroccan Arabic: Gvt 469s in French: Gvt 469s, Lie 477s in German: Gvt 469s, Lie 483 in Dutch: Gvt 484 final empty Nuclei: parameterisation of its governing and licensing abilities —• FEN/ parameterisation 535 summary parameterisation FEN and schwa 545 (FEN), 554 (schwa), 555 (both) parameters need enriched representations 323 Parole vs. Langue, Saussurian principle violated by the grounded philosophy ofOT318 Particle Phonology: privative melodic primes 41-43, 624 Peripherality Condition 357 phonological domains —• domains phonological primes internal structure of consonants general 37, 41 place vs. manner 52 velars 43 coronals 42c, 45 obstruents 37, 52 sonorants 37, 48 r 49a-d, 50 laterals 50 nasals 49e sonority consonantal 46 vocalic 54 internal structure of vowels back vowels 43b comparison consonants-vowels 42a-b general 624

place of articulation —» phonological primes positional effects vs. effects due to adjacency 567, 570 Positional Faithfulness 132 postnuclear Rhymal complement —> Rhymal Adjunct primes —> phonological primes privative melodic primes 37 Projection Principle —» structure preservation Prosodie Government —> Government

R Reduction, i.e. the elimination of any sequence "empty Nucleus plus Onset which does not have any skeletal slot", device of Kaye & Gussmann (1993) 487 redundancy of arboreal syllable structure in hybrid models (Standard GP, Harris 1994a) 1 Is, 209s, 398-400 representations their function before OT 306-308 (fighting back —• overgeneration) their function in OT (none, hence we are back to —» overgeneration) 309, especially 312s and their progressive enrichment, especially in GP 308,323 need to be rich enough in order to support the parameterisation of their body 323 eliminate an absolute neutralisation 328s resyllabification 10, 17,363 Rhymal Adjunct 6,627

s+C clusters and effects word-initial 96 word-initial and internal 368 Saussurian principle Langue vs. Parole violated by the grounded philosophy of OT 318 schwa = vowel that alternates with zero —• vowel-zero alternations parameterisation of its lateral actorship —» parameters

Subject Index Second (High German) Consonant Shift 572 Sievers' law 119 sonorants in branching Onsets —• head internal structure —» phonological primes behaviour in Codas —• Coda/ behaviour of sonorants sonority sonority sequencing 11, 37, 51, 382 restrictions on word-initial clusters —» co-occurrence restrictions in # representation of in internal structure of consonants —» phonological primes SPE SPE hocus-pocus (Lightner: sweet and hedonistic etc. from the same UR) 333 overgeneration 306 Standard GP back to SPE: unable to express the Coda context 525 Standard Government Phonology 10s, 15, 136, 153, 623 strengthening (fortition) —• Coda Mirror stress assignment —* syllable weight strict directionality, strict locality (Standard GP) 108, 626 Strong Position —» Coda Mirror structural analogy —» Dependency Phonology structure preservation 10, 17 super-heavy Rhymes —> Closed Syllable Shortening parameter on their existence in Harris1 (1994a) system 190 parameter on their existence in CVCV 232 syllabic and trapped consonants can obstruents be syllabic? 617, note 195 (294), note 386 (617) syllabic consonants provoke unvocalised prefixes in Czech 246, 263 alternations of syllabic and nonsyllabic versions of the same consonant 259

835

syllabic and trapped consonants/ syllabic consonants (continued) genesis of: loss of a preceding vowel 262 complementary distribution of aC and CI 257, 261 in classical approaches since SPE 253 representation in CVCV not in Nuclei 254 right- or left branching? 246, 260, 263 trapped consonants in Polish: description 244 provoke vocalised prefixes in Polish 247, 265 transparent to voicing 269 are extrasyllabic (J. Rubach) 270 in CVCV representation 264 trapped sonorants are in fact obstruents 271 comparison syllabic vs. trapped consonants opposite regarding stress and poetry 245,251,267 diachronic situation in Western Slavic Czech syllabic consonants are either trapped or prevocalised in Polish 278 interpretation of OCS script and pan-Slavic comparatism 279s reflexes of CS trtt in Western Slavic 286s Old Czech: trapped vs. syllabic is phonemic 288 syllabification algorithms 340-343, 382, 396 syllable structure also —• branching constituents arboreal vs. lateral: the null hypothesis is lateral 211 hybrid arboreal-lateral models: Standard GP, Harris (1994a) 208 redundancy of arboreal structure in Standard GP 1 Is, 209s, 398-400

836

Subject Index

syllable structure (continued) branching Onsets (and their translation into CVCV) also —» Government, Infrasgmental the sonorant is the head of TR clusters 38 replaced by lateral relations in CVCV 165, 399 comparison of branching Onsets and Infrasegmental Gvt 64 syllable weight Weight by Position 504-510 (classical approaches), 520 (in CVCV) closed syllables are heavy according to the sonority of the Coda consonant (Kwakwala) 506 (data), 511 (classical approaches), 521 (in CVCV) Onsets do not contribute to weight 512 alleged cases of Onset weight 512 extrametricality 509 analysis of in moraic and grid-based theories 507, 513 in CVCV 514 general pattern 516 visibility of Nuclei (enclosed within TR clusters): stress is blind for melody 517 Onset-irrelevance is in-built: Onsets are never followed by an empty Nucleus 518 expression of the Kwakwala pattern (closed syllables are heavy according to the sonority of the Coda consonant) 521 same representations for syllabic and prosodie processes 519 Weight by Position = empty Nuclei are or are not counted 520 stress is an exclusively nuclear phenomenon 518, 522

syntax, relations with similarities, unification with phonology survey of syntactic concepts that have been applied to phonology in Government Phonology 17, note 117(166) individual concepts: ECP & Proper Government 15, c-command 179, structure preservation 10, 17 Structural Analogy (John Anderson, Dependency Phonology) 2, 17 Locality Vol.2,1 differences with phonology, there is no 100% superposition 2 recursion 2, 213 co-occurrence restrictions 213 Τ Tonic Lengthening —> Open Syllable Lengthening Transconstituent Government —• Government, Interconstituent Transconstituent Licensing —• Licensing, Interconstituent trapped consonants —» syllabic consonants U underpaying (of consonants at word margins) —• extrasyllabicity/ empirical scope underspecification 41,624 UPPER HIGH, LOW V velar nasal (German) 446, 480, 587, 605 vertical structure —• branching constituents vertical causality —» latéralisation of structure and causality visibility of Nuclei from ABOVE (stress) —• syllable weight/ analysis of7 in CVCV vocalic (vs. consonantal) lenition —• Coda Mirror, theory vowel length —• long vowel

Subject Index vowel-zero alternations empirical situation in Czech prefixes 28, 31s, 94, 246 in Polish prefixes 247 Czech 31,127, —• vowel-zero alternations, - • LOWER pattern modern Slavic languages —> vowelzero alternations, LOWER pattern cross-linguistic 16, 23 French 21, 33, 66,126 Moroccan Arabic 22s, 384 transparency of intervening clusters 21s, 31-33, 70 whether identical vowels in a given language do or do not alternate cannot be predicted 78 more than one vowel alternates in a given language 79 in Standard GP general 19,630 intervening governing domains block PG (and falsification of this statement) 30s the analysis in Standard GP is disjunctive 72 alternating vowels are epenthesised (insertion strategy) 77 typical example of latéralisation in Standard GP 167 in CVCV general 69, 76 non-disjunctive analysis 71 alternating vowels are present in the lexicon (deletion strategy) 79-82 before TR clusters 32, 34,66, 70 lexical representation of alternating vowels in Standard GP: empty Nuclei and insertion 77 in CVCV: floating melody and deletion 79-82 disjunctive (SGP) vs. non-disjunctive (CVCV) analysis 21s, 71, 167 comparison Standard GP vs. CVCV 69 LOWER rule the occurrence of yers in the lexicon is unpredictable 415d

837

vowel-zero alternations/ LOWER rule (continued) yers and non-alternating vowels must be distinct in the lexicon 415e linear 327,415 ordered rules: Lower before yer deletion 415c cyclic application (Rubach 1984) 415h, 487s alternating vs. final yers 419 synchronic evidence 415i, 420 diachronic evidence: only alternating yers can originate in epenthesis 421 final yers are case markers 415e, 421s autosegmentalised 329, 423 difference with Havlik pattern 416 non-etymological yers 421 Slavicists were practising GP before it was born, and GP reinvented the Slavic analysis without knowing about it Lower describes a regressive lateral relation between two yers 412,424s, 458 the "abstract vowels" of Lower are empty Nuclei: the distribution of the latter in Standard GP and the former is exactly identical 412, 415i, 425, 441,460 LOWER pattern (= yer context) overview 326, 336 summary: the full empirical puzzle 440 two opposite patterns is the yer context self-conflicting? Yers sometimes provoke the appearance of the strong, at other times of the weak alternant 440, 468 —• Government/ in CVCV/ antagonistic forces who is who: which alternations are governed by Gvt, which by Lie? 473

838

Subject Index

vowel-zero alternations/LOWER pattern (continued) bearing on vowel-zero alternations modern Slavic languages, general 411 synchronic distributional situation 414 coincidence of analysis in the Slavic literature (abstract vowels = yers) and in GP (empty Nuclei) 412s, 425 difference with the Havlik pattern 416 non-etymological yers 421 bearing on other alternations in Western Slavic 428, 474, 476 cz Zába - iabka - iabek 430 (data), 465,475 (interpretation), 621 cz ù-o, poi ó-o 432s pol 3-ç 434 pol: implosion of /ji/, becoming a palatal nasalised glide 581s bearing on other alternations outside of Slavic Southern (Midi) French ATRness of mid vowels 437, 466, 477 French schwa-[e] 439, 463, 478 German velar nasal 446, 480, 587 consequences and analyses of overview 440, 456 here: yers (or "abstract vowels") are empty Nuclei 412s, 438, 441,460 here: the lateral relation described by Lower is (Proper) Gvt 458 Standard GP Gussmann & Kaye (1993) cyclicity, Reduction and domains 485 Rubach (1986), Rubach & Kenstowicz (1987): autosegmentalised Lower 329, 423

vowel-zero alternations/ LOWER pattern (= yer context)/ consequences and analyses of (continued) Rubach (1986), Rubach & Kenstowicz (1987): objections against empty Nuclei 452 and how they are overcome 461 Anderson (1982): schwas are empty Nuclei, Coda capture 442 Hall (1992): schwas are floating x-slots, but no Coda capture 446 Spencer (1986): yers are empty Nuclei, context-free fill-in without any intervocalic causality 449 Szpyra (1992a): insertion into unsyllabifiable clusters without any intervocalic causality 454 Havlik pattern description 417, 326 difference with the Lower pattern 418 "secondary vocalisation" from the old to the modern languages 418 non-etymological yers 421 Havlik does not vocalise yers, but empty Nuclei 421b parameter Havlik vs. LOWER empirical: there are two patterns of V-0 alternations in nature 326, 416, 470 theoretical: alternating vowels can (Havlik) or cannot (Lower) govern 326, 470, 472s comparison of three competitors: classical (Rubach 1984), Standard GP (Gussmann & Kaye 1993), CVCV 490 only one mechanism crosslinguistically 334s, 381

Subject Index W,X weak position(s) —• Coda Mirror weight —» syllable weight Weight by Position —• syllable weight word-initial clusters —> co-occurrence restrictions in #

839

y,Z yers (yer context) —• vowel-zero alternations, LOWER (rule and pattern) alternating vs. final yers —» vowel-zero alternations/ LOWER rule diachronic development of the CS vowels 416, 421

634 Language Index

-

-

-

-

-

-

-

this index refers to paragraphs §, that is the running number in the page margins. reference to a § that identifies the beginning of a chapter or a section refers to this chapter or section and to all of its sub-sections. §476c: an alphabetic character after a § number indicates the number of the paragraph within the § at hand. For example, §476c refers to the third paragraph of §476. cross-reference to other entries of the language index are introduced by "—*": look up here. cross-reference to a sub-entry of the language index uses the structure of computer files: the sub-entry "Sievers1 Law" of the entry "Common Germanic" is identified as "—• Common Germanic/ Sievers' Law ". the headings of level n+1 entries continue the headings of level η entries. For example, if "interpretation of this parameter" is followed by "in Standard GP" at the next level, the actual heading of the lower level entry reads "interpretation of this parameter in Standard GP". boldfaced § numbers indicate that the subject is most prominently studied there. § numbers in brackets indicate that the subject is only marginally touched there. "minor quotation" indicates that the language and the subject at hand are only mentioned: no further explanation or data or analysis will be found in the § indicated. individual languages of a bigger family are grouped under the entry of the family: "Palestinian Arabic" for example is a secondary entry of "Arabic". reference to footnotes is made by indicating their number, but also by mentioning the number of the § in which it occurs: "note 258 (415)" refers to footnote 258 which occurs in §415.

842

Language

Index

A Akkadian stability of the first vowel 90, 93 Arabic Classical Closed syllable shortening 154, 164, 223 Moroccan vowel-zero alternations general 16, 20, 383-385 behaviour of alternation sites before monomorphemic clusters (geminates) 22 behaviour of alternation sites before monomorphemic clusters (geminates) makes SGP disjunctive 72, 167 non-disjunctive solution in CVCV 73, 334 behaviour of two alternation sites in a row 23 are not conditioned by the nature of the following cons cluster 384 typology of vowel-zero alternations: Moroccan Arabic follows Havlik 331, 469s Gussmann & Kaye (1993): distribution of Havlik and Lower according to the vowel that alternates 486s anything-goes: absence of restrictions on initial clusters 89, 105, 353, 381 (specifically 383-385, summary 335), 396, 402 the first vowel of words alternates with zero 90 Palestinian (Levantine) alternations in vowel length Yoshida (1993): long vowels need a Licence to govern 155 in CVCV: managed by Licensing 164 overview Tonic Lengthening 224

Palestinian (Levantine) Arabic/ alternations in vowel length (continued) Kenstowicz (1994): stress placement and Open Syllable (Tonic) Lengthening are extrasyllabicity effects 360 typological situation of Palestinian Arabic: long vowels occur in final closed syllables 341, 536 Austrian German —• German Β Baltic (Balto-Slavic) minor quotation Sievers' Law 121 CS tbrt and trbt remain distinct in Baltic and Eastern Slavic 282 CS trbt and tbrt: Balto-Slavic comparatistic situation 282 reflex of IE r: ir, ur 279 Bella Coola -»· Salish Berber minor quotation word-initial #RT clusters 353, 383 geminates adjacent to a consonant (in heteromorphemic situations) note 155 (235) can obstruents be syllabic? note 195 (294)

Chilungu (Bantu) compensatory lengthening V,-V 2 —• V 2 V 2 226 Cologne German —• German Common Germanic minor quotations evolution of CG p,t,k in High German —» Old High German/ High German Consonant Shift Sievers' Law CG [j] - [ij] note 77 (121)

Language Index Common Germanic/ minor quotations (continued) early generative hocus-pocus (Lightner: sweet - hedonistic, tooth - dental etc. from the same UR) note 214 (333), Modern English Common Slavic —• Slavic Czech Old Czech CS tbrt > syllabic in ocz, but CS tn>t > trapped in ocz 286 evolution of CS tbrt, tn.t in ocz and mcz 286 secondary vocalisation of syllabic consonants (Czech, Slovak, Polish) 287 ocz phonemic opposition syllabic vs. trapped consonants 288290 vocalisation of prepositions before syllabic/ trapped consonants 291 ocz trapped consonants refuse to become syllabic word-initially in mcz 292 Havllk's Law 417s Modern Czech vowel-zero alternations general 16,31,33 in consonant-final prefixes 28, 31-33 in consonant-final prefixes when followed by a root with a syllabic consonant 246 two alternation sites in a row 23 attempts to make -eòek morphologically non-complex (domeòek) note 249 (414) Slavic (Czech) vowel-zero alternations: distributional facts (=Lower) 414 Havlik's Law 417s in prefixes when followed by RT-initial roots 94

843

Modern Czech/vowel-zero alternations (continued) in short adjectival forms (nemocny - nemocen) 430, 452 blocked after consonant clusters Kadlec - Kadlece 127 yers without etymological basis: diachronically "epenthetic" yers (pë-sn-b > plseft) 421 vowel length Closed Syllable Shortening (kráva - krav - kravka) 154, 223, 429s, (435), 465, (474), (534), 621 stem-final clusters may or may not enclose an alternation site: CVCC-a/o - CVCeC (farma - farem) vs. CVCC-a/o - CVCC (korba - korb) 465, 475,534 inalterable long vowels 232 ú-o ([uu]-[o]) (dùm - domu) synchronic situation 429, 432, 435 diachronic situation 433 analysis 474, 476 syllabic and trapped consonants synchronic situation trapped consonants in Modern Czech 293 inventory of trapped roots 294 vocalisation of prefixes before trapped roots 295 syllabic cons bear stress and count in poetry (trvat, bratr) 245 trapped cons neither bear stress nor count in poetry (lhát, rty, kftit, hfbitov) 245 word-initial trapped consonants (regular analysis: extrasyllabic) 389

844

Language Index

Modern Czech/ syllabic and trapped consonants/ synchronic situation (continued) syllabic consonants are not transparent to voicing (by contrast to Polish trapped consonants) 269 right periphery of syllabic consonants: their occurrence before consonant clusters 298-300 complementary distribution of voiced and voiceless versions of 273 real sonorants ([r,l,n,m]) do not devo ice wordfinally, but does 272s trapped consonants in Polish words (inventory: lexically trapped vs. trapped by a vowelzero alternation), with their Czech cognates 244, 278 diachronic situation CS origin of syllabic consonants 262 Western Slavic comparatistic situation: origin of prevocalised, trapped (Polish) and syllabic (Czech) consonants 278, 286 tbrt and tn.t: Balto-Slavic comparatistic situation 282 evolution of CS tbrt, trbt in ocz and mcz 286 secondary vocalisation of syllabic consonants (Czech, Slovak, Polish) 287 ocz trapped cons, refuse to become syllabic wordinitially in mcz 292

Modern Czech (continued) distribution of Vocative allomorphs (-e, -u, -i) 43 word-internal s+C effects (lest Isti) 370

D Dutch Middle Dutch Open Syllable (Tonic) Lengthening 223 Modern Dutch minor quotation "you can get a coronal for free at the end of words", a typical Germanic feature 350 South: [r,l] implode to [j] in Coda position (kaart > kaajt) 50 schwa epenthesis into Coda-Onset sequences according to the content of the following Nucleus (harep vs. harpoen vs. kar(e)per) 484, 555 E Eastern Slavic —> Slavic English Middle English Open Syllable (Tonic) Lengthening 223 Modern English minor quotations s+C clusters 97s "you can get a coronal for free at the end of words", a typical Germanic feature 350 r-vocalisation 529, 616 homorganic NC clusters: in(impossible, incredible) 593 homorganic CN clusters: spok[r|] note 371 (602) vowel length alternations: "Closed Syllable Shortening" (keep kept) (reanalysed by Kaye 1990a) 18, 154, 175, (533)

Language Index Modern English (continued) super-heavy Rhymes (shoulder etc.): distributional situation note 123 (175) Trisyllabic laxening/ shortening op[ej]k - op[ae]city note 123 (175), 307, 333 velar softening electri[k] - elect r i c i t y 308, 333 lowering effect of [r] on preceding vowels 49 breaking before [r], [1], [ç] and [x] in Orkney English 49 post-tonic syncope (cam(e)ra) 183 t-lenition facts 184s, 200 conditioned by stress (i.e., grounding Harris' 1994,1997 Licensing Inheritance) 196-205 vowel-zero alternations (fiddle fiddler) note 125 (184) floating consonants (an apple vs. a coffee) 388 American English varieties: glottalisation of Ν 382 syllabic consonants 242, 253s, 256s, 261s typical bogus clusters tl,dl (atlas) 182, 184s word-final consonants pattern with Onsets, not with Codas (Gussmann & Harris 2002) 365 +ATR patterns with long, -ATR with short vowels 477 Weight by Position: CVC syllables are heavy 505, 510 early generative hocus-pocus (Lightner: sweet - hedonistic, tooth - dental etc. from the same UR) 333, note 264 (418) level 1 vs. level 2 affixes (e.g. párent - paréntal - párenthood) 24, 249 morphologically complex wordfinal monster clusters (sixths) 24, 349, 630

845

Modern English (continued) dark 1 in Codas (Bill - Billy) note 385 (616) English loans from the times of the Norman conquest witness Old French tf < lat k / a (channel) 577 Estonian minor quotation word-final consonants are extrametrical 509

F Finnish contrast between Coda-Onset and Onset-Onset clusters ? When the strength of the post-Coda depends on the content of the Coda 185 French Old French compensatory lengthening (testa > tête) 226 1-vocalisation in internal Codas (chevals > chevaux) 357, 527 Modern French (= Metropolitan) minor quotations bogus clusters tl,dl 182 "r": apical trill [r] > uvular fricative [χ,κ] 274, 276 floating consonants (petit café vs. petift] effort) 340, 342, 388 vowel-zero alternations before clusters 21, 33 inventory of all types of clusters adjacent to schwa 126 before clusters: zero possible for some speakers (s'cret) 66 Grammont's Law (loi des trois consonnes) 469, 478 French follows the Havlík pattern: schwa can govern 469s ATR alternations of mid vowels —> Southern French

846

Language Index

Modern French (continued)

Southern French/ ATR alternations of mid vowels (continued)

schwa - [ε] alternations (app[a]ler - app[e]elle) data 439s Anderson's (1982) account: Coda capture 442 additional empty Nuclei required 463 are an instance of Licensing 478 optional loss of both schwa and R in internal TRa clusters (autrement, aut"ment), but no loss possible word-initially (grenouille, g"nouille) 90, note 128(194) protective schwa (forte = /fort-s/) 445 modern results of diachronic events evolution of Latin obstruents in French general 117 the evolution of post-Coda consonants is independent from the nature of the preceding Coda 204 in branching Onsets: the obstruent lenites (petra > pierre) 197 fortition of j: lat j > fr 3 in Strong Position 565 French palatalisation: lat k,g > J,3 (gamba > jambe) 577 genesis of nasal vowels 600 Southern French (= Midi) ATR alternations of mid vowels (bête [bet] - bêtement [betamä] bêtise [betiz]) data 427s comparison with other closed syllable effects 164, 440 requires additional empty Nuclei 466 are an instance of Licensing 477

are in fact alternations in vowel length that come along in a different coat (Rizzolo 2002) 164, 477, note 345 (555) distribution of nasals: implosion of nasals to velar [η] word-finally 597 Quebec French vowel-zero alternations blocked before clusters 21, 33, 72s, 628, 630 inventory of all types of clusters adjacent to schwa (deriving Charette's 1990 Government Licensing) 126, 128

G Gallo-Romance palatalisations in note 69 (117) Ge'ez (Ethiopian Semitic) minor quotation syllabic spelling system 527 Georgian (Kartvelian) trapped consonants (suspected) note 160 (242) German Old High German High German Consonant Shift general 574-576 stability of homorganic clusters in further evolution (mp > mpf Sumpf vs. lp > If helfen) note 64 (113) Middle High German Open Syllable (Tonic) Lengthening 223 nasals lower preceding vowels (sunne > Sonne) 49 New High German (= modern German) minor quotation "r": apical trill [r] > uvular fricative [χ,κ] 274, 276

Language Index New High German (= modern German) (continued) vowel-zero alternations general 16 German follows the Havlik pattern: schwa can govern 331,469s r-vocalisation 49, (529), 589, note 374 (604) [çMx] alternation (ich-ach Laut) 50 /g/ represented by [ç] in Northern varieties note 380 (609) inalterable vowel length 221 long vowels are +ATR, short vowels are -ATR 477 syllabic consonants alternation of syllabic and nonsyllabic versions of stemfinal consonants in Germanic: C,C2# - C,C2-V 254 complementary distribution of C and aC (synchronic) 257,259, 261 syllabic consonants come into being because a preceding vowel is lost 262 why syllabic consonants occur only after underlying or etymological schwa 612 velar nasal 446, 481-483, 587 morphologically complex wordfinal monster clusters in German (du wirbst) and other Germanic languages (engl sixths) 24, 349, 630 word-final monomorphemic RT#, TT# and RR# clusters: FEN can govern 550, 603 parameterised lateral actorship of schwa: it governs, but does not license 554, —• German velar nasal

847

New High German (= modern German)/ (continued) homorganic CN clusters: the realisation of underlying /CaN/ sequences without schwa (eigen - Eignung) 601 progressive homorganicity of nasals : data 257,602 free variation between aR and syllabic R, with mandatory homorganicity of nasals when schwa is absent (eigen - eigq) 604 resistance of the labial nasal against homorganisation (Atem - atmen) note 375 (604) when a vowel-initial suffix is attached to a /...CaN/-stem (Eignung): data 606 mandatory absence of schwa in /CaNV/ 606, 608 mandatory nonhomorganicity of the nasal in /CaNV/ 606, 607 devoicing of C in /CaNV/ 606,609,613 no devoicing of C in /CaNV/ if CN are homorganic (Redner) 606 extrasyllabicity word-final extrasyllabicity (Jagd [jaakt]) 345, 366s as elsewhere in Germanic, you can get a word-final coronal for free (Haupt, Arzt) 350 Cologne German 216 Austrian German (Salzburg) 50 Germanic —• Common Germanic Gothic Sie vers' Law 121

848

Language Index

Greek (Classical and Modern) minor quotations spirantisation of stops from Classical to Modern Greek under the influence of aspiration 47 Codas weaker than the intervocalic position ? 130 contrast between Coda-Onset and Onset-Onset clusters ? When the strength of the post-Coda depends on the melodic content of the Coda 185 Sievers' Law 121 progressive compensatory lengthening in IE: the laryngeal theory 233 boundary zoo in Modern Greek (Kaisse 1985)405 early generative hocus-pocus (Lightner: sweet - hedonistic, tooth - dental etc. from the same UR) note 214 (333) a language where the left margin of the word plays no role: wordinitial consonants are weak, and initial non-TR clusters exist 91, 353, 405 reduplication (grafo - ge-grafo) 97 H Hindi vowel-zero alternations 16 Hungarian vowel-zero alternations 16 alleged open syllable shortening (kéz kez-ek) and closed syllable lengthening (fa - fát) note 146 (225) no branching structure in Hungarian: TR clusters are two independent Onsets, long vowels two independent Nuclei (Ritter 1995) 13 I Icelandic there are branching Onsets in Icelandic, but only just (Gussmann 2003) 13

Icelandic

(continued)

Open Syllable (Tonic) Lengthening data 533 comparison with other closed syllable effects in other languages 164 among other Tonic Lengthening languages 224 Indo-European typical IE restrictions on word-initial clusters (#TR, but no #RT) 102, 335, 382s, 396-410, — Subject Index/ co-occurrence restrictions in #_ compensatory lengthening (subsequent to the loss of laryngeals), laryngeal theory 233 Sievers' Law in various IE languages and the laryngeal theory 121 equation of IE dialects establishing IE r279 IE comparatistic situation for Common Slavic tbrt < Ci/uRC and trbt < CRi/uC 282 early generative hocus-pocus (Lightner: sweet - hedonistic, tooth dental etc. from the same UR) note 214 (333), — Modern English Iranian minor quotations Sievers' Law 121 Italian minor quotation s+C sequences discussed in Kaye (1992a) 97 a language where words must end in a vowel: interpretation of this parameter in Standard GP: "FEN are not licensed" 18, 266 in CVCV: FEN are not governed 62 in the debate on whether Licensing was about the existence of x-slots and Government about the melodic properties thereof in Standard GP 145

Language Index Italian (continued) Tonic (Open Syllable) Lengthening data and analysis in CVCV 157s comparison with other closed syllable effects in other languages 164 among other Tonic Lengthening languages 224 comparison with progressive compensatory lengthening 227 Radoppiamento Sintattico 237 distribution of the masculine definite article (il treno - lo sporto) 369, 371 1 > j in branching Onsets (platea > piazza) 50, 197

J Japanese minor quotation post-SPE boundary zoo: six different boundaries (McCawley 1968)405 Κ Khalkha Mongolian typology Weight by Position: CVC syllables are light 505 Khosian Zufhoasi: positional neutralisation of the occurrence of consonants 365 Kiowa (Aztec-Tanoan) closed syllable shortening 154,164, 223 Kolami (Dravidian) vowel-zero alternations 16 Koromfe (Niger-Congo, Central Gur) syllabic nasals 256 Kwakwala (native American, AlgonquianWakashan, Northern Wakashan) stress assignment: the sonority of Coda consonants decides whether the syllable is light or heavy description (Zee 1995) 506 analysis in grid-based and moraic approaches 511 analysis in CVCV 521

849

L Lardil (Australian, Pama-Nyungan) typology Weight by Position: CVC syllables are light 505 Latin minor quotation Sievers' Law 121 compensatory lengthening (*kasnus > cl. lat kaanus) 226 evolution of lat VsC > VVC in Old French: compensatory lengthening (testa > tête) 226 IE *-se > infinitive -re: head-initial geminates (velie) 237 Latin stress rule description and situation in the typological picture 505 analysis in grid-based and moraic approaches 507-511 analysis in CVCV 516 evolution of lat j in French: j > 3 in Strong Position {#,C} 565 evolution of lat k,g in French: French palatalisation lat k,g > J,3 / a (gamba > jambe) 577 evolution of lat nasals in Coda position in French: genesis of French nasal vowels 600 evolution of lat 1 in European Portuguese note 385 (616) evolution of lat 1 in Old French: 1-vocalisation in internal Codas (talpa > taupe) 527 evolution of lat kt in Romanian (nocte > noapte) 43 evolution of lat 1 in branching Onsets in Italian (platea > piazza) 50, 197 evolution of Latin obstruents in French 117, 204 emergence of epenthetic [g,k] without Latin etymological basis in Romansch 275 Levantine Arabic —• Arabic Lithuanian assignment of prosodie features (tone) to CVC syllables according to whether the Coda consonant is a sonorant or not (Zee 1995) 506

850

Language

Index

LuGanda (Bantu) minor quotation: compensatory Lengthening note 150 (227) M Malayalam typology Weight by Position: CVC syllables are light 505 Metropolitan French —» French Modern French —• French Middle High German —• German Middle Dutch - * Dutch Modem Czech —> Czech Mon-Khmer (Austro-Asiatic) doubts on the existence of alleged syllabic obstruents (example from the Mon-Khmer language Semai) note 195 (294) Moroccan Arabic —» Arabic Ν Navaho (native American, Athapascan) minor quotation post-SPE boundary zoo: seven different boundaries (Stanley 1969, 1973)405 O Occitan (Romance) minor quotation: nasal vowels 597 Old Church Slavonic OCS Slavic Old Czech Czech Old High German —• German Oowekyala —* Salish

Palestinian Arabic Arabic Pirahä (Brazil, Mura) stress placement is Onset-sensitive 512 Polish trapped consonants synchronic situation trapped consonants in Polish: inventory of words (lexically trapped vs. trapped by a vowel-zero alternation), with their Czech cognates 244,278

Polish/ trapped consonants/ synchronic situation (continued) trapped consonants neither bear stress not count in poetry (trwaé, krwi, wiatr) 245 trapped consonants are transparent to voicing (by contrast to Czech syllabic consonants) 269 real sonorants ([r,l,n,m]) do not devoice word-finally, but does 272s complementary distribution of voiced and voiceless versions of 273 trapped consonants are obstruents and therefore impose their voice value to preceding obstruents 274 diachronic situation Western Slavic comparatistic situation: origin of prevocalised, trapped (Polish) and syllabic (Czech) consonants 278, 286 CS origin of Polish prevocalised liquids (gardlo < gbrdlo) 262, 279 CS tbrt and trtt remain distinct in Polish 281 tbrt and tn>t: Balto-Slavic comparatistic situation 282 secondary vocalisation of syllabic consonants (Czech, Slovak, Polish) 287 vowel-zero alternations Slavic (Polish) vowel-zero alternations: distributional facts 414 front yers palatalise (dzien - dnia) note 254 (415), 450, 452 #RTeC-a with alternating e (mgla mgiel) 108 vocalisation of consonant-final prefixes general 248s before roots with a trapped consonant 250, 265

Language Index Polish/ vowel-zero alternations (continued) contrary to Czech, no diachronically "epenthetic" yers in Polish, i.e. yers without diachronic basis (CS ba-sn-b > cz báseñ, pol baáií) note 271 (421) Gussmann & Kaye's (1993) interpretation: domains and Reduction 485 extrasyllabicity word-final extrasyllabicity (kadra kadr [katr]) 347, 350, 354s word-initial extrasyllabicity (rdza, brak rdzy) 354s, 367 exhaustive data: word-initial consonant clusters in Polish 375, 622 reflexes of Western Slavic vowel lengthening before voiced wordfinal consonants ó-o ([u]-[o]) (bóg - boga) synchronic situation 429, 432, 435 diachronic situation 433 analysis 474, 476 alternation of nasal vowels ^-ç (bl^d - blçdu) synchronic situation 429,432, 435 diachronic situation 434 analysis 474, 476 allophony of nasal vowels ç): implosion of the nasal consonant (—» [w]) in internal and final Codas 598 the palatal nasal ή implosion of the palatal nasal (—» |j]) in internal Codas (koniec konca) 581s alternation [n] - [j] (pan - panstwo) 583 alternation [ji] - |j] (koniec koñca) 584

8 51

Portuguese European minor quotations s+C clusters 97s genesis of nasal vowels 600 distribution of [r] and [r] ([r] in Strong Position, [r] elsewhere) note 384 (616) evolution of lat 1 note 385 (616) Brazilian 1-vocalisation in both internal and final Codas data and description 528 regular extrasyllabic analysis 341 comparison with 1-vocalisation in other languages 50, 529

Quebec French —• French

R Rheto-Romance minor quotation lat 1 in branching Onsets unaffected 50 Romance lat 1 > j in branching Onsets (e.g. platea > it. piazza) 50, 197 lenition of Latin obstruents 116s Strong Position: typical Romance lenition pattern (stability of consonants in Strong Position, lenition in weak positions) vs. the High German Consonant Shift (a little lenition in Strong Position, more lenition in weak positions) 575 genesis of nasal vowels (Portuguese and French) 600 fortition of lat j > (Î3 in Strong Position (e.g. in French: jocu > jeu, sapjam > sache) 565 Romansch (Romance, Switzerland) trapped consonants 275 emergence of epenthetic [g,k] without Latin etymological basis 275 Romanian evolution of lat kt > rom pt 43

852

Language Index

Russian minor quotations vocalisation of yers 415b Slavic metathesis note 184 (282) comparatistic situation of Slavic syllabic consonants (Czech, Slovak, Serbo-Croatian): their Common Slavic origin and their nonsyllabic cognates in Polish and Russian 262 unexpected yer-vocalisation in CS trbt-V (glfctati > rus glotat') note 183 (282), note 186 (284)

S Salish (native American Northwest) Salish family typology of word-initial extrasyllabicity 353 there are no extrasyllabic clusters in Salish note 195 (294), 376 Bella Coola obstruent-only words/ sentences, can obstruents be syllabic? 376 Oowekyala obstruent-only words/ sentences 376 Selayarese (Austronesian) minor quotations Open Syllable (Tonic) Lengthening 224 used by Piggott (1991,1999) in order to argue against Coda Licensing 538 Semai —• Mon-Khmer Semitic —• Akkadian, Arabic, Tiberian Hebrew, Tigrinya Serbo-Croatian minor quotation 1-vocalisation and comparison with 1-vocalisation in other languages 50, 529 comparatistic situation of Slavic syllabic consonants (Czech, Slovak, Serbo-Croatian): their Common Slavic origin and their nonsyllabic cognates in Polish and Russian 262

Serbo-Croatian

(continued)

only [r] can be syllabic, [1] vocalises to [o] in Codas, but to [u] when it ought to be syllabic (i.e. when flanked by consonants): vuk note 175 (262) Slavic (Serbo-Croatian) vowel-zero alternations: distributional facts 414 Slavic pan-Slavic comparatistic situation of Slavic syllabic consonants (Czech, Slovak, Serbo-Croatian): their Common Slavic origin and their non-syllabic cognates in Polish and Russian 262 Slavic vowel-zero alternations: distributional facts 414 (summary 327) result of the vocalisation of yers in different Slavic dialects 415b Slavic metathesis note 184 (282) Common Slavic CS comparatistic situation of Slavic syllabic consonants (Czech, Slovak, Serbo-Croatian): their Common Slavic origin and their non-syllabic cognates in Polish and Russian 262 CS origin of Cz syllabic & Polish prevocalised liquids (gtrdlo > pol gardlo, cz hrdlo) 278s CS tbrt and trU remain distinct in Polish 281 CS tbrt and trbt remain distinct in Baltic and Eastern Slavic 282 tbrt and trbt: Balto-Slavic comparatistic situation 282 evolution of CS tbrt and trbt in Czech 285 evolution of yers (vocalisation or loss) in Old Czech: Havlik's Law description 417, 326 "secondary vocalisation" from the old to the modern languages 418

Language Index Common Slavic/ evolution of yers (continued) non-etymological yers 421 Havlik does not vocalise yers, but empty Nuclei 421b the loss of yers in initial #C C position has caused the appearance of random wordinitial clusters in Polish and Czech: the existing vs. nonexisting #CCs in these languages are a lexical accident carried over from CS 375 Old Church Slavonic OCS OCS origin of Polish prevocalised liquids (gardto < gtrdlo) 279 CS tbrt > OCS : only a graphic merger with CS trbt 281-283 South Slavic minor quotation syllabic consonants in Slavic: Czech, Slovak and South Slavic (save Bulgarian) note 186 (284), - rus glotat') note 183 (282), note 186 (284) ESI Ce/oRC < CS tbrt invalidates the hypothesis that OCS script CRb/tC was born by regular metathesis note 184 (282) syllabic consonants in Slavic: Cz, Slovak and South Slavic (save Bulgarian), against prevocalised reflexes in Eastern Slavic and Polish note 186 (284)

853

Western Slavic minor quotation result of the vocalisation of yers in different Slavic dialects 415b Western Slavic vowel lengthening before voiced word-final consonants 431 Polish reflexes o-ó ([o]-[u]) bóg-boga —• Polish/ reflexes of Western Slavic vowel lengthening Czech reflex: o-ù ([o]-[uu]) dûm - domu —• Modern Czech/ vowel length Western Slavic comparatistic situation: origin of prevocalised, trapped (Polish) and syllabic (Czech) consonants 278, 286 CS tbrt and trbt: Balto-Slavic comparatistic situation 282 secondary vocalisation of syllabic consonants (Czech, Slovak, Polish) 287 Slovak syllabic consonants comparatistic situation of Slavic syllabic consonants (Czech, Slovak, Serbo-Croatian): their Common Slavic origin and their non-syllabic cognates in Polish and Russian 262, note 186 (284) evolution of Old Czech trapped consonants: they have become syllabic in Modern Czech (and Slovak) 285 secondary vocalisation of syllabic consonants (Czech, Slovak, Polish) 287 short vs. long syllabic consonants 300 Western Slavic vowel lengthening before voiced word-final consonants 431 Rhythmic Law (in a sequence of two long vowels, the second shortens) note 302 (461)

854

Language Index

Slovak (continued) vowel-zero alternations Slavic (Slovak) vowel-zero alternations: distributional facts 414, summary 326 more than one vowel alternates with zero 79, 423h Slovak follows the Havlik pattern: schwa can govern 469s Somali (Cushitic) vowel-zero alternations 16 allophony stop (Strong Position) unreleased stop (Coda) - fricative (intervocalic position) 118 inalterable vowel length 221 distribution of nasals: neutralisation of the opposition between [m] and [n] in favour of [n] in Codas 596 Sorbían minor quotation Western Slavic vowel lengthening before voiced word-final consonants 432 South Slavic —» Slavic Spanish Caribbean [r,l] implode to [j] in Codas (carta > cajta) 50

Tiberian Hebrew stability of the first vowel of the word 90 allomorphy of the definite article (ha+C,C,V-haaCiV) 92 spirantisation 561, 569-571 geminate integrity 594 Tigrinya the panel of languages that was used in Standard Government Phonology in order to build the theory of vowel-zero alternations included Tigrinya (as well as French, Moroccan Arabic and Tangale) 79, 461 Turkish vowel-zero alternations 16 Closed Syllable Shortening (meraaki meraktan - merak) data and analysis 154, 223 comparison with other closed syllable effects 164, 440 V Vedic Sievers' Law 121

w,x Western Slavic —• Slavic

T,U Tangale (Chadic) vowel-zero alternations 16 parametric choice regarding Government Licensing when a cluster is followed by a vowel that alternates with zero (Charette 1990): cluster simplification (against nonalternation of the vowel in French) 139, 169 the panel of languages that was used in Standard Government Phonology in order to build the theory of vowel-zero alternations included Tangale (as well as French, Moroccan Arabic and Tigrinya) 79, 461

Y,Z Yawelmani vowel-zero alternations 18 alternations in vowel length Kaye (1990a) reinterprets Closed Syllable Shortening as "shortening before an empty Nucleus" 18, 154 vowels shorten before an empty Nucleus because they fail to be licensed (Yoshida 1993): the internuclear relation was left unnamed by Kaye (1990a) 155 the empty Nucleus analysis eliminates quite some arboreal syllabic structure 170