A Guide to Morphosyntax-Phonology Interface Theories: How Extra-Phonological Information is Treated in Phonology since Trubetzkoy’s Grenzsignale 9783110238631, 9783110238624

This book reviews the history of the interface between morpho-syntax and phonology roughly since World War II. Structura

195 35 7MB

English Pages 900 [901] Year 2010

Report DMCA / Copyright

DOWNLOAD PDF FILE

Table of contents :
Frontmatter
Table of contents – overview
Introduction
Part One – Morpho-syntactic information in phonology: a survey since Trubetzkoy's Grenzsignale
1. The spectrum: what morpho-syntactic information can do to phonology
2. Trubetzkoy's Grenzsignale
3. American structuralism: juncture phonemes
4. Chomsky, Halle & Lukoff (1956)
5. SPE sets the standards for 40 years
6. The life of boundaries in post-SPE times
7. Lexical Phonology
8. Halle & Vergnaud (1987a): selective spell-out and SPE-restoration
9. Kaye (1995): selective spell-out and modification-inhibiting no look-back
10. Prosodic Phonology: on the representational side
11. Optimality Theory
12. Distributed Morphology
Interlude – Modularity
1. Introduction: the relative absence of modularity in interface thinking
2. Modularity and connectionism, mind and brain
3. The modular architecture of the mind: where it comes from
4. The modular architecture of the mind: how it works
5. Modularity of and in language, related systems
6. How modules communicate
Part Two – Lessons from interface theories
1. A guide to the interface jungle
2. Empirical generalisations
3. Issues that are settled
4. Modularity, translation, the diacritic issue and local vs. domain-based intervention: settled in verb, but not in fact
5. Open questions (general)
6. Open questions (procedural)
Conclusion – Intermodular argumentation
Backmatter
Recommend Papers

A Guide to  Morphosyntax-Phonology Interface Theories: How Extra-Phonological Information is Treated in Phonology since Trubetzkoy’s Grenzsignale
 9783110238631, 9783110238624

  • 0 0 0
  • Like this paper and download? You can publish your own PDF file online for free in a few minutes! Sign Up
File loading please wait...
Citation preview

A Guide to Morphosyntax-Phonology Interface Theories

A Guide to Morphosyntax-Phonology Interface Theories How Extra-Phonological Information is Treated in Phonology since Trubetzkoy’s Grenzsignale

by Tobias Scheer

De Gruyter Mouton

ISBN 978-3-11-023862-4 e-ISBN 978-3-11-023863-1 Library of Congress Cataloging-in-Publication Data Scheer, Tobias, 1968⫺ A guide to morphosyntax-phonology interface theories : how extra-phonological information is treated in phonology since Trubetzkoy’s Grenzsignale / by Tobias Scheer. p. cm. Includes bibliographical references and index. ISBN 978-3-11-023862-4 (hardcover : alk. paper) 1. Grammar, Comparative and general ⫺ Morphosyntax. 2. Grammar, Comparative and general ⫺ Phonology. 3. Historical linguistics. I. Title. P241.S416 2010 4151.9⫺dc22 2010040348

Bibliographic information published by the Deutsche Nationalbibliothek The Deutsche Nationalbibliothek lists this publication in the Deutsche Nationalbibliografie; detailed bibliographic data are available in the Internet at http://dnb.d-nb.de. 쑔 2011 Walter de Gruyter GmbH & Co. KG, Berlin/New York Cover image: Creatas/Thinkstock Ra´ko prˇilı´ta´ by Mogdolı´na Printing: Hubert & Co. GmbH & Co. KG, Göttingen ⬁ Printed on acid-free paper Printed in Germany www.degruyter.com

Table of contents – overview §

page

1

Table of contents – detail ................................................................... vii Table of graphic illustrations............................................................. xlv Editorial note ................................................................................... xlvii Foreword The plot, and how to use the book........................................................ il

2

3

Introduction

4

1. Procedural and representational communication with phonology........................................................................................1 2. Functional historiography................................................................3 3. The syntactic frame: minimalist phase theory.................................7 4. Definition of the object of the study..............................................16 5. Trying to get an independent handle on the interface ...................22 6. Deforestation .................................................................................27 7. Structure of the book and of Vol.2 ................................................31

7 11 24 34 42 47

Part One Morpho-syntactic information in phonology: a survey since Trubetzkoy's Grenzsignale 50 55 59 73 81 109 139 215 258 360 450 531

1. The spectrum: what morpho-syntactic information can do to phonology......................................................................................35 2. Trubetzkoy's Grenzsignale ............................................................39 3. American structuralism: juncture phonemes .................................43 4. Chomsky, Halle & Lukoff (1956) .................................................59 5. SPE sets the standards for 40 years ...............................................67 6. The life of boundaries in post-SPE times ......................................93 7. Lexical Phonology.......................................................................123 8. Halle & Vergnaud (1987a): selective spell-out and SPE-restoration............................................................................185 9. Kaye (1995): selective spell-out and modification-inhibiting no look-back................................................................................219 10. Prosodic Phonology: on the representational side.......................301 11. Optimality Theory .......................................................................385 12. Distributed Morphology ..............................................................447

vi

Table of contents – overview

§

page

Interlude Modularity 587 588 600 604 622 649

1. Introduction: the relative absence of modularity in interface thinking........................................................................................497 2. Modularity and connectionism, mind and brain..........................499 3. The modular architecture of the mind: where it comes from ......515 4. The modular architecture of the mind: how it works ..................519 5. Modularity of and in language, related systems..........................535 6. How modules communicate ........................................................557

Part Two Lessons from interface theories 656 657 666 687 719 762

1. 2. 3. 4.

A guide to the interface jungle ....................................................563 Empirical generalisations ............................................................565 Issues that are settled...................................................................573 Modularity, translation, the diacritic issue and local vs. domain-based intervention: settled in verb, but not in fact .........589 5. Open questions (general).............................................................609 6. Open questions (procedural) .......................................................647

841

Conclusion Intermodular argumentation

842 846

1. Trying to get a handle on the interface........................................705 2. Intermodular argumentation ........................................................707

863

References ......................................................................................717

864

Subject index..................................................................................785 Language index..............................................................................838 Index of phenomena ......................................................................843

865 866

Table of contents – detail § 1 2

page Table of graphic illustrations ............................................................ xlv Editorial note................................................................................... xlvii Foreword The plot, and how to use the book ....................................................... il

3

Introduction

4

1. Procedural and representational communication with phonology ... 1 1.1. Cyclic derivation and hashmarks .............................................. 1 1.2. Interface Dualism: both means of talking to the phonology are needed ................................................................................ 2 2. Functional historiography ................................................................ 3 2.1. Anderson's dualistic legacy: structure and process................... 3 2.2. Historiographic cherry-picking................................................. 5 2.3. No "external" history: only scholarly work is used .................. 6 3. The syntactic frame: minimalist phase theory ................................. 7 3.1. The inverted T delineates the scope of the book and serves as a referee ................................................................................ 7 3.2. Interactionism, selective spell-out and no look-back devices (PIC) ............................................................................ 8 3.2.1. When the generative mainstream became interactionist ................................................................... 8 3.2.2. Selective spell-out .......................................................... 9 3.2.3. No look-back devices (the PIC)...................................... 9 3.3. Intermodular argumentation ................................................... 10 3.3.1. The intermodular potential of interactionist phase theory............................................................................ 10 3.3.2. Phase theory is the bridge that forces syntax and phonology to converge ................................................. 11

5 6 7 8 9 10 11 12 13 14 15 16 17 18 19

viii

Table of contents – detail

§

page

20

3.4. Focus on the spell-out mechanism(s?).................................... 12 3.4.1. Minimalist interface orientation: spell-out marshals both morpho-syntax and phonology ............................. 12 3.4.2. The word-spell-out mystery.......................................... 13 3.4.3. We need to know more about the spell-out mechanism.................................................................... 14 4. Definition of the object of the study .............................................. 16 4.1. The book is only about interface theories that follow the inverted T................................................................................ 16 4.2. Interface theories that lie beyond the inverted T model ......... 17 4.2.1. Theories where everything is scrambled: HPSG .......... 17 4.2.2. OT and its connectionist endowment: a programmed trope for scrambling all into one ............. 17 4.2.3. Jackendoff's parallel model: all modules are structure-building ......................................................... 19 4.3. PF, an androgenic intermundia ............................................... 19 4.3.1. The minimalist dustbin: clean syntax, dirty phonology ..................................................................... 19 4.3.2. Syntax, morphology, PF ............................................... 21 4.4. Modularity is the touchstone................................................... 21 5. Trying to get an independent handle on the interface .................... 22 5.1. Intermodular argumentation, history ...................................... 22 5.2. Modularity .............................................................................. 23 5.2.1. Generative grammar deeply roots in modularity, but is often offended..................................................... 23 5.2.2. Modularity in the history of generative grammar: the GB-interlude of syntax-internal (nested) modules ....... 24 5.2.3. The refereeing potential of modularity lies waste ........ 25 5.2.4. Introduction to (Fodorian) modularity.......................... 26 5.2.5. Structuralist and generative modularity........................ 27 6. Deforestation.................................................................................. 27 6.1. The core of Government Phonology: lateral, rather than arboreal syllable structure....................................................... 27 6.2. The lateral project leaves no place for arboreal prosodic constituency ............................................................................ 28 6.3. Recursion and other expected consequences of trees are absent in phonology................................................................ 29 6.4. The lateral project predicts that phonology is non-recursive.. 30

21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46

Table of contents – detail

ix

§

page

47

7. Structure of the book and of Vol.2................................................. 31 7.1. How to access the book: the story, its relation with current syntactic theory and its thematic guide................................... 31 7.2. Vol.2: Direct Interface, One-Channel Translation and their application to CVCV .............................................................. 32

48 49

Part One Morpho-syntactic information in phonology: a survey since Trubetzkoy's Grenzsignale 50

Chapter 1 The spectrum: what morpho-syntactic information can do to phonology

51

1. Boundaries have a triggering, a blocking or no effect ................... 35 2. Blocking and triggering effects: illustration .................................. 36 2.1. Process-blocking boundaries: French gliding......................... 36 2.2. Process-triggering boundaries: obstruent voicing in Puyo Pongo ...................................................................................... 37

52 53 54

55

Chapter 2 Trubetzkoy's Grenzsignale

56

1. The cradle of the functional perspective ........................................ 39 2. Grenzsignale do not contribute anything to the mapping puzzle ... 40 3. Grenzsignale are immaterial and under morpho-syntactic control ............................................................................................ 40

57 58

59

Chapter 3 American structuralism: juncture phonemes

60

1. Introduction.................................................................................... 43 2. Level independence: no morphology in phonology....................... 44 2.1. The orthodox point of view .................................................... 44 2.2. Junctural minimal pairs: night rate vs. nitrate........................ 45 3. Morphology in a phonological guise: Moulton (1947) on German........................................................................................... 46

61 62 63 64

x Table of contents – detail §

page

65

4. "Accidental" coincidence of juncture and morpho-syntactic divisions ......................................................................................... 49 5. Structuralist views on the coincidence of juncture and morphosyntactic divisions.......................................................................... 50 5.1. Defenders of orthodoxy (Hockett, Joos, Trager) and Harris' ambiguous position................................................................. 50 5.2. Opposite view: Pike's grammatical prerequisites ................... 52 6. Whatever suits the analyst: juncture in the middle of morphemes..................................................................................... 54 7. Is there a phonetic correlate of juncture? ....................................... 55 8. Structuralist terminology ............................................................... 57 9. Conclusion: Level Independence seeds modularity and enforces translation....................................................................................... 57

66 67 68 69 70 71 72

73

Chapter 4 Chomsky, Halle & Lukoff (1956)

74

1. The structuralist cover: economy................................................... 59 2. Phonological domains built on juncture distinctions ..................... 60 3. Level Independence abolished: phonology does take morphosyntactic information into account................................................. 61 4. Two for the price of one: multifunctional juncture........................ 62 5. Morpho-syntactic control over juncture restored, and no phonetic correlate........................................................................... 62 6. Privativity, an accidental consequence of economy ...................... 64 7. Cyclic derivation (inside-out interpretation).................................. 64

75 76 77 78 79 80

81

Chapter 5 SPE sets the standards for 40 years

82

1. Introduction.................................................................................... 67 1.1. Interface Dualism.................................................................... 67 1.2. Modular seeds in SPE: the inverted T and translation............ 68 2. Translation in SPE (output: boundaries)........................................ 70 2.1. The inverted T model.............................................................. 70 2.2. Boundaries .............................................................................. 71 2.2.1. Boundaries are [-segment] segments without phonetic correlate ......................................................... 71 2.2.2. Different types of boundaries: +, =, #........................... 72

83 84 85 86 87 88 89

Table of contents – detail

xi

§

page

90

2.3. The general mapping algorithm: boundaries restore (almost) full morpho-syntactic information........................... 73 2.4. Readjustment .......................................................................... 75 2.5. Affix classes and their representational management............. 77 2.5.1. Rule-blocking boundaries (stress-shifting affixes)....... 77 2.5.2. Rule-triggering boundaries (stress-neutral affixes) ...... 78 2.6. Labelled brackets .................................................................... 79 2.6.1. How brackets and labels are used in the phonology: bláckboard vs. black bóard .......................................... 79 2.6.2. Brackets and boundaries are (not) redundant: compensation vs. condensation .................................... 80 2.6.3. Brackets and interactionism.......................................... 83 2.6.4. Modularity and modularity offenders in SPE............... 84 3. The phonological cycle and one single phonology........................ 85 3.1. The phonological cycle........................................................... 85 3.1.1. Cyclic derivation: how it is motivated and how it works ............................................................................ 85 3.1.2. Cycles are defined like boundaries: by major categories (N,V,A)........................................................ 86 3.2. Word-level rules and the unity of phonological computation ............................................................................ 87 3.2.1. Cyclic vs. word level rules ........................................... 87 3.2.2. SPE is representational: class 1 vs. class 2, cyclic vs. word-level rules ...................................................... 89 3.2.3. SPE's representationalism maintains the unity of one single computational system.................................. 90 4. Conclusion ..................................................................................... 90

91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108

109

Chapter 6 The life of boundaries in post-SPE times

110

1. Introduction: overview until the 80s .............................................. 93 2. The mapping puzzle: alas, there are no natural classes of boundaries...................................................................................... 94 3. Boundary mutation rules................................................................ 95

111 112

xii Table of contents – detail §

page

113

4. McCawley (1968): boundaries define domains of rule application...................................................................................... 96 4.1. Cyclic derivation on the grounds of phonological domains ... 96 4.2. Local vs. domain-based intervention – notational variants?... 97 5. Typology and strength of boundaries............................................. 98 5.1. Boundary zoo.......................................................................... 98 5.2. Boundary contrast: minimal pairs and phonological permeability ............................................................................ 99 6. Attempts to restrict boundary abuse and the boundary zoo ......... 100 6.1. Boundaries in an arbitrary rewrite-machinery ...................... 100 6.2. Boundary strength as a diagnostic for boundary abuse ........ 101 6.3. Boundary economy I: Basbøll's general and grounded advice.................................................................................... 102 6.4. Boundary economy II: the + boundary can be dispensed with ....................................................................................... 103 7. Internal and external SPE-revision: Kiparsky (1968-73) and Natural Generative Phonology..................................................... 104 7.1. Introduction........................................................................... 104 7.2. Morpho-phonology in SPE and the 70s................................ 104 7.3. Natural revival of structuralist juncture (abuse) ................... 106 7.4. The elimination of (word) boundaries from P-rules – a case of wishful thinking ..................................................... 108 7.5. The retrenchment of morpho-phonology mechanically reduces the number of boundaries ........................................ 109 7.6. Autosegmental structure: from representational to procedural management of the word boundary .................... 111 8. What exactly is the output of translation, if any?......................... 112 8.1. Growing uncertainty: what kind of intermundia animals are boundaries? ..................................................................... 112 8.2. Boundaries cannot be segments............................................ 114 8.2.1. If not segments, what then? ........................................ 114 8.2.2. Lass (1971): # is [-voice]............................................ 115 8.2.3. Bankruptcy of boundaries and abandon of translation: Pyle (1972) .............................................. 116 8.2.4. When boundaries are bankrupt, the alternative is direct syntax: Pyle, Rotenberg, Hyman, Kenstowicz & Kisseberth ........................................... 118 9. Conclusion ................................................................................... 119

114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137

138

Table of contents – detail

§

xiii page

139

Chapter 7 Lexical Phonology

140

1. Introduction.................................................................................. 123 2. Empirical foundations.................................................................. 124 2.1. Affix classes and affix ordering............................................ 124 2.2. Cross-linguistic reality of affix classes................................. 125 3. Lexical Phonology and the abstractness debate ........................... 126 4. The general architecture of Lexical Phonology ........................... 127 4.1. Interactionism – a new idea in the interface landscape......... 127 4.2. Strata and their procedural order: how to kill two birds with one stone ....................................................................... 128 4.3. Morpheme-specific mini-grammars ..................................... 129 4.3.1. The stratal perspective supposes selective rule application .................................................................. 129 4.3.2. Underapplication is achieved by distinct minigrammars and level ordering ...................................... 130 4.3.3. How to make mini-grammars different but not waterproof: domain assignment.................................. 131 4.4. The general picture: interactionist Lexicon → syntax → postlexical phonology ...................................................... 132 4.5. Praguian segregation: lexical vs. postlexical phonology ...... 133 4.5.1. The birth of postlexical phonology: Rubach (1981)... 133 4.5.2. Praguian segregation: syntax and morphology are different ...................................................................... 135 4.5.3. Morphologically conditioned vs. exceptionless rules ............................................................................ 135 4.5.4. Lexical vs. postlexical phonology: respective conditioning factors .................................................... 136 4.5.5. Cyclic vs. postlexical phonology: no cyclic interpretation of words ............................................... 137 4.6. Definition of interpretational units ....................................... 138 4.6.1. From brackets to strata ............................................... 138 4.6.2. Interactionism reconciles cyclic derivation with modularity................................................................... 138 5. The analysis of affix classes and its proceduralisation ................ 140 6. Rule-blocking boundaries are eliminated altogether.................... 141 6.1. Rule-blocking boundaries translate as level 1 rules.............. 141 6.2. Complete and unintended elimination of boundaries ........... 143

141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165

xiv Table of contents – detail §

page

166

7. Rule-triggering boundaries: brackets and bracket erasure ........... 144 7.1. English nasal cluster simplification ...................................... 144 7.2. Brackets and bracket erasure are needed for a stratal account.................................................................................. 147 7.3. A hybrid representational-procedural theory because of the rule-triggering pattern ..................................................... 149 7.3.1. LP-style brackets undo what was gained by interactionism ............................................................. 149 7.3.2. LP-style brackets and brackets in SPE have got nothing in common..................................................... 150 7.3.3. Brackets are boundaries that are introduced through the back door................................................. 151 7.3.4. A hybrid representational-procedural theory.............. 151 7.4. Bracket erasure ..................................................................... 152 7.4.1. No look-back devices in Lexical Phonology and elsewhere .................................................................... 152 7.4.2. SPE has bracket erasure, but no no look-back effect........................................................................... 153 8. Derived environment effects........................................................ 154 8.1. Properties and illustration of the phenomenon ..................... 154 8.1.1. Phonologically and morphologically derived environments .............................................................. 154 8.1.2. The foundational Finnish case.................................... 154 8.1.3. Non-application of rules to mono-morphemic strings ......................................................................... 155 8.1.4. Sensitivity to derived environments and to affix classes is orthogonal ................................................... 157 8.2. Derived environments are an offspring of the abstractness debate................................................................ 157 8.2.1. Absolute neutralisation and free rides ........................ 157 8.2.2. The Alternation Condition.......................................... 158 8.2.3. "Low-level", "automatic" or "phonetic" rules may apply morpheme-internally......................................... 159 8.2.4. The Revised Alternation Condition: derived environments enter the scene...................................... 160

167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187

Table of contents – detail

xv

§

page

188

8.3. Solution 1: the Strict Cycle Condition (SCC)....................... 161 8.3.1. Mascaró's SCC has got nothing to do with derived environments .............................................................. 161 8.3.2. Kiparsky (in fact Halle) adds derived environments to Mascaró's SCC ....................................................... 162 8.3.3. Deriving the SCC from the Elsewhere Condition ...... 164 8.3.4. There are rules that apply in the Lexicon but affect underived items .......................................................... 165 8.3.5. Structure-building vs. structure-changing rules.......... 165 8.3.6. Cyclicity as a property of strata vs. post-cyclic lexical rules................................................................. 166 8.3.7. The SCC-K is void of empirical content .................... 168 8.4. Solution 2: derived environments are made a lexical contrast (Kiparsky 1993) ...................................................... 169 8.4.1. Back to where we started: Kiparsky declares the bankruptcy of SCC-K ................................................. 169 8.4.2. Different lexical representations for the same segment....................................................................... 170 8.4.3. Why NDEB processes must be obligatory and neutralising ................................................................. 171 8.4.4. Posterity of Kiparsky's lexical solution ...................... 172 8.5. Solution 3: bracket-sensitive rules (Mohanan 1982) ............ 172 8.5.1. Bracket-sensitive rules can do all derived environment effects .................................................... 172 8.5.2. Brackets and SCC-K are direct competitors – but not in the literature...................................................... 174 8.6. Solution 4: derived environment effects are non-linguistic in nature (Anderson 1981) .................................................... 175 8.6.1. Fultonians: speakers use extra-linguistic evidence (spelling) in order to establish underlying forms........ 175 8.6.2. Explaining the genesis of derived environment effects does not exonerate from coming up with a synchronic scenario .................................................... 176 8.7. Conclusion ............................................................................ 177 8.7.1. A new phenomenon that is poorly understood ........... 177 8.7.2. Kiparsky's solutions miss the facts or the point.......... 178 8.7.3. Mohanan's bracket-sensitive rules – a misjudged option.......................................................................... 179

189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206

207 208 209 210

xvi

Table of contents – detail

§

page

211

9. Conclusion ................................................................................... 179 9.1. Interactionism and multiple mini-grammars......................... 179 9.2. Unprecedented proceduralisation of the interface ................ 181 9.3. Representational communication and the pernicious SCC-K................................................................................... 181

212 213 214

215

Chapter 8 Halle & Vergnaud (1987a): selective spell-out and SPErestoration

216

1. Introduction: a hermaphrodite theory with a new idea ................ 185 1.1. Unseating Lexical Phonology and restoring SPE ................. 185 1.2. Relations with Lexical Phonology........................................ 185 1.3. Relations with SPE ............................................................... 187 1.4. New ideas in interface thinking: selective spell-out and interpretation-triggering affixes............................................ 187 1.5. Relevant literature and roadmap ........................................... 188 2. Anti-interactionism ...................................................................... 189 2.1. Restoration of the inverted T: all concatenation before all interpretation......................................................................... 189 2.2. Interactionism does not imply the Lexicon and is not incompatible with the inverted T .......................................... 191 3. Selective spell-out........................................................................ 191 3.1. A new idea: affix-triggered interpretation ............................ 191 3.2. Interpretational relevance of affixes percolates to their node....................................................................................... 192 3.3. Underapplication is achieved by selective spell-out of nodes..................................................................................... 194 3.4. The management of English stress ....................................... 195 3.4.1. Selective spell-out analysis of the párent - paréntal contrast ....................................................................... 195 3.4.2. Stress copy: storing before erasing............................. 196 4. The non-interactionist architecture .............................................. 197 4.1. Cyclic vs. word-level (non-cyclic) rules............................... 197 4.2. Multiple mini-grammars: morpheme- and chunk-specific phonologies........................................................................... 198 4.3. The general architecture........................................................ 200 4.4. Terminological pitfalls.......................................................... 201 4.5. No look-back devices............................................................ 203

217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237

Table of contents – detail

xvii

§

page

238

4.6. Distinct pre- vs. post-word phonology yes, Praguian segregation no....................................................................... 204 4.7. The word as an autonomous phonological unit .................... 206 4.7.1. The word is sealed – an insuperable barrier for some processes ........................................................... 206 4.7.2. Selective impact of no look-back?.............................. 207 4.8. Interpretational units............................................................. 208 5. Anti-interactionist ammunition: bracketing paradoxes................ 209 5.1. Affix ordering turns out to be wrong: bracketing paradoxes .............................................................................. 209 5.2. Dual membership, optionality, overgeneration..................... 210 5.3. Halle & Vergnaud's analysis of bracketing paradoxes ......... 211 5.4. Affix stacking generalisations as selectional restrictions or parsing-based (Fabb 1988, Hay 2002).............................. 212 6. Empirical coverage of Halle & Vergnaud's selective spell-out ... 213 6.1. Analysis of the rule-blocking pattern (level 1 rules) ............ 213 6.2. No solution for the rule-triggering pattern (level 2 rules)..... 214 7. Stratal vs. non-interactionist architecture: two testing grounds ... 215 7.1. Syntactic information is or is not available when words are build ................................................................................ 215 7.2. Phonology-free syntax: is morphological concatenation sensitive to derived phonological properties?....................... 216 8. Conclusion ................................................................................... 216 8.1. SPE with Lexical Phonology freckles .................................. 216 8.2. Selective spell-out is a groundbreaking idea ........................ 217 8.3. Distinct computational systems yes – but which ones? ........ 218

239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257

258

Chapter 9 Kaye (1995): selective spell-out and modification-inhibiting no look-back

259

1. Introduction.................................................................................. 219 1.1. Editorial note ........................................................................ 219 1.2. Roadmap ............................................................................... 220 2. Setting the scene: Kaye (1989) .................................................... 220 2.1. Phonology exists because it enhances parsing...................... 220 2.2. Perception-oriented views of phonology and the interface... 222 2.3. Typical boundary detectors................................................... 222

260 261 262 263 264 265

xviii

Table of contents – detail

§

page

266

3. Domain structure: how it is created, what it represents and how it works................................................................................. 223 3.1. The concat- and the φ-function............................................. 223 3.1.1. General properties ...................................................... 223 3.1.2. Computation in Government Phonology.................... 223 3.1.3. Important properties of the φ-function for Kaye's interface theory........................................................... 225 3.2. Domain structure is created by interleaved concat and φ ..... 225 3.3. Analytic vs. non-analytic ...................................................... 227 3.4. Domain structure is the result of selective spell-out............. 227 3.5. Interpretation prior to concatenation: domain structure generates more than spell-out can create .............................. 228 3.6. Domain structure is interactionist, brackets are only shorthand............................................................................... 229 3.7. Kaye's procedural-only approach and morpho-phonology ... 230 4. Selective spell-out: Kaye's vs. Halle & Vergnaud's implementation ............................................................................ 231 4.1. Introduction........................................................................... 231 4.2. Underapplication is achieved by no look-back..................... 231 4.3. No automatic spell-out of roots, but systematic spell-out at the word-level ................................................................... 233 4.4. Who is interpretation-triggering – class 1 or class 2 affixes?.................................................................................. 234 4.5. Interpretation-triggering affixes: spell-out of the sister vs. their own node................................................................. 235 4.6. Morpheme- and chunk-specific phonologies........................ 236 4.7. Derived environments are a separate issue ........................... 237 4.8. Cyclic interpretation of words?............................................. 238 4.9. Summary: two ways of doing selective spell-out ................. 238 5. No look-back devices: implementations since 1973.................... 239 5.1. Like lexicalism, no look-back is born in the early 70s as an overgeneration-killer........................................................ 239 5.2. Chomsky's (1973) Strict Cycle Condition: you need to use new material ................................................................... 240 5.3. Application to phonology: Kean (1974) and Mascaró (1976)..................................................................... 241 5.4. Halle/Kiparsky's SCC-K: scrambling with derived environments......................................................................... 242 5.5. Mohanan's bracket erasure: indirect bearing on rules........... 243

267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292

Table of contents – detail

xix

§

page

293

5.6. Modification-inhibiting no look-back: first timid steps in the 70s and 80s ................................................................. 244 5.6.1. Early formulations in phonology and syntax.............. 244 5.6.2. Phonology I: stress and the Free Element Condition .................................................................... 244 5.6.3. The FEC is weak: process-specificity, parameterisation and restriction to structure that is absent from the lexicon........................................... 246 5.6.4. Phonology II: structure preservation in syllabification ............................................................. 247 5.6.5. Syntax: Riemsdijk's (1978) Head Constraint.............. 247 5.7. Modification-inhibiting no look-back in Kaye's system....... 250 5.7.1. Kaye calls on SCC-M, but applies something else..... 250 5.7.2. Kaye's modification-inhibiting no look-back ............. 251 5.7.3. Chomsky's "spell-out and forget" is too strong for phonology: "don't undo" and process-specific no look-back .................................................................... 253 5.7.4. Morpheme-specific phonologies and selective spell-out do the same job and are therefore mutually exclusive...................................................... 255 5.8. On the (modern) syntactic side: derivation by phase and Phase Impenetrability ........................................................... 255 5.8.1. Interactionism is enforced by the minimalist concern for economy of cognitive resources (active memory)......................................................... 255 5.8.2. Phase Impenetrability is the instrument of active memory economy ....................................................... 259 5.8.3. The old and the new: a memory keeper and/or diacritic marking needed?........................................... 260 5.8.4. Ancestors of multiple and selective spell-out: Bresnan (1971) ........................................................... 261 5.9. Conclusion ............................................................................ 264 6. Empirical coverage: Kaye's system and affix class-based phenomena ................................................................................... 264 6.1. The empirical testing ground: five English affix classbased phenomena.................................................................. 264

294 295 296

297 298 299 300 301 302

303

304 305

306 307 308 309 310 311

xx Table of contents – detail §

page

312

6.2. The rule-blocking pattern (level 1 rules) .............................. 266 6.2.1. Modification-inhibiting no look-back achieves underapplication at the outer cycle............................. 266 6.2.2. Nasal assimilation requires independent spell-out of the affix .................................................................. 267 6.2.3. Anti-affix ordering items cannot be done: stress violates no look-back.................................................. 269 6.3. Spell-out of terminals: the problem and a possible solution ................................................................................. 270 6.3.1. The problem: independent spell-out of affixes prior to their being merged.................................................. 270 6.3.2. Morphological adjuncts, counter-cyclic merger and intermodular predictions............................................. 270 6.3.3. Benefits: bracketing paradoxes, category selection, double affixation......................................................... 271 6.3.4. Procedural first ........................................................... 273 6.4. The rule-triggering pattern (level 2 rules)............................. 274 6.4.1. Underapplication at the inner cycle ............................ 274 6.4.2. Kaye's solution is like Mohanan's, but respects modularity................................................................... 275 6.4.3. Why string-final /gN/ and /mn/ do not survive computation ................................................................ 276 6.4.4. Prediction: processes without additional condition cannot instantiate the rule-triggering pattern.............. 276 6.4.5. Kaye's analysis predicts the word-final/class 2 disjunction .................................................................. 277 6.5. Conclusion ............................................................................ 277 7. Phonological consequences of domains and their erosion........... 278 7.1. Empty nuclei that are string-final upon computation ........... 278 7.1.1. Final empty nuclei in Government Phonology........... 278 7.1.2. From word-final to domain-final empty nuclei .......... 279 7.1.3. The well-formedness of final empty nuclei is carried over to outer domains ..................................... 280 7.1.4. Computation is right-to-left: string-final nuclei are phase-initial ................................................................ 280 7.2. Consequences of domain structure for stress and vowel reduction ............................................................................... 281 7.2.1. Dialectal and idiolectal variation due to variable domain structure ......................................................... 281

313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335

Table of contents – detail

xxi

§

page

336

7.2.2. Following SPE: "old" stress protects against vowel reduction .......................................................... 282 7.2.3. How to detect the difference between [[X] [Y]] and [[X] Y] ................................................................. 283 7.2.4. Chunk-specific phonologies: like SPE, Kaye provides for a specific word-level phonology ............ 283 7.3. Diachronic erosion of domain structure................................ 284 8. Parsing cues ................................................................................. 286 8.1. Theory-independent parsing cues: knowledge of morpheme structure .............................................................. 286 8.2. Theory-dependent parsing cues (in English) ........................ 287 8.2.1. Empty nuclei detected by phonology ......................... 287 8.2.2. Morphological interpretation of empty nuclei............ 289 8.3. When phonology is useless................................................... 290 9. Lexical access and the organisation of the lexicon ...................... 291 9.1. Lexical entries are grouped according to phonological structure ................................................................................ 291 9.2. Possible lexical entries are defined by phonology................ 292 9.2.1. How speakers decide that blick, but not lbick, is a possible word.............................................................. 292 9.2.2. The full addressing space is created, including "empty" slots .............................................................. 292 9.3. Phonology defines lexical access: look-up vs. compute....... 293 9.3.1. Introduction ................................................................ 293 9.3.2. Look-up I: related keep - kept vs. unrelated table - house ............................................................... 294 9.3.3. Look-up II: phonologically similar keep - kept vs. regular suppletion go - went ....................................... 295 9.3.4. Computation: when parsing cues are available (peeped) ..................................................................... 296 9.3.5. How parsing is done with hidden morphology (stepped) .................................................................... 297 9.3.6. The four identification patterns are more or less costly .......................................................................... 297 9.3.7. Why is direct look-up not generalised? ...................... 297 10. Conclusion ................................................................................. 298

337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359

xxii Table of contents – detail §

page

360

Chapter 10 Prosodic Phonology: on the representational side

361

1. Overview: autosegmentalised boundaries and fresh data ............ 301 2. The roots of Prosodic Phonology................................................. 303 2.1. Selkirk adapts Liberman & Prince's strong/weak arboreal structure ................................................................................ 303 2.2. A second strand that became mainstream: Nespor & Vogel.................................................................... 304 3. From boundaries to domains: a historical choice that has gone almost unnoticed .......................................................................... 305 3.1. Boundaries are diacritic and local......................................... 305 3.2. The elimination of boundaries in Lexical Phonology remained unreflected in Prosodic Phonology ....................... 306 3.3. Prosodic Phonology is a child of autosegmentalism............. 307 3.4. The (non-)discussion of boundaries in Prosodic Phonology ............................................................................. 308 3.4.1. Boundaries were not an issue anymore for Nespor & Vogel (1986)........................................................... 308 3.4.2. Looking for anti-boundary arguments........................ 309 3.4.3. References that do not contain any argument against boundaries ...................................................... 310 3.4.4. The diacritic argument................................................ 312 3.4.5. Domains have an independent motivation: stress, rhythm and musical properties ................................... 315 3.4.6. The idea of unified representations was abandoned by Selkirk herself........................................................ 316 3.4.7. Summary..................................................................... 317 4. The heart of Prosodic Phonology: Indirect Reference and its consequences................................................................................ 318 4.1. Introduction........................................................................... 318 4.2. The buffer, its construction workers and how it unloads its goods................................................................................ 319 4.2.1. Mapping rules and the black box................................ 319 4.2.2. Mapping is done in modular no man's land................ 320 4.2.3. The layers of the Prosodic Hierarchy ......................... 321

362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382

Table of contents – detail

xxiii

§

page

383

4.2.4. Geometric properties of the Prosodic Hierarchy: the Strict Layer Hypothesis ........................................ 322 4.2.5. Three ways of making reference to the Prosodic Hierarchy .................................................................... 324 4.3. The old and the new: mapping rules and Indirect Reference .............................................................................. 324 5. Mapping: its mechanics, its evolution and our understanding thereof .......................................................................................... 326 5.1. Introduction: the mapping puzzle (again) ............................. 326 5.2. Mapping and its inflational evolution................................... 328 5.2.1. Early mapping in Selkirk's work until her 1984 book ............................................................................ 328 5.2.2. Parametric variation of ω and φ in Nespor & Vogel (1986)............................................................... 329 5.2.3. More variation for the phonological phrase................ 330 5.2.4. Cross-linguistic atomisation of mapping.................... 331 5.2.5. The mapping puzzle is sometimes hidden by the clean Prosodic Hierarchy............................................ 332 5.3. What is the morpho-syntactic rationale behind mapping?.... 334 5.3.1. Selkirk (1986) puts to use the technology of the 80s: X-bar-based mapping.......................................... 334 5.3.2. End-based mapping does not solve the mapping puzzle either................................................................ 335 5.3.3. Boundaries are the relevant descriptive currency....... 337 5.4. Phonology can see morpho-syntactic structure, but not its labels..................................................................................... 337 6. Closer inspection of the buffer (Prosodic Hierarchy): what it is and what it is not ...................................................................... 338 6.1. The only purpose of the Prosodic Hierarchy is the storage of morpho-syntactic information .......................................... 338 6.2. The buffer does not include syllables and feet...................... 340 6.3. The buffer is a diacritic – an autosegmental diacritic ........... 342 6.3.1. Prosodic Phonology lays claim to boundaries: they are the old buffer, prosodic domains are the modern buffer ............................................................. 342 6.3.2. The buffer and SPE-type boundaries share all properties .................................................................... 343 6.3.3. What counts as a diacritic? ......................................... 344

384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403

404 405

xxiv Table of contents – detail §

page

406

7. Good and bad reasons for Indirect Reference .............................. 345 7.1. Direct syntax vs. Prosodic Phonology .................................. 345 7.1.1. Two approaches, their competition and their evolution..................................................................... 345 7.1.2. Is the buffer useless and redundant? A real good motivation is needed................................................... 347 7.2. A good reason: modularity ................................................... 347 7.2.1. Introduction: different modules do not speak the same language ............................................................ 347 7.2.2. Phonology-free syntax................................................ 347 7.2.3. Phonology and morpho-syntax do not speak the same language – hence communication requires translation ................................................................... 351 7.2.4. Nobody makes the modular argument in order to sustain Indirect Reference .......................................... 352 7.2.5. Modularity and Level Independence .......................... 353 7.3. A bad reason: non-isomorphism ........................................... 354 7.3.1. Non-isomorphism in the Prosodic Phonology literature...................................................................... 354 7.3.2. The phenomenon: cat-rat-cheese, phonology over sentences..................................................................... 355 7.3.3. Domain abuse I: there is no argument when phonology refers to boundaries instead of domains ... 357 7.3.4. Domain Abuse II: theoretical units are confused with descriptive categories ......................................... 358 7.3.5. Is prosodic phrasing sensitive to the length of the string? ......................................................................... 359 7.4. Conclusion ............................................................................ 361 8. Relations with Lexical Phonology and the metrical grid ............. 361 8.1. Introduction........................................................................... 361 8.2. The metrical grid................................................................... 362 8.2.1. Selkirk (1984): mapping modifies the (pre-existing) grid...................................................... 362 8.2.2. Selkirk (1986): the grid is born from mapping on the grounds of prosodic constituency ......................... 364 8.2.3. Alignment on Nespor & Vogel's peaceful (and modular) coexistence.................................................. 364

407 408 409 410 411 412 413

414 415 416 417 418 419 420 421 422 423 424 425 426 427 428

Table of contents – detail

xxv

§

page

429

8.3. Lexical Phonology I: conflict with Prosodic Phonology ...... 366 8.3.1. No concurrence above the word level ........................ 366 8.3.2. Hayes (1989 [1984]): Prosodic Phonology above, Lexical Phonology below the word level ................... 366 8.3.3. Selkirk (1984): Lexical Phonology is redundant and has to go............................................................... 367 8.3.4. Inkelas (1990): Prosodic Phonology with the empty shell of Lexical Phonology .............................. 368 8.3.5. Lexical Phonology does not violate Indirect Reference.................................................................... 370 8.4. Lexical Phonology II: peaceful coexistence (soft version)... 371 8.4.1. How labour is supposed to be divided: "direct reference to morphological structure" ........................ 371 8.4.2. There is no natural division of rules into "purely phonological" vs. "morpho-phonological" ................. 372 8.4.3. Two examples............................................................. 373 8.4.4. Peaceful coexistence supposes split mapping ............ 374 8.5. Lexical Phonology III: peaceful coexistence (radical version) ................................................................................. 374 8.6. Conclusion ............................................................................ 377 9. Prosodic Morphology................................................................... 378 9.1. Representational and non-representational incarnations ...... 378 9.2. Representational period: prosodic morphemes are units of the Prosodic Hierarchy ..................................................... 378 9.2.1. A typical example: reduplication................................ 378 9.2.2. Another typical example: (Semitic) templatic morphology................................................................. 379 9.2.3. Moras, syllables and feet do not carry any morphosyntactic information .................................................. 380 9.3. Generalized Template Theory: morphology marshals phonology ............................................................................. 381 10. Conclusion: translation yes, buffer no ....................................... 382

430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449

xxvi

Table of contents – detail

§

page

450

Chapter 11 Optimality Theory

451

1. Setting the scene .......................................................................... 385 1.1. OT makes no contribution on the representational side........ 385 1.2. Representations marshalled by constraints ........................... 385 1.3. Anti-cyclicity and the serial offspring of Lexical Phonology ............................................................................. 386 2. Adaptation of Prosodic Phonology to the constraint-based environment ................................................................................. 388 2.1. Introduction........................................................................... 388 2.2. Constraint-based instead of rule-based mapping .................. 388 2.2.1. Mapping understood as the coincidence of constituent edges: ALIGN ............................................ 388 2.2.2. Parallel mapping: translation and reference to prosodic constituency are conflated ........................... 389 2.2.3. Parametric variation: interaction of ALIGN and WRAP .......................................................................... 390 2.3. The Strict Layer Hypothesis made less strict........................ 391 2.4. Phase-based mapping............................................................ 392 2.5. Mapping: focus on new factors, but the puzzle is the same as before....................................................................... 393 3. Cyclic derivation and its relation with phonology ....................... 394 3.1. Anti-cyclicity in OT and elsewhere: cutting off deep generative roots..................................................................... 394 3.1.1. Cyclic spell-out is a bone of contention: OT vs. generative grammar .................................................... 394 3.1.2. Anti-cyclic voices from other quarters ....................... 395 3.1.3. The stake is high: inside-out interpretation ................ 396 3.2. OT could respect modular contours – but it does not ........... 396 3.2.1. Cyclic derivation and phonological computation are entirely independent as long as phonology is phonology proper........................................................ 396 3.2.2. Modularity is not an issue for classical incarnations of OT .......................................................................... 398 3.2.3. Cyclic derivation, but morphology and phonology merged in the same constraint chamber (Wolf 2008) ................................................................ 399

452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470

471 472

Table of contents – detail

xxvii

§

page

473

4. Morpheme-specific mini-grammars in OT .................................. 400 4.1. Introduction........................................................................... 400 4.2. Opacity killers, cyclicity killers and their relationship ......... 400 4.3. Morpheme-specific mini-phonologies: parallel vs. reranked incarnations............................................................ 401 5. Parallel mini-grammars................................................................ 403 5.1. Co-phonologies: two constraint rankings ............................. 403 5.1.1. Lexical material selects a specific computational system......................................................................... 403 5.1.2. How affix class-based phenomena could be analysed ...................................................................... 403 5.1.3. Are co-phonologies a version of Halle & Vergnaud (1987a)? ..................................................... 404 5.2. Indexed constraints: two grammars in the same constraint ranking ................................................................. 405 6. Reranked mini-grammars: Stratal OT, DOT................................ 407 6.1. Serial vs. parallel solutions ................................................... 407 6.1.1. Parallel OT couched in the stratal architecture of Lexical Phonology...................................................... 407 6.1.2. The critical contrast with parallel implementations is not discussed........................................................... 407 6.1.3. Indexed constraints, but not co-phonologies, are perceived as a competitor ........................................... 408 6.2. Stratal OT is more than just an OTed version of Lexical Phonology ............................................................................. 409 6.2.1. Lexical Phonology anew: unhorsing the SPE heritage ....................................................................... 409 6.2.2. Inflational access to morpho-syntactic information is a concern in Stratal OT ........................................... 410 6.2.3. The solution is the same as before: Indirect Reference and peaceful coexistence ........................... 411 6.3. How different can mini-grammars be? ................................. 412 7. Other cyclicity killers................................................................... 413 7.1. Interface constraints.............................................................. 413 7.1.1. Back to direct syntax .................................................. 413 7.1.2. Back to SPE-type morphological diacritics................ 414

474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496

xxviii

Table of contents – detail

§

page

497

7.2. Analogy (Output-Output correspondence) ........................... 415 7.2.1. OO correspondence was designed as a cyclicity killer............................................................................ 415 7.2.2. OO is not a cyclicity killer by itself............................ 416 7.3. Enriched representations (van Oostendorp).......................... 417 8. More direct syntax: representational continuity between morphology and phonology ......................................................... 418 8.1. Introduction........................................................................... 418 8.2. Coloured Containment (van Oostendorp 2006a) .................. 418 8.2.1. Faithfulness between morphology and phonology ..... 418 8.2.2. Representational identification of the old and the new ............................................................................. 419 8.2.3. Information transmission: representational vs. procedural ................................................................... 420 8.2.4. Independent representations marshal GEN and afford to be coloured .................................................. 420 8.2.5. Faithfulness between morphological and phonological structure ................................................ 421 8.2.6. Coloured Containment applied to derived environment effects .................................................... 423 8.2.7. Anti-Lexical Phonology: the interface representationalised .................................................... 425 8.2.8. Mapping: a second means of talking to the phonology ................................................................... 425 8.3. Sign-Based Morphology (Orgun 1996a) .............................. 426 8.3.1. Monostratal HPSG-style representations where syntactic, semantic and phonological information is scrambled................................................................ 426 8.3.2. Cyclic effects in SBM ................................................ 428 8.3.3. There is no interface if all is one and the same thing............................................................................ 429 9. Derived environment effects in OT ............................................. 430 9.1. Introduction........................................................................... 430 9.2. Coloured Containment and Constraint Conjunction (Łubowicz)........................................................................... 431 9.3. Comparative Markedness (McCarthy) and analogy (Burzio)................................................................................ 432 9.4. Direct syntax (interface constraints): Root Faithfulness (Anttila)................................................................................. 434

498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513

514 515 516 517 518 519 520

Table of contents – detail

xxix

§

page

521

9.5. Revival of the Elsewhere Condition (Cho, Iverson)............. 434 9.6. Co-phonologies (Yu) ............................................................ 435 10. OT is a strong modularity-offender: violations are in-built ....... 435 10.1. Modularity yes or no – this is the question, however rarely addressed in the OT literature................................. 435 10.2. Direct Syntax in OT is regular and uncontradicted .......... 438 10.3. Parallel mapping puts the Translator's Office in the phonology ......................................................................... 439 10.4. Scrambling: morpho-phonological contours are blurred .. 439 10.5. Radical scrambling: one single constraint ranking for morpho-syntax, semantics and phonology........................ 441 10.6. Two souls are dwelling in OT: generative and connectionist..................................................................... 442 11. Conclusion ................................................................................. 444

522 523 524 525 526 527 528 529 530

531

Chapter 12 Distributed Morphology

532

1. Introduction.................................................................................. 447 2. Setting the scene: Distributed Morphology vs. Lexical Phonology .................................................................................... 448 2.1. From Halle & Vergnaud (1987a) to Distributed Morphology: against the two-place approach....................... 448 2.2. The single engine approach is agnostic with respect to the phonological interpretation of the string (morphemeand chunk-specific parsing) .................................................. 449 2.3. The general architecture of Distributed Morphology ........... 450 2.4. The unity of syntax and morphology under debate .............. 452 2.4.1. Specific morphological operations: does DM live up to its ambition? ...................................................... 452 2.4.2. Voices that argue for the traditional stance: syntax ≠ morphology.................................................. 452 2.5. Earlier versions of "No escape from syntax": Selkirk (1984) and Inkelas (1990).................................................... 453 3. Direct merge and opacity ............................................................. 455 3.1. Introduction: DM also looks at interpretative effects at LF.. 455

533 534 535

536 537 538 539 540 541 542

xxx Table of contents – detail §

page

543

3.2. In DM, opacity is due to direct merge .................................. 456 3.2.1. Direct merge to a root produces opacity, merge to an xP guarantees transparency.................................... 456 3.2.2. An example of direct-merge-created PF and LF opacity: cómparable vs. compárable.......................... 457 3.2.3. The origin of the idea that direct merge causes opacity: idiosyncrasy is a property of "lower" items in the tree .......................................................... 458 3.3. Marvin's (2002) DM-analysis of condensation vs. compensation ........................................................................ 460 3.3.1. SPE's classical take: two different suffixes ................ 460 3.3.2. The SPE analysis is empirically flawed: Halle & Kenstowicz (1991) admit a lexical conditioning........ 461 3.3.3. *Transport-ate: intermediate derivational forms that happen not to exist as words................................ 462 3.3.4. Direct (transportation) vs. indirect (condensation) merge of -ate............................................................... 462 3.3.5. All xPs are phase heads, direct merge is the antilexicalist way of encoding lexicalist observations...... 464 3.3.6. Direct merge may, but does not need to produce opacity ........................................................................ 464 3.4. Phase Impenetrability à la carte? .......................................... 465 3.4.1. The strong DM claim that all xPs are spelled out imposes process-specific Phase Impenetrability ........ 465 3.4.2. We know that stress is a strange guy anyway............. 466 3.5. All xPs are spelled out vs. selective spell-out plus the PIC .. 467 3.5.1. Affix class-based phenomena in DM ......................... 467 3.5.2. Underapplication cannot be done when all xPs are spelled out................................................................... 468 3.5.3. Opposite ways to go: distinct representations (DM) vs. distinct computation (Kaye, Halle & Vergnaud) .. 469 3.5.4. A direct merge analysis for affix class-based phenomena is not viable ............................................. 471 3.5.5. Analysing cómparable vs. compárable with selective spell-out ....................................................... 472 3.5.6. Conclusion.................................................................. 473

544 545 546

547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563

Table of contents – detail

xxxi

§

page

564

3.6. Articulation of semantic and phonological opacity .............. 473 3.6.1. All logical possibilities occur ..................................... 473 3.6.2. Neither the PIC nor direct merge can account for all patterns .................................................................. 475 3.6.3. Independent LF and PF phases? ................................. 475 3.7. Conclusion: predictable vs. unpredictable opacity and associated analyses ............................................................... 476 4. Anti-lexicalism is an orientation that allows for lexicalist analyses........................................................................................ 477 4.1. The phonological side of the (anti-)lexicalist issue .............. 477 4.2. A third player: allomorphy, suppletion................................. 479 4.3. Distributed Morphology relies on the usual "phonological similarity" ............................................................................. 480 4.4. There is no semantic effect if cómparable and compárable are made of two independent Vocabulary Items ..................................................................................... 481 5. PF movement ............................................................................... 483 5.1. Syntactic motivation ............................................................. 483 5.2. Phonological motivation....................................................... 484 5.2.1. Cases where phonology impacts syntax ..................... 484 5.2.2. Phonology-internal triggers: Piggott & Newell on Ojibwa ........................................................................ 485 5.2.3. Phonologically motivated allomorphy: Lowenstamm on French ............................................. 487 5.3. Overgeneration, direct syntax, modularity and the creation of two distinct computational systems .................... 491 6. Conclusion ................................................................................... 493 6.1. Direct merge vs. selective spell-out: a geometric, rather than a computational solution ............................................... 493 6.2. Spell-out at every xP stands in the way ................................ 494 6.3. Unifying ambitions and specific tools for morphology ........ 494 6.4. Two specificities: LF and no proposal regarding representational communication ........................................... 495

565 566 567 568 569 570 571 572 573

574 575 576 577 578 579 580 581 582 583 584 585

xxxii Table of contents – detail § 586

page

Interlude Modularity

587

Chapter 1 Introduction: the relative absence of modularity in interface thinking ........................................................................... 497

588

Chapter 2 Modularity and connectionism, mind and brain

589

1. Monism vs. dualism, symbolic vs. non-symbolic representations ............................................................................. 499 1.1. Levels of representation in the standard cognitive model..... 499 1.2. A language of thought: symbolic vs. anti-symbolic views of cognition........................................................................... 500 1.3. What would adult science look like without symbols?......... 502 2. Connectionism and its representatives in linguistics ................... 505 2.1. The symbolic front line and its roots in Cognitive Science .. 505 2.2. How neural networks work................................................... 506 2.3. No distinction between storage and computation (the rule/list fallacy)..................................................................... 507 2.4. All-purpose parallel vs. specialised step-by-step computation .......................................................................... 508 2.5. What it all comes down to: connectionist computation is content-free ........................................................................... 510 3. Conclusion: peaceful coexistence at first, but not for long.......... 510

590 591 592 593 594 595 596 597 598 599

600

Chapter 3 The modular architecture of the mind: where it comes from

601

1. The brain as a set of functional units: F-J Gall's early 19th century phrenology ...................................................................... 515 2. Independent faculties, their correlation with size and the skull bone.............................................................................................. 516 3. Faculty psychology married with computation theory (von Neumann - Turing) .............................................................. 517

602 603

Table of contents – detail

§

xxxiii page

604

Chapter 4 The modular architecture of the mind: how it works

605

1. Higher and lower cognitive functions, modules and the central system............................................................................... 519 2. How much of the mind is modular?............................................. 520 2.1. Peripheral vs. massive modularity: is there a nonmodular core? ....................................................................... 520 2.2. Is the central system impenetrable for human intelligence?.......................................................................... 521 2.3. Is the mind (are modules) the result of Darwinian adaptation?............................................................................ 522 3. Core modular properties .............................................................. 523 3.1. Domain specificity................................................................ 523 3.2. Informational encapsulation ................................................. 524 3.3. Summary: how to identify a module..................................... 526 4. Specialised neurons and neural localisation of cognitive functions....................................................................................... 527 4.1. Mind-brain relationship ........................................................ 527 4.2. Functional anatomy: the existence of specialised and localisable (suites of) neurons is undisputed ........................ 528 4.3. Some literature...................................................................... 530 5. Modules can be plugged out without affecting other faculties .... 531 5.1. Double dissociation .............................................................. 531 5.2. Documented cases: face recognition, number sense............. 531 5.3. Double dissociation of language........................................... 532

606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621

622

Chapter 5 Modularity of and in language, related systems

623

1. Modularity in the early days of generative grammar: 50s-60s .... 535 1.1. A spearhead of the cognitive revolution of the 50s in language................................................................................ 535 1.2. LSLT: language is made of modules (levels), a concatenation algebra and interfaces .................................... 536 1.3. Modularity on its way: from LSLT to Aspects and SPE ...... 538 2. Modularity implies biology and innateness: the language organ ............................................................................................ 539

624 625 626 627

xxxiv Table of contents – detail §

page

628

3. Grammar itself is made of modules: GB-subtheories and their (questionable) status as cognitive modules.......................... 540 3.1. The inverted T is the baseline since the 60s ......................... 540 3.2. GB-subtheories are presented as modules, but insulated from the cognitive context .................................................... 541 3.3. Chomsky (1981): subcomponents (inverted T) vs. subsystems (theta theory etc.)............................................... 542 3.4. Are GB-subsystems cognitive modules? .............................. 543 3.5. Biolinguistics: an evolutionary argument against language-internal modularity (Hornstein 2009).................... 545 4. GB modules and their perception in non-linguistic quarters ....... 547 4.1. Chomsky (1981) calls GB-subtheories modules without comment ............................................................................... 547 4.2. Perception of GB-modules in non-linguistic quarters: puzzlement............................................................................ 548 5. Minimalism and biolinguistics do away with GB modules ......... 550 5.1. Minimalism: GB-subtheories have to go .............................. 550 5.2. Grammar reduces to morpho-syntax: PF and LF are neither language- nor species-specific .................................. 550 6. Identifying linguistic modules ..................................................... 551 6.1. How to identify grammar-internal modules.......................... 551 6.2. Dissociation: Pragmatics, Lexicon vs. morpho-syntax......... 552 6.3. Domain specificity (Starke): morpho-syntax-semantics vs. phonology........................................................................ 553 6.4. Domain specificity (Jackendoff, Chomsky): phonology is distinct............................................................................... 554 6.5. Phonology-free syntax .......................................................... 555 6.6. Late Insertion is the segregation of phonological and other vocabulary ................................................................... 555 6.7. Phonology vs. phonetics ....................................................... 556 7. Encapsulation is called inclusiveness in syntax........................... 556

629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648

Table of contents – detail

§

xxxv page

649

Chapter 6 How modules communicate

650

1. Intermodular communication requires translation ....................... 557 2. Translation of what? .................................................................... 558 2.1. Modular computation: vocabulary (input) vs. structure (output) ................................................................................ 558 2.2. Is structure, but not vocabulary, translated? ......................... 558 3. Translation is selective, and the choice of translated pieces is arbitrary........................................................................................ 559 4. Outlook: intermodular translation is the focus of Vol.2 .............. 560

651 652 653 654 655

Part Two Lessons from interface theories 656

Chapter 1 A guide to the interface jungle ...................................................... 563

657

Chapter 2 Empirical generalisations

658

1. Introduction.................................................................................. 565 2. Morpho-syntax has no bearing on the content of phonological computation.................................................................................. 566 3. Morpho-syntax and melody are incommunicado......................... 568 3.1. Morpho-syntax can neither read melody nor bear on it........ 568 3.1.1. Phonology-free syntax is in fact melody-free syntax.......................................................................... 568 3.1.2. Carriers of morpho-syntactic information do not include melody ........................................................... 568 3.2. (Floating) morphemes without linear realisation originate in the lexicon......................................................................... 569 3.3. Conclusion: vocabulary excluded from translation altogether?............................................................................. 570

659 660 661 662 663 664 665

xxxvi

Table of contents – detail

§

page

666

Chapter 3 Issues that are settled

667

1. There are no boundaries inside morphemes................................. 573 2. There is no phonetic correlate of morpho-syntactic information................................................................................... 574 2.1. Phonetic correlate of morpho-syntactic breaks: definition ... 574 2.2. Generative diacritics make no noise, except in Natural Generative Phonology........................................................... 575 3. Affix ordering is wrong ............................................................... 576 4. Interpretation is inside-out, grammar is interactionist, brackets are unnecessary relics .................................................................. 577 4.1. Inside-out interpretation and cyclic derivation ..................... 577 4.1.1. Two ways of organising inside-out interpretation: brackets vs. interactionism ......................................... 577 4.1.2. Brackets require a PF parsing device and make a procedural insight representational (and diacritic) ..... 579 4.2. The line-up of interface theories in regard of interactionism ....................................................................... 580 4.2.1. Revolution (Lexical Phonology) and counterrevolution (Halle & Vergnaud)................................... 580 4.2.2. Phonology and the interface are not the same thing ... 581 4.2.3. Derivation by phase: when generative grammar became interactionist .................................................. 582 4.3. Only interactionism makes inside-out interpretation compatible with modularity.................................................. 583 4.3.1. Modularity referees in favour of interactionism......... 583 4.3.2. Nobody used modularity in the 80s............................ 584 5. Interface Dualism......................................................................... 585 5.1. If tacitly, (almost) all theories implement Interface Dualism................................................................................. 585 5.2. Representational communication needed (contra Kaye 1995) ..................................................................................... 586 5.3. Procedural communication needed (contra orthodox OT).... 587

668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686

Table of contents – detail

§

xxxvii page

687

Chapter 4 Modularity, translation, the diacritic issue and local vs. domain-based intervention: settled in verb, but not in fact

688

1. Introduction.................................................................................. 589 1.1. Three questions: two are settled, one has never been discussed ............................................................................... 589 1.2. Modularity and diacritics are only settled in verb ................ 589 1.3. The three questions are the backbone of the representational channel ....................................................... 590 2. Translation in structuralist and generative interface theory......... 590 2.1. Interface design was done in absence of a modular/ cognitive background – but translation has always been practised................................................................................ 590 2.2. The birth and variable incarnation of diacritics .................... 591 2.2.1. Juncture phonemes and SPE-type boundaries: diacritic translation and various degrees of camouflage ................................................................. 591 2.2.2. The abandon of Level Independence makes boundaries diacritics................................................... 592 2.2.3. Since structuralism, the output of translation has always been a diacritic................................................ 593 2.2.4. No Diacritics ! – no diacritics ? .................................. 593 2.3. Modularity and translation were invented by structuralism ......................................................................... 594 2.3.1. Non-cognitive modularity: Level Independence enforces translation..................................................... 594 2.3.2. Translation affords the assessment of phonological theories according to their behaviour at the interface ...................................................................... 594 2.4. Generative modularity offenders: reference to untranslated morpho-syntactic information .......................... 595 2.4.1. Translation was not a standard in generative theory until the mid 80s.............................................. 595 2.4.2. Turning back the wheel: weak and strong modularity offenders in (more or less) recent development ............................................................... 596 2.4.3. A note on Structural Analogy..................................... 597

689 690 691 692 693

694 695

696 697 698 699 700 701

702 703 704

705

xxxviii Table of contents – detail §

page

706

3. Local vs. non-local carriers of morpho-syntactic information..... 598 3.1. Local boundaries vs. non-local domain-based intervention ........................................................................... 598 3.2. Notational variants and real differences ............................... 599 3.2.1. Domain-based can be translated into local reference and vice-versa ............................................. 599 3.2.2. The difference is conceptual, not empirical................ 600 3.3. The local baby and the diacritic bathwater ........................... 601 3.4. There can be non-diacritic boundaries, but what would a non-diacritic domain look like? ............................................ 602 3.4.1. Non-diacritic boundaries (can) exist........................... 602 3.4.2. Top-down constructions are diacritic by definition (prosodic word and higher)......................................... 603 3.4.3. Higher layers of the Prosodic Hierarchy are the projection of nothing .................................................. 604 3.4.4. Projections created by phonological computation cannot be the output of translation ............................. 605 3.5. Conclusion: possible carriers reduce to syllabic space ......... 606 4. Conclusion ................................................................................... 606

707 708 709 710 711 712 713 714 715 716 717 718

719

Chapter 5 Open questions (general)

720

1. Grammatical architecture: alternatives to the inverted T............. 609 1.1. Generative semantics ............................................................ 609 1.2. Parallel modules.................................................................... 610 1.2.1. Against syntactico-centrism (Jackendoff) .................. 610 1.2.2. Designated portions of the skeleton project morphosyntactic features (Bendjaballah & Haiden)............... 611 1.3. Conclusion: the baseline of the generative paradigm is modularity............................................................................. 612 2. PF – a strange hermaphrodite animal........................................... 613 2.1. Clean syntax, dirty phonology/PF?....................................... 613 2.1.1. Minimalism shrinks syntax......................................... 613 2.1.2. Minimalism pumps up PF .......................................... 614 2.1.3. Dumping into the PF dustbin and hoping that it is big enough .................................................................. 614 2.1.4. The syntacticians' phonology is not what phonologists call phonology....................................... 615

721 722 723 724 725 726 727 728 729 730 731

Table of contents – detail

xxxix

§

page

732

2.2. Confusion and mistiness: what does PF mean, what does it contain?.............................................................................. 616 2.3. Properties of PF: what kind of animals live in the intermundia? ......................................................................... 619 2.3.1. Internal structure of PF............................................... 619 2.3.2. What happens "at PF"................................................. 619 2.4. Trying to make sense of PF from the modular point of view....................................................................................... 621 2.4.1. PF is a cover term for a number of serially ordered computational systems................................................ 621 2.4.2. The minimalism-born intermundia violates domain specificity ................................................................... 622 2.4.3. Mixing phonology and the intermundia ..................... 623 2.4.4. The internal structure of the intermundia: two distinct derivational stages, morphosyntax and morphophonology (Idsardi & Raimy forth) ............... 624 2.5. Linearisation ......................................................................... 626 2.5.1. Introduction: no business of phonology ..................... 626 2.5.2. In minimalist times: no business of syntax either....... 626 2.5.3. Both syntax-internal and syntax-external linearisation is minimalism-compatible...................... 627 2.5.4. Everybody but Kayne does linearisation "at PF" ....... 629 2.5.5. Linearisation in phonology in order to derive phonetics (Raimy)?..................................................... 631 2.6. Conclusion: a minimalism-born monster.............................. 633 3. The balance of procedural and representational communication............................................................................. 635 3.1. Which channel for which phenomenon?............................... 635 3.2. Prosodic Phonology tries to get away with peaceful coexistence and/or random distribution................................ 636 3.3. Random use of both channels is typical (and tacit), but it may be doubted that this is the right way to go .................... 636 4. Morpho-syntactic structure may, but content (labels) may not bear on phonology........................................................................ 637 5. The mapping puzzle ..................................................................... 639 5.1. The issue(s), looked at by more or less helpless linguists .... 639 5.2. The mapping puzzle since SPE............................................. 640

733 734 735 736 737 738 739 740

741 742 743 744 745 746 747 748 749 750 751 752 753 754 755

xl

Table of contents – detail

§

page

756

6. Privativity..................................................................................... 641 6.1. On the representational side: complete or privative translation of morpho-syntactic divisions? ........................... 641 6.1.1. All theories are non-privative since SPE.................... 641 6.1.2. Unanimous non-privativity in individual theories...... 641 6.2. On the procedural side: selective vs. non-selective spellout ......................................................................................... 643 6.3. Five arguments in favour of privativity ................................ 643

757 758 759 760 761

762

Chapter 6 Open questions (procedural)

763

1. Selective spell-out and interpretation-triggering affixes.............. 647 1.1. Selective spell-out: what it is and how it works ................... 647 1.2. Phase edge and piece-driven phase....................................... 648 1.2.1. Spell out your sister: the phase edge in phonology .... 648 1.2.2. Piece-driven vs. node-driven phase ............................ 650 1.3. Systems with non-selective spell-out: Lexical Phonology (DM)................................................................... 650 1.4. Spell-out is selective in syntax, but is this an argument? ..... 651 1.5. Conclusion: a unifying perspective....................................... 653 2. Phase theory................................................................................. 654 2.1. A rapidly growing field whose diversity can be confusing at times (not only for phonologists)...................................... 654 2.2. What counts as a phase? ....................................................... 655 2.2.1. Inner- and extra-syntactic criteria for distributing phasehood................................................................... 655 2.2.2. The trend is towards atomisation................................ 656 2.2.3. Is there a lower limit for phasehood? Is spell-out selective? .................................................................... 657 2.2.4. Anti-locality of movement marshals atomisation....... 658 2.3. Extensions of the basic model .............................................. 659 2.3.1. Asymmetric spell-out: independent access of LF and PF......................................................................... 659 2.3.2. PIC à la carte: process-sensitive no look-back ........... 659 2.3.3. Phase Extension: when phasehood depends on what the head is made of ............................................ 661 2.3.4. Unification of piece-driven and node-driven phase by a lexical phasehood feature.................................... 662

764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782

Table of contents – detail

xli

§

page

783

3. No look-back (Phase Impenetrability) ......................................... 663 3.1. Limited scope of this section ................................................ 663 3.2. No look-back devices since Chomsky (1973) and their non-conflatability with derived environment effects............ 664 4. Why are there no phonological effects of the cyclic spell-out of words?...................................................................................... 665 4.1. The absence of cyclicity-induced external sandhi: a consensual fact that theories build on, but do not talk about ..................................................................................... 665 4.2. How interface theories behave: claims for and against cyclic spell-out of words, for and against its cyclic interpretation......................................................................... 666 4.2.1. The baseline position of SPE: everything is cyclic .... 666 4.2.2. Lexical Phonology: the interpretation of word sequences is not cyclic................................................ 666 4.2.3. Lexical Phonology makes no claim about spell-out and installs non-cyclic interpretation of word sequences without argument ......................... 667 4.2.4. Halle & Vergnaud and Kaye: restoration of SPE – everything is cyclic.................................................. 668 4.2.5. Distributed Morphology is entirely agnostic in phonological matters .................................................. 668 4.3. The word-spell-out mystery.................................................. 669 4.3.1. Cyclic spell-out of words but no phonological traces? ......................................................................... 669 4.3.2. Wrong data or an on/off switch for Phase Impenetrability ........................................................... 669 4.3.3. A solution: (chunk-specific) PIC à la carte ................ 670 4.3.4. PIC à la carte that does not want to be named (Samuels 2009a) ........................................................ 671 4.3.5. Phase Impenetrability is a property of the spellout mechanism, not of concatenative or interpretational systems .............................................. 672 4.4. Intonation requires cyclic spell-out of words for sure – but this does not appear to concern phonological computation ...... 674 4.4.1. Intonation is governed by syntactic structure ............. 674 4.4.2. Prosodic structure may be recursive – phonology is not ........................................................................... 675

784 785 786 787

788

789 790 791

792 793 794 795 796 797 798 799

800 801 802

xlii

Table of contents – detail

§

page

803

4.4.3. Phonology is not recursive: confusion between non-recursive phenomena and their eventual analysis with recursive constructions ......................... 676 4.4.4. Prosodic structure is not created by phonological computation ................................................................ 678 4.4.5. Phonology only interprets: Merge must be absent, hence the result of phonological computation is flat .. 679 4.4.6. Is intonation a phonological phenomenon at all? ....... 680 4.5. Conclusion ............................................................................ 681 4.5.1. If the generalisation is correct, Praguian segregation alone will not do...................................... 681 4.5.2. Chunk-specific PIC à la carte ..................................... 682 4.5.3. Outlook....................................................................... 683 5. Chunk-specific phonologies......................................................... 683 5.1. Specific world-level phonology............................................ 683 5.1.1. Everybody has a word-specific phonology ................ 683 5.1.2. Since SPE, word-specific phonologies are based on (English) stress ...................................................... 684 5.2. Specific sentence-level phonology: Praguian segregation.... 685 5.2.1. Cyclic phonology of morphemes vs. non-cyclic phonology of words.................................................... 685 5.2.2. Arguments for Praguian segregation .......................... 686 5.3. Praguian segregation and double dissociation ...................... 686 5.3.1. Double dissociation applied to Praguian segregation.................................................................. 686 5.3.2. Processes that are restricted to morpheme sequences.................................................................... 687 5.3.3. Processes that apply across the board......................... 688 5.3.4. Processes that are restricted to word sequences ......... 689 5.4. PIC à la carte......................................................................... 691 5.4.1. Multiple mini-phonologies and chunk-specific phonologies................................................................. 691 5.4.2. Process-specific PIC, rather than "don't undo!" ......... 692 5.4.3. PIC à la carte: process- and chunk-specific................ 693 5.5. Conclusion ............................................................................ 694

804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827

Table of contents – detail

xliii

§

page

828

6. Morpheme-specific mini-phonologies (level 1 - level 2)............. 695 6.1. Cyclicity analysed by morpheme-specific miniphonologies vs. by no look-back: a major front line in generative interface theory.................................................... 695 6.2. Intermodular argumentation: if the PIC exists in syntax, it must be active in phonology as well.................................. 696 7. Empirical coverage ...................................................................... 697 7.1. Introduction: three competitors and the empirical record (affix class-based phenomena)............................................. 697 7.2. Affix class-based phenomena ............................................... 698 7.2.1. The rule-blocking pattern ........................................... 698 7.2.2. The rule-triggering pattern.......................................... 699 7.2.3. Is the rule-triggering pattern less real? ....................... 701 7.3. Derived environment effects................................................. 701 7.3.1. A separate issue .......................................................... 701 7.3.2. A blooming landscape of analyses ............................. 702 7.3.3. Anti-cyclic phenomena: derived environment effects have a big brother ........................................... 703

829

830 831 832 833 834 835 836 837 838 839 840

841

Conclusion Intermodular argumentation

842

1. Trying to get a handle on the interface ........................................ 705 1.1. Looking at the interface through the prism of its history...... 705 1.2. Two more prisms used: Interface Dualism and modularity.. 705 1.3. Three theory-external ways to get a handle on interface theory .................................................................................... 706 2. Intermodular argumentation......................................................... 707 2.1. Intermodular potential of interactionist derivation by phase ..................................................................................... 707 2.1.1. Each end of the interactionist pipe may impact the other............................................................................ 707 2.1.2. Morpho-syntax can referee competing phonological analyses and theories .................................................. 708 2.1.3. Intermodular arguments can only be made through the procedural channel: translation is arbitrary .......... 708 2.2. Conditions and limitations of intermodular argumentation .. 709

843 844 845 846 847 848 849 850 851

xliv Table of contents – detail §

page

852

862

2.3. Six properties of spell-out that can be made intermodular arguments.............................................................................. 709 2.3.1. Convergence of syntactic and phonological tools ...... 709 2.3.2. Strong and weak version of the intermodular argument ..................................................................... 710 2.3.3. Import of phonology-based mechanisms into syntax.......................................................................... 711 2.3.4. Four syntactic referees for phonological theories and the special status of interactionism ...................... 711 2.3.5. Phonology must provide for a "freezing" PIC, selective spell-out and the phase edge ........................ 712 2.4. How many spell-out mechanisms are there in grammar?..... 713 2.4.1. The parallel is always between the spell-out of morphemes and words................................................ 713 2.4.2. The weak version of the argument can be made, but the strong version hinges on the unity of morphology and syntax .............................................. 714 2.4.3. The word-spell-out-mystery strikes again .................. 714 2.5. Conclusion ............................................................................ 715

863

References....................................................................................... 717

864

Subject index .................................................................................. 785 Language index .............................................................................. 838 Index of phenomena....................................................................... 843

853 854 855 856 857 858 859 860

861

865 866

Table of graphic illustrations § description 592 592 592 601 601 601 612 612 612 612

table number an atom: scientific reality (243a) a transistor: symbolic (243b) a transistor: physical (243c) Franz Joseph Gall (252) American Phrenological (253) Journal, 1848 Webster's Academic (253) dictionary, 1895 optical illusion: Zöllner (255) optical illusion: Müller- (255) Lyer optical illusion: (255) Poggendorff optical illusion: Hering (255)

copyright

page

free .........................................503 GNU Licence .........................503 GNU Licence .........................503 free .........................................515 free .........................................516 free .........................................516 GNU Licence .........................525 GNU Licence .........................525 GNU Licence .........................525 GNU Licence .........................525

All graphic illustrations are taken from Wikipedia (August 2010).

1

Editorial note This book is a piece of a manuscript that was circulated since October 2008, but (unsurprisingly) turned out to be too long to fit into the covers of a single book. The manuscript ran under the same title as the present book and was designed as the second volume of Scheer (2004a). Back in 2001 when I started working on Scheer (2004a), it was meant to be the third part of it, but in the end had to be outsourced because the project grew hopelessly out of size. The ancestor of the 2008 manuscript was thus ready in 2003, and it was intended to become Vol.2 of the book that was published in autumn 2004. At the outset of Scheer (2004a), I was naïve enough to believe that "[w]hen this editorial note was written (August 2004), the second volume was almost completed in draft. The constant reference that is made to it here should therefore reflect its divisions quite closely." Little did I know about the fact that the work on the manuscript would take another six years. And that it would end up the same way as the manuscript of Scheer (2004a): too long in order to be published within the covers of a book. The outsourcing thus continued in the way described below, and I hope that the piece which did not make it into the book that the reader holds in hands will really be published as a whole without further subdivision in the future (and earlier than six years from now, i.e. October 2010). The manuscript dated October 2008 was made of three Parts: 1. Morpho-syntactic information in phonology: a survey since Trubetzkoy's Grenzsignale 2. Lessons from interface theories 3. How translation works: Direct Interface and just one – channel

this book Vol.2

The book that the reader holds in hands encompasses the first two Parts. It thus proposes a history of the morpho-syntax → phonology interface (in this direction only, Parti I), an Interlude that presents modularity from a Cognitive Science perspective, and a catalogue of design properties that a sound theory of the interface should or must not be based on given the lessons from the historical survey, the Cognitive Science background and the current (minimalist) landscape of generative grammar (Part II). The original Part III exposes my own view of the interface and will (hopefully) make it into an independent book. For reasons explained in the foreword and in §42 below, this further book may be legitimately called

xlviii Editorial note

volume 2 of Scheer (2004a), while the present book is a stand-alone piece of work that has only an implicit relationship with Scheer (2004a). Therefore, the forthcoming volume that accommodates the original Part III of the manuscript is referred to as Vol.2 below, while Scheer (2004a) appears as Vol.1. Beyond the obvious issue regarding the size of the original manuscript (871 pages), the decision to make Part I and Part II a stand-alone book is motivated by the fact that they are thematically independent and theory-unspecific. Finally, it is worth mentioning that if the present book is largely identical to the aforementioned 2008 manuscript, the feedback from various sides that I have received (see the end of the foreword) and the final revision process have reshaped a number of sections and chapters quite significantly. For example, a piece that was completely rewritten is the history of modularity in GB (§622). Also, topics are now discussed that were absent from the draft, or underexposed: these include Prosodic Morphology (§442), the strange hermaphrodite animal PF (§726), linearisation (§741), the adaptation of Prosodic Phonology to the phase-based environment (§462) and the early history of modification-inhibiting no look-back in the 80s (§293). Thus it may still be worthwhile opening the book for readers who have already struggled with the manuscript.

2

Foreword The plot, and how to use the book The ambition of this book is twofold, corresponding to its two Parts: it first proposes a history of the interface between morpho-syntax and phonology roughly since World War II, and then tries to see where we stand today; the Interlude and its position in the book points to the central role that modularity plays in the generative (and minimalist) architecture of grammar. The questions asked are: what can we learn from previous endeavour? Is the current state of (phonological) theories of the interface really informed of earlier results? Which questions are solved, which ones remain open? Is the current state of affairs compatible with current (morpho-)syntactic theory and especially with the recent phase-based, i.e. interface-based (or even: interface-motivated) environment of the minimalist programme? How could phonological and morpho-syntactic theories of the interface converge (there is no way not to: a consistent theory of grammar must have an interface theory that is compatible with both its input and output)? How can we argue with properties of one for or against theories of the other (intermodular argumentation)? The book may thus be accessed in three different ways: through a specific theory, through the chronology of events, and thematically. That is, the reader may want to know what a specific theory has got to say regarding the interface, he may follow (a particular period of) the development of the question since Trubetzkoy, or he may use the thematic summaries that structure Part II and the keyword-based access that is offered by the thematic index (both distribute into Part I). The book is thus devised for a general audience that wants to know about the interface and its history, as much as for more specialized interests that wrestle with intermodular communication both in syntactic and in phonological quarters: one thing that became increasingly clear to me as the book was growing is the fact that syntacticians and phonologists are largely unaware of the interface theories that are developed on the other side, which however may directly impact their work. In this context, modularity will be a key concept that escorts the reader all through the book as a guiding light. On the one hand, it is the basic idea about the architecture of grammar that unites structuralist and generative thinking; on the other hand, it offers an extra-linguistic reference point which sets the cognitive frame of interface theories and is able to

l Foreword

referee them. This contrasts with the (phonological) interface literature where modularity is more or less irrelevant in practice (see §36). It was mentioned in the editorial note that the book is actually a piece (in fact the lion share) of a manuscript that was originally intended to be the second volume of Scheer (2004a). The two Parts (and the Interlude) that now appear as the present book are the result of the fact that at some point I wanted to make sure that my own view of the interface – Direct Interface – does not reinvent the wheel. Direct Interface (which is introduced in the original Part III, also Scheer 2008a, 2009a,c) is based on the idea that only truly phonological objects can be the output of translation, that is, be representational carriers of morpho-syntactic information in phonology. Unfortunately, this claim had been made in Prosodic Phonology before, where the Prosodic Hierarchy was taken to be "truly phonological" (as opposed to diacritic SPE-type boundaries) (see §§405,690). In order to find out that the "truly phonological" objects of Prosodic Phonology, i.e. the constituents of the Prosodic Hierarchy, are in fact diacritics in an autosegmental guise, I had to read through the foundational literature of this theory, which dates back to the early 80s. This reading then expanded to other "old stuff", which was growing older and older as I was gaining ground. About five years later, the result is this book on the (history of the) interface. The Introduction below provides further detail regarding the prisms that are used in the book in order to look at the interface, its structure and the delineation of its object of study. It also explains the kind of historiography that is practised. The book may thus be described as a kind of necessary prelude to Part III of the original manuscript. This Part III being absent, however, the book acquires a stand-alone virtue: it is thematically consistent, and there is no reference to Government Phonology in general (except of course in the chapter on this specific theory) or to CVCV (strict CV) in particular. The bonds with Vol.1 are thus delicate, if any. Hence the decision, counter to its genesis, to have the book run as an independent item: it is well suited to stand on its own feet, and calling it volume two of Scheer (2004a) would have been inconsistent since the reader would have found no continuation or development of CVCV. This being said, let me explain in which way the book prepares Part III, which will also stand alone, as the true Vol.2 of Scheer (2004a). Again, the bond is Prosodic Phonology and something that may be called deforestation (see §42): following the lateral core of Government Phonology, Vol.1

Foreword

li

has worked out a lateral analysis of syllable structure, which is traditionally represented by trees. That is, lateral relations among constituents (government and licensing) take over the function of arboreal structure, which is thus eliminated at the syllabic level. The present book follows up on this by doing away with the arboreal representation of morpho-syntactic information (i.e. the Prosodic Hierarchy), which is shown to be diacritic and hence unwarranted. Phonology, then, is entirely flat: no concatenation of any pieces, no Merge (or other tree-building device), no projection, no trees (trees are also absent below the skeleton if melodic representation is a matter of monovalent primes). The rationale behind deforestation is twofold. For one thing, the lateral and the tree-based approaches to syllabic structure do the same labour and are therefore mutually exclusive: (at least) one of them must be wrong (see Vol.1:§165). The other reason for deforestation is the fact that in the generative architecture of grammar PF and LF are "merely" interpretational devices: concatenation is the privilege of (morpho-)syntax. Hence PF and LF do not concatenate anything, and if Merge is the universal concatenative engine, it must be absent from phonology. This means that there is no treebuilding device, and hence that there are no trees in this module. The absence of trees, in turn, explains why there is no recursion in phonology (see §42). It was mentioned that Part III of the original manuscript exposes Direct Interface, my own view of how morpho-syntax talks to phonology. Direct Interface draws the consequences from the lessons that are established in Part II: it is introduced as an answer to deforestation, i.e. the need for non-diacritic carriers of morpho-syntactic information in phonology. This programme is supplemented with the discussion of consequences for interface theory (non-computational translation of morpho-syntactic information into phonological objects through a lexical access) and CVCV, as well as with an application to the representation of the beginning of the word in phonology (the so-called initial CV). A brief description of Vol.2 is provided in §49. While writing the book and preparing the manuscript for print, a number of pieces have been published in form of articles (or are underway). The reader may find it useful to rely on them in parallel since they condense this or that aspect of the book in a reasonably sized stand-alone item. Three articles concern the procedural side of the interface (i.e. cyclic spellout and the PIC): Scheer (2008c) (on the parallel between the edge of the phase in current syntactic theory and interpretation-triggering affixes in the classical phonological literature: in both cases the sister of actual phase

lii Foreword

head is spelled out, see §765), Scheer (2009b) (on the word-spell-out mystery, see §§786,851) and Scheer (2010b) (on the number of computational systems in phonology, see §828). Another article, Scheer (2008a), is about the diacritic issue: the Prosodic Hierarchy is as much a diacritic as the hashmark (if in an autosegmental guise) and therefore has to go (see §§402,692). Finally, two articles are outsourced from Vol.2 (Part III): Scheer (2009a) and Scheer (2009c) concern external sandhi, the phonological motivation of phases above the word level and the question of how the beginning of the word is represented (the initial CV is phase-initial, rather than word-initial). Finally, a few words are in order regarding ideas and people that were important while writing the book. The treatment of the central issue regarding modularity (actually more in the forthcoming Vol.2 than in the present book) owes to (unpublished work by) Michal Starke and his classes at various EGG summer schools (Novi Sad 2002, Cluj 2004, Olomouc 2006). On another front, the book was written in parallel with Ricardo BermúdezOtero's Stratal Optimality Theory (Bermúdez-Otero forth a), a sister project regarding the interface (that is still to appear). Ricardo's book has also a strong historical inclination. Over the years, we entertained a continuous, gentle and critical conversation about our shared interests, and occasionally our diverging views, which has had quite some effect, especially on my understanding of Lexical Phonology. It looks like in the end I will win the race for publication – which means that Ricardo will not have to quote page and section numbers of forthcoming manuscripts. A number of people have helped improving the book a lot over the six years of work that it took to get to this foreword. Namely the feedback based on the 2008 manuscript was rich and invaluable. Asking people to read 871 pages is very impolite and unrealistic – and it does not make friends. The natural reaction is to shy away when opening the parcel. Three readers were masochistic enough not only to really go through the manuscript from cover to cover, but also to write up pages and pages of detailed comments: Marc van Oostendorp, Diana Passino and Gaston Kočkourek. They are to be thanked in the first place. Terje Lohndal took the trouble to comment on the pieces of the manuscript that directly deal with syntactic theory (and its history). Since of course I am as ignorant in syntactic matters as a phonologist can be and had to work hard in order to have a remote chance not to utter monkeyshine when it comes to syntax proper, Terje's look at the text through the syntac-

Foreword

liii

tic lens and the detailed correspondence with him have made the book much less unreadable for syntacticians. Finally I am also indebted to Grzegorz Michalski, Markéta Ziková, Artur Kijak, Katérina Jourdan and Victor Manfredi for valuable feedback. Châteauneuf de Grasse, October 2010

3

Introduction

4

1. Procedural and representational communication with phonology

5

1.1. Cyclic derivation and hashmarks This book discusses how morpho-syntactic information is shipped to and processed by phonology. It has a number of characteristics, which are fleshed out in the introduction below. The most prominent characteristic of the book, regarding both its organisation and content, is certainly the procedural-representational prism that is used in order to look at the interface and at interface theories. There are two ways for morpho-syntax to bear on phonology: procedurally and representationally. The former is a genuinely generative invention that has come into being in Chomsky et al. (1956:75) and was successively known as the transformational cycle, the phonological cycle, cyclic derivation and finally today as derivation by phase (in syntactic quarters). It embodies the insight that (phonological and semantic) interpretation applies successively from the most to the least embedded piece. It will therefore be sometimes referred to as inside-out interpretation in the book. The other means by which morpho-syntax can influence phonology is through the insertion of a representational object into the linear string that is submitted to phonological computation: morpho-syntactic structure is translated into items which are processed by phonology. This is the traditional interface management which is practised (at least) since the 19th century, and in any case is shared by structuralist and generative thinking: carriers of extra-phonological information in phonology have successively incarnated as juncture phonemes, SPE-type diacritics (# and the like) and the Prosodic Hierarchy, each being the representative of its time. That is, carriers of morpho-syntactic information were (juncture) phonemes when phonemes were the basic currency in phonological theory, they were made segments in SPE (# was supposed to be a [-segment] segment) where the basic phonological units were segments, and finally became autosegmental domains (prosodic constituency) in the early 80s when all areas of phonology were autosegmentalised. One goal of the book is to show that all objects which were thought of as the output of translation thus far are diacritics and therefore do not qualify (including the Prosodic Hierarchy: recall from the foreword that

2

Introduction

this was my original motivation to dive into historiography). Direct Interface, to be introduced in Vol.2 (also Scheer 2008a, 2009a,c), is about the elimination of all diacritics that mediate between morpho-syntax and phonology.

6

1.2. Interface Dualism: both means of talking to the phonology are needed Following this orientation, the book systematically distinguishes between procedural and representational aspects of the interface. It defends what I call Interface Dualism, i.e. the idea that natural language provides for and uses both channels: theories that try to reduce interface activity to either channel are on the wrong track. The interface landscape as it stands today is structured along this fraction line anyway: roughly speaking, Lexical Phonology (and its modern offspring: DOT, Stratal OT) is the procedural theory of the interface, while Prosodic Phonology (and its modern offspring) is the representational theory of the interface. The book shows that the two aspects are complementary, rather than in competition. As we will see as we go along (see the summary in §748), the interplay of procedural and representational means of managing interface phenomena has always been vague, to say the least: typically the question is not even addressed, and phonologists have actually put much effort into not bringing it up. The result is an undetermined peaceful coexistence (don't look at me, so I won't look at you) that was installed in the 80s (see §423, save two punctual attempts at reducing the interface to the representational channel), and is still the diplomatic standard today. A discussion that defines the rule of the game in order to arrive at a proper division of labour that does away with overlap is certainly warranted: "show me your interface phenomenon, and I tell you whether it is due to procedural or representational activity" would be the ideal situation. The book is not a good place to call for this debate, though, since it does not make any contribution. I have tried to classify interface phenomena (i.e. phonological processes that are influenced by morpho-syntactic information) for several years, at least into three categories: "procedural only", "representational only" and a misty in-between that could be due to either. In the end I have given up on this because of the complexity of the task (see the summary of this issue in §748). There is just a very broad indication left that is based on the opportunity of intermodular argumentation (on which more in §§17ff): Procedural First. That is, when competing procedural and representational analyses are

Functional historiography

3

empirically equivalent, choose the former because it may make predictions in morpho-syntax that can be controlled independently of phonological evidence (see §316). On the empirical side, the issue is discussed on the occasion of what I call the word-spell-out mystery (chunk-specific phonologies §786, see §22 below), and also in Vol.2 where so-called connected speech (i.e. external sandhi, cases where phonology applies across word boundaries) is examined. In any event, one day the question will have to be seriously addressed: which interface phenomena are due to procedural activity, against which other phenomena that are the result of representational communication? In absence of a significant contribution, the book can at least call for upgrading the issue on the research agenda.

7

2. Functional historiography

8

2.1. Anderson's dualistic legacy: structure and process The reader will have noticed that the procedural-representational prism which the book uses in order to look at the interface is much like the structure-process opposition that makes the spine of Stephen Anderson's (1985) history of phonology in the 20th century. Using this prism produces results that strike close to the mark – at least in Anderson's case. Writing at the peak of the representational (autosegmental) wave and having observed the see-saw movement of phonology between process- and representation-oriented extremes, Anderson just had to extrapolate what comes next: another round of the computational extreme, in revenge to the representational excess. Phonology was thus programmed to produce OT, which entered the scene a couple of years later. Anderson predicted its arrival, but little did he know how extreme this round of "phonology is computational and nothing else" could get: SPE is by far outcompeted. This is certainly not unrelated to the connectionist roots of OT (§529): connectionism knows only computation, objects (symbols) do not exist (§593). Today there is a strand in OT (still timid but sensible, e.g. Blaho et al. 2007) which tries to rebalance the system by making representations (which are not "emergent") independent of constraints, and by crediting them with a true arbitral award (also independent, i.e. which cannot be outranked by constraints).

4

Introduction

Anderson's legacy is dualistic in nature: at the end of his book he warns phonologists to believe that representations alone provide access to felicity – but also that the next round of "computation is king" will not be any more successful than preceding attempts to reduce phonology to computation. His words were not marked, and we assisted a computational festival for about a decade. The peak of this cycle may now be behind us, and it seems that here and there representations begin to play a role again that is not controlled by computation. All this notwithstanding, it goes without saying that OT has made a valuable contribution to computational theory and the expression of parametric variation: the study of computation is not objectionable – but the ambition to make it (or rather: the fact of making it) the only thing that determines grammaticality is (see Scheer 2010a). Rather than oscillating between the procedural and the representational end of the spectrum, one may incline to believe that phonology would be better advised to break out of this cyclic movement. Each generation is reinventing the wheel of the teachers of their teachers – always in a new guise, with different vocabulary and promising that this time the progress will be unequalled. This vicious circle seems to be entirely unimpacted by the fact that Anderson has made the see-saw movement explicit, together with its prejudicial effects. The aforementioned Interface Dualism thus follows Anderson's footsteps: as much as in phonology proper, both procedural and representational activity is needed at the interface. Alongside with the request for a sound balance between structure and computation, another dualistic property of adult science is a related point of interest which, unfortunately, characterises phonology less and less in recent years. Representatives of adult science know that the scientific truth is where theoretical prediction (top-down) and empirical data (bottom-up) meet, and that one cannot exist without the other. What we are living through today, however, is a period of empiricist conquest (Katz & Bever 1974 describe the cyclic return of empiricism in linguistics), the present wave being usage-based or even "cognitive" (Bybee 2001, Langacker 1987 et passim). Sciences are characterised by an infantile empiricist period upon inception: "the more primitive the status of a science is, the more readily can the scientist live under the illusion that he is a pure empiricist" (A. Einstein). The problem with linguistics, and specifically with phonology, is that empiricist waves roll ashore every couple of decades.

The syntactic frame: minimalist phase theory 9

5

2.2. Historiographic cherry-picking It is obvious from the preceding that historiographic activity in this book is not an end in itself: it serves a purpose and has a function in a broader argumentation. In this sense, the (hi)story that is told in this book has the same ambition as what traditionally made the motivation of historians: to learn from the past in order to understand the present, and to shape the future. Knowing about the past is a good vaccination against multiple reinventions of the wheel. The book thus tries to get a handle on present-day interface theories by looking at the past: history is taken as a source of insight into those properties that a correct interface theory should have, and into those that it must not have. Given this premise, not just any way of telling about past theories of the interface will do. Everything that Part I and Part II (as well as the Interlude) report on is weighed according to its contribution to the demonstration: something is to be shown, and things that are mentioned are somehow relevant in this context. Also, a decision is made how important a given fact, analysis, period, mechanism or theory is according to the goal of the demonstration. This means that something which is taken to be a major fact about the interface may be relegated to a few lines in a sub-section, while some other item may be discussed over a whole chapter even though it does not usually appear as a relevant property of the interface in the literature. This way of proceeding may be called historiographic cherrypicking. In practice, though, the former situation where a notorious fact is hardly explored does not really occur (I hope): cherry-picking is only done in the frame of a certain ambition at exhaustivity. On the other hand, there are cases of "positive" cherry-picking, i.e. where issues that the interface literature does not (or hardly) talk about are put into the spotlight. Examples include the local (roughly, boundaries) vs. non-local (domain-based, roughly the Prosodic Hierarchy) insertion of representational objects into the phonological string (§706), modularity (§844), selective spell-out (§763) and privativity (§756). The kind of history that is told is thus not neutral or impartial: it is goal-oriented and functional. Admitting that one does not look at facts like a robot, that one does not try to marshal oneself down to strict neutrality, is but a description of reality: there is no such thing as unoriented and purely factual historiography; those who pretend that history can be told from a strictly neutral vantage point merely try to achieve a rhetorical advantage

6

Introduction

that knights their own partial view on the facts with the promise of objectivity.1 The question is not whether history-telling is partial and oriented or not; the only thing that it is worth bothering is the degree of partiality and orientedness. Everybody has his own, and the kind of bias at work may be very different. Also, the best history is not necessarily the one that is written with the least degree of partiality: Michelet's history of the French Revolution is anything but impartial, but still invaluable today. History is not a concentration of unrelated and uninterpreted facts; it is only history when it makes sense, and sense can only be made by the historian from hindsight. Science is after insight, rather than after methodological correctness, impartiality or other formal and secondary virtues: this point is constantly made by Noam Chomsky since the 60s (e.g. Chomsky 1965:20). What can be done in order to facilitate the task of the reader is to make biases explicit. This is what is done in this introduction (as much as possible), which identifies the prisms that are used in order to look at the interface.

10

2.3. No "external" history: only scholarly work is used Finally, a word is in order regarding the fact that the history of interface theories which is presented in this book is only "internal", that is, based on published (or unpublished, but written) scholarly work. Only incidentally will oral information be a relevant source. As far as I can see, there is only one case in point: Kaye's work on parsing cues (§340) and lexical access (§346), which is not available in print (for almost twenty years now). In no case is any "external" history solicited: who was friends with whom, who broke up with whom because of an unpleasant review, who was the teacher of whom and programmed his pupils to think this or that way, who was married with whom, who studied where, who has which character sketch in the opinion of whom, who wrote which letter or e-mail to whom and so on play no role in the book.

1

Regarding the history of linguistics, see Koerner (2002a:154f) for a different position that promotes a kind of enlightened positivism ("broad positivism" in Koerner's words), which is close to pure historiography (i.e. tries to show "what really happened"), declines any ambition to explain the present and to act on the future, but admits that there is no absolute objectivity, and that somebody must make sense of the "facts".

The syntactic frame: minimalist phase theory

7

Of course, this is not to deny the possibility to gain insight through the study of "external" or social factors, or to assert that they have no bearing on how interface theory developed. They certainly did to some extent. But this is simply not the way I look at the interface in this book: coming to grips with the scholarly record is serious enough a challenge; digging out social factors from oral information would require entirely different techniques of investigation.

11

3. The syntactic frame: minimalist phase theory

12

3.1. The inverted T delineates the scope of the book and serves as a referee An important backdrop of the book is the generative architecture of grammar, the so-called inverted T, which was developed in the 60s (Chomsky 1965:15ff, §§86,623 discuss its genesis at greater length) and since then stands unchallenged in generative quarters (the generative semantics interlude and Jackendoff's parallel alternative lain aside, see §24): a concatenative device (morpho-syntax) feeds two interpretative devices (PF and LF). We will follow the career of the inverted T and its impact on interface theories in Part I, but the interest is not merely historiographic: the inverted T delineates the scope of the book. That is, theories which follow a different architectural setup are mentioned, but not evaluated (see §§24,720 below). Also, the inverted T in its modern guise, i.e. phase theory, is used in Part II as a measure for the evaluation of interface theories. Finally, the minimalism-induced biolinguistic programme (initiated by Hauser, Chomsky & Fitch 2002, opposed by Pinker & Jackendoff 2005a,b) that is designed to downplay UG and properties of language that are specifically linguistic (see §§609,633,639) does not fundamentally modify the interface landscape. Even if it were true that the only specifically linguistic property of language is recursion and hence that specifically linguistic mechanisms reduce to Merge and Phase, morpho-syntax would still have to talk to phonology (and semantics), the only difference being that phonology (and semantics) are based on more general cognitive mechanisms that are not specifically linguistic (or even species-specific, see e.g. Samuels 2009a,b). If thus (what is specific to) language reduces to (certain aspects of) morpho-syntax, the question of language-internal modular structure is certainly obsolete (there is no such thing), but the relations of morpho-syntax with phonology and semantics need to be organised in the same modular

8

Introduction

environment as before. The only difference is that PF and LF are not considered linguistic anymore: morpho-syntax will talk to them like it talks to other cognitive modules such as audition and vision. The sections below discuss in which way the minimalist focus on the interface has turned the interface landscape upside down by making it interactionist and therefore allowing for what I call intermodular argumentation. Also, in necessary anticipation of Part I and Part II, we will see that a number of central properties of current syntactic phase theory have actually been invented in phonology. Note that all this only concerns the procedural side of the interface: modern phase theory is transparent for representational communication with phonology. The inverted T as such, however, impacts both procedural and representational communication.

13

3.2. Interactionism, selective spell-out and no look-back devices (PIC)

14

3.2.1. When the generative mainstream became interactionist Since its inception and until 1998/99, the inverted T was supplemented with a proviso which requires that all concatenation be done before all interpretation. That is, the morpho-syntactic derivation is first completed, and the result (S-structure) is then sent to PF and LF in one go (SPE). An alternative view of the communication between morpho-syntax and LF/PF was formulated in phonology in the early 80s: the backbone of Lexical Phonology, so-called interactionism, holds that concatenation and interpretation are intertwined. That is, first some pieces are merged, the result is interpreted, then some more pieces are concatenated, the result is again interpreted, and so on (§146). While GB-syntax of that time hardly produced any echo, generative orthodoxy in phonology reacted on this violation of "all concatenation before all interpretation": Halle & Vergnaud (1987a) propose a noninteractionist version of Lexical Phonology that restores the interface landscape of SPE on this count as well as on a number of others (§222).

The syntactic frame: minimalist phase theory 15

9

3.2.2. Selective spell-out Halle & Vergnaud (1987a) also promote a new idea: selective spell-out. Since cyclic derivation was introduced by Chomsky et al. (1956:75) and formalized in Chomsky & Halle (1968:15ff), interpretation was held to run through the bracketed string (that is inherited from S-structure) from inside out. Roughly (see §103) every morpheme break defined a cycle. Halle & Vergnaud (1987a) dispense with this definition of what an interpretational unit is: they propose to grant cyclic status only to a subset of morphosyntactic divisions. In other words, some nodes trigger interpretation, others do not. This is what I call selective spell-out, which is exactly how modern syntactic phase theory works: in more familiar terminology, nodes may or may not be phase heads, hence their material may or may not be an interpretational unit. As far as I can see, the phonological heritage is left unmentioned in the syntactic literature since derivation by phase was introduced by Epstein et al. (1998:46ff), Uriagereka (1999) and Chomsky (2000a, 2001 et passim) (§304). This is also true for interactionism: Epstein et al.'s (1998:46ff) Spellout-as-you-Merge (see §776), Uriagereka's (1999) multiple spell-out and Chomsky's derivation by phase make the generative interface architecture interactionist, exactly along the lines of Lexical Phonology: first you do some concatenation, then some interpretation, then some more concatenation etc. For (extra-linguistic) reasons of computational economy regarding the limited availability of active memory, a costly cognitive resource (e.g. Chomsky 2000a:101, 2001:15, see §305), modern phase theory applies the interactionist world view. Here again, thus, the phonological origin of the idea went unnoticed as far as I can see (let alone the anti-interactionist reaction of generative orthodoxy that was mentioned above).

16

3.2.3. No look-back devices (the PIC) The book also closely follows the footsteps of a question that is intimately related to selective spell-out and interactionism: critical for current syntactic phase theory is a device which guarantees that previously interpreted strings do not burden further computation – in Chomsky's terms, strings that are returned from interpretation are "frozen" and "forgotten" when concatenation resumes. It is only this kind of no look-back device that brings home Chomsky's promise of active memory economy.

10

Introduction

No look-back devices are around in generative thinking since Chomsky's (1973) Conditions on Transformations. Their offspring – until its recent revival in the guise of the Phase Impenetrability Condition (PIC) – was essentially phonological (e.g. Mascaró's 1976 and Kiparsky's 1982a,b Strict Cycle Condition). No look-back devices are designed to prevent computation to consider "old" strings. Depending on their precise formulation, however, they may have very different effects, which correspond to the thing that the analyst wants the computation to be unable to do. We will see that Chomsky's modern "freezing" no look-back, the PIC, is quite different from its 1973 version, and like interactionism and selective spell-out has a phonological precedent (§287).

17

3.3. Intermodular argumentation

18

3.3.1. The intermodular potential of interactionist phase theory The interactionist perspective paves the way for what I call intermodular argumentation. In contrast to GB, where the completed morpho-syntactic derivation was merely dumped into PF (and LF) with a "good bye and don't come back", phase theory establishes a two-way pipe between the morphosyntactic and the phonological (and semantic) modules. Actors on both ends are not free anymore to do what they want: their theories and analyses may make predictions on the other end. This is what Procedural First is based on (recall this notion from §6): while a particular syntactic or phonological analysis makes predictions on the other end of the pipe when the communication is procedural, representational communication does not offer this opportunity. That is, the translation of morpho-syntactic into phonological vocabulary is necessarily arbitrary and therefore never makes predictions on the other end of the pipe (§850, also Vol.2). The intermodular potential of phase theory, however, has not received much attention thus far. Syntacticians use Phase Impenetrability for syntax-internal purposes, and phase theory evolves at high speed without taking into account what happens when the parcel spends time on the phonological side. On the other hand, phonologists have barely acknowledged the existence of phase theory, let alone taken into account the predictions that it makes on the phonological side. I argue that intermodular argumentation provides stronger evidence than what can be produced by modular-internal reasoning: it offers the

The syntactic frame: minimalist phase theory

11

maximal degree of independent assessment that linguists can expect without leaving their discipline. Be it only for that reason, the new interactionist architecture that the minimalist perspective brought about is a good thing to have: after a long period of more or less waterproof coexistence, syntacticians and phonologists can talk again about things that do not concern the weather or job openings. Intermodular argumentation shines through on various occasions in the book; the different strands are then bundled in the conclusion of Part II (§841).2 19

3.3.2. Phase theory is the bridge that forces syntax and phonology to converge The existence of a single and interactionist spell-out mechanism that relates the morpho-syntactic and the phonological derivation puts pressure on both ends of the pipe that ships pieces back and forth: it is hard to see how the respective devices could not be identical. An obvious example is the fact that the pieces which are exchanged must be the same: ideally, thus, we will find phonological and syntactic evidence for the same phase boundary. This question is discussed in Vol.2: while evidence sometimes converges (e.g. for the CP), phonologically motivated chunk delineation typically remains unechoed on the other side (e.g. the word), and vice-versa (e.g. DP) (see Scheer 2009a,c). This of course does not mean that the general idea is wrong; it just points to the possibility that there are phases which do not leave traces on either side of the pipe. While this is certainly not what linguists want to hear (we expect phases to leave traces), it is an empirical question that needs to be sorted out, and the diagnostics that I am able to apply are of course hopelessly incomplete. That is, in the best case my conclusion will turn out to be wrong: every phase boundary that is based on either phonological or morpho-syntactic evidence will be found to leave traces in both phonology and syntax at least in some language. But phase theory has still a long way to go before this kind of question can be addressed, if only because it is far from being stabilised on the syntactic side. Phase theory, and especially the question how phasehood is defined, are high-ranking items on the research agenda of syntacticians, and 2

Three pieces of the book that revolve around this idea are also published separately: Scheer (2008c, 2009b, 2010b).

12

Introduction

the literature therefore produces a blooming variety of options, alternatives, extensions and refinements of Chomsky's (2000a et passim) original take according to which only vP and CP are phases. It is not easy for a phonologist to keep track of this field which evolves at high speed (probably syntacticians are puzzled as well), but I did my best to find out about the general direction, which is clearly towards the atomisation of phasehood, i.e. the recognition of smaller and smaller chunks that are granted phase status (see §771). Obviously, it is difficult to hunt down phonological evidence for syntactic phases if the set of what counts as a syntactic phase is constantly moving.

20

3.4. Focus on the spell-out mechanism(s?)

21

3.4.1. Minimalist interface orientation: spell-out marshals both morphosyntax and phonology The study of the mutual intermodular conditioning of morpho-syntax and phonology slowly shifts interest away from actual morpho-syntactic and phonological computation. Instead, the spell-out mechanism comes to stand in the spotlight. This is but to be expected in a modular perspective where morpho-syntax, phonology and semantics are input-output systems that carry out computation, but are blind for what happens before or after their activity. In this context, the following questions arise: is Phase Impenetrability a property of phonology? Or of morpho-syntax? Does phonological computation decide to ignore "old" strings? And if so, does morpho-syntax single out exactly the same "old" strings that are to be "forgotten"? This is not very likely a scenario. Rather, the central mechanism where interpretational units (i.e. phases) are defined, and where decisions about Phase Impenetrability are made, is the device that is responsible for shipping: the spell-out mechanism. This means that things which have been (or are) thought of as properties of morpho-syntax or phonology may turn out to be properties of the spell-out mechanism. Cases in point are interactionism, Phase Impenetrability and the phase edge (spell out your sister!, see §765, Scheer 2008c). Altogether six devices are discussed in §851 that could be common to morpho-syntax and phonology, or rather, which characterise the spell-out mechanism that itself is common to both modules.

The syntactic frame: minimalist phase theory

13

During the discussion of what it takes for an intermodular argument to bite, an old question crops up that is not likely to be solved tomorrow: we would need to know whether morphology and syntax are the same or two distinct computational systems. On this depends the possibility of having more than one spell-out mechanism: if morphology and syntax are independent, each could come with its own spell-out mechanism. In this case, there would be no guarantee that the properties of the spell-out of morphemes (interactionism, Phase Impenetrability etc.) are the same as what is encountered for the spell-out of words. If on the other hand morphology is just the lower piece of syntax, there can be only one spell-out mechanism, which means that intermodular predictions are much more precise (see also the word-spell-out mystery discussed next and Scheer 2009b). 22

3.4.2. The word-spell-out mystery Examples where the cyclic spell-out of morphemes leaves phonological traces are commonplace: stress placement in [[párent] hood] and [parént-al] for instance is a direct function of the fact that the root is a phonologically relevant domain in the first case, but not in the second. The same structure [[A] B], however, does not appear to have any impact on phonology if A and B are words, rather than morphemes. As a matter of fact, the literature (on external sandhi) does not document any cases where the contrast between [[A] B] and [A B] produces a phonological effect when A and B are word-sized or larger chunks. This appears to be truly mysterious: hierarchical structure is certainly not the privilege of pieces that are smaller than words. On current minimalist assumptions (interactionism), morpho-syntactic structure is sent to phonology piecemeal (derivation by phase). Could a situation then be imagined where phonology is sensitive to chunk size? That is, where it reacts under the piecemeal fire of morphemes, but simply ignores the fact that it is also hit by successive waves of words? The default assumption is certainly that an interpretational system is sensitive to its input conditions, and in any case there is no reason for one particular chunk size (morphemes) to provoke a phonological reaction, while another chunk size (words and larger items) leaves phonology unimpacted. This is what I call the word-spell-out mystery (see §§786,851). The classical solution in phonology was developed by Lexical Phonology, where an on/off switch for cyclicity is proposed for morphology and syntax (which are then of course independent computational systems):

14

Introduction

cyclic interpretation (i.e. piecemeal fire) is turned on for the interpretation of morphemes (lexical phonology), but off when words are computed (postlexical phonology). Some doubt may be reasonably entertained whether the empirical situation is really what the literature describes (i.e. that there are no cases where the cyclic spell-out of words leaves phonological traces). In case it turns out to be true, though, that phonology is insensitive to the cyclic spell-out of words, it is hard to see how chunk-specific phonologies (i.e. distinct computational systems that process distinct pieces according to their size) could be escaped. §794 explores a solution that does not rely on two distinct computational systems where an on/off switch regulates phonological computation. Rather, the on/off switch is on Phase Impenetrability. That is, the spell-out mechanism decides whether old strings are or are not submitted to phonological computation: if they are not, a PIC-effect is encountered (i.e., a phonological trace of cyclic spell-out is produced); in case they are, no PIC-effect is observed. This configuration derives what is called postlexical phonology in Lexical Phonology: word sequences get away without any phonological trace of cyclic derivation.

23

3.4.3. We need to know more about the spell-out mechanism The bottom line of all this is that the book raises a number of questions, but hardly provides any answers. The questions, however, are valuable I think, because they confront syntacticians with phonological issues that all of a sudden may become vital for them. This is what intermodular argumentation is about. The same holds true for the status of spell-out, which is a question that emerges as this introduction is written: as far as I can see, the generative literature has not worked out very detailed accounts of how spell-out actually works, what it does, what it is unable to do etc. One result of the book is certainly that the spell-out mechanism is the central piece of interface theory: much depends on its properties – and on its status. It can hardly be just some appendix to the syntactic derivation, as it is sometimes thought of. Spell-out must be able to read morpho-syntactic vocabulary and structure, and – on the assumption that interpretation is chunk-specific – to operate a distinction between different chunk sizes. The question, then, is whether this "intelligent" behaviour requires the spell-out mechanism to be a computational system in its own right

The syntactic frame: minimalist phase theory

15

which, in a modular perspective, amounts to saying that it is a module. This would come close to the architecture of Prosodic Phonology (§379) and the Jackendoffian picture (Jackendoff 1997, 2002) where an interface module mediates between the three modules of the inverted T (see Vol.2 for the latter). It may thus be the case that the procedural side of the interface coin provides evidence for an "intelligent" mediator à la Jackendoff. This is precisely what Michal Starke (and Vol.2) argue against based on the evidence of the representational side of the coin: One-Channel Translation holds that the transformation of morpho-syntactic into phonological vocabulary is done through a lexical access, rather than through computation (see §49 below). A split into computational mediation on the procedural, but lexical transmission on the representational side does not look like a viable architectural option. The alternative is to conceive of a spell-out mechanism that is non-computational, but still able to distinguish different chunk sizes (if these are really empirically relevant, i.e. if the word-spell-out-mystery turns out to be empirically correct). At this point I give up speculating: this is as far as the venture of the book takes the reader. What is for sure, though, is that the spell-out mechanism, its precise attributions and its eventually modular status deserves to be upgraded on the research agenda. Work in this area is done for example by Idsardi & Raimy (forth) and Samuels (forth), as well as in Tromsø by Michal Starke and colleagues. That the mechanism which is responsible for the shipping of pieces between morpho-syntax and phonology is understudied is also shown by the muddy waters that people (well… syntacticians) usually refrain from naming and describing when they talk about "PF": it is obvious that PF is distinct from phonology (a lot of things are supposed to happen "at PF" and to be "phonological" that are truly miraculous for poor phonologists); at the same time, the minimalist perspective on language issues a strong demand to outsource important labour into some ill-defined intermundia that is neither narrow (morpho-)syntax nor phonology (clean syntax, dirty PF/phonology, see §§30,726).

16

Introduction

24

4. Definition of the object of the study

25

4.1. The book is only about interface theories that follow the inverted T The discussion of the inverted T model in §12 has an appendix, which is what we turn to now. The book is about the interface of morpho-syntax and phonology. The title, "How morpho-syntax talks to phonology", however, is explicit on a restriction: communication is only considered in one direction. Eventual phonological influence on morpho-syntax lies beyond the scope of the study. This of course does not mean that it is not mentioned. Relevant discussion around the principle of phonology-free syntax is provided in §412 (also §§253,645) and actually produces a robust empirical generalisation that is significant in a modular perspective: melody (i.e. phonological objects below the skeleton) and morpho-syntax are entirely incommunicado – there is no conditioning in either direction (§660). Beyond that, the bearing of phonology on morpho-syntax is only relevant in this book for the sake of completeness, and insofar as it makes a contribution to the interface management in the other direction. The same is true for two other restrictions of the scope of the book: interface theories that do not follow the inverted T, and the kind of intermundia where objects of wonder such as PF movement and deletion of entire sentences are supposed to occur. These cases are discussed in §26 and §30, respectively. The core of the inverted T is syntactico-centrism: there is only one device where pieces are glued together (morpho-syntax or, if morphology is independent, morphology and syntax); this device exchanges pieces with two interpretational systems, where they are assigned a meaning (LF) and a pronunciation (PF). Morpho-syntax has the privilege of Merge and Phase, i.e. the ability to concatenate pieces and to talk to other modules (Hauser et al. 2002). Thus LF and PF do not concatenate anything: they only interpret what they receive. There are (at least) two currently entertained approaches to the interface which propose a different scenario: theories of representational continuity between morpho-syntax and phonology (and semantics), and the parallel-construction model. The former is the strand of HPSG and related work (§513), the latter is developed by Ray Jackendoff (1997 et passim).

Definition of the object of the study 26

4.2. Interface theories that lie beyond the inverted T model

27

4.2.1. Theories where everything is scrambled: HPSG

17

HPSG and Jackendoff's parallel model are not at the same distance of the Chomskian inverted T: HPSG was generative at some point, but today is quite distant from Chomskian linguistics and generative concerns. An important factor of division is the architecture of grammar and the interface of morpho-syntax with phonology: HPSG denies that there is any. In HPSG, representations are monostratal. This means that they are fully informed of morpho-syntactic, semantic and phonological information, which is available at any point in the derivation (actually, talking about a derivation is improper because of monostratalism). In practice, thus, terminals and nodes of the "syntactic" tree carry phonological information, to the effect that there is no need for lexical insertion or for any procedural or representational communication between morpho-syntax and PF/LF: everything is one big tree, and all information is continuously available. The issue (or rather: one issue) that HPSG has with the generative approach is thus about modularity: there are no modules in the HPSG landscape. This, in turn, means that there can be no interface: an interface requires two distinct entities that communicate. In a fully scrambled everything-is-one environment, though, no such entities can be identified. HPSG may thus talk about the relationship of morpho-syntax and phonology, but hardly about any interface. The phonological affiliate of HPSG is Declarative Phonology (Scobbie 1996, Coleman 2005). The book discusses the HPSG-based work of Orgun (1996a et passim) on one occasion (§512), but only for the reasons mentioned, i.e. in the interest of completeness and contrast with some other demonstration. 28

4.2.2. OT and its connectionist endowment: a programmed trope for scrambling all into one In this context, OT needs to be mentioned as well. Like HPSG, OT has a natural tendency to scramble everything into one big constraint hierarchy (see §523). Of course there are many different degrees of scrambling also among OT practitioners, but the tendency is towards a single grammatical space where anti-derivationalism is enforced globally (see Scheer 2010a).

18

Introduction

The most visible result of this trope is anti-cyclicity, i.e. the denial of inside-out interpretation (see §464). Other diagnostics for OT's misty (and largely unreflected) relationship with modularity are the customary and uncontradicted violations of Indirect Reference (i.e. the prohibition to make reference to untranslated morpho-syntactic categories, see §377), the fact that mapping (of morpho-syntactic into phonological categories) is done in the phonological constraint hierarchy (ALIGN and WRAP are interspersed with purely phonological constraints) and the common occurrence of constraints whose formulation combines phonological and morphological instructions. It is suggested in §529 that this scrambling trope is a consequence of the other half of OT's genetic code: it is certainly true the OT is a generative theory, but it is also true that its central tool, constraint interaction and parallel assessment of competitors, was conceived in direct reference to connectionism, the theory that competes with modularity for the description of the cognitive architecture (see §§36,586 below). Paul Smolensky is one of the founders of OT and of connectionist theory (e.g. Smolensky 1987, 1988a). Connectionism, however, is the polar opposite of generative and rational thinking: it is empiricist, anti-symbolic and anti-modular (see §§588,598). Knowing about the connectionist roots of OT thus helps to understand the extreme computational orientation of phonology in the past decade and a half that was discussed in §7. It is sometimes rightly recalled that OT is a theory of constraint interaction, not of constraints. This means that OT does not supply any substance itself: there are genuine vocabulary items in structuralism (phonemes), SPE (segments) and autosegmental theory (autosegmental structure), but there are no OT-specific representational items. OT uses whatever representational material comes the way, and may well produce the same result with entirely different (and mutually incompatible) vocabulary.3 It is difficult not to establish a direct relationship between the fact that OT is a purely computational theory where representations make no sovereign contribution to the definition of grammaticality (which is decided by constraint interaction alone, see Vol.1:§309) and its content-free connectionist prototype. 3

For example, Lombardi (2001:3) writes with respect to melodic representation: "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."

Definition of the object of the study

19

The trouble is that modularity is one of the deepest layers of generative thinking (see §§603,623). Two souls alas! are thus dwelling in the breast of OT, and the sparks of their encounter may be observed at the interface.

29

4.2.3. Jackendoff's parallel model: all modules are structure-building Unlike HPSG, Jackendoff's parallel model (which is described at greater length in §722) represents an alternative to the inverted T within the generative paradigm. Morpho-syntax, semantics and phonology are modules, which means that there is an interface just like among the modules of the inverted T: procedural and representational communication needs to be organised between entities that do not speak the same language (of the mind). At variance with the inverted T, however, all modules are granted the ability to build structure, i.e. to glue pieces together. Morpho-syntactic, semantic and phonological structure is built in parallel, and information is exchanged among modules at any time in the derivation when this is necessary for the construction on any of the three sides. This architecture has consequences for the communication among modules: beyond procedural and representational communication, it allows for additional types of interaction. Bendjaballah & Haiden (e.g. 2003a,b, 2007) for example argue that the parallel construction in the three modules is homomorphous, i.e. equally advanced, and that this is controlled for by constant communication among them.

30

4.3. PF, an androgenic intermundia

31

4.3.1. The minimalist dustbin: clean syntax, dirty phonology The heart of the minimalist programme is the ambition to clean syntax from everything that is not "perfect" in the Chomskian sense, i.e. that is not motivated by interface requirements ("bare output conditions"), or by a general condition on computational efficiency. This direction is the source of the notorious interface-orientation of minimalist syntax, of which we have seen an effect in §11: the interface has become interactionist, and phase theory paves the way for intermodular argumentation. This is certainly all to the good.

20

Introduction

The other side of the coin, however, is the creation of a kind of intermundia where things are unloaded that syntacticians do not want to accommodate in syntax, but which are not phonological either. In the current environment, one is well advised to add that "phonological" in this context refers to what phonologists call phonology: there is a fair amount of confusion in PF-oriented syntactic quarters where PF and phonology are used as synonyms. Typically, what syntacticians call phonological when they talk about PF-outsourced syntactic operations has got nothing to do with phonological computation (§731): there is an ill-defined, minimalism-born intermundia between spell-out and vocabulary insertion on the upper and phonological computation on the lower end. Until the mid-90s, the PF of the inverted T model was more or less coextensive with phonology; minimalism has shrunk syntax and pumped up PF, which is now made of phonology plus "something else". This murky additional workspace obeys rules that are quite different from what is known from anywhere else: locality conditions are different from syntax (§580), it violates modularity by simultaneously accessing morphosyntactic and phonological vocabulary (§738), and it operates with a strange notion of hierarchy where the nodes of the PF tree are the projections of nothing – at least not of the terminals, which are phonological (see §747 for a summary). §726 inquires on the properties and the internal structure of PF, as well as on the phenomena that PF is supposed to handle. Things that syntacticians want to make PF responsible for include head movement (Chomsky 1995a, chapter 4), various kinds of ellipsis (e.g. Merchant 2001) and clitic placement (e.g. Bošković 2001). Typically, what PF is expected to do is to make things disappear: "delete at PF" (see §§732f). Syntacticians seem to use "PF" as a magic word – pronounce it and get rid of your trouble. It may be reasonably asked whether anything is gained when a clean syntax is bought at the expense of an ill-defined buffer that emerges out of the blue and where established principles of linguistic analysis do not hold. Dumping displeasing things into PF is not analysing or solving them, and being allowed to analyse them with all kinds of ad hoc mechanisms (fission, fusion etc., the most exotic animal of the PF zoo being PF movement, §§574,738) that are unheard of elsewhere may turn out to be a remedy that is worse than the disease.

Definition of the object of the study 32

21

4.3.2. Syntax, morphology, PF On this backdrop of a minimalism-born buffer between narrow syntax and phonology, the book is only about the interface of "narrow phonology" with morpho-syntax. The PF intermundia does not appear in interface theories because it is too recent. It is therefore absent from the discussion in Part I (except in the chapter on Distributed Morphology) and appears only in Part II (§726). Of course the book does not try to find out exactly which pieces of morpho-syntax are narrow syntax and which pieces belong to the PF intermundia: this issue is debated by a large body of syntactic literature. Interface theories are phonologically oriented (they describe the influence of morpho-syntax on phonology, not the reverse), and they are made by phonologists (with Distributed Morphology being half an exception); like these, the book thus trades with morpho-syntax as a whole without participating in the debate about the division of labour between narrow syntax and PF. The situation is a little different for another question that is much debated in the syntactic literature, i.e. whether syntax and morphology are a single computational system or two distinct modules. This issue escorts the reader through a good deal of the book: it was brought to the forefront of interface discussion by Lexical Phonology (lexical vs. postlexical phonology, §153) and is the bone of contention for Distributed Morphology (§533). It also plays a central role in the aforementioned word-spell-outmystery (§§786,851).

33

4.4. Modularity is the touchstone It is not the case that the preceding pages are devised to evaluate the theories and mechanisms mentioned. They just explain what the book is about, and what it does not consider. Those theories and devices that are not in its focus may or may not be interesting or correct: the book does not argue about that, even though my personal view may shine through on occasion. There is one criterion, though, that is applied to all theories and devices discussed: following the generative tradition, the book is committed to the modular architecture of the mind, and hence of language (modularity as a principle of cognitive organisation is introduced at length in the Interlude §586, also §36 below). We will see that various generative theories have been at odds with modularity in various ways over time (see §702 for a summary). This notwithstanding, modularity is considered a necessary

22

Introduction

property of a correct interface theory: a law is not invalidated because it is disobeyed here and there. Among the theories and devices quoted, HPSG and PF movement are definitely incompatible with modularity, but Jackendoff's parallel architecture or the outsourcing-created intermundia are not. That is, it could turn out that the modular architecture of language is not syntactico-centristic, i.e. that PF and LF also enjoy the privilege of concatenation. Finally, modularity also automatically draws a red line between the book and functional approaches to language. These may have rule systems much along the lines of modular computational systems,4 but will always allow for extra-grammatical forces such as ease of articulation and communicative success to bear on grammar. A module, however, knows only its own law and carries out computation in complete absence of any external influence.

34

5. Trying to get an independent handle on the interface

35

5.1. Intermodular argumentation, history Like other theories, interface theories are built on data and on a conceptual background. The book of course moves along these lines. In addition, however, the book tries to get a handle on interface theories by three independent referees: its historical approach, modularity and intermodular argumentation. The historical work indeed affords a certain degree of independence from individual interface theories and fashions: it allows making generalisations and discovering patterns that are not usually mentioned in the literature (and it also affords not to go in circles or to reinvent the wheel). Some examples were already mentioned in §9: local (roughly, boundaries) vs. non-local (domain-based, roughly the Prosodic Hierarchy) insertion of representational objects into the phonological string (§706), selective spell-out (§763), no look-back devices (the Strict Cycle Condition in its various brands, today Phase Impenetrability) (§287), privativity (§756). The historical strategy is complemented by two other instruments: intermodular argumentation and modularity. The former was already men4

Cast in Natural Phonology, the Beats and Binding model of DziubalskaKołaczyk (2001b, 2002) for example is a lateral approach to syllable structure: bindings are (bidirectional) relations among adjacent segments..

Trying to get an independent handle on the interface

23

tioned in §§17ff (see the summary in §841): it correlates the behaviour of phonology-oriented interface theories with the requirements of current syntactic phase theory and thereby achieves an independent judgement. For example, if minimalist interface orientation and the strive for computational economy is on the right track, it must apply to all components of grammar. Active memory must thus be unburdened in phonology as much as in syntax: phonological computation must also be able to "forget" strings that have already been interpreted (Chomsky 2001:12f is explicit on this, see §306). This means that Phase Impenetrability must be active in phonology, and that interface theories which do not use this device at all such as Lexical Phonology (for the management of affix class-based phenomena) are on the wrong track (see §828). 36

5.2. Modularity

37

5.2.1. Generative grammar deeply roots in modularity, but is often offended The third independence-fostering instrument is modularity. It was already mentioned in §33 that modularity is used as a referee for interface theories in the book: the cognitive system is modular in nature (and it does not matter for interface theory whether PF and LF are language-specific or language-unspecific modules, the latter option being entertained by the biolinguistics programme, see §§12,633,639). Morpho-syntax and phonology are distinct modules that carry out computation on the basis of domain specific vocabulary. Since they are incommunicado by themselves, their communication needs to be organised: modularity requires interface activity (§650). Modularity is one of the deepest layers of generative thinking (see §623), and its presentation in the Interlude reflects its central status: in the 50s, Noam Chomsky participated in the development of the general computational paradigm (Turing - von Neumann, see §603) that underlies much modern science and grew into the standard paradigm of how the mind works (Cognitive Science). On the grounds of Artificial Intelligence and 19th century faculty psychology (F-J Gall's phrenology, see §§601f), the modern formulation of modularity is due to Fodor (1983). Language takes a prominent place in Fodor's book, which has grown out of a class co-taught with Chomsky. Despite the fact that modularity is so deeply rooted in generative thinking, it is not a loss of time to recall all this – and to apply the modular referee to phonological and interface theories. This is one thing that I had to

24

Introduction

learn while reading through the literature: modularity usually appears in introductory classes to linguistics and in first chapters of linguistic or phonological textbooks – but then disappears from the radar. It is not uncommon for theories to explicitly subscribe to the modular architecture of the mind in general and of language in particular, but then to live in overt violation of modularity. The list of modularity-violating generative interface theories (something that should be a contradiction in terms) is examined in §702: since SPE through so-called direct syntax approaches in the 80s up to OT, Distributed Morphology (PF movement), and the misty PFintermundia that results from the minimalism-driven outsourcing of phenomena into PF (§726), different theories violate modularity in different ways and for various reasons. 38

5.2.2. Modularity in the history of generative grammar: the GB-interlude of syntax-internal (nested) modules A related aspect of the problem is the narrowly linguistic horizon that generative linguists typically have when they talk about modularity. The bridge with Cognitive Science does not accommodate much traffic: linguists may not know what it takes to be a cognitive module, what its properties are, how it works, how it is defined, how it is detected and so forth. §622 traces back the history of modularity in generative grammar. Since his earliest writings (LSLT, Chomsky 1955-56), Chomsky has described the functional units of grammar as computational input-output devices that work on a proprietary vocabulary (even though he used the terms of the time, which may be unfamiliar today and blur identification) (§623). This was condensed into the inverted T model in Aspects (Chomsky 1965), and the inverted T where a central concatenative device (morpho-syntax) feeds two interpretative devices (PF and LF) is the baseline of generative thinking up to the present day (§629). Until GB, the generative architecture was thus made of three computational systems which had all modular characteristics and were described as such (see for example the quote from SPE in §613), but were not called modules. This word became a standard in Cognitive Science only with Fodor's (1983) ground-laying book, which emerged from a class that Fodor and Chomsky co-taught in fall 1980. The major innovation of the new development in generative grammar that Chomsky was about to introduce then (and which was launched in written form in Chomsky 1981, 1982) was entirely based on the modular

Trying to get an independent handle on the interface

25

idea, and now used the word module: the central idea of Government and Binding is to cut syntax down into six sub-systems, or sub-theories (theta theory, government theory etc.). This move was supposed to provide a handle on syntactic complexity, which was out of reach, Chomsky argued, if approached with a single system. In GB, syntax is thus viewed as the result of the interplay of a number of fairly simple basic systems whose workings the linguist can hope to understand (§628). This left the general architecture with a kind of nested structure, and quite tacitly so: the focus was on the GB-subtheories, or modules, and the macro-modules of the inverted T, as well as their relationship with the GBmodules, were left without much discussion. A fair question is thus what kind of status a nested modular structure has, and indeed whether GBsubtheories qualify as cognitive modules in the Fodorian sense in the first place (§632). Also, the existence of two types of quite different units that are called modules (the endpoints of the inverted T and GB-subtheories) was a source of confusion and puzzlement outside of generative quarters (§634). Finally, the minimalist (and biolinguistic) turn brought generative grammar right back to where it started in the 60s: GB-modules are done away with, and the inverted T is more important than before in an environment where syntax is shaped according to interface-induced pressure (§637).

39

5.2.3. The refereeing potential of modularity lies waste It was mentioned that modularity in generative grammar was typically considered in a narrowly linguistic (or even syntactic for the GB period) perspective. I have come across two concrete cases that are directly related to this issue, i.e. where an obvious argument from the properties of cognitive modules was left unmobilised in debates where it would have made a decisive contribution. One is the quarrel that opposed so-called direct syntax approaches to Prosodic Phonology in the 80s (see §407). The founding statement of Prosodic Phonology is the principle of Indirect Reference which says that phonological rules cannot make direct reference to morpho-syntactic categories; rather, morpho-syntactic information needs to be translated into phonological objects (the Prosodic Hierarchy) in order to be able to bear on phonological computation.

26

Introduction

This is the exact description of a modular relationship: modules cannot look into other modules because they would not understand their idiom. They can only communicate through translation (§650). As far as I can see, though, the modular argument is entirely absent from the debate: although Fodor (1983) was contemporary, it was not used by defenders of Indirect Reference (instead, the decisive argument was taken to be so-called nonisomorphism, which turns out to be a non-argument, see §416). The other case in point where the modular referee was not called upon is interactionism: Lexical Phonology introduced this way of piecemeal communication in the 80s, but even under anti-interactionist fire (see §222) did not use the argument that the interactionist architecture is the only way to reconcile inside-out interpretation (i.e. cyclicity, to which everybody subscribes) with modular requirements (more on this in §680).

40

5.2.4. Introduction to (Fodorian) modularity The point that intermodular argumentation offers the maximal degree of independence from phonology within the realm of grammar was already made. Modularity takes this independence one step further: it is able to referee linguistic theories from outside of grammar. Language is modular, and the computational systems that contribute to it (whether specifically linguistic or not), as well as their communication, are subjected to the same requirements as modules that compute other cognitive functions. The modular referee is thus taken seriously in this book: modularity will contribute its arbitral award to every issue. This of course is mainly done in Part II where theories are evaluated. In order to prepare the discussion, modularity is introduced from a Cognitive Science perspective in the Interlude (§586): where it comes from (phrenology, F-J Gall's faculty psychology), where it stands in the philosophical landscape (it is rational/mentalist), which is its competitor (empiricist connectionism), which are the issues at stake (e.g. symbolic vs. content-free), how modules are identified (double dissociation, domain specificity), examples of modular analyses of other cognitive functions (e.g. the number faculty), its application to language, the predictions it makes and the requirements it issues.

Deforestation 41

27

5.2.5. Structuralist and generative modularity Finally, it is interesting to observe the convergence of structuralist and generative thinking in regard of modularity: §692 shows that translation (of morpho-syntactic into phonological objects), an important consequence of modularity, was actually invented by structuralism, where it was enforced by Level Independence (see also §72). Just like generative theory, but without any cognitive background, structuralist Level Independence considers morpho-syntax and phonology two distinct ontological entities that are incommunicado as such. The structuralist-generative unity further adds to the weight of the modular argument.

42

6. Deforestation

43

6.1. The core of Government Phonology: lateral, rather than arboreal syllable structure It was mentioned in the foreword in which way the book is related to Vol.1. The project of Vol.1 is to build a lateral alternative to the traditional arboreal conception of syllable structure: Vol.1:§165 explains at length in which way replacing arboreal structure by lateral relations (government and licensing) is the core of the research programme of Government Phonology. In a nutshell, the idea is that the syllabic position of a segment is not defined by a constituent to which it belongs (and whose status is itself defined by the arboreal relations that it entertains with other constituents), but by lateral relations that hold among constituents. For example, a consonant does not show characteristic coda behaviour because it belongs to a constituent "coda" whose mother is the rhyme; rather, coda behaviour is due to the fact that relevant consonants occur before a governed empty nucleus which is unable to provide support (licensing). This explains why coda consonants are weak, rather than strong (while the weakness of the coda constituent does not follow from anything). Standard Government Phonology (Kaye et al. 1990) introduced the lateral project, but ran out of breath half-way: the result is a hybrid model where lateral relations cohabitate with arboreal structure that is left over from the traditional tree-based approach. On many occasions, lateral and arboreal structure do the same labour, which is an intolerable situation for

28

Introduction

sure (this was correctly observed by Takahashi 1993 early on, see Vol.1:§208): either syllable structure is lateral or it is arboreal – it cannot be both. Hence if the lateral project is worth being explored at all, it must be applied all the way down. This is what Lowenstamm's (1996) idea is about: arboreal syllable structure is done away with altogether (constituents reduce to a strict sequence of non-branching onsets and non-branching nuclei), and lateral relations alone define syllabic positions. Vol.1 works out the conditions of these premises.

44

6.2. The lateral project leaves no place for arboreal prosodic constituency The result at the end of Vol.1 is a (fully) lateral theory of phonology – or rather, of syllable-related phonology. For there are other areas in phonology where arboreal structure is traditionally assumed: below the skeleton for the representation of melody (Feature Geometry), above the skeleton for the representation of morpho-syntactic information (the Prosodic Hierarchy). While privative melodic representations (Anderson & Jones 1974 and ensuing applications in Dependency Phonology, Particle Phonology and Government Phonology) provide a non-arboreal alternative for the former, the Prosodic Hierarchy stands unchallenged in the latter area. The question is thus whether a scenario is viable where arboreal structure is absent from all areas of phonology except for the representation of morpho-syntactic information. This ties in with Lowenstamm's (1999) idea that morpho-syntactic information can be represented by an empty CV unit, i.e. a non-arboreal object that is inserted locally into the linear string. Also, the initial CV is part and parcel of the Coda Mirror (Ségéral & Scheer 2001, 2005, 2007, 2008, Vol.1:§§83,110). There is thus reason to question the arboreal standard of representing morpho-syntactic information: if it is represented in terms of objects that are inserted into the linear string rather than by prosodic constituency, phonology as a whole has a non-arboreal perspective. But there is also positive evidence that pleads against the Prosodic Hierarchy, which turns out to be a diacritic upon closer inspection (§402, Scheer 2008a). If diacritics do not qualify, this is reason enough for the Prosodic Hierarchy – and hence for the arboreal representation of morpho-syntactic information – to be counted out. On this backdrop, Direct Interface, which is introduced in Vol.2 (also Scheer 2008a, 2009a,c), is an attempt to make the representation of morpho-syntactic information 1) non-arboreal, 2) local and 3) non-diacritic. It

Deforestation

29

completes the deforestation of phonology by doing away with the last piece of traditional arboreal structure. The historical inquiry of this book is a consequence of Direct Interface: it was mentioned in the foreword that by looking at the history of interface theories I originally wanted to make sure that I am not reinventing the wheel. The causal chain of the book thus runs from the inception of the lateral project in the late 80s over Vol.1 to the deforestation of phonology, Direct Interface (Vol.2) and the history of interface theories (this book). It is therefore useful to expose this link in the introduction of the book: even though Direct Interface is only introduced in Vol.2, and although the discussion of Prosodic Phonology (§360) and the local (i.e. nonarboreal) vs. non-local (i.e. arboreal) perspective on the insertion of representational carriers of morpho-syntactic information (§687) can stand alone, the reader should be given the means to follow the global project and the role that is played by the chain link of this book. In this perspective, the following section shows that the deforestation of phonology is also independently motivated: the phenomena that are expected to result from arboreal structure (such as recursion) are absent from the record.

45

6.3. Recursion and other expected consequences of trees are absent in phonology Part and parcel of the inverted T model is that only morpho-syntax has the privilege of concatenation: phonology and semantics merely interpret; they are not equipped for gluing pieces together. In the minimalist environment, concatenation is the result of Merge. This operation is thus available in morpho-syntax, but not in phonology and semantics. Phonological theories, however, have always relied on tree-building devices, at least since autosegmental structure is used. While feature geometric trees are lexically specified, syllabic and prosodic arborescence is assumed to be the result of online tree-building activity, today as much as in the past. A classical example are syllabification algorithms, which build arboreal syllable structure on the basis of the segmental properties of the lexically unsyllabified linear string. It is true that phonological trees do not involve any concatenation of pieces (they are built on a pre-existing linear string): this is what makes them different from morpho-syntactic trees. As a consequence, though, phonological and morpho-syntactic trees are not the same thing. Hence if

30

Introduction

any, the phonological tree-building device is different from morphosyntactic Merge. Accommodating distinct Mergem-synt and Mergephon in grammatical theory of course ruins the minimalist ambition, which counts on only one universal piece-gluing (and hence tree-building) device. But there is more reason to believe that a tree-building phonological Merge cannot be the correct scenario. Neeleman & van de Koot (2006) show that trees of whatever kind have certain formal properties that make predictions on the type of phenomenon that should be found in a treebearing environment. These include projection, long-distance dependencies and recursion. Neeleman & van de Koot (2006) demonstrate that phonological phenomena do not display any of these properties. They therefore conclude that the presence of trees in phonology overgenerates: arboreal structure predicts things that are absent from the record.

46

6.4. The lateral project predicts that phonology is non-recursive The same point can also be made from the other end. There is no phonological equivalent to multiple phrasal embedding, where the only limit on the number of recursions is set by performance restrictions (see §803 for examples, also from morphology). The absence of recursion has long been recognised as a major difference that sets phonology apart from morpho-syntax. Everybody knows about the fact, which is undisputed,5 but still begs the question: there must be a reason why phonology is not recursive. Nespor & Vogel (1986) for example make the difference explicit, but leave it at that.

5

The phonological part of the recent literature on recursion which was generated by Hauser et al.'s (2002) idea that recursion (and hence Merge) could be restricted to narrow syntax often falls prey to the confusion between recursive phenomena and the analysis thereof: recursion is a phenomenon whose existence is established by pre-theoretical and pre-analytic properties, not by analyses that happen to use recursive constructions. The existence of the latter does not document recursion in phonology, but merely the fact that some analysts use recursive constructions. In syntax and morphology, recursion is a phenomenon whereby you can keep repeating the same type of item indefinitely until grammar-external limits regarding memory etc. are reached. Nothing of that kind has ever been reported from phonology. More on this confusion in §803.

Structure of the book and of Vol.2 (1)

31

"In relation to the difference between the morpho-syntactic and prosodic hierarchies, it should be noted, furthermore, that the two differ not only in the way they divide a given string into constituents. They also differ with respect to depth. That is, since the rules that construct the phonological hierarchy are not recursive in nature, while the rules that construct the syntactic hierarchy are, the depth of phonological structure is finite, while the depth of syntactic structure is, in principle, not finite." Nespor & Vogel (1986:2)

What Nespor & Vogel say is that there is no particular reason why syntactic rules are recursive, but phonological tree-building rules are not. In other words, the absence of recursion in phonology is accidental in their system: phonological rules happen not to be recursive, but could well be. By contrast in a phonology where trees are absent altogether because interpretational devices have no access to the tree-building device Merge, the absence of recursion is predicted. This is because recursion is formally defined as a node that is dominated by another node of the same kind: if a computational system is unable to build trees, there can be no domination at all, and hence no recursive phenomena (this was also pointed out in the foreword to Vol.1; further discussion is provided in §§802ff). The absence of recursion in phonology is thus predicted by the lateral project and its concomitant elimination of trees.

47

7. Structure of the book and of Vol.2

48

7.1. How to access the book: the story, its relation with current syntactic theory and its thematic guide As was mentioned, the book falls into two parts and an Interlude. Part I is historiographic: it reviews structuralist and generative interface thinking chronologically, but also proceeds theory by theory. The Interlude (§586) introduces modularity, the rationalist theory of the (human) cognitive system that underlies the generative approach to language (but is also incarnated by structuralism through Level Independence, see §692). It was explained in §36 in which way modularity is a key concept of the book. Part II serves two functions: it distils lessons from the review of interface theories on the one hand (a how-to is provided at the outset in §656), and locates the interface debate in the landscape of current minimalist syntax on the other hand.

32

Introduction

Regarding the latter, Part II introduces to ongoing debate in syntax that is by and large absent from (traditional, but also more recent) interface theories: linearisation is discussed in §741, the blown-up PF area into which minimalism outsources a whole lot of mechanisms that were previously syntactic and are still not phonological (despite the P of PF) is examined in §726, and §771 reports on phasehood in current syntactic thinking. Regarding the former, answers to the following questions are sought: which ideas are theory-resident? In which different guises does a given idea appear over time? Which are the mechanisms that have survived the verdict of time? Which empirical generalisations emerge? Which are the watershed lines that separate interface theories into different camps? Which are the questions that interface theories are constantly after? Which ones are solved, which ones remain pending? Part II is thus a thematic guide to Part I: instead of approaching the interface period by period and theory by theory, relevant topics are addressed across periods and theories, which are accessed by systematic cross-reference for each issue. Also, Part II to a certain extent abandons the journalistic style that is used as much as possible in Part I: the goal now is to evaluate, to assess and to tell good from bad, correct from incorrect, plausible from implausible, successful from unsuccessful, rather than to report. This setting inevitably introduces a certain amount of repetition: like Vol.1, the book is not expected to be read from cover to cover. The reader will rather enter the book by looking up what a specific theory proposes (access by theory), what was on the research agenda at a given point in time (chronological access), or how theories deal with a specific phenomenon or generalisation (thematic access). Repetition and cross reference are necessary in order to keep the information level constant for all types (and points) of access, but it is true that for the cover-to-cover reader this produces an annoying amount of repetition. 49

7.2. Vol.2: Direct Interface, One-Channel Translation and their application to CVCV At the end of this introduction, it may be useful for the reader to be able to get an idea of what Vol.2 looks like. Much more than this book, it is analytic in the sense that it draws conclusions from the historical survey on the representational side: how is translation (of morpho-syntactic information into phonological objects) organised, and what does its output look like?

Structure of the book and of Vol.2

33

Vol.2 falls into two theory-unspecific and two theory-specific chapters. Chapter one introduces Direct Interface, the idea that is at the origin of the overall project on the interface and defines what the output of translation looks like (it must be non-diacritic). Chapter two is about how this output comes into being, i.e. the translational process itself. Following (unpublished) work by Michal Starke, it is argued that there is only one source of phonological material that enters phonological computation, no matter whether it represents morpho-syntactic or lexical information: the lexicon. One-Channel Translation contrasts with all previous approaches to translation, which systematically distinguish between morphemic and nonmorphemic information: the former is what morphemes are made of, and everybody agrees that it comes into being through lexical (or vocabulary) insertion when morpho-syntactic structure is converted into a phonological string. The latter is what I call boundary information, i.e. morpho-syntactic properties that materialise in phonology in form of representational objects, but do not ride on morphemes: since SPE, hashmarks and more recently prosodic constituents are born through a specific mapping mechanism that is computational in kind. Starke's alternative does away with computational translation: it makes boundary information originate in the lexicon as much as morphemic information. The result is a uniform translation of morphosyntactic into phonological material through a lexical access. These two chapters are theory-unspecific in the sense that they define the design properties of a correct interface theory: they do not make any statement about specific phonological theories. What they do allow for, though, is the evaluation of competing phonological theories according to their behaviour at the interface. Traditionally, a uniform interface vocabulary that is shared by all individual phonological theories mediates between these and morpho-syntax: juncture phonemes, hashmarks, the Prosodic Hierarchy. Hence whatever variation competing phonological theories produce will be neutralised and invisible at the interface. By contrast in the perspective of Direct Interface, different phonological vocabulary that is proposed by different phonological theories makes contrasting predictions when it acts as the output of translation; also, anything cannot be the output of translation anymore since only objects qualify that make a good lexical entry (One-Channel Translation). Chapters three and four then apply Direct Interface and One-Channel Translation to the particular phonological theory that I am committed to, Government Phonology in general and CVCV in particular. It is only this last half of Vol.2 that bridges over to Vol.1. Chapter three modifies the lateral theory of phonology as it stands at the end of Vol.1 according to the

34

Introduction

requirements of the interface, i.e. Direct Interface and One-Channel Translation (see also Scheer & Ziková forth). Shaping linguistic theory according to interface requirements of course is a very minimalist thing to do – unlike in syntax-centred minimalism, however, here it is phonology that is shaped by the interface. Finally, chapter four shows the interface-readjusted system at work: the initial CV has been around in Government Phonology and CVCV for quite some time now (Lowenstamm 1999), and it has produced a reasonable amount of empirical work. It is shown that only syllabic space passes the filters that are defined by Direct Interface and One-Channel Translation: it is therefore the only possible output of translation. In CVCV, syllabic space reduces to CV units, which are thus the only possible carriers of morpho-syntactic information. On this backdrop, a case study is undertaken which reviews the empirical evidence for the initial CV and inquires on the modalities of its management. The parameterisation of the initial CV is discussed namely in the light of its behaviour in connected speech, i.e. where phonology applies across word boundaries. This chapter is about the only place in Vol.2 that is data-oriented, i.e. where genuine empirical generalisations are made.

Part I Morpho-syntactic information in phonology: a survey since Trubetzkoy's Grenzsignale 50

Chapter 1 The spectrum: what morpho-syntactic information can do to phonology

51

1. Boundaries have a triggering, a blocking or no effect We start the historical survey of interface theories with a non-historical chapter that introduces the classification of interface events which is proposed by Kenstowicz & Kisseberth (1977:83ff, 1979:407ff). Their system is particularly useful because it is pre-theoretical and a priori covers all logically possible situations. It will be constantly referred to as we go along, and it offers a rationale that the reader may use when encountering particular analyses in a specific theoretical environment. According to Kenstowicz & Kisseberth, morpho-syntactic divisions may have three and only three phonological effects. These are shown under (2) below. (2)

possible effects of morpho-syntactic structure on phonology given two morphemes M1 and M2, their concatenation may a. have no effect at all: phonology works as if there were no morpho-syntactic division, i.e. as if the sequence of sounds were mono-morphemic. b. block a process that would apply if the morpho-syntactic division were not there. c. be a condition on the application of a process that would not go into effect if the morpho-syntactic division were not there. These cases are known as derived environment effects.

Kenstowicz & Kisseberth provide rich material for the illustration of the three situations. Let us consider a prototypical representative for each of them. (2a) is trivial: it corresponds to what is sometimes referred to as a

36

Chap 1: what morpho-syntactic information can do to phonology

"late rule", a "rule of phonetic implementation".6 At the word level, a typical example would be the aspiration of English voiceless stops that occurs word-initially (phólitics, where an acute accent indicates stress) and before stressed vowels (pholithícian), but is not sensitive to any eventual morpheme boundaries inside the word: it applies across the board. But there are also less surface-oriented processes where morpheme boundaries may be ignored by the phonology. Namely, this is the case for affix class-based phenomena (§§163,166) where a given class of morphemes follows this pattern, while another class imposes a phonological trace of the concatenation. For example, the division in parént-al is phonologically invisible (stress is as penultimate as in the mono-morphemic párent), while the boundary in párent-hood is phonologically relevant: the computation of stress only takes into account the first portion of the word.

52

2. Blocking and triggering effects: illustration

53

2.1. Process-blocking boundaries: French gliding In order to illustrate boundaries that block processes, let us look at a typical case of the suffix-prefix contrast where a process applies across the "weak" suffix boundary, while it is blocked by the "strong" prefix boundary. In French, the hiatus created by a vowel-final stem and a vowelinitial suffix is resolved by the insertion of a glide in case the stem-vowel is high. The glide in question is a copy of the high vowel, thus producing the sequences [ij-V], [uw-V], [yÁ-V] (see Dell 1976:109, Selkirk 1972:385ff). Table (3) below offers illustration. (3a) shows that the glide does not belong to the lexical information of the roots in question: they appear with their final vowel when suffixed by a zero personal ending. There is no glide in the lexical representation of the vowel-initial suffixes under (3b) either, as evidenced by their vowelinitial appearance after consonant-final stems. Under (3c), however, a glide appears when a hiatus is produced by the concatenation of a vowel-final stem and a vowel-initial suffix. Note that the quality of the suffixal vowel, as well as the kind of suffix added, is irrelevant. This behaviour contrasts 6

While implementing the same idea of the phonological inertness of morphosyntactic boundaries, the notion of postlexical rules known from Lexical Phonology (see §153) is different since it is bound to a particular chunk size: it applies only to strings that are larger than words.

Blocking and triggering effects: illustration

37

with the situation encountered in identical phonological circumstances but where the hiatus is created through the assembly of a prefix and a stem, as under (3d). No glide may break up the hiatus here: *[bi-j-anyEl] "biannual". (3)

French gliding a. the stem does not contain any glide: inflected forms with zero endings je lie [li] I relate je loue [lu] I rent je sue [sy] I sweat b. vowel- initial suffixes do not contain any glide: C-final stems (chant"to sing") chant-er [Sãt-e] -e infinitive chant-ez [Sãt-e] -e 2pl pres chant-ais [Sãt-E] -E 1sg pret chant-a [Sãt-a] -a 3sg passé simple c. concatenation of a V-final stem and a V-initial suffix -er inf. -ais 1sg -ons 1sg -a 3sg passé -ez 2pl pres pret pres simple li-er [li-j-e] [li-j-a] [li-j-E] [li-j-ç)] lou-er [lu-w-e] [lu-w-E] [lu-w-ç)] [lu-w-a] su-er [sy-Á-e] [sy-Á-E] [sy-Á-ç)] [sy-Á-a] d. concatenation of a V-final prefix and a V-initial stem bi-annuel bi-annuel [bi-anyEl] anti-existentiel [ãti-EksistãsjEl] anti-existential anti-alcoolique [ãti-alkoolik] anti-alcoholic archi-ondulé very undulated [aXSi-ç)dyle] archi-ennuyeux [aXSi-ãnyÁijø] very boring

Hence the prefix boundary blocks glide-insertion. Since all prefixes behave like that, and all suffixes allow gliding to go into effect, the conclusion that is commonly drawn ranks the boundaries: suffixation creates a "weak" boundary that allows the vowels on both sides to see each other, while prefixation involves a "strong" boundary that blurs visibility. 54

2.2. Process-triggering boundaries: obstruent voicing in Puyo Pongo Let us now look at a process that applies only if a (specific) morphosyntactic division separates the agent and the patient of the phonological

38

Chap 1: what morpho-syntactic information can do to phonology

event. Kenstowicz & Kisseberth (1977:88f, 1979:407f) report the following data from Puyo Pongo (Quicha, Eastern Ecuador, cf. Orr 1962). (4)

Puyo Pongo: obstruent voicing after heteromorphemic nasals a. within a morpheme, obstruents may be voiced or voiceless after nasals voiceless T in N__ voiced T in N__ p-b pampaljina hambi skirt, poison t-d indi to stir the fire, sun tÉSuntina we, day tÉs-dÉZ ¯ukantÉSi pundÉZa k-g soot, ten SiNki tÉSuNga b. following a nasal and a morpheme boundary, obstruents are only voiced V-__ N-__ /-ta/ wasi-ta kan-da house, you atan-da meat, the frog ajtÉSa-ta puru-ta wakin-da gourd, others is it good?, you? /-tÉSu/ ali-tÉSu kan-dÉZu manioc?, is there? lumu-tÉSu tijan-dÉZu isn't it?, does he have? mana-tÉSu tÉSarin-dÉZu

(4a) shows that nasals may be followed by both voiced and voiceless obstruents if the NC cluster is mono-morphemic. In a heteromorphemic situation as under (4b), however, only voiced obstruents are observed. Concatenation to a vowel-final stem shows that the initial dentals of the objective marker /-ta/ and the question marker /-tÉSu/ are underlying voiceless: they undergo voicing when attached to a nasal-final stem. In this case, thus, a morpho-syntactic division is the critical condition for the application of a process, obstruent voicing. Or, in static terms, the voice-voiceless contrast after nasals is neutralised after a boundary.

55

Chapter 2 Trubetzkoy's Grenzsignale

56

1. The cradle of the functional perspective The first structuralist approach to the phonological effect of morphosyntactic divisions is functional. Trubetzkoy (1939:241) (see also earlier sources: Trubetzkoy 1935:30ff, 1936) holds that Grenzsignale (demarcation signals) alert the listener in order to facilitate the parsing of the phonetic continuum that reaches his ears. Without being helped by the phonology, the task of identifying morphemes would be much more difficult and take much longer. Therefore some morpho-syntactic divisions leave a material trace in the signal so that they could not be overheard. Trubetzkoy (1939:242) contends that "all languages possess specific phonological means in order to signal the presence or the absence of sentence-, word- or morpheme boundaries in a particular location in the continuous stream of speech." According to him, Grenzsignale "may well be compared to the traffic lights in the streets."7 Trubetzkoy will have many followers, structuralist and generative 8 alike. A functional component does not appear to be incompatible with a formal approach to the interface. A zoom on a functional interpretation within an overall formal frame will be undertaken in §§264,340: the classical interface theory of Government Phonology is based on so-called parsing cues (Kaye 1995). Other generative continuators of Trubetzkoy's Grenzsignale will also be briefly reviewed on this occasion.

7

8

Translations are mine. The original German text is as follows: "Jede Sprache [besitzt] spezielle phonologische Mittel, die das Vorhandensein oder das Nichtvorhandensein einer Satz-, Wort- oder Morphemgrenze an einem bestimmten Punkt des kontinuierlichen Schallstromes signalisiert" (Trubetzkoy 1939:242); "Sie dürfen wohl mit den Verkehrssignalen in den Straßen verglichen werden." (Trubetzkoy 1939:242). On the structuralist side, the idea of his Grenzsignale was applied to various languages for example by Trnka (1940), Anderson (1965), Bolinger & Gerstman (1957), Romportl (1984), and in the work on phonetic correlates of juncture (see §70, which are sometimes called boundary signals, i.e. by Lehiste 1962).

40 57

Chap 2: Trubetzkoy's Grenzsignale

2. Grenzsignale do not contribute anything to the mapping puzzle The functional issue brings up issue of the mapping puzzle for the first time. The mapping puzzle is a pervasive mystery that runs through the entire interface literature and is not any clearer today than it was twenty or thirty years ago. The question is exactly which morpho-syntactic construction can, must or must not be promoted to phonological visibility: what kind of cross-linguistic generalisations are found?9 One could expect a significant contribution from the functional perspective, which precisely defines the "need" for this or that morphosyntactic structure to leave a phonological trace. As far as I can see, however, this is not the case. There is a large amount of variation across languages, which linguists have not been able to reduce to any rationale or recurring patterns. That is, some languages seem to be eager to help the listener and hence offer a lot of (different) demarcation signals, while others appear not to care: they make masochistically little use of Grenzsignale. Trubetzkoy (1939:255ff) distinguishes positive and negative signals. The former identify the presence of a morpho-syntactic division, while the latter guarantee its absence. For example, the spiritus asper in Classical Greek unmistakably identifies the beginning of a word; in English, the velar nasal signals the end of a morpheme, and the sequences [Ts], [Dz], [tÉʃt], [tÉʃs], [sʃ] necessarily enclose a morpho-syntactic division. On the other hand, [r] and [h] only occur morpheme-internally in Efik (a variety of Ibibio, Niger-Congo).

58

3. Grenzsignale are immaterial and under morpho-syntactic control Another interesting property of Grenzsignale is that they do not appear to possess any material existence (except in the phonetic signal). Unlike in all later theories, Trubetzkoy does not note boundaries, or give them any other material or graphic existence. They appear to be just right or left margins of meaningful units. Hence Trubetzkoy does not appear to consider them linguistic objects in their own right, i.e. objects that are endowed with intrinsic properties.

9

The mapping puzzle will be discussed on various occasions in the book: §111 (post-SPE times), §§387,392f,396 (Prosodic Phonology), §463 (OT), §753 (summary).

Grenzsignale are immaterial and under morpho-syntactic control

41

As we shall see below, this contrasts with later structuralist and generative conceptions, which have sought to grant a material phonological identity to extra-phonological information. That is, American Structuralists insert juncture phonemes in the linear string of phonemes, while extraphonological information materialises in the generative paradigm as SPEtype diacritics such as # first, later on in form of the Prosodic Hierarchy. As will be shown in Vol.2 (also Scheer 2008a, 2009a,c), the purpose of Direct Interface is to do away with this diacritic tradition: juncture phonemes, #s and the Prosodic Hierarchy are diacritics, i.e. objects which do not exist in phonology in absence of the appeal to extra-phonological information. Thus structuralist as much as generative carriers of boundary information so far have only stored alien information in the phonology. Its carriers therefore do not have any phonological properties and hence do not make predictions: a # or a prosodic word can cause any phonological event and its reverse to occur (see Vol.2). Interestingly, Trubetzkoy's Grenzsignale cannot produce this kind of ghost ships in phonology: they are immaterial. On the other hand, they are of course as unable to make predictions regarding their effect as the structuralist and generative diacritics. Ségéral & Scheer (2008) (and Vol.2 at greater length) show that this is odd since the influence of morphosyntactic information on the phonology of natural language is not arbitrary; rather, it produces recurrent effects. Direct Interface thus represents a compromise between the Trubetzkoyan and later approaches: in order to have a predictable effect, morphosyntactic information must certainly materialise in the phonology, but not as a diacritic. The only option, then, is to make it a truly phonological object, i.e. one that exists in phonology in absence of extra-phonological conditioning. Finally, the fact that positive Grenzsignale signal the existence of a morpho-syntactic division makes their occurrence in the middle of morphemes impossible (see the quote of Vachek 1970 in §66, who points this out). Negative Grenzsignale do occur in the middle of morphemes (see §57), but they signal the absence of a morpho-syntactic division. Hence Trubetzkoyan carriers of morpho-syntactic information always respect morpho-syntactic divisions: morpheme-internal Grenzsignale can only signal their absence, while Grenzsignale that occur at morpheme boundaries can only signal their presence. That is, Grenzsignale are distributed according to morpho-syntactic divisions in all cases. As we will see below, this allies the Trubetzkoyan and the generative conceptions against structuralist juncture phonemes. The structuralist en-

42

Chap 2: Trubetzkoy's Grenzsignale

shrining of descriptive economy indeed led to juncture phonemes in the middle of morphemes – a move with disastrous consequences (see §69). It is only in the generative paradigm (§78) that the morpho-syntactic control over boundaries (or juncture) will be restored.

59

Chapter 3 American structuralism: juncture phonemes

60

1. Introduction In American Structuralism, morpho-syntactic information in phonology was carried by juncture phonemes. These were part and parcel of structuralist theory: among others, Hockett (1955:167ff, 1958:54ff), Trager (1962), Devine & Stephens (1976:293ff) and Aronoff (1980) offer (historical) surveys. Structuralist theory required that morpho-syntactic information be recorded as a phonological object (a phoneme), rather than as a diacritic. This is due to so-called Level Independence. Rooting in a bottom-up discovery procedure of linguistic structure, this principle prohibits the use of higher level information at any given level of description. Hence phonology being described before morphology and syntax, no look-ahead is allowed: units of the latter two levels must not be present in the former. The base line of structuralist orthodoxy is well expressed by Hockett's (1942) apodictic formulation below. (5)

"No grammatical fact of any kind is used in making phonological analysis." Hockett (1942:20) "There must be no circularity; phonological analysis is assumed for grammatical analysis, and so must not assume any part of the latter. The line of demarcation between the two must be sharp." Hockett (1942:21)

We will see below that there were also voices that pled for a less strict enforcement of Level Independence. But even in quarters where the principle was strictly obeyed, juncture phonemes were promoted for two reasons, neither of which has anything to do with the representation of morpho-syntactic information. First, they served as an environment for the prediction of allophones. Second, they enhanced economy: the inventory of phonemes could be reduced when reference to juncture phonemes demoted phonemic candidates to allophonic status. The questions that were discussed in this environment all have to do with the intuitively awkward prohibition to use morpho-syntactic information in phonology. The following pages show how structuralists tried to sell

44

Chap 3: American structuralism: juncture phonemes

morphology for phonology in order to circumvent the difficulty (Moulton's analysis of German in §64); they show how Level Independence led to the abuse of boundaries, which were placed in the middle of morphemes; finally, they discuss the debated question whether juncture phonemes (ought to) have a phonetic correlate.

61

2. Level independence: no morphology in phonology

62

2.1. The orthodox point of view The empirical basis of American structuralism was the description of native American languages. In relation with the practice of the descriptivist school of the 40s and 50s, the methodology used in fieldwork was an important factor in theory-building. That is, American structuralism ambitioned at constructing a fully fledged grammatical statement of a language from scratch, i.e. from the raw data that reach the ear of the fieldworker. In the introduction to his book, Harris (1951) is explicit on the fact that this is what his work is all about. (6)

"The whole schedule of procedures outlined in the following chapters […] is designed to begin with the raw data of speech and end with a statement of grammatical structure." Harris (1951:6)

The discovery procedure of grammatical structure, then, is strictly positivist in the sense that only recorded or previously constructed information can feed linguistic analysis. Observable data are made of sound and of nothing else; the corresponding units are called phones. The analysis of phones establishes a higher order phonemic level, whose basic units are phonemes. Phonemes cluster together as morpho-phonemes, which are the building blocks of the morpho-phonemic level of analysis. Finally, morphophonemes build clauses at the clausal level. Therefore information from a higher level of description could not possibly be present at a lower level: it does not exist yet. Hence no piece of morphological or syntactic vocabulary, or any representative thereof, could be used in phonological analysis.10

10

Lehiste (1960:47ff) is also explicit on this.

Level independence: no morphology in phonology

45

Aronoff (1980), whose line of reasoning I follow in the present section, traces back how American structuralism addressed morphological information.11 (7)

"One of the essential characteristics of the American descriptivists' phonemic level, a consequence of their theory of discovery procedures, was its autonomy from syntax, semantics, and morphology. One was supposed to be able to do a phonemic transcription which did not refer to higher levels of analysis. Indeed, according to the theoreticians, one was supposed to be able to do a phonemic analysis without having a clue as to the higher structure of an utterance. In the case of juncture, the phonemic transcription could not refer to the fact that night rate is a compound consisting of two words, while nitrate is one single morphological unit. One could not account for phonetic distribution in terms of morphology and syntax." Aronoff (1980:30)

Hence within the limits set by this approach, morphological information could either not be referred to at all in phonology, or it had to be sold as phonological. "The American descriptivists believed that junctures were phonemes because they had to", as Aronoff (1980:30) puts it. This is how and why morpho-syntactic information could only incarnate as truly phonological objects (rather than as diacritics): "to a Descriptivist, then, if a boundary has a phonemic effect, it must be represented as an element on the phonemic level: a juncture phoneme" (Aronoff 1980:30). While this position regarding the use of morpho-syntactic information in phonology was widespread, it is not the only one that can be found in structuralist thinking. Goldsmith (2008:49ff) offers an informed historical discussion: the spectrum of opinions ranged from Hockett's (1942) and Joos' (1964) unbending orthodoxy over Harris' (1951) apparent (but selfcontradicted) tolerance of some look ahead to Pike (1947) who argued that no phonological analysis can be carried out in absence of minimal morphosyntactic information (grammatical prerequisites). We will come back to these voices at greater length as we go along (§66).

63

2.2. Junctural minimal pairs: night rate vs. nitrate The following sections illustrate the kind of dilemma into which structuralists were driven by Level Independence. We set out with the most impor11

Lehiste (1960:5ff) offers a detailed survey of how juncture was treated in the structuralist period from the internal point of view.

46

Chap 3: American structuralism: juncture phonemes

tant one, which is caused by so-called junctural minimal pairs. These occur when identical phonemic sequences with contrastive meaning are phonetically distinct because of an allophonic variation that is governed by a morpho-syntactic division. The most famous example is certainly the pair night rate - nitrate. Being word-final, the dental of the former is unreleased; by contrast, since the dental of the latter occurs word-internally and before a tonic vowel, it is aspirated. Speakers consistently identify both words on the grounds of this difference, which is controlled by the word boundary. Word boundaries, however, must not exist in phonology, which means that the analyst is forced into the recognition of two distinct /t/ phonemes. The following section shows how an identical situation in German is circumvented by the introduction of juncture phonemes, whose function is to dress up illegal morpho-syntactic information in a phonological disguise. Moulton (1947) is often quoted as a typical illustration for juncture phonemes.

64

3. Morphology in a phonological guise: Moulton (1947) on German The first problem that Moulton (1947:220) comes across is aspiration. He discusses junctural minimal pairs in German. Two instances appear under (8) below. (8)

junctural minimal pairs in German: aspiration spelling surface a. dieses Kabinett [ diiz´skhabI'nEt] diese Skandale [ diiz´skan'daal´] b. ich antworte: Terrasse [/Iç /antvoåt´thE'“as´] ich antwortete: Rasse [/Iç /antvoåt´tE'“as´]

gloss this cabinet these scandals I answer 'terrace' I answered 'race'

Under (8a), the portion [ diiz´ska] is identical except for aspiration, which occurs in the first, but not in the second sequence quoted. Therefore, aspiration cannot be predicted from any environment and must be granted a phonemic status. Another case that appears to show the distinctiveness of aspiration is the true "minimal pair" under (8b) (note that the stress pattern is identical: main stress on the last, secondary stress on the first [a]).12 12

Moulton transcribes the first "e" in Terrasse and the last "e" in antwortete alike, i.e. in his phonemic transcription /e/. The standard pronunciation, however,

Morphology in a phonological guise: Moulton (1947) on German

47

Facing this evidence, the regular structuralist analysis would be to recognise a phoneme of aspiration in the phonemic inventory of German. However, Moulton suggests an alternative which is based on the following observation: aspiration is predictable everywhere but in a very reduced set of utterances, which are represented under (8). The general rule seems to be "/p t k/ are aspirated after a pause, but unaspirated after any segmental phoneme". This analysis is then simply extended to the few contrastive cases: "there are places within an utterance where /p t k/ behave as if they were preceded by a pause" (Moulton 1947:220). In order to give a phonemic interpretation to this insight, Moulton chooses to represent the "pause" by a segmental phoneme which he calls "open juncture" and refers to as /+/. As many other segmental phonemes, /+/ produces allophonic variation: "at the beginning or end of an utterance it appears as a pause of brief duration or, in free variation with this, as zero" (page 220). Hence the pair under (8b) is phonemically distinct in the following way: /+/ix /antvorte+te'rase+/ vs. /+/ix /antvortete'rase+/. Moulton admits that the solution which resorts to juncture, as it stands, is not any more attractive than the orthodox way of addressing the evidence, and perhaps even worrisome because it introduces a new concept. But there are two other phenomena in German that could be accounted for with open juncture as well: the distribution of the glottal stop and the alternation of [X] and [ç]. The reasoning is the same as before: the glottal stop must be granted distinctive status because there are minimal pairs such as those under (9) below. Den Bau erkennen under (9a) may be pronounced without glottal stop [deen''bauå kEn´n], in which case it is homophone with den Bauer kennen. However, the ambiguity may also be done away with by producing a glottal stop in den Bau erkennen [deen''bau/å kEn´n] (while no glottal stop can appear in den Bauer kennen). Therefore, the glottal stop is meaningful: it can be used in order to distinguish the two sequences. Under (9b), the two items may also be homophones in case Arbeit ersuchen is produced without a glottal stop. As before, however, the opdoes not produce the same sound ([E] in Terrasse, schwa in antwortete). Perhaps are there dialects (Moulton uses a Northern variety) where both are either schwa or [E]. In any event, I have noted [E] in both instances for the sake of argument. Following Standard German, I also transcribe r-vocalisation in codas and its uvularised versions elsewhere (Moulton only notes [r] and /r/).

48

Chap 3: American structuralism: juncture phonemes

tional presence of a glottal stop makes the two sequences distinct, both phonetically and semantically. (9)

junctural minimal pairs in German: glottal stop spelling surface a. den Bauer kennen [deen''bauå kEn´n] den Bau erkennen [deen''bauå kEn´n] [deen''bau/å kEn´n] b. Arbeiter suchen [''/arbaitå zuuXen] Arbeit ersuchen [''/aabaitå zuuXen] [''/aabait/å zuuxen]

gloss "to know the farmer" "to recognise the building" "to look for workers" "to request work"

Here again, the recognition of a juncture phoneme would afford to avoid admitting the glottal stop in the phonemic inventory of German: den Bauer kennen would be /+deen''bauå kEn´n+/, while den Bau erkennen would identify as /+deen''bau+å kEn´n+/. On this analysis, the distribution of the glottal stop is allophonic: "vowels show an allophone with glottal stop after /+/, but without glottal stop after all other segmental phonemes" (Moulton 1947:223). Finally, the distribution of [X] and [ç] (which is notorious from linguistic classrooms) is another illustration of the same pattern. [X] occurs after non-front vowels (including [a]), while [ç] is found after front vowels and consonants (that is, [l,r,n] as in Milch, durch, manch "milk, through, some"). However, a small residue of words resists this distribution, and these invariably involve the diminutive morpheme -chen [-ç´n]. This time, thus, there are real minimal pairs as under (10a,b), i.e. which do not involve word boundaries. In addition, though, Moulton (1947:223) also mentions utterances such as under (10c) where [ç] and a preceding back vowel are separated by a word boundary. (10)

junctural minimal pairs in German: [ç] - [X] spelling surface a. Kuchen [kuuX´n] Kuhchen [kuuç´n] b. tauchen [tawX´n] Tauchen [tawç´n] c. da China so groß ist [daaçiinasoog“oosIst]

gloss cake little cow to dive little rope since China is so large

"Accidental" coincidence of juncture and morpho-syntactic divisions

49

This situation normally requires the existence of two phonemes, /ç/ and /X/. But an analysis along the lines of Moulton's juncture phoneme can rescue the items which violate the distributional regularity: if the morpheme boundary of the diminutive suffix incarnates as /+/, the cake will be /+kuuX´n+/, whereas the little cow becomes /+kuu+X´n+/. On this analysis, then, /X/ is subject to allophonic variation whereby [X] appears after non-front vowels, while [ç] is found after all other segmental phonemes, including the juncture /+/.

65

4. "Accidental" coincidence of juncture and morpho-syntactic divisions Considering the evidence from a wholesale perspective, Moulton (1947:223f) favours an analysis that uses one single object, i.e. the open juncture /+/, over the regular account where three new phonemes must be recognised (i.e. /ç/, /// and /aspiration/). Economy is thus the critical argument. The juncture /+/ is multifunctional and does the job of three otherwise unrelated phonemes. However, Moulton insists on the fact that the benefit is functional, rather than numeric: a generalisation would be missed if the evidence relating to morphological boundaries were expressed by three independent items. He thus explicitly recognises the impact of morphology on phonology. In the same way, Moulton is explicit about the morpho-syntactic conditioning: all minimal pairs for aspiration and the glottal stop that one can come by are monsters, i.e. contain a word boundary that is critical for the opposition. In the case of the [X]-[ç] alternation, the same holds true for morphological boundaries. The corresponding putative phonemes would not exist if only items of word-size (of morpheme-size for [X]-[ç]) were considered. Also, Moulton makes a phonetic argument: in slow speech, /+/ may (but need not) coincide with a pause. Finally, Moulton comes to grips with the critical question: all that he has said and done so far is in overt conflict with the structuralist principle of Level Independence. That is, there must not be any coincidence of /+/ with morpho-syntactic divisions. Or if so, it can only be accidental. This is precisely the line of argumentation that Moulton adopts. Fortunately enough for him, aspiration and the glottal stop also occur morpheme-internally: the complete descriptive statements are "word-initially and before stressed vowels" for the former, "before word-initial vowels and

50

Chap 3: American structuralism: juncture phonemes

in the middle of a hiatus in case its second element is stressed" for the latter. Hence in words such as Papier [pha'phiiå] or Laterne [la'thEån´], the presence of aspirated word-medial consonants enforces the structure /+pa+pir+/ and /+la+terne+/, respectively. In the same way, Theater [the'/aatå] and Ruine [“u'/iin´] call for word-internal /+/, i.e. /+te+ater+/, /+ru+iin´+/. In the end, thus, Moulton (1947) tries to deny the phonological relevance of morpho-syntactic information for which he has provided evidence. He does so under the pressure of structuralist orthodoxy and, one may assume, against better understanding (see note 14 on page 225 of his article). The take-home message of the previous pages is the causal relation that, starting with Level Independence and the prohibition of higher level information in phonology, inevitably leads to juncture phonemes and their presence in the middle of morphemes. Before the absurd consequences of the latter are discussed in §69, the following section provides a broader overview of how morpho-syntactic divisions were treated in the structuralist literature.

66

5. Structuralist views on the coincidence of juncture and morphosyntactic divisions

67

5.1. Defenders of orthodoxy (Hockett, Joos, Trager) and Harris' ambiguous position It was mentioned in §62 that even though Level Independence was the orthodox structuralist position defended by Hockett (1955) and Joos (1964), more tolerant views that allow for morpho-syntactic information to inform phonological analysis are also found in the literature (see the discussion in Aronoff 1980:31f and Goldsmith 2008). Let us begin with Zellig Harris' (1951) position, which is ambiguous. Given the quote below from the outset of his chapter on juncture, he appears to defend true orthodoxy: morpho-syntactic traces in phonology are strictly prohibited, and the only purpose of juncture is phoneme economy.

Coincidence of juncture and morpho-syntactic divisions in structuralism (11)

51

"This procedure introduces junctures as a factor in phonemicization, but only, of course, to the extent that this is possible without knowledge of morphemes. 8.1. Purpose: Eliminating Restrictions on Sets of Phonemes We reduce the number of phonemes, and simplify the statement of restrictions upon the environments in which they occur, by considering those restrictions of environments which apply to large numbers of phonemes." Harris (1951:79)

That is, juncture phonemes do not need to comply with any requirement other than acting as an environment for the prediction of allophones. Just like all other phonemes, they thus occur at arbitrary places in the string. Accordingly, Harris (1951:87f) allows for the German word Teil [thajl] "part" to identify as /d#ajl/ in order to be able to generalise the final devoicing pattern which is found in this language: word-final underlying /d#/ appears as [t]; hence /d/ can be eliminated from the phonemic inventory if all [t]s, even in non-neutralising and morpheme-internal environments, are interpreted as the result of devoicing before "#". The treatment of German voiceless obstruents is thus an application of the economic ambition: a world with fewer phonemes is a better world.13 However, elsewhere in the book Harris (1951) departs from the orthodox point of view by allowing not only "accidental" coincidence of juncture and morpho-syntactic divisions, but also their causal relationship. In a footnote, Harris (1951:241) writes "that phonemic junctures are used for segments which occur only at morpheme (or other) boundary." Also, he appears to enjoy a perspective where juncture can be made to carry morpho-syntactic information: "[t]he great importance of junctures lies in the fact that they can be so placed as to indicate various morphological boundaries." Harris (1951:87). This inconsistency is pointed out by Aronoff (1980:34f), while Goldsmith (2008:51ff) goes at great length to show that Harris cannot be counted as a representative of orthodoxy: he provides the quotes of the 13

Hockett (1942:9) provides a general formulation of economy in structuralist theory: "Economy: if several different analyses equally satisfy the other requirements, that which establishes the smallest number of phonemes is the one to be preferred. This is a corollary of the general scientific principle that the simplest description which accounts adequately for all the facts is to be preferred."

52

Chap 3: American structuralism: juncture phonemes

preceding paragraph and many others in support, but does not mention the conflicting statement under (11). Hockett (1955) was already identified as a defender of the orthodox position: In his textbook, he is explicit about the fact that the coincidence of juncture with morpho-syntactic divisions can only be accidental. (12)

"There is one potential source of error which ought to be avoided. Juncture phonemes are not recognized in order to show grammatic boundaries of one or another kind - say boundaries between words. It often happens that grammatic boundaries fall at open junctures, particularly grammatic word boundaries." Hockett (1955:172)

Unlike Harris (1951), however, he seems to refrain from explicitly admitting that juncture may also fall in morpheme-internal position. Lehiste (1960:47) shares this view. Finally, Trager (1962) is another representative of structuralist orthodoxy: juncture phonemes are estranged from any morpho-syntactic correlate, and hence do not imply any prediction. (13)

68

"Most places where /+/ occurs also indicate the end of a word, but in some instances it is only the end of a morpheme that is present, as in shyness /šáy+nˆs/; while in still others there is not even a morpheme-end, as in the pronunciation of center as /sén+t´r/. Moreover, there are word sequences without /+/: postman may be /pówsm´n/ as well as /pówst+m´n/."14 Trager (1962:19f)

5.2. Opposite view: Pike's grammatical prerequisites While the orthodox attitude was widespread and dominant, it also faced principled opposition. Pike (1947) leads this move. His article is entitled "Grammatical prerequisites to phonemic analysis", and he argues that phonemic analysis cannot be done without morpho-syntactic information. In a follow-up article five years later (Pike 1952), he adduces further support for his position and discusses the literature on the topic that has appeared in the meantime. 14

Trager does not indicate why this particular pronunciation requires juncture. The transcriptions in slashes are approximate: it is difficult to reproduce the multiple diacritics on all sides of the basic symbols that are carried by structuralist phonetic records.

Coincidence of juncture and morpho-syntactic divisions in structuralism (14)

53

"In recent years various phonemicists seem to have set as an ideal of phonological description and analysis the elimination of all reference or reliance upon facts about the grammatical structure of the language being investigated. […] The present article holds that it is impossible for such claims to be realized completely, and that even were it possible it would at times prove undesirable." Pike (1947:155)

Pike also challenges economy: in a perspective that airs a generative flavour, he favours insight over description. (15)

"We are not after simplicity first, but rather a representation of the structure of the language as it functions, whether the result be simple or complex." Pike (1947:172)

Vachek (1970) builds on Trubetzkoy's Grenzsignale in order to support Pike's grammatical prerequisites: if juncture has the Trubetzkoyan demarcative function, it cannot exist in the middle of morphemes; indeed, there is nothing to be marked morpheme-internally or, worse, the listener could be put on the wrong track. (16)

"W. Haas arrived at a conclusion analogous to ours. In answering the question of how much grammatical information is necessary for an adequate phonemic analysis he insists on the necessity of a certain minimum of grammatical prerequisites for this purpose, and as such minimum he specifies precisely the knowledge of the placement of word and morphemic limits." Vachek (1970:963) "This zero character of the supposed juncture phoneme is in full conformity with the classical Prague conception of Trubetzkoy's frontier signals. […] Such a potential pause can never occur inside a morpheme but only between words (and, though less frequently, between morphemes constituting a word)." Vachek (1970:965)

Wells (1947:107) is also a critical voice. He writes that "linguists find themselves tempted to institute 'junctures' simply as notational devices for reducing the number of phonemes." In his view, "juncture, whenever it occurs, is a morpheme - though often with no detectable meaning" (page 108). This indeed rules out morpheme-internal juncture. Unlike in his later textbook that was quoted above, Hockett (1949) also appears to argue for syntactic control over juncture phonemes.

54

Chap 3: American structuralism: juncture phonemes

(17)

"If some peculiar phenomenon is predictable […], and if its successive occurrences seem to mark fairly well the borders between phonological words, then that phenomenon is junctural. […] A juncture phoneme is then a grouping of such phenomena which makes for unambiguous and simple linear transcription. If this last statement seems arbitrary, it is no more so than is, for me, the definition of any kind of phoneme." Hockett (1949:35f)

In sum, it seems that Level Independence was challenged especially in older sources: the orthodox position became only dominant in the 50s.

69

6. Whatever suits the analyst: juncture in the middle of morphemes Let us now have a closer look at the disastrous consequences when juncture phonemes are estranged from morpho-syntactic control. The structuralist dilemma is embodied in the word "juncture" itself: if juncture phonemes may occur in the middle of morphemes, one wonders in which way they deserve their name. The case of Hill (1954) provides good illustration. Hill (18)

"define[s] juncture […] as a lengthening of the preceding phoneme by one half-unit, where a full unit is equal to the average length of a sound as member of a phoneme." Hill (1954:440)

This general equation of juncture with lengthening comes from the evolution of Latin where several modifications of length occurred. For example, Hill ascribes the gemination of lat. labra in it. labbra to the existence of a juncture phoneme, because "no explanation […] is as simple as postulating the sequence /-b + r-/ […], where the juncture resulted in a lengthening of the preceding consonant." That is, it. labbra is derived from /lab+ra/. The juncture phoneme is thus exclusively defined by its effect, and its effect is attributed to its existence. The analysis is overtly circular. The only thing to which Hill seems to commit himself is a stable effect of juncture, i.e. lengthening. One could therefore expect that juncture is present in all lengthening environments, but absent elsewhere. Alas, Hill (1954:441) overthrows this principle as well: one page later, the presence of juncture is responsible for the stability of a cluster, while its loss triggers assimilation. Latin ad ferō, *quid-pe were /ad+ferō, quid+pe/, but become afferō, quippe in Vulgar Latin because the juncture was lost: /adferō, quidpe/ opens the way to assimilation.

Is there a phonetic correlate of juncture?

55

While this could still be said to bear on the length of some segment, Hill (1954:444) completely gives up on that notion as he tries to make juncture responsible for the difference between [i] and [j]. On his analysis, [j] is an allophone of /i/ when juncture was lost: lat. ais "thou sayest" is /a+is/ and therefore comes out as ['ais], while lat. aes "brass" is /ais/ and appears as ['ays]. It may certainly be argued that Hill (1954) is not representative of the structuralist literature: a framework is not responsible for bad analyses that are done in its name. The point, however, is that the theory does not object against this kind of absurd analysis. What is more, it actually pushes phonologists to posit juncture phonemes in the middle of morphemes, as we have seen: under Level Independence, their coincidence with morphosyntactic divisions must be purely accidental. As a result, juncture phonemes are demoted to a meaningless diacritic that may be used whenever it suits the analyst, and without any cost. Since their phonetic realization is "zero" most of the time, having or not having a juncture in a phonemic transcription is for free. Why, then, should phonologists bother to look for explanations and causal relations? Whatever the problem, an invisible juncture phoneme is the solution: as a deus ex machina, it can be made responsible for any phenomenon and its reverse. In other words, being able to implement juncture phonemes in the middle of morphemes is a license for printing banknotes – worthless banknotes that do not buy anything.

70

7. Is there a phonetic correlate of juncture? The other major issue that is discussed in the structuralist literature concerns an eventual phonetic manifestation of juncture. The regular position is that, just as for all other phonemes, a stable phonetic correlate is required. Recall that Moulton (1947) proposes allophonic variation between a "pause" and "zero". However, a fraction that one may call abstract does not care for any phonetic grounding. Hockett's early writings represent this position. (19)

"So we see that the failure of all the allophones of some juncture phonemes to have some (articulatory or acoustic) property in common is no logical defect." Hockett (1949:38f)

Harris (1951:79ff) also favours this position, and Wells (1947:108) sets out to make juncture a true morpheme: "juncture, whenever it occurs,

56

Chap 3: American structuralism: juncture phonemes

is a morpheme – though often with no detectable meaning." In this case indeed, there is no place for any phonetic correlate: "the validity of juncture phonemes is open to grave doubts on phonetic grounds" (page 107). Shocked by Harris' (1951) juncture abuse that was reported earlier (German Teil analysed as /d#ajl/, §66), Hockett stands on the traditional concrete side by 1955. (20)

"Juncture phonemes achieve their power precisely because of their phonetic heterogeneity. […] There is one way of speaking of juncture which retains phonetic homogeneity of all its allophones. This is Harris's way. Harris sets up a juncture as a 'zero' phoneme – a phoneme having no phonetic properties at all (and, because of this, having identical phonetic properties in all environments). The only function of the 'zero' phoneme, then, is to function as environment for ordinary phonemes: English /b/ is represented by different allophones when flanked by this 'zero' phoneme and when not so flanked, and by different allophones depending on whether the 'zero' phoneme precedes or follows. This seems to the present writer a most unfortunate and misleading kind of hocus-pocus; he feels that setting up junctures as always involving identifiable phonetic material, no matter how diverse, is much better." Hockett (1955:171f)

The phonetic reality of juncture has also been studied instrumentally. Examples are Bolinger & Gerstman (1957), Lehiste (1960, 1962, 1965) and Hoard (1966). All work in this area indeed found "phonetic factors that indicate the presence of internal open juncture" (Lehiste 1960:39). Lehiste (1960:5ff), Devine & Stephens (1976:294ff) and Aronoff (1980:32ff) offer an informed survey of the question. Aronoff reports on attempts to make juncture a suprasegmental object: still a phoneme, but a suprasegmental, rather than a segmental one (Lehiste 1960:48 also argues along these lines). In a comment on Moulton (1947) and Leopold (1948), Martin Joos writes in his Readings in Linguistics that "finally one must assign juncture to a phonemic status: otherwise it is nothing. By hypothesis it can't be segmental: no room there. Hence we are forced to the Hockett solution […]: it is suprasegmental" (Joos ed. 1957:216). This may look like an escape hatch into a mushy category, especially since the difference between segmental and suprasegmental phonemes, beyond intuition, is not very clear. Rather, what Joos means is that suprasegmentals do not need to have any specific phonetic existence; therefore we are fine: juncture is a phoneme, but nobody can hear it. The suprasegmental alternative thus simply appears to be a rhetorically enhanced way to have your cake and eat it too.

Structuralist terminology 71

57

8. Structuralist terminology For the sake of completeness, the terminological jungle, which is typical for the structuralist literature, needs to be mentioned. There were end- and meaningless debates on the vocabulary that should be used in order to refer to juncture. Lengthy discussion of terms such as transition, disjuncture, schismeme or contiguity without much, if any, empirical or theoretical content were led for example in Welmers (1947), Hall (1948), Kepke (1948), Bolinger & Gerstman (1957), Trager (1962). Also, different juncture phonemes were recognised (see Harris 1951:86f for a survey): internal vs. terminal juncture, internal vs. external juncture (Trager & Bloch 1941), close juncture vs. disjuncture (Hall 1946:81), level terminal vs. upturn terminal vs. downturn terminal juncture (Potter 1962) or plus vs. internal open vs. terminal juncture (or even "semijuncture") (Stockwell et al.1956). Trager (1972:45) for example ends up with five different junctures.

72

9. Conclusion: Level Independence seeds modularity and enforces translation The foregoing pages have discussed a number of issues that were debated in the structuralist literature. Important for the following steps of the survey below are Level Independence, the morpho-syntactic control over juncture and its phonetic correlate. One could be tempted to say that structuralism was wrong on all of these counts. That would be too rash a conclusion, though. It is true that nobody today believes anymore that morpho-syntactic divisions leave a phonetic trace. It is also true that admitting juncture in the middle of morphemes is a disaster: there is no point in doing phonology anymore since all facts observed can be ascribed to arbitrarily placed and arbitrarily chosen ghosts. The frequent mocking of the structuralist approach in modern retrospection, however, is undue and misevaluates an important contribution that hides behind unfamiliar vocabulary. Level Independence expresses the idea of what will later be known as Fodorian modularity, at least partly: although miles away from a cognitive perspective and ambitioning to achieve descriptive accuracy rather than understanding, it embodies the insight that phonology and morpho-syntax (the structuralist literature commonly opposes phonology and grammar) are two ontologically distinct

58

Chap 3: American structuralism: juncture phonemes

systems, and a priori incommunicado. Therefore morpho-syntactic information can only play a role in phonology if prior to its action it is translated into truly phonological units (rather than into diacritics). The impossibility to use untranslated morpho-syntactic units in phonology (and vice-versa) follows from what is known as domain specificity in the modern modular perspective (§§610,640). True, structuralist phonology and morpho-syntax are only waterproof in one direction: the bottom-up discovery procedure uses phonological units in order to build morphosyntactic structure. Also, the claim that no information transits in the opposite direction is of course wrong (but even structuralists did not really believe in this dogma: this is what the camouflage of juncture in a phonemic guise is all about). Beyond these issues, however, the idea that we face two different levels of analysis which use two distinct sets of vocabulary and possess their own computation is a central insight expressed by Level Independence. It also lies at the heart of Fodorian modularity which is introduced at length in the Interlude (§586) and will become the cognitive reference for the generative architecture of grammar in the 80s (see §415). In short, Level Independence enforces intermodular translation of morpho-syntactic into phonological vocabulary (see §649 and the summary regarding translation in §692). Finally, it is interesting to observe that all subsequent theories of the interface (with the exception of so-called direct syntax, §§137,407) have followed the track that was opened by structuralist translation. Since the only phonological vocabulary that was available in structuralist theory were phonemes, morpho-syntactic information could only incarnate as juncture phonemes. The output of translation then changed as the vocabulary of phonological theories evolved: phonemes were followed by segments in early generative phonology (sic: # and + were supposed to be [-segment] segments, §87) and autosegmental domains (the Prosodic Hierarchy) in the early 80s after the advent of autosegmental structure (§368). The interface invariant, then, is that the carriers of morpho-syntactic information in phonology always correspond to the basic representational units that current phonological theory manipulates.15 15

OT is an exception: being a theory of computation (and of nothing else), it has no genuine representational vocabulary (§452). As in other areas (such as segmental representation), the old furniture of the 80s, prosodic constituency, is therefore still the relevant interface currency in present-day constraint-based environments.

73

Chapter 4 Chomsky, Halle & Lukoff (1956)

74

1. The structuralist cover: economy In the 50s when generative analysis was exotic in structuralist-dominated phonology, Chomsky et al. (1956) tried to introduce what later will become orthodox ideas of SPE (significantly, in a Festschrift for Roman Jakobson). Since the environment was by and large hostile, they attempted to clothe their work on boundaries in the structuralist strive for economy.16 That is, any move that reduces the number of phonemic entities was welcome (see §§60,65). In this perspective, Chomsky et al. (1956) propose to replace the four different stress phonemes that were commonly recognized for English (and numbered 1 to 4) by one single opposition, "accented" vs. "unaccented". All the rest of the labour, they submit, can be done by juncture. The examples that commonly served for the establishment of four stress phonemes are junctural minimal pairs that had been largely discussed in the structuralist literature. A good example is the oft-quoted light house keeper (e.g. Bolinger & Gerstman 1957). This item possesses four stressable vowels, and according to the distribution of stress can mean different things ("for some speakers" according to Chomsky et al. 1956). Table (21) below illustrates the contrasts reported. (21)

light house keeper: contrastive readings according to stress a. [2134] a housekeeper who is light in weight b. [1324] a person who keeps a light house c. [3134] a person who does light housekeeping

Chomsky et al. (1956) propose to achieve the contrast by juncture, rather than by stress phonemes. They use two different junctures, "internal" "–" and "external" "=". Phonemically speaking, the threefold "minimal pair" is thus structurally, rather than prosodically distinct in the way shown under (22) below.

16

St. Anderson (1985:313f) describes the socio-political background and impact of Chomsky et al. (1956).

60 (22)

Chap 4: Chomsky, Halle & Lukoff (1956) light house keeper: identities according to Chomsky et al. (1956) a. [2134] /light = house – keeper/ a housekeeper who is light in weight b. [1324] /light – house – keeper/ a person who keeps a light house c. [3134] /light house keeper/ a person who does light housekeeping

The distribution of stress, then, is allophonic according to juncture. With reference to Hockett (1955:158, 168), Chomsky et al. (1956) explicitly advertise simplicity and economy as the trump of their analysis: junctures "are introduced for the purpose of reducing the number of physical features that must be considered phonemic" (pages 66 and 68).

75

2. Phonological domains built on juncture distinctions Pretending to follow the track of economy allows Chomsky et al. (1956) to introduce genuinely generative ideas into the structuralist environment. Among these features most prominently the claim that not only syntax and semantics, but also phonology is hierarchically organised. (23)

"Constituent structure has always been considered a characteristic feature of the higher levels of morphology and syntax. We are suggesting here that it exists on the phonological level as well. Every linguistic level, then, has the basic form of a linear system of symbols, organised into a hierarchical arrangement." Chomsky et al. (1956:78f)

The phonological constituent structure that Chomsky et al. (1956:70) appeal to is built on the juncture phonemes that were introduced in order to reduce the number of phonemes. That is, all junctures are ranked according to their relative prominence: some are stronger, others are weaker. This move foreshadows the classical generative treatment of boundaries where the morphological + is weak, against the stronger syntactic # (boundary strength, §§87,116). On these grounds, Chomsky et al. (1956:70) construct a "phonemic clause" which "will be bounded […] by zero-order junctures. Within this domain, the two junctures discussed above [i.e. internal and external] suffice to determine stress in a simple and significant manner." As far as I can see, this is the first occurrence of the idea that junctures define phonological domains. Credit for this conception is usually given to McCawley (1968:52ff) (see §§113,373).

Level Independence abolished: phonology with morpho-syntactic information 76

61

3. Level Independence abolished: phonology does take morphosyntactic information into account Another thing that Chomsky et al. (1956:67) do is to overthrow the structuralist mantra of Level Independence (§61). They argue that there is no reason why morphological and syntactic considerations should not be brought to bear in order to establish a phonemic transcription. Halle (1959) secures this position with an argument that draws on the analogy with other sciences: theoretical constructs are independent of the procedure that leads to their postulation. Hence any evidence, including morpho-syntactic, is good evidence for the establishment of the phonemic string.17 (24)

"The theoretical constructs which make up the representations discovered by the different types of analysis are, however, postulated within the framework of the individual sciences without regard for the procedures whereby they can be discovered in the data. Theoretical constructs are never introduced because of considerations that have to do with analytic procedures. Thus, for instance, it is inconceivable that chemistry would establish substances that can be identified by visual inspection as a category distinct from substances that require more elaborate techniques for their identification. Yet this is precisely the import of Condition (3a), for it sets up a distinction between phonemes and morphophonemes for the sole reason that the former can be identified on the basis of acoustic information alone, whereas the latter require other information as well.[…] The abolition of Condition (3a) is not as much at variance with traditional practice as might at first appear. It is hardly an accident that in the phonological descriptions of E. Sapir, and to some extent also in those of L. Bloomfield, Condition (3a) played no role." Halle (1959:23f)

Another question related to structuralism is whether the phonemic transcription allows, or should allow, to arrive at a unique representation for every utterance heard. Chomsky et al.'s (1956:67) answer is no.

17

Condition (3a) that is mentioned in the quote below is Level Independence: "Condition (3a): A phonological description must include instructions for inferring (deriving) the proper phonological representation of any speech event, without recourse to information not contained in the physical signal" Halle (1959:21).

62 77

Chap 4: Chomsky, Halle & Lukoff (1956)

4. Two for the price of one: multifunctional juncture Liberated from the two structuralist straitjackets discussed in the previous section, Chomsky et al. (1956) can take full advantage of what Moulton (1947) only hinted at: it can hardly be accidental that the distribution of juncture phonemes, which are established on purely phonological grounds, by and large coincides with morpho-syntactic divisions. Chomsky et al. (1956) now are able to emphasize the multifunctional character of juncture. (25)

"Foremost among these advantages is the fact already mentioned that the constituent organisation imposed from purely phonological considerations (i.e., from considering the simplest way to state stress) correlates quite closely with the constituent organisation that is required for the description of English on other levels. This correspondence leads to an overall simplification of the grammar of the language, since the constituent structure once stated can be made to serve a variety of functions." Chomsky et al. (1956:78)

Multifunctional juncture also responds to an objection that could be levelled against Chomsky et al. (1956) on phonological grounds: what kind of economy do we achieve if four stress phonemes are traded against one accent phoneme and two junctures (internal and external)? If the junctures that do the job of stress distribution are needed at higher levels anyway, they are not introduced into the grammar for the mere purpose of stress. Rather, Chomsky et al. (1956) only fertilise for phonological analysis a structure that is needed anyway.

78

5. Morpho-syntactic control over juncture restored, and no phonetic correlate A direct consequence of restoring morpho-syntactic control over the distribution of juncture is that juncture abuse as described in §69 is expunged. That is, phonologists are not free anymore to distribute juncture whenever it suits their analysis; especially, no juncture can ever occur in the middle of morphemes. (26)

"Junctures should be distributed in a manner that is significant on higher levels. Specifically, junctures should appear only at morpheme boundaries, and different junctures should correspond, by and large, to different morphological and syntactical [sic] processes." Chomsky et al. (1956:67)

Morpho-syntactic control over juncture restored, and no phonetic correlate

63

This, in turn, leads to the purely diacritic definition of boundaries that is practised in SPE: the phonetic correlate that was called for because of the phonemic status of juncture (see the structuralist discussion in §70) is now definitely done away with. Recall, for example, Trager's (1962:17) antipathy for utterance-initial juncture because "there is nothing, phonemically speaking, before the beginning". In contrast to this position, Chomsky et al. (1956) deny any phonetic status to boundaries. (27)

"The segmental phonemes represent physical entities and, therefore, each manifestation of a phoneme must have certain stateable physical properties. […] The junctures, on the other hand, do not represent physical entities, but are introduced for the purpose of reducing the number of physical features that must be considered phonemic." Chomsky et al. (1956:66)

On the first page of his book and in a chapter entitled "Segments and Boundaries", Halle (1959) consolidates this view. (28)

"In phonology, speech events are represented as sequences of entities of two kinds : segments, to which specific phonetic (articulatory as well as acoustical) properties are assigned, and boundaries, which are characterized solely by their effects on the former)." Halle (1959:19), emphasis in original

This distinction will later appear as the SPE feature [±segmental] which opposes juncture (negative value) and phonetically expressed segments (positive value) (see §87). In a footnote depending on the quote above, Halle (1959) also justifies the introduction of the term boundary. (29)

"Boundaries are analogous to what some linguists have termed 'junctures'. Since the latter term has recently been used in a very special sense, the more neutral 'boundary' has been adopted here." Halle (1959:19)

Hence it seems that the non-phonetic conception of juncture was at the origin of the introduction of the word boundary into phonology: Halle wanted an alternative to the phoneme-bound juncture.

64 79

Chap 4: Chomsky, Halle & Lukoff (1956)

6. Privativity, an accidental consequence of economy Another aspect that is relevant for the discussion below is privativity. We have seen that Chomsky et al. (1956) require juncture to coincide with some morpho-syntactic division. But is it also true in turn that all morphosyntactic boundaries are projected onto phonology? The answer is no. (30)

"Since junctures are introduced for the purpose of reducing the number of physical features that must be recognized as phonemic, we do not require that every morpheme boundary be marked by a juncture. […] Only those morpheme boundaries are marked by a juncture where actual simplifications in the transcription are achieved. In other words, junctures are postulated only where phonetic effects can be correlated with a morpheme boundary." Chomsky et al. (1956:68)

Chomsky et al. (1956) thus argue for the privative representation of morpho-syntactic information in phonology: only those divisions which indeed produce a phonological effect are projected. SPE will revert back to a non-privative mapping (see §90): all morpho-syntactic divisions land in phonology no matter whether they are phonologically relevant or not. Like most other architectural properties of SPE, this stance abides in more recent theories, among which namely Prosodic Phonology. In contrast to this unanimous generative tradition, it is argued below (§756, also in Vol.2) that translation of morpho-syntactic information into phonological objects must be privative.

80

7. Cyclic derivation (inside-out interpretation) Finally, due mention needs to made of the fact that cyclic derivation, a landmark in interface theory, was born in Chomsky et al. (1956). Cyclic derivation is unprecedented in structuralist or neogrammarian thinking; it has known a number of different incarnations according to the environment of particular theories and their take on the architectural design of grammar. The principle itself, however, stands (almost) unchallenged today (see §672). The idea is that the linear string is not interpreted as a whole. Rather, phonological (and semantic) computation applies to successive, morphosyntactically defined chunks along the hierarchical morpho-syntactic structure. In other words, strings experience inside-out interpretation along the morpho-syntactic tree. For example, given an embedded structure such as

Cyclic derivation (inside-out interpretation)

65

[[[A] B] C], phonology applies to all domains, starting with the most embedded one and moving outwards: first phonology assesses [A], then [AB], and finally [ABC]. That is, embedded pieces experience the same computation several times: first alone, then in the company of other pieces (in our example A is computed three times). Cyclic derivation appears in Chomsky et al. (1956) as a local and anecdotic algorithm that derives stress placement. Chomsky et al. do not provide any further discussion, and no more general ambition is declared. (31)

"Rule 4: Given a phonemic clause, (i) assign the value 1 to all accented words; (ii) then apply each rule pertaining to accented vowels no more than once to each constituent, applying a rule to a constituent of order n only after having applied it to all constituents of order n+1; i.e. beginning with the smallest constituents and proceeding to larger and larger constituents; […]" Chomsky et al. (1956:75)

The anecdotal way in which cyclic derivation was introduced into linguistic thinking contrasts with the landmark qualities that it has acquired in further evolution. This treatment is characteristic for the pre-SPE period, though: discussion only concerned the representational side of the interface coin. Nobody had thought of the possibility that there may also be a procedural way for morpho-syntax to talk to phonology. Starting with Bromberger & Halle (1989:65ff), the idea that phonology is made of procedurally (and extrinsically) ordered instructions is often quoted when it comes to explain in which way structuralism and generativism are different. A whole body of literature has pointed out, however, that ordered rules are not a generative invention: they are also present in Bloomfield's (1939) Menomini Morphophonemics and Wells (1949) (e.g. Huddleston 1972, Encrevé 2000, Koerner 2004). Goldsmith (2008) has recently studied the question whether the transition from structuralist to generative phonology is to be seen as a rupture (as is constantly claimed by generativists) or a continuity. Regarding ordered rules and other candidate properties that may be claimed to set generative phonology apart from its predecessor, Goldsmith concludes that all major generative ideas were present in structuralist thinking (even though they may not have been in the focus of structuralist attention). On this backdrop, cyclic derivation may be a successful candidate for a property that was completely absent from linguistic thinking before the

66

Chap 4: Chomsky, Halle & Lukoff (1956)

advent of generative grammar: inside-out parsing is a genuinely generative idea. In any event, in 1956 we are still miles away from the generative standard that I call Interface Dualism (§683), i.e. the idea that there are two means for morpho-syntax to bear on phonology, one procedural, the other representational. The rebalancing of the exclusively representational view on the interface will only be operated in SPE.

81

Chapter 5 SPE sets the standards for 40 years

82

1. Introduction

83

1.1. Interface Dualism SPE has set the standards for interface theory that are still valid today. This will be obvious as more recent theories are discussed below. Beyond labels and vocabulary, which have changed since the 60s, concepts have remained the same; they may have been worked out in greater detail, couched in various syntactic and phonological environments and cover a larger empirical record, but the overall architecture has remained unchanged. The headstone of the interface architecture of SPE is what I call Interface Dualism. That is, morpho-syntax talks to phonology through two channels, one procedural, the other representational. Cyclic spell-out is the procedural way of bearing on a phonological process. It was introduced in Chomsky et al. (1956:75) (see §80) and is known as derivation by phase today (Chomsky 2000a et passim) (see §§304,673, Bermúdez-Otero forth b provides an overview). Note that SPE did not provide for any no look-back device. The ancestor of today's Phase Impenetrability Condition (PIC) was only introduced in Chomsky's (1973) Conditions on Transformations under the header of the Strict Cycle Condition. It was then applied to phonology by Kean (1974) and Mascaró (1976) in the 70s, before becoming an important tool of Lexical Phonology (see §287 for a survey). The representational way of talking to phonology is through objects that carry morpho-syntactic information and are inserted into the phonological module. These objects were juncture phonemes in structuralism (§72), they are boundaries in SPE, and the autosegmental equivalent of the 80s will be the Prosodic Hierarchy (see §360). I argue that the dual nature of communication is critical for a correct interface theory: the post-SPE development has produced models which have maximised either the procedural (Lexical Phonology, §139, Kaye 1995, §258) or the representational channel (Prosodic Phonology, §360). These monocultures cannot account for all patterns found.

68 84

Chap 5: SPE sets the standards for 40 years

1.2. Modular seeds in SPE: the inverted T and translation The generative approach to language is modular at heart, both regarding the interaction of grammar with other cognitive systems (physiology, perceptual psychology, acoustics, vision etc.) and its internal organisation proper. Modularity is introduced at greater length in the Interlude (§586, also §410). The reason why it needs to be mentioned here is that the modular idea is part and parcel of the generative enterprise since the 60s, even though modularity as a general theory of the cognitive system (Fodor 1983) was still in an embryotic state in the 60s (see §623). As we will see below, it clearly shines through in SPE (without the word modularity appearing, though), but is also happily violated when it comes to labelled brackets. Modularity holds that the human cognitive system in general, and language in particular, is made of several independent computational systems that are specialised for the execution of a particular task and autistic in the sense that they do not know, or care for, what is going on outside of their narrow limits. Also, modules are non-teleological, i.e. they pursue no other goal than successfully carrying out their computation. This idea may be traced back to Chomsky (1955-56) (see §§623,628, where the development of modularity in generative grammar is discussed at greater length). The following quote from SPE shows that modularity is really what Chomsky and Halle meant when they were talking about phonology, even though they were lacking the modern vocabulary. (32)

"The rules of the grammar operate in a mechanical fashion; one may think of them as instructions that might be given to a mindless robot, incapable of exercising any judgment or imagination in their application. Any ambiguity or inexplicitness in the statement of rules must in principle be eliminated, since the receiver of the instructions is assumed to be incapable of using intelligence to fill in gaps or to correct errors." Chomsky & Halle (1968:60)

Modularity has shaped the general architecture of generative grammar, which at least since Aspects (Chomsky 1965:15ff) recognises three modules (syntax, phonology and semantics). The corresponding inverted T model is presented in §86 below. This architecture is still in place today (e.g. Chomsky 2000a et passim, §304), and the loops that it made in the development of interface theory will be a loyal companion through subsequent chapters. The second obvious impact of modularity on how SPE conceives the interface with morpho-syntax is the existence of translation: SPE bothers transforming morpho-syntactic structure into items that are part of the in-

Introduction

69

ventory of phonological objects, #s. In the view of SPE, indeed, hashmarks are regular segments ([-segment] segments, see §87). Were morpho-syntax and phonology one, and could they understand the vocabulary that the other uses, no such translation would be needed in the first place. The existence of translation is thus a clear trace of modular thinking (§85 describes how it is organised in SPE). It was already argued in §72 that the modular idea was seeded by the structuralist principle of Level Independence (notwithstanding the fact that it was born in absence of any reference to the cognitive system, see §61). Early generative work was of course embossed by the overarching methodological principle of Level Independence that guided structuralist work: this is what Chomsky et al. (1956) were wrestling with (§76). Also, there is a very clear continuance of translation since structuralist days: successive phonological theories translate morpho-syntactic information into their genuine phonological vocabulary: first into (juncture) phonemes, then into ([-segment]) segments, finally into autosegmental structure (the Prosodic Hierarchy) (see §§379,692). It is thus certainly true that the existence of translation witnesses modular thinking – but this thinking is not a generative privilege. Generativists built the modular inverted T model, which requires translation, on the foundations of structuralist Level Independence. Hence generative grammar may not get credit for the idea that morpho-syntax and phonology are distinct waterproof worlds. What it does get credit for, however, is the institutionalisation of translation whose existence, recall, was denied by the official structuralist mantra, but circumvented in practice. This aspect of the continuity between structuralist and generative thinking is taken up at greater length in §§415,693. The discussion also shows that it was only the implementation of a translational mechanism that raised the diacritic issue (§78), which will run all through the remainder of the book. Finally, it is worth mentioning that the headstone of Prosodic Phonology, Indirect Reference, is also an offspring of modularity, even though for some reason this went unnoticed in the 80s (see §414). We will see that Indirect Reference roots in SPE: its ancestor is the Readjustment Component. To date the literature motivates Indirect Reference with the original SPE example (cat-rat-cheese, see §91): the pattern is called nonisomorphism (see §416).

70

Chap 5: SPE sets the standards for 40 years

85

2. Translation in SPE (output: boundaries)

86

2.1. The inverted T model Modularity conceives the internal organisation of grammar as a collaboration between distinct and independent computational systems. This architecture was made explicit in Aspects (Chomsky 1965:15ff): (morpho-)syntax is the central concatenative system, whose output is interpreted by phonology (PF) and semantics (LF); these produce form and meaning, respectively. The architecture is shown under (33) below; it is known as the inverted T model (or the Y model). (33)

the inverted T model: all concatenation before all interpretation morpho-syntax

PF

LF

In SPE and further practice, morpho-syntax and the two interpretative modules are procedurally ordered so that words and sentences are pieced together before being shipped to interpretation at PF and LF. That is, all concatenation is done before all interpretation.18 While the inverted T model stands unchallenged up to the present day as a whole, the full completion of concatenative activity before interpretation can begin will be a bone of contention in further development. It is therefore important to realise that the requirement "all concatenation before all interpretation" does not follow from the inverted T: it is an additional plug-in that needs to be motivated independently. It is certainly true that no such motivation was provided in SPE and the post-SPE literature, where "all concatenation before all interpretation" was assumed without discussion. The competing view is so-called interactionism, according to which concatenation and interpretation are interspersed. Interactionism was introduced by the stratal architecture of Lexical Phonology (§146); it then 18

Bresnan (1971) is an early and isolated case where this principle is violated (by a syntactician). Bresnan's analysis of English intonation is discussed in §308.

Translation in SPE (output: boundaries)

71

prompted Halle & Vergnaud's (1987a) reaction, which attempts to restore "all concatenation before all interpretation" (§216). Modern phase theory, however, is interactionist at heart (§304), and so are Kaye (1995) (§275) and Distributed Morphology (of which Morris Halle is a founder, §532).

87

2.2. Boundaries

88

2.2.1. Boundaries are [-segment] segments without phonetic correlate In order to see how translation works in SPE, let us begin by introducing its output: boundaries. The generative approach to language conceives grammar as a set of basic symbols, which are modified by a computational system. In SPE, binary distinctive features (which are constitutive of segments) are the basic symbols, which are manipulated by (ordered) rules. Whatever is found in natural language must be formalised as an interplay of structure (features/segments) and processes (rules). Boundaries are no exception: SPE considers them to be ordinary segments. Their only peculiarity is the fact that unlike /p/, /e/ and the like, they do not have any phonetic manifestation. SPE thus continuates the nonphonetic generative view on juncture that Chomsky et al. (1956) have argued for (see §78). Boundaries now root in morpho-syntactic structure alone: their job is to carry its information into the phonology. Economy (of phonemes) is not an issue anymore (there are no phonemes), and there cannot be any boundary in absence of a morpho-syntactic division (i.e. in the middle of morphemes, see §78). This non-phonetic scenario for boundaries, however, faces the same conceptual trouble as its juncture-based equivalent: all segments are supposed to have a phonetic correlate, and the properties of segments are defined by their featural make-up. Since only features can make segments different, then, the peculiar non-phonetic status of boundary segments must be due to some feature. Therefore Chomsky & Halle (1968:66f) use the feature [±segment] in order to distinguish regular segments that have a phonetic realisation ([+segment]) from boundary segments that do not ([-segment]). They are explicit on the absence of any stable phonetic correlate of [-segment] segments: "boundary features do not have universal phonetic correlates, except perhaps for the fact that word boundaries may optionally be actualised as pauses" (Chomsky & Halle 1968:364).

72

Chap 5: SPE sets the standards for 40 years

SPE thus practices exactly the same solution for the integration of diacritics into phonological vocabulary as structuralism (where, recall from §64, juncture phonemes were said to be realised as a "pause of brief duration or, in free variation with this, as zero").

89

2.2.2. Different types of boundaries: +, =, # SPE recognises three different boundaries: "#", "+" and "=". This contrast is motivated as before in structuralism: different boundaries are needed because they provoke different phonological effects. However, unlike Chomsky et al. (1956) and McCawley (1968), SPE does not establish any hierarchy among the three boundaries at hand; rather, they remain an amorphous set of unranked and unordered [-segment] segments (Chomsky & Halle 1968:371). The three boundaries fall into two groups: + and = are "morphological", while # is "syntactic". That is, the former are recorded in the lexicon together with morphemes (all and only those lexical entries that are morphologically complex bear a = or a +), while # is absent from the lexicon: it represents information that was created online by morpho-syntactic computation. Hashmarks thus by and large reproduce the morpho-syntactic tree and are inserted into the phonological string by a general mapping algorithm (on which more in the following section). The difference between + and = is one of learned (Latinate) vs. regular vocabulary: the latter only occurs with prefixes in words such as per=mit, de=signate, inter=dict, ad=voc+ate, con=de=scend and the like (Chomsky & Halle 1968:94f), while the former is the morphological default. It occurs at all morpheme boundaries within a lexical entry, as well as at edges, hence for example /+para+site+/. Clusters of # and + that arise after lexical insertion are reduced to one single # or + by convention (Chomsky & Halle 1968:12f, 364f). The = boundary was exotic since the heydays of SPE (it did not do much labour), and was rapidly abandoned (e.g. Siegel 1980). As for all other segments, the contrast among boundaries could only be achieved by features, and a three-way distinction requires two binary features. Therefore SPE introduced the distinctive features [±word boundary (WB)] and [±formative boundary (FB)]. The segment #, then, is specified as [+WB, -FB], + is made of [-WB, +FB], while = identifies as [-WB, -FB] (Chomsky & Halle 1968:66f). Basbøll (1975:110) points out

Translation in SPE (output: boundaries)

73

that the fourth logical possibility, [+WB, +FB], begs the question – an issue which is not taken up by Chomsky & Halle (1968).

90

2.3. The general mapping algorithm: boundaries restore (almost) full morpho-syntactic information In practice, then, SPE works down all the way from syntax and morphology to phonology. Deep syntactic structure is first converted into surface syntactic structure, which serves as the input to phonology. Independently of lexical insertion that treats only morphemic information (through a lexical access), morpho-syntactic structure is converted into boundaries according to a fixed algorithm that works on the grounds of morpho-syntactic criteria alone: # is inserted into the linear string at the beginning and at the end of each major category (i.e. nouns, verbs and adjectives), and also on each side of higher constituents that dominate major categories, i.e. NPs, VPs and so forth (Chomsky & Halle 1968:12f, 366ff). The distribution of # boundaries over the linear string restores all morpho-syntactic information in phonology except for one particular situation: successive morphemes that belong to the same major category are not separated by a #. Their morpho-syntactic relationship thus remains unreflected in the phonology. A case in point is the root [tele+graph] under (34) below: tele- and -graph being both nominal under (34b), they are left unseparated by a # boundary in the phonological output under (34c). We will see in §103 that the distribution of brackets, i.e. of cycles, is done by the same algorithm as the one that places # boundaries: brackets and #s are exactly isomorphic. The process of translating morpho-syntactic structure into representational objects that are then inserted into phonological representations will later be called mapping (§379). While the mapping algorithm is simple and identical for all languages in SPE, it will expand into a whole mapping component with a host of parameterised mapping rules in further development (§386). Table (34) below reproduces a simplified version of Chomsky & Halle's (1968:13) favourite telegraphic example that shows how morphosyntactic structure is converted into boundaries.

74 (34)

Chap 5: SPE sets the standards for 40 years mapping in SPE: exhaustive translation of morpho-syntactic structure into linear boundaries a. morpho-syntactic and phonological representations of the sentence "we established telegraphic communication" b. morpho-syntactic structure and phonological terminals [S# [NP#[N# we #]#] [VP#[V'#[V# establish #] past #] [NP#[A#[N# tele+graph #] ic#] [N#[V# communicate #] ion #]#] #] #] c. phonological result of the translation alone ### we ##### establish # ed ## ## telegraph # ic ### communicate # ion ####

Of course, (34b) is not an actual state of the grammar; it just helps visualising (less well than the tree structure in Chomsky & Halle's original) where #s are placed in recognition of which morpho-syntactic node. This mechanism produces important clusters of #s, which are reduced to maximally two consecutive items by convention (see Selkirk 1972:12, 1974:578). Translation in SPE thus produces a string of morphemes where the "syntactic distance" of two adjacent items may be read off by looking at hashmarks: the more of them separate two morphemes, the more distant they are. Finally, all boundaries that are left at the end of the phonological derivation are erased by convention before the output of phonology is shipped off to the phonetic component. Unlike Chomsky et al. (1956), SPE thus promotes a non-privative management of morpho-syntactic information. That is, all higher level divisions are injected into the phonology, including those that are useless. These are then either erased by conventional #-cluster reduction, or transit phonology in order to be eliminated at the end of the derivation This raises the issue of economy: why should the computational system translate phonologically irrelevant material into boundaries? Why should phonology store and carry useless objects on its back all through the phonological derivation? And why, finally, should additional computational effort be conceded in order to eliminate the irrelevant? Whether translation should be privative or not is a pervasive issue in interface theory; although not in a prominent position, it will escort the reader through the remainder of the historical survey. A summary is then provided in §757.

Translation in SPE (output: boundaries) 91

75

2.4. Readjustment We have seen that SPE translates morpho-syntactic structure into boundaries, that is into vocabulary that can be parsed by the phonology. Most of the time, the equivalence is one-to-one, and an invariable mapping algorithm carries out the translation; nothing more needs to be said in this case. Sometimes, however, (translated) syntactic surface structure does not qualify as the input for phonological interpretation: it needs to be modified (readjusted) before phonology can make use of it. (35)

"We have two concepts of surface structure: input to the phonological component and output of the syntactic component. It is an empirical question whether these two concepts coincide. In fact, they do coincide to a very significant degree, but there are also certain discrepancies. These discrepancies […] indicate that the grammar must contain certain rules converting the surface structures generated by the syntactic component into a form appropriate for use by the phonological component. In particular, if a linguistic expression reaches a certain level of complexity, it will be divided into successive parts that we will call 'phonological phrases', each of which is a maximal domain for phonological processes. […] It appears that the syntactic component of the grammar generates a surface structure Σ which is converted, by readjustment rules that mark phonological phrases and delete structure, to a still more superficial structure Σ'. The latter then enters the phonological component of the grammar." Chomsky & Halle (1968:9f)

This quotation contains a good deal of what will be at the forefront of interface discussion in the 80s when Prosodic Phonology (§360) is developed. In the environment of SPE, the proviso introduced is that translation always inserts boundaries into the phonological string, but that these boundaries may represent syntactic surface structure (in most cases), or a readjusted version thereof (exceptionally). The reason for readjustment is the same as for what will be called Indirect Reference in Prosodic Phonology later on (§§378,406): the existence of cases where phonology obviously relies on a constituent structure that is different from the morpho-syntactic output. This pattern is called non-isomorphism in Prosodic Phonology (see §§416ff). The quote below shows how it was introduced in SPE on the grounds of the cat-rat-cheese example.

76

Chap 5: SPE sets the standards for 40 years

(36)

"Consider, for example, sentences such as (124), where the three bracketed expressions are the three noun phrases in the predicate: (124) This is [the cat that caught [the rat that stole [the cheese]]] Clearly, the intonational structure of the utterance does not correspond to the surface structure in this case. Rather, the major breaks are after cat and rat ; that is, the sentence is spoken as the three-part structure this is the cat – that caught the rat – that stole the cheese. This effect could be achieved by a readjustment rule which converts (124), with its multiply embedded sentences, into a structure where each embedded sentence is sister-adjoined in turn to the sentence dominating it. The resulting structure appears then as a conjunction of elementary sentences (that is, sentences without embeddings). This allows us to say that intonation breaks precede every occurrence of the category S (sentence) in the surface structure and that otherwise the ordinary rules prevail." (emphasis in original) Chomsky & Halle (1968:371f)

Chomsky & Halle thus suggest to dis-embed the cat-rat-cheese sentence: the constituent structure that syntax has produced ([the cat that caught [the rat that stole [the cheese]]]) is readjusted so that three sister nodes are created: [[the cat] [that caught the rat] [that stole the cheese]]. Note that readjustment and mapping into boundary structure are two different things: the former manipulates brackets, not boundaries, and takes place before mapping translates brackets into boundaries. Phonology then uses the readjusted and translated structure. Finally, due mention needs to be made of the fact that readjustment is an unpredictable distortion of the original morpho-syntactic structure. That is, the content and behaviour of the readjustment component is idiosyncratic: it does not depend on either morpho-syntax or phonology. In the 80s, Prosodic Phonology will reproduce the SPE architecture one-to-one, including the readjustment component and its unpredictable action (see §380).

Translation in SPE (output: boundaries) 92

2.5. Affix classes and their representational management

93

2.5.1. Rule-blocking boundaries (stress-shifting affixes)

77

Chomsky & Halle (1968) were aware of the fact that English has two affix classes: SPE is the source of this empirical generalisation. Its most prominent effect is the (non-)bearing of affixes on stress placement. This is identified as follows in SPE. (37)

"Alongside of the affixes that affect stress placement […], there are other 'neutral affixes' which characteristically play no role in the placement of stress, for example, the adjective-forming affixes -y, -like, -able, -ish and affixes such as -ing, -past tense, -hood, -ness, -ly, -wise. We can indicate the fact that an affix is neutral by making use of the # boundary." Chomsky & Halle (1968:84)

For Chomsky & Halle, the challenge raised by this situation is to make sure that the Main Stress Rule (which shifts stress one vowel right upon every concatenation of a suffix, i.e. on every cycle) does not apply to strings that contain stress-neutral affixes. Stress-shifting affixes, on the other hand, are regular in regard of the Main Stress Rule. As shown by the quote, the solution adopted is representational: stress-neutral affixes (also called class 2) come with a #, while stressshifting affixes (class 1) are only equipped with a + boundary. The Main Stress Rule, then, is supplied with a proviso in prose which says that the rule can apply only to strings where no # boundary is found (Chomsky & Halle 1968:85). This makes sure that the application of the Main Stress Rule is blocked when it applies to [root # affix] (class 2 strings). Hence in [[parent] #hood], the Main Stress Rule assigns stress to the first vowel on the initial cycle, i.e. [párent], but cannot shift stress any further to the right on the second cycle although the structural description is met: -hood comes with a # boundary, which thus occurs in the middle of the second cycle [párent # hood]. Therefore the Main Stress Rule is blocked, and the output of the derivation is párent-hood. By contrast, the stressshifting suffix -al (class 1) comes with a + boundary. The Main Stress Rule being insensitive to +, it also applies on the second cycle of [[parent] +al] and hence shifts stress one vowel right, which produces parént-al.

78 94

Chap 5: SPE sets the standards for 40 years

2.5.2. Rule-triggering boundaries (stress-neutral affixes) The same lexically encoded contrast also accounts for all other affix classbased phenomena. The strategy is always the same: phonological rules are made sensitive to #, which either blocks or triggers the process at hand. This distinction corresponds exactly to Kenstowicz & Kisseberth's (1977, 1979) typology of morphological impact on phonological processes that was introduced in §51: morpho-syntactic divisions may be rule-blocking or rule-triggering. We have seen how # blocks stress assignment in SPE; all other cases of rule-blocking boundaries (which will be encoded as level 1 rules in Lexical Phonology) work along the same lines. Rule-triggering boundaries (level 2 rules in Lexical Phonology), on the other hand, appear in the structural description of rules: the processes at hand only go into effect if the string contains the boundary. English features three processes of this kind, all of which control the simplification of a stem-final cluster that contains a nasal: gn-n (sign, sign-ing2 vs. sign-ature1), mn-n (damn, damn-ing2 vs. damn-ation1), mb/ŋg-m/ŋ (sing, sing-ing2 vs. long-er1, bomb, bomb-ing2 vs. bomb-ard1). Chomsky & Halle (1968:85) point out that "the inflectional affixes which are neutral with respect to stress also characteristically affect final clusters in the same way as word boundary does." That is, cluster simplification occurs word-finally and before stress-neutral (i.e. class 2) affixes. This disjunction is good reason indeed to suppose that the identity of stressneutral (class 2) affixes is to bear a # boundary: whatever the identity of the right edge of the word, it must also characterise stress-neutral (class 2) affixes. In practice, then, the rule that is responsible for the gn - n alternation for example will be as follows: g → ø / __n #. The velar in [sign#] and [sign#ing], but not in [sign+ature], is thus deleted. An additional benefit is that the rule blocks deletion morpheme-internally, where all of the clusters mentioned systematically survive: ignore, amnesia, finger. The inventory of affix class-based phenomena that are found in English will be introduced with greater care in §§162f,166: their analysis is foundational for Lexical Phonology. For the time being, it is enough to bear in mind that they enjoy an entirely representational analysis in SPE.

Translation in SPE (output: boundaries) 95

2.6. Labelled brackets

96

2.6.1. How brackets and labels are used in the phonology: bláckboard vs. black bóard

79

SPE also makes constant and crucial use of brackets which indicate morpho-syntactic constituent structure. Consider under (38) below the contrast between the noun blackboard and the NP black board that Chomsky & Halle (1968:16) propose (the notation is theirs). (38)

cohabitation of brackets and boundaries in SPE a. bláckboard [N# [A# black #]A [N# board #]N #]N b. black bóard [NP# [A# black #]A [N# board #]N #]NP

The only difference is the morpho-syntactic label of the outermost constituent, i.e. N for blackboard, NP for black board. This contrast is reflected in the phonology by different stress contours: the noun has initial stress (bláckboard), while the NP is stressed on the second element (black bóard). Based on (38), which is the input to phonological computation, Chomsky & Halle (1968:16f) derive this contrast by the cyclic application of the following three rules.19 (39)

compound and nuclear stress rule a. stress assignment in monosyllables, the vowel receives primary stress. b. compound rule assign primary stress to a primary-stressed vowel in the context __...Vˊ …]N c. nuclear stress rule assign primary stress to a primary-stressed vowel in the context Vˊ …__...]NP

Based on (38), the derivation now proceeds from inside out, i.e. by applying phonology to all domains that are identified by brackets, starting with the most embedded domain and moving outwards. When the deriva19

For the sake of exposition, Vˊ indicates primary stressed, Vˋ secondary stressed vowels (instead of the integer-superscripts of SPE). Further distinctions that are made in SPE (3-stress, 4-stress) are irrelevant for the discussion.

80

Chap 5: SPE sets the standards for 40 years

tion moves on to larger domains, inner (old) brackets are erased. This procedure is called the transformational cycle in SPE and will later be known as the phonological cycle, or simply as cyclic spell-out (or cyclic derivation). This concept is introduced at greater length in §100 below. In the respective inner domains, then, only (39a) is applicable. After the erasure of inner brackets, the result is as under (40) below. (40)

intermediate representations a. bláckboard [N# # bláck # # bóard # #]N b. black bóard [NP# # bláck # # bóard # #]NP

All vowels have thus been assigned primary stress. The rules under (39) now reapply to the outer domain, and this time (39b) and (39c) are applicable: given the explicit mention of N and NP in their structural description, the former applies to the noun (but not to the NP), while the latter transforms the NP (but not the noun). Chomsky & Halle then recur to a stress demotion convention which avoids that a given domain contains more than one primary stress: "when primary stress is placed in a certain position, then all other stresses in the string under consideration at that point are automatically weakened by one" (Chomsky & Halle 1968:16f). The application of (39a,b) and this convention, then, produces the correct result: bláckboard and black bóard. 97

2.6.2. Brackets and boundaries are (not) redundant: compensation vs. condensation The derivation shown in the preceding section is but an example of how SPE works in general: phonological derivations have (almost) complete and constant access to both morpho-syntactic structure (brackets) and morpho-syntactic labels (N, NP etc.). Brackets are needed in order to run cyclic derivation: in their absence, embedded domains could not be identified. At the same time, having phonological rules refer to morpho-syntactic labels is the easiest way of making phonology sensitive to the nature of morpho-syntactic nodes when these are phonologically relevant (as in the case of bláckboard vs. black bóard). Two questions arise when looking at this setup: why is morphosyntactic structure translated into boundaries with rules making reference to

Translation in SPE (output: boundaries)

81

the output of the translation (rather than to the original structure)? This seems superfluous given that morpho-syntactic labels can be made direct reference to. Second, given that brackets and # boundaries are exactly isomorphic (see §90), isn't the double translation of morpho-syntactic structure into both redundant? If phonology has (almost) full access to constituent structure via brackets, what are boundaries and their complicated translational mechanism good for? Chomsky & Halle (1968:370f) are explicit on the fact that brackets and boundaries are not the same thing. The difference boils down to the fact that the former are hard-wired, while the latter are ordinary ([-segment]) segments which for this reason can be freely manipulated by phonological rules: # can be erased or transformed into + by rule (see §112). By contrast, brackets are invisible and untouchable for phonological computation: they may not appear in either the structural change or the structural description of rules. Also, they can only be modified by readjustment, which is done prior to phonological action. The only thing that brackets do is to guide the derivation along the embedded morpho-syntactic structure. Chomsky & Halle are explicit on this. (41)

"In our treatment, boundaries are units in a string, on a par in this sense with segments. Like segments, each boundary is a complex of features. Boundaries function rather differently from the various types of constituent markers (labelled brackets) that play a role in determining the application of the phonological rules of the transformational cycle." Chomsky & Halle (1968:371)

Chomsky & Halle (1968:370) also call on an example in order to show that the distortion which is introduced by an eventual rule-triggered deletion of boundaries makes brackets the only visible trace of morphosyntactic structure ("there are instances where word boundaries must be deleted but constituent structure maintained", Chomsky & Halle 1968:370). The two words compensation and condensation (a pair which is often quoted up to the present day, and on which more in §547) look much the same at first sight. However, the second vowel of the former, but not of the latter, may be reduced to schwa. Chomsky & Halle (1968:39, 370) point out that this correlates with a contrast in the derivational history of the two words: while condensation is derived from the verb to condénse, there no such thing as *to compénse. Rather, compensation is derived from to compensáte, which means that the two nouns do not bear the same suffix: condens-ation vs. compensat-ion. Their respective structure is thus as under (42) below.

82 (42)

Chap 5: SPE sets the standards for 40 years condensation vs. compensation a. condensation [N# [V# condens #]V at+ion #]N b. compensation [N# [V# compensat #]V ion #]N

Therefore, on the inner cycle, [condéns] receives stress on the second vowel (counting from the left margin), while [compensát] is stressed on the third vowel. According to Chomsky & Halle, this is the reason why the reduction of the e is blocked in condensátion: the vowel was stressed on an earlier cycle and is therefore protected. This is not the case for compensátion, whose second vowel has never been stressed. Crucial for this analysis is the independent interpretation of the inner cycle of [#[#condens#] at+ion#]. The relevant chunk is identified by brackets, but also by boundaries. Chomsky & Halle (1968:370) point out that the latter, however, cannot do the job of delineating the inner cycle since the right boundary of [#condens#] must be erased. This is because of stress, which shifts to the right on the outer cycle: -ation is a stress-shifting (class 1) suffix. In SPE, the difference between stress-shifting (class 1) and stressneutral (class 2) affixes is that the latter, but not the former, come with a # boundary (see §92). This means that the # boundary, which was initially inserted between the root and -ation, must be deleted by rule: otherwise stress would not shift, and *condéns-ation would be derived. After the elimination of the boundary, however, boundary structure cannot identify the inner cycle anymore. Therefore, Chomsky & Halle (1968:370) argue, "there are instances where word boundaries must be deleted but constituent structure maintained." Now it could be argued that boundary erasure takes place only after the boundary has served for cycle identification, i.e. once the inner cycle has been processed. As far as I can see, this option is not considered by Chomsky & Halle (1968:370), who merely state that "the elimination of internal # […] can be taken care of by a lexical rule which will be automatic with these and various other affixes and which will affect the boundary but not the constituent structure." Be that as it may, the fact is that SPE deliberately organises a twofold representation of morpho-syntactic structure in phonology: in terms of brackets and in terms of boundaries.

Translation in SPE (output: boundaries) 98

83

2.6.3. Brackets and interactionism In sum, SPE provides for a hard-wired and a soft representation of morphosyntactic structure. Boundaries are a perfect incarnation of the modular idea: they represent morpho-syntactic information that was translated into regular phonological vocabulary by a mapping mechanism, and they do not carry any labels: the only thing that they inject is the geometric structure of the morpho-syntactic tree. Labelled brackets, on the other hand, are modularity offenders: they give full and permanent access to morpho-syntactic structure and labels, that is to objects that are non-phonological in kind and have not been translated. As was pointed out earlier (§97), however, phonological computation (rules) cannot make reference to brackets, whereas they are able to refer to their labels, and also to boundaries. Access to morpho-syntactic structure and labels is thus organised by two distinct objects of the linear string: boundaries provide the former, while brackets carry the latter information. We have seen that Chomsky & Halle are also explicit on the fact that the only labour which is done by brackets is to identify relevant chunks for cyclic derivation. In other words, brackets seem to be representational objects, but they serve the purpose of organising the procedural communication with phonology. Hence they somehow representationalise the procedural channel. Given the modern vantage point, one may ask whether SPE really means that brackets are present in the phonological string when they graphically appear in the midst of segments. They could also be used in a metaphorical sense, i.e. as a shorthand notation for morpho-syntactic structure which is still accessible, and along which spell-out moves. What exactly, then, is the difference between modern derivation by phase (Chomsky 2000a et passim, §304) and SPE-style cyclic spell-out? Could the SPE mechanism just be shorthand for the modern idea of multiple spell-out where successive pieces of the morpho-syntactic tree are shipped off to interpretation? The answer, I think, is yes in principle: nothing in the architectural setup needs to be changed in order to eliminate brackets from the linear SPE-string, interpreting them as successive phases. But this is true only in principle – in practice, the plugin to the inverted T model according to which phonology applies only once the morphosyntactic derivation is complete (§86) stands in the way. In other words, interactionism is the bone of contention. We will see in §146 that Lexical Phonology introduces the idea that morphology and

84

Chap 5: SPE sets the standards for 40 years

phonology are interleaved: first you concatenate some pieces, then you interpret them, then you add some more pieces, then you interpret the new string and so on. The inverted T model as introduced in §86 is incompatible with interactionism; we will see that this is the major argument that Halle & Vergnaud (1987a) use against Lexical Phonology in their attempt to restore SPE in the environment of the 80s (§222). Since Epstein et al. (1998:46ff), Uriagereka (1999) and Chomsky (2000a et passim), however, and for reasons that have got nothing to do with phonology, generative grammar has become interactionist: the inverted T model is still in place, but the additional proviso that the complete morpho-syntactic derivation must be achieved before phonology can run is abandoned (§304). A summary of the issue involving brackets and interactionism is provided in §672.

99

2.6.4. Modularity and modularity offenders in SPE Let us now look at SPE's ambiguous relationship with modularity. On the one hand, the generative architecture of grammar (the inverted T, §86) is an application of modularity to language. Also, Chomsky & Halle have a typically modular conception of phonological computation (see the quote in §84 and compare with the description of modular activity in the Interlude §586).Finally, like its structuralist predecessor, SPE provides for translation (of morpho-syntactic structure into phonological boundaries), and translation is a necessary consequence of modularity (domain specificity: modules do not speak the same language). Another fundamental property of SPE, however, is a strong modularity offender: labelled brackets. Brackets are phonological aliens (unlike boundaries, they are invisible for rules because they are non-segments) and therefore should not be able to be parsed by the phonological component. However, phonology must somehow take brackets into account when processing the linear string: this is the only way to know how cycles are delineated. It was mentioned in the previous section that SPE may probably receive an interactionist interpretation from hindsight, in which case brackets disappear and the inverted T is fully compatible with modularity (see §680). This does not do away with the even more offending property of brackets to be labelled, though: rules can make direct reference to untranslated morpho-syntactic labels (N, V, NP etc.). This should be excluded in a

The phonological cycle and one single phonology

85

modular environment since phonology cannot parse this kind of alien vocabulary. In sum, SPE certainly sensed modularity and applied its major consequence, translation, "intuitively". That is, there is no explicit discussion of the matter in SPE: modularity as such remains unnamed and undefined; and it is violated by a central device of the theory, labelled brackets. Only in the 80s will the reference to untranslated morpho-syntactic information and especially to morpho-syntactic labels be explicitly prohibited under the header of Indirect Reference (§§377,406) – but even then without explicit mention of modularity as such (§414).

100

3. The phonological cycle and one single phonology

101

3.1. The phonological cycle

102

3.1.1. Cyclic derivation: how it is motivated and how it works The phonological cycle is a hallmark of generative phonology: it runs through the entire literature from the earliest contribution until the most recent versions of minimalism where it appears in the guise of phase theory. In actual fact, the idea that there is a procedural communication between morpho-syntax and phonology is only carried into linguistic thinking by the generative approach: since the 19th century, this option had never been considered. The introduction of cyclic derivation thus creates what I call Interface Dualism (see §683 for a summary): it adjoins a procedural channel to the traditional representational channel of communication. Cyclic derivation was introduced in Chomsky et al. (1956:75) as a local (and nameless) algorithm for the derivation of stress (see §80). In the 60s, the generative architecture of grammar in form of the inverted T model was developed (§86). In this environment, SPE makes cyclic derivation the general mechanism that controls the communication between morphosyntax and phonology/semantics. While cyclic derivation is referred to as the Transformational Cycle in SPE, it is better known as the phonological cycle in the 70s and the 80s (a label that is due to Mascaró 1976, see §189). Today, the same principle has reincarnated as derivation by phase (Chomsky 2000a et passim), an important tool at the forefront of the minimalist programme (§549). The following quote illustrates the general value that Chomsky & Halle grant to cyclic derivation.

86

Chap 5: SPE sets the standards for 40 years

(43)

"Investigation of English and other languages confirms this expectation and permits us to formulate the principle of the transformational cycle in full generality, applying to all surface structure whether internal or external to the word." Chomsky & Halle (1968:27)

The cyclic interpretation of the string that is submitted to the phonology proceeds step by step from inner domains outwards. SPE offers the following description. (44)

"Regarding a surface structure as a labeled bracketing […], we assume as a general principle that the phonological rules first apply to the maximal strings that contain no brackets, and that after all relevant rules have applied, the innermost brackets are erased; the rules then reapply to maximal strings containing no brackets, and again innermost brackets are erased after this application; and so on, until the maximal domain of phonological processes is reached." Chomsky & Halle (1968:15)

Chomsky & Halle (1968) insist on the fact that the phonological cycle expresses a natural and intuitive generalisation. (45)

"Notice, once again, that the principle of the transformational cycle is a very natural one. What it asserts, intuitively, is that the form of a complex expression is determined by a fixed set of processes that take account of the form of its parts." Chomsky & Halle (1968:20)

Illustration of how cyclic derivation works was already provided in §§96f (bláckboard vs. black bóard, compensation vs. condensation). Despite its generalization to the entire grammar in further development, the intimate relationship that cyclic derivation entertains with stress should be borne in mind: the very idea was born in Chomsky et al. (1956:75) on the basis of English stress, and only rules related to stress placement are cyclic in SPE. 103

3.1.2. Cycles are defined like boundaries: by major categories (N,V,A) It is important for SPE (but not for more recent models of cyclic derivation such as Distributed Morphology, see §550) that the result of every cycle is an actual word which may be pronounced and has a meaning. Hence the word theatricality that Chomsky & Halle (1968:88f) discuss is made of five morphemes that are distributed over three cycles: [[[theatr]N ic + al]A i + ty]N. The delineation of cycles follows the distribu-

The phonological cycle and one single phonology

87

tion of # boundaries: it is determined by major categories. Recall from §90 that nouns, verbs and adjectives as well as projections thereof are flanked by # boundaries in phonology. The output of each cycle, then, is an actual word: théatre, theátr-ic-al, theatr-ic-al-i-ty. As far as I can see, the fact that brackets (i.e. cycles) only delineate transitions between major categories is not made explicit in SPE. Examples like theatricality, however, show that successive morphemes that belong to the same major category are not separated by brackets. The # boundary and brackets are thus exactly isomorphic (at a certain point in the derivation: right after translation, i.e. before eventual rule-governed modifications of boundaries, see §§97,112). As a consequence, the restoration of morpho-syntactic structure in phonology is only almost complete: transitions between morphemes that belong to the same major category remain invisible – invisible in terms of #s and brackets; they are of course marked by the default boundary + that accompanies every morpheme.

104

3.2. Word-level rules and the unity of phonological computation

105

3.2.1. Cyclic vs. word level rules Cyclic derivation along the transformational cycle concerns all pieces of a linguistic derivation, from the most embedded morpheme up to the largest syntactic unit. SPE is explicit on the fact that there is no difference between morphemes and words: "the principle of the transformational cycle […] appl[ies] to all surface structure whether internal or external to the word" (Chomsky & Halle 1968:27). This notwithstanding, SPE distinguishes between two kinds of rules: cyclic rules20 and rules of word-level phonology.21 Rather than an expression of a theoretical premise, this contrast is acknowledged because it is necessary for the analysis of English that SPE proposes. For example, unstressed vowels are reduced to schwa by a context-free rule – rule (103) in the quotation below. Chomsky & Halle (1968) explain why the Vowel Reduction Rule cannot apply to individual morphemes or clusters thereof that 20 21

Which at that time were called cyclical, just as syntactic was syntactical. This distinction actually shapes the architecture of SPE: chapter three is called "The transformational cycle in English phonology", while chapter four is entitled "Word-level phonology".

88

Chap 5: SPE sets the standards for 40 years

do not make a word: it must apply at the word level, i.e. after all morphemes have been concatenated. (46)

"We see at once, incidentally, that the Vowel Reduction Rule cannot itself be cyclical. Once a vowel has been subject to rule (103) [the Vowel Reduction Rule], its original underlying form is unrecoverable. Therefore, if this rule were to apply at any point in the first cycle, for example, certain vowels will be reduced to [ə] even though in some later cycle they may receive primary stress. Evidently, rule (103) must apply only after the process of stress assignment within the word is complete. Within our framework, this means that the rule of Vowel Reduction is restricted to the level of word boundaries." Chomsky & Halle (1968:113)

Hence vowel reduction must not apply to the inner cycle of [[parent] al] since this would reduce the /e/ to schwa (main stress falls on the /a/). This /e/, however, will end up being stressed on the outer cycle (parént-al), and thus must not be reduced (stressed vowels are never reduced). The insight behind the prohibition of vowel reduction to apply to chunks that are smaller than the word is simply that vowel reduction is a function of word stress, and hence must "wait" until the last morpheme is considered for the assignment of word stress. Word level rules thus do not apply to chunks that are smaller than a word – but they do not apply to chunks that are bigger than a word either. (47)

"The Vowel Reduction Rule (103), cannot be permitted to apply cyclically beyond the level of word boundary. [… It] must apply only once in the course of a derivation; within our framework this means that [it] must be [a] noncyclical rule restricted to the level of word boundary." Chomsky & Halle (1968:133f)

This is because stress is further manipulated above the word level, i.e. in intonation (sentence/phrasal stress), but this never leads to vowel reduction. The decision which vowels are reduced is taken at the word level, no matter how much stressed vowels may be weakened by intonation later on. Word-level rules thus apply exactly once in a derivation – when words are completed: they do not affect either smaller or bigger chunks. The definition of word-level rules is thus evidently procedural: they apply to a certain chunk in the course of cyclic derivation. Curiously enough, though, their implementation is representational in SPE: given the general mapping mechanism that translates syntactic surface structure into clusters

The phonological cycle and one single phonology

89

of hashmarks (§90), the word is defined as a string that is preceded and followed by ##. (48)

"The surface structure specifies that each word constitutes a stage of the transformational cycle. By a word, we mean an element of the form ##...##, where … contains no occurrence of ##. […] A rule restricted in application to contexts meeting this condition is what we call a rule of word phonology. Evidently, such rules will not reapply at successive stages of the transformational cycle, even if interspersed freely among the cyclic transformational rules." Chomsky & Halle (1968:163)

Chomsky & Halle are certainly right in pointing out that the restriction of word level rules to this context produces the desired effect. The question, however, is whether a representational disguise of a procedural event is warranted. 106

3.2.2. SPE is representational: class 1 vs. class 2, cyclic vs. word-level rules Just as for word-level rules, SPE proposes a representational identification of the contrast between class 1 and class 2 affixes. Recall from §92 that the former come with a + boundary, while the latter are identified by a # boundary. The heart of Lexical Phonology will be to replace this representational management of affix classes by a procedural solution where class 1 affixes are concatenated (and interpreted) before class 2 affixes. This supposes the existence of distinct mini-phonologies (one set of rule for the assessment of class 1 strings, another set of rules for the interpretation of class 2 strings, see §148). In the 80s when the scene will be dominated by Lexical Phonology, Halle & Vergnaud (1987a) undertake the restoration of the SPE-setup. However, they follow the procedural trend of the 80s: just like the distinction between affix classes, the implementation of word-level rules will be procedural (in terms of a distinct block of so-called non-cyclic rules, see §233). Finally, it should be borne in mind that SPE's word-level rules do not recover the same notion as what will be known as postlexical rules in Lexical Phonology; also, the SPE-supported contrast between cyclic and wordlevel rules does not correspond to either the distinction between level 1 and level 2 rules, or to the opposition between lexical and postlexical rules. Rather, it reappears in the 80s under the label of post-cyclic lexical rules

90

Chap 5: SPE sets the standards for 40 years

(Lexical Phonology, see §194) or the aforementioned non-cyclic rules (Halle & Vergnaud, see §§233,236).

107

3.2.3. SPE's representationalism maintains the unity of one single computational system It is argued below that the major conceptual division among interface theories on the procedural side is between those that work with multiple miniphonologies and those where all strings are assessed by the same computational system (see §828). A number of other properties (such as the existence of a no look-back device, today called Phase Impenetrability) depend on this initial choice. It is therefore important to be explicit on the situation in SPE: Chomsky & Halle use one single set of phonological rules that assesses all strings, no matter what their size. The last quote in §105 shows that word level rules are not set apart from cyclic rules in either formulation or location: both types of rules are interspersed in the unique sequence of ordered rules that is active in SPE. That is, the unity of the computational system is achieved by a representational (rather than a procedural) identification of word-level rules (a string is a word iff it is enclosed in ##...##). This definition will be abandoned by Halle & Vergnaud's (1987a) incarnation of the architecture of SPE, where cyclic and word-level rules are separate computational systems (§233).

108

4. Conclusion About all major issues of interface theory are discussed in SPE: Chomsky & Halle define the benchmarks for further development which, as we will see, make the research agenda up to the present day. SPE installs representational and procedural communication with morpho-syntax, organises translation, raises the question of readjustment and direct reference to untranslated morpho-syntactic categories, introduces affix class-based phenomena and analyses them according to a strictly representational line of attack, argues for a specific word-level phonology, but couches its existence in the unique computational system that the theory provides for.

Conclusion

91

Finally, SPE also contains the seeds of modularity, which however remains unspoken and undefined as an architectural principle of grammar: the conception of the computational system as an ignorant and autistic robot, the inverted T model, readjustment and the organisation of translation are the fruit of modular thinking. However, modularity is also violated: brackets and the checkless reference to their labels are non-modular elements of the theory, which did not raise any objection. We will see how the modular idea wins more conscious recognition in generative theory in the 80s as we go along (§406).

109

Chapter 6 The life of boundaries in post-SPE times

110

1. Introduction: overview until the 80s Let us now look at the use that was made of linear boundary symbols in SPE itself as well as in the subsequent development of generative phonology. The leading idea of the following pages is to document the growing dissatisfaction and frustration as boundaries were put to use during the 70s, and to identify the reasons for this development. Foremost among these is the fact that, just like on the segmental side, SPE does not impose any restrictions on the occurrence of boundaries (except that they must coincide with a morpho-syntactic division). Parallel to the development on the segmental side, this gave rise to boundary abuse and a whole boundary zoo (§116). Reactions on this situation were numerous, both from inside and outside the SPE paradigm; they were also embedded in the general critique of SPE that ran through the 70s under the header of the abstractness debate and gave rise namely to Natural Generative Phonology, the external challenger of SPE (§§119,124). This movement culminates in an article by Pyle (1972), who properly demonstrates that boundaries are unsuited for carrying morphosyntactic information: they are supposed to be segments but do not behave as such. That is, their diacritic identity is merely hidden behind the [-segment] delusion. "Culminating" in this context of course means logically culminating, since Pyle's demonstration of the bankruptcy of boundaries was available as early as 1972 – but had no significant echo. Another function of this chapter is to provide an outlook on the leading issues of the 80s. In fact, the scene that will be the background of Lexical Phonology and Prosodic Phonology is set in the 70s: at least as far as the representational side of the interface is concerned, all relevant questions are already asked, and all possible answers given. The 80s, then, just select proposals according to new theoretical premises (GB, modularity), and work out what was seeded. One case in point is the question whether morpho-syntactic intervention should be local (boundaries) or domain-based (the domains of the Prosodic Hierarchy). This issue was introduced by McCawley (1968) (§113); it

94

Chap 6: The life of boundaries in post-SPE times

will be critical when Elisabeth Selkirk sets up Prosodic Phonology (§373) in the early 80s (although very quickly nobody will remember this debate anymore). Local vs. non-local intervention will come back to the focus of interest in §706 below. Another case is what will be known as direct syntax approaches in the 80s, i.e. the main competitor of Prosodic Phonology. This perspective dispenses with the translation of morpho-syntactic structure into phonological objects altogether. Instead, phonological rules are favoured that make direct reference to morpho-syntactic material. Direct syntax approaches were erratically voiced on several occasions in the 70s without reference to each other (§§123,132,137); but in all cases that I have come across, they were the result of the frustration that boundaries generated: if boundaries cannot do the job, they have to go, and so has thus translation. Translation itself is poorly understood in the 70s, because it is poorly studied, both theoretically and empirically. Its place in the overall interface architecture is vague (§131), and we encounter first frustrated comments on the mapping puzzle (§111). These two issues will be central targets of Prosodic Phonology, which affords substantial progress regarding the former (§377), and at least the most serious empirical effort to date regarding the latter (§§387,392f,396). Summaries concerning translation and the mapping puzzle are available in §687 and §753, respectively.

111

2. The mapping puzzle: alas, there are no natural classes of boundaries Since boundaries root in morpho-syntactic divisions, linguists have tried to find out which particular morpho-syntactic (or semantic) configuration produces a phonological effect. And in turn, which phonological effects are induced by boundaries. That is, some pattern should emerge as empirical coverage and relevant studies grow: either for a given language, but even more plausibly on a cross-linguistic count. It can hardly be imagined that the correlation between morpho-syntactic causes and phonological effects is arbitrary. This is what I call the mapping puzzle, following the terminology of Prosodic Phonology where the process of translation from morpho-syntax to phonology is called mapping (see §380). In the post-SPE environment where boundaries are the interface currency, trying to come to grips with the mapping puzzle amounts to looking for natural classes of boundaries. That is, for a set of boundaries that share phonological behaviour crosslinguistically and also have a stable morpho-syntactic identity.

Boundary mutation rules

95

I have only come across relatively few instances in the literature of that period where this issue is discussed: Kenstowicz & Kisseberth (1977:103ff), Basbøll (1975:121ff, 1978a:165f) as well as Devine & Stephens (1976, 1980). True, the empirical basis was anecdotic and vanishingly small at that time, but there is hardly any less frustration today after an important descriptive effort during the 80s and early 90s (due to Prosodic Phonology, §463) than there was in post-SPE times. This is to say that the mapping puzzle has timidly emerged when interface was done with boundaries. It appeared for the first time in the discussion of Trubetzkoy's Grenzsignale (§57), and will run through the rest of the historical survey (§§387,392f,396,463). Unfortunately, though, at the end of the book mapping will remain as mysterious as it was in the 70s (see the summary in §753).

112

3. Boundary mutation rules So-called boundary mutation rules (or readjustment rules), contextsensitive or not, can modify the boundary structure inherited from higher levels. This possibility was initiated by Chomsky & Halle (1968:366ff), and further developed for example by Selkirk (1972, 1974) and Sag (1974). Boundary mutation rules can achieve transformations such as # → + or ## → # and so forth (including + → #, cf. Sag 1974:603f). Selkirk (1972, 1974) makes extensive use of boundary mutation rules for the description of French liaison, which is largely sensitive to morphological and syntactic information (e.g. ils *([z]) ont "they have" where liaison is mandatory, against ont-ils *[z] eu "have they had?" where liaison is impossible). That is, Selkirk first simplifies the maximum boundary cluster ## that can come down from syntax in liaison environments (## → #); then she applies a rule that triggers liaison with a single intervening #, but which is blocked by double ##. The effect is that the environment for liaison is defined in the phonology on the grounds of the boundary strength that phonology inherits from higher levels (Elordieta 2008:212ff provides a comprehensive overview of Selkirk's early work). Boundary mutation rules were also given a socio-linguistic role according to the general observation that "the more casual or 'reduced' the style level becomes, the more grammatical boundaries lose their effect" (Basbøll 1975:114). This perspective tacitly assumes that the "regular" effect of boundaries, lost in low style, is rule-blocking, rather than the enhancement of rule application. Following this line, Selkirk (1972) depletes

96

Chap 6: The life of boundaries in post-SPE times

the linear string of more and more boundaries, or transforms the stronger # into less harmful +, = as the style level decreases.

113

4. McCawley (1968): boundaries define domains of rule application

114

4.1. Cyclic derivation on the grounds of phonological domains Chomsky et al. (1956) have introduced the idea that boundaries define domains which correspond to relevant morpho-syntactic chunks of the linear string (the "phonemic clause") and serve as a measure for the application of rules in cyclic derivation (see §75). This concept was further developed by McCawley (1968:52ff) who proposes to order boundaries along a strength hierarchy, something that was also done by Chomsky et al. (1956), but is rejected in Chomsky & Halle (1968:371). In doing so, boundaries acquire a "ranking function": each division has a hierarchical status and thereby ranks the rules in whose structural description it occurs. Boundaries, then, define the domain of application of rules in that "the juncture gives the limits of the stretches of utterance to which certain rules apply" (McCawley 1968:55). Basbøll (1975:111, 1978b) and Selkirk (1980a) are along the same lines. Chomsky & Halle (1968:371) recognise the appeal of McCawley's idea, but argue against the exclusive delimiting function of boundaries. McCawley's mechanism indeed reminds of Trubetzkoyan Grenzsignale (§55), albeit in a non-functional perspective. Devine & Stephens (1976:292f) put it this way: "while some morphosyntactically conditioned phonological rules do signal beginnings and ends, others, just like morphologised rules proper, evidently signal the morpho-syntactic category and not its delimitation." In the example provided by McCawley, a rule that voices intervocalic /t/ will not apply to the first /t/ of the sequence /#pat#atak#/, but does affect the second one. This is because the rule applies separately to the two chunks #pat# and #atak#. Since /t/ is intervocalic in the latter, but not in the former domain, it will be transformed only in #atak#, and the overall result is [patadak]. This is obviously an application of cyclic derivation (§80) where also some no look-back device (§287) is needed: if there is an outer cycle, i.e. [[pat][atak]], something must prevent the voicing rule from applying to the second /a/ of the overall string at the outer cycle.

McCawley (1968): boundaries define domains of rule application

97

If there is cyclic derivation based on regular cyclic structure (i.e. brackets), the question arises whether McCawley's boundary-based chunks are any different from those that are delineated by cyclic structure. If they are not, boundary-defined domains of rule application appear to be redundant. Another way of looking at this situation is to recognise that there is a competition between a representational (boundaries) and a procedural (cyclic derivation) solution. As far as I can see, this issue was not discussed in the contemporary literature that debates McCawley's proposal. 115

4.2. Local vs. domain-based intervention – notational variants? McCawley's (1968) idea remained anecdotic in the SPE-dominated scene by the time it was aired. However, the further development of interface theory shows that it has an important potential: it introduces the central question whether morpho-syntactic intervention in phonology is local or domain-based. In the former case, a punctual object (# in SPE) is inserted into the linear string at a specific point and hence can only bear on its immediate environment. By contrast, in McCawley's domain-based perspective no effect is caused by the presence of one single object in the linear string. Rather, only the association of two identical items (the stretch delineated by #...#) can bear on the phonology. And even then is the effect not due to the neighbourhood action of a boundary; rather, the domain as such impacts the phonology. McCawley's mechanism is thus a linear notation of what will later be known as the Prosodic Hierarchy: by the time it was put forth, the linear environment did not allow for a formulation in terms of autosegmental domains. The heart of the proposal, though, is precisely this: morphosyntax bears on phonology through the action of domains, rather than of singleton items that are locally inserted into the linear string. When Elisabeth Selkirk set up Prosodic Phonology in the early 80s, she was embarrassed by McCawley's domain-defining boundaries (Selkirk 1980a): her purpose is to show that domains are a better vector of morphosyntactic information than boundaries. If both turn out to be notational variants, the major break in interface tradition that she advertises would look less fresh (see §373). The real issue, though, which is truly critical and which we come across for the first time here, is the alternative between local and non-local intervention in phonology. The former is represented by Trubetzkoy's

98

Chap 6: The life of boundaries in post-SPE times

Grenzsignale, structuralist juncture and SPE-type boundaries, while the latter is advocated by McCawley (1968) and Prosodic Phonology, the to date dominant interface theory. It is argued in §706 that the influence of one module on another through the representational channel can only be local.

116

5. Typology and strength of boundaries

117

5.1. Boundary zoo Linguists have always recognised different types of juncture. Some aspects of the structuralist typology were reported in §71. Chomsky et al. (1956) operate with "internal" and "external" juncture, while [±word boundary] and [±formative boundary] are the basic items in SPE (§87). In post-SPE times, this limited inventory becomes a true boundary zoo. It is not uncommon to find treatments where up to seven different boundaries are identified in the grammar of a given language. These are then usually ranked on a scale of boundary strength.22 Thus Basbøll (1975) acknowledges five different boundaries: ###, ##, #, + and $ (see also Basbøll 1981). For the description of Japanese, McCawley (1968:57ff) recognizes $, #, #i, :, & and *, which appear in order of decreasing strength. Stanley (1969) identifies seven different boundaries in Navaho, #, =, *, !, ", + and - (also in decreasing strength). Kaisse (1985:109ff) presents data from Modern Greek that need more than three boundaries in order to be readily described.23

22

23

More recent applications of the notion of boundary strength can be found in Loporcaro (1999) and Bertinetto (1999). These authors assign increasing numerals to boundaries that represent increasing morpho-syntactic distance (on a scale from -6 to +6, hence yielding a zoo of 13 different diacritics). Rotenberg (1978:20f) also elaborates on boundary strength. But rather than adding new boundaries to the inventory of SPE, which would have been in line with the general movement, Kaisse concludes on the failure of boundary theory. Instead, she argues for direct reference to morpho-syntactic categories. This so-called direct syntax approach is further discussed in §407.

Typology and strength of boundaries 118

99

5.2. Boundary contrast: minimal pairs and phonological permeability On which grounds, then, are different boundaries distinguished? The best evidence are "minimal pairs" that contrast two different boundaries in an otherwise identical segmental environment. For example, there are two homophone suffixes -er in English: the comparative -er (long - long-er, quick - quick-er), and the agentive -er (sing - sing-er). The latter does, the former does not trigger a rule which is presented as /g/-deletion in Chomsky & Halle (1968:85f, 369f): /long-er/ → lo[Ng]er vs. /sing-er/ → si[N]er. If the analyst is lucky enough, a given root supports both affixes, as is the case in English where (at least some) speakers can derive the agentive longer "a person who is longing" from the verb to long. Hence the minimal pair longer lo[Ng]er "comparative of long" vs. longer lo[N]er "a person who is longing". The contrast must thus be ascribed to the existence of two different boundaries, i.e. agentive /sing#er/, /long#er/ vs. comparative /long+er/. On the account of Chomsky & Halle (1968:369f), a boundary mutation rule has transformed the original comparative /long#er/ into /long+er/. The rule that deletes /g/, then, contains # in its structural description and is therefore inoffensive in regard of /long+er/. Of course, it also leaves morphologically simplex forms such as finger /finger/ untouched. Note that the /g/ of /sing##/, where ## represents the word boundary, will also be deleted.24 Hence one may expect that seven different boundaries rely on a seven-way contrast. But of course this is not the case. Stanley (1973:192f) admits that "phonologically parallel contexts can seldom be found to distinguish more than two or three types", and further says "'minimal pairs' for boundary phenomena are difficult to find, in general, and less direct methods of justifying boundary types must be used." The less direct method in question is the various degree of phonological permeability of morpho-syntactic divisions: Stanley (1973:193) explains that in practice "it will soon be found that some [boundaries] are actually weaker than others in the sense that certain phonological rules will apply across some and not others." The same method is applied by McCawley (1968:55), who assumes "the existence of distinct junctural elements /#/ and /:/, corresponding to the fact that such rules as the p → h rule and the h → / rule apply to different-sized chunks of utterance." Establishing different boundaries on the grounds of their phonological permeability is indeed the common method that allows for arriving at 24

More on English agma when Lexical Phonology is presented in §167.

100 Chap 6: The life of boundaries in post-SPE times the aforementioned boundary zoo. Examples are Chomsky & Halle (1968:84ff), Siegel (1974), Aronoff (1976), Kenstowicz & Kisseberth (1977:103ff), Booij (1977), Allen (1980). Devine & Stephens (1976:298f) offer a general overview. To the best of my knowledge, nobody has ever tried to establish a cross-linguistic inventory of boundaries that is based on their sensitivity to certain rule types or some other cross-linguistically resident property. What has been done, though, is to define families according to the formal properties of boundaries. Stanley (1973) for example distinguishes three classes of boundaries on the grounds of three different rule classes: 1) rules that are blocked by boundaries of strength x or stronger, 2) rules that apply only when a specific boundary is present, 3) rules that apply only when a boundary of strength x or stronger is mentioned at the beginning or the end of its structural description.

119

6. Attempts to restrict boundary abuse and the boundary zoo

120

6.1. Boundaries in an arbitrary rewrite-machinery In the 70s, the major objection levelled against SPE was its unrestrictedness: SPE can describe those processes that occur in natural language as much as those that do not. Indeed, nothing in the rewrite-machinery or in other parts of the theory marshals overgeneration. In chapter nine of SPE, Chomsky & Halle (1968) identify this problem. In their view, markedness is a prime candidate for a solution. During the 70s, then, things did not quite go their way: two strands tried to derive the impossibility of outlandish processes. One from formal properties of the theory (the movement initiated by Kiparsky 1968-73 and which led to Lexical Phonology), the other by cutting down the formal machinery to a minimum, and by retrenching the scope of phonology (Natural Generative Phonology, see §124 below). This evolution is known as the abstractness debate (e.g. Kenstowicz & Kisseberth 1977:1-62, 1979:204ff, also §§125f and Vol.1:§305). Part of the critique was also that SPE-technicians, not unlike their structuralist peers, were only after the "correct" surface forms and did not care much for the way they are produced. This challenge also extends to how boundaries were used: analysts would appeal to them if they helped achieving the correct result, and otherwise dismiss them. True, everybody was bound by the morpho-syntactic corset of boundaries, which was put

Attempts to restrict boundary abuse and the boundary zoo 101

into force by Chomsky et al. (1956) (i.e., no boundaries in the middle of morphemes, see §78). But beyond that, no principle defined when the analyst could, should or must not call on a boundary. The ensuing practice is questioned by Devine & Stephens (1976). (49)

Boundaries were "considered justified by their proponents merely because they produced the correct output and were more or less consistent with background assumptions; whereas in fact a hypothesis should have empirical consequences different or in addition to its explanandum." Devine & Stephens (1976:309f)

For this reason and also because of the awkward boundary zoo (§117), some authors tried to define criteria which are able to restrict the distribution of boundaries, and to control their use. These efforts are reviewed below. 121

6.2. Boundary strength as a diagnostic for boundary abuse Kenstowicz & Kisseberth (1977) focus on the relative strength of boundaries, which is established according to their phonological permeability in regard of a given process. This ranking, then, must be consistent with respect to all other rules: boundary X may not be weaker than boundary Y in the conditioning of a one process, but stronger in regard of some other. (50)

"Boundaries must be […] ordered from 'weakest' (i.e., expressing the closest bond between morphemes and thus having the least inhibiting effect on phonological rules) to 'strongest' (i.e., expressing the weakest bond between morphemes and thus having the greatest inhibiting effect on phonological rules)." Kenstowicz & Kisseberth (1977:105ff)

Also, "the relative strength of boundaries must conform to grammatical facts" (page 105). That is, the boundary hierarchy established on phonological grounds should match, or at least not fall foul, of the more or less close relationship of the morphemes in question. Boundary strength can thus be used as a diagnostic for misplaced boundaries or boundary abuse.

102 Chap 6: The life of boundaries in post-SPE times 122

6.3. Boundary economy I: Basbøll's general and grounded advice Basbøll (1975, 1978a,b, 1981) attempts to make boundaries a last resort strategy. He argues that boundaries should not be called on in presence of an alternative solution that does the same job without them. An example of this strategy is Basbøll's (1975:115, 1978a:154) critique of SPE and Selkirk (1972, 1974), who recur to the boundary = only in learned (i.e. Latinate) vocabulary (see §89). That is, Selkirk for example justifies the existence of =, which must be different from +, by the fact that the former, but not the latter triggers the loss of the prefixal nasal consonant (and ensuing compensatory lengthening of the following liquid) in French: /in=légal/ and /con=mémoratif/ come out as illégal [ilegal] and commémoratif [kçmemoratif] (where the gemination of the liquid is only possible in high style), respectively. Basbøll argues that the diacritic [+learned] could do the same job as =, and hence should be preferred. Basbøll also makes a number of general remarks that call on external evidence in order to identify boundaries in the linear string. That is, psycho-linguistic experimentation (Basbøll 1975:122ff) and a putative phonetic identity (Basbøll 1975:123, see §70 on the structuralist debate regarding this issue) could betray the (non-)existence of boundaries. Finally, Basbøll (1975:128) conditions the occurrence of boundaries to productivity: he claims that productive, but not unproductive affixes, come with a boundary (# in his system). On more phonological grounds, Basbøll (1978a:164f) also argues that boundaries should not be used if their presence leads to a more complicated formulation of other rules. And he proposes to require multiple evidence for boundaries, which should only be introduced if needed for at least two phonological processes (Basbøll 1975:121, 1978a:165). Finally, Basbøll (1975:122) and Devine & Stephens (1976:301) recall a property of boundaries that should be self-evident: # boundaries represent morpho-syntactic, not lexical information; hence they may not be stored in the lexical entry of roots. By contrast, they may come with affixes since these precisely carry morpho-syntactic information (regarding class membership for example). Basbøll's ban on boundaries in roots is a reaction on an article by Hoard (1973), who proposes lexical entries of roots such as /dhadh#/ for Sanskrit.

Attempts to restrict boundary abuse and the boundary zoo 103 123

6.4. Boundary economy II: the + boundary can be dispensed with Boundary economy is also the concern of Hyman & Kim (1973), Hyman (1975:197f, 1978:457ff) and Strauss (1979). These authors argue that the formative boundary + of SPE can and should be done away with. Hyman (1978:459) reacts on "the abuses seen in such works as Stanley (1973), where boundaries are unnecessarily proliferated" (also Hyman 1975:197f and Hyman & Kim 1973). He claims that whatever the morpho-syntactic reality of +, it can never have a phonological effect. This is because all cases where + is held to be phonologically active that "we have investigated have either used the + boundary when the # boundary would have done as well, or have used the + boundary diacritically, and could just as well have used ad hoc boundaries such as $, % or ¢ or referred directly to the morphemes involved" (Hyman 1978:459).25 In replacement of +, Hyman (1978:459) proposes to write rules that make direct reference to the particular morphemes involved. This option did not arouse any contemporary echo; but together with Pyle (1972) and Rotenberg (1978:16f) (see §136, which Hyman leaves unmentioned), it constitutes the body of early and erratic formulations of what will be known as direct syntax approaches in the 80s. §407 below reports on the contention between this line of thought and Prosodic Phonology: the cornerstone of the latter is Indirect Reference, that is the proscription of making reference to morpho-syntactic categories in phonological rules. Strauss (1979) also argues that all English facts can be accounted for if the SPE-created # and + are merged into one single concatenation operator.26 On his analysis, the distinctive function of # and + is taken over by the computational component (precyclic vs. postcyclic affixes) and the lexicon.

25 26

Stevens (1980) discusses this issue and presents eventual counter-evidence. Following Siegel (1974:25), Strauss (1979:387) holds that = has no independent existence anyway: SPE was misled not to merge it with + (see also Siegel 1980).

104 Chap 6: The life of boundaries in post-SPE times 124

7. Internal and external SPE-revision: Kiparsky (1968-73) and Natural Generative Phonology

125

7.1. Introduction The previous pages have reviewed initiatives against uncontrolled proliferation and use of boundaries: the relevant literature made a case for a selfapplied boundary hygiene. The proposals at hand, however, had little, if any impact on the field. Part of the reason for this is certainly the fact that people simply lost interest in manicuring boundaries within a system that everybody knew needed a profound revision in central areas: overgeneration, abstractness and morpho-phonology were generative teething troubles that Chomsky & Halle (1968) had already lucidly spotted in the ninth chapter of SPE. The research agenda of the 70s, then, was dominated by the competition of an internal and an external revision movement. The former (socialist) was led by Kiparsky (1968-73) and attempted to reform SPE from inside, while the latter (communist) was initiated by Stampe (1972) and argued that there is no possible cure without completely overthrowing the system (see Scheer forth). The purpose of the following pages is not to review the history of generative phonology in the 70. Rather, I aim to show what kind of consequences the critique of SPE had on interface theory, i.e. on boundaries at that time. As we will see, the bearing is on the quantity, rather than on the quality of boundaries.

126

7.2. Morpho-phonology in SPE and the 70s Morpho-phonology is the idea, tacitly admitted in all quarters of early generative phonology, that etymologically, paradigmatically or semantically related items contract a synchronic derivational relationship (Jonathan Kaye calls this attitude the Central Dogma). Otherwise, it was argued, a generalisation would be missed. Therefore, any two such items had a common underlying form, and their surface appearance was derived by rule. For the sake of illustration, it is useful to look at the wildest case on record: the work by Theodore Lightner. Lightner is certainly not representative for the average post-SPE phonologist; but the fact that theory allowed him to do what he did without raising any objection is an unmistakable indication that something was wrong.

Internal and external SPE-revision: Kiparsky (1968-73) and NGP 105

Lightner (1978:18f, 1981, 1985) holds that each one of the following pairs are derived from a common underlying form: 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. Since the alternations h-k (heart - cardiac), d-T (third - fourth) and s-h (sweet - hedonism) suppose Grimm's Law, Verner's Law and the Ancient Greek s > h shift, Lightner concludes that all these processes are performed by the grammar of present-day English natives. The origin, development and (non-)justification of morphophonology is discussed in a large body of literature (St. Anderson 1985:331f provides an informed overview). As a piece of the discussion regarding overgeneration and abstractness (see §144), it has played an important role in the evolution of generative phonology. Overgeneration is at the origin of the internal and external critique of early generative phonology that has made the phonological agenda of the 70s. The internal (i.e. non-revolutionary) revision was led by Paul Kiparsky and the abstractness debate that he initiated in 1968 (Kiparsky 1968-73: how abstract is phonology?). This strand will open out in Lexical Phonology in the 80s. The external (i.e. system-overthrowing) critique was initiated by David Stampe (1972) and Natural Phonology, on which more shortly.27 The broad result of this evolution was that all phonological theories that individuated in the early and mid 80s have to some extent learned the following lesson: many alternations that early generativists believed were produced by online computation do not represent any synchronically active process at all. Two etymologically, paradigmatically or semantically related forms do not necessarily stand in a derivational relationship: they may as well be recorded as two independent lexical items, or represent allomorphic activity. Hence sweet and hedonism, but also, perhaps, electric and electricity or sane and sanity, may represent two distinct lexical entries that have not been modified by any rule before they reach the surface.28 27 28

Vol.1:§305 offers a more detailed review of the development of the abstractness debate and overgeneration (also Scheer 2004b:8ff, forth). Velar softening (electric - electricity) and Trisyllabic Shortening (or Laxening) (sane - sanity, see §164) are two debated cases which up to the present day produce both defenders and adversaries of the single-underlying-form analysis. Relevant literature includes Chomsky & Halle (1968:219ff, 426f), Hooper (1975:544f), Kiparsky (1982a:40f), Halle & Mohanan (1985), Harris (1994:21ff), Kaye (1995:312, 328), Coleman (1995:375ff), Halle (2005) and McMahon (2007). Hayes (1995a) and Green (2007:172ff) provide a documented overview of the question.

106 Chap 6: The life of boundaries in post-SPE times It goes without saying that the question where exactly the red line runs between the computational and the lexical option is open: some cases are lexical for sure, and the online computation of others is beyond doubt. But the swampy midfield is large enough for much debate: relevant discussion today runs under the header of (anti-)lexicalism; after a decidedly lexicalist period in the 80s (both in syntax and phonology), anti-lexicalist analyses in the spirit of the 60s have gained ground again in certain minimalist quarters. This issue is further discussed in §569. It may be doubted that a formal criterion will be able one day to decide whether a given alternation represents a common underlying form or two separate lexical entries (at least in phonology). Carvalho (2002:134ff) provides extensive discussion of this question. 127

7.3. Natural revival of structuralist juncture (abuse) Natural Generative Phonology (Vennemann 1974, Hooper 1976) is born from a reaction against overgeneration, abstractness and morphophonology.29 The central tenets of this theory distil into the True Generalization Condition and the No-Ordering Condition. According to the former, only phonetically accessible information can be used in the formulation of phonological rules, while the latter prohibits rule ordering. Alternations are divided into two types. Those that do not suffer any exception in the entire language and exclusively appeal to phonetically retrievable information are called natural and granted phonological status (P-rules). By contrast, the statement of morpho-phonemic rules (MP-rules) requires non-phonetic information such as reference to morpho-syntactic

29

Laks (2006) provides a historical overview of Natural Generative Phonology; see also Scheer (2004b:12ff) for a critical review of its phonetically oriented evolution. Note that there are two brands of Natural Phonology, one generative, the other not. Natural Phonology grew out of David Stampe's (1972) dissertation and is non-generative: the major fraction line is between the formal and the functional approach to language (in the sense of Newmeyer 1998): Natural (nongenerative) Phonology is functionalist at heart. Natural Generative Phonology also calls on Stampe's dissertation, but follows the formal line of thought. Nongenerative Natural Phonology is represented by, among others, Donegan (1978), Donegan & Stampe (1978, 1979), Dressler (1974, 1984), Hurch & Rhodes (eds.) (1996) and Dziubalska-Kołaczyk (2001a, 2002).

Internal and external SPE-revision: Kiparsky (1968-73) and NGP 107

categories, and the alternations at hand are typically riddled with exceptions. This division enforces a corresponding split among boundaries: since #, + and the like can hardly be said to represent phonetic information (they do not have any phonetic correlate, see §88), they are banned from P-rules (Vennemann 1974:360ff). However, even natural (i.e. completely regular) rules need to make reference to boundaries such as the beginning of the word. This is why Hooper (1975:545) introduces "phonetic boundaries": the "syllable boundary $" and the "pause boundary ║". Elsewhere, Hooper (1976:14) also calls them "phonological boundaries" or "grammatical boundaries". The following quote shows how Hooper tries to tell boundaries that do from boundaries that do not have a phonetic manifestation. (51)

"'Phonetic terms' refer to phonological features (that have intrinsic phonetic content) and phonological boundaries (that have a necessary and consistent phonetic manifestation). The phonological boundaries are the syllable boundary (S) and the pause boundary. Both of these boundaries are determined by phonetic means. On the other hand, the word boundary (## and #) and the morphemic boundary (+) are determined by syntactic and semantic means. These latter boundaries are counted as nonphonetic information." Hooper (1976:14)

The same dilemma is referred to with different vocabulary by Anderson (1974:3), who talks about "boundary elements" in morphophonemic rules, against "phonetically realizable (i.e., word or phrase) boundaries" that may occur in natural rules.30 No doubt this perspective is much more akin to the structuralist conception of juncture than to generative boundaries. Anderson (1974) and Hooper (1975, 1976) are forced to grant a phonetic identity to their boundaries because otherwise they would not be entitled to use them. Concomitantly, their boundaries are explicitly denied any morpho-syntactic identity – precisely the property that paved the way for structuralist boundary abuse (§69): nothing prevents this kind of boundaries from occurring in the middle of morphemes. 30

Rather than trying to make (some) boundaries phonetic objects in order to have the right to use them, Rhodes (1974) attempts at doing away with them altogether (through global and transderivational reference: a process that does not actually apply in a derivation could be "felt" and hence cause an effect in that derivation because it is active in some other derivation where the same morphemes are involved).

108 Chap 6: The life of boundaries in post-SPE times Another obvious point of contact with structuralism is Level Independence (§61): In Natural Generative Phonology, just as two decades ago, an alternation that requires the appeal to morpho-syntactic information cannot be phonological; it is rejected into morphology (or morphophonology, MP-rules).

128

7.4. The elimination of (word) boundaries from P-rules – a case of wishful thinking Devine & Stephens (1980:57ff) and Clayton (1981) point out that boundaries are an excellent testing ground for the central claim of Natural Generative Phonology, i.e. that alternations fall into two (major) classes of processes, P-rules and MP-rules. The former are productive, phonetically transparent and entirely regular, while the latter are typically non-productive, admit exceptions, may be non-transparent and governed by morphosyntactic conditions. The corresponding division of boundaries into those that are phonetically expressed ($, ║) and those that are not (#, ##) was described above. A prediction is thus made to the effect that non-phonetic boundaries such as the word boundary will never be necessary for the statement of Prules. Devine & Stephens (1980:71ff) and Clayton (1981:577ff) demonstrate that this is not true: many alternations in natural language are entirely regular, productive and make exclusive reference to the phonetic signal except for the fact that they crucially appeal to the word boundary. Devine & Stephens (1980:59f) also show that the elimination of word boundaries from phonological rules is wishful thinking, if anything, in Natural Generative Phonology practice. A processes as ordinary and frequent as apocope cannot be phonological because it needs to make reference to the right edge of the word. In spite of this, Hooper (1976:106) desperately tries to describe a case of apocope in Spanish without mentioning the word-final location. She attempts at obviating this context by setting up a structural description that appeals only to the preceding environment. Hooper contends, then, that the vowel in question is not lost in word-internal position because it is never preceded by the relevant triggering environment. Devine & Stephens show that this analysis does not stand up to fact: an apocope is called an apocope because it concerns the last vowel of the word. This case is instructive because it shows that even Natural Generative Phonology practitioners refrained from eliminating entirely productive,

Internal and external SPE-revision: Kiparsky (1968-73) and NGP 109

regular and phonetically transparent processes from phonology on the sole grounds of their making reference to word boundaries. The result is a theory that warrants the absence of # from phonology, but ends up reintroducing it through the back door (something that Devine & Stephens 1980:59f call covert boundaries).

129

7.5. The retrenchment of morpho-phonology mechanically reduces the number of boundaries Rather than eliminating morphologically conditioned processes from phonology altogether, the reformist strand, which is represented by Kiparsky's work, set out to simply retrench morpho-phonology. To which extent was a matter of debate. The general goal, i.e. the reduction of abstractness, could be achieved either by restricting the computational facilities, or by having a restrictive definition of the relationship between underlying and surface forms. The latter strategy is relevant for boundaries since it shifts burden from the computational part of grammar into the lexicon. Trivial examples were already mentioned in §126 above: instead of deriving dental from tooth by a rule that also computes a boundary in dent-al, the two items are independent lexical entries, and no phonological computation is performed when they are used. Kiparsky has attempted to formalise this orientation. The Alternation Condition (Kiparsky 1968-73, 1973a) (see §185) defines what a possible underlying representation is: in case a morpheme shows no alternation on the surface, it must not be any different in its underlying form. This results in a ban against absolute neutralisation. In a further step, Kiparsky (1973a) 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 is produced either by the concatenation of two morphemes or by the application of a phonological rule. This proviso rules out a number of abstract SPE analyses such as for instance the one regarding Trisyllabic Shortening (div[aj]ne - div[I]nity, op[ej]que - op[æ]city etc., see §164 and note 28 for further discussion and literature). In order to account for the fact that this process fails to apply to mono-morphemic items such as n[aj]ghtingale (nightingale), [aj]vory (ivory) and [ɔw]maha (Omaha), SPE recurs to abstract underlying structures such as /nixtVngael/ for nightingale: since there is no long vowel in

110 Chap 6: The life of boundaries in post-SPE times the underlying representation of this word, it cannot be subject to shortening (and is ultimately turned into [aj] by independent rules). When assuming the Revised Alternation Condition, this kind of abstract underlying representation can be dispensed with: nightingale and the like remain untouched by Trisyllabic Shortening because they are underived (Kiparsky 1982a, Booij 1987:54ff). In the same way, the theory gives up on the ambition to rule over "exceptions" that fail to correspond to any morphological regularity. For example, vowel shortening applies to mean [miin] – meant [mEnt] but not to paint [pejnt], pint [pajnt], mount [mawnt] because the latter are morphologically simplex, and their long vowel is specified as such in the lexicon: /miin+t/ vs. /pejnt/. Shortening, then, applies only to derived environments (Kiparsky 1985a:87). On the other hand, Kiparsky's struggle for defining a reasonable line of division between computation and the lexicon also meant that certain parts of morpho-phonology need to be defended against concretist ambition. In §144 below, it is argued that Lexical Phonology, i.e. the theory that grew out of Kiparsky's struggle with abstractness, may be viewed as an attempt to maintain as much morpho-phonology as possible in the computational device of phonology while cutting away the wildest outgrowths of unrestricted SPE. Whatever the balance between computation and the lexicon, it is for sure that boundaries are eliminated from analysis every time a phonologist decides to freeze an alternation by recording two distinct lexical entries, rather than deriving it by rule. The evolution in the 70s and early 80s clearly goes that way: some possibly significant variation notwithstanding, all phonological theories that individuated in the 80s gave up on some computation that was practised in SPE.31 The effect in the 80s thus was a phonological landscape where computation was not king anymore, and boundaries less important: a negotiated and more balanced relationship was established with both the lexicon and representations. The following section looks at the effect of the latter on boundaries.

31

This concerns theories as diverse as Lexical Phonology (§144), Feature Geometry, Government Phonology, Dependency Phonology and even Halle & Mohanan (1985), the direct continuators of SPE in the new autosegmental environment.

Internal and external SPE-revision: Kiparsky (1968-73) and NGP 111 130

7.6. Autosegmental structure: from representational to procedural management of the word boundary The advent of autosegmental structure in the late 70s had an important effect on the number of boundaries that are needed in order to state phonological processes. While in the linear environment of SPE all phonologically relevant morpho-syntactic divisions were represented by a boundary in the structural description of rules, the (re)introduction of syllable structure in form of an autosegmental arborescence replaced much reference to boundaries by reference to syllabic constituents. The most vigorous effect is produced by the coda, which was introduced in place of the disjunction __{#,C}. That is, rules now make reference to the coda, rather than to "#,C" when a coda phenomenon is described. Since the coda context is very frequent in the statement of different rules and across a variety of languages, a great deal of #s disappeared from phonology. An interesting question is whether the coda is just a non-linear surrogate of the end of the word, which continues to require a positive (and diacritic) statement in the phonology. That is, syllable structure is established by a syllabification algorithm that parses the items of the linear string. However, no # is parsed since the coda status of a consonant follows from the fact that there is no vowel to its right (rather than from the presence of a #). This, in turn, supposes that the phonology "knows" that there is "nothing to its right", even though another word may follow. In other words, the syllabification algorithm needs to know the limits of the domain to which it applies. This information of course is morpho-syntactic in nature; but it does not appear in the phonology (i.e. in a structural description). Rather, it defines the stretch of the linear string to which phonology applies – an entirely different perspective. Obviously, cyclic derivation, or the segmentation of the linear string into phases in more modern terms, shines through in this discussion. As a matter of fact, then, autosegmental structure shifted the grounds of the coda context __{#,C} from a representational boundarybased to a procedural analysis that relies on the segmentation of the linear string into autonomous chunks. Therefore, it is correct to say that autosegmental structure eliminated a good deal of boundaries – or rather: replaced their function by a procedural device. In any event, the procedural solution is not diacritic in nature, a fact that may be viewed as an improvement over the diacritic boundary solution. This could well have been used as an argument for the introduction of autosegmentalism.

112 Chap 6: The life of boundaries in post-SPE times It appears, however, that the articulation of procedural and representational intervention was not a topic of great interest in the 70s. It is maybe for this reason that the representational-procedural trade-off and hence the boundary-eliminating side-effect of autosegmental structure went by and large unnoticed (this was also pointed out in Vol.1:§86). Another reason why nobody used the coda in order to argue against awkward diacritic boundaries is probably that there was no specific antipathy against boundaries in the first place: they were not considered aliens, and their diacritic character was not judged unwarranted. We will see shortly (§§136f) that there were exceptions to this attitude: namely Pyle (1972) and Rotenberg (1978) called the diacritic character of boundaries into question, albeit against the mainstream and without much impact on contemporary practice. The diacritic character of boundaries only became an issue when Elisabeth Selkirk set up Prosodic Phonology against the boundary record (§373).

131

8. What exactly is the output of translation, if any?

132

8.1. Growing uncertainty: what kind of intermundia animals are boundaries? Devine & Stephens (1976) formulate the question that is at the heart of the debate on boundaries in the late 70s. (52)

"What ARE phonological boundaries? […] What are the relations and distinctions in function between phonological boundaries and morpho-syntactic rule environments?" Devine & Stephens (1976:285f), emphasis in original

The question is about the diacritic character of boundaries: phonologists felt (and some even demonstrated, see §136) that they are not properly phonological objects: neither phonemes nor segments nor anything else that would be known in phonological quarters. But obviously they are not morpho-syntactic objects either, even if they carry higher level information. Hence "the question of whether the juncture is a phonological entity or merely a list of morpho-syntactic boundary sequences then presents itself as usual" (Devine & Stephens 1976:307). The mushy identity of boundaries, which seem to be a mysterious blend of morpho-syntactic and phonological properties, led Kenstowicz & Kisseberth (1977:103ff, 1979:401ff, 421) to introduce a question that will be known as the direct syntax approach in the 80s.

What exactly is the output of translation, if any? 113 (53)

"Does the fact that a given rule applies only to structures containing a particular morpheme or morpheme class constitute evidence that a boundary different from the morpheme boundary is present? In other words, are ALL instances where the grammatical (particularly, morphological) identity of an element is relevant to a rule's application to be analysed in terms of boundaries, or is there a distinction between phenomena properly described by making reference to boundaries and phenomena properly described by means of direct reference to morphological identity? And if there is a distinction, how do we know when we are dealing with boundaries and when we are not?" Kenstowicz & Kisseberth (1977:103f), emphasis in original

Kenstowicz & Kisseberth's suggestion is one of the isolated and unsystematic occurrences in the 70s of the direct syntax idea that will be the main competitor of Prosodic Phonology in the 80s (§407). Along the same lines, Larry Hyman's work was already quoted (§123), and we will see in §137 below that direct syntax is also the natural escape hatch for those who really want to eliminate boundaries from phonology because of their diacritic character. Finally, the non-privative stance of SPE (§90) also contributed to the confusion: it led the literature to distinguish between "phonological boundaries" and "morpho-syntactic boundaries" (e.g. Devine & Stephens 1976:286f, 1980:75). The former are those that are phonologically relevant, while the latter represent the exhaustive morpho-syntactic structure. SPE supposes both to be shipped to phonology, which means that at some point the subset of morpho-syntactic boundaries that are phonologically relevant mutate into phonological boundaries: "syntactic boundary markers are replaced by phonological ones, when significant, or deleted" (Saltarelli 1970:37). Symptomatic for the unease with boundaries, then, is Greenberg (1970:10), who talks about "such shadowy but indispensable concepts as phonological word boundaries". Devine & Stephens (1976) conclude by striking the same sour note. (54)

"All of this [the treatment of boundaries in early generative times] took place within a theoretical framework which seemed content to relegate boundaries to a sort of ill-defined intermundia between the syntax and the phonology" Devine & Stephens (1976:298)

The frustration that shines through the quotes of this literature is largely due to the feeling that unbeloved boundaries are without alternative.

114 Chap 6: The life of boundaries in post-SPE times The following sections discuss attempts at breaking free from the straitjacket imposed by boundaries.

133

8.2. Boundaries cannot be segments

134

8.2.1. If not segments, what then? McCawley (1968) initiated another line of attack against boundaries and their diacritic character: he doubts that boundaries are segments. Recall that SPE accommodates boundaries in a particular sub-species of segments that is invented just for their purpose, i.e. [-segment] segments (§87). (55)

"There are two aspects of this viewpoint [the general use of juncture by 1968] which could be contested: the notion of juncture as a phonetic/ phonemic (rather than morpho-phonemic) entity and the segmentalness of juncture." McCawley (1968:52f)

In §136 below, arguments are reviewed which give flesh to this critique: boundaries do not behave like /p/, /u/ or any other regular segment. If this is true, then there are only two possible consequences: either there is no translation at all, or its output must be truly phonological objects, that is objects which are known in the phonology in absence of any issue related to the interface. We have already seen two approaches that follow the former option: structuralist Level Independence and Natural Generative Phonology. These not only deny the existence of translation (top-down: from morpho-syntax to phonology): they exclude the presence of any kind of morpho-syntactic information in phonology altogether. However, it is also possible to refuse translation while still allowing for morpho-syntactic information to bear on phonology: this is the so-called direct syntax approach that we have come across in Hyman's work which was reported in §123, and also in the previous section where Kenstowicz & Kisseberth's worry was discussed. On this count, the structural description of phonological rules directly refers to morpho-syntactic categories (structure as much as labels). As we will see below (§136), this is what Pyle (1972) goes for after having shown that boundaries are meaningless diacritics, in any event not any kind of segment. The other option is to maintain both the influence of morpho-syntax on phonology and translation: the output of translation, then, must not be diacritics or rhetorically wrapped versions thereof (such as juncture pho-

What exactly is the output of translation, if any? 115

nemes or SPE-type boundaries); rather, morpho-syntactic information must incarnate into truly phonological objects which exist in the phonology for phonology-internal reasons anyway. In the environment of SPE, truly phonological objects are (real) segments and their ingredients, distinctive features. These are then the only possible output of translation if the output must be non-diacritic. It is argued in Vol.2 (see also Scheer 2008a, 2009a,c) that this solution is correct: it is the heart of Direct Interface. The headstone of this option is the definition of the term "truly phonological": those who invent diacritics always invent some phonological-looking disguise and haste to assure that their diacritics are no diacritics: this is true for juncture phonemes, [-segment] segment boundaries and later in the 80s for prosodic words and the like. The aforementioned independent and pre-theoretical criterion for alienness in phonology is discussed at greater length in §405: something is truly phonological iff it can be observed in the phonology in absence of any morpho-syntactic conditioning. Understandably enough, though, this logical option was not pursued in the 70s when people were fed up with boundaries and diacritics: rather than trying to amend the existing system by modifying the output of translation, translation as such was thrown over board. This, it is argued below, is throwing out the baby with the bathwater. When reading through the literature of the 70s, I have come across one single attempt to translate morpho-syntactic information into a truly phonological item: Lass (1971). The following section introduces his problem and reasoning.

135

8.2.2. Lass (1971): # is [-voice] Lass (1971) tries to get a handle on Old English fricatives, whose voice value is entirely predictable from their environment: fricatives are voiced between vowels and sonorants, but voiceless if geminate, adjacent to a voiceless obstruent or to a word boundary. Lass interprets geminates as a particular instantiation of the context where a voiceless obstruent precedes or follows. Voiceless consonants and # thus form a single distributional class in Old English. The question is whether it is a natural class as well. Since its action is to make adjacent fricatives voiceless and half of its members precisely are defined by the feature [-voice], Lass (1971:16) concludes that the

116 Chap 6: The life of boundaries in post-SPE times boundary # must also bear this feature: "# is really a voiceless obstruent (albeit one with no articulatory features besides voicelessness)." This enables Lass (1971:17) to write the rule under (56) below, which earns the merit of avoiding a disjunctive statement where causes are obviously uniform. (56)

Fricative voicing assignment [+ continuant] → [α voice] / [α voice]

Lass (1971) is perfectly aware of what he is doing: he knows that making a boundary incarnate into a real feature, [-voice], is outlandish and unwarranted in the environment of SPE. (57)

"Is it justified to talk in these terms, and refer to boundaries as something that segments can assimilate to?" Lass (1971:16)

His answer is positive: "if in fact we can say that # is marked for [-voice], then the voicelessness of fricatives can reasonably be considered an assimilation" (Lass 1971:17). Needless to say, Lass' (1971) contribution to the question what exactly boundaries are and what precisely is the output of translation remained without any echo. 136

8.2.3. Bankruptcy of boundaries and abandon of translation: Pyle (1972) Rather than merely another voice airing the discomfort and frustration with boundaries, the article by Pyle (1972) properly demonstrates their definitive bankruptcy. Pyle shows that it is absurd to treat boundaries as regular segments by pointing out the unwarranted predictions of this perspective. Just like Lass' contribution, though, Pyle's article did not have any impact on the contemporary literature. Were boundaries ordinary segments like /p/, /i/ etc., they would have to be subject to ordinary SPE rewrite rules. That is, they should be able to occur as any of the variables in the universal rule format X → Y / A__B. One such participation of boundaries in the rule component was indeed proposed: boundary mutation rules (§112). These place boundaries in X and Y (e.g. # → +), including the possibility for them to be eliminated. But what about cases where boundaries would be inserted by some rule ø → + / A__B?

What exactly is the output of translation, if any? 117

One could argue that SPE distributes boundaries according to morpho-syntactic divisions. Therefore, no boundary can be inserted by rule in the middle of morphemes. However, this would require a filter that is active after the application of rules. The same filter could be said to be responsible for the absence of rules that change the linear order of boundaries in the way known from segmental metathesis +C → C+ / A__B (Pyle 1972:520). That is, strings where boundaries have been moved inside morphemes would be ill-formed. An additional convention would be needed in case some rule eliminates an entire morpheme, thereby creating a cluster of boundaries. Hence A+B+C becomes A++C by the application of B → ø / A__C. Obviously, ++ is not a relevant linguistic object, and no rule ever appeals to it. But again, it could be argued that boundary clusters are eliminated by the general SPE convention that reduces any cluster of #s to two (§90), and which is extended to clusters of +s (that are reduced to one). However, the need for special provisions that regulate the life of boundaries, but not of regular segments, is precisely the point that Pyle (1972) makes: pretending that boundaries are just a little peculiar kind of segments is a trick. They are fundamentally different from /p/, /u/ etc. Still more serious than the special status of boundaries are the two following things that grammar cannot do to boundaries, but should be able to. Since anything can be turned into anything in SPE, and since boundaries are ordinary segments, some process should be able to transform a boundary into a regular segment: + → a / C__C. Hence, say, /dog+s/ would come out as /dogas/. This, of course, is unheard of in natural language. The other impossibility that Pyle (1972:524) points out is the peculiar "invisible" status that Chomsky & Halle (1968:364ff) assign to boundaries. That is, rules are supposed to apply irrespectively of boundaries unless a specific boundary condition is mentioned in their structural description. That is, any rule which applies to the string XYZ also applies to X+YZ, XY+Z and X+Y+Z. If boundaries are not any different from regular segments, the latter should also be able to be "invisible" at times. But of course, there is no rule in natural language that ignores, say, /p/s unless a /p/ is explicitly mentioned in its structural description. Pyle (1972) identifies the fundamental crux as follows. (58)

"All of the rules in question here would have the effect of making the characterisation of formatives [i.e. boundaries] partly phonological. In these terms, the weakness of BM theory [boundary marker theory] is that it would allow phonological rules to redefine formatives in phonological terms." Pyle (1972:521)

118 Chap 6: The life of boundaries in post-SPE times In other words, in SPE boundaries represent morpho-syntactic information, but beyond the phonological effect produced are also supposed to be phonological objects with phonological behaviour (i.e. segments). It is this autonomous phonological life that is incompatible with their actual behaviour, and also with the special status that they are assigned ex cathedra in SPE. 137

8.2.4. When boundaries are bankrupt, the alternative is direct syntax: Pyle, Rotenberg, Hyman, Kenstowicz & Kisseberth The conclusion that Pyle (1972) draws from the bankruptcy of SPE-style boundaries is their non-existence. Phonology does make reference to morpho-syntactic information, but this information is not translated into items that are inserted into the phonological string. Instead, Pyle advocates a transderivational mechanism whereby phonological rules "can look back in the derivation" and thereby detect the phonologically relevant morphosyntactic structure even though it has already been erased at earlier derivational stages (i.e. upon lexical insertion). This of course is at the expense of violating the basic principle of ordered rule application whereby the input of rule n+1 must be the output of rule n. In any case, though, Pyle's reaction on the failure of boundaries is a version of the direct syntax approach that we have already come across: he looks back into the derivation until he leaves phonology and recovers morpho-syntactic structure. That phonology ought to make direct reference to morpho-syntactic structure and labels is also the conclusion of Rotenberg (1978), albeit following a somewhat different motivation: in a chapter called Against Boundaries, Rotenberg (1978) raises principled objections against the presence of any kind of objects in phonology that carry morpho-syntactic information. Particularly the diacritic character of boundaries is at stake, and a finger is pointed at phonologists, who do not behave according to regular linguistic (and scientific) standards: it would not cross the mind of any syntactician to invent some deus ex machina just in order to be able to refer to it when he is unable to do his job on syntactic grounds. That is, why do phonologists tolerate apples and bananas, when the idea of having aliens in their theory is judged outlandish by all other linguists?

Conclusion 119 (59)

"Before I say anything, I note the naked fact that phonological and morphological rules have idiosyncratic domains of application: The morpheme, the syllable, the word, possibly the phrase, and the sentence come to mind. Rules of phonology and morphology are not very different from other kinds in this respect; syntactic and semantic rules of course show analogous boundedness, proving themselves to be limited to certain phrasal categories, sentences, utterances, or discourses. In syntax and semantics, this unavoidable observation is perceived as an incitement to discover properties of the various rules, or, better, of the various components to which they belong, from which their limitations will follow. A fatalistic and slightly empty solution to the problem, which no one even thinks to propose, would be to set up ad hoc boundary symbols flanking each sort of domain, which our rules can now pay attention to as it appears necessary. By doing merely this we seem to be condemning ourselves to a lack of insight into the several systems of rules. This same solution, however, is precisely the one generally accepted among phonologists. In order to implement it […], one quickly finds the need for a great deal of theoretical machinery to place boundaries, to delete most of them when they pile up, and to ignore the rest of them when they get in the way. All of this comes from assuming that boundaries exist as items of vocabulary on a par with the others." Rotenberg (1978:16f)

Pyle (1972) and Rotenberg (1978) are thus representatives of the direct syntax alternative that is born from the obvious inadequacy of boundaries (Elordieta 2008:217ff also describes this evolution). Recall the discussion of other proposals along these lines in §§123,132: Hyman (1978:459), Kenstowicz & Kisseberth (1977:103ff, 1979:401ff, 421) and also Szpyra (1989:11) call for the elimination of translation, and its replacement by direct reference to morpho-syntactic structure and labels. The condensation of these isolated voices into a major player of interface theory in the 80s is discussed in §407.

138

9. Conclusion The discussion in the post-SPE period was completely monopolised by the representational side of the interface coin. The procedural management of morpho-syntactic information had just been introduced in SPE (its original mention in Chomsky et al. 1956 had had no real impact), but cyclic derivation went by and large unnoticed in the 70s until Lexical Phonology was established in the early 80s. The following chapter zooms in to this theory and to the procedural means of impacting phonology.

120 Chap 6: The life of boundaries in post-SPE times On the dominant representational front, thus, the question of boundaries was discussed from top to bottom. At the end of the decade, a general feeling of dissatisfaction was widespread, but viable alternatives were not really fleshed out. Except perhaps the idea to abandon representational intervention in phonology altogether: what will later be known as direct syntax approaches was present in the literature in an embryotic from through several isolated and uncoordinated voices. Interestingly, though, the overall plot was set by the end of the 70s: all problems that are prompted by the representational question, and all logically possible solutions, were laid out. Up to the present day, all that followed is just more of the same, if using different vocabulary and a different rhetorical packaging. Let us review the three critical questions in their logical order. The first issue is whether or not phonology makes reference to morpho-syntactic information. There can be no reasonable doubt that this question is settled: higher level information informs phonology all over the place and is necessary for phonological computation to be able to run. The second issue is whether this reference is direct or through the mediation of translation. While direct reference could be reasonably advocated in the 70s where modularity was not an issue, it is ruled out if modularity is taken seriously: different modules do not speak the same language. While generative theory is a broad application of modularity to language, its first incarnation as SPE violates modularity in central areas (see §99). The contours of modern modular theory were only defined in the 80s (see the Interlude §586, the modular argument is discussed in §§649,692). This is when Indirect Reference, the founding statement of Prosodic Phonology, appeared. While it did not explicitly rely on modularity (§414), it grew on the quarrel with direct syntax approaches (§407). If thus reference to morpho-syntactic information is needed but must not be direct, the final question is what it is made to. That is, what exactly is the output of the translational process? Boundaries do not qualify for the reasons shown by Pyle (1972). Nor do any other diacritics, for the same reason. Morpho-syntactic information can only incarnate into objects that are truly phonological, i.e. which are used by the phonology in absence of any interface issue. In principle, features are good candidates (see Lass' 1971 proposal regarding [-voice]). However, we will see in §§653,663 below that a pervasive empirical generalisation points to the fact that melodic primes (i.e. everything that is located below the skeleton) are never carriers of morphosyntactic information in phonology.

Conclusion 121

We have come across other attempts to make the output of translation a piece of the vocabulary that is offered by whatever the current phonological theory. The incarnation into phonemes that structuralists promoted failed as much as the incarnation into segments, the regular phonological currency of SPE. In both cases, arbitrary diacritics were rhetorically wrapped into a phonemic or a segmental guise. True incarnation into a real phoneme or a real segment is by and large outlandish: except Lass' (1971) attempt, nobody has ever found a truly featural identity of #. Hence there are only two solutions: either the entire reasoning regarding Indirect Reference and translation is wrong, or the attempts to incarnate morpho-syntactic information into truly phonological vocabulary have failed because phonological theories have not provided the right vocabulary thus far. It is argued in Vol.2 that the latter conclusion is correct: rather than phonemes or segments, syllabic space – a CV unit – is the phonological carrier of morpho-syntactic information.

139

Chapter 7 Lexical Phonology

140

1. Introduction The following pages introduce the basic ideas of Lexical Phonology in its classical stratal skin. The ambition is not to cover all aspects of the theory (this would go way beyond the scope of the book); rather, the focus rests on the treatment of extra-phonological information in phonology. More detailed introductions from an informed look-back position are offered by Giegerich (1999:7ff), McMahon (2000:35), Rubach & Booij (2003) and Bermúdez-Otero (forth a). Primary texts that have made important contributions to the theory include Aronoff (1976), Allen (1978), Pesetsky (1979), Kiparsky (1982a,b,c, 1985a) Mohanan (1982, 1986), Rubach & Booij (1984, 1987), Rubach (1985, 1986, 1993), and an early overview is provided by Kaisse & Shaw (1985). Echoing the previous chapter, the presentation looks at Lexical Phonology through the prism of boundaries in order to see to which extent the rather frustrating result of the post-SPE period can be relieved. Lexical Phonology achieves a maximal turnout from the procedural management of morpho-syntactic information. This is the exact opposite take of previous endeavour in the interface since the 19th century: the idea of procedural interface management was only introduced by Chomsky et al. (1956) (see §80); prior to this article, morpho-syntactic information was only hooked on some representational object that appears in the phonology. The maximisation of procedural labour is also the opposite take of SPE, which (unlike in the segmental area) was by and large representational at the interface (§§92,107). Cyclic derivation (of which Bermúdez-Otero forth b provides an overview) is thus a specifically prominent feature of Lexical Phonology, a theory that has shaped its face under the stratal banner for generations of phonologists. The following pages also try to show the price that has to be paid when the trade-off between representational and procedural interface management is shifted to the latter extreme.

124 Chap 7: Lexical Phonology 141

2. Empirical foundations

142

2.1. Affix classes and affix ordering The critical discovery for the establishment of Lexical Phonology was the existence of two classes of affixes in English, and of their non-arbitrary ordering with respect to the stem. Relevant generalisations for the former insight were made by SPE (Chomsky & Halle 1968:84ff, see §92), for the latter by Dorothy Siegel's (1974) MIT dissertation. English affixes appear to fall into two classes whose members share a number of morphological and phonological properties that are not observed for items of the other class. Table (60) below shows class membership (according to Mohanan 1986:16; see §245 for discussion).32 (60)

class membership of English affixes class 1 class 2 inun-ity -ness -ic -less -ian -hood -ory -like -ary -dom -ion -ful -ate -ship -al (adjective-forming) -ed -y -ing (noun-forming)

(adjectival) (noun-forming)

There are several diagnostics for class membership. On the morphological side, distributional evidence is brought to bear: class 1 affixes occur closer to the stem than class 2 affixes (this is where their name comes from). That is, affixes of both classes can freely attach to stems that already contain an affix of the same class (class 1: atom-ic1-ity1, univers-al1-ity1, class 2: atom-less2-ness2, beauty-ful2-ness2, guard-ed2-ness2). In addition, class 2 affixes can hook on a class 1 affix (univers-al1-ness2). However, sequences of class 2 - class 1 affixes do not occur (*atom-less2-ity1,

32

The two classes at hand appear under different headings in the literature: level 1 vs. level 2, stress-shifting vs. stress-neutral, neutral vs. non-neutral, cohering vs. non-cohering, cyclic vs. non-cyclic. In the remainder of the book, they are consistently referred to as class 1 vs. class 2.

Empirical foundations 125

*piti-less2-ity1, *guard-ed2-ity1 etc.). This observation is known as the affix ordering generalisation (Siegel 1974).33 Another morphological property that sets the two classes apart is the ability of their members to attach to bound stems. Class 1 affixes can be concatenated to stems that have no independent existence (in-ert, in-trepid) as well as to independently existing words (in-tolerable), while class 2 affixes can only attach to words (un-aware, against *un-ert, *un-trepid). This morphological categorisation is mirrored in the phonology. In SPE, affix class distinction is expressed representationally: class 1 affixes are introduced by the + boundary, while class 2 affixes come with the "stronger" # boundary (see §92). The phonological effects encountered most prominently concern stress placement, and this criterion is commonly admitted as the basic diagnostic for class membership. Class 1 affixes are reputed to be stressshifting, while class 2 affixes are stress-neutral. Classical examples are words such as párent, válid and átom, which appear with right-shifted stress when occurring with a class 1 suffix (parént-al, valíd-ity, atóm-ic), but conserve their lexical pattern when followed by a class 2 suffix (párent-hood, válid-ness, átom-ise). Further phonological affix class-based effects are reviewed as we go along.

143

2.2. Cross-linguistic reality of affix classes A question of course is whether affix classes are a general phenomenon or merely a specific property of English. Although the equivalence is not complete, it is obvious that in English the two classes in question reflect different lexical strata in the historical development of the language: class 1 affixes are of Romance origin, while class 2 affixes represent the Germanic heritage. The question whether affix classes exist in other languages is thus legitimate. Over the years, studies of other Germanic languages such as Dutch and German, but also of unrelated languages such as Malayalam (Dravidian, Mohanan 1986), Basque or Dakota (native American) have been mobilised in order to show that affix classes have a cross-linguistic reality and are therefore relevant for linguistic theory as such. The situa33

See for example Mohanan (1986:15ff), Szpyra (1989:40ff, 178ff), Giegerich (1999:11ff) for more detailed evidence. The empirical validity of affix ordering is further discussed in §243 below.

126 Chap 7: Lexical Phonology tions found confirm the English pattern: typically, affix classes correspond to the import of various strata of vocabulary in different historical periods of the language. Also, the number of lexical strata varies across languages, and in principle is unbounded. Booij (2000 [1996]:297) offers an overview of these questions. Before looking at how the concordant phonological and morphological effects of affix classes were condensed into Lexical Phonology, the following section recalls that the theory is also a result of a conceptual issue, the abstractness debate.

144

3. Lexical Phonology and the abstractness debate Lexical Phonology also roots in the abstractness debate that was initiated by Kiparsky (1968-73) and dominated the 70s (§125f, also Vol.1§307). This origin is made explicit in Kiparsky (1982a:34ff). Given the utterly overgenerating SPE mechanics, it was clear that the expressive power of the grammar needed to be marshalled somehow. Restrictions on possible underlying representations ((Revised) Alternation Condition, Kiparsky 1973a, see §§185ff) and the computational component (Strict Cycle Condition, Chomsky 1973, Kean 1974, Mascaró 1976, see §188) flanked Chomsky's (1970) cut-down of transformational power in morphology: Remarks on Nominalization discharge derivational morphology from words such as reduction, transmission and recital. Since their formation is unproductive and semantically opaque, Chomsky argues, they are stored in the lexicon as a whole.34 On the phonological side of this movement towards a less permissive grammar, the critical question is the attitude towards alternations that are subject to more or less heavy morphological conditioning. In the name of phonological realism, Natural (Generative) Phonology bans a rule from phonology as soon as its formulation involves the slightest bit of a morphological condition (see §127). Contrasting with this radical position, Kiparsky tried not to throw out the baby with the bath water: abstractness is not evil in itself; therefore only a certain kind of abstractness needs to be done away with (see §129 for examples).

34

This move heralds the era of lexicalism, which will dominate the 80s, but is called into question in certain versions of minimalism (and in Distributed Morphology). The lexicalist issue is discussed at greater length in §§539,569.

The general architecture of Lexical Phonology 127

In this sense, Lexical Phonology may be viewed as an attempt to maintain as much morpho-phonology as possible in the computational device of phonology while cutting away the wildest outgrowths of SPEinduced overgeneration. The early concern for overgeneration and a restricted grammar (of which Szpyra 1989:15ff, Giegerich 1999:2f, 100ff and McMahon 2000:35f provide an informed overview) was revived in the late 80s when Halle & Mohanan (1985) proposed a version of Lexical Phonology which untied much of the restrictions that the Lexicon had established: there was an inflation of the number of English strata, and derivations could "loop" back into earlier levels. McMahon (2000:50ff) considers this a major setback for the theory.

145

4. The general architecture of Lexical Phonology

146

4.1. Interactionism – a new idea in the interface landscape The empirical and conceptual challenges mentioned, i.e. affix classes and affix ordering on the one hand, abstractness on the other, were condensed into what may be called classical Lexical Phonology. This move is described at the outset of Kiparsky (1982a,b). The critical innovation that afforded to kill two birds with one stone was the idea of interspersing word formation rules with phonological rules: first you apply phonology to a piece, then you concatenate an affix, then you do some more phonology on the new string created, then you concatenate another affix etc. This procedural scrambling of morphology and phonology was called interactionism later on. It was first proposed by Pesetsky (1979), Booij (1981), Williams (1981) and Kiparsky (1982a,b).35 Interactionism indeed allows capturing the phonological effect of affix classes, which was known since SPE, and the fresh facts from morphology regarding affix ordering (this is where Kiparsky's 1982b title "From Cyclic Phonology to Lexical Phonology" comes from). Interactionism therefore certainly deserves to be recognised as the founding statement of Lexical Phonology (also in its self-understanding, e.g. Rubach & Booij 1984:1). The basic stratal architecture of the theory is

35

See §308 for discussion of Bresnan (1971), an early analysis along the lines of interactionism.

128 Chap 7: Lexical Phonology a consequence, as much as the introduction of morpheme-specific minigrammars. Interactionism is a new idea in the interface landscape: it is the only major concept that was not heralded in SPE. As a genuine contribution of Lexical Phonology, it will be assaulted by the generative mainstream in the 80s (Halle & Vergnaud 1987a, see §215), before making a swift career in current minimalist theory, where it appears as derivation by phase (§304) (see the summary of this evolution in §676).

147

4.2. Strata and their procedural order: how to kill two birds with one stone Interactionism accounts for affix ordering by assuming the existence of several procedurally defined levels (or strata – both terms are used synonymously in this book) that the morpho-phonological derivation runs through without being able to loop back: class 1 affixation is done at level 1, hence before class 2 affixes are concatenated at level 2 (see §152 for a schematic overview). This guarantees that a stem may take on a (number of) class 1 affix(es) and then move on to level 2 where class 2 affix(es) may join. The result are sequences of class 1 - class 2 affixes. The reverse order, however, cannot be generated because this would imply a return to level 1 once an affix was added at level 2. A critical ingredient of Lexical Phonology is thus the existence of procedurally ordered strata. The prohibition of returning to an earlier stratum is called level ordering. Note that level ordering is logically independent from interactionism; it is only with this additional stipulation that affix ordering can be derived. On the phonological side, two-step affixation allows for a two-step interpretation, hence capturing the insight of the phonological cycle: starting with the stem, phonological interpretation works its way outwards through the affixes (see §100). The outward-bound movement is guaranteed by the fact that level 1 accommodates class 1 affixes, and level 2, class 2 affixes. Recall from §141 that the former occur closer to the stem than the latter. Table (61) below illustrates how interactionist Lexical Phonology works on the grounds of the familiar contrast between paréntal and párenthood (see §93).

The general architecture of Lexical Phonology 129 (61)

párent - parént-al vs. párent-hood in Lexical Phonology parent parént-al lexicon parent parent level 1 concatenation — parent-al stress assignment párent parént-al level 2 concatenation — — rule application — —

párent-hood parent — párent párent-hood —

Since the derivation strictly follows the path lexicon → level 1 → level 2 without possibility of looping back, class 2 affixes may not come to stand closer to the root than class 1 affixes. Table (61) shows that level ordering is needed on the phonological side as well: the stress assignment rule is only active at level 1. Párent-hood retains stress on the root because the stress rule is absent from level 2. Were párent-hood able to go through level 1 after -hood is concatenated, *parént-hood would be produced by the reapplication of the stress rule.36 148

4.3. Morpheme-specific mini-grammars

149

4.3.1. The stratal perspective supposes selective rule application An important consequence of strata is the split of phonology into two distinct computational systems. That is, the stratal model only works if there are distinct phonological mini-grammars, one for each stratum: phonology 1 assesses the string that was concatenated at stratum 1 (and thus contains the stem plus class 1 affixes), while phonology 2 interprets the result of the concatenation that was made at stratum 2 (see (62) below). Phonology 1 and phonology 2 are thus morpheme-specific: they are designed to apply to strings that are only made of a subset of morphemes. The example under (61) shows why the stratal approach requires distinct morpheme-specific mini-phonologies. The class 1 affix -al is concatenated at level 1, while the class 2 affix -hood comes in at level 2. In order to 36

The possibility of looping back to previous levels was actually proposed by Halle & Mohanan (1985), together with an inflation of strata for the analysis of English. This option unties the bonds that strata are supposed to introduce – it is a kind of SPE system in an empty stratal shell that overgenerates as wildly as before (see §144). For that reason, the model has not found many followers: McMahon (2000:50ff) provides an overview of the critique.

130 Chap 7: Lexical Phonology achieve the correct result, stress assignment needs to apply at level 1 – but it must not apply at level 2. Were it also active after the concatenation of -hood, *parént-hood with regular penultimate stress would be produced. Therefore the phonological computation that assesses level 1 strings must be different from the one that interprets level 2 strings. In classical Lexical Phonology where computation is done in terms of rules, this contrast is expressed by the fact that some rules are present at level 1, but absent at level 2 (this is the case of stress assignment in our example), or viceversa.37 That is, the set of rules that applies to level 1 strings is distinct from the set of rules that interprets level 2 strings. Rules are therefore identified as level 1 or level 2 rules: they must "know" where they apply. This is done by some lexical specification (domain assignment, see §151).

150

4.3.2. Underapplication is achieved by distinct mini-grammars and level ordering Distinct morpheme-specific mini-grammars are needed because a subset of the phonological computation which is active in the language (stress assignment under (61)) must not apply to strings that are created by the attachment of a subset of affixes: class 2 in our example.38 What affix class-based phenomena require, thus, is the underapplication of phonology: a subset of the active phonology of the language must be precluded from applying to strings that were created by the attachment of a certain affix class. In a stratal perspective, this effect is achieved conjointly by the split of phonology into distinct morpheme-specific mini grammars and level ordering. Both are indeed necessary: §147 has shown that distinct minigrammars are toothless in absence of level ordering (hence if the derivation may loop back to previous levels). On the other hand, level ordering alone is unable to enforce underapplication: were the stress rule present at level 2, the incorrect *parént-hood would be derived.

37

38

In more recent constraint-based environments, the contrast between two distinct mini-grammars is expressed in terms of different constraint rankings, which are typically achieved by "constraint reranking" between strata (see §473). There are also phenomena that require the non-application of certain phonological processes to strings that were created by the concatenation of class 1 affixes (so-called level 2 rules). These are discussed in §166.

The general architecture of Lexical Phonology 131 151

4.3.3. How to make mini-grammars different but not waterproof: domain assignment In addition to distinct computational systems on the phonological side and level ordering, classical Lexical Phonology features a third device that participates in achieving underapplication: domain assignment (which Mohanan 1986:21 calls the Stratum Domain Hypothesis). That is, rules are augmented with a diacritic that specifies their domain of application; they thus identify as level 1 or level 2 rules. Domain assignment is not something that the stratal environment requires. Rather, it is a particular view on how mini-grammars should be organised. In order to be able to maintain a certain unity of grammar, classical Lexical Phonology holds up a single pool of rules. According to Mohanan & Mohanan (1984) and Mohanan (1986:13f), these then freely "intervene" at a particular stratum or postlexically: rules may apply in any location, and any location may host any rule. Also, a given rule may happily apply in several locations, for example at stratum 1 and postlexically, but not at stratum 2. In practice, the effective application of rules in different locations is an empirical question: rules apply where their contribution is needed. However, since they do not apply everywhere (this is the whole point of underapplication), they have to know where exactly they intervene. This information is recorded by a diacritic, domain assignment. Were a single pool of rules not a virtue by itself, domain assignment could be dispensed with. The alternative, then, would be to accept that there are two distinct mini-grammars which do not "see" each other, and where the rules of one stratum do not entertain any relationship with the rules of another stratum, notwithstanding the presence of the same rule in different locations. This option makes the distinct computational systems waterproof, as opposed to Mohanan's non-waterproof mini-grammars: in Mohanan's system, the same rule-individual intervenes in distinct strata, while different clones of the same rule are active in different strata in the waterproof option. In one case there is one single pool of rules, while in the other two distinct pools act without their members ever "seeing" each other.39 39

Whether mini-grammars are waterproof or not is an issue in OT, where it is known as the "co-phonology proliferation problem" (Orgun 1996a:114, see §492). It was also discussed in the context of Kiparsky's (1982a,b) attempt to derive the Strict Cycle Condition (SCC) from the Elsewhere Condition (see §190). Mohanan & Mohanan (1984) argue against two independent set of rules on the grounds of rules that apply both in the Lexicon and postlexically.

132 Chap 7: Lexical Phonology 152

4.4. The general picture: interactionist Lexicon → syntax → postlexical phonology Illustrating the above discussion, table (62) below shows the general architecture of classical Lexical Phonology. (62)

underlying representations: underived roots stratum 1 morphology

stratum 1 phonology

stratum 2 morphology

stratum 2 phonology

stratum n morphology

stratum n phonology

Lexicon 1. morphological word-formation rules 2. phonological rules that are sensitive to morphological information

Output: words (i.e. items that have a meaning and a pronunciation) syntax postlexical module 1. phonological rules that are sensitive to syntactic information 2. "automatic" phonological rules, i.e. which are sensitive to phonological information only Output: sentences (i.e. that have a meaning and a pronunciation) towards phonetic interpretation

This overall architecture encompasses the labour that is classically attributed to phonology, morphology and syntax. For the time being, only

The general architecture of Lexical Phonology 133

the interaction between morphology and phonology (interactionist) was considered: the output of stem-affix concatenation and (phonological) interpretation of the strings created are words. Words are stand-alone items that possess a pronunciation and a meaning. Therefore the word factory is called the Lexicon: it produces words. It is opposed to the postlexical area, which is discussed below. For the time being, it is important for the discussion below to make explicit what a (lexical) stratum (or level) is in Lexical Phonology. The Lexicon is made of strata, and a stratum is the association of a certain type of concatenative activity (the attachment of affixes that belong to a given affix class) with a certain type of phonological (and semantic) interpretation (the computation of exactly the string that was created by the concatenative activity described). Within each stratum, concatenation necessarily precedes interpretation, and the phonological grammar that assesses the string is necessarily different from the phonology that interprets strings at other strata (multiple mini-grammars). In the remainder of this book, the word "stratum" (or "level") is used only in this sense, namely when the stratal perspective is opposed to the classical inverted T model where all concatenation precedes all interpretation (§86). Terminology is relevant here because over the years "strata" and "stratal" have become a kind of cover term which makes broad reference to Lexical Phonology as such or, as we shall see, simply to the existence of multiple mini-grammars.40 Let us now look at the major division that appears under (62), which was not discussed yet: the Lexicon is followed by a postlexical module.

153

4.5. Praguian segregation: lexical vs. postlexical phonology

154

4.5.1. The birth of postlexical phonology: Rubach (1981) It was mentioned in §146 that Kiparsky (1982a,b) takes Lexical Phonology to be the synthesis of three strands: affix classes, affix ordering and the abstractness debate. In actual fact, he also quotes a fourth source: cyclic phonology. This is in reference to Mascaró (1976): Kiparsky will adapt his Strict Cycle Condition to the stratal environment (see §188). 40

Halle & Vergnaud (1987a) for example use the word "stratum" in the latter sense (see §235): even though they are anti-interactionist and hence have done away with strata, they call their mini-grammars (cyclic and word-level) strata.

134 Chap 7: Lexical Phonology In this context, Kiparsky quotes Rubach (1981). As far as I can see, Rubach (1981) is to be credited for the formal introduction of Praguian segregation (more on this term in the following section) into generative theory. On the backdrop of the systematic difference between regular and surface palatalisation in Polish, Rubach (1981:18ff) concludes that phonological rules should be split into two separate and derivationally ordered blocks: first all regular rules apply; the output of this computation is then further assessed by "automatic" or surface-oriented rules (such as the distribution of aspiration in English, or surface palatalisation in Polish). This is the birth of postlexical rules, which is how phonological rules that apply to chunks above the word level are called in Lexical Phonology and up to the present day. They are opposed to word-constructing lexical rules, which Lexical Phonology places in the Lexicon. Note that the distinction between lexical and postlexical rules is different from the contrast that opposes level 1 and level 2 rules. While Rubach (1981) only distinguished between the former, fully-fledged Lexical Phonology implements both oppositions (see §152). Finally, it is useful for the discussion below to be explicit on the fact that SPE's word-level phonology (§104) has no equivalent in stratal Lexical Phonology. It is not the same thing as postlexical phonology at all since it concerns only the word-level: SPE explicitly excludes the application of word-level rules to chunks that are larger than the word (§105). A specific version of Lexical Phonology will reintroduce word-level rules later on (in the guise of post-cyclic lexical rules, see §194). In sum, then, Lexical Phonology accommodates (at least) three distinct computational systems: (at least) two in the Lexicon (level 1 and level 2), and one more after syntax has applied (postlexical phonology). There is an important difference in kind regarding the way in which the three computational systems are opposed, though. Lexical strata (i.e. levels) are different because of their morpheme-specificity: we are facing morphemespecific mini-phonologies. By contrast, the opposition between the postlexical and the lexical system is entirely unrelated to the quality of the chunks considered: it is their size alone that matters. Computational systems are thus either morpheme-specific (class 1 vs. class 2) or chunkspecific (smaller chunks, i.e. morphemes, vs. bigger chunks, i.e. words). It is argued below (§212) that the introduction of multiple minigrammars (together with interactionism) is the genuine footprint of Lexical Phonology that makes this theory different from all others. Historically, then, Rubach (1981) was the first to depart from the unitary block of rules

The general architecture of Lexical Phonology 135

that was inherited from SPE – this is how the idea of multiple minigrammars entered the scene.

155

4.5.2. Praguian segregation: syntax and morphology are different Regarding the relationship between syntax and morphology, Lexical Phonology follows the traditional view whereby the two components are distinct computational systems with distinct grammars and possibly distinct vocabulary. Lexical Phonology grants a formal status to this perspective: it introduces an architectural distinction between the Lexicon (i.e. morphology, where words are constructed) and syntax (where sentences are constructed). Only is the order reversed: on the traditional view, syntax is done first, and its output is the input to morphology. The roots of the architecture that Lexical Phonology promotes are ultimately phonological: Booij (1997:264, note 3) traces the distinction between lexical and postlexical phonology back to the Prague Linguistic Circle (Circle 1931) where an distinction is made between the phonology of words (phonologie du mot) and the phonology of sentences (phonologie de la phrase). Below I therefore use the term Praguian segregation in order to refer to this conception (rather than the theory-laden notion of postlexical phonology). Whether syntax and morphology are distinct computational systems or one is a resident issue in linguistics that has gained renewed interest in recent years (see §537, also on (anti-)lexicalism).

156

4.5.3. Morphologically conditioned vs. exceptionless rules The difference between word- and sentence phonology defines the position of Lexical Phonology in the debate regarding morpho-phonology (see §§126,129): how much morphological information, if any, should be allowed for in the statement of phonological rules? Following the take of Lexical Phonology, all that can be expressed by strata, but no more. The effect of morphological conditions on phonological rules is that the rules in question appear to have exceptions: the phonological conditions may be met, but the process may still not go into effect (underapplication: the stress pattern of párent-hood is non-penultimate and hence opaque). On the other hand, there are phonological processes that apply any time their phonological conditions are satisfied, i.e. irrespectively of mor-

136 Chap 7: Lexical Phonology phological conditions. These are then exceptionless (or automatic, i.e. the equivalent of natural P-rules in Natural Generative Phonology, see §127). Lexical Phonology expresses this opposition in terms of the Praguian distinction between word-level and sentence-level phonology: the former is subject to morphological conditions because it builds words and thus interacts with morphology, while the latter describes the life of adult words, hence with no morphology intervening anymore. Finally, Lexical Phonology grants an architectural and procedural reality to the Praguian observation: word- and sentence phonology are done at autonomous derivational levels where either morphemes or words are concatenated. The question, then, to be discussed in §158 below, is why in the architecture of Lexical Phonology morphology bears on the phonology of words (via interactionism), while syntax does not influence the phonology of sentences in the same way (postlexical phonology is noninteractionist, i.e. non-cyclic). 157

4.5.4. Lexical vs. postlexical phonology: respective conditioning factors Beyond the Lexicon, words serve as the input to syntax. On the account of Lexical Phonology, syntax concatenates words, not morphemes; therefore the internal structure of words is invisible to syntax (no look-back, see bracket erasure in §174). Upon the availability of syntactic structure, i.e. after syntax has applied, the string is assessed by the postlexical block of (phonological and semantic) rules. This is Praguian sentence phonology, which concerns processes that are either insensitive to any kind of extra-phonological information (morphological and syntactic alike), or that are conditioned by syntactic structure alone (the rule makes reference to syntactic information). In the latter case, they could not apply in the Lexicon because syntactic information had not yet been constructed. Conversely, rules that are sensitive to morphological information cannot apply postlexically: morphological structure is erased at the end of each stratum (by the aforementioned bracket erasure, see §174 below). According to this system, lexical phonological processes interpret morphological structure and allow for exceptions, i.e. may be opaque, while postlexical rules interpret syntactic structure and are exceptionless. In sum, thus, Lexical Phonology splits the grammatical architecture into two (almost) waterproof components: morphology (the Lexicon) and syntax; both concatenate pieces, and each has its own associated phonol-

The general architecture of Lexical Phonology 137

ogy. Neither the two concatenative devices nor the two associated phonologies need to share the same computation (they could in principle, but in practice they never do). 158

4.5.5. Cyclic vs. postlexical phonology: no cyclic interpretation of words In reference to SPE (§105), rules that apply in the Lexicon are called cyclic in Lexical Phonology because they are said to apply cyclically, i.e. along the lines of the transformational cycle (§100). At the same time, they apply according to the interactionist architecture of the Lexicon, that is in alternation with concatenative activity. Cyclic interpretation in Lexical Phonology thus takes on a meaning that it did not have in SPE: it is interactionist (more on this in §188 below when Kiparsky's 1982a,b Strict Cycle Condition SCC is discussed). By contrast, rules that apply to the string that is returned by syntax are called postlexical and are non-interactionist: there is no interleaving with the concatenative activity of words. In the terminology of Lexical Phonology, postlexical rules are thus non-cyclic. The contrast between lexical phonology that is interactionist and postlexical phonology that is not departs from SPE's conception of cyclic derivation, which concerns morphemes as much as words ("the principle of the transformational cycle […] appl[ies] to all surface structure whether internal or external to the word", Chomsky & Halle 1968:27, see §105). Interestingly, Lexical Phonology imposes the non-cyclic interpretation of words without discussion. As may be seen in the following quote, Kiparsky (1982b) simply decrees that the concatenation of words is not cyclic. (63)

"The former, the rules of lexical phonology, are intrinsically cyclic because they reapply after each step of word-formation at their morphological level. The latter, the rules of postlexical phonology, are intrinsically noncyclic." Kiparsky (1982b:131f, emphasis in original)

The question why sequences of morphemes should, but sequences of words should not be derived cyclically is an important issue that has received almost no attention as far as I can see. Further discussion is provided in §§432,786 below in the light of modern phase theory (the word-spell-out mystery).

138 Chap 7: Lexical Phonology 159

4.6. Definition of interpretational units

160

4.6.1. From brackets to strata A question that comes with cyclic derivation is how cyclic boundaries are defined. Every cycle delineates a portion of the linear string that is interpreted by phonology (and semantics). In SPE, these interpretational units are delineated by brackets, which roughly coincide with morpheme boundaries but in fact are only inserted at every transition of major categories (noun, verb, adjective, §103).41 Lexical Phonology has a different take on the definition of interpretational units: instead of a rigid calculus in terms of morpho-syntactic labels, affix classes define the chunks of the linear string to which phonological (and semantic) computation applies. That is, all affixes of the same affix class are concatenated at a given stratum, and the resulting string is then interpreted. In Lexical Phonology, strata are thus the relevant interpretational units. In order to illustrate the contrast with SPE, consider the word origin-al-ity, which identifies as a three-cycle item [[[origin]N al]A ity]N in SPE, while it represents only one single interpretational unit in Lexical Phonology (both -al and -ity are stress-shifting class 1 affixes): [origin-al-ity]. The question how cyclic boundaries are defined runs through the entire literature: different theories provide (slightly) different answers. In modern times, the issue is called phasehood: the question how phase boundaries are defined is at the forefront of the research agenda (§773) since Chomsky's (2000a et passim) derivation by phase has admitted an interactionist architecture (see §304).

161

4.6.2. Interactionism reconciles cyclic derivation with modularity In SPE, cyclic (inside-out) interpretation is assured by brackets that are inserted into the linear string and guide the derivation as it proceeds (§95). Brackets are needed because all concatenation is completed before interpretation begins (§86): like a sponge they store morpho-syntactic structure (and also the labels associated) in order to make it available in phonology. 41

Hence the word theatr-ic-al-i-ty identifies as [[[theatr]N ic + al]A i + ty]N: five morphemes make three cycles. Cyclic structure is thus a proper subset of morpho-syntactic structure in SPE, even though most of the time both coincide.

The general architecture of Lexical Phonology 139

This raises the issue of untranslated morpho-syntactic information in phonology, which should be prohibited on modular grounds (see §§98f). In short, the prohibition of intertwining concatenative and interpretational activity that was in place in SPE enforces brackets, which are incompatible with modularity. Interactionism does away with this prohibition and therefore opens the way for an entirely different perspective: interpretational units (i.e. the chunks of the string to which phonological computation applies) are delineated by strata, rather than by brackets. Unlike brackets, strata are procedurally defined: at stratum 1, only the root and class 1 affixes exist – class 2 affixes have not yet been concatenated. Therefore, at any given point in the derivation, phonological computation simply applies to the maximal string that is currently available. Table (64) below depicts the two ways of implementing inside-out interpretation. (64)

organisation of inside-out interpretation a. brackets (SPE) interpretation along the fully constructed morpho-syntactic spine [[[root] class 1 affixes] class 2 affixes] PF, LF PF, LF PF, LF b. interactionist/stratal perspective (Lexical Phonology) interpretation upon concatenation 1. [root] → PF, LF 2. [root - class 1 affixes] → PF, LF 3. [root - class 1 affixes - class 2 affixes] → PF, LF

In modern phase theory (Chomsky 2000a et passim, see §§304,771), one would say that spell-out occurs at every phase head: phases define interpretational units. The interactionist perspective of Lexical Phonology (and later on of derivation by phase) thus makes SPE-type brackets superfluous: at no point in the derivation does phonological computation process phonological aliens such as brackets because interpretational units are delineated procedurally, rather than representationally. In other words, interactionism reconciles inside-out interpretation with modularity: it dispenses with the presence of untranslated morphosyntactic information in phonology (more on this in §§170,305,680). This is an important consequence of interactionism. Rubach (1984:225) for ex-

140 Chap 7: Lexical Phonology ample is explicit on the fact that interactionism replaces brackets, but as far as I can see the advance that this represents from the modular point of view is not a concern in the literature.

162

5. The analysis of affix classes and its proceduralisation Affix class-based phenomena in general, and those that are found in English in particular, have already been used above in order to illustrate the behaviour of SPE and Lexical Phonology. They will escort the reader all through the book since they are a valid testing ground for interface theories: affix class-based phenomena are historically relevant, well known and accessible to all readers. The basic facts are introduced in §163 and §166 below. It was already mentioned in §140 that Lexical Phonology proceduralises the interface: morpho-syntactic information that was transmitted to the phonology by representational devices (boundaries and brackets in SPE, see §§92,95,107) is now encoded procedurally. Roughly speaking, boundaries (like brackets) are replaced by lexical strata. We have seen in §97 that boundaries and brackets are orthogonal in SPE: while the former may be modified during phonological computation, the latter are set in stone and will always guide the cyclic derivation through the string. In this environment, SPE proposes a representational management of affix classes: class 1 affixes come with a + boundary, while a # boundary accompanies class 2 affixes (§92); rules then make reference to #, which either blocks or triggers the process at hand. In Lexical Phonology by contrast, affix classes have a procedural management: they are different because they belong to different strata, which means that they are concatenated at different derivational stages. Obviously there is no point in encoding affix class identity twice, i.e. once representationally (by boundaries) and another time procedurally (by strata). Since the whole point of Lexical Phonology is to capture morphological (affix ordering) and phonological effects of affix classes by (procedural) interactionism, the representational treatment of SPE had to go. Lexical Phonology is thus characterised by the proceduralisation of the analysis of affix classes. The competition between procedurally ordered strata and boundaries is perfectly explicit in the earliest source of Lexical Phonology.

Rule-blocking boundaries are eliminated altogether 141 (65)

"To intrinsically order the levels of the morphology as they apply and to identify them uniquely with boundaries at the same time, would be overkill, since the boundaries themselves can do the work of ordering affixation processes. […] I will, therefore, take the perhaps uncautious step in this section of assuming that boundaries are not linguistic units, and will generally assume an ordering hypothesis." Pesetsky (1979:16f)

The detail of this general evolution, however, needs to be sorted out. The following pages discuss the new procedural management of affix classbased phenomena.

163

6. Rule-blocking boundaries are eliminated altogether

164

6.1. Rule-blocking boundaries translate as level 1 rules In order to see the cleaning-up effect that Lexical Phonology produces on the representational side, it is useful to follow the distinction between ruleblocking and rule-triggering boundaries that was introduced in §51. Let us first look the former. A rule-blocking boundary is a specific morphosyntactic division that prevents a phonological process from going into effect, while this process would have applied in its absence. In Lexical Phonology, rule-blocking boundaries are useless altogether. This was already shown in §147: on the SPE-analysis of the contrast between paréntal and párenthood, stress assignment is blocked by #, which characterises class 2 affixes. In a stratal environment, however, the derivation runs without contribution of any boundary. That is, underapplication to class 2 strings is assured by the conjoint action of distinct morphemespecific mini-grammars and level ordering (§150). Crucially, stress assignment is a level 1 rule in this environment. English Trisyllabic Shortening (or Laxening) may provide further illustration of the pattern. Notwithstanding the fact that there is good reason to doubt that this process is synchronically active (see note 42), it was a flagship of the Lexical Phonology literature (e.g. Kiparsky 1982a:35ff, Mohanan 1986:18ff, see also Myers 1987). Trisyllabic Shortening is a process whereby the long vowel or diphthong of bisyllabic words is shortened when a class 1 suffix is added as

142 Chap 7: Lexical Phonology under (66a). By contrast, class 2 suffixes as under (66b) do not provoke any reaction.42 (66)

Trisyllabic Shortening (or Laxening) non-trisyllabic item a. class 1 suffix sane [sejn] Christ [krajst] b. class 2 suffix maiden [mejd´n] wild [wajld]

trisyllabic item san-ity Christ-ian maiden-hood wild-ness

[sQnɪtɪ] [krIstÉS´n] [mejd´nhəd] [wajldnEs]

An informal description will thus state that class 2 suffixes prevent Trisyllabic Shortening from applying. In SPE, this blocking effect is attributed to the presence of a # boundary that comes with class 2 affixes: the context of the phonological rule which is responsible for Trisyllabic Shortening requires that the morpheme break be represented by a + boundary (Chomsky & Halle 1968:180). On the other hand, the procedural architecture of Lexical Phonology does not need to appeal to boundaries. Trisyllabic Shortening is a level 1 rule. As is shown under (67) below, san-ity1 is pieced together at level 1, hence shortening will apply. By contrast, level 2 -hood2 has not yet been attached to maiden when shortening applies; therefore maiden remains unmodified: the trisyllabic condition is not met. A third example, less controversial than Trisyllabic Shortening, is nasal assimilation in the two prefixes un- and in-: the nasal of the latter takes on the place of articulation of the following obstruent (im-possible, i[ŋ]-credible), while the nasal of the former remains (or may remain for certain speakers) unaltered (u[n]-predictable, u[n]-comfortable).

42

Trisyllabic Shortening encounters quite a number of counterexamples such as obese [çwbiis] - obese-ness [çwbiisnEs] (class 2), which should but does not react when the class 1 suffix -ity is added: obes-ity [çwbiisitɪ]. A given root may even produce reacting items along with derivatives that remain unaffected: wild-ness [wajldnEs] and wilderness [wIldånɛs] bear the same class 2 suffix but show contrasting behaviour. Also, Trisyllabic Shortening does not appear to be productive, and additional doubt has been cast on its synchronic reality by psycho-linguistic evidence. Hayes (1995a) and Green (2007:172ff) provide an informed review of the status of Trisyllabic Shortening today. Relevant literature that is concerned with the phenomenon since the 70s is mentioned in note 28.

Rule-blocking boundaries are eliminated altogether 143 (67)

Trisyllabic Shortening in Lexical Phonology san-ity maiden-hood lexicon sejn mejd´n level 1 concatenation — sejn-ɪtɪ Trisyll. Short. — sQn-ɪtɪ level 2 concatenation — mejd´n-hʊd rule application — —

Independent diagnostics such as the ability to attach to bound stems identify in- as a class 1 affix, while un- is class 2 (in-ert, in-trepid, but *un-ert, *un-trepid etc., see §142). In terms of SPE, as before, the # boundary that comes with class 2 un- blocks the application of the nasal assimilation rule. In Lexical Phonology, on the other hand, nasal assimilation is a level 1 rule. The contrastive behaviour of the two prefixes follows, as is shown under (68) below. (68)

nasal assimilation in Lexical Phonology im-possible lexicon possible level 1 concatenation in-possible nasal assimilation im-possible level 2 concatenation — rule application —

un-predictable predictable — — un-predictable —

The logic is as before: level 2 affixes escape whatever the process at hand (which is thus "blocked") because this process is not active anymore by the time they are concatenated. Level 1 phonology underapplies to class 2 strings. In sum, rule-blocking boundaries translate as level 1 rules in a stratal environment.

165

6.2. Complete and unintended elimination of boundaries The previous section has shown that the stratal account of rule-blocking boundaries does not mention any boundary anymore. The boundaryeliminating effect of the stratal architecture is clearly identified in the literature: Mohanan (1982:24f, 94), Kiparsky (1982a:11, 1982b:131), Halle & Mohanan (1985:64), Szpyra (1989:24, 27) and Mohanan (1986) are explicit on the fact that boundaries are completely banned from phonological rules in Lexical Phonology..

144 Chap 7: Lexical Phonology (69)

"In Lexical Phonology, boundary symbols are replaced by references to the beginning and end of forms, using morphological bracketing, and no special conventions are needed for blocking the application of rules." Mohanan (1986:20) "Boundary symbols such as + and # can be entirely eliminated from phonological representations. The requisite information is carried by the appropriate ordering of levels and the morphological bracketing of the string." Kiparsky (1982b:139)

The quote below shows that the eradication of boundaries was indeed understood as an unintended side effect, rather than as a goal of the theory. (70)

"Originally postulated in order to account for morphological distribution, the conception of lexical strata also yields a way of dealing with morphological information in phonology. SPE makes use of boundary symbols like +, # and ## to refer to morphological information. Instead of using such symbols, the phonological rules in Lexical Phonology (a) refer to the beginning and the end of morphological forms, and (b) are specified for their domain of application in terms of lexical strata." Mohanan (1986:18)

In the last part of the quote, Mohanan points out that the elimination of boundaries is achieved at the expense of two new devices: the diacritic specification of every rule for the level(s) at which it applies (domain assignment, §151), and edge-indicating brackets (interestingly, no mention is made of the split of phonology into multiple mini-grammars §148, and of level ordering §147). Brackets as used in Lexical Phonology (and bracket erasure) are introduced on the following pages on the occasion of the discussion of ruletriggering boundaries.

166

7. Rule-triggering boundaries: brackets and bracket erasure

167

7.1. English nasal cluster simplification Let us now consider rule-triggering boundaries and their fate in Lexical Phonology. Recall from §51 that a rule-triggering boundary represents a specific morpho-syntactic division which triggers the application of a phonological process that would not have gone into effect in absence of this division.

Rule-triggering boundaries: brackets and bracket erasure 145

English features three processes of this kind, all of which control the simplification of a stem-final cluster that contains a nasal (see §94). In all cases, the cluster is simplified word-finally and before class 2 suffixes, but survives before class 1 suffixes and in morpheme-internal position: gn-n (sign, sign-ing2 vs. sing-ature1, ignore), mn-n (damn, damn-ing2 vs. damn-ation1, amnesia), mb/ŋg-m/ŋ (sing, sing-ing2 vs. long-er1, finger, bomb, bomb-ing vs. bomb-ard, Hamburg). The latter alternation, postnasal plosive deletion, suffers from a number of exceptions.43 The following discussion therefore focuses on the former two cases. Data and analysis under (71) below are taken from Mohanan (1986:21ff) (the facts are also discussed by, among many others, Halle & Mohanan 1985:95f, Borowsky 1986:232ff). Mohanan (1986:21ff) points out that the absence of word-final [gn, gm, mn] is certainly not due to syllabic reasons since parallel word-final clusters such as [-zmÿ ], [-klÿ], [-sn̩] do occur with a syllabic sonorant in very common words such as prism, pickle, listen. There is no reason why column and sign could not be colu[mn̩] and si[gn̩], respectively. Mohanan also bolsters the synchronic reality of the alternations by showing that they are productive: when asked to produce the infinitive and the -ing form of nonce words like limnation [lImnejS´n], native speakers return [lIm], [lImIN], rather than [lImn], [lImnIN]. Nasal cluster simplification illustrates the pattern of rule-triggering boundaries since the underlying form must contain a cluster (/gn/, /gm/ and /mn/), which is simplified by a phonological process when class 2 suffixes are attached. That is, class 2 suffixes trigger cluster simplification where class 1 suffixes are inert.

43

For one thing, some class 1 affixes such as -ology and -ese trigger deletion: bomb-ólogy, Peking-ése. Their class membership is evidenced by the fact that they shift stress (as shown by the examples) and are able to combine with bound stems as in phren-ology, Portugu-ese (see §142 for the bound stem diagnostic). Also, a number of words bear [ŋ] in morpheme-internal position: dinghy, hangar, Birmingham. Bermúdez-Otero (2008, forth a) shows that the two aspects in which post-nasal plosive deletion misbehaves are related along the lines of a generalisation that was established by Chung (1983:63). Based on evidence from explicit statements by 18th century orthoepist James Elphinston, Bermúdez-Otero (forth b) reconstructs the evolution of /ng/ simplification, showing that it crept into the language "from outside-in", i.e. affecting first larger, then smaller chunks (phrase level, word level, stem level).

146 Chap 7: Lexical Phonology A further challenge for analyses is due to the fact that the word-final context systematically behaves like class 2 suffixes: it triggers simplification. The relevant rule must thus somehow make reference to the stringfinal location as a condition for the process to go into effect. This is confirmed by the fact that morpheme-internal /gn/ and /mn/ do not simplify: a[gn]ostic, a[mn]esia. (71)

English nasal cluster simplification a. gN - n __# class 2 suffixes n/m n/m sign sign-ing resign resign-ed assign assign-ment design design-ed, design-s malign malign-ing, malign-ed benign – paradigm – b. mn - m __# m solemn damn condemn hymn

class 2 suffixes m – damn-ing condemn-ing hymn-ing, hymn-ed

column autumn

column-s, column-ed –

__-V class 1 suffixes gn/gm sign-ature, sign-al, sign-ify resign-ation assign-ation design-ate malign-ant, malign-ity benign-ity, benign-ant paradigm-atic __-V class 1 suffixes mn solemn-ity damn-ation condemn-ation hymn-al, hymn-ology, hymn-ary, hymn-ic column-al autumn-al

Recall from §94 that the disjunction "word-finally and before class 2 suffixes" is reduced to # in SPE, which is what the deletion rule simply makes reference to: g → ø / __N#. This rule will not apply to clusters before class 1 suffixes because these come with a +, rather than with a # boundary.

Rule-triggering boundaries: brackets and bracket erasure 147 168

7.2. Brackets and bracket erasure are needed for a stratal account The stratal architecture alone cannot account for the rule-triggering pattern: in our examples, the deletion rule must be present at level 2 in order to touch strings that bear class 2 suffixes (e.g. sign-ing). Since all strings that are present at a given stratum must run through all subsequent strata on their way to the surface, however, strings that bear class 1 affixes (e.g. sign-ature) would also have to undergo deletion at level 2, which is not what they do. Some additional device is thus needed in order for Lexical Phonology to be able to account for the rule-triggering pattern. This is when the cycle-delineating brackets known from SPE (see §95) enter the scene. As we will see, though, the use that Lexical Phonology makes of brackets and bracket erasure is only remotely akin with the devices of the same name that were introduced in SPE. The analysis of the rule-triggering pattern in terms of brackets and bracket erasure is due to Mohanan (1982:24f) (see also Mohanan & Mohanan 1984, Halle & Mohanan 1985, Mohanan 1986). In Lexical Phonology, it has become the consensual analysis of the phenomenon at hand. We will see in §203 that Kiparsky's (1982a,b) Strict Cycle Condition (SCC) is a direct competitor that does the same job. For the time being, though, let us introduce Mohanan's mainstream analysis. Mohanan (1986) explains the distribution and function of brackets as follows. (72) "We use the the notation of brackets to refer to morphological concatenation: [ __ refers to the beginning of a form, __ ] to the end, and ] [ __ or __ ] [ to the junction between two forms (the end of a form followed by the beginning of another)." Mohanan (1986:18)

Brackets thus delineate morpheme boundaries, and the structural description of phonological rules makes direct reference to them. Nasal cluster simplification occurs in word-final position (sign) and before class 2 affixes (sign-ing). It was mentioned in §167 that SPE captured this disjunction with the # boundary, which follows the cluster in both environments (§94). On Mohanan's analysis, the two cluster-reducing contexts share the property of being morpheme-final. Since morpheme-boundary information is represented by brackets, the relevant rules thus need to go into effect before a (closing) bracket. They appear under (73) below (from Mohanan 1986:22).

148 Chap 7: Lexical Phonology (73) a. g deletion, domain: level 2 g → ø / __ [+nasal] ] b. n deletion, domain: level 2 n → ø / [+nasal] __ ]

Deletion thus takes place in word-final /mn/ and /gn/, which are followed by a bracket: /[damn], [sign]/ → damn, sign. However, brackets are unable to distinguish between class 1 and class 2 suffixes: [[sign] [ing]] and [[sign] [ature]] should both lose their /g/. The distinction between affix classes therefore needs to be assured by some other means. This job is done by bracket erasure, as described under (74) below.44 (74)

Bracket Erasure Erase internal brackets at the end of each level.

Table (75) below shows how the derivation of nasal cluster simplification works when brackets and bracket erasure are in place. Note that ruletriggering boundaries translate as level 2 rules in the stratal environment. (75)

44

nasal cluster simplification in Lexical Phonology underived class 2 damn damn-ing lexicon [damn] [damn] level 1 concatenation – – bracket erasure – – level 2 concatenation – [[damn] [ing]] cluster simplification [dam] [[dam] [ing]] bracket erasure – [dam ing]

class 1 damn-ation [damn] [[damn] [ation]] [damn ation] – – –

The version of bracket erasure under (74) is Mohanan's (1982) original take which he called the Opacity Principle, and which was taken over by Kiparsky (1982b:140). Mohanan's (1982:23) original formulation is as follows: "the internal structure at one stratum is invisible to the processes of another." In later work, Mohanan (1986:25) goes back to the stronger version of SPE where brackets are erased at the end of each cycle (instead of each stratum only, Mohanan 1986:59f, note 7 explains why). This move is based on the option of distinguishing between cyclic and non-cyclic strata (Mohanan & Mohanan 1984, Halle & Mohanan 1985, see §194) and relies on an idiosyncratic definition of what counts as a cycle (which is different from the definition that is in place in SPE).

Rule-triggering boundaries: brackets and bracket erasure 149

By way of bracket erasure, [[damn] [ation]] becomes [damn ation] at the end of level 1. It is handed down as such to level 2, where cluster simplification is active. However, the elimination of the inner brackets bleeds the deletion rule, which applies only to clusters that are followed by a bracket. Hence [damn ation] is left undamaged, just like mono-morphemic items (amnesia). By contrast, /-mn/ before class 2 suffixes is always followed by a "fresh" bracket since class 2 suffixes only join in at level 2: [[damn] [ing]] still bears inner brackets when cluster simplification applies. In the same way, /-mn/ in underived stems is always followed by a bracket, which is peripheral and therefore never erased: [damn] arrives at level 2 with the critical bracket following the cluster, which is therefore simplified. As a result, [damn] and [[damn] [ing]] lose their nasal, but [damn ation] and [amnesia] do not. That is, nasal cluster simplification underapplies to class 1 strings. 169

7.3. A hybrid representational-procedural theory because of the ruletriggering pattern

170

7.3.1. LP-style brackets undo what was gained by interactionism It was mentioned in §161 that interactionism and the stratal architecture have the beneficial effect of doing away with the modularity-violating brackets of SPE (see also §§161,305,680). Rather than by brackets (which are untranslated morpho-syntactic information in phonology), interpretational units are now identified by strata. Against this backdrop, though, Mohanan's reintroduction of brackets into the stratal architecture destroys the gain that interactionism had achieved. Lexical Phonology with brackets accommodates the old awkward representational delineation of morpho-syntactic structure on top of the procedural (stratal) definition of interpretational units. Or, in other words, Lexical Phonology holds that phonologically relevant morpho-syntactic divisions do not coincide with interpretational units: the latter, strata, are subdivided into the former, morphemes. That is, origin-al-ity is only one interpretational unit (a root plus two class 1 affixes that are concatenated at the same stratum), but accommodates internal structure through brackets: [[origin] [al] [ity]].

150 Chap 7: Lexical Phonology This is at variance with SPE, where brackets and interpretational units coincided: every chunk delineated by brackets – a cycle – was subject to independent interpretation, and vice-versa. 171

7.3.2. LP-style brackets and brackets in SPE have got nothing in common Let us now look at how LP-style brackets relate to their ancestors in SPE. In fact, they have barely anything in common beyond the name. The more the original literature is removed from the observer's perspective, the easier they may therefore be confused. We have already seen that brackets in Lexical Phonology can be mentioned in phonological rules (this is actually the reason why they exist in the first place, see (73)). In SPE they cannot: their exclusive purpose is to guide the cyclic derivation through the linear string (§98). We have also seen that precisely this function – the definition of interpretational units – is not what brackets do in Lexical Phonology: interpretational units are defined by procedurally ordered strata. Their labour is simply to indicate morphological divisions. This is, then, another difference with respect to SPE. Recall from §§103,160 that it is not true that brackets delineate all morphological divisions in SPE: only those morpheme boundaries are armed with brackets which represent a transition between major categories. Hence the structure of [[[theatr]N ic + al]A i + ty]N. Mohanan's brackets, however, mark each and every morpheme (see the quote in (72)). Originality will thus be represented as [[origin] [al] [ity]] where two class 1 morphemes cohabitate with a root in the same interpretational unit (stratum 1). Finally, brackets in SPE make the hierarchical morpho-syntactic structure available in the phonology: brackets reproduce morpheme breaks and the embedded status of morphemes. By contrast, Lexical Phonology does not care for the hierarchical status of morphemes (because this information is already procedurally encoded in strata): all that Mohanan's brackets do is to indicate that there is a morpheme boundary. For example, Mohanan's structure of theatricality (supposing the same morphological analysis as in SPE) is flat: [[theatr] [ic] [al] [i] [ty]].

Rule-triggering boundaries: brackets and bracket erasure 151 172

7.3.3. Brackets are boundaries that are introduced through the back door LP-style brackets appear to be an attempt to reintroduce boundaries through the back door. Recall from §165 that the elimination of boundaries is advertised as an achievement of Lexical Phonology: it would be difficult to concede that they are still needed for the analysis of the rule-triggering pattern. A theory that combines procedural and representational elements for the management of affix class-based phenomena is not really what one may expect given the research programme of Lexical Phonology that has set out to replace representational by procedural solutions. If representational boundaries are needed anyway and can do all the job, what is the procedural technology good for? Despite the attempts to camouflage the representational travesty, stratal Lexical Phonology is in fact a hybrid representational-procedural theory. Bermúdez-Otero (2008) calls the use of brackets a cheap trick. He points out that just like SPE, Lexical Phonology needs a trigger for the ruletriggering pattern. The only difference with respect to the analysis in SPE, then, is the name of the trigger: a boundary here (g → ø / __N#), a bracket there (g → ø / __N]). Recall that the difference between boundaries and brackets in SPE was precisely the ability of rules to refer to the former, but not to the latter. The bracket in Mohanan's rule is thus no more than a boundary in rhetorical disguise.45

173

7.3.4. A hybrid representational-procedural theory Mohanan's brackets, however, do their job less well than boundaries in SPE. Recall from §§94,167 that the rule-triggering pattern in English raises the challenge of the disjunction "word-finally and before class 2 suffixes" – this is where nasal clusters are simplified. The disjunction is reduced to # in SPE, which is what the deletion rule makes reference to. We have seen in §168 that hybrid Lexical Phonology cannot capture this disjunction even when equipped with Mohanan's brackets. This is because there is only one kind of bracket: unlike boundaries in SPE, which fall into + and #, brackets are unable to distinguish class 1 and class 2 suffixes – both bear the same bracket. The rule g → ø / __N] will thus incorrectly delete the velar also in [[sign] [ature]]. This problem is solved by the procedural architecture (and 45

Kaisse & Shaw (1985:10ff) and Halle & Mohanan (1985:59, 64f) provide more comparative discussion regarding boundaries and brackets.

152 Chap 7: Lexical Phonology bracket erasure, to be further discussed shortly), which is able to distinguish class 1 ("old") from class 2 ("fresh") brackets. In other words, Lexical Phonology has split the information which in SPE is supported by the # boundary alone into a representational and a procedural piece: the location of a morphological division is signalled by a bracket, but whether this particular division is phonologically relevant or not, and if so, whether it has a blocking or a triggering effect, is controlled procedurally by level ordering. This fact is frankly acknowledged by Mohanan (1982). (76)

"In SPE, the use of boundary symbols serves two functions: (i) If a rule contains a boundary symbol such as + or # in its structural description, it applies to a string only if the string contains the required boundary. (17) shows how this function of the boundary symbol is taken care of in Lexical Phonology by defining the domain of the rule as the stratum associated with the boundary, and replacing the boundary with brackets." Mohanan (1982:24)

Finally, another aspect of brackets is that they make Lexical Phonology a representative of the non-privative take of SPE (§90). Brackets are the representational means of Lexical Phonology to refer to morphosyntactic information in phonology. Since they do not reflect the hierarchical aspect of morpho-syntactic structure, one must assume that they are the output of translation (or at least of readjustment: this question is not addressed in the Lexical Phonology literature). In these non-hierarchical limits, then, brackets transmit full morpho-syntactic information, rather than just the information that is phonologically relevant. 174

7.4. Bracket erasure

175

7.4.1. No look-back devices in Lexical Phonology and elsewhere Let us now have a look at bracket erasure and the labour that it does in the analysis of the rule-triggering pattern. Bracket erasure destroys the morphological structure of a string before handing it down to the next stratum where further morphemes are concatenated. When the phonological computation of a later stratum applies, then, the original morphological structure of the "old" string is invisible: a string made of several "old" morphemes is treated exactly like a morphologically non-complex item.

Rule-triggering boundaries: brackets and bracket erasure 153

This means that at any given point of morphological construction, phonological computation cannot "look back" into previous strata – it sees only those morphological divisions that were created in its own stratum. This is the first time that we come across what I call a no look-back device in this book. We will see that no look-back devices are accommodated in various theories and occur in a number of flavours. The ancestor of no look-back devices is Chomsky's (1973) Strict Cycle Condition. Various theories have implemented different brands of no look-back devices over the years, including namely Kiparsky's SCC-K, an important brick of Lexical Phonology (on which more in §188). A survey of no look-back devices is provided in §287. In Lexical Phonology, bracket erasure is often advertised as a major insight (e.g. Kaisse & Shaw 1985:4f, Mohanan 1986:23ff, Bermúdez-Otero forth a:45ff): it is held to be the root of Praguian segregation (§153). That is, word- and sentence phonology are different because phonological rules which apply across words have no access to word-internal structure, which has been erased. Finally, due mention needs to be made of the fact that bracket erasure has a competitor in Lexical Phonology which does the same no look-back job: Kiparsky's (1982a,b) adaptation of the Strict Cycle Condition. Comparative discussion is provided in §203. 176

7.4.2. SPE has bracket erasure, but no no look-back effect Note that just like Mohanan's brackets, bracket erasure in Lexical Phonology shares nothing but the name with the device of the same name that is known from SPE. Unfortunately, the constant reference of the Lexical Phonology literature to the SPE-ancestor in this context contributes a lot to terminological and general confusion. Chomsky & Halle (1968:15) have indeed coined the term bracket erasure, but this device does not produce any no look-back effect. In SPE, bracket erasure effects the deletion of the brackets that delineate a cycle (i.e. an interpretational unit) at the outset of the next cycle (see §100). Hence the structure [[A] B] is processed in such a way that first phonology applies to [A]; then the next cycle, [[A] B], is considered, but before phonological rules apply, inner brackets are erased: the input to the phonological computation is [A B]. Bracket erasure itself is thus identical in SPE and Lexical Phonology (see (74), with the proviso regarding the different distribution of brackets, see §171).

154 Chap 7: Lexical Phonology However, recall that brackets serve only one purpose in SPE, i.e. the delineation of cyclic domains and hence the application of cyclic derivation. Phonological rules can make reference to boundaries, but crucially not to brackets (§§98,170f). This is why brackets and bracket erasure are unable to produce a no look-back effect in SPE: in order for phonological computation to be unable to look back to erased brackets, it would have to be able to see brackets in the first place. This, however, is impossible: brackets may not be mentioned in phonological rules. Hence it is only because Mohanan allows rules to make reference to brackets that a no look-back effect is produced in Lexical Phonology on the grounds of the same bracket erasure convention that was active in SPE.

177

8. Derived environment effects

178

8.1. Properties and illustration of the phenomenon

179

8.1.1. Phonologically and morphologically derived environments Lexical Phonology has coined the notion of so-called derived environment effects (e.g. Kiparsky 1982b, Rubach 1985).46 Processes that fall into this category apply only to strings that have already undergone some phonological or morphological operation; underived items on the other hand remain unaffected. On the classical definition (e.g. Kiparsky 1982b, Rubach 1985:157), an item counts as derived iff it is either morphologically complex (that is, if some concatenation has taken place) or the result of the application of a phonological rule.

180

8.1.2. The foundational Finnish case The canonical example for a phonologically derived environment is Kiparsky's (1973a,b) analysis of Finnish where underlying /t/ surfaces as [s] if followed by a "derived" i. Table (77) shows relevant data.

46

Much like the pages below, Bermúdez-Otero (forth a:§1.2) traces back the history of derived environment effects, and of Kiparsky's Strict Cycle Condition.

Derived environment effects 155

(77a) shows that /t/ is affected by a following -i if both segments are separated by a morpheme boundary: this is a morphologically derived environment. The -i in the examples shown is the past tense marker, and the underlying stem-final /t/ is witnessed by morphemes such as the 3rd person impers. marker -koon, which produces /halut-koon/ → halutkoon "to want 3rd impers." (compare with /halut-i/ → halusi "wanted"). Underlying /t/ also undergoes the t → s rule under (77b), i.e. in case the following i is an underlying /e/ that was raised in word-final position by way of an independent rule (the /e/ is witnessed by the essive forms where it escapes raising because it is not word-final). (77)

Finnish t → s / __i in derived environments a. morphologically derived: the -i belongs to a different morpheme /halut-i/ → halusi wanted /hakkat-i/ → hakkasi hewed /turpot-i/ → turposi swelled b. phonologically derived: the i is an underlying /-e/: e → i / __# essive nominative vete-nä vesi water käte-nä käsi hand mete-nä mesi honey c. underived environments: stem-internal /i/ essive nominative neiti-nä neiti young lady äiti-nä äiti mother koti-na koti home

Finally, (77c) shows that /t/ remains unaffected by a following i if this i is really underlying (i.e. not the result of a phonological process) and does not belong to a different morpheme.

181

8.1.3. Non-application of rules to mono-morphemic strings It appears that phonologically derived environments have received far less attention and empirical support than morphologically derived environments.47 One reason is certainly the fact that the application of a rule is much subject to analysis and hence to debate, while the existence of a morphological boundary is a hard fact. In a constraint-based environment for 47

Some more discussion of phonologically derived environment effects is provided in §517.

156 Chap 7: Lexical Phonology example where computation is parallel and hence no ordered phonological events exist, the entire notion of phonologically derived environment is obsolete anyway, or at least unstateable. On the other hand, effects of morphologically derived environments have been reported regularly. To date the discovery of this phenomenon and its formalisation is taken to be a major, for some the major contribution of Lexical Phonology: "The lasting result of Lexical Phonology is that phonological generalizations may be bound by derived environments" Rubach & Booij (2003:446), "One important aspect of the phonology-morphology interface which has been discovered in the second half of the twentieth century" van Oostendorp (2007:135). Typically, sensitivity to morphologically derived environments is displayed by segmental rules, i.e. where a trigger and a target are clearly identifiable in the linear string, and some melodic property transmitted (which is not the case for prosodic phenomena such as stress placement). An example is the palatalisation in Polish discussed by Rubach (1984:59ff) and Rubach & Booij (1984:3f). In Polish, a certain kind of palatalisation targets preceding dental consonants, but only if the palatal trigger and the dental patient belong to two different morphemes. This is shown under (78) below.48 (78)

48

Polish palatalisation dental with no palatal agent spelling [] a. derived gwut g»ód grymas grɨmas b. underived dinosaur dinɔsawr protest prɔtEst desant dEsant c. both servis sɛrvis

dental + palatal agent spelling [] głodz-ić gwɔdɸ-itɲ grymas-ić grɨma˛-itɲ

servis-ie

sɛrvi˛-ɛ

Glosses (left-to-right, top-down): "hunger, to starve somebody, grimace, to grouch, dinosaur, protest, landing". All cases where palatalisation does not apply morpheme-internally are recent loans; concomitantly, there are no native words with morpheme-internal [di] or [si]. Hence one could be tempted to regard the pattern as a mere problem of loanword adaptation: loans are borrowed without amendment. However, Rubach & Booij point out that loanwords regularly inflect and palatalise in the relevant environment (e.g. protest - protesc-ie [prɔtɛstɲ-ɛ] "protest, id. LOCsg"), which is not an argument in favour of this perspective.

Derived environment effects 157

As may be seen under (78a), stem-final [s,d] turn into [˛,dɸ] before the infinitive morpheme [-itɲ]. By contrast, (78b) shows that no palatalisation occurs in the same phonological environment if the triggering vowel and the dental target belong to the same morpheme. The word under (78c) demonstrates both effects on the same lexical item: the underlying sequence /se/ shows no effect when it is mono-morphemic, but the fricative palatalises before the LOCsg marker -ie [-ɛ]. Different analyses of this pattern are discussed in §§188ff below.

182

8.1.4. Sensitivity to derived environments and to affix classes is orthogonal Another aspect of derived environment effects is that they are entirely orthogonal to affix class-sensitivity. Trisyllabic Shortening for example is a process which combines the derived environment restriction (it only applies to plurimorphemic strings: nightingale and ivory remain unaffected, more on this shortly) with a selective application to a particular affix class (class 1: s[aj]ne - s[æ]n-ity1 vs. m[aj]den-hood2, see §164). This, however, is a mere coincidence. Other affix class-sensitive rules also apply to mono-morphemic strings. For example, just like Trisyllabic Shortening, nasal assimilation is a level 1 rule (see §164): the class 1 affix in- reacts (im-possible), while the class 2 affix un- does not (un-predictable). Unlike Trisyllabic Shortening, however, nasal assimilation is active in mono-morphemic strings: all mono-morphemic nasalobstruent clusters are homorganic in English (climb, hint, finger, etc.). Trisyllabic Shortening and nasal assimilation thus share the selective application to strings that are created by class 1 affixation, but go separate ways regarding the derived environment parameter.

183

8.2. Derived environments are an offspring of the abstractness debate

184

8.2.1. Absolute neutralisation and free rides The idea that phonological processes do not apply within morphemes, i.e. when the trigger and the target belong to the same morpheme, has emerged from the abstractness debate of the 70s (§§125f) and Kiparsky's continuous attempts to restrict the variability of underlying forms (§129). Cole (1995) and Bermúdez-Otero (forth a:§1.2) provide a good overview of this move-

158 Chap 7: Lexical Phonology ment, which is also explained by Kiparsky himself in Kiparsky (1982b, 1993), and by Anderson (1981:530ff). In SPE, underlying forms could be at any distance of what they sound like on the surface; whatever suited the analyst in order to get things right was allowed. Namely, the underlying form of morphemes that do not alternate at all could be different from their surface form. For example, mono-morphemic items such as n[aj]ghtingale (nightingale) and [aj]vory (ivory) systematically resist Trisyllabic Shortening even though they satisfy its structural description (§164). SPE reacts in a non-systematic way that misses the obvious morphological generalisation. That is, the application of the rule is eluded simply by destroying either the target or the triggering context of each individual lexical item: instead of /aj/, nightingale was said to have an underlying /i/, i.e. /nixtVngael/ (Chomsky & Halle 1968:234); and instead of /i/, the last segment of ivory was made a glide, i.e. /ivorj/ (Chomsky & Halle 1968:181, which makes ivory miss the trisyllabic condition). Independent rules that are ordered after Trisyllabic Shortening, then take /ix/ to [aj] (via /i/), and vocalise the final glide of ivory. Another trouble when the underlying form of invariable morphemes may be distinct from their surface form are so-called free rides. In our example, these concern the converse surface situation, i.e. cases where the third but last vowel of a mono-morphemic item is short. That is, Kiparsky (1982b:148) points out that on the SPE analysis, the underlying form of [æ]libi, c[æ]mera or P[æ]mela cannot be determined: the third but last vowel of these words could either be faithfully short and hence appear as such on the surface, or it could be underlyingly long, i.e. /aj/libi, /kaj/mera, /paj/mela, in which case the /aj/ will be shortened into [æ] via Trisyllabic Shortening – a free ride, i.e. one without consequences. In the same way, nasal assimilation (§164) can give rise to free rides. English has no non-homorganic [NC] clusters at all. The nasal assimilation that is observed in i[m]-possible, i[ŋ]-credible, then, can also be said to be responsible for mono-morphemic [NC] clusters: lamp and think may be construed as /lænp/ and /Tɪnk/, which then take a free ride on nasal assimilation in order to surface with a homorganic cluster. 185

8.2.2. The Alternation Condition Kiparsky (1968-73:14f) coins the term absolute neutralisation, which describes a situation where a morpheme has an underlying sequence that may

Derived environment effects 159

never be inspected on the surface because it is modified by a rule in all occurrences. That is, the situation of nightingale and ivory on the analysis of SPE. Opposed to absolute neutralisation are contextual neutralisations, which transform only a subset of the occurrences of a morpheme. For example, /in-/ appears with a modified nasal in i[m]-possible and i[ŋ]-credible, but surfaces as such in i[n]-applicable where the quality of the nasal may not be said to be the result of assimilation since the consonantal trigger is missing. On this backdrop, Kiparsky argues that absolute neutralisation and free rides must be disallowed on principled grounds: grammar cannot tolerate underlying forms of non-alternating morphemes which are different from their surface form. This is done at first by the Alternation Condition (Kiparsky 1968-73:14ff), which states that "obligatory neutralization rules cannot apply to all occurrences of a morpheme" in Kiparsky's (1982b:148) formulation. 186

8.2.3. "Low-level", "automatic" or "phonetic" rules may apply morphemeinternally What the Alternation Condition amounts to is the insight that phonological rules never apply inside morphemes or, conversely, that phonology only applies across morpheme boundaries. The question is whether this statement is empirically true in its overarching and somewhat brutal generality. Quite unsurprisingly, the answer is no: there are many cases where an underlying mono-morphemic string is transformed by phonological computation. This is recognised by Kiparsky (1968-73:18) and Kiparsky (1973a:64), who concedes that "if a form appears in a constant shape, its underlying form is that shape, except for what can be attributed to lowlevel, automatic phonetic processes" (Kiparsky 1968-73:18). The trouble, then, is to tell "real" phonological rules from low-level, automatic and phonetic processes. Kiparsky (1968-73:18) writes that "these can be defined as processes which do not cause neutralization of distinct representations" and quotes the great vowel shift together with the loss of /g/ in sing [sɪŋ] as examples. In Kiparsky (1982b:154), he adds velar softening (electri[k] electri[s]-ity, see §126, note 28). "Automatic" aspiration of English voiceless plosives (roughly word-initially and before stressed vowels) would certainly be another candidate, as much as the distribution of suprasegmen-

160 Chap 7: Lexical Phonology tal properties such as stress, which always concerns underived roots as much as plurimorphemic strings. The question whether it can be predicted which rules obey and which rules disregard the prohibition to apply to mono-morphemic strings will escort the reader on the pages below. Having run Kiparsky's (1968-73, 1973a) proposal against a more substantial set of empirical evidence, Kenstowicz & Kisseberth (1977) conclude, pessimistically, that (79)

"Kiparsky's principle may be too strong in that some rules of nonautomatic neutralization apply in nonderived contexts. If so, it is not immediately clear that there is a way to predict which rules will apply only in derived contexts and which will apply in nonderived contexts as well." Kenstowicz & Kisseberth (1977:214)

Halle (1978) agrees. (80)

"To conclude it would appear that the special conditions on rules discovered by Kiparsky are unconnected with the automatic or nonautomatic character of the rule." Halle (1978:132)

We will see below that the same class of "low level", "automatic" or "phonetic" processes – which may therefore be called the bad guys – also violates some versions of no look-back devices that have got nothing to do with derived environments (§315). 187

8.2.4. The Revised Alternation Condition: derived environments enter the scene The definition of what exactly is low-level, phonetic and automatic will haunt Lexical Phonology all through, as will be evident from the discussion below. For the time being, with this question blinded out, the derived environment condition emerges. For a number of reasons that Kiparsky (1982b:148ff) explains at length, he upgrades the Alternation Condition by building on derived environments. The Revised Alternation Condition (RAC), then, is as under (81) below. (81)

Revised Alternation Condition (RAC) Obligatory neutralization rules apply only in derived environments. Kiparsky (1982b:152)

Derived environment effects 161

As before, mono-morphemic strings may thus not be affected by the kind of rules that are non low-level, non-automatic and nonphonetic. Given the RAC, Trisyllabic Shortening for example cannot apply to nightingale and ivory because the strings at hand are underived.

188

8.3. Solution 1: the Strict Cycle Condition (SCC)

189

8.3.1. Mascaró's SCC has got nothing to do with derived environments After having introduced the Revised Alternation Condition, Kiparsky (1982b:153ff) goes on to advertise Mascaró's (1976) Strict Cycle Condition (SCC) ("a major step forward"). According to Kiparsky, the advance offered by Mascaró is the identification of the location in the grammar where the two types of rules live, i.e. "real" rules that apply only in derived environments and "low-level" rules that are automatic. Mascaró proposes that the former are cyclic (in the sense of SPE, see §105), and Kiparsky concludes that the latter must then be non-cyclic, which in terms of Lexical Phonology means postlexical. Mascaró's (1976) SCC, however, has got nothing to do with derived environments at all: like Kean (1974), Mascaró works at importing Chomsky's (1973) Strict Cycle Condition (see §289) into phonology – something that he calls the Phonological Cycle, and which is known as such since then (SPE only talked about the Transformational Cycle, see §100). In adaptation to syntactic strict cyclicity, then, Mascaró's Phonological Cycle is the introduction of a no look-back device into phonology. This notion was introduced in §174 and is further discussed in §287. Mascaró's (1976) SCC appears under (82) below. (82)

"Strict Cycle Condition (SCC) Proper application of cyclic rules For a cyclic rule to apply properly in any given cycle j, it must make specific use of information proper to (i.e. introduced by virtue of) cycle j." Mascaró (1976:7)

162 Chap 7: Lexical Phonology Mascaró (1976) explains that (83)

"this condition ensures that no 'improper' cyclic application, that is, multiple application of a rule, opposite rule ordering, etc. on the same cycle results. In other words, it makes it impossible for rules to 'return to earlier stages of the cycle after the derivation has moved to larger, more inclusive domains' (Chomsky, (1973), 243)." Mascaró (1976:7f)

Multiple application of a rule to a given string in two different cycles is thus prohibited by the requirement for rules to use at least some material that was added on the outer cycle. Hence given [[X]i Y]j, a cyclic rule may apply first to X on the inner cycle i (it takes into account material that was introduced on this cycle); however, on the outer cycle it may apply to [XY] j again only if it takes into account some material of Y. Mascaró's SCC is thus perfectly unable to account for derived environment effects: the only thing that it does is to introduce a no look-back device into phonology. For example, SCC does not prevent Trisyllabic Shortening from applying to mono-morphemic items such as nightingale and ivory at all: on the innermost cycle, i.e. [nightingale], the rule uses material of this cycle and hence applies "properly". Just as much as in [[san] ity] where the material of the outer cycle is needed in order to satisfy the structural description.

190

8.3.2. Kiparsky (in fact Halle) adds derived environments to Mascaró's SCC Let us now look at how Kiparsky (1982a,b) introduces Mascaró's SCC. He writes that (84)

"with some simplification, his [Mascaró's] proposal was: (47) Strict Cycle Condition (SCC): a. Cyclic rules apply only to derived representations. b. Def.: A representation φ is derived w.r.t. rule R in cycle j iff φ meets the structural analysis of R by virtue of a combination of morphemes introduced in cycle j or the application of a phonological rule in cycle j." Kiparsky (1982b:153f)

Derived environments have thus appeared overnight in what Kiparsky sells as Mascaró's slightly "simplified" Strict Cycle Condition. Also,

Derived environment effects 163

the notion of phonologically derived environment has joined in where Mascaró was only talking about morphological conditions. Mascaró faithfully applies Chomsky's (1973) idea that a rule can only apply if it uses material that was freshly introduced on the latest cycle. It was shown in the previous section that this has got nothing to do with derived environments. Cole (1995:72) correctly points out that Kiparsky's SCC does two things which are logically unrelated: it restricts the application of cyclic rules to derived environments under (84a) and encodes Chomsky's (1973) no look-back device under (84b). The former continuates Kiparsky's Revised Alternation Condition, while the latter introduces the no look-back idea (see also Iverson & Wheeler 1988 on this distinction). In actual fact, however, the fusion of Kiparsky's concern for derived environment effects on the one hand, and of Chomsky's Strict Cycle Condition (in Kean's 1974 and Mascaró's 1976 guise) on the other, was operated by Morris Halle (1978:129ff). Halle writes: (85)

"the version of the constraint on cyclic rule application that I propose below is a combination of certain suggestions made by Kiparsky (1973[a]:60), with others due to Mascaró (1976:9)." Halle (1978:131)

As far as I can seen, Kiparsky's SCC is identical to Halle's version of the Strict Cycle Condition (which is discussed at greater length in §291). The existence of Halle (1978), however, has hardly left any trace in the literature. 49 Kiparsky (1982a) and Kiparsky (1982b), where (84) is introduced, are almost identical texts (the latter is a subset of the former). Halle (1978) is absent altogether from Kiparsky (1982b); it is mentioned in the reference section of Kiparsky (1982a), but does not appear in the text (or the notes). Also, Lexical Phonology in general and Kiparsky in particular are always credited for the SCC in the subsequent (overview) literature (Cole 1995 for example does not mention Halle 1978). Beyond the issue regarding the absence of Halle (1978) from the literature and from common phonological belief, the fusion of the management of derived environments with the no look-back device may or may 49

I am aware of two exceptions: prior to Kiparsky's (1982a,b) articles that have shaped the field, Rubach (1981:18ff) reports on Halle (1978) and is explicit about the fact that Halle's version of no look-back introduces derived environment effects into Chomsky's Strict Cycle Condition, which was not concerned with this phenomenon. Szpyra (1989:17) also mentions Halle (1978) as a precursor of Kiparsky's derived environment-containing SCC, but mistakenly lines up Kean (1974) and Mascaró (1976) as well.

164 Chap 7: Lexical Phonology not be a good idea. This is a separate question, and from today's post-hoc perspective the answer appears to be a clear no (see §§197,214). Finally, note that we have already come across another way of implementing a prohibition on look-back: bracket erasure (§174); the competition between the SCC and this mechanism is discussed in §203. The takehome message of the present section is that Lexical Phonology (or actually Halle 1978) has integrated Chomsky's no look-back device in order to make it cover derived environment effects. Contrary to what Kiparsky suggests, however, Mascaró's SCC does not afford any labour regarding derived environments: this clause needs to be explicitly added in form of (84a), which amounts to restating the Revised Alternation Condition (81). 191

8.3.3. Deriving the SCC from the Elsewhere Condition Kiparsky's (1973a,c) Elsewhere Condition regulates the competition between different rules that are applicable to a given form and produce different results. In such a situation, Kiparsky argues, only the more specific rule applies. Kiparsky (1982a:46ff, 1982b:159ff) proposes to derive the SCC from the Elsewehre Condition. In order for this to be done, so-called lexical identity rules need to be introduced: every lexical entry constitutes a phonological rule by itself whose output is identical to the input. Thus /niitingale/ is "transformed" into /niitingale/, /dog/ becomes /dog/ and so forth. Identity rules then compete with regular phonological rules in the Lexicon. In case there is true competition, i.e. when a real rule tries to modify an underlying item which is also represented by its identity rule, the latter always wins because it is more specific. The effect is that mono-morphemic items cannot be modified at all since the modifying rule will always lose against the identity rule. In the case of /niitingale/ for example, Trisyllabic Shortening will try to transform the /ii/ into /i/, but is blocked by the identity rule. In plurimorphemic strings such as /sææn-ity/, however, Trisyllabic Shortening can go into effect because there is no lexical item sanity, hence no identity rule and no competition. Under the pressure of Mohanan & Mohanan (1984) (see also Iverson & Wheeler 1988) and others, however, Kiparsky (1985b) abandons this line of attack altogether in subsequent work (as much as the SCC, see §197). The idea to analyse derived environment effects by enforcing lexical identity for isolated morphemes, however, will be revived in OT (see §521).

Derived environment effects 165 192

8.3.4. There are rules that apply in the Lexicon but affect underived items Kiparsky (1982b:154) is seduced by the perspective that the SCC affords expressing a correlation between the way a rule is ordered and the way it applies: "early" rules that live in the Lexicon are cyclic and hence can only apply to derived environments, while "late" postlexical rules that apply after syntactic computation are those that were called low-level (or automatic, or phonetic) above: they apply across the board without any condition on derived environments. The distribution in the architecture of Lexical Phonology is thus clear – but it remains to be seen whether it is really true that no postlexical rule respects the derived environment condition, and that no cyclic (lexical) rule violates it, i.e. applies even to mono-morphemic items. The former question has not received much attention in the literature. It amounts to asking whether cyclic derivation is needed for sequences of words as well (it is documented for morphemes). This is an important issue that impacts Praguian segregation and lexicalism; it was already mentioned in §158 and is discussed at greater length in §786 under the header of the word-spell-out mystery. The latter question is intimately related to the one that was discussed in §186 regarding "low level" processes that apply to mono-morphemic strings. It has a clear answer: it is not true that all rules which apply in the Lexicon are unable to modify mono-morphemic items. An example is English stress assignment, which must take place in the Lexicon because it is sensitive to affix classes (párent, parént-al vs. párent-hood, §147), but at the same time applies to underived /parent/, which is lexically unstressed and surfaces as párent.

193

8.3.5. Structure-building vs. structure-changing rules In the face of this evidence, a consensual move was to weaken the SCC so that cyclic rules may also apply to underived environments, but only if they add information (as opposed to the elimination or modification of existing information). This split of cyclic rules into a group of SCC-obeying (structure-changing) and SCC-violating (structure-building) items is introduced by Kiparsky (1982a:46ff, 1982b:160ff). Hence stress may be assigned to non-derived items in the Lexicon because stress assignment only involves the supplementation of prosodic structure: nothing in the input representation is modified. The same is true

166 Chap 7: Lexical Phonology for rules that assign syllable structure to lexically unsyllabified items. By contrast, a rule such as the Polish palatalisation discussed in §181 will be unable to apply to mono-morphemic strings since it involves the modification of the featural makeup of the input. An associated debate that was triggered by Kiparsky's introduction of the class of structure-building rules concerned the question whether such rules can destroy structure that was built on previous cycles. As far as I can see, this discussion only concerned two typical phenomena where structure was held to be built from scratch on the basis of lexical items that lack constituency: stress (McCarthy 1980, Steriade 1988, Halle 1990:158ff) and syllabification (Steriade 1982, 1984). We will see in §293 below that this kind of structure preservation may be interpreted as a herald of modification-inhibiting no look-back (today known as the Phase Impenetrability Condition, PIC). A further complication is the question whether structure-building applications of rules create (phonologically) derived environments. Originally, Kiparsky (1982a:47) said yes, but Kiparsky (1985a:91) argues for a negative answer: only structure-changing rules allow other structurechanging rules to apply in mono-morphemic items. In this context, underspecification also plays a role: autosegmental representations that have expanded in the early 80s allow for segments that are not fully specified for all melodic properties at the underlying level (e.g. Steriade 1987, Archangeli 1988). The missing information is then filled in by some (default) rule during the derivation. Filling in an underspecified gap may be interpreted as a structure-building, rather than a structurechanging operation. By contrast, modifying the melodic setup of a segment is always structure-changing when all segments are fully specified underlyingly.

194

8.3.6. Cyclicity as a property of strata vs. post-cyclic lexical rules But the structure-building vs. structure-changing proviso was still not found to be able to cope with the empirical situation: Rubach (1984), Rubach & Booij (1984, 1987), Rubach (1990) and Halle & Mohanan (1985) present cases from English, Polish and French where structure-changing rules whose application in the Lexicon is beyond doubt do apply to underived items. These rules must then be lexical, but cannot be cyclic. This further soaks the SCC: recall that Kiparsky was seduced by the fact that it seemed to be possible to predict from the lexical character of a

Derived environment effects 167

rule that it can only apply to derived environments. This now turns out to be wrong. In the face of this evidence, the SCC is further weakened in order to be rescued. Two strategies are developed that appear to be distinct and competing at first sight (but see below). Both further restrict the application of the SCC, this time not to a certain type of rule, but to a sub-area of the Lexicon. Mohanan & Mohanan (1984) and Halle & Mohanan (1985) propose that cyclicity is a property of strata: each individual stratum may or may not be cyclic, that is, may or may not respect the SCC. Halle & Mohanan (1985:96f) for example argue that in English, stratum 1 (level 1) is cyclic – that is, its rules are blocked in underived environments. By contrast, stratum 2 (level 2) is non-cyclic, which means that its rules may freely transform underived strings. Halle & Mohanan (1985:95ff) motivate this move with English nasal cluster simplification (§167): mono-morphemic /mn/ sequences are reduced to [m] when their morpheme occurs in isolation (damn) and before class 2 suffixes (damn-ing), but remains unaffected before class 1 suffixes (dam[n]-ation). Underapplication must thus be organised so that the deletion rule (that is, no doubt a structure-changing rule) applies to the two former, but not to the latter context. If of all lexical rules only those observe the SCC (i.e. are blocked in underived environments) that are active at stratum 1, the cluster of [damn] and [[damn] [ing]] may happily be reduced by the stratum 2 rule n → ø / [+nasal]__ ]. 50 The alternative strategy is represented by Rubach & Booij (1984, 1987) and Rubach (1990): their scenario weakens the SCC even more than Halle & Mohanan's (1985). They present evidence to the end that cyclicity is not tied to strata at all. That is, Rubach & Booij (1987:4) give up on any kind of predictability: there is nothing that allows us to distinguish lexical rules that do from lexical rules that do not respect the SCC. Instead, they simply set up a new category of rules that applies in the Lexicon, but is ordered after cyclic rules: lexical postcyclic rules. On their count, then, cyclic lexical rules respect the SCC and apply only to derived environments, while post-cyclic lexical rules violate the SCC and apply across the board. Both types of rules are lexical and thus continue to be opposed to postlexical rules. 50

Halle & Mohanan (1985) use Mohanan's (1982, 1986) brackets and also otherwise follow Mohanan's (1986) analysis of nasal cluster simplification (see §168).

168 Chap 7: Lexical Phonology Rubach & Booij's lexical post-cyclic rules revive SPE's word-level rules: recall from §105 that while SPE's cyclic rules iteratively apply to each cycle, word-level rules apply only once to word-sized strings. Also, just like Rubach & Booij's lexical post-cyclic rules, word-level rules in SPE do not apply to chunks that are bigger than the word. We will see in §§233f that – quite logically – Halle & Vergnaud's (1987a) attempt to restore SPE and to unseat interactionism implements the same idea. In this model, SPE's word-level rules are called non-cyclic (as opposed to cyclic rules, which continuate the cyclic rules of SPE). 195

8.3.7. The SCC-K is void of empirical content Given this general landscape where the SCC is weakened every time it meets an obstacle, it is difficult to see what remains of its empirical content. Or rather, what remains of Kiparsky's SCC. For Mascaró's original version had got nothing to do with derived environments, and (contrary to what Kiparsky says) did not try to identify cyclic rules as applying only in the Lexicon: there was no Lexicon by the time Mascaró (1976) wrote, not any more than there were Lexical Phonology or strata. Therefore it is useful to disentangle the terminological confusion that Kiparsky's annexation of the SCC has produced: the Lexical Phonology literature typically refers to "SCC" as Kiparsky's device, but one may also find reference to the offspring of Chomsky's (1973) and Mascaró's (1976) Strict Cycle. Talking just about the "SCC" is thus abetting confusion: lexical phonologists will think of Kiparsky's version, i.e. the tool for derived environment effects, while people familiar with the other tradition will think of the no look-back device that is known as Phase Impenetrability today and has got nothing to do with derived environments. For this reason, the remainder of the book carefully distinguishes between Mascaró's (SCC-M) and Kiparsky's (SCC-K) version of the SCC. The latter does, the former does not make reference to derived environment effects.

Derived environment effects 169 196

8.4. Solution 2: derived environments are made a lexical contrast (Kiparsky 1993)

197

8.4.1. Back to where we started: Kiparsky declares the bankruptcy of SCC-K Under the pressure of the empirical and conceptual problems reviewed, Kiparsky (1993) ends up acknowledging the failure of all attempts to build derived environment effects (which he calls NDEB: nonderived environment blocking) into the theory, or to derive them from some more general principle. In particular, he declares the bankruptcy of SCC-K:51 it is simply not true that the behaviour of a rule with respect to derived environments can be predicted from its location in the grammatical architecture. Kiparsky (1993:280ff) reviews cases which show that NDEB is neither specific to cyclic nor to lexical rules: there are cyclic lexical rules with no NDEB, and also NDEB in word-level and postlexical rules. He therefore abandons the strategy to add patch after patch to a theory that is void of empirical content: the SCC-K is on the wrong track and discredits the overall architecture of the theory; it must therefore be thrown over board. We are thus back to where we started: derived environment effects exist, but they are mysterious and escape an implementation in formal terms. Kiparsky (1993) declares the failure of all attempts, past and future, at expressing NDEB by constraints on underlying forms (Alternation Condition) or phonological computation (Revised Alternation Condition, SCC-K, Elsewhere Condition). What we are left with, then, are the questions that Kiparsky (196873) already struggled with (§186): 1) what distinguishes ("real") rules that apply only to derived environments on the one hand from "low-level", "automatic" or "phonetic" rules that apply across the board on the other? 2) assuming that the answer is known, how do we implement it into a formal theory of phonology? Kiparsky argues that the empirical record, after 20 years of research on the matter, has produced a picture that allows for an accurate description of the derived environment animal: the class of obligatory neutralisation rules are NDEB. This is the original Revised Alternation Condition (§187), which "is really no more than a descriptive generalization dressed up as a 51

This is also Cole's (1995:89f) conclusion, who visibly did not hold Kiparsky (1993) in hands by the time she wrote.

170 Chap 7: Lexical Phonology principle and is unstateable as a formal condition on phonological rules", but empirically "has struck much nearer the mark" (Kiparsky 1993:277f) than SCC-K. 198

8.4.2. Different lexical representations for the same segment Kiparsky (1993:285ff) then goes on to propose a solution that is based on two assumptions: 1) learners construct the simplest grammar (what "simple" means is left undefined, though, just as in the literature on the evaluation measure, e.g. Kiparsky 1974); 2) lexical entries are maximally underspecified: a feature can have three values (positive, negative, unspecified). What his proposal really relies on, however, is the idea that since the encoding of NDEB as a constraint on underlying forms and on computation has failed, the derived vs. non-derived contrast must be achieved by different lexical representations. That is, morpheme-internal occurrences of a segment will have a melodic (featural) setup that is distinct from the one of the same segment in morpheme-marginal position (actually, in morphemefinal position). The tool that allows Kiparsky to express this lexical distinction is underspecification (e.g. Steriade 1987). In the Finnish case of NDEB that was discussed in §180, the /t/ of the essive vete-nä "water" turns into [s] before the infinitival -i in the corresponding nominative form ves-i. By contrast, mono-morphemic /ti/ sequences surface without modification, as for example in neiti-nä "young lady, essive". Kiparsky (1993:286) proposes that the relevant assibilation rule adds the specification [+continuant] to segments that are lexically unspecified for this feature: [0cont] → [+cont] / __i (where [0α] means that the value of the feature is unspecified). /t/s that occur before /i/ in their lexical entry, then, are fully specified as [-cont], while /t/s that stand elsewhere in the string are unspecified and hence bear [0cont]. A late default rule then fills in all unspecified instances of [cont] with a negative value (hence making /[0cont]-t/ a real [t]). This system blocks the application of the assibilation rule to morpheme-internal /ti/ strings because the /t/, occurring before a tautomorphemic /i/, is in fact a /[-cont]-t/: assibilation only targets /[0cont]-t/. Hence /neiti-nä/ → [neiti-nä]. By contrast, the underspecified /t/ in /veT-i/ (upper case letters indicate underspecification) will be supplemented with the posi-

Derived environment effects 171

tive value for its /[0cont]/ feature because the assibilation rule targets [0cont]; the result is an [s].52 In support of his analysis, Kiparsky argues that maximally underspecified lexical representations are the result of the acquisitional process because learners target the simplest grammar.

199

8.4.3. Why NDEB processes must be obligatory and neutralising The question, then, is why only obligatory and neutralising rules may be sensitive to derived environments. Recall that Kiparsky believes that this is the correct empirical generalisation.53 Regarding the former issue, Kiparsky (1993:287) argues for an acquisition-based scenario: in order to learn a rule that is sensitive to derived environments and optional at the same time, negative evidence would be required. Such a rule would produce two possible results in derived environments, but only one result in non-derived environments (where it cannot apply). Hence, Kiparsky argues, the absence of variability in the latter context cannot be inferred from distributional evidence or the stimulus, which means that it could only be understood upon an explicit instruction not to apply the rule morpheme-internally – a piece of negative evidence. Nega52

53

In Kiparsky's system, morpheme-internal instances of /t/ that do not occur before /i/ are also underspecified. Assibilation does not target them, but they are filled in with the negative value by the late default rule. Hence /…aTa…/ → […ata…]. Kiparsky (1993) also discusses another condition that involves the distinction between structure-building and structure-changing rules (see §192). In an autosegmental environment, autosegmental representations (below and above the skeleton) are considered to be structure. In an underspecification perspective, then, filling in an underspecified segment is a structure-building process, just like the construction of syllable- or stress-relevant structure (Kiparsky assumes feet). By contrast, structure-changing processes are those that delete or delink autosegmental objects (Kiparsky 1993:288 is not really explicit on this point and offers no illustration of a structure-changing process that cannot be NDEB). On Kiparsky's analysis, the contrast at hand is relevant foremost for suprasegmental phenomena (syllable structure, stress). As far as I can see, however, he does not provide any explanation why structure-changing processes cannot be sensitive to derived environments, or why the distribution of NDEB is not the reverse among structure-building and structure-changing processes. This aspect of Kiparsky's theory is not really relevant for the discussion, though.

172 Chap 7: Lexical Phonology tive evidence, however, is excluded on general learnability-theoretic grounds (e.g. Marcus 1993). The fact that rules must be neutralising in order to experience the derived environment restriction follows from the simplicity request. In order to show this, Kiparsky (1993:287) discusses English aspiration, which is absent from underlying representations and added by rule in entirely predictable and "automatic" fashion (aspiration is thus a prototypical case for a postlexical rule in terms of Lexical Phonology): voiceless stops are aspirated word-initially and in onsets of stressed syllables (with of course a fair amount of dialectal variation). This rule is not neutralising since its output does not coincide with any segment of the underlying inventory. Kiparsky argues that distributional evidence allows learners to construct underlying forms without the added feature (here aspiration), and that they will end up with unaspirated underlying forms across the board because this is the simplest grammar of English. It is not clear to me, however, why uniformly non-aspirated underlying forms should prevent the aspiration rule from applying only in derived environments.

200

8.4.4. Posterity of Kiparsky's lexical solution Even though Kiparsky's (1993) idea to lexically prespecify the invariability of morpheme-internal material had an offspring (Inkelas & Cho 1993, Oh 1995, Inkelas & Orgun 1995, Inkelas 2000), it was not followed by the mainstream of Lexical Phonology and follow-up theories. The "orthodox" strand (e.g. Rubach & Booij 2003:444) as much as the continuators of Halle & Vergnaud (1987a) (§237, e.g. Halle & Nevins 2009) to date maintain the SCC-K and all the patches mentioned that are necessary in order to make the empty shell float. The eventually pernicious character of the SCC-K is further discussed in §214.

201

8.5. Solution 3: bracket-sensitive rules (Mohanan 1982)

202

8.5.1. Bracket-sensitive rules can do all derived environment effects Mohanan's (1982:24f) bracket-sensitive rules are certainly a much simpler way of building derived environment effects into a formal system than the SCC-K and Kiparsky's (1993) lexical solution. Mohanan-type brackets and their rule-triggering function were discussed and illustrated in §166; they

Derived environment effects 173

are put to use in Mohanan & Mohanan (1984), Halle & Mohanan (1985) and Mohanan (1986:21ff). Mohanan-type brackets delineate each and every morpheme upon concatenation (see §168) – what they signal are thus simply (all and only) morpheme boundaries. Therefore, a rule whose application is conditioned by the presence of a bracket automatically distinguishes between monoand plurimorphemic strings. In order to restrict the application of a rule to derived environments, then, the only thing that needs to be done is to make it sensitive to brackets. A rule with a bracket in its structural description will never apply to mono-morphemic items. In the Polish case discussed in §181, /s/ palatalises before /e/, but only if the two segments are separated by a morpheme boundary: /servis-e/ "service LOCsg" bears two instances of /se/, one mono-morphemic, the other separated by a morpheme boundary. The result is [sɛrvi˛-ɛ]: the latter is, the former is not subjected to palatalisation. In order to block palatalisation within morphemes, it is thus enough to write a rule as under (86). (86)

s → ˛ / __ ] [ e

All cases of morphologically derived environment effects could be managed like this. Bracket-sensitive rules also have an important advantage in comparison with the SCC-K that relies on computation: while brackets can be built into rules à la carte, computational principles such as the SCC-K marshal all rules alike. What we observe, though, is precisely that some rules do obey the derived environment condition, while others do not. All attempts at understanding why rules behave in this or that way have been rather unsuccessful: the distinction between structure building and structure changing rules (see §193) is not any more operative than the lexical vs. postlexical distinction (§193) or any other criterion that relies on the place in the architecture where rules apply (this is what Kiparsky 1993 shows, see §197). Bracket-sensitive rules thus offer a very simple solution for derived environment effects that allows for a process-specific distribution of derived environment sensitivity. This seems to be exactly what the empirical record requires. The more surprising is the fact that this option seems to be absent from the literature. It appears that nobody has thought of applying the bracket-based technology to derived environment effects. Mohanan (1986) for example makes ample use of brackets and bracket erasure, but does not address the question of derived environments at all.

174 Chap 7: Lexical Phonology The following section tackles another point that seems to have gone unnoticed in the literature: the competition between the two no look-back devices that Lexical Phonology has produced, i.e. bracket erasure and the SCC-K.

203

8.5.2. Brackets and SCC-K are direct competitors – but not in the literature It was shown in §190 that Kiparsky's version of the Strict Cycle Condition, SCC-K, is designed for covering both derived environment effects and no look-back effects. It is thus a direct concurrent of Mohanan's (1986) brackets and bracket erasure (see §168). The previous section has shown that bracket-sensitive rules are well suited to account for derived environment effects. Together with bracket erasure, they also constitute the no look-back device in Mohanan's system (see §174). Curiously enough, though, it seems that the concurrence of Kiparsky's SCC-K and Mohanan's bracket-based mechanism has gone unnoticed in the literature: I have not come across any work where both ways of going about derived environment effects and no look-back are discussed. Although writing on the backdrop of the SCC-K tradition that Lexical Phonology lived with in the early 80s, Mohanan (1986) for example does not even mention derived environment effects and the Strict Cycle Condition. He diplomatically writes that "in the model of Lexical Phonology presented in this book, […] information about concatenation is represented in terms of brackets" (Mohanan 1986:127f, emphasis mine). Be that as it may, the fact that brackets and the SCC-K do the same job for derived environments was shown in the previous section. That bracket erasure and the other side of the SCC-K coin (strict cyclicity) are functionally equivalent is demonstrated below. Table (87) reproduces the SCC-K for convenience. (87)

Strict Cycle Condition (SCC-K) a. Cyclic rules apply only to derived representations. b. Def.: A representation φ is derived w.r.t. rule R in cycle j iff φ meets the structural analysis of R by virtue of a combination of morphemes introduced in cycle j or the application of a phonological rule in cycle j.

(87a) covers derived environment effects, while (87b) takes care of no look-back phenomena: a rule may apply to a string only if it uses morphological material that is introduced at the latest cycle.

Derived environment effects 175

Comparatively, recall from §168 that the bracket is the critical ingredient in Mohanan's rule "g → ø / __ [+nasal] ]" that prevents the /gn/ sequence in sign-ature from being reduced: at level 2 where it applies, the inner brackets of the level 1 concatenation [[sign][ature]] have been erased by virture of bracket erasure. By contrast, the level 2 concatenation [[sign][ing]] (as well as the nonderived [sign]) still bears inner brackets upon the application of the rule, which may thus go into effect. On the count of the SCC-K, the stratal organisation is identical, but the reason for the non-application of g-deletion to sign-ature is not the same: -ature being concatenated at level 1, the level 2 rule g-deletion is blocked from applying to the level 1 string sign-ature because it does not use any material that was added at the level of its application. By contrast, the rule may apply to sign-ing because -ing is merged at level 2. The systematic non-reduction of morpheme-internal /gn/ (and /mn/) as in ignore (and amnesia) is also covered by the SCC-K, this time by the side of the coin that takes care of derived environment effects (87a). The combined no look-back and derived environment effect of the SCC-K thus covers the rule-triggering pattern for free. The more surprising is the fact that the literature does not seem to relate level 2 rules to the SCC-K – or derived environment effects to bracket-sensitive rules. Of course, the functional equivalence and hence competition of Mohanan's bracket-based system and Kiparsky's SCC-K does not make any statement about the intrinsic merits or problems of the devices at hand. If the SCC-K is to be abandoned because it is simply wrong empirically as Kiparsky (1993) suggests (§197), it is out of business for the analysis of the rule-triggering pattern as well. And the same goes if Mohanan's brackets are riddled by the concerns that are discussed in §169. In this case, however, Lexical Phonology has lost all instruments for the analysis of the ruletriggering pattern, which begs the question. 204

8.6. Solution 4: derived environment effects are non-linguistic in nature (Anderson 1981)

205

8.6.1. Fultonians: speakers use extra-linguistic evidence (spelling) in order to establish underlying forms Anderson (1981) inquires on the relationship between the linguistic system (grammar) and extra-linguistic factors such as acoustics, physiology, perception, general cognitive constraints and so forth. In this resident discussion since Saussure introduced the distinction between Langue and Parole,

176 Chap 7: Lexical Phonology Anderson defends the Saussurian/Chomskian autonomy perspective. He concludes that "an adequate account of the phonological systems of natural languages must accord a central role to a set of principles that have no direct foundation in extralinguistic considerations" (Anderson 1981:535). Part of his argumentation is to establish criteria that are able to detect patterns which are due to extra-linguistic causalities. One of the examples discussed are derived environment effects. Anderson (1981:530ff) shows that the claim of the (Revised) Alternation Condition is counterfactual: English stress reduction provides unquestionable evidence to the end that speakers lexicalise forms which never appear on the surface. Anderson's example is the proper name Fúlton, whose second syllable is unstressed and hence reduced to schwa. There is no way to ever hear anything else than a schwa in this name. According to the (Revised) Alternation Condition, the underlying form should thus be established with a schwa in the second syllable. Contra to that prediction, speakers produce an [ow] as soon as they are asked to talk about followers of Mr Fulton on the hypothesis of him entering politics and creating his party, the Fultónians. The [ow] of course is the regular result of a stressed /o/, which speakers thus had lexicalised without prior exposure to the word Fultonians. Anderson concludes that there are sources of evidence other than the phonetic signal and alternations that speakers use in order to establish underlying forms. In the case of Fulton, the relation with spelling is obvious: speakers who produce Fultonians with an [ow] have lexicalised an /o/ because they know how Fulton is spelt. 206

8.6.2. Explaining the genesis of derived environment effects does not exonerate from coming up with a synchronic scenario On these grounds, Anderson argues that the restrictive application of a rule to derived environments does not need to be a property of the linguistic system: it could as well be an artefact of the learning situation, which prevents speakers from coming across evidence for alternations. That is, alternations are absent within stems, just like in Fulton, but unlike in this case, the ordinary situation is that speakers happen to have no access to extralinguistic evidence such as spelling or neighbouring dialects. Hence they will be left without evidence for an alternation, and therefore do not produce any. Kiparsky (1993:278) discusses Anderson's alternative. While he agrees with the acquisitional scenario as an explanation for the genesis of

Derived environment effects 177

derived environment effects, he maintains the need for a grammatical management of the result in adult grammar. It is indeed not obvious how Anderson's solution could be implemented into the synchronic functioning of a grammar. The non-alternating mono-morphemic items must be somehow lexicalised, and will then have to undergo phonological rules. If there is a rule that concerns an underlying sequence /XY/, and /XY/ occurs within a stem, it will undergo this rule. If this rule is sensitive to derived environments, however, /XY/ will resist. Hence nothing has been gained: we are still looking for a means of preventing rules from applying to mono-morphemic /XY/. Anderson (1981) does not explain what the lexicalised form of segments of monomorphemic items looks like: if their underlying form is faithful to their invariable surface, there is no reason why they should not undergo relevant rules. If on the other hand invariable items (may) have an underlying form that is different from their surface form, they may escape relevant rules – but then we are back exactly to what Kiparsky (1993) proposes. And, as will be discussed in §209 below, essentially to the SPE solution of making nightingale /nixtVngael/. 207

8.7. Conclusion

208

8.7.1. A new phenomenon that is poorly understood No doubt Lexical Phonology in general and Paul Kiparsky in particular are to be credited with the discovery of the derived environment phenomenon, which is certainly an important property of natural language and was a major concern for phonological theory over the years. "I regard the discovery of the existence of rules of this kind as one of the most significant empirical finds in modern phonology", Halle (1978:128) says for example. One the one hand, the study of derived environment effects has produced a stable empirical record, and has helped characterising those processes that apply only in derived environments, as opposed to those that apply across the board. On the other hand, however, the phenomenon is not really understood: no operative criterion that allows for a waterproof split of processes into those that are and those that are not restricted to derived environments has emerged, and the various attempts to account for the phenomenon by grammatical tools have not yielded convincing results. Lexical Phonology has produced three strategies for the analysis of derived environment effects: 1) a constraint on rule application (SCC-K,

178 Chap 7: Lexical Phonology Kiparksy 1982a,b, Rubach & Booij 1984, 1987); 2) a lexical contrast between morpheme-internal and morpheme-peripheral variants of segments (Kiparsky 1993); 3) Mohanan-style brackets and bracket-sensitive rules (following Mohanan's 1982, 1986 system, but which for some reason was never applied to derived environment effects). All three solutions are unsatisfactory. 209

8.7.2. Kiparsky's solutions miss the facts or the point The SCC-K is void of empirical content (§195): it is simply not true that the location of a rule (its application in the Lexicon) determines the way it applies (cyclically, i.e. only to derived environments). Building the sensitivity to derived environments into the lexicon (Kiparsky 1993) comes at the cost of doubling the segmental inventory: the same object, i.e. the same phoneme and the same segment, may now have two distinct lexical representations, one underspecified at morpheme edges, the other not (in morpheme-internal position). Technically, this may produce the correct result. However, it is hard to trust a solution where the same phoneme (or segment) has two distinct lexical representations on the grounds of a parameter that has got nothing to do with lexical contrast. Still more worrisome is that the lexical solution misses the morphological generalisation that is at the origin of derived environment effects. That is, the environment that blocks the application of a rule is not just any environment: only mono-morphemic items block derived environment rules. Since the distribution of items is free in the lexicon, this would be entirely accidental on Kiparsky's (1993) count: the underspecified and the fully specified versions of the same phoneme could as well have the opposite distribution, which would produce underived environment effects – a phonological monster. Finally, Kiparsky's (1993) lexical solution is strikingly similar to the strategy that was used in SPE, which also relies on a lexical contrast. Recall that in order to prevent Trisyllabic Shortening from applying to nightingale (and only in order to do so), the underlying form /nixtVngael/ is set up (§184). Independent rules then take /ix/, which does not satisfy the structural description of the rule, to [aj]. On Kiparsky's (1993) account, the underlying representation /neiti/ with a morpheme-internal [-cont] /t/ is set up; this blocks the Finnish assibilation rule, which applies only to [0cont] /t/. By lexical specification (accident?), the [0cont] version of the dental is only

Conclusion 179

encountered morpheme-finally (as in /veT/, which is turned into [ves-i] by the rule in question). In SPE as much as on Kiparsky's (1993) analysis, an artificial lexical contrast is introduced that has no motivation other than producing the correct derived environment effect. In both cases, a late (and automatic) default rule turns the underlying item that was made artificially different from its surface form back to what it really sounds like.

210

8.7.3. Mohanan's bracket-sensitive rules – a misjudged option The simplest way of analysing derived environment effects are Mohanan's (1982, 1986) bracket-sensitive rules. On this count, rules are directly made sensitive to morphological divisions, which appear in structural descriptions as such. The unwarranted burden that brackets represent from the modular and theory-internal point of view was discussed in §169. An obvious advantage of this solution, however, is its flexibility: the presence of brackets in a rule is an idiosyncratic property of this rule. Hence rules may or may not be sensitive to derived environments – this is exactly what we observe: some processes are restricted to derived environments, others are not. Unless this split may be predicted (all attempts have failed thus far), an idiosyncratic solution that allows for process-specific sensitivity to derived environments hits closest to the mark. The discussion of derived environment effects will be resumed as we go along: their status in Kaye's system is discussed in §284, and solutions that are proposed in OT are reviewed in §§509,516.

211

9. Conclusion

212

9.1. Interactionism and multiple mini-grammars Lexical Phonology is to be credited for two important contributions to grammatical thinking and linguistic theory: interactionism and multiple mini-grammars. Interactionism is the idea that concatenative and interpretational activity is interspersed (§146). It affords an elegant reconciliation of cyclic derivation with the modular request of not representing untranslated morpho-syntactic information in the phonology: this is how the awkward brackets of SPE can be done away with (see §§161,170).

180 Chap 7: Lexical Phonology Traditional grammar and generative orthodoxy (SPE on the phonological, Aspects up to GB on the syntactic side) have always thought of a strict ordering of morpho-syntax and phonology (semantics): morphosyntactic concatenation is entirely completed before phonological and semantic interpretation begins. This view was long associated with (but in fact does not follow from) the inverted T model (§86). Interactionism is a revolution in this environment: it opens an entirely different view of how the three pieces of the inverted T interact. We will see in the following chapter that interactionism aroused a strong reaction on the side of generative orthodoxy (Halle & Vergnaud 1987a). A more remote posterity is discussed in §§304,676: since Chomsky's (2000a et passim) phase theory, interactionism has become the central architectural device of grammar. Multiple mini-grammars are the other outstanding contribution of Lexical Phonology to linguistic theory. As far as I can see, the idea that phonology is made of several distinct computational systems was never expressed before.54 Here as well, the innovation unseats classical (generative) thinking, at least partly: we will see in §234 below that it is useful to distinguish between two types of multiple computational systems: morpheme-based miniphonologies (level 1 vs. level 2 rules in Lexical Phonology) and chunk size-based mini-phonologies (SPE's word-level phonology, Praguian segregation). Today morpheme-specific mini-phonologies stand unchallenged in OT (see §§477,483). In §828, it is argued that current syntactic theory, where Phase Impenetrability is critical, cannot work with morphemespecific computational systems at the other end of the phasal pipe (see also Scheer 2010b).

54

One could think of loanword phonology as a precedent: a traditional view is that loanwords, which in some languages do not obey the same phonological constraints as native words, are governed by a specific computational system. This is certainly true, but the parallel appears to be spurious: the application of regular or loanword phonology is determined idiosyncratically for each lexical item; by contrast, multiple mini-phonologies are fully grammatical: they concern the same lexical items. Also, loanword phonology has got nothing to do with the interface. As far as I can see, the loanword parallel has played no role in the development of Lexical Phonology. By contrast, there is an obvious relationship between parallel incarnations of multiple mini-phonologies (cophonologies, indexed constraints, see §478) and loanword phonology.

Conclusion 181 213

9.2. Unprecedented proceduralisation of the interface In this book, interface theories are considered in the light of the balance between procedural and representational means of talking to phonology. In this perspective, the fate of traditional SPE-style boundaries was examined. Kenstowicz & Kisseberth (1977:83ff, 1979:407ff) have established a pretheoretical classification of phonological patterns that are sensitive to extraphonological information (§51): morpho-syntactic divisions may either be rule-blocking or rule-triggering. This contrast is faithfully reproduced by the stratal architecture of Lexical Phonology: it translates as the action of two different morpheme-specific mini-grammars. The set of rules that is active at level 1 (level 1 phonology) is responsible for the rule-blocking pattern, whereas level 2 phonology produces the rule-triggering pattern. While Kenstowicz & Kisseberth talked about rule-blocking and ruletriggering boundaries, the analysis that Lexical Phonology proposes encodes the same contrast in terms of procedurally ordered mini-grammars. This equivalence is a good witness of the proceduralisation of the interface that Lexical Phonology has undertaken. Since the 19th century, the only way for morpho-syntactic information to reach phonology that linguists have thought of was representational. Chomsky et al. (1956) and SPE have introduced cyclic derivation, i.e. the idea that information may also be transmitted procedurally (§100). However, SPE remained by and large a representational theory (as far as the interface is concerned): affix classes have a representational management in terms of boundaries (§92), and the idea of a specific word-level phonology is also implemented representationally (despite its obviously procedural character, see §105). Lexical Phonology is thus the first theory that really gives flesh to the procedural idea. It affords an unprecedented proceduralisation of the interface, which is an appreciable advance on the diacritic front: boundaries are phonological aliens.

214

9.3. Representational communication and the pernicious SCC-K The central problem of Lexical Phonology, in my view, is that it cannot stand up to its non-representational ambition: Lexical Phonology has set out to maximise the turnout of procedural communication with morphosyntax. Accordingly, the representational component should have shrunk, which it did indeed – without however disappearing completely.

182 Chap 7: Lexical Phonology The project to proceduralise the interface at the expense of representational devices was a side-effect of the stratal architecture, rather than a primary goal (see §165). Once it was set in motion, however, boundaries were reputed to be out of business. It is true that the stratal system affords a drastic reduction of (diacritic) boundaries in interface management. This offers an entirely new perspective to the many who were frustrated by the oddities of boundaries in the late 70s (see §131). Unfortunately, the procedural tools of the new theory cannot do everything: the naked stratal architecture cannot account for two core phenomena: the rule-triggering pattern (§166) and derived environment effects (§177). The patches that were proposed in order to bring them home were discussed, namely Kiparsky's SCC-K (§188) and Mohanan's bracket-based system (§§168f). 55 Curiously enough, the former was not applied to the rule-triggering pattern, while the virtues of the latter for derived environment effects were not explored (§201). Kiparsky (1993) himself has then declared the bankruptcy of the SCC-K, which is simply empirically wrong (§197, but Rubach & Booij 2003:444 continue to advertise an SCC-K-based version of Lexical Phonology). On the other hand, Mohanan's brackets reintroduce SPE-type boundaries through the back door. The bracket-based remedy, which is designed to cover the rule-triggering pattern, may however turn out to be worse than the original disease: §169 draws the list oddities that one must be prepared to go along with when brackets and bracket erasure are taken seriously. The result is a hybrid theory where boundaries, which come in the rhetorical guise of brackets, are just as critical as they have always been in SPE, and also do the same labour (they trigger rules). Bermúdez-Otero (2008, forth a:§1.2, §2.6.3) offers a much more detailed overview of stratal Lexical Phonology. He reaches a very similar conclusion, expressed with greater vigour, though: all the adornment that the practitioners of the 80s were inveigled to add to the basic stratal idea – namely the SCC-K and brackets – was the fall of man of Lexical Phonology (§1.2.2 of his book manuscript is called "The deleterious effects of Strict Cyclicity"). In Bermúdez-Otero's view, this is what lured Lexical Phonology on to decline (§1.2 is called "Lexical Phonology crippled by the legacy of SPE"), and Stratal OT is about to reinitialise the motion on the grounds of the original setup.

55

But also Rubach & Booij's (1984) lexical post-cyclic rules or the option of noncyclic strata that Halle & Mohanan (1985) provide for, see §194.

Conclusion 183

In conclusion, Lexical Phonology (in its classical or OTed skin, see §§477,483) either buys into additional machinery in order to be able to account for the rule-triggering pattern, or cannot cover this phenomenon. Also, the two disputed patterns – derived environment effects and rule-triggering boundaries – have in common that they call for a no lookback device. Today the no look-back issue is at the forefront of the research agenda in minimalist syntax: the modern device is called Phase Impenetrability. It is shown in §828 below that morpheme-specific mini-phonologies are incompatible with no look-back since both devices do the same job. If Phase Impenetrability is necessary in syntax, then, morpheme-specific mini-grammars have to go. Finally, it should be borne in mind that if derived environment effects and the rule-triggering pattern are both about no look-back, they are distinct phenomena. Kiparsky's SCC-K is precisely an unsuccessful attempt at unifying them. The two patterns are logically independent (§182), and we will see that while the rule-triggering problem may have a solution in the the classical approach to computation which does not provide for morpheme-specific mini-phonologies, derived environment effects beg the question as much today as they did when Kiparsky gave up on the SCC-K.

215

Chapter 8 Halle & Vergnaud (1987a): selective spell-out and SPE-restoration

216

1. Introduction: a hermaphrodite theory with a new idea

217

1.1. Unseating Lexical Phonology and restoring SPE Halle & Vergnaud (1987a) (or actually Halle 1986, see §221) is the starting point of a new tradition in the procedural communication between morphosyntax and phonology. It combines the basic architecture and tools of SPE with an entirely new idea: spell-out of morpho-syntactic nodes is selective; whether a node is spelled out or not depends on a lexical specification of affixes. Morris Halle has always defended SPE: he admits some developments such as autosegmentalism, but considers the basic architecture correct. This is also true for the interface: Halle & Vergnaud (1987a) is the attempt to restore SPE in this area. We have seen in the preceding chapter how profoundly the advent of Lexical Phonology has overthrown the architecture (interactionsim) and functioning (strata) of the system that SPE had defined. Lexical Phonology completely dominated the field by the mid-80s, and Morris Halle at first adopted the general framework (Halle & Mohanan 1985), if with the amendment that strata may be non-cyclic – a herald of selective spell-out as we will see. Halle & Vergnaud (1987a) then break with Lexical Phonology by attacking its foundations, i.e. interactionism and the stratal architecture.

218

1.2. Relations with Lexical Phonology In the self-understanding of Halle & Vergnaud (1987a:77), the model that they propose was "much influenced by the ideas of [Lexical Phonology] […], as well as by such critics of this approach as Aronoff & Sridhar (1983) and Sproat (1985)" (more on the latter in §243). Ten years later, Halle (1997b:302) talks about "the Halle & Vergnaud 1987a version of lexical phonology".

186 Chap 8: Halle & Vergnaud (1987a): selective spell-out and SPE-restoration Halle & Vergnaud (1987a:77) list the features that their model takes over from Lexical Phonology: strata and the distinction between a lexical and a postlexical phonology. They then make the bone of contention explicit: interactionism. On the face of it, Halle & Vergnaud's theory may thus be thought of simply as a non-interactionist version of Lexical Phonology – which is the wrong way to look at the picture: both aspects of continuity with Lexical Phonology turn out to be empty word shells. Halle & Vergnaud in fact eliminate strata and the lexical - postlexical distinction. In Lexical Phonology, the latter covers two different things: the distinction between a pre-word and a post-word phonology and the idea that these represent different computational systems (i.e. are made of different sets of rules). We will see in §238 that Halle & Vergnaud subscribe to the former, but reject the latter idea. Also, the mention of strata as a tool that was taken over from Lexical Phonology is misleading – or rather, a terminological confusion. While a stratum in Lexical Phonology is the association of a certain type of morphological concatenation (e.g. class 1 affixes at level 1) with the associated phonological computation (level 1 rules) (see §152), strata in Halle & Vergnaud's terminology are simply serially ordered blocks of rules: "we adopt from Lexical Phonology the organization of phonological rules into a number of blocks called strata" (Halle & Vergnaud 1987a:77, emphasis in original). No morphological activity is implied: Halle & Vergnaud could not interleave the serial ordering of rule blocks with concatenation since they precisely reject interactionism. Hence the object that is called stratum in Lexical Phonology has got nothing to do with Halle & Vergnaud's "strata" – a regrettable terminological confusion that does not help determining who is who. Another point is that two relevant aspects of the relationship with Lexical Phonology are not mentioned. Halle & Vergnaud (1987a) follow the proceduralising programme of Lexical Phonology (§213): the interface is only thought of in terms of procedural communication with morphosyntax, which means that there is no place for boundaries. Also, morphemespecific multiple mini-grammars – a major innovation of Lexical Phonology (see §212) – are not mentioned. On the face of it, Halle & Vergnaud (1987a) have also taken over this idea: they distinguish between a cyclic and a non-cyclic pool of rules. These, however, do not match the level 1 level 2 distinction of Lexical Phonology; rather, they replicate the SPEdistinction between cyclic and word-level rules (more on this in §232 below).

Introduction: a hermaphrodite theory with a new idea 187 219

1.3. Relations with SPE This much for the relationship of Halle & Vergnaud with Lexical Phonology: proceduralisation, multiple mini-grammars, Praguian segregation yes, interactionism (and hence strata) no. Again, this may look like a mere noninteractionist version of Lexical Phonology. That Halle & Vergnaud's model is more than that – in actual fact a movement that is directed against Lexical Phonology – appears when comparing it with SPE. Boundaries set aside, Halle & Vergnaud restore the complete technology of SPE. Recall from §105 that SPE entertained a contrast between cyclic and word-level rules. This is exactly what is reproduced under slightly different labels: cyclic rules still run under the same name, but word-level rules are now called non-cyclic rules (more on the use of the word cyclic in §233). Another important aspect of SPE-restoration are brackets, which are reintroduced into phonological representations, and which play exactly the same role as they did in SPE (the delineation of interpretational units, see §98). The revival of brackets is an automatic consequence when interactionism is abandoned: recall from §161 that interactionism is an advance from the modular point of view because it allows eliminating brackets, which are phonological aliens. A third aspect of SPE-restoration is the dismissal of Praguian segregation – a genuine finding of Lexical Phonology (see §238): Halle & Vergnaud (1987a) acknowledge distinct pre-and post-word interpretation, but like in SPE hold that these are carried out by the same computational system. Finally, SPE is also restored when Halle & Vergnaud (1987a) grant cyclic interpretation to sequences of morphemes as much as to sequences of words (see §238): recall from §158 that the derivation of words is noncyclic in Lexical Phonology (postlexical phonology is non-interactionist).

220

1.4. New ideas in interface thinking: selective spell-out and interpretationtriggering affixes All this being said, Halle & Vergnaud's model is not just pieced together from the bricks that previous theories put on the market. It also makes a genuine contribution to interface theory: the idea that the concatenation of an affix may or may not trigger interpretation, and hence that morpho-

188 Chap 8: Halle & Vergnaud (1987a): selective spell-out and SPE-restoration syntactic nodes are spelled out only selectively (see §308 for discussion of Bresnan 1971, an early analysis along these lines). This mechanism, though in a different guise as we will see (§§277,304,763), is what is known today as derivation by phase: some nodes (phase heads) do, others (non-phase heads) do not trigger interpretation (although phasehood is not – at least in current syntactic theory – a lexical property of the piece that is merged, an issue that is discussed in §§767,782).

221

1.5. Relevant literature and roadmap The body of work that we are talking about in this chapter is relatively small. It was mentioned at the outset of the chapter that the idea of selective spell-out and affix-triggered interpretation originates in a 1986 manuscript by Morris Halle (Halle 1986), which has never been published, but is sometimes quoted in the literature (not by Halle & Vergnaud 1987a, though).56 In this book, the theory exposed in the present chapter is referred to in terms of its first manifestation in print, i.e. Halle & Vergnaud's (1987a). Halle & Vergnaud (1987a) is a book about stress, not about the interface. The interface as such only plays a role in chapter three, which is called Stress and the Cycle. The related interface theory only really emerged in subsequent work which includes Halle & Vergnaud (1987b), Steriade (1988), Halle (1990), Halle et al. (1991), Halle & Kenstowicz (1991) and Odden (1993). It was applied to various areas, and has a modern offspring; relevant literature includes Halle (1997b), Halle & Matushansky (2006), Halle & Nevins (2009). The following pages discuss the anti-interactionist foundations of Halle & Vergnaud's theory (§222), introduce the new idea, selective spellout (§232), and describe the design properties of its general architecture (§232). In a second step, empirical issues are discussed: affix-ordering, which turns out to be wrong (a fact that offers ammunition for antiinteractionism) (§243), the empirical coverage of Halle & Vergnaud's model in regard of affix class-based phenomena (§248), and conflicting predictions made by the stratal and the non-interactionist architecture (§251). 56

Despite intensive research (that included asking phonologists who were active in the 80s, as well as a request on Linguist List in June 2008, #044441), I could not put my hands on a copy of the manuscript. Morris Halle himself said he cannot find any copy.

Anti-interactionism 189 222

2. Anti-interactionism

223

2.1. Restoration of the inverted T: all concatenation before all interpretation In his Ph.D, Sproat (1985) argues in favour of the idea that morphology and syntax are expressions of the same computational system whose difference lies only the size of the pieces that are concatenated.57 On the other hand, he calls into question the difference between the properties of lexical and postlexical phonological rules. This perspective is incompatible with interactionism where word formation, but not syntactic construction, is interspersed with phonology: either all concatenative activity is interleaved with phonology, or none is. Also, if there is no significant difference between lexical and postlexical rules, there is no reason to provide for a split application of phonology before and after syntax. Sproat (1985) thus attacks the basic architecture of Lexical Phonology from all sides: his target is the Lexicon – the locus of interactionism –, which he argues does not exist. This means that the unique concatenative device – morpho-syntax – must be wholly located before phonology applies. This view thus restores the original interpretation of the inverted T model where all concatenation precedes all interpretation (§86). Table (88) below opposes this view and the macro-architecture of Lexical Phonology. (88) inverted T (SPE) vs. interactionist architecture (Lexical Phonology) a. inverted T model (§86) b. interactionist architecture (LP) (§146) lexical material

syntax

57

lexical material morphology: word forLexicon mation phonology: interpretation of roots and morphemes

morphology

syntax

phonology

postlexical phonology

This is also the central claim of Distributed Morphology, see §533.

190 Chap 8: Halle & Vergnaud (1987a): selective spell-out and SPE-restoration Halle & Vergnaud (1987a:78) quote Sproat (1985) as their source for anti-interactionism (but keep morphology and syntax distinct, as under (88a)). (89)

"We deviate from most proponents of Lexical Phonology in that, following Sproat (1985), we do not assign the rules of morphology-prefixation, suffixation, reduplication, compounding, and so o n – t o particular phonological strata. Instead, we make the traditional assumption that these rules are the province of a special module, the morphology. In our theory, then, as in SPE, morphology is distinct and separate from phonology. Morphology interacts with phonology in that it creates the objects on which the rules of phonology operate." Halle & Vergnaud (1987a:78, emphasis in original)

While this only concerns "most proponents of Lexical Phonology", the anti-interactionist stance has a stronger expression in Halle et al. (1991).58 (90)

"A fundamental tenet of all versions of Lexical Phonology is that affixation processes and other rules of word formation, traditionally thought to make up a separate module of the grammar (namely, the morphology) are interleaved among the rules of phonology. Counterexamples to this interleaving were noted already in Aronoff (1976) but have been widely disregarded. They are taken seriously, however, in EOS [An Essay on Stress, i.e. Halle & Vergnaud 1987a], where morphology is reinstated as a separate component of the grammar, ordered before the phonological component." Halle et al. (1991:142)

Odden (1993:111ff) is another voice that calls for a non-interactionist architecture (non-interactive in his terms). He also provides an informed comparison with the stratal perspective: necessary and optional pieces of both approaches are distinguished. In the 90s, Sproat's view according to which morphology and syntax are expressions of the same computational system will become the central claim of Distributed Morphology (Halle & Marantz 1993 et passim) (the single engine approach, see §536). The issue that Morris Halle had with Lexical Phonology in the 80s, then, is the same as the one that Distributed Morphology has with this theory today: the Lexicon has to go (§533).

58

The counter-examples mentioned in the quote are bracketing paradoxes that invalidate affix ordering. This issue is discussed in §243.

Selective spell-out 191 224

2.2. Interactionism does not imply the Lexicon and is not incompatible with the inverted T An important point for the discussion below (which was already made in §86) is the fact that the inverted T model itself is not incompatible with interactionism. It is only its classical interpretation that rules out interleaving of concatenative and interpretative activity: since Aspects and until Epstein et al. (1998:46ff), Uriagereka (1999) and Chomsky (2000a et passim), it was thought that all concatenation is done before all interpretation. This assumption, however, is independent of the inverted T architecture. We will see in §§304,771 that an architecture where the output of a central concatenative system (morpho-syntax) is interpreted by semantic and phonological computation may well derive a sentence by iteratively piecing chunks together and shipping them off to interpretation. This is the description of Chomsky's (2000a et passim) derivation by phase. Also, there is no necessary relationship between interactionism and the Lexicon: the latter supposes the former, but an interactionist architecture can well live without the Lexicon. Again, derivation by phase instantiates such a system: the derivation is interactionist, but there is no Lexicon in sight.

225

3. Selective spell-out

226

3.1. A new idea: affix-triggered interpretation The headstone of Halle & Vergnaud's (1987a) system is a new idea in interface thinking: affix-triggered interpretation. Affixes are lexically specified for triggering interpretation: upon concatenation, they may or may not provoke spell-out. That is, there are interpretation-triggering and interpretation-neutral affixes. Halle & Vergnaud call the former cyclic, the latter noncyclic.59 The following quote illustrates this point. (91)

59

"I shall assume that whether or not an affix is cyclic is not a property of the morphological rule by which it is assigned, but is rather an idiosyncratic and variable property of the affix." Halle (1986:6), quoted after Kaye (1992a:142)

There is a trap associated to this terminology: non-cyclic affixes do not trigger non-cyclic phonology, which only applies at the word-level when all wordinternal concatenation is completed.

192 Chap 8: Halle & Vergnaud (1987a): selective spell-out and SPE-restoration As a result, spell-out is selective: two pieces may – [[X] Y] – or may not – [X Y] be separated by an interpretational division (which is exactly what today is called a phase boundary). That is, they may or may not constitute an interpretational unit. Which string exactly cyclic affixes trigger the spell-out of is an important, but a secondary question that does not vitiate the basic idea of selective spell-out. In Halle & Vergnaud's system, the string that is spelled out upon the concatenation of a cyclic affix is made of all the material that was previously present in the derivation, plus the material of the affix itself. We will see in §277 below that Kaye (1995) proposes a different take: in his model, which follows the idea of selective spell-out, the concatenation of an interpretation-triggering affix provokes only the spell-out of its sister (i.e. excluding its own material). Applied to (English) affix classes, selective spell-out builds on the class membership of affixes, which is specified in their lexical entry. Class 2 affixes, then, do not trigger spell-out (they are non-cyclic), while class 1 affixes do (they are cyclic). We will see in the following section that this perspective prompts the need for considering morpho-syntactic structure as such, i.e. as a real structure in the morpho-syntactic sense of the word. Recall that Lexical Phonology works without any (or with hardly any) reference to the morpho-syntactic tree: class membership determines at which stratum the affix is merged, and this is all that needs to be understood in order to run the derivation. Otherwise hierarchical morpho-syntactic structure is irrelevant and typically left unmentioned. 227

3.2. Interpretational relevance of affixes percolates to their node Considerations regarding morpho-syntactic structure are absent from Halle & Vergnaud (1987a). As far as I can see, Halle et al. (1991:142) are the first to introduce the idea that the lexically specified class membership of affixes is inherited by morphological constituent structure upon concatenation. While Halle & Vergnaud (1987a) are not really explicit on this issue, Odden (1993:114ff) is unmistakable regarding the role of morphological structure and spell-out: unlike Halle et al. (1991), he shows actual trees and supplements them with affix class information.

Selective spell-out 193

On Halle & Vergnaud's analysis, then, affixes divide into cyclic and non-cyclic items, and this specification percolates to the nodes that dominate them. Table (92) below depicts this organisation.60 (92) Halle & Vergnaud (1987a): analysis of affix classes a. univérs-al-ness b. govern-mént-al γ class 2

γ β

class 1

phon α

x

class 1

phon β

class 2 root

α x

root

spell-out

spell-out

[root - class 1] class 2 [univers-al] ness

[root - class 2 - class 1] [govern-ment-al]

Under (92a), the word [[[univérs]al]1ness]2 illustrates the pattern where a class 1 (cyclic) suffix precedes a class 2 (non-cyclic) suffix, while (92b) shows the reverse order of affix classes, illustrated by the word [[[govern]ment]2al]1. Under (92a), β is thus a class 1 node (a cyclic constituent in Halle & Vergnaud's terminology), whereas γ is a class 2 node (a non-cyclic constituent). Under (92b), the distribution is the reverse. Following Halle & Vergnaud's non-interactionist perspective, the derivation then moves on until the structure of the full sentence is built. Upon completion of the concatenation, spell-out then transforms the tree 60

The representations show an "x" as the sister of roots. This follows the practice in Distributed Morphology (see §543) where roots are lexically unspecified for category (noun, verb, adjective), which is supplied by their sisters. Categorysupplying x's are not conceptually necessary in the present environment and of course are absent in Halle & Vergnaud (1987a) and the related literature. They merely show that affixes are not sisters of roots. As will be seen shortly, this is an explicit requirement of Halle & Vergnaud, since roots must be spelled out in isolation. Also note that according to affix ordering (§142), the sequence class 2 - class 1 affix as under (92b) should not exist. This issue is discussed in §243 below.

194 Chap 8: Halle & Vergnaud (1987a): selective spell-out and SPE-restoration into a bracketed linear string, which is the input to phonological computation. Like in SPE, brackets thus transport morpho-syntactic structure into the phonology. The genuine modification that Halle & Vergnaud (1987a) apply to this system, though, is that the spell-out mechanism ignores class 2 (non-cyclic) nodes: brackets are only inserted at class 1 (cyclic) nodes. Translated into modern (interactionist) terminology, class 1 nodes are phase heads, but class 2 nodes are not. That is, a string that is dominated by a class 1 node will be subject to phonological (and semantic) interpretation, while a string that is dominated by a class 2 node remains uninterpreted.

228

3.3. Underapplication is achieved by selective spell-out of nodes In Halle & Vergnaud's (1987a) system, the difference between phase heads and non-phase heads is converted into contrasting bracketing: univérsalness comes down to phonology as /[univers-al] ness/, while governméntal appears as /[govern-ment-al]/. Like in SPE, interpretational units are thus defined by brackets – but these have been readjusted during spell-out according to the contrast between class 1 and class 2 affixes: the latter is invisible for the spell-out mechanism and therefore leaves no trace in the phonology: a [root+affix 2] string is treated exactly like a mono-morphemic string. The bracketed string is then parsed by cyclic rules (i.e. at stratum 1 in Halle & Vergnaud's terms). In our example, the rules in question assign penultimate stress to every bracket-enclosed interpretational unit. Hence /[univers-al] ness/ will come out as univérs-al-ness, while /[govern-ment-al]/ is interpreted as govern-mént-al. Note that crucially, cyclic rules must not reapply to the entire word in the former case, otherwise *univers-ál-ness would be produced. This is guaranteed by the fact that the entire word is not enclosed in brackets, hence escapes interpretation. This also supposes, however, that the stress rule is absent from wordlevel rules (non-cyclic, stratum 2 in Halle & Vergnaud's terminology), which always apply to the result of morphological concatenation. Underapplication, then, is achieved by the node-sensitive spell-out mechanism: nodes that are skipped do not leave any trace in the phonology. This is the central idea of Halle & Vergnaud's work; it is unprecedented in linguistic thinking (including SPE and Lexical Phonology). I therefore refer to the model as the selective spell-out perspective. This also does justice to the fact that Halle & Vergnaud, like SPE but unlike Lexical Phonology, take

Selective spell-out 195

the morpho-syntactic tree seriously: their spell-out works exactly like modern multiple spell-out, which also skips non-phase heads – the only difference is Halle & Vergnaud's anti-interactionism, which requires translation into brackets prior to interpretation.

229

3.4. The management of English stress

230

3.4.1. Selective spell-out analysis of the párent - paréntal contrast We have already seen an aspect of Halle & Vergnaud's analysis of English stress. Let us now look at the complete mechanism. The reference point is the contrast between párent and parént-al which bear transparent penultimate stress on the one hand, and opaque párent-hood on the other (the Lexical Phonology analysis of this pattern was discussed in §147). The first thing to be considered are roots: the discussion in §226 has only indicated that affixes may or may not trigger spell-out – roots were not mentioned. The fact that bare roots such as párent receive stress supposes that they experience the application of cyclic rules (we already know from the analysis in §226 that stress assignment is a cyclic rule). That is, the structures shown under (92a,b) are incomplete: the root-node α also triggers interpretation, a fact that, after spell-out, leaves the root enclosed by brackets: the root is an interpretational unit of its own. Calling on Halle & Mohanan (1985:66), Halle & Vergnaud (1987a) are explicit on the cyclic status of roots. (93)

"As noted by Halle and Mohanan (1985), in a cyclic stratum 'the relevant phonological rules apply to every morphological constituent in the stratum – to the basic stem […] as well as to every constituent created by morphological processes.' " Halle & Vergnaud (1987a:78)

Given that on Halle & Vergnaud's analysis class 1 affixes are cyclic, but class 2 affixes are not, table (94) below shows the structure of parént-al and párent-hood. Paréntal is thus produced by a two-step derivation: first stress is assigned to [parent], yielding [párent], then the cyclic stress rule reapplies to [párent al], which produces páréntal. In order not to derive a monster with two primary stresses, Halle & Vergnaud (1987a:83) introduce a Stress Erasure Convention (see also Halle & Kenstowicz 1991:460f) which adds a proviso to the stress rule: prior to its application, all previously assigned stresses are eliminated. Hence the application of the stress rule to the outer

196 Chap 8: Halle & Vergnaud (1987a): selective spell-out and SPE-restoration cycle [párent al] first produces [parent al] before the string receives penultimate stress. (94) Halle & Vergnaud (1987a): analysis of affix class-based stress a. parént-al b. párent-hood β al1

phon α

n

phon parent

spell-out [[parent] al]

β hood2

α n

phon parent spell-out

[parent] hood

On the other hand, párenthood identifies as /[parent] hood/ and thus involves only one pass of the cyclic stress rule, which applies to [parent]. This produces the correct result párent-hood. 231

3.4.2. Stress copy: storing before erasing Stress erasure prompts a concern regarding the contrast between condensation and compensation, which until today is an oft-quoted piece of the evidence that SPE has accumulated in regard of English stress (see §546). The classical SPE analysis (§97) crucially needs stress to be carried over from earlier to later cycles: the e of condensation does not reduce to schwa (while the e of compensation does) because it bore main stress on the earlier cycle that produced the existing verb to condénse. By contrast, there is no verb *to compénse. That is, the e of the associated verb to cómpensate was never stressed in its derivational history; it therefore remains unprotected and may reduce (Chomsky & Halle 1968:116). This analysis is incompatible with Stress Erasure. Halle & Vergnaud (1987a:104ff) react by introducing an additional device called Stress Copy, which stores stress information of previous cycles in a parallel structure (the metrical grid) before Stress Erasure takes place (see also Halle & Kenstowicz 1991:490f, who offer more discussion). A further complication is that English features items such as conservátion and consultátion (cf. to consérve, to consúlt) that have the same

The non-interactionist architecture 197

derivational history as condensátion (cf. to condénse), but do allow for the reduction of the pretonic vowel. This contrast (which was unrecorded in SPE) appears to be unpredictable and hence an idiosyncratic property of words. Since secondary stress is supposed to protect vowels against reduction, Halle & Kenstowicz (1991:460f, 490f) propose that Stress Copy, which conserves primary stresses of earlier cycles in the grid-storage, is lexically restricted: rather than being automatic, it applies only on the basis of a lexical diacritic that condensátion does, but conservátion does not possess.

232

4. The non-interactionist architecture

233

4.1. Cyclic vs. word-level (non-cyclic) rules SPE distinguishes cyclic and word-level rules. Recall from §105 that the former iteratively assess every cycle (i.e. every chunk that is delineated by brackets), while the latter apply only once in a derivation, at the word level. Also recall that this distinction was implemented representationally, rather than procedurally (§107): the structural description of word-level rules was flanked by ##. This made sure that they could not apply to strings either below or above the word-level. It also allowed to maintain all phonological rules in one single computational system: there were no multiple minigrammars in SPE. Halle & Vergnaud (1987a) restore this distinction, but abandon the representational management. This option was developed in Halle & Mohanan (1985:66ff), an article that was still couched in regular interactionist Lexical Phonology. The innovation that Halle & Mohanan proposed was a contribution to the debate which rules are cyclic – and hence apply only to derived environments –, and which rules are not restricted in this way. Recall from §188 that Kiprasky's SCC-K held that all lexical rules (i.e. those that contribute to the construction of words) are cyclic, while postlexical rules are not. This generalisation was riddled with counter-examples and finally abandoned by Kiparsky (1993) himself (see §197, more on the different meanings of cyclic in §236 below). In this context, it was already reported in §194 that Halle & Mohanan (1985) introduced the idea that cyclicity is a property of individual strata, rather than of the Lexicon as such. On their analysis, level 1 in English (where class 1 affixes are concatenated) is cyclic, but level 2 (where class 2 affixes are merged) is not. This was the birth of the idea that the

198 Chap 8: Halle & Vergnaud (1987a): selective spell-out and SPE-restoration difference between cyclic and non-cyclic rules is one of serial ordering: the former simply apply before the latter. Non-cyclic rules (which apply at Halle & Mohanan's 1985 non-cyclic stratum) were thus mechanically rejected at the end of the derivation of the word – and hence exactly recover the word-level rules of SPE. Word-level rules in Halle & Vergnaud's (1987a) architecture do the same labour as they did in SPE, and are motivated by the same phenomena. In English, word-level rules are typically related to stress: a number of processes make reference to the location of main stress in a word and thus can only apply once the locus of main stress is determined (see §105). Processes that take the vowel that bears primary stress as a reference point include vowel reduction and stress clash (the Rhythm Rule, Liberman & Prince 1977:309ff). The former reduces certain unstressed vowels (Páris vs. P[ə]rísian), the latter retracts final main stress when the following word has initial stress (thirtéen vs. thírteen mén). Halle & Vergnaud (1987a) and the subsequent literature also discuss cases from other languages where word-level rules do critical labour (see Ziková 2008 for discussion of Halle & Vergnaud's 1987a:82f analysis of Russian yers).

234

4.2. Multiple mini-grammars: morpheme- and chunk-specific phonologies An important question for further discussion is whether Halle & Vergnaud's theory implements multiple mini-grammars or not. Recall that multiple mini-grammars are a landmark property of Lexical Phonology (§§148,212). No doubt the answer is yes since the block of cyclic rules and the block of non-cyclic (word-level) rules constitute two distinct computational systems that are made of different sets of rules. The fact that they do not correspond to any division of the classical stratal architecture is a different question. In other words, the contrast between Halle & Vergnaud's cyclic and non-cyclic rules has got nothing to do with either the pair level 1 vs. level 2 rules or the pair lexical vs. postlexical rules. Table (95) below recapitulates the three pairs of computational systems that we have come across thus far.

The non-interactionist architecture 199 (95) multiple mini-grammars: three different types type specific phonologies for a. morpheme-specific phoaffix classes nologies b. chunk size-specific pho1. chunks below vs. at the word level nologies 2. chunks until the word level vs. beyond

The major line of division that characterises the pairs of computational systems is the property to which they are sensitive: either to specific (classes of) morphemes, or to chunks of a specific size. The former is the basic idea of Lexical Phonology where different affix classes are assessed by different phonologies (level 1 vs. level 2). The latter falls into two categories according to the size of the chunks that are opposed: the most familiar distinction is between word- and sentence phonology (i.e. Praguian segregation); this is what is known as the opposition between lexical vs. postlexical phonology. The other opposes chunks below and at the word level. Both have been implemented in Lexical Phonology, the latter only in some varieties of the theory under the label of post-cyclic lexical rules (Rubach & Booij 1984, see §194). Table (96) below shows which theory implements which opposition. (96) multiple mini-grammars: three different types SPE Lexical Phonology a. morpheme-specific – level 1 vs. level 2 b. chunk size -specific A – lexical cyclic vs. lexi(below vs. at word level) cal post-cyclic (§194) c. chunk size -specific B – lexical vs. postlexical (morpheme vs. word sequences)

Halle&Vergn. – cyclic vs. non-cyclic –

Recall from §106 that the distinction between cyclic and word-level rules existed in SPE, but was managed representationally. As a consequence, SPE formally accommodates only one computational system. Halle & Vergnaud (1987a) took the step of implementing this distinction serially in terms of different blocks of rules, hence admitting the existence of multiple computational systems. However, we will see below that this is the only type of multiple mini-grammars that Halle & Vergnaud (1987a) accept: morpheme-specific effects are handled by cyclic rules alone (§248), and the same computational system assesses sequences of morphemes and sequences of words (§238).

200 Chap 8: Halle & Vergnaud (1987a): selective spell-out and SPE-restoration 235

4.3. The general architecture It was shown in §233 that Halle & Vergnaud's (1987a) revival of SPE's word-level rules in the guise of non-cyclic rules is based on Halle & Mohanan's (1985) non-cyclic strata which were developed in the environment of interactionist Lexical Phonology. It was also mentioned in the previous section that Rubach & Booij (1984) (also in further work: Rubach & Booij 1987, Rubach 1990) had already proposed the existence of lexical, but post-cyclic rules (as opposed to lexical cyclic rules, see §194). All these – word-level rules (SPE), lexical post-cyclic rules (Rubach & Booij 1984) and non-cyclic rules (Halle & Vergnaud 1987a) – are indeed the same thing. The only difference between the two former and the latter is that Rubach & Booij's version of word-level rules come in addition of the regular interactionist strata (level 1 and level 2), while on Halle & Vergnaud's (1987a) count they replace stratum 2. This is shown under (97) below. As may be seen, Halle & Vergnaud (1987a:78) and Halle et al. (1991:142) call the locus where cyclic rules apply stratum 2, while noncyclic rules are active at stratum 3. 61 Stratum 1 accommodates preword allomorphy: this is where allomorphy of the kind that is found in English irregular verbs is computed (e.g. sing - sang - sung). In Chomsky & Halle's (1968:238ff) summary of rules, preword allomorphy corresponds to readjustment rules, which are ordered before cyclic rules (word-level rules are distinguished from cyclic rules by an asterisk in this summary). In sum, thus, Halle & Vergnaud (1987a) reproduce exactly the architecture of SPE. The only thing that is modified is the switch from SPE's representational perspective to the procedural perspective of Lexical Phonology: rather than a representational, word-level rules now have a procedural definition, and boundaries are done away with.

61

But recall from §218 that "stratum" here does not mean the same thing as in Lexical Phonology.

The non-interactionist architecture 201 (97) general architectures with a block of non-cyclic word-level rules (SPE) a. Halle & Vergnaud (1987a) b. Lexical Phonology (incomplete) (Rubach & Booij's 1984 version) underived roots

underived roots

stratum 1 (preword allomorphy) Lexicon stratum 2 cyclic rules interpretation of class 1 nodes (before: level 1)

stratum 3 word-level rules (= non-cyclic rules) (before: level 2)

concatenation of class 1 affixes

interpretation of the resulting string

concatenation of class 2 affixes

interpretation of the resulting string

word-level rules (= lexical post-cyclic rules)

words

236

words

4.4. Terminological pitfalls Given the great many labels, some of which refer to the same thing, it may be useful to follow up on (96) in order to recapitulate the situation. Table (98) below shows relevant equivalences, also anticipating on the discussion below where more vocabulary further muddies terminological waters. One more terminological pitfall is the word cyclic, which means various things in different theories. Originally, SPE opposed cyclic and word-level rules because only the former were working through the bracketed string: cyclic rules apply iteratively to all chunks of the string that are delineated by brackets – three times for example when the string [[[A]B]C] is interpreted. On the other hand, word-level rules apply only once after the parsing of the last cycle. They are thus non-cyclic in the sense that cycles

202 Chap 8: Halle & Vergnaud (1987a): selective spell-out and SPE-restoration are irrelevant for their application, which is triggered by ## (see §105). In SPE, being cyclic for a rule thus merely means to apply to cycles. (98) equivalences across theories

Lexical Phonology

class 1 class 2 readjustment + # rules – level 1 rules level 2 rules

Halle & Vergnaud (1987a) Kaye (1995) Stratal OT

preword allomorphy stratum 1 – –

SPE

cyclic rules

–

word-level word-level rules lexial post-cyclic (Rubach & Booij 1984) (§194) non-cyclic rules

stratum 2 stratum 3 non-analytic analytic – stem-level word-level –

In (classical) Lexical Phonology, the word cyclic is used in a completely different way. It now is a synonym of lexical: all rules that apply in the Lexicon are cyclic, as opposed to postlexical rules, which apply in noncyclic fashion. In other words, cyclicity simply means that the rule in question is interactionist: lexical rules are interspersed with concatenative activity, but postlexical rules are not. This sense was coupled with another meaning. In classical Lexical Phonology, all lexical/cyclic/interactionist rules obey Kiparsky's Strict Cycle Condition (SCC-K, see §188), which means that they apply only to derived environments. By contrast, postlexical rules apply across the board. In Lexical Phonology, then, the original meaning that the word cyclic had in SPE is entirely inoperative: given interactionism, there are no brackets (Mohanan's brackets are not the same thing, see §171) that could delineate cycles (§161). Instead, the relevant interpretational units are strata (§159). Since specific mini-grammars apply to the different strata, there is no type of rule that could iteratively apply to all interpretational units. In other words, the stratal architecture and multiple mini-grammars leave no place for the SPE-concept of cyclic rules. In (classical) Lexical Phonology, lexical, cyclic, interactionist and "applying only to derived environments" were thus synonymous at first – until Rubach & Booij (1984) and Halle & Mohanan (1985) introduced rules that were lexical, but not cyclic. What both Rubach & Booij's lexical postcyclic rules and Halle & Mohanan's rules at the non-cyclic stratum have in common is that they do not obey strict cyclicity. In the two approaches at hand, then, cyclic rules are necessarily lexical, but all lexical rules are not

The non-interactionist architecture 203

cyclic. Being a cyclic rule, however, still means to be interactionist and to apply only to derived environments. Halle & Vergnaud (1987a) now restore the architecture and the tools of SPE, also regarding the word cyclic: just like in SPE, a cyclic rule in their model is simply a rule that applies to cycles. Given the noninteractionist architecture, cycles are defined by brackets, exactly like in SPE. At this point it may only be hoped that the panorama provided helps to disentangle terminological pitfalls, and does not further confuse the reader. Bringing up terminology in this context is inescapable: anybody who reads through the relevant literature will sooner or later be confronted to this question. 237

4.5. No look-back devices While the Lexicon and its stratal interactionist architecture is dismissed by Halle & Vergnaud altogether, Halle & Vergnaud (1987a) (see also Halle et al. 1991:141f) take over the Strict Cycle Condition from Lexical Phonology: "Strict Cyclicity governs the application of rules in cyclic strata but not elsewhere" (Halle & Vergnaud 1987a:78, emphasis in original). The rules that do not obey strict cyclicity are thus the same as in Lexical Phonology: word-level rules (for those versions of Lexical Phonology that provide for them) and postlexical rules. In Halle & Vergnaud (1987a) and subsequent publications, the no look-back device is always referred to as strict cyclicity, rather than as Kiparsky's well-known Strict Cycle Condition. One might thus suspect that there is a difference, but this turns out not to be the case: what Halle & Vergnaud implement is the SCC-K (rather than the SCC-M or some other no look-back mechanism). For example, Halle & Vergnaud (1987a:80) follow the classical analysis according to which Trisyllabic Shortening (§164) does not apply to items such as ivory and nightingale because the alternation is managed by cyclic rules, which according to the SCC-K (but not according to the SCC-M 62 ) cannot apply to mono-morphemic items (see §189). 62

Recall from §§189f that the difference between Mascaró's (1976) SCC-M and Kiparsky's SCC-K are derived environment effects: the SCC-K prevents processes from applying to mono-morphemic items (roots), while the SCC-M does not.

204 Chap 8: Halle & Vergnaud (1987a): selective spell-out and SPE-restoration Given the well-known counter-examples (§186), they also follow Kiparsky (1982b:160ff) in assuming that the SCC-K is only violated by structure-changing rules: if a rule merely adds items to an autosegmental representation, it is structure-building and does not conflict with strict cyclicity (see §193). An example that demonstrates this contrast is stress assignment in English and Vedic. In the former language, stress is not distinctive,63 which means that it is absent from lexical representations. The stress-assigning rule then adds metrical grid marks: this is adding new structure, rather than modifying existing structure. Therefore the SCC-K does not prevent stressassignment from applying to underived roots such as párent. By contrast, stress is distinctive in Vedic, which means that it is encoded in lexical representations. Stress rules that apply to underived roots thus meet existing grid structure, which would be modified by their application. Therefore, Halle & Vergnaud (1987a:86f) argue, the SCC-K blocks the application of stress rules to underived roots in Vedic. The question which phonological processes may and which may not look back into previous cycles is resident in the literature on no look-back devices. Whether the structure-building vs. structure-changing division is the correct generalisation is further discussed in §§554,780.

238

4.6. Distinct pre- vs. post-word phonology yes, Praguian segregation no It was mentioned in §218 that Halle & Vergnaud (1987a) do not subscribe to Praguian segregation: the same interpretational device – the cyclic and non-cyclic block of rules – assesses both sequences of morphemes and sequences of words. Hence there is no such thing as a distinct word- and sentence phonology. This notwithstanding, Halle & Vergnaud (1987a) do acknowledge that the word is an autonomous unit which enjoys an independent pronunciation and an independent meaning. Like Lexical Phonology, thus, Halle & Vergnaud consider that there are two interpretational stages: first morphemes are interpreted; the computation is then interrupted when the word 63

This is shorthand for a more complicated situation: récord (noun) and recórd (verb) is a stress-based minimal pair. However, Halle & Vergnaud (1987a:105) consider that "in languages such as English and Spanish, where stresses need to be indicated in lexical representations only exceptionally, not systematically as in Vedic or Lithuanian, stress is not distinctive and the cyclic stress rules of English therefore apply freely to underived stems."

The non-interactionist architecture 205

level is reached; the string at hand is assessed by a specific computation – word-level rules (the non-cyclic block in Halle & Vergnaud's terminology) –, which "seals" the unit and makes it inaccessible for further modification (more on this in the following section). Interpretation then resumes in order to assess larger chunks, that is sequences of words. Interpretation thus proceeds in two steps like in Lexical Phonology where lexical and postlexical phonology is distinguished with the same delineation at the word level. But unlike in Lexical Phonology, the content of "lexical" and "postlexical" phonology is the same in Halle & Vergnaud's architecture. This system is implemented into the formal architecture by simply adding extra "strata" (in the sense of Halle & Vergnaud) to the general architecture. The version under (97) above is incomplete: it only shows the window up to the word level. Table (99) below completes the picture (Halle & Vergnaud 1987a:78, Halle et al. 1991:141f). In this system, strata 2 and 3 on the one hand and strata 4 and 5 on the other have the same content: the rules that apply at strata 2 and 4 (cyclic), as well as at strata 3 and 5 (non-cyclic), respectively, are the same. The only difference between strata 2/3 and 4/5 is the fact that a chunk of a specific size – the word – has been selected at which the cyclic interpretation (of stratum 2) is interrupted. Halle & Vergnaud (1987a) thus restore the classical picture, once again: Praguian segregation was unknown in SPE. And they restore the SPE-setup yet in another respect: all chunks are subject to cyclic interpretation, no matter what their size. Recall from §102 that Chomsky & Halle (1968:27) hold that "the transformational cycle [is formulated] in full generality, applying to all surface structure whether internal or external to the word." Contrary to this principle, Lexical Phonology had introduced selective cyclic interpretation, which applied to sequences of morphemes (lexical phonology), but not to sequences of words (postlexical phonology) (see §158).

206 Chap 8: Halle & Vergnaud (1987a): selective spell-out and SPE-restoration (99) Halle & Vergnaud (1987a): general architecture (complete) underived roots stratum 1 preword allomorphy stratum 2 cyclic rules (interpretation of class 1 nodes) word-internal strata stratum 3 non-cyclic rules (= word-level rules) words stratum 4 cyclic rules word-sequence strata stratum 5 non-cyclic rules sentences

239

4.7. The word as an autonomous phonological unit

240

4.7.1. The word is sealed – an insuperable barrier for some processes It was mentioned in the previous section that the word is "sealed": its phonological properties are constructed during the interpretation of morpheme sequences, but cannot be further modified when word sequences are interpreted. This is a basic insight from SPE which was discussed in §105: in English stress-assignment is cyclic and applies to all cycles until the word level. Once this chunk-size is reached, primary stress is identified, and secondary, ternary etc. stresses are distributed according to this anchor. Vowel reduction is also effected according to primary stress. All this is done by word-level (non-cyclic in Halle & Vergnaud's terminology) rules which do not apply until the word level is reached. Chomsky & Halle (1968:133f) insist that word-level rules do not apply to chunks that are bigger than a word: in the whole derivation of a sen-

The non-interactionist architecture 207

tence, they are active exactly once – at the word level. The basic observation that motivates this is the inactivity of the process that determines word stress at larger chunks: whatever happens when phonology is done on word sequences, stress placement does not reapply, and the same is true for the vowel reduction that depends on it. That is, a word will have the same stress pattern no matter whether it receives strong or weak intonation (phrasal stress): word stress and intonation are independent.64 Therefore phonological theory must somehow guarantee that at least some phonological properties of the body of words cannot be further modified when phonology is done on word sequences. This is a typical motivation for Praguian segregation: the rules that apply to sequences of morphemes (word phonology) are not the same as the rules that apply to sequences of words (sentence phonology). In Lexical Phonology, thus, the solution is simple: the stress-assigning rule is present in the Lexicon, but absent from postlexical phonology. Systems that reject Praguian segregation are in trouble: on the assumption that the same computational system interprets morpheme- and word sequences, the stress rule that applies to the former will also have to be active when the latter are assessed. As far as I can see, none of the theories that implement the continuity of phonological interpretation between morphemes and words addresses this issue, let alone offers a mechanism that "freezes" words. This is true for SPE, Halle & Vergnaud (1987a), Kaye (1995) (on which more in the following chapter) and Distributed Morphology. 241

4.7.2. Selective impact of no look-back? In practice, what needs to be eluded in the English example is the application of the stress rule when the pieces [fóllows] and [Jóhn] of the sentence Peter follows John are interpreted. The rule that assigns word stress is blind for the status of the pieces that it considers. Hence [fóllow-Jóhn] should be treated exactly like [órigin-al]: main stress is calculated anew, and vowels that bore main stress on earlier cycles are destressed. That is, the first vowel of fóllow should lose its main stress and we should observe *follów John, just like we observe oríginal. 64

Word stress itself may be modified upon the concatenation of another word (stress clash, thirtéen vs. thírteen men), but this is done by a different phonological process: the point is that the process that places word stress does not reapply to larger chunks.

208 Chap 8: Halle & Vergnaud (1987a): selective spell-out and SPE-restoration In absence of Praguian segregation, what is thus needed is a way of making words immune against further modification – a no look-back device. This device, however, must make a difference between the wordinternal and the word-external situation: stress notoriously violates no lookback word-internally (the concatenation of another suffix erases previous stress), but strictly observes no look-back beyond the word level (no further concatenation can erase previous stress). For the sake of completeness, let us also mention that of course there are phonological processes which apply across word boundaries, i.e. where the patient belongs to one word, and the trigger to another. Flapping in American varieties of English is a well-known case in point (on which more in §821): it concerns /t/ in city as much as in invite Olivia. If words trigger phases and are reputed to be "sealed", it must be concluded that the sealing-effect applies selectively to processes: while stress assignment is blocked, flapping freely applies to word-internal material (see BalognéBérces 2004, 2005 on this issue). Again, Praguian segregation offers a simple solution: the stress rule will simply be absent from postlexical phonology, while flapping will be present both in lexical and postlexical phonology. The take-home message at this point is that the word appears to be a barrier for some processes, but not for others, and all theories need to be able to account for this process-specific sealing effect. Praguian segregation offers a straightforward solution, which is not available in systems that provide for the interpretation of sequences of morphemes and words by the same computational system. These approaches need a sophisticated no look-back device that applies selectively to different phonological processes. Poser (1986, also 1989:135ff) is the earliest source that I could identify where process-specific no look-back is proposed. The processspecificity of no look-back is further discussed in §§296,302,823 (see also Scheer 2009a,b,c). 242

4.8. Interpretational units Let us now have a brief look at how interpretational units are defined in Halle & Vergnaud's (1987a) theory. Interpretational units are the chunks of the linear string that are interpreted at the same time. Recall from §103 that the interpretational units of SPE, bracket-delineated cycles, are identical to morphemes, except that two successive morphemes of the same major

Anti-interactionist ammunition: bracketing paradoxes 209

category (noun, verb, adjective) cohabitate in the same cycle. Hence theatricality identifies as [[[theatr]N ic + al]A i + ty]N. In Lexical Phonology, strata are interpretational units (§159). That is, all affixes of the same affix class are interpreted together. A root with two class 1 affixes such as origin-al1-ity1 for example represents just one interpretational unit: [origin-al-ity] (while it identifies as three units in SPE: [[[origin]N al]A ity]N). Here as well, Halle & Vergnaud (1987a) restore the take of SPE – but it is not clear to me whether they abandon the proviso regarding major categories. On the one hand, they are explicit about the fact that indeed they do: every morpheme is a new cycle. (100) "As noted by Halle and Mohanan (1985), in a cyclic stratum 'the relevant phonological rules apply to every morphological constituent in the stratum – to the basic stem […] as well as to every constituent created by morphological processes.' " Halle & Vergnaud (1987a:78)

On the other hand, however, they work with representations that embody the state of affairs in SPE: for example, one may come across the word ungrammaticality, which on Halle & Vergnaud's (1987a:81) account identifies as [[un-[grammat + ic + al]A ]A ity]N. 243

5. Anti-interactionist ammunition: bracketing paradoxes

244

5.1. Affix ordering turns out to be wrong: bracketing paradoxes Interactionism (supplemented with level ordering) enforces affix ordering (§147), Siegel's (1974) generalisation according to which class 1 affixes never occur outside of class 2 affixes. Recall from §146 that Siegel's discovery played a crucial role in the establishment of the stratal architecture. Affix ordering is therefore a prime target for anybody who wants to do away with interactionism: in its absence, there is no reason for affixes to observe a specific order. This is precisely Halle & Vergnaud's line of attack: they call on work by Aronoff (1976) and Aronoff & Sridhar (1983, 1987) who have collected empirical evidence against affix ordering since 1976 – so-called bracketing paradoxes –, but which for some reason did not have much impact. Kiparsky (1982a:28f) for example acknowledges their existence, but has got nothing much to say, except that they are lexically marked as exceptional. The empirical argument along the lines of the counter-examples that Mark

210 Chap 8: Halle & Vergnaud (1987a): selective spell-out and SPE-restoration Aronoff has accumulated since 1976 is made in Halle & Kenstowicz (1991:459) and Halle et al. (1991:142) (Halle & Vergnaud 1987a:77 contains only a hint at Aronoff & Sridhar 1983). Today there can be no doubt that affix ordering is indeed empirically unsustainable: it is simply a wrong generalisation. Bermúdez-Otero (forth a:29f) provides a documented overview of the issue from a post-hoc perspective. Bracketing paradoxes are thus cases where a class 2 affix occurs closer to the root than a class 1 affix. Items of this kind are actually not infrequent; some examples appear under (101) below. (101) affix ordering: counter-examples a. patent-abíl2-ity1 b. develop-mént2-al1 c. organ-iz2-at-ion1 d. un-2grammatic-al1-ity1

Under (101a-c), the class 2 suffixes -able, -ment and -ize occur before the class 1 suffixes -ity, -al and -ion. Class membership of affixes is identified by their behaviour in regard of stress (which is taken to be a conclusive diagnostic in the literature): the former are stress-neutral, while the latter are stress-shifting. The most famous bracketing paradox is certainly (101d) un-grammatic-al-ity: the class 1 affix -ity attaches to un-grammatical, which already contains the class 2 affix un-. That un- is concatenated to grammatic-al, rather than to the fully suffixed grammatic-al-ity is warranted by two observations: 1) we know independently that un- only attaches to adjectives, and 2) ungrammaticality means "the fact of being nongrammatical", not "non-grammaticality" (hence un- has scope only over grammatical). 245

5.2. Dual membership, optionality, overgeneration Another argument that may be brought to bear against affix ordering is socalled dual membership, i.e. the fact that some affixes seem to belong to several affix classes. Affixes may indeed show class 1 behaviour in regard of some phenomenon, but come along as class 2 affixes on other occasions. Cases in point are -ment and -ation (see the detailed discussion in Giegerich 1999:21ff and Szpyra 1989:40ff, 186ff).

Anti-interactionist ammunition: bracketing paradoxes 211

Minimal pairs such as cómparable "roughly the same" vs. compárable "able to be compared" also illustrate dual membership: stress shift is usually taken to be a conclusive diagnostic for class 1 membership (class 2 affixes are stress-neutral). Accordingly, there should be two -able suffixes, one in each class (more on this issue in §545). Dual membership of course weakens the empirical content of affix ordering. Optionality is also an issue: Trevian (2007:435) points out that some reportedly stress-shifting affixes (púpil - pupíll-ary) may or may not behave along these lines (free variation) when combined to certain stems: frágment - frágment-ary or fragmént-ary; dísciplin - dísciplin-ary or discíplin-ary. In addition to these questions, other empirical and conceptual objections have been raised against the interactionist architecture of Lexical Phonology, for which the analysis of English affix always played an important role. Giegerich (1999:21ff), McMahon (2000:55ff), Marvin (2002:74ff) and Bermúdez-Otero (forth a:§2.5) provide more detailed material from a settled post-hoc perspective.

246

5.3. Halle & Vergnaud's analysis of bracketing paradoxes Since there is no interactionism in Halle & Vergnaud's system, there is no reason why affixes of both classes should not freely combine in any order. We have already seen in §226 how both class 1 - class 2 and class 2 - class 1 sequences are analysed according to Halle & Vergnaud's selective spellout: univérs-al1-ness2 and govern-mént2-al1 identify as /[univers-al] ness/ and /[govern-ment-al]/, respectively. Stress is then simply penultimate within the domain that is delineated by brackets. Let us now see how the notorious un2-grammatic-ál1-ity1 is derived (Halle & Vergnaud 1987a:81). Recall from §244 that for semantic (un- has only scope over grammatical) and selectional (un- attaches only to adjectives) reasons the morpho-syntactic structure must be [[un2 [[grammatic] al1]A]A ity1]N, which according to class-sensitive selective spell-out of affix boundaries is handed down to phonology as /[un [grammatic-al] ity]/ (recall that each class 1 node produces an interpretational unit). Antepenultimate stress then applies first to the inner domain (yielding grammátical), and on a second pass (after stress erasure) to the outer domain, which produces antepenultimate ungrammaticálity. In interactionist Lexical Phonology, bracketing paradoxes have been treated by carrying a representational component into the analysis. §440 shows how Rubach & Booij (1984) make use of the Prosodic Hierarchy:

212 Chap 8: Halle & Vergnaud (1987a): selective spell-out and SPE-restoration the morpho-syntactic structure (where affix ordering is respected) is transformed into prosodic constituency (where affix ordering is not respected); phonological rules then make reference to the latter, rather than to the former.

247

5.4. Affix stacking generalisations as selectional restrictions or parsingbased (Fabb 1988, Hay 2002) An entirely different outlook on affix class-based phenomena is profiled by Fabb (1988) and further developed by Hay (2002), Hay & Plag (2004) and Plag & Baayen (2009). The initial fact observed by Fabb (1988) is that affix ordering quite largely overgenerates: given the number of suffixes, their class membership and all other combinatorial restrictions, there should be 459 possible suffix pairs. The number of attested pairs, however, only amounts to fifty. Based on this observation, Fabb (1988) argues that Siegel's (1974) affix ordering generalisation is certainly too strong and empirically wrong, but that this does not mean that there are no restrictions on affix combination (affix stacking) at all. There are robust generalisations, and grammar must somehow be at the origin of the restrictions observed. He suggests that the mechanism which decides whether an affix can attach to a string is based on the selectional properties of this affix: individual affixes may (or may not) be sensitive to the fact that a potential host is already affixed. That is, a large number of affixes do not attach to already affixed words. Hay (2002), Hay & Plag (2004) and Plag & Baayen (2009) have further developed this idea, shifting it on extra-grammatical grounds: they argue that affix stacking is controlled by parsing and perception along a hierarchy of processing complexity, rather than by selectional properties. On their view, the degree of internal complexity of a host that an affix tolerates is determined "by how much structure that affix itself creates. Phrased in terms of processing, an affix that can be easily parsed out should not occur inside an affix that cannot" (Hay 2002:528, emphasis in original). Plag & Baayen (2009) complement this strand by psycholinguistic evidence (lexical decision, word naming) which they argue shows that productivity and memory are also relevant factors. This approach thus first insulates affix class-based phenomena from phonology (selectional properties do not explain why there is a classspecific phonological effect), and then from grammar altogether (perception and parsing are the reasons why some combinations do, but others do

Empirical coverage of Halle & Vergnaud's selective spell-out 213

not occur). Whether this is the correct way to go about affix class-based phenomena is left an open question in the book.

248

6. Empirical coverage of Halle & Vergnaud's selective spell-out

249

6.1. Analysis of the rule-blocking pattern (level 1 rules) We have seen how Halle & Vergnaud's system covers English stress: the selective spell-out analysis is the same whatever the order of affixes. Also, the opacity of párent-hood is derived with the same mechanism (§230). Recall from §164 that English stress assignment is a representative of the rule-blocking pattern, which in Lexical Phonology translates as level 1 rules. In Halle & Vergnaud's system, all other rule-blocking phenomena are analysed along the same lines. That is, relevant rules are always cyclic, but absent from the non-cyclic (word-level) block. The structure that selective spell-out of morpho-syntactic structure produces, then, is always the same. Trisyllabic Shortening for example applies to san-ity1, but not to maiden-hood2: the former reaches phonology as /[[san] ity]/, while the latter comes down as /[maiden] hood/. The trisyllabic condition prevents the rule from applying to [san], the inner cycle of sanity (as much as to the bare root sane). Since the whole word is a cycle as well, though, Trisyllabic Shortening tries to reapply at the word level, and this time goes into effect because the trisyllabic condition is met. By contrast, the only cycle of maidenhood, [maiden], does not allow for the application of the rule, and the word as a whole is not a cycle, which prevents Trisyllabic Shortening from reapplying. The same mechanism correctly derives the nasal assimilation of in-, against the non-assimilation of un-. Im-1possible identifies as /[in [possible]]/, while un-2predictable enters phonology as /un [predictable]/. The cyclic rule of nasal assimilation then applies to the former because the prefix together with the root is a cycle and hence experiences common interpretation; by contrast, there is no outer cycle in the latter, which means that the string which contains the /…np…/ sequence is never assessed by nasal assimilation. Note that it is crucial for this analysis of the rule-blocking pattern that the relevant rules are absent from the word-level (non-cyclic) block.

214 Chap 8: Halle & Vergnaud (1987a): selective spell-out and SPE-restoration 250

6.2. No solution for the rule-triggering pattern (level 2 rules) Let us now look at rule-triggering boundaries (level 2 rules, see §166). As a matter of fact, Halle & Vergnaud's system is as unsuited for the analysis of this pattern as the stratal architecture of Lexical Phonology where, recall, additional machinery is needed (Mohanan's brackets or the SCC-K, see §§168,203). The cyclic structure that Halle & Vergnaud's affix class-sensitive selective spell-out produces for the three relevant items sign, sign-ing2 and si[g]n-ature1 is /[sign]/, /[sign] ing/ and /[[sign] ature]/, respectively. The cluster-reducing rule must be absent from the word-level (noncyclic) block: otherwise sign-ing and sign-ature will experience the same treatment. That is, no inner brackets are left at the word-level (the noncyclic stratum); therefore bare unstructured strings are assessed. A rule such as g → ø / __n will thus reduce the cluster of sign-ature as much as of sign-ing (and sign). Making the rule cyclic does not help either. It will be blocked from applying to the root-cycle by the SCC-K to which Halle & Vergnaud subscribe (§237). The blocking effect is real since the cluster reducing rule is structure changing (it deletes an entire segment), not structure building (see §237). This means that the bare root sign will remain unreduced (also at the word level from which the cluster reduction rule must be absent as we have seen) – this is an incorrect result. Worse, Halle & Vergnaud's system distinguishes class 1 si[g]n-ature1 and class 2 sign-ing2 by contrasting cyclic structure, but the result after computation is exactly the reverse of what is attested: /[sign] ing/ has no outer cycle and will thus remain unreduced *si[g]n-ing, while the complete string of /[[sign] ature]/ undergoes cyclic computation at the outer cycle, where cluster reduction goes into effect and produces *sign-ature. This "reverse result" is characteristic for Halle & Vergnaud's system: it indicates that in fact the method is correct – selective spell-out –, but that it may need to be applied in the opposite way: class 2 affixes do, class 1 affixes do not insert brackets into the phonological string. Distribution of brackets along these lines would produce /[[sign] ing]/ and /[sign] ature/; after computation, the cluster would be reduced in the former case because there is an outer cycle, but would remain untouched in the latter since there is no computation of the complete string.

Stratal vs. non-interactionist architecture: two testing grounds 215

We will see in §281 that this is exactly the direction option taken by Kaye (1995): selective spell-out yes, but reverse distribution of brackets (today one would say: of phase heads). In sum, then, Halle & Vergnaud's system seems to be structurally unable to come to grips with the rule-triggering pattern. The situation is thus parallel to one that prevails in Lexical Phonology: in both cases the basic mechanism (i.e. the stratal architecture there, selective spell-out here) can cover the rule-blocking pattern without problem, but fails when it is confronted with the rule-triggering pattern. Unlike in the case of Lexical Phonology, however, the additional machinery that was proposed will not help out Halle & Vergnaud. The string cannot be enriched by Mohanan's brackets (§168) at every morpheme boundary since the principle of Halle & Vergnaud's system is precisely to spell-out morphological boundaries selectively. The other tool that can account for the rule-triggering pattern in Lexical Phonology, the SCC-K (§203), will not help either since it is already operative. As far as I can see, the rule-triggering pattern is not discussed in the literature that subscribes to Halle & Vergnaud's (1987a) selective spell-out model at all. In Halle & Mohanan (1985:62f), nasal cluster simplification is analysed along Mohanan's bracket-sensitive rules that are not an option anymore two years later in Halle & Vergnaud's anti-interactionist environment. The silence regarding the fact that a well-known pattern cannot be accounted for by Halle & Vergnaud's theory is strange and does not help finding out about the comparative merits of the theories of procedural interface management.

251

7. Stratal vs. non-interactionist architecture: two testing grounds

252

7.1. Syntactic information is or is not available when words are build The stratal architecture and Halle & Vergnaud's non-interactionist model contrast in a number of properties that may give rise to different predictions. Two such testing grounds are discussed in the literature in some detail: the possibility for derived phonological properties to bear on the concatenation of morphemes, and the availability of syntactic information for the construction of words. Let us first consider the latter. Lexical Phonology makes the prediction that syntactic information cannot be used for the construction of words: morphological activity takes place in the Lexicon, which is ordered

216 Chap 8: Halle & Vergnaud (1987a): selective spell-out and SPE-restoration before syntax (§152). By contrast, syntax precedes morphology in the traditional view that Halle & Vergnaud (1987a) subscribe to. The contrasting predictions are the central issue of Odden (1993:117ff), who presents a case from Kimatuumbi where certain rules that must be lexical are also informed by syntactic properties. As far as I can see, this strand of the discussion was not further developed in the literature.

253

7.2. Phonology-free syntax: is morphological concatenation sensitive to derived phonological properties? On the backdrop of Zwicky & Pullum's (1986a,b) principle of Phonologyfree Syntax (which is discussed at greater length in §412), Lexical Phonology has identified another testing ground. On interactionist assumptions, morphological concatenation can be done after some phonological rules have applied (to earlier strata). Therefore, morphological activity may be sensitive to derived phonological properties, i.e. to properties that are created by phonological computation (but are absent from the lexicon). By contrast, in the traditional non-interactionist perspective where all concatenation is completed before phonological interpretation begins, morphosyntax can never be influenced by derived phonological properties. These contrasting predictions are run against the empirical record in the state-of-the-art volume edited by Hargus & Kaisse (1993): Booij & Lieber (1993), Booij (1997:262f) and Inkelas (1993) argue in favour of interactionism (also Szpyra 1987). Especially Hargus (1993) collects relevant cases where derived phonological properties appear to influence morphology, but Odden (1993) proposes a reanalysis in a non-interactionist perspective. While the predictions on both sides are sufficiently distinct and explicit, their confrontation with the empirical record appears to be inconclusive, at least on the basis of the literature mentioned.

254

8. Conclusion

255

8.1. SPE with Lexical Phonology freckles One result of the discussion above is that Halle & Vergnaud's (1987a) model is more akin with Lexical Phonology in the self-understanding of the

Conclusion 217

authors than in actual fact. That is, it is rather misleading to think of Halle & Vergnaud's theory as a non-interactionist version of Lexical Phonology: Halle & Vergnaud (1987a) dismiss all key properties of the dominant interface framework of the 80s, i.e. of course interactionism (and hence strata, despite the fact that they also talk about "strata", see §218), but also Praguian segregation and the management of affix class-based phenomena by multiple mini-grammars. On the other hand, Halle & Vergnaud (1987a) adopt two ideas that were coined by Lexical Phonology: the proceduralisation of the interface and the SCC-K. That is, boundaries have as much disappeared from the analysis of interface phenomena in Halle & Vergnaud's perspective as they have in Lexical Phonology. This cut-down of representational communication is at variance with the situation that prevailed in SPE (see §105). Halle & Vergnaud (1987a) also take over Kiparsky's (1982a,b) version of the Strict Cycle Condition, which roots in Chomsky (1973) and Mascaró (1976), but adds the concern for derived environment effects (§§189f). A characterisation of Halle & Vergnaud's project that strikes closer to the mark is certainly one that mentions SPE: the basic goal is to restore the SPE-setup. Without really being explicit on that, Halle & Vergnaud put back to office a non-interactionist architecture ("all concatenation before all interpretation"), the distinction between cyclic and word-level rules, the cyclic interpretation of word sequences (which was abandoned by the noncyclic conception of postlexical phonology in Lexical Phonology), the use of brackets (i.e. untranslated morpho-syntactic information in phonology), preword allomorphy and the SPE-definition of what counts as an interpretational unit. 256

8.2. Selective spell-out is a groundbreaking idea Beyond the question how much genetic material Halle & Vergnaud's model has inherited from SPE, and how much is due to Lexical Phonology, its genuine contribution to interface theory needs to be praised: selective spellout. We will see below that this idea has become part and parcel of the generative interface architecture today (§§304,763): in modern terminology, not every node is a phase head, and only phase heads trigger interpretation. Interestingly, the idea itself says little about its implementation: once it is understood that not all nodes are spelled out, it needs to be decided which nodes exactly do, and which do not trigger interpretation. Under the

218 Chap 8: Halle & Vergnaud (1987a): selective spell-out and SPE-restoration label of phasehood, this is an important issue in current syntactic debate (see §771). We have seen that Halle & Vergnaud's choice to make class 1 nodes cyclic (phase heads), while class 2 nodes are transparent for interpretation, produces correct results for the rule-blocking pattern, but fails when confronted with the rule-triggering pattern (§250). The following chapter shows that completely different results are produced when 1) the root is not made a cycle per se, 2) the same (i.e. cyclic), rather than a different (noncyclic) phonology reapplies at the word level, 3) a no look-back device does actual labour in the derivation of affix class-based phenomena and 4) this device is reduced to its original expression, that is the SCC-M. The system described is Kaye's (1995): selective spell-out yes, distinct word-level phonology no, derived environment management by the no look-back device no.

257

8.3. Distinct computational systems yes – but which ones? Multiple mini-grammars will be a resident issue in the book. The question, however, is not only whether phonology should accommodate distinct computational systems – it is also, and perhaps foremost, which type of multiple mini-grammars should be provided for. A distinction that will prove relevant was introduced in §234: distinct computational systems may be specific to morphemes (or morpheme classes), or they may be sensitive to the size of the chunks that are assessed. This issue is related to the status of the word as a chunk size that requires particular care. Halle & Vergnaud (1987a) admit the distinction between pre- and post-word phonology (in continuity of Lexical Phonology, §238), but reject Praguian segregation (in discontinuity of Lexical Phonology). This prompts the issue of the word as "natural barrier" for phonological processes (§239). It is a characteristic property of theories which reject Praguian segregation – hence where the same computational system assesses morphemeand word sequences – to be in trouble when the application of wordinternal processes is blocked across words (as is the case for English word stress). Halle & Vergnaud (1987a) do not explain how they handle this phenomenon any more than more recent theories which provide for an interpretational continuity of morphemes and words. Another case in point, Kaye (1995), is discussed in the following chapter.

258

Chapter 9 Kaye (1995): selective spell-out and modificationinhibiting no look-back

259

1. Introduction

260

1.1. Editorial note The following pages attempt to provide a comprehensive overview of how the interface has been thought of in Government Phonology. Or rather, by Jonathan Kaye, who has initiated and worked out relevant ideas. According to him, a conference presentation together with Jean-Roger Vergnaud (Kaye & Vergnaud 1990) was the initial spark of the programme that was then developed throughout the 90s. A pervasive problem is that Kaye has only published one article on the subject, Kaye (1995), and that this article contains only a subset of the ideas and the empirical material that were developed over the years. True, his 1989 book (Kaye 1989) gives the direction of much of the enterprise (parsing cues, following Trubetzkoy's Grenzsignale), and Kaye (1992a) is also relevant (regarding affix-triggered interpretation and the φ-function). These sources, however, are either programmatic (the book) or marginal for the purpose (the 1992a article). One can then rely on secondary sources, that is work by SOAS students of the 90s who summarise Kaye's interface design on a few pages at the outset of their own articles and dissertations. The available material here, however, is also sparse (e.g. Ploch 1996:76f, Cobb 1996:30f, 1997). In short, anybody who tries to reconstruct the complete picture of Kaye's interface theory from sources that are available in print will be given a hard time. This is especially true for the sections on parsing cues and lexical access (§§340,346). The presentation below therefore relies on unpublished sources as well: Kaye's teaching at the EGG Summer School (Niš 2001, Novi Sad 2002, Lublin 2003), and personal discussion with him. Finally, a it is probably useful to make explicit the fact that although I am personally committed to Government Phonology, Kaye's interface theory is treated like all others in the book: its properties are reported, and connections are made with other theories. It is not necessarily the theory that I am defending, and I may or may not personally agree with it.

220 Chap 9: Kaye (1995), selective spell-out and freezing no look-back 261

1.2. Roadmap The roadmap is as follows: §262 introduces the foundations of Kaye's approach to the interface, which are functional and perception-oriented. The formal apparatus that he uses, domain structure, is then presented in §266. The heart of the chapter are §266 and §277: it is shown that Kaye follows Halle & Vergnaud (1987a) in implementing selective spell-out – but he adds a no look-back device that, unlike all others (Chomsky 1973, SCC-M, SCC-K, Mohanan-style Bracket Erasure), inhibits the modification of "old" strings. The contrast of this mechanism with previous no look-back devices is discussed in §287. Kaye's instrument turns out to be the ancestor of Chomsky's Phase Impenetrability Condition (PIC). Its empirical coverage is interesting: unlike Halle & Vergnaud (1987a), the rule-triggering pattern is fully covered; the rule-blocking pattern is also accounted for, however with an exception regarding cases where the affix, rather than the root is modified (§310). Also, Kaye's position in the debate regarding multiple minigrammars is discussed: his system makes the same choices as Halle & Vergnaud's (1987a) model, that is morpheme-specific phonology no, wordlevel-specific phonology yes, word-sequence-specific phonology no. In a second step, the (diachronic) consequences of domain structure are discussed (§328), Kaye's conception of the organisation of the lexicon is introduced (§§347f), and his view on lexical access exposed (§351).

262

2. Setting the scene: Kaye (1989)

263

2.1. Phonology exists because it enhances parsing The driving force behind Kaye's approach to the interface is functional. In his view, the evolution of the species has produced phonology because it makes communication faster, more effective and more reliable. That is, it facilitates the parsing of the continuous and indistinct phonetic signal into morphemes. 65 Morpheme recognition is the first thing that the linguistic system has to do when it is exposed to a signal – a critical step in the access

65

In his teaching, but not (yet) in print as far as I can see, Kaye also mentions a second raison d'être: phonological structure is the spine of the addressing system in the lexicon. This issue is discussed in §346 below.

Setting the scene: Kaye (1989) 221

to meaning. Kaye calls the indications that phonology provides for morpheme delineation parsing cues. Kaye (1989:11ff) compares natural and programming languages: the latter have a syntax, eventually a morphology and semantics, but no phonology. Why is that? Of course, an obvious answer is that unlike programming languages, natural language is actually pronounced, and the variation in the signal comes from the mechanic and phonetic processes (coarticulation, aerodynamics and the like) that occur when people speak. Kaye (1989:42ff) rejects this phonetic hypothesis. 66 Instead, he adopts a perception-oriented position: the linguistic system supplies the signal with predictable variation on purpose in order to facilitate the morpheme-recognition task. That is, it flies a flag in order to tell the listener where morphemes begin, and where they end. In order to show that the demarcative function of phonology is real, Kaye proposes two experiments. A thought experiment to begin with (Kaye 1989:49f): imagine a stretch of speech from which all effects of phonological processes, segmental (e.g. final devoicing) and suprasegmental (e.g. stress) alike, have been removed. Played back to a native, such a doctored signal would most probably be incomprehensible; or at least, could it be understood at all, it would drastically slow down the rate of transmission. Kaye (1989:50) therefore concludes that "human linguistic capacity is certainly an enormous advantage to our species, doubtless essential to our survival. Would a communicative system that functioned at, say, onefifth our speed offer the same adaptive qualities?" The second experiment (Kaye 1989:51f) is to show how important sandhi information is in perception: when used out of purpose, it can easily fool listeners and corrupt communication altogether. In English, the sequence /t-j/ may be pronounced [tÉʃ] provided its members straddle a word boundary, and the second item is a function word (e.g. pronoun, possessive). Hence the process may go into effect in the sentence I know what [tÉS] you want and I hit [tÉS] your brother, but not in I want *[tÉS] universal freedom, I hit *[tÉS] Yorick because here the [j-] does not belong to a function word. 66

See Ploch (2003) on this notion and its (ir)relevance for phonological theory. The central methodological claim of Kaye's approach to phonology (which Kaye refers to as the phonological epistemological principle) is that "the only source of phonological knowledge is phonological behaviour: Thus, phonetics […] plays no role in the postulation of phonological objects nor the interaction of such objects" (Kaye 2005:283).

222 Chap 9: Kaye (1995), selective spell-out and freezing no look-back Given this process, it is understandable why speakers fail to come up with any interpretation at all when presented with the input [ajnçwwatÉSuajwant]. The intended sentence is I know what ewe I want, only that the sandhi process is wrongly applied to the (rare) non-function word ewe [ju], which is a homophone of you. Parsing, then, is corrupted because the phonological system that is used for parsing visibly enforces a function-word interpretation of [ju], which can only be you. Even under strong external urge (the result of this parsing leads to the breakdown of communication), speakers are unable to consider the alternative parse that ewe would offer: they cannot "plug off" their phonological knowledge and allow for a non-functional interpretation. We will see below how this general line of reasoning shapes the interaction with morphology, where boundaries precisely may or may not fly a flag in the signal.

264

2.2. Perception-oriented views of phonology and the interface Kaye's approach falls into the large body of perception-oriented approaches and theories of the interface. These may be traced back at least to Trubetzkoy's Grenzsignale (§55), which had a rich offspring across disciplines and theories. Cutler (1996) provides a review of how speech recognition is done in the psychological literature, as well as of the role that is played by signalbased models of segmentation. In phonology, listener-oriented approaches are pursued more or less explicitly in the literature since Trubetzkoy. They are particularly prominent in the branch of OT that favours grounded (i.e. functional) constraints. Relevant literature includes Napoli & Nespor (1979:839), Booij (1983), Basbøll (1986), Bertinetto (1999), Loporcaro (1999), Hume & Johnson (eds.) (2001) and Boersma (1998, 2005).

265

2.3. Typical boundary detectors Kaye (1989:50ff) reviews phonological processes that are well suited to serve as flags for morphological parsing, and which are widely used for that purpose across languages. Two families of phenomena seem to be well equipped: suprasegmental indicators and harmony processes. The demarcative function of stress

Domain structure: how it is created, what it represents and how it works 223

for example is identified in the classical structuralist literature (e.g. Junković 1980, 1990). Stress is an accurate boundary detector: it either directly falls on edges and thereby identifies word boundaries, or it is edgeanchored. That is, its location is calculated with reference to the left or the right edge of words. Interestingly, languages do not appear to place stress with reference to the middle of the word, although this would be just as easily computable as edge-anchoring. Also, Kaye points out that harmony processes have a domain of application and thereby naturally indicate edges.

266

3. Domain structure: how it is created, what it represents and how it works

267

3.1. The concat- and the φ-function

268

3.1.1. General properties In Kaye's (1995:302ff) system, the concat function concatenates two morphemes and returns them as a linear sequence. Hence given morphemes A and B, the action of concat is to produce the output [AB]. This is quite comparable with what modern Merge does, only that no particular label is attached to the output (whose hierarchical status is indicated by brackets). The φ-function is the set of phonological processes that are synchronically active in a given language and computed online whenever phonology is done on some string. Kaye's (1995:302) shorthand for the φfunction is the order "do phonology!". Thus far, talking about the φ-function is just using a specific vocabulary in order to talk about the phonological grammar in general: there is no theory-specific aspect, and the φ-function could be whatever is the favourite computation of phonologists: a set of ordered rules, a set or ranked constraints or any other set of instructions that modify a string etc.

269

3.1.2. Computation in Government Phonology Government Phonology is often referred to as a representation-oriented theory of phonology, and there is certainly good reason for this characterisation. A correlate of the representational focus is the fact that computation in Government Phonology was not explicitly regulated for quite some time:

224 Chap 9: Kaye (1995), selective spell-out and freezing no look-back the only indication that could be found in print was Kaye's (1992a:141, 1995:291) statement according to which processes "apply whenever the conditions that trigger them are satisfied." Kaye avoids talking about rules since he strongly opposes SPE-type derivationalism where rules are extrinsically ordered (Kaye 1990b, 2008). Alongside with Optimality Theory (Prince & Smolensky 1993) and Declarative Phonology (Scobbie 1991, Scobbie et al. 1996), Government Phonology (Kaye et al. 1990) is thus one of the three theories that emerged in the early 90s (or mid-late 80s) on the grounds of a strong antiderivational mantra. The computation in all three cases is based on constraints, which however do not have the same status: while they are ranked and violable in OT, they are absolute (i.e. non-violable) in Declarative Phonology. While it was always clear that Government Phonology rejects derivationalism (i.e. extrinsically ordered instructions), how exactly computation works and what a computational instruction looks like was not made explicit until the mid-90s when the constraint-based character of computation in Government Phonology became obvious: computation is done by socalled Licensing Constraints (Charette & Göksel 1994, 1996, Kaye 2001).67 Following Kaye's (1992a:141) general philosophy mentioned, constraints in Government Phonology apply whenever a form may be modified by them, but with no extrinsic ranking or ordering, and without being able to be violated: constraints are (simultaneously and) iteratively applied to the string that is submitted to interpretation, and computation ends when no further modification can be made.68 Using serial vocabulary, this system is thus able to handle a feeding relationship (the input for the application of a constraint is created by the modification of the string by another constraint), but no other type of rule interaction (bleeding, counter-feeding, counter-bleeding). A difference must 67

68

Scheer (2010a, forth) discusses the approaches to computation in phonological theory that are available on the market in general, and how computation works in Government Phonology in particular (Scheer 2010c). This is an obvious parallel with Harmonic Serialism, which was introduced by Prince & Smolensky (1993) as a serial alternative to strictly parallel assessment of candidates (the output of EVAL loops back into GEN until no harmonic improvement can be made anymore). John Goldsmith's Harmonic Phonology (Goldsmith 1992, 1993, Larson 1992) and Harmonic Grammar (Legendre et al. 1990, Smolensky & Legendre 2006) are further implementations of Harmonic Serialism that are more closely inspired by connectionism. Finally, the most recent version of Harmonic Serialism is McCarthy's (2007) OT-CC.

Domain structure: how it is created, what it represents and how it works 225

therefore be made between serial computation (GP computation is serial in the sense that constraints may apply to the same string several times) and serialism (there is no extrinsic or logical ordering of instructions, i.e. classical extrinsic rule ordering). Also, there is no ranking or prominence relationship among constraints: all instructions are equally important. I am not aware of discussion in the GP literature of what happens when different constraints conflict, or whether such a situation is provided for at all (see discussion of this issue in Scheer 2010c). 270

3.1.3. Important properties of the φ-function for Kaye's interface theory While indications how computation really works in Government Phonology are still sparse in print (Gussmann 2007 is probably the most complete source), the two properties of the φ-function that are important for our purpose are clear: all processes are equal-righted, and all of them apply when phonological interpretation is performed. Equal-rightedness means that there is no extrinsic ordering of instructions (no sequential/chronological application, no hierarchy/constraint ranking): all processes apply simultaneously. The fact that none of them can be left unapplied excludes selective rule application (see §149): there is no way for just a subset of the phonological processes to apply at a given time and to a given string. Phonology is not divisible, and "do phonology!" means "do all the phonology!". Kaye thus restores the view of SPE on phonological processing: there is only one computational system (§107), and multiple mini-grammars of whatever kind (§234) are ruled out. Whether Kaye's system really stands up to this strict commitment is discussed in §338 below. Technically speaking, the φ-function "has one argument, a phonological string, and returns the application of the phonology to this argument, also a phonological string. The expression φ(X) means, 'apply phonology to the string X'. φ(X) returns the phonological string which results from the application of phonology to its argument" (Kaye 1995:302).

271

3.2. Domain structure is created by interleaved concat and φ Interpretational units are called domains in Kaye's terminology. Domains are defined as those chunks of the linear string that are phonologically rele-

226 Chap 9: Kaye (1995), selective spell-out and freezing no look-back vant, i.e. to which phonology applies. They are created by the interaction of the concat- and the φ-function. These are not interleaved when phonology is done on a morphologically simplex input, φ(X), or when it applies to a morphologically complex object, φ(concat(X,Y)). However, concat and phonological interpretation may also be intertwined, a situation that creates more complex structures. Given two morphemes X and Y, either may be subjected to φ before concatenation takes place. This situation corresponds to the expressions φ(concat(φ(X),Y)) and φ(concat(X,φ(Y))). In the former case, phonology operates over morpheme X, the result is concatenated with morpheme Y, and phonology again applies to the output. That latter configuration is the symmetric counterpart. Finally, the system affords a third interleaved configuration, i.e. when both X and Y are interpreted before concatenation; this creates the compound structure φ(concat(φ(X),φ(Y))). Table (102) below recapitulates the four logically possible configurations when two morphemes X and Y are interpreted in Kaye's system. (102) possible domain structure for two morphemes X and Y formal brackets number of domains a. φ(concat(X,Y)) [X Y] one: [X Y] b. φ(concat(φ(X),Y)) [[X] Y] two: [X] and [X Y] c. φ(concat(X,φ(Y))) [X [Y]] two: [Y] and [X Y] d. φ(concat(φ(X),φ(Y))) [[X] [Y]] three: [X], [Y] and [X Y]

name non-analytic analytic analytic analytic

In the simplest configuration under (102a) phonology is done only once, i.e. it applies to the result of the concatenation of X and Y. Under (102b) and (102c), it applies twice: to one of the two morphemes in isolation, and to the result of their concatenation. Finally, the compound structure under (102d) affords three passes of the φ-function, which applies to both of the morphemes in isolation, and to the result of their concatenation. In this system, the number of domains and the number of applications of the φ-function is thus always identical. Finally, it is to be noted that Kaye (1995:305f) is inclined to dismiss (102c): while [X [Y]] is a logically possible structure, it does not appear to be attested. Domain structure is further discussed by Gussmann & Kaye (1993), Cyran & Gussmann (1998, 1999) and Gussmann (1998, 2002:45ff).

Domain structure: how it is created, what it represents and how it works 227 272

3.3. Analytic vs. non-analytic Kaye talks about analytic and non-analytic domains (see the last column of table (102)). The former (e.g. [[X] Y]) contain another domain and may therefore be further analysed, while the latter (e.g. [X Y]) have no internal structure. Following the same logic, Kaye refers to a string of morphemes that has no internal structure as non-analytic morphology, while strings that subdivide into further domains are called analytic morphology. Whether two morphemes end up as an analytic or a non-analytic structure depends on their lexical properties (this is how affix classes are distinguished, see §310). That is, some affixes are specified for not creating a single domain of interpretation with their host, such as Y in [[X] Y]; they are called analytic (or interpretation-triggering in the terminology that was used in the previous chapter). On the other hand, non-analytic affixes are interpreted together with their host (Y in [X Y], they are (they are interpretation-neutral). Looked at from the vantage point of spell-out, the concatenation of analytic affixes triggers the spell-out of their sister (rather than of their own node, more on this contrast in §282 below), while non-analytic affixes have no influence on the interpretation of their sister.

273

3.4. Domain structure is the result of selective spell-out Domain structure is the result of selective spell-out in the sense of Halle & Vergnaud (1987a) (§225): given the morpho-syntactic tree, affixes may or may not trigger spell-out. If they do as Y under (103a) their sister is spelled out (i.e. enclosed by brackets in the linear notation).69 In case they do not as Y under (103b), their sister is not subjected to interpretation (i.e. will not be a bracket-delineated domain in phonology).

69

Halle & Vergnaud's (1987a) take is slightly different: instead of the sister, they spell out the material that is dominated by the projection of the interpretationtriggering affix, i.e. β under (103a). This contrast is further discussed in §282 below.

228 Chap 9: Kaye (1995), selective spell-out and freezing no look-back (103) domain structure is impoverished morpho-syntactic structure a. interpretation-triggering Y b. interpretation-neutral Y β Y

β α

phon

x

Y

X

α x

X

spell-out

spell-out

[X] Y

XY

Put differently, the morpho-syntactic division between X and Y is made visible to the phonology under (103a) but not under (103b): spell-out is selective. 274

3.5. Interpretation prior to concatenation: domain structure generates more than spell-out can create An interesting formal property of Kaye's domain structure that will turn out to be empirically relevant in §§314f is the possibility for pieces to be interpreted before they are concatenated. Consider the compound structure such as [[X] [Y]] under (104) where both X and Y are interpreted in isolation before any concatenation takes place. (104) domains that are not created by spell-out a. Y is interpreted in isolation b. Y is interpreted in isolation β Y

phon α

phon

x

X

β Y

α x

spell-out [[X] [Y]]

X spell-out

X [Y]

Domain structure: how it is created, what it represents and how it works 229

Recall from §271 that [[X] [Y] and [X [Y]] (where X is a root and Y an affix) are possible domain structures when two morphemes are concatenated. Neither, however, can be created by the (affix-triggered) spell-out of nodes, as introduced in the previous section. Under (104a) the spell-out of α makes X a domain, and the spell-out of β defines the entire string [XY] as an interpretational unit. There is no way, however, to achieve the isolated interpretation of Y by spell-out. Likewise, no node under (104b) can be spelled out so that Y becomes an interpretational unit by itself. This means that Kaye's system allows for the spell-out of nodes (α, β) as much as of terminal elements (Y). That is, domain structure is shaped by spell-out, but does not reduce to the action of this device: some domains are created by independent means. In absence of additional options for the origin of domain structure, these non-spell-out-created domains must be a lexical property of the pieces at hand. This is consistent with affix-triggered interpretation: it is the lexical properties of affixes that define interpretational units (i.e. domain structure), either by projecting their spell-out properties into the tree, or directly by specifying their own body as a domain. This may be illustrated by an interpretational structure such as [[prefix] [root]], which cannot be generated by spell-out, but which is needed (as we will see in §314) in order to derive the contrast between im-possible (where the prefix-final nasal assimilates) and un-predictable (where it does not). The status of independently spelled-out terminals and the fact that this cannot be expressed by regular spell-out is not discussed by Kaye (1992a, 1995). Kaye only looks at domain structure from the phonological (linear) point of view: domain structure is whatever phonological phenomena require it to be. In syntax, the interpretation of pieces prior to their being merged has been proposed as a consequence of counter-cyclic merger (late adjunction). This formal parallel and its consequences for phonologically motivated independent spell-out of terminals is discussed in §316.

275

3.6. Domain structure is interactionist, brackets are only shorthand Kaye always uses fully-fledged bracketed representations such as [[X] Y] – exactly what is known from SPE and Halle & Vergnaud (1987a). Unlike in these approaches, however, Kaye's structure is explicitly interactionist (§271).

230 Chap 9: Kaye (1995), selective spell-out and freezing no look-back Also, brackets do not exist: they are not parsed or referred to by any phonological mechanism. The bracketed notation merely indicates the derivational history of the string at hand as it was created by successive waves of spell-out. It is the way phonologists look at morpho-syntactic structure: they see only phonologically relevant information, and they feel at home in a linear, rather than in an arboreal representation. Domain boundaries thus represent procedural, not representational information. Recall from §§161,170 that brackets in non-interactionist systems such as SPE and Halle & Vergnaud represent untranslated morpho-syntactic information in phonology and are therefore unwarranted from the modular point of view. Interactionism is the only way to reconcile inside-out interpretation with modularity. 276

3.7. Kaye's procedural-only approach and morpho-phonology Kaye seems to reject any representational communication between morphosyntax and phonology. This may be concluded from the section "morphological effects in the phonology" in Kaye (1995:301f): the only means for morpho-syntax to talk to phonology that Kaye mentions is procedural. That is, he does not provide for boundaries, LP-style brackets (§§170f), prosodic constituency or any other kind of representational carriers of morphosyntactic information. Let us have a brief look at the consequences of this procedural-only approach for the question of morpho-phonology. Recall from §126 that the following issue was debated in post-SPE times: to which extent are alternations that make reference to morpho-syntactic information the result of online computation and suppose a single underlying form? Theories of the 70s and 80s vary as to how much of these alternations are unloaded into the lexicon; all approaches, however, impoverish the rule component of SPE. Lexical Phonology operates only a moderate lexicalisation of alternating patterns (some illustration was provided in §129). The most radical response comes from Natural (Generative) Phonology, which – in vain – attempts to eliminate morpho-syntactic information from phonology altogether (§127f). Kaye's system is also located on the far end of the spectrum. Like Natural (Generative) Phonology, Kaye does not allow for boundaries or any other kind of representational means to introduce morpho-syntactic information. The only way to make phonology sensitive to extra-phonological information, then, is to encode this information into domain structure. Any

Selective spell-out: Kaye's vs. Halle & Vergnaud's implementation 231

alternation that refers to morphological information but cannot be framed in this way must be recorded in the lexicon. Typical SPE-style alternations such as Velar Softening (electri[k] electri[s]-ity, relevant literature is mentioned in note 28) and Trisyllabic Shortening (§164) are also dismissed by Kaye's restrictive view of what phonology is able to do. The vocalic alternations involved in the latter pattern, Kaye (1995:312f) argues, cannot be afforded by the φ-function, which otherwise would be able to compute all kinds of alternations that do not occur in natural language. Today indeed, most phonologists will probably not want to consider Trisyllabic Shortening a productive phonological process anymore (see Hayes 1995a and note 42).

277

4. Selective spell-out: Kaye's vs. Halle & Vergnaud's implementation

278

4.1. Introduction The following pages introduce Kaye's implementation of Halle & Vergnaud's (1987a) selective spell-out. The central instrument that sets Kaye's approach apart from Halle & Vergnaud's is a no look-back device that disallows the modification of phonological properties which were acquired on a previous cycle (domain). It is really the formulation of the no look-back device that matters: Kaye's is unlike all previous versions. The different brands of no look-back devices that have been proposed since Chomsky (1973) are discussed at length in §287. For the time being, it is enough to recall that Halle & Vergnaud (1987a) did use a look-back device (§237), Kiparsky's SCC-K, but that this instrument is entirely irrelevant for their analysis of affix class-based phenomena (§248). The pages below first show how Kaye's no look-back device organises underapplication in affix class-based phenomena, and then moves on to identify a number of other important differences with Halle & Vergnaud's system.

279

4.2. Underapplication is achieved by no look-back The difference between class 1 and class 2 affixes was encoded by different boundaries in SPE (§§92,106), by strata in Lexical Phonology (§147) and by selective spell-out in Halle & Vergnaud's (1987a) model (§228).

232 Chap 9: Kaye (1995), selective spell-out and freezing no look-back Kaye (1995) follows the latter option. As was shown in §273, he expresses the contrast between class 1 and class 2 affixes by what he calls non-analytic vs. analytic domain structure. In English, then, class 1 affixes are non-analytic (i.e. do not provoke the spell-out of their sister), while class 2 affixes is analytic (i.e. triggers the spell-out of their sister). In this perspective, the relevant domain structure for the familiar example from English stress (párent, paréntal vs. párenthood, see §147) is [parent], [parent al1] and [[parent] hood2]. Penultimate stress is then assigned in each domain, from inside out. On the innermost pass, this produces [párent], [parént-al] and [[párent] hood].70 The derivation ends for the former two words, which do not accommodate any other domain. When phonology interprets the outer domain of [párent-hood], so-called robustness (Kaye's term) blocks the reassessment of stress (to *parént-hood): strings that have already experienced interpretation cannot be further modified. The precise formulation of Kaye's no look-back device is discussed and compared to other options in §287 below. The example at hand requires underapplication of the stress rule to the class 2 string [párent hood]: the stress rule must not reapply to the outer cycle. Unlike in all other approaches to affix class-based phenomena, underapplication is thus afforded by a no look-back device in Kaye's analysis.

70

In SPE, vowels escape reduction to schwa when they bore main stress on a previous cycle (see §97). Kaye follows this line of attack (see §334 below). Since -hood2 does not experience vowel reduction, he concludes that it must have been stressed, which in turn requires it to be a domain (only domains are stressed). Therefore the actual domain structure that Kaye (1995:308, 313) proposes is [[parent][hood]], where the competition between the two inner domains for main stress is decided in favour of the leftmost domain by virtue of the fact that English compounds are always stressed on the first element (bláckbòard, see §334). In addition of triggering the spell-out of the root, -hood2 has thus the lexical property of requiring its own spell-out prior to being concatenated. That is, we are facing a case of the kind that was discussed in §274: a terminal may be subject to independent spell-out. This additional complication does not interfere with the purpose of the demonstration, for which the simplified domain structure [[parent] hood] will do.

Selective spell-out: Kaye's vs. Halle & Vergnaud's implementation 233 280

4.3. No automatic spell-out of roots, but systematic spell-out at the wordlevel It was already mentioned that Kaye (1995) follows Halle & Vergnaud's (1987a) idea of selective spell-out, but ends up with quite different results. Let us now look at the factors that produce the contrast when selective spell-out is implemented. Three properties are relevant: the automatic spellout of roots (Halle & Vergnaud: yes, Kaye: no), the systematic spell-out at the word level (Halle & Vergnaud: no, Kaye: yes) and the management of derived environment effects (Halle & Vergnaud: integrated into the management of affix class-based phenomena, Kaye: a separate issue). The latter is only discussed in §284 below. Let us start by comparing the interpretational structure of the same itmes that is proposed by Halle & Vergnaud and Kaye. Note that for the sake of comparability, specific word-level phonologies that eventually apply to the result of the cyclic derivation (non-cyclic rules in Halle & Vergnaud's system) are disregarded for the time being. That is, brackets under (105) below define domains of application of cyclic rules on the one hand (Halle & Vergnaud), of the φ-function on the other (Kaye). (105) interpretational units: Halle & Vergnaud vs. Kaye Halle & Vergnaud Kaye a. párent [parent] [parent] b. parént-al [[parent] al] [parent al] c. párent-hood [parent] hood [[parent] hood]

In Halle & Vergnaud's system, all roots are spelled out in isolation: they form a cycle of their own (see §230). On Kaye's count, this is not the case: roots have no intrinsic virtue of being spelled out. The spell-out of all strings, roots included, is decided by the lexical properties of pieces that are concatenated: analytic affixes (cyclic in Halle & Vergnaud's terminology) provoke the spell-out of their sister, while non-analytic affixes (non-cyclic) do not. This is shown under (106) below.

234 Chap 9: Kaye (1995), selective spell-out and freezing no look-back (106) Kaye: analytic affixes provoke the spell-out of their sister a. Y is analytic: α is spelled out b. Y is non-analytic: α is not spelled out β Yanalytic

β α

x

phon X

spell-out [[X] Y]

Ynon-analytic

α x

X spell-out

[X Y]

This procedure produces /[X] Y/ for (106a) and /X Y/ for (106b). The fact that the entire word ends up being delineated by brackets (i.e. as an interpretational unit) is due to the aforementioned fact that the word as a whole is always spelled out no matter what. Hence [[X] Y] and [X Y] under (106), respectively, and [[parent] hood] vs. [parent al] for the example mentioned. This also explains the fact that párent in isolation is an interpretational unit: it is a domain by virtue of being a word with Kaye, while its being a cycle with Halle & Vergnaud is due to the fact that it is a root. The treatment of the root and the word as interpretational units that automatically trigger spell-out is thus opposite: while Halle & Vergnaud grant this status to the former but not to the latter, Kaye implements the reverse choice.

281

4.4. Who is interpretation-triggering – class 1 or class 2 affixes? Table (105) also shows that Kaye (1995) makes the opposite choice of Halle & Vergnaud (1987a) regarding the type of affix that triggers interpretation (the following table only mentions brackets that are created by the concatenation of an affix).

Selective spell-out: Kaye's vs. Halle & Vergnaud's implementation 235 (107) interpretation-triggering affixes a. Halle & Vergnaud (1987a) /[parent al1]/ vs. /parent hood2/ class 1 affixes trigger interpretation – they are cyclic class 2 affixes do not trigger interpretation – they are non-cyclic b. Kaye (1992a, 1995) /parent al1/ vs. /[parent] hood2/ class 2 affixes trigger interpretation – they are analytic (cyclic) class 1 affixes do not trigger interpretation – they are non-analytic (noncyclic)

As far as I can see, the fact that Kaye (1995) proposes the exact reverse of Halle & Vergnaud's implementation of English affix classes has left no traces in the literature. Kaye (1992a) mentions that there may be differences with respect to Halle (1986), but does not make them explicit. (108) "I accept the notion of the cycle along with the PSC [Principle of Strict Cyclicity, cf. below] as crucial parts of a theory of UG. The cyclic domains which I posit do not necessarily agree with earlier assumptions, however. I do agree with Halle (1986) regarding the cyclic or non-cyclic status of affixes. I shall assume that whether or not an affix is cyclic is not a property of the morphological rule by which it is assigned, but is rather an idiosyncratic and variable property of the affix. Halle (1986:6)" Kaye (1992a:142)

In 1992, Kaye thus still uses the SPE-terminology of Halle & Vergnaud which refers to cycles and cyclic vs. non-cyclic affixes. It is only later that cycles will appear as domains, and cyclic/non-cyclic as analytic/nonanalytic. Maybe this move in terminology is a way to mark the different – actually opposite – use of selective spell-out. 282

4.5. Interpretation-triggering affixes: spell-out of the sister vs. their own node The contrast between /[parent al1]/ vs. /parent hood2/ (Halle & Vergnaud) and /parent al1/ vs. /[parent] hood2/ (Kaye) that is evidenced under (107) also shows another fundamental difference between the two competing implementations of selective spell-out (which was already mentioned in §226 and §§272f): the string that is spelled out when an interpretation-

236 Chap 9: Kaye (1995), selective spell-out and freezing no look-back triggering affix (cyclic/analytic) is added is not the same. This is shown under (109) below. (109) interpretation-triggering affixes: what exactly is spelled out a. Halle & Vergnaud (1987a): b. Kaye (1995): cyclic affixes trigger the spellanalytic affixes trigger the spellout of their own constituent β out of their sister α β Ycyclic

phon α

x

β Yanalytic

X spell-out

[X Y]

α

phon

x

X spell-out

[X] Y

On the other hand, the concatenation of affixes that do not trigger spell-out has the same effect in both systems: no interpretational unit is created.

283

4.6. Morpheme- and chunk-specific phonologies Let us now look at Kaye's position regarding the issue of multiple computational systems in phonology. Kaye (1995:302ff) is explicit on the fact that all instructions for phonological processing are grouped together in the φfunction, and that all instructions of this function apply when the function applies (§267). Hence there is no selective rule application, and there is only one computational system that contains phonological instructions. This clearly rules out any version of multiple mini-grammars, be they morpheme- or chunk-specific. Recall from §234 that Halle & Vergnaud (1987a) reject morpheme-specific phonologies (level 1 vs. level 2 rules in Lexical Phonology) and a chunk-specific phonology for the sentence level (Praguian segregation, i.e. lexical vs. postlexical phonology in Lexical Phonology). Also recall from §238 that this notwithstanding, Halle & Vergnaud (1987a) make a difference between pre-word and post-word phonology by interrupting the derivation at the word level, where a phonology specific to that chunk size applies.

Selective spell-out: Kaye's vs. Halle & Vergnaud's implementation 237

Kaye's (1995) take is largely, maybe exactly along these lines. He rejects for sure any kind of morpheme-specific phonology, and also Praguian segregation, i.e. the application of a specific phonology to sequences of words (as opposed to sequences of morphemes). However, just like Halle & Vergnaud, Kaye does implement a formal break in the derivation at the word level: recall from §280 that the word is always spelled out as such no matter what the affixal situation and the derivational history. That is, the word is granted the status of an interpretational unit by default. The question is whether the recognition of the word as a chunk that has a special phonological status is also accompanied by the application of a specific word-level phonology. We have seen that in theory the answer is no – but in practice it may well turn out to be yes. This issue is discussed in §338 below.

284

4.7. Derived environments are a separate issue Kiparsky (1982a,b) has introduced derived environment effects into Chomsky's (1973) no look-back device (§§189f), which until then was only designed to prevent the application of rules to "old" material. It was shown in §182 that derived environment effects and no look-back effects in affix class-based phenomena are logically and empirically independent. Also, Kiparsky (1993) himself has declared the bankruptcy of his SCC-K (§197). The logical and empirical independence notwithstanding, Halle & Vergnaud (1987a) maintain the incorporation of derived environment effects into the no look-back device: they implement the SCC-K (§237). Unlike in Kaye's (1995) system, though, nothing in Halle & Vergnaud's analysis of affix class-based phenomena is due to the SCC-K, which does not intervene in the organisation of underapplication (§228). We have seen in §279 that the no look-back device is critical in Kaye's system: it is responsible for underapplication. It was also mentioned that Kaye uses a (modification-inhibiting) version of the no look-back device that is different from previous blends (but not unprecedented in the literature). Its precise nature is discussed in §299 below. In any event, derived environment effects have got nothing to do with it: Kaye (1995) considers that they are a separate issue (but does not offer any specific analysis). His no look-back device has no ambition to account for them. This is also obvious from the explicit reference that Kaye (1992a:142, 1995:307, 330 note 23) makes to Chomsky (1973), Kean (1974) and Mascaró (1976) (but not to Kiparsky 1982a,b).

238 Chap 9: Kaye (1995), selective spell-out and freezing no look-back 285

4.8. Cyclic interpretation of words? Finally, let us consider the question whether only morphemes, or also words experience cyclic (inside-out) interpretation. While SPE was explicit on the fact that the Transformational cycle applies to morphemes and words alike ("the principle of the transformational cycle […] appl[ies] to all surface structure whether internal or external to the word", Chomsky & Halle 1968:27, see §105), a headstone of Lexical Phonology is the idea that only morphemes enjoy cyclic derivation: lexical rules are cyclic, but postlexical phonology is not (§158). Regarding this issue, Kaye (1995) restores SPE as much as Halle & Vergnaud (1987a) before him (§219). He quotes French nasalisation as a case where cyclic derivation of words is required (Kaye 1992a:142ff, 1995:306ff, the evidence is discussed in §301). This issue also has an impact outside of phonology: it is relevant for the debate regarding lexicalism and raises what I call the word-spell-out-mystery in §786 below.

286

4.9. Summary: two ways of doing selective spell-out We have seen that Kaye's (1995) and Halle & Vergnaud's (1987a) ways of implementing selective spell-out are quite different. The number of factors that were compared makes it worthwhile to propose an overview. Table (110) below first gathers the properties that both systems share. (110) properties shared by Halle & Vergnaud (1987a) and Kaye (1995) Halle & Vergnaud a. multiple computational systems 1. morpheme-specific no 2. sentence-specific (Praguian segreg.) no 3. word-level-specific yes b. cyclic (inside-out) interpretation of words yes c. selective spell-out yes d. interpretation triggered by a lexical propyes erty of affixes upon concatenation

Kaye no no yes? yes yes yes

Table (111) below recalls the factors that make Kaye's and Halle & Vergnaud's system different.

No look-back devices: implementations since 1973 239 (111) differences between Halle & Vergnaud (1987a) and Kaye (1995) Halle & Kaye Vergnaud a. the root is an interpretational unit yes no b. the word is an interpretational unit no yes c. morpho-syntactic terminals may be interno yes pretational units d. interpretation-triggering affixes trigger their own their sister the spell-out of node e. English affix classes: type that triggers class 1 class 2 interpretation f. underapplication is achieved by cycles cycles and no look-back g. no look-back device 1. type of device used SCC-K modificationinhibiting 2. derived environment effects built in yes no 3. no look-back is relevant for underapno yes plication

The following section shows in which way exactly modificationinhibiting no look-back is different from other incarnations that have been entertained since no look-back devices were introduced by Chomsky (1973).

287

5. No look-back devices: implementations since 1973

288

5.1. Like lexicalism, no look-back is born in the early 70s as an overgeneration-killer Cyclic (inside-out) derivation was introduced by Chomsky et al. (1956:75) (§80) and became a central piece of the generative interface architecture in SPE (§100). However, SPE did not feature any no look-back device yet. The idea that linguistic computation cannot look back to previously interpreted strings was only introduced by Chomsky (1973) under the heading of the Strict Cycle Condition. In the context of the early 70s, this was a contribution to the attempt at marshalling the generative power of the unrestricted derivational (transformational) mechanism. Ordered transformations in syntax overgenerated as much as ordered rules in phonology (see §126). Overgeneration was a plague, and the antidote into which generative theory engaged was a restriction on both underlying forms (see §124), i.e.

240 Chap 9: Kaye (1995), selective spell-out and freezing no look-back lexicalism (on which more in §§537,569) and computation, i.e. no lookback. In Remarks on Nominalization, Chomsky (1970) had argued that words such as reduction, transmission and recital are not the result of online computation because their derivation is unproductive and semantically opaque; rather, they are stored in the lexicon as a whole. Therefore their pronunciation does not require any concatenation. On the phonological side, the same point was made regarding for example Velar Softening: some considered electri[s]ity a single lexical entry because the alleged palatalisation of the /k/ in electri[k] into [s] before /-i/ is riddled with exceptions and not productive (see §126).

289

5.2. Chomsky's (1973) Strict Cycle Condition: you need to use new material Chomsky's (1973) Strict Cycle Condition is designed to prevent rules from applying if they do not use material that was introduced into the derivation on the current cycle. The original formulation appears under (112) below. (112) Strict Cycle Condition "No rule can apply to a domain dominated by a cyclic node A in such a way as to affect solely a proper subdomain of A dominated by a node B which is also a cyclic node." Chomsky (1973:243)

The effect is that rules are blocked whose structural description is met by a string that is made exclusively of material that belongs to a previous cycle. That is, given [[AB]i C]j, a rule that is triggered by AB can apply at cycle i, but not at cycle j. In other words, rules must use material that was introduced on the latest cycle. In Chomsky's (1973:243) terms, "rules cannot in effect return to earlier stages of the cycle after the derivation has moved to larger, more inclusive domains."71 Kean (1974:179) also discusses a more complicated situation where the Strict Cycle Condition bites: a rule whose context is not satisfied in some cycle, but which is fed by the application of another rule that only 71

Following or parallel to Chomsky (1973), a number of mechanisms that restrict computation have been proposed in syntax. These include the Freezing Principle (Culicover & Wexler 1973) and the Binary Principle (Hamburger & Wexler 1975, Wexler & Culicover 1980:119ff).

No look-back devices: implementations since 1973 241

applies at a later cycle, cannot go into effect if its trigger belongs to the earlier cycle. Suppose a rule such as A → B / __Y, which cannot apply in the inner cycle of [[A X] Z]. Suppose another rule X → Y / __Z, which will go into effect once Z was added on the outer cycle. This feeds the former rule, which should be able to apply to [A Y Z], transforming A into B. The Strict Cycle Condition blocks its application, though, since none of its ingredients, A and Y, was introduced on the outer cycle.

290

5.3. Application to phonology: Kean (1974) and Mascaró (1976) During the 70s, strict cyclicity was applied to phonology by Kean (1974) and Mascaró (1976).72 Mascaró is at the origin of the term phonological cycle, which replaces the SPE-based notion of the transformational cycle in phonological quarters. Mascaró's SCC-M is repeated below (from §161). (113) Strict Cycle Condition (SCC-M) "Proper application of cyclic rules For a cyclic rule to apply properly in any given cycle j, it must make specific use of information proper to (i.e. introduced by virtue of) cycle j." Mascaró (1976:7)

Mascaró (1976:7f) explains that "this condition ensures that no 'improper' cyclic application, that is, multiple application of a rule, opposite rule ordering, etc. on the same cycle results." He also calls his version "a more general formulation of the basic idea of the Strict Cycle Condition [, which] was proposed in Chomsky (1973) for syntax" (Mascaró 1976:8). 72

The earliest (published) sources that I could identify are two articles by Morris Halle from 1973 and Wilkinson (1974) where the authors refer to an unpublished manuscript by Mary-Louise Kean (dated 1971: Kean 1971, maybe the forerunner of Kean 1974). In the "addendum" at the end of Halle (1973a), Halle advertises the fact that the revisions of the stress system of SPE that he has undertaken in this article also avoid violations of strict cyclicity that were incurred by the original SPE rules. Halle says that the unpublished manuscript by MaryLouise Kean dated 1971 drew his attention to this issue. In Halle (1973b:319), the author also refers to Kean's manuscript as the source of "a general principle of rule application which in recent discussions has been termed STRICT CYCLICITY" (emphasis in original). In order to block the application of a rule in Telugu, Wilkinson (1974:261f) uses Chomsky's strict cyclicity, quoting the manuscript of Chomsky (1973), which he dates 1971.

242 Chap 9: Kaye (1995), selective spell-out and freezing no look-back In any case, it appears to be more precise: "make crucial use" of new material is less clear than "make specific use" of new material.

291

5.4. Halle/Kiparsky's SCC-K: scrambling with derived environments It was already mentioned in §190 that the SCC-K, a headstone of Lexical Phonology, was in fact introduced by Morris Halle (1978), rather than by Paul Kiparsky (1982a,b) as is commonly believed. Unlike to Kiparsky (1982a,b) who talks about his SCC in terms of a "simplified" version of Mascaró's (1976) SCC-M ("with some simplification, his [Mascaró's] proposal was", Kiparsky 1982b:153), Halle (1978:131) is explicit on the fact that "the version of the constraint on cyclic rule application that I propose below is a combination of certain suggestions made by Kiparsky (1973[a]:60), with others due to Mascaró (1976:9)." Paul Kiparsky was concerned with the (restriction of) rules that appear to look back in a derivation since Kiparsky (1968-73) and Kiparsky (1973a). The development that led to his version of the Strict Cycle Condition was reported in §157. Halle now modifies the formulation of Chomsky's Strict Cycle Condition in one crucial aspect in order for it to also cover derived environments: instead of imposing only new material to be used by a rule, he requires that new and old material be accessed. (114) "A cyclic rule R applies properly on cycle j only if either a) or b) is satisfied: a) R makes specific use of information, part of which is available on a prior pass through the cyclic rules, and part of which becomes first available on cycle j. […] b) R makes specific use of information assigned on cycle j by a rule applying before R." Halle (1978:131)

Recall from §189 that the SCC-M has got nothing to do with derived environment effects: it is unable to prevent rules from applying to monomorphemic strings. For example, it does not prevent Trisyllabic Shortening from applying to underived items such as nightingale and ivory: on the innermost cycle, i.e. [nightingale], the rule would use material of this cycle and hence apply "properly". The move of (114a) now introduces the proviso that in addition of new material, "properly" applying rules must also make use of old material.

No look-back devices: implementations since 1973 243

This indeed blocks the application of any rule to the innermost cycle, which by definition cannot contain any old material. This is how Chomsky's Strict Cycle Condition is made to cover derived environment effects. The subsequent literature, as well as common phonological belief of generations that have not lived through this period, constantly confuse Chomsky's original "use new material!" requirement with Kiparsky's concern for derived environments (but see note 49). It is therefore important to make their independence explicit. Also recall from §182 that derived environment effects and no look-back effects in affix class-based phenomena are logically and empirically independent. Kiparsky's (1982a,b) brand of the SCC does exactly the same labour as Halle's, but affords a formulation that is quite different: while the fact of preventing rules from applying to derived environments is a mere consequence of Halle's version, derived environments now are explicitly mentioned as such. 292

5.5. Mohanan's bracket erasure: indirect bearing on rules Mohanan's (1986) bracket erasure is an entirely different no look-back mechanism. Recall from §168 that it was originally designed for the ruletriggering pattern of affix class-based phenomena. The fact that it could also be – but quite surprisingly does not appear to have been – used for derived environment effects was discussed in §201. In any event, it has nothing to do with Chomsky's original idea to require that material from the latest cycle must be used by rules. With Mohanan's bracket erasure, rules may or may not use material from the latest cycle, and they may or may not use material from previous cycles. Bracket erasure does not impose any restriction on the kind of material used. Its effect is different, and bears only indirectly on the application of rules. Bracket erasure makes all morphological divisions that were not created by concatenation at the latest stratum invisible for phonological rules. Hence a string like [[[A] B]1 C]2 where subscripts indicate strata, and brackets morphological divisions, will appear as [[AB] C] on stratum 2 since the internal structure of [[A] B] was erased at the end of stratum 1. Whether a rule appeals to A, B, C or any combination thereof is entirely irrelevant: bracket erasure makes no statement at all. On the other hand, bracket erasure makes only sense when it works in tandem with bracket-sensitive rules – otherwise it has no effect at all.

244 Chap 9: Kaye (1995), selective spell-out and freezing no look-back That is, only rules that bear brackets in their structural description will be affected by the absence of "old" brackets that were erased. In sum, Mohanan's bracket erasure stands aside in the concert of no look-back devices.

293

5.6. Modification-inhibiting no look-back: first timid steps in the 70s and 80s

294

5.6.1. Early formulations in phonology and syntax In §279 we have come across a version of no look-back that is different from all others: Kaye (1995) achieves underapplication of stress assignment at the outer domain of [[parent] hood] by virtue a mechanism that makes previously computed strings immune against further modification on later cycles. As we will see in §304, this is precisely the description of the Phase Impenetrability Condition that is assumed in current minimalist syntax. In this book, the idea that previously interpreted strings cannot be further modified is referred to as modification-inhibiting no look-back. Before looking at Kaye's implementation, this section identifies earlier incarnations of the idea that may be found in the literature under the headers of structure preservation and the Free Element Condition in phonology, and as Riemsdijk's (1978) Head Constraint in syntax.

295

5.6.2. Phonology I: stress and the Free Element Condition There is a body of literature in the early autosegmental period where autosegmental structure that encodes stress and syllabic generalisations, i.e. metrical and syllabic trees, are subject to structure preservation (at least in some languages, see Steriade 1988 for the parametric variation assumed). Rules that build metrical and syllabic trees are thus held to be unable to "destroy" structure that was already erected on previous cycles. Let us briefly review one analysis for each phenomenon, stress and syllable structure. McCarthy (1980) (which is about stress and syncope in Damascene Arabic) and Poser (1986, 1989) (on stress in Diyari, South Australian) set aside, all cases for structure preservation of foot structure that I have come across are about the same data that were introduced by Steriade (1988):

No look-back devices: implementations since 1973 245

stress in constructions with an enclitic element in Latin and Greek (also Halle 1990, Halle & Kenstowicz 1991). In Latin, word stress is antipenultimate in trisyllabic or longer words (penultimate in bisyllabic and ultimate in monosyllabic items) unless the penultimate syllable is heavy, i.e. contains a long vowel or a closed syllable, in which case stress falls on this penultimate syllable. Hence fá.ce.re "to make" and có.lu.bra "blindworm" (penultimate light) vs. ha.bée.re "to have" and a.rís.ta "ear (of corn)" (penultimate heavy). In case an enclitic joins, however, stress shifts and invariably falls on the last syllable of the word (i.e. the syllable immediately to the left of the enclitic). Hence límina "thresholds" and éa "this" end up as liminá-que "and the thresholds" and eá-propter "for this reason". That the overall string is not treated like a word is shown by the fact that the forms *limína-que (penultimate light) and *ea-própter (penultimate heavy), which are expected when the string is computed as a whole, are not produced. The standard analysis of the Latin pattern adopted by Steriade (1988:297) is that the last syllable is extrametrical; on the remaining string binary, left-dominant feet are erected from right to left. Stress then falls on the last foot: on its first syllable by default, on the second syllable in case it is heavy. Hence fá.ce.re and có.lu.bra identify as (fa.ce)re and (co.lu)bra, respectively (with -re and -bra extrametrical); since the second syllable of the foot is not heavy, the first syllable receives stress. The footing of ha.bée.re and a.ris.ta is identical, i.e. (ha.bee)re and (a.ris)ta, but this time the second syllable of the foot is heavy and therefore attracts stress. The analysis of words that occur with an enclitic is then based on the idea that the foot which was acquired on the word-internal cycle cannot be erased when the word is computed together with the enclitic. Computational domains thus identify as [[word] enclitic]. In order to achieve this effect, Steriade (1988:286) calls on Prince's (1985) Free Element Condition (FEC). The FEC restricts rules that erect foot structure to strings that do not possess any such structure yet. In our Latin examples, the last syllable is systematically extrametrical and therefore available for foot construction on the outer cycle when the enclitic is added. The rest of the word, however, is already footed and hence immune against an eventual reparse by rules that build foot structure. As a result, the string that is subject to footing on the outer cycle is made of the last syllable of the word and the body of the enclitic (which of course is also unfooted). The same stress rules then apply again to the input (li.mi)na-que and (e)a-propter: the last syllable is made extrametrical, and a foot is constructed on the remaining material; the result is (li.mi)(na)-que

246 Chap 9: Kaye (1995), selective spell-out and freezing no look-back and (e)(a-prop)ter. Stress then falls on the rightmost foot (i.e. liminá-que), and in case this foot is branching, always on its first syllable. This latter aspect is at variance with respect to the stress assigning rule that applies word-internally: the eventual heaviness of the second syllable is disregarded. Therefore eaá-propter is produced.

296

5.6.3. The FEC is weak: process-specificity, parameterisation and restriction to structure that is absent from the lexicon We are thus facing an analysis where previously assigned structure is immune against further modification on later cycles. But this is only true for stress and hence metrical structure: stress is the only thing that the Free Element Condition is competent for (even if it was later applied to syllable structure as well, see the following section). There is no more general ambition regarding other phenomena or other types of structure. Also recall that the FEC only applies in certain languages: it is not a property of computation as such. Finally, we are only talking about previously erected autosegmental structure (i.e. structure that is absent from the lexicon) – not about previously computed strings. Already computed strings will be happy to receive metrical structure provided that they do not already possess such structure. The Free Element Condition is thus much weaker than the modern Phase Impenetrability Condition that is known from current minimalist syntax (and which is introduced at greater length in §304). We have already come across a hint that process-specific no look-back may be required in phonology, and this was precisely on the occasion of the analysis of (English) stress (§241): while stress placement is strictly bound by the limits of the word, English shows external sandhi phenomena such as t-flapping. Therefore the word can be no general barrier for phonological processes: for some its boundaries are insuperable, while for others they are invisible. Hence the radical Chomskian "compute-and-forget" that is expressed by Phase Impenetrability is much too strong: it does not fit the situation that is encountered in phonology. More evidence to this end is adduced below (§§302,823). For the time being, we have seen a modificationinhibiting look-back mechanism that is specific to stress assignment, without this process-specificity being intended, though: there is no comparison in the literature quoted with other processes. Modification inhibition is postulated because it can solve a particular problem set.

No look-back devices: implementations since 1973 247 297

5.6.4. Phonology II: structure preservation in syllabification The other process to which modification-inhibiting no look-back was applied in the 80s is syllabification, i.e. the erection of syllable structure over strings that are lexically unsyllabified. The idea is the same as before: "old" syllable structure that was built on a previous cycle cannot be erased or modified by computation on later cycles. Examples typically involve suffix-prefix asymmetries that are explained by a difference in domain structure: while suffixes share a domain with the root ([root - suffix]), the prefix is located outside of this domain ([prefix [root - suffix]]). As a consequence, the syllabification of the root cannot be undone or modified on the outer cycle when the prefix is added. This produces the "unnatural" syllabic parse CVC.VC when the root is vowel-initial and the prefix consonant-initial: the FEC applied to syllable structure prohibits resyllabification across prefix boundaries. Data along these lines are analysed by Steriade (1982:84ff, 1984, 1988:205, Greek and Latin), van Oostendorp (1994, Dutch) and J. Harris (1993, Spanish). In Spanish for example, the rule that takes s to h in coda position is a diagnostic for syllable structure. Hence the contrast between deseo [deseo] "desire" and des-echo [deheo] "waste" must be due to contrasting syllabification: the s is an onset in the former (de.se.o), but a coda (des.e.o) in the latter case. In other words, there is no syllabification across the prefix boundary of des-. J. Harris (1993:182ff) explains this lack of optimising CV parses by the fact that the prefix-final consonant cannot be syllabified into the onset of the root since this would modify the syllable structure that was built on the root cycle. Steriade (1982:355) is explicit on this mechanism: "syllabification rules […] do not change already assigned syllable structures." As for stress, modification-inhibiting no look-back in these analyses is taken to be specific to a particular process, syllabification, and only concerns autosegmental structure that is absent from the lexicon.

298

5.6.5. Syntax: Riemsdijk's (1978) Head Constraint Let us now look at a syntactic precursor of modification-inhibiting no lookback. Chomsky's (1973) original formulation of the SCC was meant to cut down the generative power of syntactic computation. It initiated a theory of movement whereby long-distance displacement is cyclic, i.e. where items leapfrog through a number of intermediate landing sites before reaching

248 Chap 9: Kaye (1995), selective spell-out and freezing no look-back their final location. Abels (2003:16ff) traces back the evolution of cyclic movement since 1973, distinguishing theories where intermediate jumps are uniformly defined (uniform path) from those where they follow a nonuniform track (punctuated path). The latter have three main incarnations: the Extended Standard Theory (Chomsky 1973 and developments thereof) where movement was defined by bounding (or cyclic) nodes and went through designated escape hatches, Barriers in GB (Chomsky 1986) and today the Phase Impenetrability Condition (PIC, Chomsky 2000a et passim, see §304 below). While Chomsky's (1973) initial formulation of the SCC does not include any modification-inhibiting no look-back, Riemsdijk's (1978:160) Head Constraint, also couched in the Extended Standard Theory, does exactly the same labour as the modern Phase Impenetrability Condition. This is what Abels (2003:41) shows. Consider both conditions under (115) below. (115) comparison of the Head Constraint and the PIC a. "The Head Constraint No rule may involve Xi/Xj and Yi/Yj in the structure …Xi…[Hn…[H'…Yi…H…Yj…]H'…]Hn…Xj… (where H is the phonologically specified (i.e. non-null) head and Hn is the maximal projection of H […])." Riemsdijk (1978:160) b. "Phase Impenetrability Condition In phase α with head H, the domain of H is not accessible to operations outside α, only H and its edge are accessible to such operations." Chomsky (2000a:108)

Abels (2003:41) concludes that both conditions are identical: "no relation can be established between an element outside of (phase) XP and the complement of the phase head or anything properly contained in the complement of the phase head." Additional information that is needed in order to make either constraint work is the kind of XP to which it applies. While Riemsdijk (1978) provides a list (VP, NP, AP, PP), the definition of phase heads in phase theory is subject to debate (see §773). If the effect of the two conditions is identical, though, their motivation is not, and hence their applicability to phonological matters is not either. The PIC roots in an interface-based reasoning whereby, following the minimalist perspective of third factor explanations (Chomsky 2005), an extra-linguistic property of the cognitive system marshals linguistic mechanisms. That is, Chomsky hypothesises that active memory (work bench

No look-back devices: implementations since 1973 249

memory) is too small in order to compute an entire sentence in one go. The sentence thus needs to be cut into pieces (that is, phases) and is computed in several successive waves. At the end of the computation of each phasechunk, the output is sent to LF and PF for interpretation and then "forgotten": it comes back "frozen" and may not be accessed anymore by further computation (phase theory is introduced at greater length in §304). The PIC under (115b) then merely specifies which piece exactly is spelled out and "forgotten": the complement of the phase head XP (the edge of the phase, i.e. the head of the XP and the Specifier, are only spelled out at the next higher phase). In this scenario, it is the interactionist architecture of phase theory that makes PF (and LF) enter the scene: something can be "forgotten" only because it was spelled out. This interpretation-based definition of "old unmodifiable" and "new modifiable" chunks is applicable to phonology: strings that have already been subject to interpretation are "frozen" and not further modifiable. We have seen that this is precisely what defines modification-inhibiting no look-back: on Kaye's analysis (§279), the stress that is assigned to the inner cycle of [[parent] hood] cannot be further modified on the outer cycle. By contrast, phonology is not touched or concerned at any point in the reasoning of the 70s: the motivation and definition of the chunks that are not available for further computation is purely syntax-internal. It cannot be extended to phonology no matter how much good will is put into the venture since interpretation plays no role, and anyway phonology will be unable to interpret the arboreal configuration that defines the unmodifiable "old" chunk. Only an interpretation-based definition of unmodifiable chunks provides a common language for syntax and phonology. The reason why interpretation was out of business in the 70s is the strictly non-interactionist architecture that was inherited from SPE: all concatenation was done before all interpretation (§86). Hence there was no way to even formulate an interpretation-based definition of "old" and unmodifiable chunks.

250 Chap 9: Kaye (1995), selective spell-out and freezing no look-back 299

5.7. Modification-inhibiting no look-back in Kaye's system

300

5.7.1. Kaye calls on SCC-M, but applies something else Let us now look at how modification-inhibiting no look-back is implemented in Kaye's system. It was shown in §279 that modification-inhibiting no look-back is critical in Kaye's analysis of affix class-based phenomena: it achieves underapplication. This contrasts with all other analyses, where no look-back plays no role. Also recall that Kaye (1992a:142, 1995:307) makes explicit reference to Chomsky (1973), Kean (1974) and Mascaró (1976) when he introduces his no look-back mechanism, but leaves Kiparsky's (1982a,b) SCC-K unmentioned. This is because derived environment effects are nothing that Kaye ambitions to account for (§284): the scope of his theory are only affix class-based phenomena. In a note, he is explicit on the fact that Kiparsky's version of no look-back is an entirely different object. (116) "I mean strict cyclicity in its original sense as proposed by Chomsky (1973) and applied to phonology by Kean (1974). It has subsequently been used in a very different sense, for example, in lexical phonology." Kaye (1995:330, note 23, emphasis in original)

While the non-reference to Kiparsky's derived environments corresponds to Kaye's practice, his claim to use Chomsky's original strict cyclicity turns out to be unsubstantiated. What Kaye really applies is modification-inhibiting no look-back: whatever comes back from interpretation is "frozen" (Chomsky's terminology, see §306), which means that it is entirely invisible and may not be counted or accessed anymore by further computation. That Kaye does not apply Chomsky/Mascaró's SCC-M is obvious from the simple fact that the SCC-M is perfectly unable to do the labour required. Recall from §279 how [[parent] hood] is derived: in the inner domain, regular penultimate stress assignment produces párent. The same stress rule also applies when the outer domain is assessed – but is blocked since the result of its application, *parént-hood, would require to undo the stress-assignment that was achieved on the inner domain. The SCC-M, however, would not prevent stress from being reassigned at all: the requirement that the rule must use material which is introduced on the latest cycle is met: -hood is part of the string that the rule uses in order to compute stress assignment. It is only when no look-back prohibits the modifica-

No look-back devices: implementations since 1973 251

tion of previously interpreted strings that underapplication of the stress rule in párent-hood is guaranteed.

301

5.7.2. Kaye's modification-inhibiting no look-back Given that Kaye (1992a, 1995) explicitly calls on the SCC-M, it is not immediately obvious to understand that what he really does is quite different. Let us therefore look at the example from French that he uses both in Kaye (1992a:142ff) and Kaye (1995:306ff). Data and analysis are originally due to Prunet (1986, 1987). French shows a contrast between mon ami [mç)n ami] "my friend" and bon ami [bçn ami] "good friend": the vowel of the possessive (mon "my") is nasalized, while the vowel of the adjective (bon "good") is not. Both determiners bear a liaison consonant at their right edge: the [n] is only present when the following noun begins with a vowel. It is absent before consonants (mon café [mç) kafe] "my coffee", bon café [bç) kafe] "good coffee") and in case the words are pronounced in isolation (mon [mç)] "my", bon [bç)] "good"). Following standard autosegmental assumptions on French liaison (Encrevé 1988), liaison consonants are lexically floating. The critical contrast, Prunet and Kaye argue (on non-phonological grounds), is only in domain structure. While mon ami is the complex [[mon] ami], bon ami lacks internal structure: it identifies as [bon ami]. Domain structure and the lexical ingredients, then, produce the structures under (117) below. (117) French mon ami vs. bon ami: input to the phonology a. mon ami b. bon ami O N O N O N O N O N O N | | | | | | | | | | | | [[ x x ] x x x x ] [ x x x x x x | | | | | | | | | | m o n a m i b o n a m i

]

Two processes now apply: floating consonants associate to available consonantal positions (the floating nasal here behaves like all other floating consonants), and nasalization of vowels is effected by nasals that occur domain-finally or before a consonant (this is a specific behaviour of nasals, but general in the language).

252 Chap 9: Kaye (1995), selective spell-out and freezing no look-back The derivation then proceeds cyclically: phonology is first done on the inner domain of [[mon] ami]; the result is vowel nasalization, i.e. the association of the floating nasal to the preceding nucleus. On the outer cycle, liaison also associates the nasal to the following onset, which is now available. In the end, the nasal consonant enjoys a double association: it contributes to the pronunciation of the preceding nucleus and the following onset; the result is [mç)n ami]. By contrast, the nasal in [bon ami] will undergo liaison, but fails to nasalize the preceding vowel because bon is not a domain by itself – hence the nasal is never domain-final. Therefore the triggering environment for nasalization is never met, and the result is [bçn ami]. Table (118) below shows the output of both derivations. (118) French mon ami vs. bon ami: output of the phonology a. mon ami [mç)n ami] b. bon ami [bçn ami] O N O N O N O N O N O N | | | | | | | | | | | | x x x x x x x x x x x x | | | | | | | | | | m o n a m i b o n a m i

Critical for this analysis is that the association of the nasal to the preceding nucleus that is achieved on the inner domain under (118a) is not undone when the outer domain is subjected to the φ-function. In his 1992a article, Kaye comments that (119) "the PSC [Principle of Strict Cyclicity] will prevent tampering with the internal structure of the previous cycle and so N will remain linked to the nucleus." Kaye (1992a:144)

In his 1995 article, Kaye provides a more explicit formulation, which appears under (120) below. (120) "When phonology is done on the external domain, an empty onset is available for the n. However, the principle of strict cyclicity states that the association created in the inner domain cannot be undone in an external domain. The association remains and the n also links to the available onset." Kaye (1995:307, emphasis in original)

No look-back devices: implementations since 1973 253

This is what Kaye's no look-back device really does. A formulation that is neutral with respect to particular phenomena appears under (121) below. (121) Strict Cyclicity Kaye a string that has already been subject to interpretation (on a previous cycle) cannot be further modified on a later cycle.

It may be objected for the specific analysis of French that no particular (no look-back) mechanism is needed in order to maintain the association line between the nasal and the preceding nucleus on the outer cycle of mon ami. Indeed, there is no phonological process in the language that destroys association lines: there is no reason why the nasal should be disassociated. Hence whatever association is achieved at some point in a derivation simply piles up with all other associations. While this objection may call into doubt the necessity for modification-inhibiting no look-back in the particular French example, it does not discredit the idea itself: other cases such as stress assignment (§279) do require modification-inhibiting no look-back in Kaye's system.

302

5.7.3. Chomsky's "spell-out and forget" is too strong for phonology: "don't undo" and process-specific no look-back The formulation under (121) is obviously too strong in regard of external sandhi phenomena, i.e. phonological processes that apply across word boundaries, which violate modification-inhibiting no look-back by definition: words are always spelled out in Kaye's system. They should therefore be "frozen" no matter what, and hence any further phonology that applies across word boundaries should be prohibited. This is obviously not the case: external sandhi (postlexical phonology) exists. This problem raises the issue of the word-level hurdle that we have already come across in §241: the word level seems to be an insuperable barrier for some processes, but not for others. Also recall that processspecific no look-back is practised in the early literature that made modification-inhibiting no look-back emerge (§296): structure preservation and the Free Element Condition concerned only stress and/or syllabification, and only structure was concerned that is absent from the lexicon (segmental structure for example is left unsubjected to no look-back). It therefore appears that phonological phenomena are unsuited for Chomsky's fully radical "spell-out and forget": a weaker version of modifi-

254 Chap 9: Kaye (1995), selective spell-out and freezing no look-back cation-inhibiting no look-back is needed (something that Chomsky actually senses: see §306). One way to weaken the formulation is indicated by the quote under (120): Kaye says that properties of the string that were achieved by phonological computation on previous cycles cannot be undone later on. This is also implied by structure preservation and the Free Element Condition: only those properties of the string that were achieved by previous computation (metrical and syllable structure) are immune. Other properties may happily be modified. We are thus in presence of the two candidates for weakening the excessive strength of Chomsky's "spell-out and forget" that appear under (122) below. (122) adapting Chomsky's "spell-out and forget" to phonology a. process-specific no look-back the prohibition to modify a previously interpreted string applies selectively to individual processes: every process is specified for being subject to the no look-back restriction or not. b. don't undo! modification-inhibiting no look-back only concerns properties of the string that were acquired by previous computation. Lexical properties may be modified at any time even if the string was already interpreted.

Both candidates do the job for the pattern that was discussed in §241: word stress is strictly bound by the limits of the word (it remains unmodified by stress assignment when strings grow larger than the word), but there is external sandhi such as t-flapping across word boundaries (hi(t)[ɾ] Ann). Process-specific no look-back (PIC à la carte) solves the problem by making stress assignment, but not t-flapping, subject to the prohibition to consider "old" strings. On the other hand, (122b) is also successful: word stress cannot be undone when chunks grow bigger than the word because it was achieved by phonological computation. External sandhi, however, may happily go into effect since the word-final dental of hi(t)[ɾ] Ann is lexical: none of its properties is the result of previous computation. The latter solution may run into trouble with stress clash (thirtéen vs. thírteen men) where word stress is modified (if by a different process, see §240, note 64). But let us leave it at that for the time being. The take-home message is that the minimalist PIC "spell-out and forget" is too strong for phonology, and that there are two candidates for a weaker, phonologyfriendly version of modification-inhibiting no look-back. The issue is further discussed in §825 below.

No look-back devices: implementations since 1973 255 303

5.7.4. Morpheme-specific phonologies and selective spell-out do the same job and are therefore mutually exclusive The challenge raised by affix class-based phenomena is the organisation of underapplication: a process that is active in the language must be prevented from applying to strings that are created by the attachment of a particular affix class: to class 2 strings in the rule-blocking, to class 1 strings in the rule-triggering pattern. Theories propose various means of organising underapplication, and this is what makes them different. We have seen three ways of approaching the problem: the one of Lexical Phonology, of Halle & Vergnaud and of Kaye. The front line runs between the former and the two latter: Lexical Phonology achieves underapplication by distinct morpheme-specific phonologies (§§147,150); on the other hand, multiple computational systems play no role in the account of Halle & Vergnaud (§228) and Kaye (§279), which is based on selective spell-out and supplemented with modification-inhibiting no look-back in the case of Kaye. There are thus two philosophies of how to approach affix class-based phenomena: multiple computational systems that are morpheme-specific do the same job as selective spell-out (eventually supplemented with modification-inhibiting no look-back). The empirical coverage is not exactly the same, and the conceptual problems that are met by each system are also different (see §310 below). But what is for sure – the take-home message of this section – is that no system can reasonably implement both tools. Either underapplication is organised by morpheme-specific mini-phonologies, or by selective spell-out: no theory can feature both, except if redundancy is not an issue.

304

5.8. On the (modern) syntactic side: derivation by phase and Phase Impenetrability

305

5.8.1. Interactionism is enforced by the minimalist concern for economy of cognitive resources (active memory) The minimalist concern for extra-linguistic factors and the economy of cognitive resources has significantly modified the perspective and practice of generative grammar: syntax proper (where explanations are now sought at the interface, rather than in syntax itself) and the interface with LF and PF (which has become interactionist) are directly impacted; also, a no look-

256 Chap 9: Kaye (1995), selective spell-out and freezing no look-back back device, the Phase Impenetrability Condition (PIC), has become a central tool of the theory. This broad and far-reaching evolution was initiated by Chomsky's (2000a et passim) derivation by phase, all aspects of which of course cannot be covered here. Some more discussion is provided in §§672,771 (overview literature includes Frascarelli 2006, Lasnik & Lohndal 2009 and Freidin & Lasnik forth). Let us first look at interactionism. The generative interface architecture, which stands unchallenged to date, was designed in the 60s. The inverted T model (§86) divides grammar into three computational devices: one where concatenation is done (morpho-syntax), and two others that interpret its result (LF and PF). Until Epstein et al. (1998:46ff), Epstein (1999), Uriagereka (1999) and Chomsky (2000a), it was taken for granted in all successive and competing blends of syntax that all concatenation is done before all interpretation (§86). 73 That is, morpho-syntax first completes the entire derivation, and then ships off the result to PF and LF. Interactionism is a fundamentally different take on how morphosyntax and the interpretational modules communicate: it holds that concatenation and interpretation are interleaved. This idea was carried into linguistic thinking by Lexical Phonology, of which it is an inescapable headstone (see §§146,212). Shortly after its introduction, interactionism provoked a strong reaction on the side of generative orthodoxy: Halle & Vergnaud (1987a) tried to restore the take of SPE by proposing an analysis of affix class-based phenomena under the anti-interactionist banner (§222). Until Epstein et al. (1998:46ff), Epstein (1999) and Uriagereka (1999), the debate on interactionism was confined to phonological quarters. All of a sudden, then, it enters the syntactic scene in the late 90s, with a different name, though (multiple spell-out, derivation by phase), and as far as I can see without any reference to the phonological precedent. Uriagereka's multiple spell-out motivates the come and go between syntax and interpretational devices on economy-driven minimalist grounds. Uriagereka explains how the (radically) derivational perspective of an interactionist architecture is made possible by the minimalist abandon of the distinction between deep and surface structure: while there was no place for derivational mechanisms in the representational GB-times of the 80s, the new perspective actually relates back to the more derivational 70s.

73

Except by Bresnan (1971), to be discussed in §308 below, and in Jackendoff's (1997 et passim) parallel model (on which more in §722) where the inverted T is rejected.

No look-back devices: implementations since 1973 257 (123) "An older model provides an intriguing way of going about the general deduction of the LCA [Linear Correspondence Axiom]. For reasons that become apparent shortly, I refer to it as a dynamically split model. The origins of this outlook are discussions about successive cyclicity and whether this condition affects interpretation: are the interpretive components accessed in successive derivational cascades? Much of the debate was abandoned the moment a single level, S-structure, was postulated as the interface to the interpretive components. Now that S-structure has itself been abandoned, the question is alive again: what would it mean for the system to access the interpretation split in a dynamic way?" Uriagereka (1999:255, emphasis in original)

On the other hand, Samuel Esptein and colleagues arrive at the same result (actually in its most extreme incarnation, Spell-out-as-you-Merge, see §776) on principled grounds: their derivational approach to syntax is incompatible with single LF and PF levels. (124) "[W]e depart from the minimalist conception of linguistic levels and develop a form of the derivational model of syntax, in which the derivational process not only determines syntactic relations but also provides information directly to the interface systems. Under this derivational model, the syntactic relations in question are deduced from the way these relations are created by CHL [Chomsky's computational system of human language, see §638]; furthermore, upon creation, these relations enter into the interpretive procedures without mediation of linguistic levels." Epstein et al. (1998:47) "[E]ach transformational rule application constitutes a 'phase', which we believe to be the null hypothesis. Like Epstein et al. (1998), […] and Uriagereka (1999), Chomsky (2000[a] […]) abandons the Y-model's single split to PF and LF in which interface interpretation applies only after all transformational rules have applied." Epstein & Seely (2002:77)

Multiple spell-out was then worked out in Chomsky's (2000a, 2001 et passim) derivation by phase, which today is the standard way of conceiving of the communication between morpho-syntax and PF/LF in generative quarters. After almost 50 years of existence, the generative mainstream has thus abandoned "all concatenation before all interpretation" and became interactionist. As was mentioned, it appears that this about-turn was made without reference to Lexical Phonology and the quarrel regarding interactionism that took place in the 80s. If the result of the minimalist move is exactly what Lexical Phonology proposed in the 80s, its motivation is very different: it was already

258 Chap 9: Kaye (1995), selective spell-out and freezing no look-back mentioned that Uriagereka's multiple spell-out is motivated by the minimalist concern for economy. Chomsky argues that linguists are invited to cut the derivation of a whole sentence into smaller chunks – phases – if extralinguistic properties of the cognitive system are taken seriously: morphosyntactic computation uses active memory ("work bench memory", "workspace"), a very limited cognitive resource. Chomsky's (2000a:101) minimalist take is that the faculty of language has "optimal" design properties and operates with economy principles. This means, among other things, that computational complexity is unwarranted. (125) "There is mounting evidence that the design of FL [faculty of language] reduces computational complexity. That is no a priori requirement, but (if true) an empirical discovery, interesting and unexpected. One indication that it may be true is that principles that introduce computational complexity have repeatedly been shown to be empirically false." Chomsky (2001:15)

The bias against computational complexity is thus assumed to have extra-linguistic causes, i.e. the restriction on the availability of active memory, an expensive cognitive resource. In a linguistic derivation, then, a whole sentence is too big and too demanding computationally speaking in order to be able to be processed in one go. If sentences are built step by step, the burden imposed on active memory by the computation of the successive pieces is reduced. The interactionist come and go between syntax and LF/PF is a consequence of this implementation-based approach, which Chomsky first explicitly mentions in his Minimalist Inquiries (Chomsky 2000a:106). Chomsky advertises the result of the interactionist architecture as follows. (126) "If such ideas prove correct, we have a further sharpening of the choices made by FL [faculty of language] within the range of design optimization: the selected conditions reduce computational burden for narrow syntax and phonology." Chomsky (2001:15)

Chomsky thus explicitly extends the invisibility of previously interpreted strings to phonological computation. This is also shown by the following quote. (127) "The computational burden is further reduced if the phonological component too can 'forget' earlier stages of derivation." Chomsky (2001:12f)

No look-back devices: implementations since 1973 259

In the recent development of generative theory, the global architecture of grammar has thus become interactionist under the pressure of economy-orientated design and the concern for cognitive (extra-linguistic) plausibility (third factor explanations, Chomsky 2005), which are the driving force of the minimalist programme.

306

5.8.2. Phase Impenetrability is the instrument of active memory economy The economy of active memory supposes that morpho-syntactic computation is unburdened with the management of the structure of previous phases: on each pass, only the material of the current phase is subject to morpho-syntactic computation. Chomsky (2001) refers to the inertness of previous phases by saying that they are "frozen in place" (Chomsky 2001:6) or "forgotten" (Chomsky 2001:12) by later computation. The formal expression of this inertness, then, is the Phase Impenetrability Condition (PIC), which is thus understood as a protection against the overheat of active memory.74 (128) "The whole phase is 'handed over' to the phonological component. The deleted features then disappear from the narrow syntax. […Uninterpretable features] have been assigned values (checked); these are removed from the narrow syntax as the syntactic object is transferred to the phonology. The valued uninterpretable features can be detected with only limited inspection of the derivation if earlier stages of the cycle can be 'forgotten' – in phase terms, if earlier phases need not be inspected. The computational burden is further reduced if the phonological component too can 'forget' earlier stages of derivation. These results follow from the Phase-Impenetrability Condition (PIC) (MI [Minimalist Inquiries, i.e. Chomsky 2000a], (21)), for strong phase HP with head H, (7) The domain of H is not accessible to operations outside HP; only H and its edge are accessible to such operations." Chomsky (2001:12f)

Note that according to Chomsky phonology is concerned by computational economy just like syntax. Chomsky (2004) reiterates the need for the PIC to exist in phonology; interestingly, though, he senses that "spell-

74

The notion of phase edge that is mentioned in the quote below, as well as its eventual equivalent in phonology, is discussed in §765.

260 Chap 9: Kaye (1995), selective spell-out and freezing no look-back out and forget" may be too strong a condition on phonological computation. This is indeed what we have seen in §302. (129) "Φ [the phonological component] is greatly simplified if it can 'forget about' what has been transferred to it at earlier phases; otherwise, the advantages of cyclic computation are lost. Although the assumption may be somewhat too strong, let us assume it to be basically true " (Chomsky 2004:107f)

Phase Impenetrability is thus a cognitive-economy-born no lookback device. Curiously enough, the syntactic literature does not appear to make reference to the phonological no look-back devices that are an important piece of phonological thinking. Also, the literature typically does not mention that Phase Impenetrability is a revival of Chomsky's (1973) Strict Cycle Condition (but see Müller 2004): the idea of no look-back was born in syntax. Just as on the phonological side some time ago, there is ongoing debate in current syntax what exactly "frozen" means. While this question is beyond the scope of the book, it is touched on in §778. Phase Impenetrability is thus an old acquaintance, but comes in a new guise: motivation from cognitive economy was completely absent in earlier incarnations of no look-back devices, which were exclusively based on grammar-internal considerations.

307

5.8.3. The old and the new: a memory keeper and/or diacritic marking needed? An obvious consequence of Phase Impenetrability that I could not find any trace of in the literature is the need for a memory keeper and/or diacritic marking of "old" strings. The PIC requires that the computational system make a difference between "old" and "new" strings, i.e. between those that have already been interpreted (frozen, not a possible input to computation), and those which are still uninterpreted and therefore a good input to computation. There are two ways of implementing this idea: either the "old" string is reintegrated into the workspace when it comes back from interpretation, and the computational system is able to distinguish it from freshly concatenated material; or it never comes back to the computational workbench and instead is stored in some intermediate memory. In this case, all stored pieces are concatenated (in the correct order), and the sentence is pronounced, when the last piece of the highest CP is interpreted.

No look-back devices: implementations since 1973 261

The former solution requires some diacritic marking of "old" strings: otherwise the computational system could not know that a string is "frozen". The latter solution does not, but instead needs a memory keeper and workbench-independent storage (more on this in §§797f). The memory keeper option appears to be in line with the philosophy of Phase Impenetrability: the PIC is motivated by the discharge of active memory. The question is left open for the time being. We will return to it at greater length in §794. We know already that the old-new distinction is also relevant in phonology: this is what modification-inhibiting no look-back is all about (§§293,299). In the chapter on OT we will see that it is also relevant in systems that do not directly relate to modification inhibition or the PIC (McCarthy's 2003a Comparative Markedness §519, van Oostendorp's 2006a colours §505). What may be said already here is that the memory keeper option is not workable in phonology for the reasons discussed in §302: processes whereby pieces of the "old" string are modified under the influence of a piece of a "new" string are trivial and commonplace: e.g. external sandhi, but also regular palatalisations where a suffix-initial front vowel alters a stem-final velar (provided that it can be independently shown that there is a phase boundary between the stem and the suffix). Therefore the "old" string must be somehow available for further computation, even if (parts of it) cannot be modified. This leaves us with the two options (and possibly still others) that are discussed in §302: process-specific PIC and "don't undo!". On the latter count (but not on the former), computation needs to be able not to identify which part of the string that is submitted to computation is "old", but which properties are the result of previous computation. Or, in other words, "old" in this context does not mean "a previously uninterpreted morpheme", but "a property of the string that was not created by previous computation, i.e. which was present as such in the lexicon." Again, some diacritic marking seems to be necessary in order for the computational system to be able to identify non-lexical properties of the string. Implications of processspecific PIC are further discussed in §825. 308

5.8.4. Ancestors of multiple and selective spell-out: Bresnan (1971) Uriagereka (1999:277), Chomsky (2000a:150) and historical overviews of generative grammar such as Lasnik & Lohndal (2009:48) and Freidin & Lasnik (forth) identify Bresnan (1971) as an early case of multiple and

262 Chap 9: Kaye (1995), selective spell-out and freezing no look-back actually selective spell-out of PF properties. On the LF side, they mention the Ph.Ds of Ray Jackendoff (1969) and Howard Lasnik (1972) as early sources of interpretation that occurs in the midst of the syntactic derivation. Like (phonological) interactionism and its modern syntactic version derivation by phase (or multiple spell-out), these analyses thus dispense with the principle "all concatenation before all interpretation" that accompanied the inverted T since the 60s (see §86). Halle & Vergnaud (1987a) are credited with the paternity of selective spell-out in §§220,225; let us therefore have a brief look at Bresnan's argument. Bresnan's article is well-known in syntactic quarters since Bresnan is a syntactician and writes about intonation (sentence stress), which is the phonological phenomenon that syntacticians are confronted with first if they meet any phonology at all. Bresnan's (1971:258) central example is reproduced under (130) below (based on an observation by Newman 1946; underscored words are intonationally prominent). (130)

a. Helen left directions for George to follow. b. Helen left directions for George to follow.

In these examples, sentence stress impacts meaning, and the impact is not optional: under (130a) directions is the object of the verb to follow, while under (130b) directions has no argumental involvement with to follow. That is, Helen has left directions which George is supposed to follow under (130a) while Helen has left directions to the effect that George should follow under (130b). SPE has defined a Nuclear Stress Rule (NSR) that distributes primary intonational prominence to the rightmost item that bears primary word stress in the current cycle (Chomsky & Halle 1968:17f). In case of successive applications in several cycles, the latest application of the NSR defines primary stress: all previously distributed intonations are demoted by one on a scale of prominence. Bresnan argues that the contrast under (130) cannot be derived if the NSR is blindly applied to the overall output of the syntactic derivation: the two sentences under (130) involving identical words and word order, the intonation should always be as under (130b), i.e. rightmost. If on the other hand the NSR applies every time an S is completed, the derivational history of embedded S's matters. Bresnan (1971:260) assumes the following structures.

No look-back devices: implementations since 1973 263 (131)

a. [S Helen left [NP directions [S for George to follow directions S] NP] S] b. [S Helen left [NP directions [S for George to follow S] NP] S]

She then derives the intonational difference from the fact that follow has a direct object under (131a) while it does not under (131b). Being the rightmost constituent throughout the entire derivation, follow receives stress under (131b). By contrast under (131a) the rightmost directions is deleted after the completion of the embedded S, and this is why it bears primary stress in its surface position. Or, in other words, the surface intonation that we see on directions was crucially assigned before the word was moved. The detail of the derivation is more intricate (and not really made explicit by Bresnan): it appears that some PIC-like device is needed in order to make the material of the embedded S under (131a) invisible when the matrix S is computed – otherwise the moved copy of directions would not be the rightmost item, and its primary stress would be overridden when the matrix S is computed. Be that as it may, the take-home message is that Bresnan (1971) indeed practised a form of interactionism since in her analysis a phonological rule, the NSR, crucially applies to a word in situ and before the syntactic derivation is completed. In modern vocabulary, one would say that S is a phase head. Note that Bresnan calls the NSR a cyclic rule, but what she calls a cycle has got nothing to do with the cycle as defined in SPE (where roughly every morpheme is a cycle §101). Rather, she appeals to the transformational cycle as defined by Chomsky (1970). This means that she also practised selective spell-out in the sense of Halle & Vergnaud (1987a) (§225): not only is concatenation interspersed with interpretation, but also only a subset of nodes (S in this case) is a spell-out point. In the end, though, Bresnan's analysis may turn out not to involve any interactionism, or selective spell-out for that matter. This is because intonation is maybe defined by a syntactic, rather than by a phonological operation. The surface marker of intonation is certainly a modulation of sound, but this effect may as well be achieved if PF interprets a string where intonational prominence is encoded due to syntactic marking. This issue is further discussed in §800.

264 Chap 9: Kaye (1995), selective spell-out and freezing no look-back 309

5.9. Conclusion In conclusion of this section, it is useful to point out how a number of concepts relate historically: there appears to be quite some confusion about no look-back devices (and it is true that one can think of less intricate matters). On the one hand, different incarnations of no look-back may not be related to each other (for example, bracket erasure is not related to the SCC-K, see §201). On the other hand, when they are, they are often held to be somehow equivalent. For example, Halle/Kiparsky's SCC-K is taken to be a development of Chomsky/Mascaró's SCC-M, even though the two mechanisms are quite different in design, scope and effect. Modification-inhibiting no look-back is an entirely different way of doing no look-back that emerged in the 80s in phonology and has also a syntactic precursor (Riemsdijk 1978). While the general idea is that previously interpreted strings are immune against further modification on later cycles, what exactly they are immune against is subject to debate. Chomsky's modern "spell-out and forget" is too strong for phonological matters, and there are at least two ways of weakening the formulation (§302). In any event, Chomsky/Mascaró's strict cyclicity has got nothing to do with modification-inhibition. In minimalist syntax, modification-inhibiting no look-back surfaces when Chomsky (2000a, 2001) describes previously interpreted strings as "frozen": further computation may "forget" about them. Even if the motivation is entirely different (computational economy, cognitive resources), it is quite surprising that Chomsky does not make reference to his own 1973 article, which introduced the idea of no look-back into linguistic thinking (Chomsky 1973 is absent from Chomsky 2000a, 2001), and also that the minimalist literature does not appear to mention the phonological precedent regarding modification-inhibition.

310

6. Empirical coverage: Kaye's system and affix class-based phenomena

311

6.1. The empirical testing ground: five English affix class-based phenomena Let us now run Kaye's version of selective spell-out against the record that was used in order to evaluate the empirical coverage of affix class-based phenomena that Lexical Phonology (§§163,166) and Halle & Vergnaud's (1987a) system (§248) offer. Table (132) below recalls the five relevant

Empirical coverage: Kaye's system and affix class-based phenomena 265

affix class-based phenomena that are found in English (the record was established in §§163,166) ("yes" means that the process goes into effect, "–" that it does not). (132)

English affix class-based phenomena monomorphemic situation a. Trisyllabic Shortening – b. nasal assimilation: un- yes /inc. stress placement transparent d. nasal cluster simplification 1. /gn/ → n 2. /mn/ → m

yes yes

class 1 affix

class 2 type affix

yes yes

– –

transparent

opaque rule-blocking

– –

yes yes

rule-blocking rule-blocking

rule-triggering rule-triggering

The table indicates the situation with class 1 and class 2 affixes, as well as in mono-morphemic situations. The latter are understood as inclusive of all triggering conditions that are unrelated to affix classes. That is, Trisyllabic Shortening goes into effect with class 1 affixes (s[æ]n-ity, against class 2 m[ej]den-hood), but requires a string of three syllables; it fails to apply, however, to mono-morphemic three-syllable strings (nightingale, ivory, cf. §129). Nasal assimilation has no other condition than the adjacency of a nasal with an obstruent and the affix class contrast between un- and in-: it always goes into effect morpheme-internally (there are no non-homorganic NC clusters in English mono-morphemic items). The situation of nasal cluster simplification is a little more complicated since it only goes into effect if an additional condition (other than affix class requirements) is met. That is, clusters are only simplified in morpheme-final position (§203): while /sign/ and /damn/ undergo the process and appear as sign and damn, the cluster remains undamaged morpheme-internally (i[gn]ore, a[mn]esia). Therefore the morpheme-final condition must be included in the statement of the process. The table indicates that it goes into effect in mono-morphemic situations because this is what sign and damn are. It does not in i[gn]ore and a[mn]esia because the morpheme-final condition is not fulfilled – not because of affix class related matters. This also shows that the resistance of morpheme-internal /gn/ and /mn/ against reduction has nothing to do with derived environments: sign

266 Chap 9: Kaye (1995), selective spell-out and freezing no look-back and damn are perfectly underived, but still undergo the process. The situation is thus quite different from the one where ivory and nightingale resist Trisyllabic Shortening, which is a real derived environment effect. The last column of table (132) classifies the five processes according to Kenstowicz & Kisseberth's typology that was introduced in §51: Trisyllabic Shortening, nasal assimilation and stress placement instantiate the rule-blocking pattern (level 1 rules in Lexical Phonology), while nasal cluster simplification illustrates the rule-triggering pattern (level 2 rules). Following the same procedure as before when Lexical Phonology and Halle & Vergnaud's (1987a) system were examined, the discussion below proceeds in this order. Finally, the special status of stress placement needs to be mentioned. Strictly speaking, this process is a category of its own: whether it goes into effect or not is not a question since it always does (there are no unstressed words). Rather, what is at stake is the transparent vs. opaque application of stress placement: mono-morphemic items (párent) and class 1 strings (parént-al) are in the former case (penultimate stress), while class 2 strings (párent-hood) receive opaque stress (non-penultimate). Stress placement is mentioned as a rule-blocking process because it is prevented from reapplying (rather than from applying) to class 2 strings. This is also the reason why it is treated as a level 1 rule in Lexical Phonology (§§147,164). 312

6.2. The rule-blocking pattern (level 1 rules)

313

6.2.1. Modification-inhibiting no look-back achieves underapplication at the outer cycle It was shown in §279 that in Kaye's system modification-inhibiting no look-back achieves underapplication for a phenomenon of the rule-blocking pattern, stress assignment. While the affixation of a class 1 affix is neutral with respect to spell-out, the concatenation of a class 2 affix triggers the interpretation of its sister, the root in [[parent] hood]. The stress that [parent] acquires at PF is then unmodifiable when the word is spelled out as a whole. By contrast, the word-level spell-out of [parent-al] is not confronted to any previously assigned stress; this allows for regular (i.e. transparent) stress placement.

Empirical coverage: Kaye's system and affix class-based phenomena 267

Trisyllabic Shortening is analysed along the same lines:75 it applies to class 1 strings (s[Q]n-ity1), but not to class 2 strings (m[ej]den-hood2) or to mono-morphemic items (nightingale). Underapplication to class 2 strings is a consequence of modification-inhibiting no look-back at the word level. Trisyllabic Shortening did not go into effect on the inner cycle of [[maiden] hood] because the trisyllabic condition was not met. At the outer cycle, then, the previously interpreted [ej] cannot be further modified. By contrast, the root of [san ity] has not experienced previous spell-out and is therefore free to undergo Trisyllabic Shortening at the word level. On the other hand, underapplication to mono-morphemic items such as [aj]vory or n[aj]ghtingale is a derived environment effect that, recall (§284), lies beyond the scope of Kaye's analysis of affix class-based phenomena. 314

6.2.2. Nasal assimilation requires independent spell-out of the affix Nasal assimilation is a case where the affix needs to be spelled out independently before it is merged. Recall from §274 that this is a possibility that Kaye's system provides for, but which is not available in regular spellout where only nodes of the morpho-syntactic tree, not terminals, can be interpretational units. In order to see this, it is enough to consider the result that is produced on the basis of [in possible] and [un [predictable]]: the class 2 affix un-, but not the class 1 affix in-, triggers the spell-out of the root. Nasal assimilation makes /nC/ sequence homorganic: it applies to [in possible], and nothing more needs to be said. Upon the derivation of [un [predictable]], the inner domain [predictable] is interpreted first: since there is no /nC/ cluster, nothing happens. At the outer domain, then, the string [un predictable] is computed, and nasal assimilation goes into effect. No look-back is toothless: the root was spelled out before and hence is untouchable – but the prefix was not. Its nasal thus happily undergoes assimilation, and the unwarranted *um-predictable is produced. In order to get things right, the affix needs to be spelled out before being merged. At the word-level domain, modification-inhibiting no lookback would then guarantee the non-assimilation of the nasal. 75

Though not by Kaye, who considers the alternations at hand non-phonological (see §276). The analysis below merely shows how Trisyllabic Shortening could be done in Kaye's system, were it declared phonological.

268 Chap 9: Kaye (1995), selective spell-out and freezing no look-back This is precisely the analysis that Kaye (1995:305) proposes: he identifies un-predictable as the analytic [[un] [predictable]], while im-possible (and of course all morpheme-internal clusters) are non-analytic: [in-possible] (and [lanp] for lamp for those who go along with the abstract analysis of an underlying dental). The nasal, then, assimilates to the following obstruent in [im-possible] and [lamp], while un- remains unaffected: it was already spelled out in its own domain and hence cannot be further modified when nasal assimilation applies to the outer domain [un predictable]. Kaye (1995:305) also discusses the candidate structure [un [predictable]] where un- does not sit in a domain of its own. He rejects this option on the grounds of a phonological argument that is specific to Standard Government Phonology – and which does not go through. In Government Phonology, lexical items always end in a nucleus: in case they are consonant-final on the surface, the final syllabic constituent is an empty nucleus (Kaye 1990a). Since un- is an independent lexical item, it ends in an empty nucleus, whose Empty Category Principle (ECP) needs to be satisfied. In case [unø-] is a domain of its own, the final empty nucleus is licensed by virtue of being domain-final (see §330). No look-back guarantees that this phonological property – the well-formedness of the final empty nucleus – remains unmodified on later cycles. If on the other hand un- does not constitute a domain, Kaye argues, [unø [predictable]] is phonologically ill-formed because the only way for empty nuclei that are not domain-final to be silenced is through Government from the following contentful nucleus. The intervening branching onset [pr], however, blocks Government, to the effect that the overall structure is ungrammatical. This argument is not conclusive: along the same lines im-probable, which identifies as [inø probable], would be ill-formed since the empty nucleus is unable to receive Government from the first root vowel over the [pr] cluster. Kaye (1995) thus chooses the correct compound structure [[un] [predictable]] for the wrong reason. Un- must indeed come with a domain of its own in order for nasal assimilation to be unable to target its nasal, which will be protected by no look-back. But there is no independent phonological reason.

Empirical coverage: Kaye's system and affix class-based phenomena 269 315

6.2.3. Anti-affix ordering items cannot be done: stress violates no lookback Finally, it needs to be mentioned that there is one sub-pattern of stressplacement (132c) that Kaye's system cannot account for. This is when a class 1 suffix is merged to a class 2 suffix: the result is regular stress shift as everywhere else when class 1 affixes are concatenated. Given that the class 2 affix, which is lower in the tree, has already triggered interpretation; however, what is expected is that stress does not move: it should be prevented from doing so by modification-inhibiting no look-back. The cases in question are exactly the ones that invalidate Siegel's (1974) affix ordering generalisation, and which are also known as bracketing paradoxes. The relevant pattern was introduced in §243, and Halle & Vergnaud's analysis was discussed in §246. An example is govern-mént2-al1: -ment identifies as a regular class 2 suffix by not triggering stress shift – its concatenation produces góvern-ment2. According to Kaye's system, this is because the suffixation of class 2 affixes triggers the spellout of the root, which acquires a stress pattern that no look-back prevents from being modified on the outer cycle of the relevant structure [[govern] ment]. Up to this point, everything is fine, and the derivation is exactly as in §313. When -al1 is added, then, the relevant domain structure is [[govern] ment2 al1]. The root was already spelled out through the action of -ment2 and therefore bears stress in góvern-ment – it should therefore be immune against the "wish" inherent to -al1 to shift stress. But in fact it is not: the merger of -al1 causes stress to move one vowel right in violation of no look-back. There is nothing more to be said, except perhaps that the process which produces the violation of no look-back is stress. Stress, however, is a well-known problem child, which appears to be particularly apt to violate no look-back devices. Recall from §192 that it was already stress which violated the generalisation regarding derived environments, i.e. Kiparsky's SCC-K, another no look-back mechanism. We will see in §554 that stress is also declared exceptional in regard of no look-back by Marvin (2002). Ultimately, what this is all about is to find a rationale for determining the class of processes that (may) violate no look-back. This was already attempted in the 80s when structure-building were opposed to structure changing processes: the former were supposed to be able to violate the SCC-K, while the latter were held to obey no look-back (§192).

270 Chap 9: Kaye (1995), selective spell-out and freezing no look-back Without this being very original (see Rubach 1984:222), one rationale that is suggested in §554 below is that suprasegmental processes such as stress and tone (may) violate no look-back. A summary of the quirky properties of stress is provided in §814.

316

6.3. Spell-out of terminals: the problem and a possible solution

317

6.3.1. The problem: independent spell-out of affixes prior to their being merged Underapplication of the rule-triggering pattern is no problem when the process modifies the root: given [[root] A] (or [A [root]] for that matter), the root is protected by modification-inhibiting no look-back on the outer cycle since it was already spelled out. In a compound structure [[A] [root]] (or [[root] [A]] for that matter), however, the fact that [A] is a domain of its own requires its independent spell-out, crucially without the root being spelled out at the same time (which is what happens on the outer cycle). A, though, is necessarily dominated by a node that also dominates the root, which means that what the spell-out mechanism must be able to do is to send A to interpretation before it is merged into the tree (§274). Otherwise there is no way for it to be spelled out in absence of the root: on regular assumptions only nodes can be phase heads. Kaye's system thus requires the independent spell-out of terminals if an affix, rather than a root, is concerned by a no look-back effect, that is, in case the underapplication of a process to an affix needs to be organised. This is precisely what must be done in nasal assimilation: the prefix-final nasal in un-predictable does not assimilate and hence must be protected from nasal assimilation by a domain of its own: [[un] [predictable]] (§314).

318

6.3.2. Morphological adjuncts, counter-cyclic merger and intermodular predictions Clearly, a solution to this problem can only be syntactic – phonology has nothing to say about spell-out. Taking a closer look at the syntactic literature shows that the independent spell-out of terminals has actually been

Empirical coverage: Kaye's system and affix class-based phenomena 271

proposed – of course for purely syntactic reasons that have nothing to do with our phonological motivation.76 The idea of counter-cyclic merger (or late adjunction) in syntax is that items may be merged into the syntactic structure without extending the root of the tree. A specific version of this approach holds that the status of a phrase as an adjunct (or subject) entails interpretation at PF prior to counter-cyclic merger into the core syntactic tree (e.g. Lebeaux 1988, Stepanov 2001). Adjuncts are therefore a separate phase in the terminology of modern phase theory. This indeed looks like a description of the behaviour of un-. Let us therefore explore the hypothesis that un- (but not in-) is a (morphological) adjunct. In the perspective of the literature mentioned, adjuncts enjoy a certain autonomy and may be merged into syntactic structure countercyclically, that is to a non-root node. An adjunct may thus be absent from a syntactic structure at a level of construction that has already grown higher than its own location: it may join in later. If the analysis of un- as an adjunct turns out to be plausible, and if morphological adjuncts are really spelled out before they are merged into the tree, the problem that is raised by Kaye's requirement that terminals must be able to be spelled out independently disappears. Or rather, it turns into an interesting cross-modular prediction, which has all for being wrong, but turns out to be true. The overall prediction is that all affixes which are subject to a phonological no look-back effect are in fact (morphological) adjuncts – this is something that may be controlled on the morpho-syntactic side. Newell & Scheer (2007) make an argument in favour of this perspective that builds on the benefits which are produced on the syntactic as much as on the phonological side. That is, the interpretation of un- as an adjunct explains so-called bracketing paradoxes where a mismatch between morphological and phonological structure seems to be produced by un-, but not by in-. A digest of the argumentation is provided below.

319

6.3.3. Benefits: bracketing paradoxes, category selection, double affixation The phonologically contrasting behaviour of un- and in- is paralleled on the morpho-syntactic side. Visible effects are so-called bracketing paradoxes, which were already discussed in §§243,440: the structure of un-happi-er 76

The discussion below is a digest of Newell & Scheer (2007).

272 Chap 9: Kaye (1995), selective spell-out and freezing no look-back should be [[un [happy]] er] since it means "more unhappy", rather than "not more happy". That is, un- has only scope over the root. Phonologically, however, the structure should be [un [[happy] er]] since the synthetic comparative -er selects maximally bisyllabic stems (big - bigger, happy - happier); adjectives with more syllables have an analytic comparative (beautiful - *beautifuller, more beautiful). The synthetic un-happi-er, then, is only possible if -er is concatenated before un- is merged. The same invisibility of un- (but not of in-) for comparative allomorphy selection is responsible for the fact that un- allows for unlikelier (likelier), while in- cannot build *impoliter (politer, more impolite). These phenomena fall out if un- is a counter-cyclically merged adjunct: it is absent when the comparative selects for maximally bisyllabic roots. By contrast, in- is present upon suffixation and thus blocks the derivation. Another known property of adjuncts is that they do not project features (e.g. Lebeaux 1988, Stepanov 2001). This is precisely the case of un-, which adjoins to various syntactic categories: verbs (unlock vs. *inlock), nouns used as adjectives (unBritney: "she is the unBritney" vs. *inBritney) and adjectives (unhappy, intolerable). By contrast, in-, being a non-adjunct, projects adjectival features and thus only attaches to adjectives (intolerable). Un- being an adjunct, on the other hand, it extends the base to which it attaches. Other morphological adjunction phenomena such as double affixation in cases like eater upper (vs. *eater up, *eat upper) are covered by the same analysis. Here the particle is argued to adjoin counter-cyclically. The derivation therefore includes an interruption of the one-to-one mapping between linear proximity and hierarchical structure of affixes, inducing respell-out of the agentive morpheme. Counter-cyclic merger in general and the analysis of un- in particular also raise a number of concerns. For one thing, counter-cyclic merger violates Chomsky's (1995a) Extension Condition. Also, it may be argued that in fact there are two distinct un-'s: un- appears to be negative when combined with an adjective, but reversative when combined with a verb. In defence of the analysis proposed, it may also be argued, though, that the meaning of un- is reversative even in combination with an adjectival host (Kennedy 2001); therefore all instances of un- represent one and the same item. Newell (2005a,b, 2008) expands on the syntactic issues related to the analysis of un- as an adjunct.

Empirical coverage: Kaye's system and affix class-based phenomena 273 320

6.3.4. Procedural first The appendix to this discussion is about a topic that is eluded elsewhere in the book: the balance of procedural and representational solutions for interface phenomena (see §748). Suppose there are empirically equivalent procedural and representational solutions for a given interface phenomenon – which one should be preferred? Newell & Scheer (2007) argue that a procedural analysis is always more interesting because it makes predictions on the morpho-syntactic side. By contrast, representational analyses cannot have any impact beyond phonology for principled reasons. In order to make the point, let us look at the purely representational analysis of the nasal assimilation facts that is proposed in Prosodic Phonology. Rubach & Booij (1984:11f) and Vogel (1991) derive the contrasting behaviour of un- and in- from contrasting prosodic structure: un- is mapped into a prosodic word (PW) of its own ([un]pw[predictable]pw), while in- is merged with the PW of the stem ([in-possible]pw). The assimilation rule, then, is restricted to apply within prosodic words. Unlike the procedural alternative, this analysis makes no claim at all regarding the morpho-syntactic properties of the affixes involved: it can run with any (syntactic) derivational history of in- and un-. Eventual morphosyntactic differences between the two affixes are thus unexpected and unexplained. It was shown earlier, however, that in- and un- also contrast on the morpho-syntactic side. It is quite obvious that the difference between the two prefixes is given lexically and then produces effects both on the morpho-syntactic and on the phonological side. A correct analysis must thus be able to do exactly that: derive all contrastive behaviour from a lexical property of the items at hand. Since phonology only interprets morpho-syntactic structure, this lexical property must produce a morpho-syntactic effect before the result can make a difference in phonological interpretation. That is, the key lies in morpho-syntax, not in phonology – hence analyses such as the one that is based on contrastive translation into prosodic words do not qualify: they create the (PW-) contrast during the translation of morpho-syntactic structure into phonological objects, rather than simply handing down an existing morpho-syntactic contrast. The thing is that this example where a representational analysis is unable to do justice to morpho-syntactic facts is not just an example. It is indicative of the principled inability of representational solutions to say anything about morpho-syntax at all. That is, the translation of morphosyntactic structure into phonological objects is arbitrary: a basic property of

274 Chap 9: Kaye (1995), selective spell-out and freezing no look-back translation is that anything can be translated into anything (more on this in Vol.2). By contrast, in a phase-based environment, analyses on one end of the pipe that relates morpho-syntax and LF/PF may also impact the other end. Therefore Newell & Scheer (2007) argue for procedural first: given a phonological effect that is impacted by morpho-syntactic information, and given competing procedural and representational solutions, always choose the former. The procedural analysis will make predictions on the morphosyntactic side and hence allow for an extra-phonological control of a phonology-based analysis. There is certainly reason to believe that the strongest argument for analyses in a given module is the (correct) prediction that it makes in another module. Phonological phenomena that are not conditioned by extra-phonological information cannot benefit from nonphonological refereeing. Phonological analyses that are in a position to be refereed outside of phonology should try to go for this arbitral award. This pattern – intermodular argumentation – is further developed on various occasions in the book; a summary is provided in §841.

321

6.4. The rule-triggering pattern (level 2 rules)

322

6.4.1. Underapplication at the inner cycle The rule-triggering pattern requires underapplication to strings that are the result of the concatenation of a class 1 affix. In the English examples from nasal cluster simplification, this means that the root-final cluster must be reduced in [[root] class 2] (sign-ing) as well as in roots that occur in isolation (i.e. [root], sign), but not in [root class 1] (si[g]nature). It is obvious that modification-inhibiting no look-back is perfectly unable to describe this kind of underapplication: inside-out interpretation requires that a process which affects more embedded domains (the root in sign-ing) also applies to higher interpretational units, i.e. the word-level in our case. Unless, of course, like in Halle & Vergnaud's system, the wordlevel is not systematically subjected to the same (cyclic) rules as more embedded interpretational units. We have seen in §280, though, that Kaye rejects this option: the φ-function systematically applies to word-sized chunks. Hence there is no way to have the cluster simplification rule apply to the root of sign-ing, but not to the root of si[g]nature at the outer domain.

Empirical coverage: Kaye's system and affix class-based phenomena 275

Kaye's approach therefore seems to be structurally unable to account for the rule-triggering pattern. Unless, however, there is a phonologyinternal reason for the underapplication to class 1 strings. As a matter of fact, English nasal cluster simplification does feature such a reason: in order for the clusters /mn/ and /gn/ to reduce to [m] and [n], respectively, they not only need to be followed by a class 2 affix – they also need to be final. Indeed, morpheme-internal clusters such as in i[gn]ore and a[mn]esia remain unreduced (recall that this has nothing to do with derived environments since the cluster happily simplifies in non-derived environments as well: sign). Therefore the question is what the clusters need to be final of in order to reduce. Morpheme-final is not a solution: the morpheme-final cluster of class 1 strings (si[gn]-ature) does not reduce. The only thing that "final" can mean is thus "string-final upon phonological computation". That is, the cluster reduces iff it is the last item in the string that is submitted to PF. In other words, if it is domain-final (or phase-final, see §329). Applying this rule to Kaye's domain structure produces the correct result: [sign] and [[sign] ing2] reduce, but [sign ature1] and [ignore] do not because their /gn/ is never string-final upon PFcomputation.

323

6.4.2. Kaye's solution is like Mohanan's, but respects modularity Note that this solution follows the same spirit as Mohanan's (1986) bracketsensitive rules (§168): Mohanan's rule is sensitive to "]" where Kaye's process applies only to string-final elements. In both cases, the structure that is submitted to phonological computation contrasts in the same way: while sign-ature is one single domain, sign-ing has ("fresh" according to Mohanan) internal structure. The difference is twofold, though: 1) with Kaye, the contrasting domain structure is the result of a principled mechanism, i.e. selective spell-out (while Mohanan has no equivalent); 2) Kaye does not use any untranslated morpho-syntactic objects (Mohanan's brackets). Kaye's solution is thus unvitiated of the conceptual shortcoming that was pointed out in §170: interactionism having successfully done away with SPE-style brackets that violate modularity, Mohanan reintroduces them through the back door. The information that the process really makes reference to is not morpheme-finality or any diacritic, but simply the fact of being string-final when phonological computation applies.

276 Chap 9: Kaye (1995), selective spell-out and freezing no look-back The question why string-finality should cause the reduction of the clusters at hand is discussed in the following section.

324

6.4.3. Why string-final /gN/ and /mn/ do not survive computation The question is thus why unlike other string-final clusters of the same sonority profile, /gN/ and /mn/ do not survive phonological computation. As Mohanan (1986:21ff) points out, the clusters if prism [-zmÿ ], pickle [-klÿ], listen [-sn̩] do, with their sonorant becoming syllabic. All analyses must thus somehow grant a particular status to /gN/ and /mn/, which is also to be related to the "weak" behaviour of the voiced stops in /ng,mb/ (bomb-ing, see note 43). Whatever the reason for the weakness of the clusters that do not survive phonological computation in string-final position, it is an observational fact that they survive in presence of a vowel to their right (upon computation). It is thus reasonable to hold the absence of such a vowel responsible for their implosion. In Government Phonology, the relevant generalisation is expressed in terms of the empty nucleus that follows all consonant-final words: the domain-final empty nucleus in [signø] and [[signø]ing] is unable to licence the preceding cluster; this task requires a full nucleus. In [ignorance] and [sign ature], the nucleus following the cluster is contentful (to the extent that the -a of -ature is floating and lands in the final empty nucleus of the root) – therefore the cluster survives. On this analysis, even the domain-final condition that Kaye (1995) systematically mentions is superfluous: clusters simply reduce iff followed by an empty nucleus upon computation. If the mention of the domain-final condition could be suspected to draw on a (Mohanan- or SPE-type) diacritic, reduction before empty nuclei is entirely unambiguous: the interactionist architecture assures that the phonological process at hand makes only reference to truly phonological objects.

325

6.4.4. Prediction: processes without additional condition cannot instantiate the rule-triggering pattern We have seen in §322 that Kaye's system alone is unable to account for the rule-triggering pattern. It predicts that there will always be an additional phonological condition that has nothing to do with affix classes and re-

Empirical coverage: Kaye's system and affix class-based phenomena 277

stricts the application of relevant processes (domain finality in the English case). This prediction is falsified by a process that follows the ruletriggering pattern and whose phonological condition is the mere existence of an underlying sequence, without any additional proviso. For example, nasal assimilation (un-predictable vs. im-possible) could not follow the rule-triggering pattern since on the phonological side the bare existence of an /nC/ string is enough to trigger assimilation.

326

6.4.5. Kaye's analysis predicts the word-final/class 2 disjunction Kaye's system makes yet a different prediction: if there is any natural kinship between contexts that are defined by affix classes and the end of the word, only class 2 strings can line up with the word-final situation. This is because the root in [[root] class 2] shares with the simplex root [root] the fact of being domain-final (as opposed to [root class 1] where the root is not domain-final). This is the consequence of Kaye's take that class 2, rather than class 1 affixes, trigger interpretation. Recall from §281 that Halle & Vergnaud subscribe to the opposite distribution where class 1 affixes do, but class 2 affixes do not trigger spell-out. Kaye's result also supposes that what is actually spelled out is not the node that dominates the affix, but the affix' sister (§282): otherwise the root would not be a domain in [[root] class 2]. SPE has observed that English roots in word-final position and before class 2 affixes systematically behave in the same way (see §94): "the inflectional affixes which are neutral with respect to stress also characteristically affect final clusters in the same way as word boundary does" (Chomsky & Halle 1968:85). What was an unexplained coincidence then now follows from selective spell-out and the two additional choices mentioned. Note that these were not made in order to get Chomsky & Halle's observation right: the English system as a whole is only consistent with this configuration.

327

6.5. Conclusion In conclusion, Kaye's system affords a complete empirical coverage of affix class-based phenomena. A special provision, however, needs to be made regarding the possibility for morpho-syntactic terminals to be spelled out

278 Chap 9: Kaye (1995), selective spell-out and freezing no look-back prior to being merged (§274). This is necessary in Kaye's system, but nothing that regular spell-out provides for: only nodes can be phase heads on standard assumptions. It was shown in §316, however, that the independent spell-out of terminals was proposed independently in syntactic analysis under the header of countercyclic merger (or late adjunction). Minimally, what Kaye's system allows, or rather: what it provokes, is a dialogue between phonologists and syntacticians: his system, which is exclusively based on phonological evidence, can only work if a specific operation exists in morpho-syntax. I take this kind of intermodular argumentation to be a valuable touchstone for analysis on both ends of the pipe that was established by phase theory (more on this in §841).

328

7. Phonological consequences of domains and their erosion

329

7.1. Empty nuclei that are string-final upon computation

330

7.1.1. Final empty nuclei in Government Phonology The following pages discuss some consequences of Kaye's domain structure. One thing that domains afford is a more precise definition of the peculiar status that finality confers: empty nuclei are allowed to remain empty when they occur in domain-, rather than in word-final position. In other words, the latter is a special case of the former. In Government Phonology, word-final consonants are onsets of empty nuclei in all languages (Kaye 1990a). The ban on final codas also extends to lexically stored items: consonant-final morphemes end in an empty nucleus as well. Empty nuclei, however, are not for free: they must have a reason to be empty; the conditions under which empty nuclei are tolerated are defined by the Empty Category Principle (ECP, see Kaye et al. 1990:219ff, Vol.1:§60). Government from a following full nucleus is typically responsible for the emptiness of nuclei. Word-final empty nuclei, however, cannot be governed since there is no governor available to their right. Rather, it is their final location that allows them to remain empty. This observational statement is the way in which Government Phonology encodes the notoriously deviant behaviour of the right edge of the word (e.g. Broselow 2003, see Vol.1:§§266, 524 for further discussion). Final empty nuclei, then, are governed by virtue of being final (more on this parameter in Vol.2, see also Scheer & Ziková forth).

Phonological consequences of domains and their erosion 279

Finally, note that final empty nuclei are not tolerated in all languages: there are languages where all words must end in a vowel. The tolerance of final empty nuclei is thus the matter of a parametric choice that determines whether or not languages can have consonant-final words (Kaye 1990a).

331

7.1.2. From word-final to domain-final empty nuclei One feature that makes the right edge of the word peculiar is the occurrence of massive consonant clusters which are not found elsewhere, (i.e. wordinternally or word-initially). Germanic languages offer typical illustration. In English for example, the word sixths affords the monster cluster [-ksTs]#, and German is perfectly happy with [-ntkst]# in a word like röntgst "you do x-ray". What all final monster clusters have in common, however, is the fact of being heteromorphemic. Mono-morphemic wordfinal sequences are severely restricted and observe ordinary sonority sequencing.77 There is thus a direct relationship between morphological and phonological complexity: monster clusters can exist only because they are heteromorphemic. It is therefore safe to conclude that the morphological divisions in six-th-s are phonologically relevant. In Kaye's system, phonological visibility of morphological information can only be due to analytic morphology, that is to domain structure. Six-th-s is therefore made of three domains: [[[six]th]s]. Since all consonant-final morphemes end in an empty nucleus, though, the complete structure at hand is [[[sixø]thø]sø] (Kaye 1995:304). Three empty nuclei in a row would of course be ill-formed in regular circumstances: one is word-final and may remain empty for that reason. The other two, however, beg the question. On the other hand, all of them share the fact of being domain-final, and we know that this domain boundary is responsible for the well-formedness of the cluster. The conclusion, then, is that being word-final is irrelevant: what really counts for empty nuclei is domain-finality. That is, domain-final empty nuclei may remain unpronounced.

77

The typical Germanic licence for an extra dental set aside. Goldsmith (1990:140ff) and Harris (1994:66ff) provide detailed discussion of the English situation, Hall (1992:110) offers exhaustive facts for German (see also Vol.1:§350).

280 Chap 9: Kaye (1995), selective spell-out and freezing no look-back 332

7.1.3. The well-formedness of final empty nuclei is carried over to outer domains As a result, word-final monster clusters lose their exotic flavour: sixths [[[sixø]thø]sø] is not any more exceptional than, say, sit, which identifies as [sitø]. Both words possess domain-final empty nuclei whose phonetic absence is due to their domain finality. The only difference is their number: three in the former, one in the latter case. This, however, is irrelevant: a sequence of three, five, or sixteen empty nuclei in a row is well-formed as long as all of them are domain-final. This means that just like other phonological properties, the autonomy of domain-final empty nuclei is carried over to later cycles by virtue of modification-inhibiting no look-back. The final empty nucleus of the inner domain in [[[sixø]thø]sø] is sealed when it experiences interpretation and then "forgotten" in subsequent computation on outer domains: the string [sixøthøsø] contains three empty nuclei in a row, two of which are "invisible": they have received their phonological properties upon the interpretation that they were subject to at earlier cycles. In conclusion, then, there is no upper limit for the size of word-final clusters as far as phonology is concerned: any number of consonants in a row is grammatical and may happily be executed. Only morphology marshals the size of clusters: it will not keep adding analytic suffixes indefinitely.

333

7.1.4. Computation is right-to-left: string-final nuclei are phase-initial Being domain-final means being the rightmost object of a cycle (or a phase in modern terms). That is, domain-final (empty) nuclei are the rightmost object of the string that is submitted to phonological computation. Its special status is an observational fact, but of course we would like to know why the rightmost, rather than, say, the leftmost (empty) nucleus of an interpretational unit is special. A relevant consideration here is certainly the fact that in CVCV phonological computation is done from right to left: all lateral relations are head-final (see Vol.1:§218). For two items that are related by a lateral relation, this logically supposes to know the result of the computation of the head, i.e. .the rightmost item, before the target, further to the left, can be assessed: its value depends on the value of the head. The parsing of strings from right to left has also an obvious empirical basis: the overwhelming

Phonological consequences of domains and their erosion 281

majority of phonological processes is right-to-left, i.e. the patient is followed by the trigger (e.g. palatalisation). In this perspective, the fact that the phonological value of the phaseinitial object (i.e. the first item to be computed) is determined by a parametric choice that decides whether or not it may be empty is not really surprising: in order to parse the string, phonology needs to know about the first item since all the rest depends on its status. This issue is further discussed in Vol.2 and Scheer & Ziková (forth).

334

7.2. Consequences of domain structure for stress and vowel reduction

335

7.2.1. Dialectal and idiolectal variation due to variable domain structure Recall from §271 that out of the three logically possible analytic structures, Kaye (1995:305f) believes that [X [Y]] does occur. The present section discusses two contrasts that involve the two remaining analytic configurations: the compound structure [[X] [Y]] is compared to [[X] Y] on the one hand, and to the non-analytic configuration [X Y] on the other. Kaye (1995:313) considers English compounds whose second member is -metre. Table (133) below shows some cases in point. Primary stress is indicated by an acute, secondary stress by a grave accent. Vowel reduction, then, applies to vowels with no degree of stress. (133) English compounds in -metre a. no variation across British and American varieties b. variation between British and American varieties c. idiolectal variation

[[X] [Y]] míllimètre

[X Y] thermómetre

áltimètre (British) kílomètre

àltímetre (American) kilómetre

Since vowel reduction (to either schwa or nothing) is entirely predictable from stress, it would certainly be ill-advised to ascribe the variation under (133b,c) to contrasting lexical recordings of the vowel quality. Rather, the morphological complexity that was a property of the words at some stage of their evolution may or may not have survived into the present-day lexicon. The stress-assigning mechanism, then, was stable over time. In the lefthand column of (133), the original compound structure is still in place, which ensures the typical compound stress pattern: primary

282 Chap 9: Kaye (1995), selective spell-out and freezing no look-back stress falls on the first member, while secondary stress (and accordingly no vowel reduction) is observed on the second member. In the righthand column, however, primary stress is located further to the right, and there may or may not be a sizable secondary stress. The distribution of the contrast over grammatical, dialectal or idiolectal parameters appears to be random: sometimes it witnesses a real grammatical difference to which all English natives subscribe (or, at least, both British and American speakers): under (133a) míllimètre is necessarily stressed on the initial vowel, while thermómetre must not. On the other hand, áltimètre under (133b) identifies British pronunciation, whereas àltímetre is characteristic for Americans. Finally, whether an English speaker pronounces kílomètre or kilómetre as under (133c) is a matter of a personal choice, and a given speaker may probably produce variable results from utterance to utterance.

336

7.2.2. Following SPE: "old" stress protects against vowel reduction That secondary stress protects against vowel reduction (even if only present on an earlier cycle, viz. the famous contrast between compensation and condensation) was proposed in SPE (see §§96f). Kaye's treatment of stressdetermined vowel reduction is a close match of the SPE analysis. Following SPE (see §96), thus, Kaye (1995:313) holds that both parts of the word receive stress in the lefthand column of (133) because they constitute a domain of their own. Therefore phonology applies to them in isolation, which, among other things, makes them receive stress. By contrast, the words in the righthand column of (133) are parsed as one single chunk by the φ-function: main stress falls on the penultimate syllable. In all cases, unstressed vowels are reduced when the outer domain is computed, that is at the word level. The difference is that the original morphological complexity of the items in the righthand column has become invisible to the phonology (the internal structure of the initial compound [[X] [Y]] was lost), while it is still alive in the lefthand column. Items of the former kind are thus recorded as one single lexical entry with the nonanalytic structure [X Y]. They are therefore not any different from monomorphemic objects. This kind of diachronic domain erosion is further discussed in §339 below.

Phonological consequences of domains and their erosion 283 337

7.2.3. How to detect the difference between [[X] [Y]] and [[X] Y] Let us now turn to the contrast between the compound structure [[X] [Y]] and [[X] Y]. The words in the righthand column of table (133) have been interpreted as non-analytic items, i.e. [X Y]. But is there any reason why they could not be instances of [[X] Y]? For the particular cases at hand, the answer is no as far as I can see: [[kilo]metre] would produce the same result kilómetre. This is because stress would be assigned to [kilo], but not to metre since the latter is not a domain of its own and therefore fails to be considered by the stress-distributing φ-function. Vowel reduction then follows as before. Other cases, however, allow for distinguishing [[X] [Y]] and [[X] Y]. A word like póstman [pɔ́wstm´n] conveys two critical pieces of information that may be used in perception: -man is unstressed and must therefore have been subject to vowel reduction, and [stm] is not a possible monomorphemic sequence in English. The former property rules out a compound analysis: were -man a domain of its own, it would bear secondary stress (as is the case, at least in Southern British English, for súpermàn, i.e. [[super] [man]]). On the other hand, the presence of [stm] disqualifies a solution with no internal structure at all: [stm] can only be [[…stø]m…]. In sum, then, phonological properties of the input (parsing cues, on which more shortly in §340) necessarily identify póstman as [[post] man].

338

7.2.4. Chunk-specific phonologies: like SPE, Kaye provides for a specific word-level phonology Given his treatment of stress-defined vowel reduction that follows SPE's line of attack, Kaye also validates the existence of a specific word-level phonology. Recall from §104 that SPE provides for a specific word-level phonology, i.e. a set of rules that applies only to word-sized chunks, not to smaller or bigger pieces. Word-level phonology appears as post-cyclic lexical rules in Lexical Phonology (or at least some versions thereof, see §194), and as non-cyclic rules in Halle & Vergnaud (1987a) (see §233). Kaye (1995) explicitly rejects any brand of multiple computational systems in phonology: he insists on the fact that the φ-function is one, i.e. accommodates all phonological instructions of a language without exception. It was mentioned in §267 that his practice, though, seems to provide for a specific word-level phonology.

284 Chap 9: Kaye (1995), selective spell-out and freezing no look-back What was meant is Kaye's treatment of stress-driven vowel reduction in English. The argument is exactly the same as in SPE (§105): vowel reduction cannot apply to smaller chunks because it is calculated with respect to the main stress of the whole word, and it cannot apply to larger pieces because word stress is entirely insensitive to whatever happens after the word level.78 The fact that Kaye subscribes to a specific word-level phonology is an element in the discussion of multiple computational systems in phonology in general, and of the contrast between morpheme-specific and chunkspecific phonologies in particular (§§234,283). While the former are specific only to Lexical Phonology (level 1 vs. level 2), the latter is practised by all interface theories: SPE, Lexical Phonology, Halle & Vergnaud and Kaye provide for a specific word-level phonology. On the other hand, Praguian segregation, another chunk-specific phonology, is only practised by Lexical Phonology; it is rejected by all other theories. 339

7.3. Diachronic erosion of domain structure When looking at English words that represent the combination of two independent words (nouns) which has taken place at some point in the history of the language, the following generalisation emerges: the older the compound, the more internal structure is lost. That is, compounds start out their life as the compound structure [[X] [Y]]. This is when the compound is not yet a separate lexical entry: it is the result of online concatenation of X and Y, which have been extracted by two independent lexical accesses. Diachronic erosion then eliminates either of the inner domains, or both. The regularity of this process actually affords to make predictions on the relative age of words just by looking at their phonological behaviour. Consider words such as blackboard, blueberry, postman, shepherd and cupboard. Two criteria identify bláckbòard [blǽkbç$çd] as a true compound: both of its members are stressed, and the sequence [kb] is nothing that English allows for morpheme-internally (more on clusters that signal morpheme-boundaries in §340). Hence the original domain structure [[black] [board]] is still alive today, and we can bet that this word is not very old. Its lexicalisation is witnessed, though, by the unpredictable meaning: a blackboard is not just any board that is black (and this is also true for the other compounds discussed). 78

Stress clash (thirtéen vs. thírteen men) is a different issue, see §240, note 64.

Phonological consequences of domains and their erosion 285

Blueberry is pronounced [blúb´r´] in Southern British varieties, but comes out as [blúbE$ri] in American dialects. The latter pronunciation is the result of the original structure: Americans have maintained [[blue] [berry]], while all internal structure was erased in British English, where [blue berry] is executed. Or rather, we cannot be sure whether British speakers interpret [blue berry] or [[blue] berry] since no trans-morphemic cluster provides evidence. By contrast, we have already seen that a word like postman identifies as [[post] man] (rather than as [post man]) due to its medial cluster (§337). There was thus a diachronic change from [[post] [man]] to [[post] man]. Finally, words like shepherd [Sɛ́p´d] and cupboard [k√́b´d] (Southern British) signal a complete loss of internal structure: they do not bear any secondary stress and accordingly show vowel reduction. Crucially, though, they have eliminated the original trans-morphemic clusters [ph] and [pb], which could never be mono-morphemic in English. In other words, the erasure of the internal structure has provoked cluster simplification. The words at hand have thus evolved from [[X] [Y]] to [X Y] (possibly via [[X] Y], which corresponds to the pronunciations [shiiph´d] and [k√́pb´d], respectively). All words discussed share a semantic fact that is typical for compounds: their meaning is non-compositional. Interestingly, the degree of estrangement from the original meaning may be more or less important. The sense of blackboard for example is fairly close to "a board that is black": the compound almost matches the meaning of the ingredients. The semantic drift is no doubt greater for blueberry and postman, since the former is not just any berry that is blue, and the latter is not just any man that has to do with the post. Finally, the sense of cupboard is fairly opaque if one were to reconstruct it from "a board that carries cups". The meaning of shepherd is not as opaque as the one of cupboard, but it is certainly true that a shepherd is not just a dog that herds sheep. What this means is that our phonological measure of the age of the words at hand could be worse: it is a fairly good match of their semantic distance.

286 Chap 9: Kaye (1995), selective spell-out and freezing no look-back 340

8. Parsing cues

341

8.1. Theory-independent parsing cues: knowledge of morpheme structure We are now in a position to address parsing: how can phonology help identifying morphemes in the unstructured linear signal? Recall from §263 that according to Kaye, the enhancement of parsing is one of the two reasons why phonology exists (the other being lexical access, to be discussed in §346 below). Parsing cues follow the perception-oriented logic of Trubetzkoy's Grenzsignale (§55), which counts a number of modern incarnations (see §264). When the phonological system of the perceiver runs over the linear string, it is able to tell whether or not a given sequence is well-formed according to its standards. That is, a sequence is ill-formed if it does not conform to what the application of the φ-function would have produced. In other words, speakers know what a morpheme may look like, and what it cannot look like. Classically, this knowledge is covered by the notion of Morpheme Structure Rules. Anything that constitutes a phonological anomaly or is not compatible with morpheme structure, then, alarms the parsing system: it signals the presence of a morpheme boundary. This of course supposes that a morpheme cannot accommodate just any sequence of sounds: in order for the system to have any turnout at all, some clusters that occur in the language must be absent from monomorphemic strings. English (and arguably all other languages) follow this pattern. Let us look at a few examples. The cluster [mz] is a possible sequence in English (it seem-s, dream-s etc.), but does not occur within a morpheme. The same is true for the cluster [stm] that is found in postman (see §334). In both cases, a monomorphemic parse of the string is not compatible with the phonological knowledge of the listener: there is no morpheme in the language that accommodates the clusters at hand. Hence the parses [[seem] s] and [[post] man] are enforced. Another example is the sequence [ksTs] in the word sixths that was discussed in §331. Since neither [ksT] nor [sTs] nor [sT] nor [Ts] can be tautomorphemic, the correct structure must be [[[ksø]Tø]sø]. Darkness and enlargement could not be the result of a computation over a single domain either since [rkn] and [rdÉZm] are not good tautomorphemic clusters. This is

Parsing cues 287

how the suffixes -ness and -ment are identified by pure phonology, i.e. without look-up in the lexicon or any morphology intervening.

342

8.2. Theory-dependent parsing cues (in English)

343

8.2.1. Empty nuclei detected by phonology What a listener interprets as an alarm signal thus depends on his phonological knowledge: something that cannot be produced by the φ-function cannot be mono-morphemic. Finding out about the precise properties of the φfunction, however, is what all phonological theories attempt to do. Hence what counts as a parsing cue to a certain extent is theory-dependent. Below the theory-specific parsing flags that are flied by Standard Government Phonology are discussed. In Standard Government Phonology, a surface cluster [CC] may or may not be separated by an empty nucleus: it either represents a coda-onset sequence, a branching onset or two independent onsets, which are then separated by an empty nucleus. For example, [ŋk] in Czech may be a true or a spurious (Kaye's terminology) cluster, whose identity is only betrayed by its behaviour. The cluster appears unaltered when the word banka [baŋka] "bank NOMsg" appears in GENpl where the case marker is zero: bank [baŋk]. By contrast, the GENpl of linka [liŋka] "line" is linek [linɛk]: a vowel-zero alternation splits up the cluster, and the nasal ceases to be homorganic. Hence banka identifies as /baŋka/, while linka is /linøka/ underlyingly. The detection of empty nuclei is part of the morpheme recognition task: in order to identify a morpheme, the listener must run it against the lexicon and therefore needs to correctly identify its phonological structure. Since part of this structure (such as certain empty nuclei) cannot be predicted from the surface, it needs to be identified by some other means. Below three parsing cues are discussed that reveal the existence of empty nuclei in the linear string. The following section then shows how it is decided whether or not these "hidden" empty nuclei are morphologically relevant (i.e. whether they are domain-final or not). The first parsing cue is based on the take of Government Phonology on superheavy rhymes: they do not exist (in English or elsewhere). Kaye (1990a) shows that the consonant of VVC sequences is always an onset, even when it is word-final or followed by another consonant. Hence speakers know that the only possible parse of VVCC# sequences contains two

288 Chap 9: Kaye (1995), selective spell-out and freezing no look-back empty nuclei, /VVCøCø#/. 79 While the word-final empty nucleus is enforced by Coda Licensing anyway (see §330), its cluster-internal peer is the only possible parse given the ban on superheavy rhymes. This means, then, that a universal property of syllable structure betrays the existence of empty nuclei in words such as seem-ed [siimd], peeped [piipt], seep-ed [siipt] and fak-ed [fejkt], whose only possible parse is /siimødø/, /piipøtø/, /siipøtø/ and /fejkøtø/, respectively. In each case, the long vowel allows only for an onset interpretation of the following consonant, which in turn can only be followed by another onset consonant. Another claim of Standard Government Phonology is that NC clusters which are adjacent at the skeletal level are homorganic. On this take, the homorganicity of the nasal is an automatic consequence of the Interconstituent Government that it experiences as the patient of a coda-onset sequence. Since adjacency at the skeletal level automatically induces homorganicity, non-hormorganic NC clusters cannot be coda-onset sequences.80 A listener who comes across words such as dream-s [driimz] (plural), seem-ed [siimd], wing-ed [wINd] ("she winged her way home"), un-clip [√nklIp] or un-pleasant [√nplEz´nt] thus knows that the NC cluster involved is separated by an empty nucleus: /driimøzø/, /siimødø/, /wINødø/, /unøclip/, /unøpleasant/. The third phonological property that allows for the detection of empty nuclei concerns sonorant clusters. In English, groups where the first member is either r or l are fine (-rl- darling, -rn- earn, -rm- harm, -lmcalm, -ln- vulnerable, -lr- walrus81), while sonorant clusters beginning with n do not occur (*-nl-, *-nr-). According to Kaye et al. (1990:218), this restriction is due to the relative complexity of the segments involved: r is the least complex sonorant and thus a perfect governee, while n is the most complex object and hence the best governor. Since sequences of sonorants 79 80

81

Only in case the final cluster does not qualify as a branching onset of course. Otherwise a /VV.TRø/ parse is possible (though not obligatory). While spurious NC clusters (/NøC/) may or may not be homorganic: in about the Northern half of Poland, Gienku "proper name, VOCsg" is pronounced [gjɛnku] (NOMsg Gienek [gjɛnɛk]), while Southern Polish speakers say [gjɛŋku]. The same goes for the contrast between Czech banka and linka which was discussed earlier in this section. In conclusion, then, non-homorganicity offers a parsing cue, but homorganicity does not. The reality of -ln- and -lr- may be doubted on the basis of their rarity: only three instances of each cluster seem to exist (kiln, vulnerable, ulna, cavalry, chivalry, walrus).

Parsing cues 289

can only instantiate coda-onset clusters (where the onset governs the coda), the heavy n will never be able to be governed by the light l,r. Knowing about this, phonology flies a flag when listeners come across words such as un-real or un-lawful. Since the sonorant clusters at hand do not qualify for either a branching onset or a coda-onset sequence, the only way to parse them is as independent onsets with an intervening empty nucleus: /unølawful/ and /unøreal/ are the output of the phonological filter. The three theory-dependent parsing cues discussed certainly do not exhaust the flags that Standard Government Phonology flies upon parsing. And of course, other theories predict other parsing cues in different situations. This is all to the good, for it makes sure that different theories make different predictions. And it is to be expected that all theories make predictions regarding well- and ill-formed mono-morphemic clusters.

344

8.2.2. Morphological interpretation of empty nuclei Let us now turn to the morphological interpretation of the "hidden" empty nuclei that have been detected on phonological grounds. In Government Phonology (Kaye et al. 1990:219), the Empty Category Principle regulates the interpretation of empty nuclei. That is, nuclei may be empty iff they are either governed or domain-final. 82 Speakers who are equipped with this knowledge know that items with two empty nuclei in a row are ill-formed. That is, [peepødø] and [dreamøzø] cannot be the correct structure for the input peep-ed and dream-s: the internal empty nucleus cannot be governed because the following nucleus is empty as well. Were the φ-function to interpret [peepødø], the first empty nucleus, escaping Government, would be vocalised. This is indeed what happens when wicked [wickødø] and naked [nakødø] are executed: they appear as [wIk´d] and [nejk´d], i.e. with a schwa (*[wIkt] and *[nEjkt] are not possible pronunciations). Since in peep-ed and dream-s, however, no schwa appears, the only way to make sense of the internal empty nucleus is to conclude that it is

82

Other circumstances that allow for empty nuclei have also been proposed in Standard Government Phonology (Magic Licensing, Interonset Government). They are irrelevant here (see Vol.1:§§96,106 for discussion).

290 Chap 9: Kaye (1995), selective spell-out and freezing no look-back domain-final. The only parse that is consistent with the φ-function and the ECP is thus [[peepø]dø] and [[dreamø]sø], respectively.83 This, in turn, means that a morpheme boundary must separate the word-final cluster: domain boundaries always coincide with morpheme boundaries. In sum, then, a piece of morphological structure is detected on purely phonological grounds. 345

8.3. When phonology is useless While phonology may help unpacking the acoustic signal, it is not something that morpheme recognition can always rely on. Morphologically complex strings may as well pass the phonological filter without any alarm bell being rung. How many parsing cues are actually produced depends on a set of heterogeneous and language-specific factors such as the type of morphology, phonological inventories and morpheme structure (see Trubetzkoy's take on this in §57). The English past tense morpheme -ed for example will produce three alarm signals in seem-d: the existence of a long vowel before a word-final cluster, the non-homorganicity of [md] and the fact that [md] is not a possible tautomorphemic sequence. Elsewhere, the cluster in seep-ed [siipt] gets away without arousing suspicion ([pt] is a possible tautomorphemic sequence: apt, adopt), but the morpheme boundary will still be detected due to the preceding long vowel. Still in other cases, however, -ed will remain undetected: since [kt] and [pt] are possible morpheme-internal clusters (act, adopt), lack-ed [lækt], pack-ed [pækt] and stepp-ed [stEpt] are undistinguishable from mono-morphemic items. This is also true for words such as back-s (either the plural of back or the third person singular of to back): [ks] is a good tautomorphemic sequence (wax, lax). Phonology is useless here, and morpheme recognition must solely rely on regular non-phonological mechanisms. 83

Note that the presence of an empty nucleus in un-real [unøreal] and un-lawful [unølawful] (see §3430) does not allow us to conclude that it is also domainfinal: its absence from the surface could also be due to its being governed since there is a valid governor to its right. Recall from §314 that Kaye's (1995:305) argument in favour of the empty nucleus being also domain-final is flawed. The only parsing cue that is able to detect that un- is a domain of its own is the nonassimilation of its nasal.

Lexical access and the organisation of the lexicon 291 346

9. Lexical access and the organisation of the lexicon

347

9.1. Lexical entries are grouped according to phonological structure Let us now look at the second reason for the existence of phonology that Kaye has identified (in his teaching, but not yet in print as far as I can see): the fact that it provides an addressing system for the lexicon. What Kaye means is that phonology defines the structure of the lexicon and thereby organises lexical access. That is, phonology helps speakers finding lexical entries more efficiently and more rapidly. Let us first consider the former aspect. The organisation of and access to ordinary (man-made) dictionaries depends on the spelling system used. A consequence of the Latin spelling for a dictionary of English for example is that the section "k" reduces to a small number of pages, while words beginning with "c" or "b" represent a substantially bigger chunk. Now suppose that the same dictionary is written in Arabic characters. While the section "b" will have the same size as before, it will be located elsewhere in the volume. The Latin sections "c" and "k", however, will be merged under one single heading (since there is only one letter for the sound [k]). Given that the spelling system matters, then, Kaye contends that natural language lexica use the phonological code for storing purposes. This means that phonologically similar words are stored next to each other: there will not be any difference between "k" and "c" as in English dictionaries. Rather, everything that begins with a [k] will be stored under this heading, like in the Arabic version of the English dictionary. The rule that phonological similarity determines the organisation of the dictionary also extends to syllable structure: the internal structure of the heading [k] subdivides into different storage spaces according to whether [k] belongs to a simplex onset, to a branching onset or, in languages that allow for word-initial #kt, #pt (such as Classical Greek), to sequences of two independent onsets (which are separated by an empty nucleus). Finally, s+C clusters constitute a fourth batch of phonologically homogeneous words: according to Kaye (1992b), the s belongs to a coda and is preceded, word-initially, by an empty onset and an empty nucleus.

292 Chap 9: Kaye (1995), selective spell-out and freezing no look-back 348

9.2. Possible lexical entries are defined by phonology

349

9.2.1. How speakers decide that blick, but not lbick, is a possible word Speakers know what a possible word is in their language, and what is not. This knowledge is independent of any syntactic or semantic property since judgements may be produced upon simple exposition to nonce words. Therefore phonology alone must be the basis on which speakers decide whether something is a possible word or not. Since SPE, the pair blick - lbick is often quoted for the sake of illustration. Both items are not actual English words. However, the former could be one because its initial cluster is well formed (sonority increases), while the latter could not: its initial cluster violates sonority sequencing. That sonority sequencing is a property of English grammar is shown by the different attitude that natives adopt in regard of the two candidate words: lbick is not a possible word for any speaker, while blick could enter the language at any time, did it acquire a meaning.84 According to Kaye, though, the speaker's decision regarding lbick is not achieved by performing a phonological assessment of the candidate. Rather, the speaker simply looks up lbick in the lexicon, and finds no match. By contrast, blick is an entry of the lexicon of every English native, if one that lacks non-phonological properties.

350

9.2.2. The full addressing space is created, including "empty" slots Kaye thus contends that the mental lexicon contains not only those items which are indeed used by the language (that is, which have syntactic and semantic properties associated to their phonological address). It also provides for "empty" entries which are only specified for their phonological properties. Empty entries may become real words at any time – they just need to be filled in with syntactic and semantic features. On this view, then, universal principles and language-specific parameters allow for the precise calculation of the overall addressing space as soon as mature phonology is available. That is, every possible address is 84

See Scheer (2007) and Vol.2 for illustration in the Slavic family where wordinitial sonorant-obstruent clusters exist, but are far from exhausting the logical possibilities in any individual Slavic language. The gaps are argued to be accidental (hence that can be filled at any time), rather than systematic (i.e. grammar-driven).

Lexical access and the organisation of the lexicon 293

not only imagined, it is actually created: all entries, filled and empty alike, are fully specified for non-predictable phonological properties. 85 Empty addresses are ordered according to the same principle of phonological similarity as their filled cousins, and may be retrieved by lexical access. This means that speakers are not actually "learning" new words. Rather, they associate syntactic and semantic properties to a lexical address that already exists. Also, the lexicon as such never increases: the number of entries is stable during the entire life of a speaker, and identical among all natives of the same language. In this perspective, the lexicon is thus a cognitive implementation of the notion "possible word (morpheme)", and phonology decides what a possible lexical address (hence a possible word/morpheme) is. 351

9.3. Phonology defines lexical access: look-up vs. compute

352

9.3.1. Introduction Given that phonological information contributes to the identification of morpheme boundaries, how exactly are phonological parsing cues used when the segmented string is compared with the lexicon in order to retrieve the associated morpho-syntactic and semantic information? And how is lexical information accessed in case of inputs such as stepp-ed [stEpt] which do not fly any phonological flag that betrays their morphological complexity? These questions are closely related to the issue regarding the identification of those alternations which represent actual synchronically active, online computed processes. Recall from §126 that the crux of phonology (as opposed to syntax, see Scheer 2004b:23ff), is that all alternations observed do not represent the result of a computation over a common underlying form (Kaye 1995:313). Different theories have adopted different delineations of alternations that are due to online processing and alternations which are lexicalised (see §124). It was reported in §276 that Government Phonology is located on 85

Only infinite morpheme length would produce an infinite set of possible lexical entries. Many languages impose restrictions on the maximal length of morphemes, and those that do not (English can accommodate verbs of virtually any size: to abracadabrate etc. is certainly workable) may be thought of as limiting the projection of the addressing space to a certain length (e.g. calculated in terms of the number of x-slots). Kaye does not discuss this issue.

294 Chap 9: Kaye (1995), selective spell-out and freezing no look-back the far end of the spectrum where actual phonological online computation covers only a small subset the alternations observed. Namely, classical phenomena of English phonology such as irregular past tense formation (keep - kept), Velar Softening (electri[k] - electri[s]-ity) and Trisyllabic Shortening (div[aj]ne - div[I]nity) have nothing to do with phonology: they are not the result of any computation performed by the φ-function. Rather, they represent two independent lexical entries (more on this debate in §569). On this backdrop, the following pages provide a more fine-grained picture of what it means for two morphologically, paradigmatically or semantically related items to be distinct lexical entries: if keep and kept are in this situation, do they have the same lexical structure as two entirely unrelated words such as, say, table and house? The answer is no. There is a sense in which keep and kept are related, but this relation is not a matter of computation. Rather, it involves a "lexical pointer", i.e. something that ran under the label of via-rules in Natural Generative Phonology. 353

9.3.2. Look-up I: related keep - kept vs. unrelated table - house Morphologically complex kept is indistinguishable from mono-morphemic items, as is evidenced by words such as adopt and apt. This is also the case for most irregular past tense forms: left is just like soft and rift, no phonological property betrays wove (cf. clove, stove), and sang cannot be told from bang or fang.86 Since these items do not provide any parsing cue, the listener can only rely on a direct look-up in the lexicon in order to find out about the associated morpho-syntactic and semantic information. The difference between keep - kept and a pair such as table - house is the fact that the latter two items are not grammatically related. According to Kaye, this contrast is mirrored by the shape of the lexical entries at hand: table and house list the full set of morpho-syntactic and semantic properties, and so does the lexical entry keep. By contrast, the entry kept is empty, except for the information that it is the past tense form of keep. That is, the user who hits the entry kept remains uninformed of the morpho-syntactic 86

When comparing the past tense of regular and irregular verbs, it appears that the former provide generous parsing cues, while the latter seem to put some effort into "hiding away": they (have) transform(ed) their body so as to be undetectable by the phonological filter. Kaye (1995:310ff) further discusses this "conspiracy" strategy.

Lexical access and the organisation of the lexicon 295

and semantic features that are necessary in order to interpret this item. These are only listed in the entry keep, and a "pointer" in the entry kept indicates where they may be looked up. The pointer at hand does the same labour as so-called via-rules in Natural Generative Phonology (Hooper 1976:47f). This theory denies the existence of any kind of morpho-phonology (only phonetically expressed information may be used by phonological processes, see §127) and hence considers keep and kept two separate lexical entries which, however, are related by a via-rule (while table and house are not). 354

9.3.3. Look-up II: phonologically similar keep - kept vs. regular suppletion go - went Let us now consider cases of regular suppletion such as go - went. Just like keep - kept, they entertain a grammatical relationship and hence are different from table - house. Is keep - kept thus a case of suppletion that works in the same way as go - went? Kaye says no: the difference is that the members of the former pair are phonologically related, while the members of the latter are not. That is, keep and kept begin with a [k] and possess a [p], whereas go and went do not share any phonological property. In diachronic terms, the former two items have contracted a derivational relationship at some point in the history of the language, while the latter two have never been derived from one another. Phonological relatedness, Kaye argues, plays a role when a lexical entry is looked up and accessed. During this process, it is loaded into shortterm (active) memory; its immediate neighbours, however, accompany this move. Kaye uses the metaphor of vinyl records on a shelf: when looking for a particular record in an unordered collection, rather than going item by item, one pulls out (or rather: pulled out) a batch of records that probably contains the target. This neighbourhood method, Kaye contends, is also practised in lexical access: the target entry is loaded into short-term memory together with some previous and some following items. Short-term memory is very efficient but costly: access to its content is immediate (as opposed to a lengthy look-up procedure in long-term storage), but has only a small capacity. Hence keep will be in the batch that is loaded into short-term memory when kept is looked up: because of their phonological similarity and the fact that the lexical addressing system is based on phonological structure, keep and kept are fairly close neighbours. Activating the pointer in the entry of kept will lead to keep where the rele-

296 Chap 9: Kaye (1995), selective spell-out and freezing no look-back vant morpho-syntactic and semantic features may be retrieved. Since, however, keep is already present in the short-term memory, there is no need for a heavy look-up procedure: the access is immediate. By contrast, go will not end up in the short-term memory batch when went is looked up: the addresses of both items are too distant. How, then, are the relevant morpho-syntactic and semantic features retrieved? Kaye argues that in cases of true suppletion, there is no lexical pointer. Rather, morpho-syntactic and semantic features are repeated in both grammatically related entries, which also contain an explicit statement regarding their relationship: "past tense see the entry 'went'" for go, "present tense see the entry 'go'" for went.

355

9.3.4. Computation: when parsing cues are available (peeped) The three patterns that have been discussed thus far (table - house, keep kept, go - went) share the property of not offering any parsing cue. This complicates the retrieval of the associated morpho-syntactic and semantic features for keep - kept: kept is morphologically complex (other than the lexical meaning of keep, it possesses a past tense feature), but this complexity cannot be detected by phonological means. In order to access the associated morpho-syntactic and semantic information, a two-step procedure is thus necessary whereby the first lexical entry that is hit (kept) does not provide the relevant features. These are only available in the lexical entry (keep) that is identified by the pointer which is found in kept. Compared to this two-step look-up procedure, things are less complicated when parsing cues allow for the identification of morphemes: the associated morpho-syntactic and semantic information may be directly retrieved from the first lexical entry that is looked up. This is what Kaye calls identification by (phonological) computation. For example, the structure [[peepø]dø] is identified by phonological means when the input peep-ed [piipt] is submitted because there are no long vowels before [pt] in mono-morphemic items. Since domain boundaries identify morpheme boundaries, the stem [peepø] is isolated and may be submitted to look-up. This leads to a match which grants direct access to the associated morpho-syntactic and semantic information.

Lexical access and the organisation of the lexicon 297 356

9.3.5. How parsing is done with hidden morphology (stepped) Finally, the fifth pattern to be considered is hidden morphology that enjoys a phonological exponent, e.g. stepp-ed [stɛpt]. Like for keep - kept, a stem and a past tense feature need to be identified; this time, however, the latter has an independent lexical representative (while there is no morpheme -ed ever identified when kept is parsed). Since [stɛpt] does not provide any phonological parsing cue, look-up of the entire item is performed, which however returns no match: there is no lexical entry stept. Therefore alternative parsing mechanisms need to be called on; these may rely on morpho-syntactic information that is available from other chunks of the signal, or on look-up of substrings of stept.

357

9.3.6. The four identification patterns are more or less costly The four patterns discussed correspond to four different ways of identifying the morpho-syntactic and semantic information that is associated to the morphemes in question. The pathways at hand are more or less complex, and hence should be more or less costly. The easiest way of accessing the relevant lexical information is when no phonological computation is performed on an item, whose look-up then produces immediate success: this is the case of unrelated words (table house) as well as of suppletive forms (go - went). In the case of kept, the first lexical access leads to an empty entry, which only provides a pointer. This pointer needs then to be followed in order for the relevant information to be retrieved. The parsing of peeped supposes additional phonological computation before look-up can be performed. Finally, the failure of phonological identification of the morphologically complex stepped produces certainly the most complex procedure: non-phonological morpheme identification needs to be performed before look-up can be successful.

358

9.3.7. Why is direct look-up not generalised? A question that is prompted by this situation is why the less costly procedure, direct look-up, is not generalised. That is, why is the entire lexicon not organised along the lines of went? A possible answer is the amount of storage space that would be needed: all grammatical forms of a root, including inflection, would have to

298 Chap 9: Kaye (1995), selective spell-out and freezing no look-back be listed separately. Kaye rejects this perspective because storage capacity is not something that humans are short of: they can learn three, four or more foreign languages and accordingly mobilise three or four times more storage space than if they just spoke their mother tongue. As may be seen from result, the human brain is able to satisfy this significant demand of extra storage space: the learning of new words does not cause the loss of old recordings. Rather than calling on storage, Kaye hints at acquisition, which would be slowed down: were all grammatical forms of a word lexically recorded, no paradigms could be created on the grounds of a given form. That is, a child could not infer the third person of a verb in case the root is known.

359

10. Conclusion Kaye's approach to the interface combines a central contribution of Lexical Phonology (interactionism, §275), Halle & Vergnaud's idea of selective spell-out (§277) and modification-inhibiting no look-back (§§293,299). This makes his system utterly modern: today all three takes are standard in the interface-oriented minimalist architecture. Since the 60s indeed, generative theory (Generative Semantics and Lexical Phonology set aside) has always respected the principle "all concatenation before all interpretation" – until Chomsky's derivation by phase has made it interactionist (§305). Since the 60s, either all nodes were spelled out (SPE), or spell-out was absent from the radar (Lexical Phonology) – today phase theory is based on Halle & Vergnaud's selective spellout (§§256,763). Since Chomsky (1973), a number of no look-back devices have been proposed, but the modification-inhibiting no look-back that first appeared in the 80s (§293) and which is generalised by Kaye (§299) is entirely different. Chomsky's Phase Impenetrability Condition today does the same thing: it "freezes" previously computed strings, which are then unavailable for further computation (§306). Even if the phonological version(s) of modification-inhibiting no look-back need to be worked out (the phonological situation requires a weaker formulation, §302), the general idea is the same. Of course, the fact that minimalist syntax has made the same choices about ten years later does not mean that they are correct. The coincidence, however, is quite striking, and the more so given that the modern syntactic literature makes no reference to phonological precedents at all (Chomsky

Conclusion 299

does not even quote his own 1973 article when he introduces his freezing no look-back). Kaye has not invented modification-inhibiting no look-back (§293), but he has generalised the idea, making it a property of computation as such (this is the case in current syntactic theory as well). Also, he introduced it into the analysis of affix class-based phenomena. Finally, he proposed that affixes do or do not trigger interpretation, but rather than the node that dominates the affix, the sister of the affix is spelled out (§282, see Scheer 2008c). This gives Halle & Vergnaud's selective spell-out an entirely new face, one that coincides with modern phase theory (or rather, modern phase theory coincides with Kaye's system) where the edge of the phase also defines the sister of the head X° (i.e. the complement) as the string that is sent to interpretation (more on this in §765). Like Halle & Vergnaud, Kaye promotes the restoration of SPE. This is true for example regarding the resident issue of multiple computational systems. The take of SPE is that there is a specific word-level phonology, but no morpheme-specific or sentence-specific phonology (Praguian segregation). Halle & Vergnaud and Kaye hold the same position. Regarding affix classes, Kaye's system is unlike others in two important respects. For one thing, class 2 affixes (rather than class 1 affixes as with Halle & Vergnaud) trigger interpretation (§281). On the other hand, as was mentioned, the driving force of affix class-based phenomena is a no look-back device: no other approach achieves underapplication by this means (§279). On the empirical side, Kaye's system offers a fairly good coverage of English affix class-based phenomena. When compared to Lexical Phonology and Halle & Vergnaud, it appears to be a step ahead. This is true conceptually regarding the former, where the rule-triggering pattern is accounted for at the expense of reintroducing untranslated morpho-syntactic objects (Mohanan's brackets) that were blessedly eliminated by interactionism (§170). On the empirical side, Kaye's system fares better than Halle & Vergnaud's, who cannot account for the rule-triggering pattern at all (§250). A concern is the syntactic viability of Kaye's claim that terminals can be spelled out independently (§274). The syntactic echo of this prediction is discussed in §316. Kaye's domain structure (phase structure in modern terms) follows the general idea that some morpho-syntactic boundaries are visible in phonology, while others are not. The former are phase heads, the latter are not. Beyond the morpho-syntactic footprint that domain structure represents in

300 Chap 9: Kaye (1995), selective spell-out and freezing no look-back phonology, its role as a lexical element, and hence in diachronic evolution, was discussed (erosion of domain structure, §328). Kaye's approach to the interface as a whole is perception-oriented and hence functionalist: according to him, domain structure (and phonology as such) exist in order to provide parsing cues that enhance morpheme identification (§340). Another function of phonology is to provide an addressing space in the lexicon, which organises lexical access (§346).

360

Chapter 10 Prosodic Phonology: on the representational side

361

1. Overview: autosegmentalised boundaries and fresh data Prosodic Phonology has been developed some 30 years ago and today stands still unchallenged87 as the (representational) phonological interface theory. After the advent of Optimality Theory, Prosodic Phonology has been translated into the new constraint-based environment, but this move did not modify its basic tenets and mechanisms: in OT, the interface with morpho-syntax is as much managed through prosodic constituency as it was before (see §455). As we will see below, Prosodic Phonology carries over the interface architecture of SPE into the autosegmental environment of the 80s. The only true evolution concerns the units that carry morpho-syntactic information: local boundaries are replaced by autosegmental domains (the Prosodic Hierarchy). Also, SPE concepts are sometimes borrowed without reference. This view on the matter is not quite what the early Prosodic Phonology literature offers:88 the new theory is advertised as an antithesis of SPE. In some cases, counterfactual statements are even made in order to show that the new theory is radically different. For instance, Nespor & Vogel (1986:37) write that "in traditional generative theory it was supposed that these [morpho-syntactic] domains directly correspond to syntactic constituents (see Chomsky & Halle, 1968)" (more on this in §385). The pages below thus show that the genuine contribution to interface theory made by Prosodic Phonology is the autosegmentalisation of the carriers of morpho-syntactic information: instead of boundaries, prosodic domains are inserted into phonological representations. I argue that this development was a result of the general autosegmentalisation that has affected all areas of phonology in the late 70s and early 80s: the Prosodic Hierarchy is the specific implementation of autosegmentalism in the realm 87

88

Or almost: some voices in Distributed Morphology quarters revive the direct syntax approach and hence do away with the Prosodic Hierarchy: Pak (2008), Samuels (2009a:239ff), see §580. In the later literature (since about 1986), SPE is not a reference anymore. Nespor & Vogel (1986) for example take prosodic constituency for granted without discussion of the transition with SPE.

302 Chap 10: Prosodic Phonology, on the representational side of the interface (see §368). It represents a major historical break: boundaries have always been the interface currency since the 19th century. The difference between boundaries and domains is their local vs. non-local character. That is, a morpheme cannot belong to a boundary (as it can belong to a prosodic constituent), and it does not make sense to talk about a prosodic constituent that intervenes between two morphemes (while a boundary always separates items in the linear string). As far as I can see, the relative merits of local and non-local intervention have never been discussed (§366). Rather, the few occasions in the early Prosodic Phonology literature on which the competition with boundaries is addressed dismiss them on the grounds of their diacritic character. It was taken for granted, then, that unlike boundaries, the Prosodic Hierarchy is not a diacritic. This tacit assumption, however, does not stand up to fact: the Prosodic Hierarchy is just as much a diacritic as SPE-type boundaries, if an autosegmental one (§402). Later in the book it is argued that representational communication with phonology must not be through diacritics (§692) – but it needs to be local (§706). What is required, thus, is some kind of non-diacritic boundaries. That this is not a contradiction in terms is shown in Vol.2 and Scheer (2008a, 2009a,c). On the credits side, Prosodic Phonology can certainly claim the fact that it has made translation systematic and consensual. Recall that the frustration of the late 70s has prompted calls from different quarters in favour of direct reference to morpho-syntactic categories (§131). Prosodic Phonology has fought out this issue in the mid 80s (see §407 below). As a result, the need for a Translator's Office where morpho-syntactic information is transformed into the phonological idiom has become the backbone of representational interface theory (Indirect Reference). A curious aspect of this evolution is that it was made on the grounds of an argument that turns out to be vacuous upon inspection: nonisomorphism supposes that domains are taken for granted – if the same reality is looked at through the prism of boundaries, non-isomorphism evaporates (§416). On the other hand, Prosodic Phonology has never called on modularity, the reason that makes Indirect Reference really indispensable (§410). This is strange indeed since the development of Prosodic Phonology and Fodorian modularity was contemporary in the 80s. On the empirical side, Prosodic Phonology has aroused an important activity, to which phonologists owe a great amount of fresh and detailed descriptions, namely regarding the interface with syntax. Like other theo-

The roots of Prosodic Phonology 303

ries, however, Prosodic Phonology could not elucidate the mapping puzzle (see §111). The new empirical wealth has rather added to the tessellate character of the picture (§§387,392f,396). Prosodic Phonology is certainly the most influential interface theory of the past 40 years. Its resounding success may be measured when realising that up to the present day there is hardly any phonological theory that does without the Prosodic Hierarchy. Also, prosodic categories have ceased to engage phonologists theoretically: the prosodic word and its peers have become descriptive categories, much in the way syllabic vocabulary such as onset and coda is used when informally talking about languages. As is obvious from this overview, Prosodic Phonology is entirely devoted to the representational side of the communication between morphosyntax and phonology. It does not make any claim as to whether there is a procedural management, or how it could work. In actual fact, the question is eluded. By default, a peaceful cohabitation with Lexical Phonology, the dominant contemporary procedural theory, was engaged. However, Selkirk (1984) and Inkelas (1990) have pointed out that both theories actually compete below the word level. The relationship between Prosodic Phonology and Lexical Phonology (which continues to be an issue in OT, see §488) is discussed in §423. The following pages thus address the issues mentioned and try not to lose track of the general picture.

362

2. The roots of Prosodic Phonology

363

2.1. Selkirk adapts Liberman & Prince's strong/weak arboreal structure The earliest ancestors of Prosodic Phonology are commonly taken to be Liberman (1975) and Liberman & Prince (1977). The central idea of these authors is that segments are dominated by a multi-layered arboreal structure (syllables, feet and words) which expresses rhythmic (linguistically "musical") properties of the linear string and assign relative prominence (strong vs. weak status) to individual chunks. Based on this line of thought, Elisabeth Selkirk developed the first model of Prosodic Phonology: she fertilised the autosegmental arboreal idea for the interface of phonology with other modules, something that was not originally intended by Liberman & Prince (1977).

304 Chap 10: Prosodic Phonology, on the representational side In Selkirk's early work (Selkirk 1978, 1980a,b, 1981 [1978], 1981) and especially in the ground-laying article Selkirk (1981 [1978]),89 the sixlayer Prosodic Hierarchy which is still in place today was introduced: the syllable, the foot, the phonological word, the phonological phrase, the intonational phrase and the phonological utterance. Selkirk's proposals were influential, and mainstream phonology rapidly integrated them as a major contribution to the general expansion of the autosegmental idea. Her 1984 book (Selkirk 1984), though, which was advertised in all of her articles since 1978, finally took a different turn under the influence of Prince's (1983) grid-only approach: autosegmental prosodic constituency was evacuated altogether in favour of the metrical grid (more on this in §426). Nevertheless, Selkrik (1984) is a landmark of early Prosodic Phonology, also because it assures a bridging function between the linear SPE environment and the new autosegmental interface – an aspect that is discussed at length in §369. Finally, two years later, Selkirk (1986) returns to prosodic constituency, now arguing for a "peaceful coexistence" of the original Prosodic Hierarchy and the metrical grid (thus following Liberman & Prince 1977 and Nespor & Vogel 1982:226, 1986, see §427).

364

2.2. A second strand that became mainstream: Nespor & Vogel Building on Selkirk's work but mobilising fresh data from languages such as Italian and Greek, Marina Nespor and Irene Vogel have grounded a parallel stream of inquiry: after an exploratory period (Nespor & Vogel 1979, 1982, 1983, Napoli & Nespor 1979, Nespor 1985, 1986, Vogel 1982, 1985, 1986), their 1986 book (Nespor & Vogel 1986) concentrates the insights gained. It rapidly became the authoritative reference in Prosodic Phonology, and indeed the standard theory of how morpho-syntax communicates with phonology. Other authors that have contributed to the early period which prepared Selkirk (1984) and Nespor & Vogel (1986) include Hayes (1989 89

This article is commonly quoted as Selkirk (1978), hence as the oldest ancestor of Prosodic Phonology. Since chronology sometimes matters below, I add this date in square brackets. The content of this article was first presented at an Amherst conference in 1978; a manuscript almost identical to the 1981 publication was then circulated at least since 1980. The first published version, however, is the text that appeared in the proceedings of the Nordic Prosody conference in 1981.

From boundaries to domains: a historical choice gone almost unnoticed 305

[1984]), Booij (1983, 1985a,b, 1986), Neijt (1985), Dell (1986), Itô (1986), Gvozdanović (1986). Namely Hayes (1989 [1984]) was influential.90 The framework that emerged in Hayes (1989 [1984]) and Nespor & Vogel (1986), as well as in the strand represented by Selkirk (1984), stands unchallenged today (but see note 87). Almost thirty years have gone by without any substantial modification of its basic tenets. A fair means to judge the impressive success of Prosodic Phonology is the fact that its genuine units – the foot, the phonological word, the phonological phrase, the intonational phrase and the phonological utterance – have started out as theoretical constructions but today are common descriptive categories. Thus just like for syllables, there may be different opinions on how exactly a phonological word is built and what it encompasses, but its existence is beyond doubt: there is an arboreal structure above the syllable that encodes morpho-syntactic information.

365

3. From boundaries to domains: a historical choice that has gone almost unnoticed

366

3.1. Boundaries are diacritic and local Since the 19th century, morpho-syntactic influence on phonological processes has always been conceived in terms of boundaries. Vocabulary may have varied according to the period and the theory, but the way information is transmitted through the representational channel remained stable over neogrammarian analysis, structuralism (Trubetzkoy's Grenzsignale, juncture phonemes) and SPE (boundaries): the carrier is diacritic and local. Boundaries are local because they define the relationship between two adjacent morphemes or words. This is the fundamental property that makes them different from domains, which are non-local by definition: they span a number of elements of the linear string, thereby creating labelled clusters. That is, an individual element of the linear string belongs to a domain, but it cannot belong to a boundary. On the other hand, a boundary is precisely located in the linear string and can influence only adjacent objects: the one immediately preceding and the one immediately following. It does not make sense to talk about domains that intervene, or are located 90

This article was circulated as a manuscript since 1984, but only published in 1989. For this reason, it does not mention Nespor & Vogel (1986), but is referred to for example in Selkirk (1986).

306 Chap 10: Prosodic Phonology, on the representational side between two elements of the linear string. Domains, by definition, are nonlocal, while the essence of boundaries is to be local. The locality of intervention is well incarnated by the traditional notion of sandhi: sandhi phenomena occur at the break of two morphemes or words, and are triggered (or uninhibited) by this division. Boundaries are also diacritic: when it comes to give them a material identity, arbitrary symbols are chosen. The diacritic character of boundaries is recognised in the literature, and its undesirable consequences are pointed out: see §§69f (structuralists), §§119,133 (generativists), also Scheer (2009a,c), Vol.1 §§84,87 (discussion of the phonological identity of the beginning of the word). It is argued below that the true difference between boundaries and prosodic domains is their local vs. non-local action, rather than their (non-)diacritic character. The scarce Prosodic Phonology literature that talks about the transition from boundaries to domains at all, however, has never disentangled the local and the diacritic issue: the latter was pinpointed, while the former went unnoticed. There can be no doubt that we need to do away with diacritics – but this does not mean that locality has to go as well. This is the whole point of Direct Interface (see Vol.2, Scheer 2008a, 2009a,c): to find a means of making representational intervention local and non-diacritic. 367

3.2. The elimination of boundaries in Lexical Phonology remained unreflected in Prosodic Phonology Lexical Phonology is relevant for the debate regarding the competition between boundaries and domains insofar as it has its own ways of eliminating the former. Recall from §162 that the existence of procedurally ordered lexical strata makes boundaries toothless. Lexical Phonology pinpoints the redundancy of boundaries given the stratal environment, but does not produce any positive argument against them. Boundaries are just useless because the procedural mechanism does the labour and, according to Lexical Phonology, does it better since it is independently motivated (e.g. Mohanan 1982:94ff). As far as I can see, the fate of boundaries in Lexical Phonology – and hence the argument with the procedural side of the interface – is entirely absent from the Prosodic Phonology literature. This ties in with the tormented (non-)relationship between Lexical Phonology and Prosodic Pho-

From boundaries to domains: a historical choice gone almost unnoticed 307

nology: §423 discusses the peaceful coexistence that was officially proclaimed, as well as the backstage conflict. In the small body of early works that is discussed below, Prosodic Phonology has thus started the discussion of boundaries anew. This time, boundaries are not a competitor of serially ordered lexical strata, but of autosegmental domains. Unlike in the Lexical Phonology literature, this debate produced some concern for boundaries as such and for their properties.

368

3.3. Prosodic Phonology is a child of autosegmentalism When reading through the Prosodic Phonology literature, the almost complete absence of reference to work from before the 80s (structuralist and generative alike) is quite striking. The only exceptions are a few early articles by Elisabeth Selkirk, two articles by Geert Booij (1983, 1985a) (on which more shortly), and most notably Selkirk's 1984 book, which assumes a pivotal role between SPE and Prosodic Phonology. Let us start by having a closer look at how Selkirk (1984) explains the transition from boundaries to domains. Selkirk has worked on the interface since the early 70s, mainly on the grounds of French liaison (Selkirk 1972, 1974). During that period, her work was couched in the SPE framework, thus using boundaries. Living through the advent of autosegmentalism, it seemed clear to her that the interface with morpho-syntax needs to join the move: a situation where all areas of phonology (syllable structure, stress, internal structure of segments) except its upper interface are progressively autosegmentalised would have been strange. The quote below shows that Selkirk clearly identifies the elsewhere winning arboreal approach as the motor for the introduction of domains.

308 Chap 10: Prosodic Phonology, on the representational side (134) "The syllables of phonological representation are arranged in some kind of hierarchical organization. […] By 'hierarchical organization' we mean, very roughly speaking, the organization of the units of phonological analysis into layers, vertically arranged on the same plane. […] This conception of phonological representation as having its own hierarchical structure(s) demands a radical rethinking of the relation between syntax and phonology. […] Thus the interpretation question – the question of the mapping between phonological representation and syntactic representation – takes on a much greater importance than in the standard theory, and has an entirely different quality to it. It must be viewed as a characterization of the relation between the syntactic hierarchy, on the one hand, and the phonological hierarchy (or hierarchies), on the other." Selkirk (1984:7f)

Therefore, (135) "the junctural properties of sentences should be somehow represented 'suprasegmentally' rather than as the segmental boundaries of the standard theory. […] Thus the theory of phonological representation that we will advocate here eliminates segmental boundary elements altogether." Selkirk (1984:8)

This is how the domain enters the scene: the idea that morphosyntactic structure defines the domain of application of phonological rules had already been voiced early on in the linear environment, namely by McCawley (1968) (see §113). Selkirk now calls on this perspective: domains are non-local by definition, and they have an obvious autosegmental identity in terms of arboreal structure. 369

3.4. The (non-)discussion of boundaries in Prosodic Phonology

370

3.4.1. Boundaries were not an issue anymore for Nespor & Vogel (1986) Selkirk's (1984:7f) description of the transition from boundaries to domains is precious since the historical break at hand has remained by and large unreflected in the Prosodic Phonology literature. This is witnessed, among other things, by the fact that prosodic domains were taken for granted without discussion in Nespor & Vogel (1986), the book that is considered the central reference of Prosodic Phonology. In the very first sentence of their book indeed, Nespor & Vogel merely state their disagreement with the linear SPE system that uses boundaries. They then go on talking about domain theory without making any argument regarding the transition.

From boundaries to domains: a historical choice gone almost unnoticed 309 (136) "In early generative theory, phonology was characterized by a linear organization of segments and a set of phonological rules whose domains of application were implicitly defined in terms of the boundaries of the surface morpho-syntactic constituent structure. […] It is our contention that this view of phonology is fundamentally inadequate. […] The subsystem [of phonology] we will be concerned with in the present study is the prosodic subsystem, and in particular, the theory of domains." Nespor & Vogel (1986:1)

Hence in 1986 boundaries are not considered a serious competitor anymore: they are a mere vestige of an earlier theory which today is overcome. In the remainder of the book, boundaries are only occasionally mentioned: once on page five where their inadequacy is said to follow from non-isomorphism, and another time on page 35s. 371

3.4.2. Looking for anti-boundary arguments Since the transition from boundaries to domains is a fundamental break in phonological culture, I have read through the early Prosodic Phonology literature (that is, until 1986) in order to find out whether this critical issue was discussed at all. The relevant literature falls into three categories: most of the time the issue regarding boundaries is not mentioned at all; sometimes boundaries are declared inadequate without argument but referring to other articles; finally, arguments against boundaries are made in less than a handful of articles. Selkirk (1978), Nespor & Vogel (1979), Nespor (1985, 1986), Vogel (1985, 1986), Gvozdanović (1986), Neijt (1985), Booij (1985b, 1986), Dell (1986) and Itô (1986) fall into the first category. Reference to articles that are supposed to contain anti-boundary arguments is made in Nespor & Vogel (1982:226) (to Napoli & Nespor 1979 and Clements 1978), Selkirk (1980b:580) (to Selkirk 1981 [1978], 1984), Hayes (1989 [1984]:203) (to Selkirk 1980a) and Selkirk (1986:376) (to Selkirk 1980a and Rotenberg 1978). Most of the target articles, however, contain no argument against boundaries at all. This is the case of Napoli & Nespor (1979), Selkirk (1981 [1978]) and Selkirk (1984) (see §368). Among the references quoted, only Selkirk (1980a) and Rotenberg (1978) really argue against boundaries. This is also the case of some sources that are not mentioned: Hayes (1989

310 Chap 10: Prosodic Phonology, on the representational side [1984]), Booij (1985a) and the aforementioned occasion in Nespor & Vogel (1986:35f). Let us first look at those references which are supposed to offer arguments against boundaries, but in fact do not.

372

3.4.3. References that do not contain any argument against boundaries Napoli & Nespor (1979:818ff) talk about boundaries only when they review an argument made by Selkirk (1972). Their goal is to show that phonological rules can only make reference to surface-, not to deep syntactic structure. Working on French liaison, Selkirk (1972) has collected evidence to the end that a possible liaison between two phonetically adjacent words is blocked if a trace intervenes. On her analysis, movement has taken away the lexical item, but left its boundaries in place. Liaison, then, is computed on the grounds of the total number of boundaries that occur between the two words, including those that originate in the trace. Napoli & Nespor (1979:818ff) show that in a parallel situation regarding Raddoppiamento Sintattico in Italian, traces are irrelevant. This discussion is also reflected at greater length in Nespor & Vogel (1986:48ff), where the authors defend the position that phonetically absent syntactic constituents can never influence phonological processes. The issue at hand is thus quite independent from boundaries as an interface currency: an argument is made against the phonological incidence of phonetically absent objects, not against boundaries as such. Nothing is said about which type of carrier should be chosen in order to represent morpho-syntactic information in phonology. Referring to this article with the promise to find arguments against boundaries is misleading at best. The case of Clements (1978) is parallel. True, this author admits that his previous analysis of Ewe (Kwa, West Africa) which used "tone group boundaries" is unwarranted because "a set of boundary-insertion conventions unmotivated in the phonology of other languages" is required in order to describe the tonal alternations in external sandhi situation (p.22). But Clements then develops an alternative analysis that does away with tone group boundaries in favour of phonological rules whose structural description makes direct reference to syntactic constituents. Again, there is no argument against boundaries as such in this article, nor is it in any way Clements' intent to eliminate boundaries from phonological theory. It is only suggested that for the particular data set at hand an analysis without

From boundaries to domains: a historical choice gone almost unnoticed 311

boundaries fares better. This analysis then follows the lines of direct syntax (§136), which is at odds with Prosodic Phonology (§407). Selkirk (1981 [1978]) does not contain any argument against boundaries either: after having introduced the six layers of the Prosodic Hierarchy, the author simply says in the conclusion that if prosodic constituency is accepted, boundaries are superfluous. (137) "Once prosodic categories form part of the phonological representation, the motivation for boundaries (cf. SPE, Selkirk (1972)) as part of a phonological representation disappears. Boundaries are none other than an encoding in the string of segments (the standard generative representation) of the higher order prosodic structure organizing that string. At one time thought of as the appropriate formal mechanism for delimiting domains of application of phonological rules, boundaries are rendered otiose by the introduction of prosodic structure into phonological representation and, in particular, the inclusion of prosodic categories which effectively designate those domains. Looking at it in a different light, we can think of the focus being taken away from boundaries, which were originally viewed as markers of juncture between syllables and other prosodic units, and being put on the prosodic units themselves." Selkirk (1981 [1978]:136ff)

Booij (1983) makes the same point. (138) "In the theory of prosodic phonology, grammatical boundaries can be dispensed with in phonological representations." Booij (1983:268)

No doubt Selkirk and Booij are right – but this does not tell us why boundaries should be replaced by domains in the first place, or in which way domains are superior. Declaring the existence of domains ex cathedra is not making an argument. Rather, Selkirk (1981 [1978]) and Booij (1983) follow the line of reasoning that is adopted later on by Selkirk (1984) (see §368): the general trend towards autosegmentalism imposes autosegmental structure for the interface as well; since boundaries are not autosegmental but local, they do not qualify. Finally, Nespor & Vogel (1986:35f) try to make an argument against boundaries on the grounds of a rule in Modern Greek. They show that in certain circumstances, the nasal of a nasal-obstruent sequence assimilates to the obstruent in place, while the obstruent, if voiceless, becomes voiced. The process at hand applies across morpheme boundaries as well as between a negative element and a verb, and between an article and a noun. It is blocked, however, between an auxiliary and a verb. Nespor & Vogel say that a solution using boundaries would have to place a morpheme boundary

312 Chap 10: Prosodic Phonology, on the representational side between articles/negative elements and nouns/verbs. This, however, the argument goes, is syntactically unwarranted since the division between words cannot be a morpheme boundary. This argument is not conclusive: there are no morpheme boundaries between words indeed – but a solution using boundaries does not need to assume that there are any. All theories need to somehow group morphosyntactic divisions into a triggering and a blocking family. It is perfectly regular to see that morphological and syntactic boundaries are distributed over both of these groups. Hence there is no point in talking about morpheme boundaries between words; morpheme boundaries just happen to produce the same effect as some word-level boundaries.91

373

3.4.4. The diacritic argument Unlike in the references mentioned in the previous section, boundaries are really under fire in Rotenberg (1978), Selkirk (1980a), Booij (1983, 1985a) and Szpyra (1989). The point made by these authors concerns the diacritic character of boundaries.92 In a chapter called Against Boundaries (that was already quoted in §136), Rotenberg (1978:16ff) makes a convincing argument against boundaries: phonology interprets phonological objects (just like syntax syntactic objects), not bananas or pink panthers (see also Pyle 1972 on this issue, as reported in §136). This is a surprisingly modern reasoning: it looks at the interface from the modular perspective, which was little developed by the time Rotenberg wrote. That is, a module can only parse and understand vocabulary that belongs to its own idiom. This idea is called domain 91

92

Kaisse (1985:109ff) also argues against boundaries. Instead of replacing them by prosodic domains, however, she proposes to have their labour done by direct reference to syntactic categories (see §4070). Her argument relies on the maximal number of different boundaries that SPE and Selkirk (1972, 1974) allow for, i.e. three. She presents Greek evidence which, according to her analysis, requires four different boundaries. Hence the argument does not concern boundaries as such; rather, a particular implementation thereof is at issue. It is noteworthy that linear diacritics have also been proposed in Prosodic Phonology: after having constructed the metrical grid through regular mapping rules, Selkirk (1984:314) argues for "Silent Demibeat Addition" at morphemeand word-intersections. Phonological rules, then, make reference to the overall grid structure. Selkirk (1986:376) admits that these silent demibeats are diacritic boundaries in a different guise.

From boundaries to domains: a historical choice gone almost unnoticed 313

specificity (see §610). Its application enforces translation from morphosyntactic into phonological categories (§649), an issue that lies at the heart of Prosodic Phonology (Indirect Reference, see §§377,406). It is thus certainly correct to call on Rotenberg as a voice against boundaries – but not in order to promote the domain-based alternative. Rotenberg tells us nothing at all about the comparative merits of boundaries and domains. The case of Selkirk (1980a:126ff) is similar: she reviews the arguments made by Pyle (1972) regarding the overgeneration that boundaries introduce. Boundaries indeed allow for the formulation of outlandish rules that lie outside of what natural language can do for sure. The reason again is their diacritic character: unlike true linguistic objects, arbitrary symbols have arbitrary effects.93 Szpyra (1989:11, 182f) and Booij (1983:268f, 1985a) fall into the same category as Rotenberg (1978) and Selkirk (1980a). In his conclusion, Booij (1985a:34) writes that the theory of Prosodic Phonology "excludes the rather arbitrary use of boundaries made possible in the SPE-framework" (see §393 for further discussion of Booij 1985a). Selkirk (1980a) also considers a competitor that is more serious than just simple boundaries: McCawley's (1968) idea that boundaries define the domain of rule application, and that different boundary strengths determine different domains (see §113). Obviously, this is the direct translation of autosegmental domains into linear vocabulary – or rather, the new autosegmental perspective is the translation of McCawley's (1968) proposal. If both are just notational variants, the new Prosodic Hierarchy is in trouble. At this point, Selkirk says that (139) "the relations among boundaries that are captured in the strength hierarchy must be stipulated in the theory. They do not follow from anything inherent to the notion of boundary in the theory." Selkirk (1980a:128)

This is certainly true – but does the layering of prosodic constituents follow from anything in the theory of domains? Prosodic constituents exist 93

The diacritic-overgeneration issue is usually brought up when the historical evolution from boundaries to prosodic constituency is described in more recent (overview) literature, e.g. Inkelas & Zec (1995:537f). The question whether prosodic constituency also overgenerates, however, is not addressed on these occasions. It is shown in §402 below that the Prosodic Hierarchy is as much a diacritic as SPE-type boundaries.

314 Chap 10: Prosodic Phonology, on the representational side because the related facts exist, and their size adapts to whatever is found out there. Hence just as much as boundary strength, domains merely record what happens. Finally, Hayes (1989 [1984]:203ff) makes a similar point. He argues that boundary-based theories need two independent stipulations in order to cover the facts: 1) if a phonological rule can apply across one kind of boundary, it can also apply across all "weaker" boundaries; 2) if a rule applies before or after one kind of boundary, then it also applies before or after all "stronger" boundaries. Domain-based theories, on the other hand, need only one stipulation for the same job, i.e. the Strict Layer Hypothesis (see §383 below). This argument does not take account of the non-monolithic character of the Strict Layer Hypothesis, which is pointed out for example by Inkelas (1990), Itô & Mester (2003 [1992]) and Ladd (1986, 1996). Selkirk (1996) herself admits that the Strict Layer Hypothesis is but a cover term for a heterogeneous set of four primitive component constraints. Hence Hayes' argument may well turn out to be counter-productive if four constraints are needed in order to capture the generalisations that are handled by two principles in boundary theory. In sum, the diacritic character of boundaries is certainly a serious argument against them: the fact that diacritics do not qualify is an important lesson of the historical survey that is provided in §692. Everybody is certainly called to look for a way to encode extra-phonological information in a nondiacritic fashion, and domains are the solution proposed in the early 80s. Strangely enough, though, at no point does the Prosodic Phonology literature consider the eventuality that prosodic constituents are also diacritic. It is shown in §402 below that the Prosodic Hierarchy indeed is a diacritic, if an autosegmental one (also Scheer 2008a). If this is true, we have nothing gained in the transition from boundaries to domains. And we are still left with the pending question of local vs. non-local intervention. Recall from §366 that this contrast is what I take to really distinguish boundaries and domains – precisely because they may not be told apart on the grounds of the diacritic issue. Only is the local vs. nonlocal aspect never discussed in the literature when it comes to compare the two competitors. Since the eventually diacritic character of the Prosodic Hierarchy is not discussed any more than the local vs. non-local perspective, boundaries could be disqualified by simply pointing out their diacritic character. This is

From boundaries to domains: a historical choice gone almost unnoticed 315

exactly what happened. Or, in other words, the local baby was thrown out with the diacritic bathwater (more on this in §706).

374

3.4.5. Domains have an independent motivation: stress, rhythm and musical properties Selkirk (1980a:126ff, 1984:8ff) certainly levels the most serious argument against boundaries. She argues that since Liberman (1975) and Liberman & Prince (1977) domains are independently motivated by stress, rhythm and musical aspects of speech. Boundaries, on the other hand, are unable to encode these properties. The former thus allow for a unified coverage of both interface information and stress/rhythm/musical aspects, while the latter multiplies representational devices. Let us first address stress and its associated prosodic category. For one thing, it is not the case that Liberman & Prince (1977) (or any of their followers for that matter) offer a uniform representation of stress: they build a tree based on weak/strong oppositions, but then from this arboreal structure derive the metrical grid, which is responsible for stress clash (thirtéen vs. thírteen men). Also, the unifying perspective holds only if arboreal structure is indeed the correct way of representing stress. This, however, cannot be taken for granted: Prince's (1983) grid-only approach competes with feet for the representation of stress. Still another option is explored by Szigetvári & Scheer (2005) who develop unified representations for the syllable and stress which are lateral, rather than arboreal or grid-based. Finally, it may not be a correct move to grant identical representations to things that are different in kind. Stress is a domestic property of phonology that, unlike those phonological phenomena for which prosodic domains are needed, owes nothing to extra-phonological information. Were there no interface, feet would still exist, while the five higher levels of the Prosodic Hierarchy would not. Just like syllables, feet are a bottom-up construction and not, like higher prosodic categories, objects that are created by top-down mapping. In other words, the Prosodic Hierarchy is a blend of heterogeneous phonological objects, rather than a homogeneous arboreal representation. This is pointed out by Rice (1990:292 note 3) and is actually admitted by Nespor & Vogel (1986:109) (see §401). In this light, then, it is worth wondering whether ontologically distinct categories should be represented by identical devices. Inkelas (1990) for example says no: it is not right to have properly prosodic constituency

316 Chap 10: Prosodic Phonology, on the representational side only from the word level on – the Prosodic Hierarchy ought to be extended to smaller units (hence the title of her dissertation "Prosodic Constituency in the Lexicon", see §433). Let us now have a look at rhythm and musical properties of speech. Since Hayes (1984), rhythm is regarded as an emanation of metrical poetry and music, rather than of the linguistic system. Hayes (1984:65) writes that "grids are not strictly speaking a linguistic representation at all" and concludes on page 69 that rhythm on the one hand and linguistic structure such as stress or the Prosodic Hierarchy on the other belong to separate cognitive domains. Nespor (1988:228) adds that rhythm "is, in fact, not properly a phenomenon of language, but rather of all temporally organized events." If rhythm is not a linguistic object, then, linguistic theory must not attempt at representing it. This view was followed by all subsequent work including Nespor & Vogel (1986, 1989:87f) and Selkirk (1986): rhythmic structure materialises as the metrical grid, which is produced by a secondary mapping that takes the Prosodic Hierarchy as an input (more on this in §427). It thus appears that the unification argument involves properties which according to subsequent work must not be unified. 375

3.4.6. The idea of unified representations was abandoned by Selkirk herself The unification argument was abandoned in later work by Selkirk, who reverted to distinct representations for stress and rhythm on the one hand, and morpho-syntactic information on the other. In her early work, Selkirk promoted two different means of melting domestic phonological properties and interface information. Before her 1984 book, the allround representations were autosegmental and arboreal. Selkirk (1984), however, gives up prosodic constituency in favour of Prince's (1983) grid-only approach: among the prosodic constituents that were assumed before, only the intonational phrase survives.94 The function of all other layers is taken over by the metrical grid (Selkirk 1984:29ff). Two years later, Selkirk (1986:376) joins Nespor & Vogel's (1986) perspective according to which three types of representations and distinct computational mechanisms live in peaceful coexistence above the syllable: prosodic constituency, the metrical grid (which is responsible for stress and rhythm) and Lexical Phonology (see §427). In such a landscape, the unifi94

Selkirk (1984:27f) holds that intonation is an emanation of semantics alone ("sense unit") which lacks any syntactic conditioning (see also §4250).

From boundaries to domains: a historical choice gone almost unnoticed 317

cational argument is lost: it is tacitly admitted that domestic (stress, rhythm) and non-domestic phonological information (morpho-syntactic conditions on phonological processes) is different in kind and should not be encoded in a single set of representations.

376

3.4.7. Summary The arguments that have been levelled against boundaries reduce to two items: their diacritic character and the absence of independent motivation. Discussion of the transition from boundaries to domains is scarce, and there has never been any pro-boundary party that would have responded to the critiques, or compared the putative gain when domains are used. The replacement of boundaries by domains is a major break in the history of interface theory. Rather than by the two arguments mentioned, it was motivated by the general trend towards autosegmental structure that was ambient in the late 70s and early 80s. In this context, the transmission of morpho-syntactic information could not stand back: "the junctural properties of sentences should be somehow represented 'suprasegmentally' rather than as the segmental boundaries of the standard theory", as Selkirk (1984:8) explains. The diacritic argument is certainly valid: diacritics do not qualify. However, it was taken for granted without discussion that the alternative, the Prosodic Hierarchy, is not a diacritic. §402 shows that it is: diacritic boundaries were replaced by diacritic domains, and nothing was gained on this count. The other argument may have been an issue in the early days when prosodic constituency and the metrical grid were mutually exclusive (Selkirk's work until 1984). It has disappeared when Prosodic Phonology admitted the peaceful coexistence of both types of representation (Selkirk 1986, Nespor & Vogel 1986). Finally, an important property of boundaries has gone entirely unnoticed: unlike domains, boundaries have a local action. This issue is further discussed in §706 below.

318 Chap 10: Prosodic Phonology, on the representational side 377

4. The heart of Prosodic Phonology: Indirect Reference and its consequences

378

4.1. Introduction The following pages introduce Prosodic Phonology from the inside: its founding statement (Indirect Reference), the consequences thereof (the Translator's Office, mapping rules, the Prosodic Hierarchy), and how the various devices interact technically. The principle of Indirect Reference holds that morpho-syntactic categories are invisible to the phonology: phonological processes cannot make direct reference to them. Rather, morpho-syntactic structure needs to be translated into phonological vocabulary, the Prosodic Hierarchy, to which phonological processes may then appeal. Indirect Reference is present since the earliest incarnations of Prosodic Phonology, and germane to all of its versions. Theories that disagree with it are called direct syntax approaches (on which more in §407). If it is true that phonology has no direct access to morpho-syntactic structure, higher level information needs to be transformed into items that are part of the phonological world. This translation is done by so-called mapping rules, whose input is the morpho-syntactic structure (and also what I call the black box, on which more below). They return the Prosodic Hierarchy, which lies inside the phonological module. I call this output the buffer because its only purpose is to take stock of morpho-syntactic information, and to restore its load inside the phonology. The buffer is therefore the instrument of Indirect Reference: phonological rules that are sensitive to morpho-syntactic information make reference to the buffer, rather than to morpho-syntactic categories themselves. What the buffer is made of and how exactly translation works are secondary questions.95 What is important to understand for the time being 95

The latter is discussed in the following section and in §387. Regarding the former, the content of the buffer indeed is not pre-empted by the overall architecture of Prosodic Phonology. Even though arboreal structure in general and Selkirk's (1981 [1978]) prosodic categories in particular have rapidly become the only kind of buffer used, other implementations are possible. Selkirk (1984) for instance abandons all arboreal prosodic structure except the intonational phrase. Instead, following Prince's (1983) grid-only approach, she holds that the content of the buffer is the metrical grid. That is, mapping rules produce units of the grid or modify existing grid structure; phonological rules then make reference to the grid.

The heart of Prosodic Phonology: Indirect Reference and its consequences 319

is that the existence of the buffer is the heart of Prosodic Phonology – and that the buffer is the consequence of Indirect Reference.

379

4.2. The buffer, its construction workers and how it unloads its goods

380

4.2.1. Mapping rules and the black box Table (140) below shows the place of the buffer in the overall architecture of Prosodic Phonology. (140) the general architecture of Prosodic Phonology Morpho-Syntax

Interface mapping rules

?

Black Box

Phonology

the buffer: prosodic constituency

phonological rules that are sensitive to morphosyntactic information make reference to the buffer x x x x x x x x x x x x x x x

320 Chap 10: Prosodic Phonology, on the representational side The buffer is created by mapping rules, which do their job on the grounds of two sets of information: the output of morpho-syntax and what I call the Black Box. The existence of the Black Box is critical for Prosodic Phonology: it is the consequence of so-called non-isomorphism, the (only) argument that grounds Indirect Reference (and hence the buffer). Non-isomorphism is discussed at length in §§416ff: following SPE, Prosodic Phonology holds that the domains to which phonological rules make reference sometimes do not match any morpho-syntactic division: phonologically relevant and morpho-syntactic domains may be nonisomorphic. Thus the output of morpho-syntax sometimes is not ready for phonological use. It therefore needs to be modified according to phonological demands (in SPE, this job was done by the so-called readjustment component, see §91). Mapping rules transform morpho-syntactic structure into prosodic constituency, which is legible by phonological rules. However, they do not simply replicate morpho-syntactic structure in a different (prosodic) vocabulary. On the one hand, not all properties of morpho-syntactic structure are phonologically relevant. Mapping rules therefore impoverish the morpho-syntactic output. On the other hand, non-isomorphism enforces the rearrangement, eventually the augmentation of morpho-syntactic structure. Mapping rules thus work on the output of morpho-syntax, but also make sovereign readjustment decisions – it is the latter that I call the Black Box. This label expresses the mystery that continues to be associated with readjustment decisions up to the present day: linguists have a poor understanding of the particular groupings of morpho-syntactic units that is relevant for the phonology. This issue was referred to as the mapping puzzle above (§111); it is further discussed in §§387,392f,396. 381

4.2.2. Mapping is done in modular no man's land In a modular perspective, mapping and the Black Box cannot be located either in morpho-syntax or in phonology. Mapping rules have access to both morpho-syntactic and phonological structure: they can read the former, and create part of the latter (the Prosodic Hierarchy). Since modules do not understand the language of other modules, the interface is necessarily located in modular no man's land, i.e. somewhere after morpho-syntax and before phonology. This is indicated under (140), but deserves to be made explicit (see §413).

The heart of Prosodic Phonology: Indirect Reference and its consequences 321

The development of modular theory in the early 80s (Fodor 1983) was contemporary to Prosodic Phonology. Quite surprisingly, then, the Prosodic Phonology literature does not mention the modular issue and the fact that the architecture of Prosodic Phonology is a faithful implementation of modular principles (see §414). Although Nespor & Vogel (1986) are not explicit on the location of the mapping activity throughout their book, their final diagram on page 302 places mapping rules both outside of morpho-syntax and outside of phonology, thereby confirming their neutral status. 382

4.2.3. The layers of the Prosodic Hierarchy Since Selkirk (1981 [1978]) and up to the present day, the units of the Prosodic Hierarchy have been remarkably stable. Table (141) below shows Selkirk's initial inventory. (141) the Prosodic Hierarchy according to Selkirk (1981 [1978]) phonological utterance (U) | intonational phrase (IP) | phonological phrase (φ) | phonological word (ω) | foot (Σ) | syllable (σ)

sentence intonational chunk NP, VP, AP word

Variation concerns the number and label of prosodic constituents: the phonological word is sometimes called prosodic word, and additional layers that have been proposed are moras and the clitic group (intermediate between the prosodic word and the prosodic phrase). 96 As may be seen, the Prosodic Hierarchy parallels the relevant morpho-syntactically defined chunks of the linear string: the phonological ut96

The clitic group, proposed by Hayes (1989 [1984]) and Nespor & Vogel (1986), was challenged early on (e.g. Inkelas 1990, Zec 1993, Booij 1995, 1996, Monachesi 1996, Peperkamp 1996), and today has quite consistently disappeared from the inventory of prosodic constituents.

322 Chap 10: Prosodic Phonology, on the representational side terance roughly corresponds to one (or several) sentences, the phonological phrase is about of the size of an XP, and the prosodic word encompasses a regular word. No direct equivalents may be quoted for the intonational phrase, feet and syllables, but they are also inserted in the hierarchy according to their size. What exactly belongs to a specific prosodic constituent is subject to language-specific variation: a prosodic word for example may or may not include prefixes, and compounds may or may not count as one prosodic word. These choices are managed by mapping rules, which convert morpho-syntactic into prosodic structure. Their parameterisation is discussed at greater length in §388 below.

383

4.2.4. Geometric properties of the Prosodic Hierarchy: the Strict Layer Hypothesis The Prosodic Hierarchy is a fairly unconstrained arboreal structure, comparable to feature geometric trees below the segmental level. That is, restrictions on branching of the X-bar kind do not apply. A prosodic tree may host nodes that do not branch or that branch two, three or more times. Under (142) below appears an example of a fully prosodified English sentence taken from Hayes (1990:86).97 (142) complete prosodification of an English sentence U IP | φ | C ω | On

97

ω | Wednesday,

IP φ C ω | he

ω | told

φ | C

C ω | the

ω | stories

ω | to

ω | the

ω | children

Note that the prosodification does not include the two lowermost layers of the Prosodic Hierarchy, syllables and feet. This is because they are bottom-up constructions to which mapping rules do not contribute anything (see §401). Abbreviations: U (phonological utterance), IP (intonational phrase), φ (phonological phrase), C (clitic group, see note 96), ω (phonological word).

The heart of Prosodic Phonology: Indirect Reference and its consequences 323

The only well-formedness condition on prosodic constituency is laid down in the Strict Layer Hypothesis (SLH), which was originally formulated by Selkirk (1981) (the label Strict Layer Hypothesis only appears in Selkirk 1984:26f).98 It expresses the idea that a prosodic constituent of a given layer only dominates constituents of the immediately lower level, and is exhaustively contained in a constituent of the immediately higher level. Hence an IP for example is always the daughter of an U, and only dominates φs. In the environment of OT, Selkirk (1996:189ff) has factored out the SLH into four more primitive component constraints, which appear under (143) below (more on this in §461). (143) the Strict Layer Hypothesis cut into pieces a. Layeredness a node of layer n can only dominate a node of layer n+1, and can only be dominated by a node of layer n-1. b. Headedness each node of layer n must dominate at least one unit of layer n-1. c. Exhaustivity association lines may not bypass any layer: no association of two units that belong to non-adjacent layers is allowed. d. Nonrecursivity nested structures are prohibited: no node may dominate a node of the same label.

Selkirk (1996) believes that (143a,b) are universal (undominated in terms of OT), while (143c,d) may or may not be observed by languages. Finally, it is noteworthy that a by-product of the SLH is nonprivativity in the sense of §79 (see also §400 and the summary in §756): all strings are exhaustively parsed at all prosodic levels independently of whether there is any associated phonological effect or not (Hayes 1989 [1984]:220, 1990:86, Nespor & Vogel 1986:11). That is, the full prosodic arborescence is always constructed, and all expressions in all languages have the same number of levels of embedding.

98

Literature on the SLH abounds; it includes Nespor & Vogel (1986:7ff), Itô & Mester (2003 [1992]) and Selkirk (1996).

324 Chap 10: Prosodic Phonology, on the representational side 384

4.2.5. Three ways of making reference to the Prosodic Hierarchy Finally, let us look at how morpho-syntactic information is restored in the phonology. The structural description of phonological rules may make reference to the Prosodic Hierarchy in three different ways (Selkirk 1980a:111f, Vogel 1985, Nespor & Vogel 1986:77). The three types of rules are identified under (144) below. (144) rule types in Prosodic Phonology a. domain span rule b. domain juncture rule c. domain limit rule

A domain span rule applies in a certain segmental environment iff this environment is contained within a given domain. Domain juncture rules apply when some part of the segmental string needs to be adjacent to a boundary that separates two items of the same prosodic unit. For example, English aspiration has been said to occur "foot-initially" (e.g. Nespor & Vogel 1986:90ff). As indicated by their name, domain juncture rules endorse the function of regular linear boundaries: a specific morpho-syntactic division is targeted. Finally, domain limit rules apply when certain segmental conditions are met, but only at the beginning and at the end of a given unit of the prosodic constituency.

385

4.3. The old and the new: mapping rules and Indirect Reference When looking at the evolution of generative interface theory, it appears that all properties of the Prosodic Phonology architecture have been formulated by Chomsky & Halle (1968). In SPE, mapping is done by a fixed algorithm that transforms morpho-syntactic structure into clusters of boundaries (see §90). This job is done by mapping rules in Prosodic Phonology. The readjustment rules of SPE (see also Langendoen 1975) appear as the Black Box under (140). Also, the idea that phonology does not make direct reference to the output of the syntactic component originates in SPE. The following quotation is repeated from §82 for convenience (see also Dresher's 1996:41ff overview article on this heritage).

The heart of Prosodic Phonology: Indirect Reference and its consequences 325 (145) "We have two concepts of surface structure: input to the phonological component and output of the syntactic component. It is an empirical question whether these two concepts coincide. In fact, they do coincide to a very significant degree, but there are also certain discrepancies. These discrepancies […] indicate that the grammar must contain certain rules converting the surface structures generated by the syntactic component into a form appropriate for use by the phonological component. In particular, if a linguistic expression reaches a certain level of complexity, it will be divided into successive parts that we will call 'phonological phrases', each of which is a maximal domain for phonological processes. […] It appears that the syntactic component of the grammar generates a surface structure Σ which is converted, by readjustment rules that mark phonological phrases and delete structure, to a still more superficial structure Σ'. The latter then enters the phonological component of the grammar." Chomsky & Halle (1968:9f)

Of course it is true that Indirect Reference is more than just SPEreadjustment, which in the Prosodic Phonology architecture corresponds to the Black Box. Indirect Reference bans phonological rules that make direct reference to morpho-syntactic structure and labels – this kind of rule was common practice in SPE (see §95). As will be shown below (§410), Indirect Reference is thus a real progress from the modular point of view. This being said, the Prosodic Phonology literature often tries to insulate the new perspective from SPE so radically that wrong or incomplete statements are made. For example, I was unable to find any reference in the Prosodic Phonology literature that establishes the heritage relation between the readjustment rules of SPE and Indirect Reference. Readjustment rules are sometimes mentioned (regularly in the work by Selkirk), but never as the ancestor of Indirect Reference.99 Worse, SPE is sometimes presented as a system where only direct reference to morpho-syntactic structure is made. (146) "According to SPE, phonological rules apply to the linear surface structure of a sentence, that is, to the output of the syntactic rules." Nespor & Vogel (1982:225)

99

Significantly, this connection is made explicit by an opponent of Prosodic Phonology: arguing for the direct syntax approach (see §4070), Kaisse (1985:110 note 1) asks whether "the rules building up metrical structure are or are not significantly different from readjustment rules."

326 Chap 10: Prosodic Phonology, on the representational side "In traditional generative theory it was supposed that these [morphosyntactic] domains directly correspond to syntactic constituents (see Chomsky & Halle, 1968)." Nespor & Vogel (1986:37) "In an SPE-type model of phonology, the only way of representing the domains of a phonological rule is in terms of morphosyntactic constituents, the implicit claim being that such constituents are, in fact, the only domains in which phonological rules may apply." Vogel (1986:59)

In the first two quotes, Nespor & Vogel forget to mention the fact that there was translation in SPE: morpho-syntactic surface structure was transformed into phonological boundaries (§90). This translational process is exactly equivalent to mapping, except that the output categories are different: boundaries in SPE, prosodic domains in Prosodic Phonology. Vogel's (1986) statement is simply counterfactual: she seems to have overlooked the existence of readjustment in SPE.

386

5. Mapping: its mechanics, its evolution and our understanding thereof

387

5.1. Introduction: the mapping puzzle (again) How exactly does the mapping mechanism work? Is it possible to establish cross-linguistic generalisations that tell us which portions of the morphosyntactic structure are phonologically relevant, and which ones are not? This aspect of the interface – the mapping puzzle – is certainly the most poorly understood of all. For one thing, as Kaisse & Zwicky (1987) put it, (147) "the study of postlexical rules sensitive to syntactic or prosodic structure is still in its infancy. Phonologists know comparatively little about the range of phenomena that can be encompassed by such rules, compared for instance with what is known about word-internal phonological processes or rules of syntax. At the moment theories must be advanced on the basis of data that are, from the language-internal point of view, rich and complex, but are also, from the cross-linguistic point of view, sparse and diverse." Kaisse & Zwicky (1987:4)

Odden (1987) and Ladd (1986:311) make the same observation. But also, the data that linguists are aware of seem to project a picture of relative anarchy: it was noted in §§111f,379,388ff (see also the summary of the mapping puzzle in §753) that the morpho-syntactic contexts which trigger

Mapping: its mechanics, its evolution and our understanding thereof 327

or block phonological processes on many occasions do not make sense (morpho-syntactically speaking). In many cases the only thing that linguists are able to do is to write down an amorphous list of triggering and blocking boundaries. The practice of Prosodic Phonology, then, is to equate the size of the chunks that are delineated by blocking boundaries to some level of the prosodic constituency. In the next step, a mapping rule is written that specifies which parts of the morpho-syntactic structure are grouped together in order to build the phonologically relevant chunk of the linear string. This chunk is then called a unit of the Prosodic Hierarchy (e.g. a phonological phrase). The mapping job, thus, amounts to boundary grouping. Why a mapping rule groups this or that chunk of the linear string rather than others remains unanswered in most cases. In actual fact, the literature hardly asks the question in the first place. Nobody knows whether the mapping mechanism is restricted in any principled way, or according to which rationale mapping decisions are made. Another aspect of the mapping puzzle is the fact that cross-linguistic generalisations do not appear to emerge. The evolution of mapping from the earliest sources of Prosodic Phonology up to the present day has progressively made allowance for a highly variable empirical picture. That is, the mapping literature went through an evolution from firm universal statements with little or no tolerance for parametric variation to the recognition of a bewildering variety of language-specific situations. Quite frustratingly, the availability of more and more data seems to make the overall picture mistier and mistier, instead of clearing it up. The above quote from Kaisse & Zwicky (1987:4) provides good testimony of this situation. The following pages attempt to document this evolution, that is both aspects of the mapping puzzle that were mentioned: the growing parametric variation that goes hand in hand with the absence of cross-linguistic generalisations, and (the absence of) a morpho-syntactic rationale that governs mapping decisions. The former issue is addressed in §388, the latter is discussed in §394. To round off the discussion of mapping, §398 discusses the observation that phonology may react on geometric properties of morpho-syntactic structure, but not on its labels.

328 Chap 10: Prosodic Phonology, on the representational side 388

5.2. Mapping and its inflational evolution

389

5.2.1. Early mapping in Selkirk's work until her 1984 book Recall from §90 that SPE did mapping on the grounds of a universal algorithm that does not admit any parametric variation: #s are inserted at both edges of major categories as well as of XPs, before the entire string is shipped off to the phonology. By convention, clusters of hashmarks are then reduced to a maximum of two. In Selkirk's first article, the genesis of the phonological word is not addressed (Selkirk 1981 [1978]:124f). The phonological phrase is constructed according to the following instructions, which Selkirk believes to be the "major principles". (148) "The Phonological Phrase: Constituency (i) An item which is the specifier of a syntactic phrase joins with the head of the phrase. (ii) An item belonging to a "non-lexical" category (cf. Chomsky 1965), such as Det, Prep, Verbaux, Conjunction, joins with its sister constituent." Selkirk (1981 [1978]:126)

Two years later, Selkirk (1980b:580) still writes that "the conception of mapping between syntax and phonology is an eminently simple one." Nespor & Vogel (1982) make the following statement. (149) "[W]e propose a series of mapping conventions for Italian which we will then argue are sufficiently general to be able to account for any X-bar type language once the values of certain syntactic parameters have been assigned." Nespor & Vogel (1982:227)

In her 1984 book (where the buffer is made of metrical grids, rather than prosodic constituents, see §426), Selkirk proposes that mapping boils down to four universal classes of rules: (150) "The theory of metrical grid construction has two major components. The first is a set of text-to-grid alignment (TGA) rules. Through these rules, the particular properties of the text may impose requirements on the rhythmic realization of the sentence. […] Quite certainly, one of the major descriptive tasks for a metrical grid theory of stress is to characterize the TGA rules available to language. As we will show, it seems that these rules fall into just four classes." (Selkrik 1984:54)

Mapping: its mechanics, its evolution and our understanding thereof 329

Some of these four classes enclose further possible parameters (see note 92). Hence the realm of the mapping system has already expanded quite significantly when compared to Selkirk's first article. 390

5.2.2. Parametric variation of ω and φ in Nespor & Vogel (1986) Let us now look at the variation regarding the construction of the phonological word and the phonological phrase that Nespor & Vogel (1986) initially provide for (see also Nespor & Vogel 1983:124f, as well as the variation acknowledged by Hayes 1989 [1984]:206ff and Gvozdanović 1986:40ff). According to their typology (p.117ff), phonological words may be coextensive with the constituent dominated by the terminal node of the morpho-syntactic tree, i.e. encompass the stem and all affixes as well as all elements of compounds. According to Nespor & Vogel's analysis, this is the case in Latin and Modern Greek. Another option for the phonological word is to not exactly mirror morpho-syntactic structure. In this case, two groupings are observed: the phonological word may be formed by the stem and all affixes, while compounds represent two separate units: stem one with all prefixes, and stem two with all suffixes (Sanskrit, Turkish). The alternative grouping is as before, except that in compounds the prefixes of stem one form a separate domain (Hungarian). The cross-linguistic variation that Nespor & Vogel provide for is thus threefold as far as the phonological word is concerned. It is expressed by language-specific mapping rules which decide, for each level of the Prosodic Hierarchy, which units of the morpho-syntactic input are grouped together. As a general measure, Nespor & Vogel (1986:5) argue that the degree of universality is proportional to the rank in the Prosodic Hierarchy: crosslinguistic variation is most important for the phonological word (mapping rules do not intervene in the definition of syllables and feet, which are bottom-up constructions, see §401) and decreases as we move up in the prosodic tree: the phrasing of the phonological utterance, then, is almost universal.

330 Chap 10: Prosodic Phonology, on the representational side Accordingly, the variation that Nespor & Vogel (1986) have found for the phonological phrase is as follows.100 (151) "Phonological Phrase Formation I. φ domain The domain of φ consists of a C which contains a lexical head (x) and all Cs on its nonrecursive side up to the C that contains another head of the maximal projection of X. II. φ construction Join into an n-ary branching φ all Cs included in a string delimited by the definition of the domain of φ." Nespor & Vogel (1986:168)

391

5.2.3. More variation for the phonological phrase Subsequent studies have shown, however, that there is much more crosslinguistic variation than that. Cho (1990) for example considers a phonological process in Korean whereby obstruents voice in intervocalic position depending on the type of adjacent boundary.101 He has found the following triggering and blocking environments (Cho 1990:48ff). (152) in Korean, an underlying word-initial after a vowel-final word is voiced in the context determiner - noun adjective - noun possessive noun - noun (John's book) verb of a relative clause - noun object - verb verb - verb

plain voiceless obstruent that occurs remains voiceless in the context subject - verb subject - object object - object conjunction - noun topic-NP - S' (apples, they throw away) subject - sentential adjective subject - sentential adverb

On these grounds (and also considering additional data that do not need to be reproduced here), Cho concludes that the mapping rule under

100 101

"C" refers to the clitic group, i.e. the prosodic layer immediately below the phonological phrase in Nespor & Vogel's system (see note 96). Korean has three contrastive sets of obstruents, so-called plain, aspirated and tense. Only the former undergoes the process at hand.

Mapping: its mechanics, its evolution and our understanding thereof 331

(153) below is needed in order to correctly phrase the morpho-syntactic output. (153) mapping rule for phonological phrases in Korean (Cho 1990:57) apply the following rules cyclically to all maximal projections, proceeding from the bottom up. At any given stage (a) applies before (b). Let the maximal projection under consideration on a given cycle be M. a. if M branches, combine the head of M into a phonological phrase with all adjacent unphrased material, up to and including the closest XP, or if no such phrase is present, the left edge of M. b. phrase any focused word with the next word, unless that word is already phrased. After (a) and (b) have applied in all possible environments, (c) applies. c. unphrased words form phonological phrases of their own.

The mapping operation for phonological phrases in Korean is thus clearly more complicated than what was foreseen in the early days of Prosodic Phonology.

392

5.2.4. Cross-linguistic atomisation of mapping Detailed studies of the mapping mechanism in other languages have produced similar results: the variational space is atomised, and every language seems to have its own idiosyncratic way of grouping morpho-syntactic divisions into prosodic constituents. Examples are Neijt (1985, Dutch), Cowper & Rice (1987, Mende), Vogel (1988, Hungarian), Vogel (1990), Condoravdi (1990, Greek), Poser (1990, Japanese), Bickmore (1990, Kinyambo, Bantu), Kidima (1990, Kiyaka, Bantu), Peperkamp (1995, Italian), Selkirk (2000, English), Kanerva (1990, Chichewa, Bantu), the latter focusing specifically on the issue regarding the unpredictability of the mapping mechanism. The most ambitious attempt at boxing the mapping mechanism on universal grounds is certainly Selkirk's (1986) end-based (or edge-based) model, which is presented in §§395f. Unfortunately, it does not really stand up to the evidence that was produced in subsequent years. Bickmore (1990) for example, who adheres to Selkirk's end-based perspective, concludes that (154) "we are still in search of a parameterized cross-linguistic phonological phrase-construction rule with descriptive adequacy (let alone explanatory adequacy)." Bickmore (1990:17)

332 Chap 10: Prosodic Phonology, on the representational side Clearly, rigid and universal mapping rules that were proposed in the early days of Prosodic Phonology are unable to describe the highly variable cross-linguistic picture. The same holds true for the mapping mechanisms suggested by Nespor & Vogel (1986) where timid parametric elements appear. Finally, Selkirk's (1986) end-based model does not offer significant progress on this front either.102 Building on Selkirk (1986), OT has later expressed mapping in terms of constraint interaction (ALIGN and WRAP, see §457), which opens a larger parametric space: under the pressure of more and more empirical variation, Selkirk (1996) for example cuts down the Strict Layer Hypothesis (§383) into four independent constraints, two of which may be violated (see §461). In sum, the claim that mapping rules define a reasonable amount of cross-linguistic variation which is marshalled by universal properties of prosodic constituency does not stand up to fact: it is fair to say that still today linguists simply do not understand how mapping works. All that they can do is record amorphous lists of blocking and non-blocking environments that are produced by empirical observation. Therefore, the frustration that is generated by the mapping puzzle has not calmed down since the late 70s, in spite – or perhaps because – of the fact that the Prosodic Phonology literature has produced an impressive amount of new empirical wealth. As a reaction, the literature has rather turned away from the study of cross-linguistic variation in morphosyntactically conditioned prosodic phrasing. Further discussion of this move is provided in §463. 393

5.2.5. The mapping puzzle is sometimes hidden by the clean Prosodic Hierarchy The highly variable and uncontrolled mapping mechanism contrasts with the uniform Prosodic Hierarchy: Prosodic Phonology indeed aims to en102

Kanerva (1990:150f) and Bickmore (1990:3ff) provide an interesting overview of the various attempts that have been made to define the phonological phrase. Quite surprisingly, Inkelas & Zec 1995:539) agree that the mapping mechanism is by and large mysterious for intonational phrases and phonological utterances, but claim that the two smaller units, the phonological word and the phonological phrase, "exhibit impressive cross-linguistic similarities; […] the attested range of variation appears sufficiently small to be viewed as parametric in nature." This is not exactly what one would conclude from the inspection of the relevant literature.

Mapping: its mechanics, its evolution and our understanding thereof 333

code the non-phonological information that is used in phonology in terms of a universal arboreal structure: the buffer is the same in all languages (Nespor & Vogel 1986:11). That the clean and universal Prosodic Hierarchy is the result of mushy mapping activity, however, is sometimes left unmentioned in the Prosodic Phonology literature. This is when prosodic constituency is meant to appear as a maximally simple, constrained and unified theory which, unlike direct syntax approaches and SPE, does not need to recur to flourishing devices that get out of hand. One example is Nespor & Vogel's (1986) presentation of intervocalic s-voicing in Italian (ISV). They write that (155) "we can formulate ISV as a ω span rule in a maximally simple way as seen in (64), whereas the expression of the domain of application of this rule in terms of morphological constituents would not amount to more than a list of disparate environments." Nespor & Vogel (1986:129)

Nespor & Vogel forget to mention that in order to achieve the maximally simple structural description of the ω span rule, they first need to create the ω by a maximally complicated and unnatural mapping rule which transforms the list of disparate environments into an ω. More of the same is found in Booij (1985a). (156) "This theory [Prosodic Phonology] is superior to both standard and natural generative phonology in its approach to the influence of morphological boundaries on phonological processes. […] It excludes the rather arbitrary use of boundaries made possible in the SPE-framework." Booij (1985a:34)

It is certainly true that SPE is much too powerful a system and wildly overgenerates – this has been spotted in the post-SPE period and actually by Chomsky & Halle (1968) themselves in their ninth chapter. However, Booij does not mention that the clean prosodic guise was built by an entirely unconstrained construction worker: mapping rules can do anything its reverse. They are able to group any two adjacent objects into a prosodic constituent and hence overgenerate just as much as the SPE formalism. The embarrassment regarding the mapping puzzle is thus as real in Prosodic Phonology as it has always been – perhaps even more pressing since nothing was gained by the fresh data that came in. The only thing that has changed is the vocabulary that is used for the description of the embarrassment: mapping rules now, the absence of natural classes of boundaries then.

334 Chap 10: Prosodic Phonology, on the representational side 394

5.3. What is the morpho-syntactic rationale behind mapping?

395

5.3.1. Selkirk (1986) puts to use the technology of the 80s: X-bar-based mapping This situation is the starting point of Selkirk (1986:384): "up to now there has been no general theory of the mapping between syntactic structure and prosodic structure." On this backdrop, Selkirk develops a mapping mechanism that significantly restricts the availability of morpho-syntactic structure. Her system is called end-based mapping (later often called edge-based theory), and Selkirk (1986:388, 400) suggests that it is universal. That is, end-based mapping is supposed to account for all possible mappings between morphosyntactic and phonological structure. In subsequent work, Selkirk's mapping mechanism was applied to a number of languages: Hale & Selkirk (1987), Cowper & Rice (1987), Selkirk & Tateishi (1988), Selkirk & Shen (1990). Dresher (1994:16ff) and Elordieta (2007:137ff, 2008:245ff) provide an informed overview of this project. In more recent minimalist times Elordieta (1997, 2007:157ff, 2008:255ff) entertaines an alternative that is based on the idea that feature checking expresses primitive grammatical relationships (morphosyntactic feature chains), which define mapping relations. Most notably, however, it is Selkirk's focus on edges that has given rise to the dominant mapping mechanism when Prosodic Phonology was transported into Optimality Theory: McCarthy & Prince (1993, 2001:vii) are explicit on the fact that alignment (on which more in §455) is directly inspired by Selkirk (1986). Building on Chen (1985), Selkirk's end-based theory holds that "the syntax-phonology mapping can be defined simply by reference to the ends of syntactic constituents" (Selkirk 1986:386, emphasis in original). Assuming X-bar theory and the irrelevance of morpho-syntactic labels (i.e. whether an X-bar structure is headed by a verb, a noun or some other lexical category, see §398), only the segmentations under (157) are allowed for. (157) possible phonologically relevant segmentations of the linear string according to Selkirk's (1986) end-based theory left end right end result on the phonological side a. Word[ ]Word phonological word b. Xhead[ ]Xhead "small phonological phrase" c. Xmax[ ]Xmax phonological phrase

Mapping: its mechanics, its evolution and our understanding thereof 335

That is, each node of the X-bar structure may represent a division that is phonologically relevant. Or, looked at from the phonological perspective, phonologically relevant divisions of the linear string necessarily correspond to and end of one of the three X-bar levels. Whether the left or the right end determines the segmentation depends on a parametric choice. According to Selkirk (1986), chunks between two ends of X0 items (what she calls the word level, i.e. (157a)) are prosodic units known as the phonological word. On the other end of the scale, the chunks located between the two ends of Xmax domains are phonological phrases. Given X-bar structure, Selkirk must also come up with a phonologically relevant intermediate chunk that is determined by the ends of Xhead. This prosodic unit, Selkirk (1986:394) claims, "has been discerned in a variety of languages. It is one which includes the head of a phrase and preceding or following specifier material, but not any complements (arguments) to the head" (also, functional words do not count as words). She calls this unit the "small phonological phrase", for which she provides evidence from French liaison. Selkirk's end-based theory thus only concerns the middle area of the Prosodic Hierarchy: the phonological utterance and the intonational phrase, as well as feet and syllables, are not considered. This is because Selkirk believes that intonational phrases are semantically defined (rather than morpho-syntactically): her position on this issue has not varied since Selkirk (1984:27f). Syllables and feet, on the other hand, are not part of the Prosodic Hierarchy for the reasons exposed in §401.103

396

5.3.2. End-based mapping does not solve the mapping puzzle either Selkirk thus claims to have found the key to the mapping-puzzle: all amorphous lists of triggering or blocking boundaries fall into place when looked at through the prism of (157). A reanalysis of vowel length in Chi Mwi:ni (Bantu, originally studied by Kisseberth & Abasheikh 1974) shows how her system works: compare Hayes' (1989 [1984]) formulation of the relevant mapping with Selkirk's (1986:397) under (158) below.

103

Against her earlier work, Selkirk (1984:29ff) held that mapping rules produce (or modify) the metrical grid, rather than prosodic constituency (see §426). In Selkirk (1986), she reverts to the Prosodic Hierarchy in its middle area, but continues to believe that stress is encoded by the metrical grid (and hence that feet do not exist: Selkirk 1986:385).

336 Chap 10: Prosodic Phonology, on the representational side (158) mapping in Chi Mwi:ni a. relation-based (Hayes 1989 [1984]:211) 1. in [X0Y''…]X'', where X0 is the head of X'' and Y'' is an adjacent complement, the sequence X0Y'' forms a P-phrase. 2. all clitic groups unaffected by 1) form P-phrases. b. end-based (Selkirk (1986) ]Xmax that is, a new phonological phrase begins after the last word of each maximal projection.

The gain of simplicity is indeed obvious. Selkirk (1986) thus levels her end-based approach against the usual way of grouping morphosyntactic units into prosodic constituents. She calls the latter relation-based because it invokes the linear order of constituents, as opposed to her mechanism, which does not appeal to relations between successive constituents in surface syntactic structure. Of course, the question is whether Selkirk's end-based mapping really stands up to its universal ambition – an empirical question. Some successful reanalyses in the end-based perspective notwithstanding,104 the answer is clearly negative: Selkirk's X-bar-based mechanism cannot reduce the flourishing lists of amorphous triggering and blocking boundaries (this issue was already discussed in §392). In most cases, mapping remains as obscure as before. Chen (1990) for example demonstrates this fact. He ends up arguing that phonology needs both end-based and relation-based mapping. In sum, then, Selkirk's (1986) ambitious programme may not have solved the mapping puzzle; its merit, however, is to have carried the mapping question to the forefront of the scene. Or, in other words, the new Xbar technology of the 80s has not prompted a significant advance over the state in which the mapping puzzle was left in the post-SPE literature (see §111): natural classes of boundaries do not emerge.

104

Thanks to the end-based mechanism, Cowper & Rice (1987) for example are able to reanalyse certain data from Kimatuumbi (Bantu) according to regular prosodic constituency. These data had been presented by Odden (1987) in support of the direct syntax approach.

Mapping: its mechanics, its evolution and our understanding thereof 337 397

5.3.3. Boundaries are the relevant descriptive currency Selkirk's (1986) article also shows that boundaries are the relevant units for the description of morpho-syntactic contexts that trigger or block a phonological process. "End-based" indeed means "at a given boundary": phonologically relevant chunks of the linear string are defined by a local transition from some morpho-syntactic category to another. Kaisse & Zwicky (1987) put it as under (159) below. (159) "Reference to constituent edges also plays a central role in end-based approaches to prosodic domain determination […]. These approaches represent a return in spirit to the word-boundary theory of Chomsky & Halle (1968) and Selkirk (1972), which held that only an impoverished subset of the information provided by syntax is available to phonology." Kaisse & Zwicky (1987:10)

That is, linguists do not know why boundaries are grouped in the way they are, but it is safe to say that boundary grouping is the heart of the mapping mechanism: relevant generalisations are expressed in terms of local transitions, rather than in reference to domains.

398

5.4. Phonology can see morpho-syntactic structure, but not its labels Unlike the unedifying situation regarding the mapping puzzle, another property of the mapping process has appeared quite clearly in the Prosodic Phonology literature. It concerns the kind of morpho-syntactic information that phonology may use: the tree geometry (immediate constituent hierarchy) is often relevant for phonological processes, while the particular labels of the nodes do not appear to play any role. In other words, phonology sees morpho-syntactic structure, but not its content. The observation that category membership is irrelevant to the phonology goes back to Selkirk (1974) and Rotenberg (1978:111ff). It is discussed by, among others, Kaisse & Zwicky (1987:7), Hayes (1989 [1984]:205, 1990:87), Inkelas & Zec (1990a:xiii) and Chen (1990). Chen (1990) actually argues against this generalisation: he presents evidence from Chinese (Xiamen) tone sandhi to the end that whenever "tree-geometry and functional relations make different predictions about phonological phrasing, it is the latter that prevails" (Chen 1990:45). Namely, he identifies the contrast between arguments and adjuncts as a phonologically relevant distinction. Some years later, however, Bao (1996)

338 Chap 10: Prosodic Phonology, on the representational side proposes a reanalysis of the data whereby phonology does not need to look at the distinction between arguments and adjuncts. In sum, the selective visibility of morpho-syntactic information for phonology is an interesting generalisation. Prima facie counter-examples such as English stress assignment which depends on whether the item is a noun (récord) or a verb (recórd), however, need to be addressed. This issue is discussed at greater length in §752. Also, we will see in §653 that the generalisation becomes meaningful beyond the descriptive level in a modular environment: modules are domain specific, which means that they use different vocabulary – they speak different languages (of the mind). Therefore the vocabulary of a given module (or projections thereof, i.e. node labels) cannot be read by other modules, while structure (the tree itself) is not handicapped for intermodular communication. Finally, we will see in §660 that the generalisation at hand ties in with similar observations from other areas and may turn out to be a piece of a more general property of how morpho-syntax and phonology communicate.

399

6. Closer inspection of the buffer (Prosodic Hierarchy): what it is and what it is not

400

6.1. The only purpose of the Prosodic Hierarchy is the storage of morphosyntactic information The Prosodic Hierarchy is process-defined: prosodic domains exist only because there are phonological processes that make reference to them. Selkirk (1986:385) is explicit on this fact: "constituents like prosodic words or higher seem to have no role in the phonology other than the providing of domains." A common formulation therefore is "prosodic constituents define the domains of application of phonological rules" (e.g. Hannahs 1996:52). In turn, this means that there is only one way to detect a domain: the units of the Prosodic Hierarchy are those, and only those, that constitute the domain of application of a phonological rule. Nespor & Vogel (1986) put it the following way.105 105

Nespor & Vogel (1986:59) also mention two other ways of establishing a phonological constituent: when a given string is the locus of phonotactic restrictions, and when it expresses relative prominence relations. These two criteria may be safely disregarded in the context of the interface since they concern the

Inspection of the buffer (Prosodic Hierarchy): what it is and what it is not 339 (160) "A string is considered a constituent in phonology, as in syntax, if a) there are rules of the grammar that need to refer to it in their formulation, or b) there are rules that have precisely that string as their domain of application." Nespor & Vogel (1986:59)

Hence prosodic domains exist only because they carry extraphonological information that is phonologically relevant. It is true that the Prosodic Hierarchy has also been used for purposes of domestic phonology that have no relation with the interface.106 This is a different debate, though: once prosodic constituency is created, it may be used – but its exclusive purpose is the storage of morpho-syntactic information; also, no piece of phonological information ever contributes to its construction (see §§405,421). Finally, it is noteworthy that even though phonological processes are the only evidence for prosodic constituents, all levels of the Prosodic Hierarchy are always constructed in all languages – even in absence of any phonological consequence (e.g. Hayes 1989 [1984]:220, Nespor & Vogel 1986:11, see also §383). Like SPE and Lexical Phonology, Prosodic Phonology is thus a non-privative interface theory (see §§79,756).

106

two layers of the prosodic hierarchy that do not carry any extra-phonological information: syllables (phonotactic restrictions) and feet (prominence relations). The heterogeneous character of the buffer is further discussed in the following section. One case in point is Rubach's (1997 et passim) account of so-called trapped consonants in Polish (see Vol.1:§240, Scheer 2008b, 2009d), which he argues are extrasyllabic (even in the middle of words). On Rubach's analysis, these consonants end up being directly attached to the prosodic word. One suspicious aspect of this scenario is the existence of extrasyllabic consonants in the middle of morphemes (in violation of the Peripherality Condition, e.g. Clements, 1990:290, Hayes, 1995b:57f). Also, a wrong prediction is made to the effect that some natural language could allow for three, five or eleven extrasyllabic consonants in a row: since nobody has ever defined co-occurrence restrictions for segments that are directly attached to prosodic words (obviously because this is also in violation of the Strict Layer Hypothesis, see §383), any number of consonants should be able to be accommodated. In sum, cases where the Prosodic Hierarchy is used in the analysis of domestic phonology, i.e. in absence of reference to morpho-syntactic information, are suspicious by themselves. If the Prosodic Hierarchy is a diacritic and has to go, they must be misanalysis anyway.

340 Chap 10: Prosodic Phonology, on the representational side 401

6.2. The buffer does not include syllables and feet The two lowermost layers of the prosodic constituency, syllables and feet, stand out among the six layers of the Prosodic Hierarchy. As was shown in the previous section, the only purpose of the four uppermost layers is to store morpho-syntactic information. Also, they are constructed by mapping rules on the grounds of morpho-syntactic structure and readjustment decisions without any participation of phonological information. That is, the four uppermost layers are pure top-down constructions.107 On the other hand, syllables and feet have a motivation that is independent from the interface: the former is a bottom-up construction based on the sonority of its terminal elements, the segments, while the latter is a function of stress. Neither the sonority of segments, which is recorded in the lexicon, nor the distribution of stress relies on any structural description of some phonological rule. Also, no mapping activity of any kind contributes to the creation of syllables and feet. In other words, were there no interface, syllables and feet would still exist, but the four higher layers of the buffer would not. Nespor & Vogel (1986) admit the ontological difference between the syllable and feet on the one hand and the four uppermost layers of the Prosodic Hierarchy on the other. (161) "The phonological word is the lowest constituent of the prosodic hierarchy which is constructed on the basis of mapping rules that make substantial use of nonphonological notions." Nespor & Vogel (1986:109)

The quote from Nespor (1999) below confirms this split into two types of prosodic constituents. (162) "The three lowest prosodic categories are the syllable (σ), the foot (Σ) and the phonological word (ω). […] Of these categories, the first two are purely phonological in nature, while the third incorporates morphosyntactic information." Nespor (1999:119)

107

What I mean is that no phonological property encoded in the lexical makeup of morphemes participates. Information from phonological processes – the structural description of rules – of course defines the division of the linear string into prosodic constituents, at least when readjustment is needed. But this is not a bottom-up construction in the sense that syllables are a function of the lexical properties of segments.

Inspection of the buffer (Prosodic Hierarchy): what it is and what it is not 341

Inkelas (1990:37f) and Downing (2006:8f) also make this distinction: bottom-up constructed items (mora, syllable, foot) are evicted from the Prosodic Hierarchy altogether. Downing specifically argues from the perspective of Prosodic Morphology (more on this in §447). The fact that syllables and feet owe nothing to mapping rules is also pointed out by Selkirk (1986:385) and Rice (1990:292 note 3) (see also Chen 1990:36 on this issue). What really is behind the division at hand, however, is the seam between morphology and syntax: the phonological word and higher layers manage the interaction of words, while the categories below the word level define the relationship between morphemes. Hayes (1989 [1984]) draws the following conclusion from this situation. (163) "The Prosodic Hierarchy should be construed solely as a theory of syntactic juncture, with word-internal juncture handled within the Theory of Lexical Phonology." Hayes (1989 [1984]:207)

That is, Prosodic Phonology defines the phonological relationship of words, while Lexical Phonology is competent for morphemes. This division of labour thus reduces Hayes' (1989 [1984]) Prosodic Hierarchy to only four layers (instead of six). This perspective, however, has not enjoyed any echo, and the regular practice in Prosodic Phonology was to claim a peaceful coexistence between prosodic constituents and Lexical Phonology below the word level. §429 discusses the relationship between the two theories at greater length. In conclusion of this section, we can record that the Prosodic Hierarchy as commonly understood is a heterogeneous assembly of autosegmental constituents which may or may not carry morpho-syntactic information, and which may be top-down or bottom-up constructions. It is thus inappropriate to talk about the whole prosodic constituency in terms of a buffer: only the four uppermost layers store and restore morpho-syntactic information. On the pages below, I make allowance for this fact: the buffer refers only to a subset of the Prosodic Hierarchy.

342 Chap 10: Prosodic Phonology, on the representational side 402

6.3. The buffer is a diacritic – an autosegmental diacritic

403

6.3.1. Prosodic Phonology lays claim to boundaries: they are the old buffer, prosodic domains are the modern buffer Vogel & Kenesei (1990:344) review the arguments in favour of Indirect Reference. One point they make is historical: all interface theories have been indirect thus far, so there is probably something to this approach. They single out SPE as a forerunner of Indirect Reference. (164) "Working within the SPE framework, Selkirk [1972] modifies the original proposal by showing that at least in certain types of phonological phenomena, interaction between the two components is only indirect. Word boundaries (#'s) inserted into a string on the basis of syntactic structure determine where external sandhi rules apply. Phonological rules thus do not directly 'see' syntactic structure, but rather access only strings of segments and boundaries." Vogel & Kenesei (1990:344)

Hence representatives of Prosodic Phonology lay claim to the equivalence of #s and the modern prosodic constituency. The same line of reasoning is found in the overview article by Inkelas & Zec (1995). The authors call p-structure the level of representation that mediates between morpho-syntax and phonology; they explicitly identify boundaries as the ancestor of this mediating structure, whose more recent incarnation is the Prosodic Hierarchy. (165) "An early version of p-structure was proposed in SPE and developed in subsequent work (Selkirk, 1972, 1974; Rotenberg, 1978). According to this view, domains of phonological rules are expressed in terms of phonological boundary symbols, generated by rules. […] Far more constrained is the 'prosodic' view of p-structure. Under this view, p-structure occupies a level with its own hierarchical organization and a high degree of autonomy." Inkelas & Zec (1995:537f)

If, thus, prosodic constituency is but a more advanced version of boundaries that presents a number of advantages, it must have the same formal properties as its predecessor. The two quotes clearly show that prosodic constituency, just as hashmarks, is a diacritic: it serves no other purpose than replicating phonologically relevant morpho-syntactic information in phonology (see §400). Although proponents of Prosodic Phonology are prompt to point out that boundaries are odd because they are diacritic

Inspection of the buffer (Prosodic Hierarchy): what it is and what it is not 343

(§373), nobody wonders whether the Prosodic Hierarchy may be a diacritic as well, even when it is advertised as the direct surrogate of arbitrary and diacritic boundaries (Scheer 2008a). 404

6.3.2. The buffer and SPE-type boundaries share all properties That the Prosodic Hierarchy is a diacritic may also be seen when comparing its birth and life with boundaries. Just as these, it is the output of a translational process: the fixed mapping algorithm (plus eventually readjustment) in SPE (§§90f), mapping rules (including the Black Box) in Prosodic Phonology. Just like boundaries, the Prosodic Hierarchy is a UFO in the phonological module: it is injected for the exclusive purpose of storing extraphonological information (§400). Domestic phonology, i.e. the one that runs without extra-phonological conditioning, does not need to refer to either boundaries or prosodic constituency. Finally, just like boundaries, the units of the Prosodic Hierarchy are arbitrarily chosen and named: "ω" (the phonological word), "φ" (the phonological phrase) etc. are not any less arbitrary than "+" and "#". For some reason, however, phonologists always point out the arbitrariness of the typewriting symbol #, but do not mind talking about omegas and phis. Saying that an omega is only shorthand for a real linguistic object, the phonological word, does not help: the same may be said about + and # – only that a regular scientific-sounding terminology has never been introduced for these objects. Finally, pointing out that omegas and phis represent certain stretches of the linear string which coarsely correlate with morpho-syntactic divisions does not make them less arbitrary. For one thing, this merely shows that their only purpose is to replicate morpho-syntactic structure in phonology. But more importantly, the same may be said about boundaries – and actually was said (by McCawley 1968, see §113): + and # represent two different boundary strengths, the latter dividing larger chunks of the linear string. If more boundaries are added to the list, they may also be correlated with increasing chunks of the linear string that coarsely represent morphosyntactic divisions.

344 Chap 10: Prosodic Phonology, on the representational side 405

6.3.3. What counts as a diacritic? In order to assess the diacritic status of the Prosodic Hierarchy, one can also try to define what a diacritic is. A formal definition must rely on the alien status of the object in question in the environment where it evolves. A workable definition appears under (166) below. (166) definition of the term "diacritic" a diacritic is a non-native object in module X: it is only used when information from outside of X is processed. It is absent from events that do not appeal to extra-Xal information.

Hashmarks and omegas alike meet these conditions: they are nonphonological intruders in the phonological world which exist only when phonology appeals to extra-phonological information, and are systematically absent from phonological processes that do not use extraphonological information (such as ordinary palatalisations). Hashmarks and prosodic constituents are two ways of fulfilling the promise of Indirect Reference. For some strange reason, though, the former are stigmatised as arbitrary diacritics, while the latter are advertised as "truly phonological objects" (e.g. Selkirk, 1984:32, 409f, Nespor & Vogel, 1986:27ff, 110ff). Nespor & Vogel (1986:3) for example call boundaries "pseudo-phonological terms" and argue that phonology should only be able to refer to truly phonological objects (just as syntax can only make reference to truly syntactic objects). This is quite ironic since nobody has ever examined the question whether the Prosodic Hierarchy is a diacritic as well. Reading through the literature, I have only come across one voice that clearly identifies the buffer for what it is. Without surprise, this voice comes from direct syntax quarters (see §407) where of course no buffer is needed: Kaisse (1990:128) calls attention to the redundant and diacritic character of prosodic constituency. She points out that the direct syntax option "does not require the postulation of constituents that are needed only to describe the sandhi phenomena in question." In sum, thus, it appears that there is just one difference between hashmarks and omegas: the former are linear and local, while the latter are autosegmental, hence non-local units. The progress since SPE, then, does not concern Indirect Reference or the diacritic notation of phonologically relevant morpho-syntactic units; rather, it comes down to a change from a linear to a non-linear encoding of higher level information. This takes us back to the genesis of Prosodic Phonology: I have argued in §§368f that Prosodic Phonology is a child of autosegmentalism; it

Good and bad reasons for Indirect Reference 345

is the application of the general autosegmental trend of the early 80s to the interface.

406

7. Good and bad reasons for Indirect Reference

407

7.1. Direct syntax vs. Prosodic Phonology

408

7.1.1. Two approaches, their competition and their evolution In the years following the introduction of Prosodic Phonology, the main debate in the realm of the interface was about the buffer as such: so-called direct syntax approaches held that the extra layer of mediating prosodic representations is 1) superfluous and 2) complicates the statement of phonological rules. Instead, phonological rules should be allowed to make direct reference to morpho-syntactic structure. Direct syntax was common in early generative work on the interface, namely as a response to the frustration that was generated by boundaries: Rotenberg (1978) (§137), Clements (1978), Pyle (1972) (§136), Hyman (1978:459) (§123) and Kenstowicz & Kisseberth (1977) (§132) are examples. Interestingly, an early representative of the Prosodic Phonology tradition (where Indirect Reference visibly was not an issue yet) also bought into the direct syntax idea: Napoli & Nespor (1979). In the 80s, the direct syntax approach was most prominently entertained by Kaisse (1983, 1985, 1990), Lobeck & Kaisse (1984), Manzini (1983), Odden (1987, 1990), Rizzi & Savoia (1993) and Chen (1990) (Elordieta 2007:126ff, 2008:211ff provides an informed overview). A more recent representative of direct syntax is Seidl's (2001) attempt to marry Indirect Reference with direct syntax: Seidl provides for direct reference to morpho-syntactic structure by "early" rules (at a level she calls morphosyntactic), but maintains indirect reference to prosodic constituency by "late" rules (at a level called Prosodic Representation). Her Minimal Indirect Reference model is thus a hybrid entity which allows for direct syntax under the pressure of certain phenomena. Since there is no conceptual difference between doing a little or a lot of direct interface, though, her system may be considered a mere variant of direct syntax. Beyond this fact, it is argued in §409 below that mixing both direct and indirect reference is actually the worst of all systems since the existence of the direct syntax option makes the Prosodic Hierarchy superfluous.

346 Chap 10: Prosodic Phonology, on the representational side The competition of direct syntax approaches and Prosodic Phonology was launched in a special issue of the Phonology Yearbook (edited by Kaisse & Zwicky 1987), where both options appear with their pros and cons. Three years later in the volume on the phonology-syntax interface edited by Inkelas & Zec (eds.) (1990c), the balance is clearly in favour of Prosodic Phonology: all contributions but those by Odden and Chen either argue that reference to syntax may only be indirect via the buffer, or take this position for granted. 108 New data that the direct syntax approach is supposed to be unable to get a handle on are presented (e.g. Cho 1990, Poser 1990, Selkirk & Shen 1990, Condoravdi 1990). On the other hand, evidence which was brought forth in favour of direct syntax is reanalysed (see already Cowper & Rice 1987). Hayes (1990) for example admits that there is a "direct residue", but only in order to propose a "theory of the residue" within the prosodic frame where the relevant alternations are dealt with in the lexicon by some kind of allomorphy (precompilation theory). This trend is confirmed in the volume edited by Kleinhenz (1996) (e.g. Hannahs 1996) and the special issue of The Linguistic Review on the "phonology between words and phrases" edited by Marina Nespor in 1996: the Prosodic Hierarchy is by and large accepted as the correct approach to the interface. Its success is based on one single argument that is rehearsed over and over again: the units that are relevant for the statement of phonological rules are determined by, but not isomorphic with, morpho-syntactic structure. In other words, the need of readjustment (in terms of SPE, the Black Box under (140)) was decisive. In the younger OT-based literature, however, Indirect Reference is systematically violated and on many occasions (quite tacitly) replaced by direct syntax (§§494,501,525). As a consequence, the existence of the Prosodic Hierarchy as such is called into question, at least by those defenders of direct syntax who care for not having done the same labour several times (Pak 2008:60ff and Samuels 2009a:284ff offer relevant discussion regarding the uselessness of the Prosodic Hierarchy in a direct syntax environment).

108

The introduction to the volume by Inkelas & Zec (1990a) offers a wellinformed overview of the debate regarding direct syntax approaches, and of the arguments that have led to their discredit.

Good and bad reasons for Indirect Reference 347 409

7.1.2. Is the buffer useless and redundant? A real good motivation is needed Prosodic Phonology holds that phonology is burdened with extra arboreal structure plus extra computation, the mapping rules. These devices have no other purpose than making morpho-syntactic structure available inside the phonological module (see §400). By contrast, the alternative solution that makes direct reference to morpho-syntactic categories achieves the same result without any additional structure or computation. A bare assessment of both directions will thus disqualify Prosodic Phonology without any hesitation: why having a significant amount of additional structure and extra computation without any need? It is thus vital for Prosodic Phonology to be able to dismiss this obvious objection, which is made for example by Kaisse (1985:156 note 1, 110 note 1, 1990:128f). That is, Prosodic Phonology needs to find a real good reason why phonology should be unable to directly refer to morpho-syntactic structure.

410

7.2. A good reason: modularity

411

7.2.1. Introduction: different modules do not speak the same language In order to make the modular argument, the following pages need to anticipate on the proper introduction of modularity that is provided in the Interlude (§586). Modularity (Fodor 1983 et passim) is certainly a good reason for Indirect Reference. The modular perspective holds that grammatical activity – as other cognitive faculties – is organised in terms of a number of ontologically distinct, non-teleological and specialised computational systems: the modules. Modules are domain specific and encapsulated (§610). That is, they are devoted to a specific task, which they carry out using a specific vocabulary. Since they are ontologically distinct and speak different languages (of the mind), they are unable not understand, or even parse, what is going on in other modules.

412

7.2.2. Phonology-free syntax An illustration of the incommunication between different modules is the well-known notion of phonology-free syntax. There is no syntactic move-

348 Chap 10: Prosodic Phonology, on the representational side ment on record that would be triggered only if, say, the candidate begins with a labial. The same holds true for other categories that are relevant in phonology such as palatality, occlusion etc. Building on this observation, Zwicky & Pullum (1986a,b) have introduced the principle of phonologyfree syntax, which holds that phonology is entirely invisible to syntax. That is, conditioning is only top-down: syntax bears on phonology, but there is no communication in the other direction. Zwicky & Pullum's principle originally concerned only syntax; however, it was rapidly extended to morphology: no concatenation of two morphemes is supposed to be conditioned by the phonological properties of the items involved.109 Phonology-free syntax has rapidly become the standard view of the macro-landscape regarding modular identities, also in the Prosodic Phonology literature. Relevant references in this context include Pullum & Zwicky (1988), Vogel & Kenesei (1990:346ff), Miller et al. (1997) and Guasti & Nespor (1999). There is evidence, though, that the generalisation is too strong as it stands. Its scope is probably restricted to a subset of phonological items, i.e. those that occur below the skeleton (melody). That is, only melody is unable to influence concatenation (while items above the skeleton may impact the workings of morpho-syntax). Another issue is a putative difference in behaviour between morphology and syntax: while nobody doubts that syntax is be melody-free (Zwicky & Pullum's original observation), morphological concatenation is sometimes claimed to be conditioned by phonological factors that include melody. Let us first look at the literature that challenges the invisibility of phonological properties for morpho-syntax as such. These include Inkelas (1990), Inkelas & Zec (1990b, 1995), Hargus (1993), Neeleman & Reinhart (1998), Szendrői (2001, 2003, 2004) regarding syntax, Szymanek (1980), Ackema & Neeleman (2004:2), Burzio (2007) regarding morphology. Szymanek (1980), Vogel & Kenesei (1990) and Inkelas & Zec (1995) provide surveys of phenomena that are frequently quoted in support of the fact that phonology may have bearing on morphology and syntax. When looking at the inventory of phenomena that are argued to induce a bottom-up conditioning, a clear regularity appears, though. Everybody indeed agrees with Zwicky & Pullum's (1986a,b) original observation that segmental properties of sound never affect a syntactic derivation; Vogel & Kenesei (1990:346) as well as Inkelas & Zec (1990b:366, 1995:547) for 109

Recall from §253 that phonology-free syntax played a role in the discussion of interactionism, a key notion of Lexical Phonology.

Good and bad reasons for Indirect Reference 349

example are explicit on this. That is, nobody has ever seen anything like "verbs that begin with a velar are raising verbs". On the other hand, recurring candidates for bottom-up conditioning are located above the skeleton. This observation is also made by Kaisse & Hargus (1993:4) in the debate on interactionism: "if an affix subcategorizes for a base with certain derived phonological properties, those properties are almost always supra-segmental (e.g. stress)." A more documented record of this intuition can be gained when parsing the literature that argues against phonology-free syntax: the phenomena quoted are intonation and stress (Szendröi 2001, 2003, 2004, Hargus 1993), tree-geometric properties of the prosodic constituency (for example the existence or branchingness of constituents, Inkelas & Zec 1988, 1990b:372ff), the size of lexical items (minimal word constraints: number of syllables or moras, e.g. Inkelas & Zec 1990b:372ff, Hargus 1993, Bendjaballah & Haiden 2005, forth) and rhythm (Guasti & Nespor 1999). In other words, the literature has identified a red line that cuts the phonological space into two areas, above and below the skeleton (see Vol.1:§216). While the latter is invisible for syntax for sure, there is reason to believe that the former may be a factor in syntactic computation. Let us now turn to the putative distinction between morphology and syntax. The literature discusses cases where melodic properties impact the concatenation of morphemes (e.g. the aforementioned Szymanek 1980 and Ackema & Neeleman 2004:2, Burzio 2007). Hargus (1993:54ff) presents evidence for phonology-sensitive morphology from segmental processes, but points out herself (p.69) that most of these unexpectedly share the fact of involving non-concatenative morphology (Semitic, reduplication, infixation). This is certainly a correct observation. Let us have a closer look at phonologically conditioned infixation, which appears to be a particularly harsh violation of phonology-free morphology and therefore is typically quoted in this context. Based on Moravcsik (2000), Samuels (2009a:147ff) provides an overview of phonological factors that are known to condition infixation cross-linguistically. The list of anchor points that infixes look at in order to determine their landing site falls into two categories: edgeoriented and prominence-oriented. For the left edge for example, documented situations are "after the first consonant (or consonant cluster)", "after the first vowel", "after the first syllable" and "after the second consonant". Prominence-based attractors are stressed vowels, stressed syllables or stressed feet.

350 Chap 10: Prosodic Phonology, on the representational side In no case, however, is melody reported to be relevant for the definition of the landing site. Hence cases where infixes are inserted after, say, the first labial consonant of the word (and in absence of labials are prefixed) do not seem to be on record. Only Zuraw (2007) has found evidence for the influence of major categories, which however is based on loanword adaptation, and which is only gradual, rather than categorical. In her Tagalog data (Austronesian, Philippines), word-initial stop-glide clusters are significantly more often split than stop-liquid clusters. Tagalog does not have native word-initial CC clusters, and hence speakers must make a decision to insert relevant infixes (which normally land after the first consonant of the word) either after C1 or C2 (e.g. graduate can come out as g-um-raduate or gr-um-aduate). The statistically relevant impact of cluster types, however, does not need to harm the generalisation that infixation is blind to melody: the most obvious analysis is to interpret the difference between stop-liquid and stop-glide as a contrast in (syllable) structure, rather than in melody. It thus seems that Zwicky & Pullum's generalisation also holds for infixation if the melodic proviso is observed: infixation does not react on melodic properties of phonology. On a more general note, defenders of phonology-free syntax have proposed reanalyses of bottom-up conditioning, or place it outside of grammar (discourse-related): among others, Zwicky & Pullum (1986b), Vogel & Kenesei (1990), Odden (1993), Miller et al. (1997), Guasti & Nespor (1999), Bošković (2001, 2005a) and Revithiadou (2006) have worked in this direction.110 In conclusion of this survey, there is reason to believe that phonology-free syntax is still a valuable generalisation which however needs to be restricted to melody. Melody-free syntax then also seems to extend to morphology. It will be shown in §660 that the non-communication between the area below the skeleton and morpho-syntax also holds true for the reverse 110

Bendjaballah & Haiden's (2005, forth) analysis of Kabyle Berber also falls into this category: it puts a phenomenon which looks like a phonological conditioning of preposition selection back under the roof of phonology-free syntax. That is, in the variety of Berber examined, "small" prepositions can only occur with the Construct State not because syntax looks at their size, but because they are small enough to be hosted by a CV unit that is provided only by the Construct State. On this analysis, thus, it is true that a phonological property of the prepositions at hand decides on whether a given item can occur in a specific syntactic context – but at no point of the process does syntax "look into phonology".

Good and bad reasons for Indirect Reference 351

direction: when morpho-syntax ships off information to phonology, it never accesses the area below the skeleton. That is, melodic properties of sound are never targeted by any higher level intervention (of the kind "p becomes r before this or that morpho-syntactic division", or "raising verbs palatalise"). Morpho-syntax can only bear on the phonological structure above the skeleton. 413

7.2.3. Phonology and morpho-syntax do not speak the same language – hence communication requires translation Given this situation, no doubt there is an ontological gap between (at least the melodic part of) phonology and morpho-syntax. This is confirmed when comparing the basic vocabulary that is used in phonology and other linguistic areas: number, person, animacy, quantification, aspect and so forth are categories that are used, understood and processed by syntax as much as by morphology and semantics. In this sense, syntax, semantics and morphology speak the same language. Much unlike phonology, which does not know what number, person etc. is. The reverse is also true: labiality, occlusion and the like play no role in either syntax, morphology or semantics.111 This privacy of vocabulary is called domain specificity in modular theory. It is the basic argument which allows us to decide that two computations are located in distinct modules. Jackendoff (1997) for example is explicit on the fact that a module can only understand its own language (of the mind), and that two different languages cannot coexist within the same module.112 (167) "'Mixed' representation[s] should be impossible. Rather, phonological, syntactic and conceptual representations should be strictly segregated, but coordinated through correspondence rules that constitute the interfaces." Jackendoff (1997:87ff)

In sum, there are two macro-modules, morpho-syntax and semantics on the one hand, phonology on the other, which manipulate mutually exclu-

111 112

This argument is developed by Michal Starke in unpublished work. Further discussion is provided in §641. This issue is discussed at greater length in §§640,650.

352 Chap 10: Prosodic Phonology, on the representational side sive categories.113 We know for sure, however, that they communicate, at least top-down: morpho-syntax bears on phonology. Modularity thus enforces translation: there must be a Translator's Office where the morphosyntactic language is translated into the phonological idiom (mapping rules and the black Box in Prosodic Phonology). Were there no translation, poor phonology could not react on orders that are issued in the morpho-syntactic language. This also implies that the translational process cannot take place in either morpho-syntax or phonology: the Translator's Office has access to the structure and the labels of both sides. It therefore must stand in modular no man's land (§381).

414

7.2.4. Nobody makes the modular argument in order to sustain Indirect Reference If implicitly and timidly on most occasions, modularity and the fact that phonology does not speak the same language as other modules is invoked in the Prosodic Phonology literature (e.g. Nespor & Vogel 1986:27ff, 110ff). Some relevant quotes from Selkirk (1984) appear below. (168) "The syntax and the phonology are entirely autonomous components, the rules of syntax making no appeal to phonology and vice versa. Mediating between these components, however, are two others that define the relation between the syntactic and phonological representations of a sentence. The principles of these components have a mixed vocabulary. They appeal to certain limited aspects of syntactic representation in constructing the hierarchies of phonological representation." (Selkirk 1984:410f) "Phonological rules are blind to syntactic structure. In the general case, then, phonological representation, richly structured itself, mediates between syntactic structure and the phonological rules that ultimately specify the details of the phonetic realization of the sentence." (Selkirk 1984:32)

113

Eventual internal modular structure within the former is a debated issue, on which more in §533.

Good and bad reasons for Indirect Reference 353 "Our hypothesis is that phonological rules make no direct reference to syntactic structure, their application being governed only by the phonological representation itself.[…] We are saying that the phonological component of the sentence is syntax-free, or that phonology, defined in this way, is 'postcyclic'. […] It is the rules of the syntax-phonology mapping […] that are unalterably syntax-dependent. […] The rules look at syntax and build phonology." (Selkirk 1984:409f)

Typically, modularity is put to use when the Prosodic Hierarchy is compared with boundaries: Nespor & Vogel (1986:3) for example call boundaries "pseudo-phonological terms" and argue that phonology should only be able to refer to truly phonological objects (just as syntax can only make reference to truly syntactic objects). As far as I can see, however, on no occasion is modularity used in order to sustain Indirect Reference. In a modular environment where phonology and morpho-syntax are two distinct modules, though, the direct syntax perspective is definitely ruled out. Morpho-syntactic information can only be parsed by phonology if it was translated into the phonological language – exactly what is required by Indirect Reference. Hence the general interface architecture (140) that Prosodic Phonology has established is a modular architecture whose central element is the Translator's Office (mapping rules and the Black Box). Strangely enough, though, it seems that this was a coincidence with, rather than an application of, the contemporary modular idea: the modular argument in favour of Indirect Reference is missing in the Prosodic Phonology literature.

415

7.2.5. Modularity and Level Independence The modular idea of distinct ontological entities itself, as well as its concomitant need for intermodular translation, signs the return of Level Independence.114 Recall from §§60f that this was the major architectural principle in structuralist theory: the use of morpho-syntactic information in phonology was prohibited. Bypassing this proscription more or less deliberately and more or less consciously, the only way to have morpho-syntactic information bear on phonology was to put it in a phonological guise (§§64ff). Since

114

The relationship between Level Independence and modularity is also discussed in §§72,699.

354 Chap 10: Prosodic Phonology, on the representational side phonemes were the relevant phonological currency, extra-phonological information was carried by juncture phonemes in structuralist times. Level Independence thus expresses the idea that phonology and morphology are two distinct ontological entities which work on different vocabulary – this is called domain specificity in modular theory. Given this rigid frame, structuralists were forced into translational activity that transforms morpho-syntactic information into phonemes. Of course, Level Independence has never been formulated this way. Of course, it had an unfortunate implementation namely because the phonemic principle enforced a free distribution of juncture phonemes, including in the middle of morphemes (§69). Of course, its motivation was not even remotely related to the cognitive ambition of modularity: Level Independence roots in the descriptive tradition of American structuralism that seeks a discovery procedure. And of course, it is wrong in its literal meaning ("there is no morpho-syntactic information in phonology"). But all this does not belittle the fact that the basic idea of modularity and its consequence, translation, was present in structuralist thinking: Level Independence does not reduce to the proscription of morpho-syntactic information in phonology. Looked at from this perspective, Level Independence (which was abandoned in generative times, see §76) appears to be quite modern: after a period of twenty years (that was initiated by SPE) where direct reference to morpho-syntactic labels was common practice, generative theory applies Level Independence in the guise of modularity since the advent of Prosodic Phonology and Indirect Reference in the mid 80s.

416

7.3. A bad reason: non-isomorphism

417

7.3.1. Non-isomorphism in the Prosodic Phonology literature The argument in favour of Indirect Reference that is made in the Prosodic Phonology literature – the only argument that one comes across – is nonisomorphism. This notion embodies the observation that in some cases, the domain to which a phonological rule makes reference is not co-extensive with any morpho-syntactic domain. Or, in other words, some phonological rules make reference to information that is absent from the syntactic surface structure. Therefore, the argument goes, there need to be a readjusting mechanism that further transforms the output of morpho-syntax so to create the

Good and bad reasons for Indirect Reference 355

divisions which are needed for phonology. This job is done by the mapping rules whose output, the buffer, provides the adequate target for the reference of all phonological rules that are sensitive to morpho-syntactic information. That is, mapping rules preserve all morpho-syntactic divisions that are necessary for phonological rules to apply, and add new information of their own (this is the function of the Black Box under (140)). The notion of non-isomorphism goes back to SPE (though it was not called like that then), where it motivated the readjustment component. The relevant quote from Chomsky & Halle (1968:371f) is discussed in §91. Non-isomorphism has then served to motivate Indirect Reference since Selkirk's (1981 [1978]:138) first article and runs through the entire Prosodic Phonology literature.115 The following quote from Selkirk (1984) is an example. (169) "An intonational phrase typically contains material belonging to a sequence of words and/or phrases and it is not necessarily isomorphic to any constituent of syntactic structure. […] The existence of the intonational phrase is motivated primarily by the necessity of defining intonational contours with respect to some unit of representation that is both larger than the word and variable in extent. That unit cannot be a syntactic one, because the syntactic sequence with which an intonational contour is associated may not be a constituent of syntactic structure. And the metrical grid alignment of the sentence defines no such unit in the representation." Selkirk (1984:27)

418

7.3.2. The phenomenon: cat-rat-cheese, phonology over sentences Nespor & Vogel (1986, all through the book: 4f, 34ff, 124ff etc.) are very conscious that their theory stands and falls with non-isomorphism. They therefore spend a lot of ink on making the argument, which is illustrated by empirical evidence from various languages. Two cases are discussed below. They are chosen on account of their significance and notoriousness. The cat-rat-cheese example was used by Chomsky & Halle (1968:371f), and since Nespor & Vogel (1983:130ff, 1986:57f) has become the standard data set which is always repeated when

115

Other references of the early literature include Selkirk (1980a:110), Nespor & Vogel (1982:226), Nespor & Vogel (1986:37ff), Hayes (1989 [1984]:201), Nespor (1985), Neijt (1985:180), Booij (1985b:149). Vogel (1999:117ff) is an example from the more recent overview literature.

356 Chap 10: Prosodic Phonology, on the representational side non-isomorphism is discussed (e.g. Vogel & Kenesei 1990, Nespor et al. 1996, Dresher 1996:42). Compare the major syntactic divisions of the sentence under (170a) with its intonational structure under (170b). (170) a. This is [the cat that caught [the rat that stole [the cheese]]] b. [This is the cat] [that caught the rat] [that stole the cheese]

The two lines of division do not coincide. Hence, goes the argument, whatever drives phonology to decide that the intonation is as under (170b), it is not the output of the syntactic module. There is no node in the syntactic tree that uniquely dominates every intonational span of the sentence. Also, syntactic theory could not evolve in a way so to achieve this goal since this would mean that syntactic structure gives up on the embedding of sentences: the middle intonational chunk [that caught the rat] is not uniquely dominated by any syntactic node because its CP also dominates the embedded third chunk [that stole the cheese]. Therefore the relevant intonational structure must be created outside of the syntax by some interface activity: mapping rules flatten their syntactic input under the influence of the Black Box (see (140)). The second phenomenon targets another unmodifiable property of syntactic structure: the fact that two independent sentences are not dominated by any node. It is therefore impossible to describe a phonological event at the break of two sentences in terms of domains that are isomorphic with syntactic structure: the relevant domain, that is the string encompassing the two sentences, does not exist in the syntactic tree. The witness for the existence of cross-sentence phonology that Nespor & Vogel (1986:4f) present is linking r. In certain non-rhotic varieties of English such as Received Pronunciation, etymological r remains unpronounced utterance-finally as well as before consonant-initial morphemes and words (mothe*[r], mothe*[r]ly, my mothe*[r] comes), but appears before vowel-initial morphemes and words if the intervening boundary allows for the linking effect (mothe[r]ish, my mothe[r] is coming). The critical fact for the demonstration at hand is that linking r may also appear at the break of two sentences, as shown under (171) below.116

116

Examples are from Nespor & Vogel (1986:4f, 46f), data and argument are developed at length in Vogel (1986). The same pattern is produced by t-flapping in certain American English varieties, cf. Nespor & Vogel (1986:46f, 223ff).

Good and bad reasons for Indirect Reference 357 (171) a. There's my mothe[r]. I've got to go. b. There's my mothe*[r]. I've got two cats.

Nespor & Vogel contrast (171a) where the r may be linked with (171b) where according to them no r may be pronounced. The difference, they argue, is semantic: the relationship between the two sentences under (171b) is distant, while it is intimate under (171a). This is the critical factor for the appearance of the linking r in otherwise identical conditions. Nespor & Vogel (1986) conclude that phonology makes reference to a domain which does not exist in syntax. The relevant phonological domain, the phonological utterance, must therefore be constructed by the interface mechanism, that is mapping rules. On this count, then, r is only linked within a phonological utterance, and too much semantic distance does not allow a single phonological utterance to span two sentences.

419

7.3.3. Domain abuse I: there is no argument when phonology refers to boundaries instead of domains The point made on the grounds of non-isomorphism is certainly correct – but only if it is taken for granted that phonological rules make reference to domains, rather than to (local) boundaries. That is, (non-)isomorphism in the Prosodic Phonology literature is always understood as a comparison between domains: morpho-syntactic and phonological. While it is certainly true that morpho-syntactic organisation is arboreal, nothing affords the a priori that the interface currency on the phonological side are non-local domains, rather than local boundaries. This puts us back to the issue of whether higher level intervention in phonology is local or non-local (§§366,706). We are thus invited to look at the same facts also through the lens of local boundaries. The effect is quite striking: non-isomorphism evaporates. In the cat-rat-cheese example under (170), intonational chunks simply begin at every CP. Hence it is enough to say that the intonation-building mechanism starts a new unit every time it hits the phonological translation of the CP-boundary. Intonational and syntactic structure are thus perfectly isomorphic, and no extra constituency on the phonological side is needed. The same holds true for linking r under (171): the point made by Nespor & Vogel relies on the absence of a domain that encompasses two sentences. If boundaries are the units to which phonological rules make reference, however, no sentence-spanning domain is needed: the contrast

358 Chap 10: Prosodic Phonology, on the representational side observed stems from the different boundaries that separate the two sentences at hand and denote their variable semantic relationship. When going through the evidence that has been produced in order to support non-isomorphism (among many others, Selkirk 1981 [1978], 1981, 1984:27f, 1986, Nespor & Vogel 1982, Nespor 1985, Hyman et al. 1987, Rice 1987, Vogel & Kenesei 1987, Nespor & Vogel 1986:36ff, Nespor et al. 1996:7), it appears that the result is always the same: the regularity which is formulated in terms of domains can as well be described with boundaries and then turns out to be perfectly isomorphic. That is, boundaries necessarily correspond to some morpho-syntactic division. Thus case after case, non-isomorphism turns out to be an artefact of the domain-preempted analysis, rather than a fact about the relation between phonology and morpho-syntax. In sum, prosodic constituency and the claim that phonological rules make reference to domains is a self-fulfilling promise: once domains exist, the argument based on non-isomorphism can be made to the effect that we need the Prosodic Hierarchy. Non-isomorphism disappears, however, if higher level intervention in phonology is local. Since non-isomorphism is the only argument for the buffer, the entire parallel world of prosodic constituency turns out to be superfluous if local intervention turned out to be correct. 420

7.3.4. Domain Abuse II: theoretical units are confused with descriptive categories It was already mentioned that the success of Prosodic Phonology may be measured by the fact that generalisations are commonly stated in terms of prosodic constituents, rather than in terms of morpho-syntactic categories or descriptions of the divisions at hand. The following quote is an example of this practice: for the authors, "applying across words" is synonymous with "applying within a domain". (172) "When a phonological rule applies across words, it is necessary to be able to specify across which types of words it may apply and across which it may not, or in other words, within which domain it applies." Vogel & Kenesei (1987:243)

Good and bad reasons for Indirect Reference 359

Using domains as a descriptive category without considering the local alternative may introduce a fatal bias into the analysis.117 This may be illustrated by Nespor & Vogel's (1986:42ff) analysis of Spanish nasal assimilation: the nasal of a nasal-obstruent cluster that straddles a word boundary may come to agree in place with the obstruent. Nespor & Vogel provide two examples where homorganisation goes into effect when the two words in question are a verb followed by an object NP, and a verb followed by an adjective, respectively. From this, Nespor & Vogel conclude that the domain of application of the homorganisation rule is "within the VP". They then introduce additional evidence that appears to violate this generalisation: there is no homorganisation between an object NP and a following preposition. They conclude that Spanish provides evidence for the conditioning of phonological processes by constituent size: processes are more easily blocked when their domain of application, the VP in the Spanish case, is long (i.e. contains many words). This analysis is inconclusive: in order to tell whether it is correct we would need to examine the case of an object NP-Prep boundary that does not make the VP too long. If this is not possible, the matter may not be decided since the blocking effect may either be due to the length of the VP or to the object NP-Prep boundary. In any event, declaring that the domain of application of a rule is the VP without having tested all possible VPinternal boundaries is confusing boundaries – the only descriptive unit that is given to the analyst – and domains, a derived category. 421

7.3.5. Is prosodic phrasing sensitive to the length of the string? Since Nespor & Vogel (1986), length sensitivity of prosodic phrasing runs through the literature. Given its initial illustration, however, it may be worthwhile checking whether it is an artefact of domain-biased analysis, rather than a fact about the interface. Four points can be made. For one thing, it is never the case that the alleged "short" and the alleged "long" portions of the string are morpho117

It is true, however, that work in Prosodic Phonology usually avoids this kind of pitfall: the inventory of boundaries which have a (non-)blocking effect in regard of some phonological process is established first. These lists are then converted into prosodic categories. Work along these lines includes Selkirk & Tateishi (1988), Hayes (1989 [1984]), Selkrik & Shen (1990), Cho (1990), Condoravdi (1990), Bao (1996).

360 Chap 10: Prosodic Phonology, on the representational side syntactically identical. This is the flaw in Nespor & Vogel's (1986) analysis of Spanish: the "longer" string has also more internal morpho-syntactic structure; hence these additional boundaries, rather than length, could well be responsible for the effect observed.118 This is precisely what Wagner (2005a,b) shows for intonational units: following the question "who was at the party?", the answer "[Morgan and Joey]" is just one intonational unit, while the answer "[Morgan] or [Joey and Ronny]" bears two intonational spans. Contra Watson & Gibson (2004, 2005) who hold that the probability of the existence of a boundary depends on the size of the constituents that are adjacent to it, Wagner shows that "length" is responsible for nothing at all – its alleged effects are due to the (recursive) syntactic structure of the "long" items. Second, it is not clear in which way "size" is a phonological category. Phonology has no means of knowing how long a word or a morpheme is. And "size" is nothing that is known in phonology proper either (i.e. when no extra-phonological information intervenes): there is no phonological process, say palatalisation, vowel harmony or the like, that depends on the length of something. Hence to the extent that anybody decides what is long and what is short, this job is surely not done by the phonology. Third, alleged size restrictions always seem to concern either the phonological phrase or the intonational phrase. There is no good reason why this should be so: if mapping may be size-sensitive, this option should at some point be visible at all levels of the Prosodic Hierarchy. Finally, the original size-based generalisation has long been reanalysed in the literature either along the lines discussed, or according to eurhythmic properties of speech. Ghini (1993) is at the origin of the latter line of thought, while Sandalo & Truckenbrodt (2002) for example argue for the former solution (the authors use a constraint WRAP-XP which requires that

118

The same holds true at the morphological level. Ricardo Bermúdez-Otero (pc) has brought up the following contrast: in the frame "… requires us to think again about these issues", a phrase-initial NP subject such as "contemporary thought" is less likely to constitute a separate intonational unit than "contemporary antidisestablishmentarianism". He concludes that size matters: larger items which are otherwise identical tend to be cut down into several prosodically relevant portions. However, antidisestablishmentarianism and thought may be identical on the syntactic side – they are certainly not identical morphologically speaking. Hence nothing guarantees that it is size, rather than the additional internal structure, which produces the effect. A test case would be two monomorphemic items that display an important contrast in length.

Relations with Lexical Phonology and the metrical grid 361

each XP be contained in a phonological phrase, thereby exploiting the additional structure of "long" items). In sum, then, there seems to be little evidence for and yet less appeal to analyses where the size of a string contributes to the construction of prosodic constituency.

422

7.4. Conclusion In sum, Prosodic Phonology has done exactly the right thing – introducing Indirect Reference as the major principle of interface architecture and installing a no-man's land-based Translator's Office and mapping rules – but for the wrong reason. Non-isomorphism is a non-reason: it exists only if the unsupported a priori is assumed that non-local domains are the operative interface currency; it evaporates when higher level information is handed down through local intervention at the seam between morphemes and words. The reason that really enforces the existence of Indirect Reference, the Translator's Office and mapping is modularity and the concomitant fact that phonology does not speak the same language as morpho-syntax. That is, any direct transmission of morpho-syntactic information to phonology would simply remain without effect since it could not be interpreted.

423

8. Relations with Lexical Phonology and the metrical grid

424

8.1. Introduction Let us now look at the relations that Prosodic Phonology entertains with other approaches to the representation of morpho-syntactic information in phonology. Two cases need to be examined: Lexical Phonology and metrical grids. The question is whether the scope and instruments of Prosodic Phonology are in concurrence with these approaches, or whether they are complementary.119 119

Beyond the issue of Prosodic Phonology and hence the interface with other modules, there has also been some debate on purely phonology-internal grounds regarding the competition between (s/w-labelled) metrical trees and the grid. Arguing that the coexistence of both devices is redundant, Kiparsky (1979) proposes a tree-only solution. By contrast, Prince (1983) and Selkirk (1984), for the sake of the same economy argument, favour a grid-only ap-

362 Chap 10: Prosodic Phonology, on the representational side Roughly speaking, two positions have been taken, one proposing a peaceful coexistence, the other arguing for a clear division of labour (with metrical grids) and an overt conflict with Lexical Phonology ("either must go"). The former view is defended by Nespor & Vogel (1986) (see also Nespor 1988, 1990) and Rubach & Booij (1984), Booij (1988, 1992), while the latter appears in Selkirk (1984:412ff) and Inkelas (1990). Inkelas is less radical than Selkirk in that she empties Lexical Phonology of its content, but formally maintains the existence of the empty shell (while Selkirk holds that Lexical Phonology has to go). As in other areas related to Prosodic Phonology, it is Nespor & Vogel's (1986) view that was widely adopted by phonologists. That is, the assumption of prosodic constituency, metrical grids and the architecture of Lexical Phonology are taken to be orthogonal issues: phonologists can design phonological theories that implement either to various degrees without this prompting a conflict with other items of the list. 425

8.2. The metrical grid

426

8.2.1. Selkirk (1984): mapping modifies the (pre-existing) grid After having introduced the idea of arboreal prosodic structure into the field in earlier work (Selkirk 1981 [1978], 1980a,b), Selkirk (1984) abandons prosodic constituency. Now adhering to Prince's (1983) grid-only approach, she believes that metrical grids, the new competitor of the Prosodic Hierarchy, is superior and makes most of the prosodic categories obsolete (Selkirk 1984:9ff). Selkirk (1984:29ff) holds that the function of all prosodic categories except the intonational phrase is taken over by the metrical grid. She argues that the metrical grid is independently needed because of the parallel between linguistic and musical/rhythmic structure. Also, grids express prominence relations (strong vs. weak) among elements of the linear string. These arguments have originally been made by Liberman (1975), Liberman & Prince (1977) and were developed by Prince (1983). They were also discussed in §§374f. proach. Liberman & Prince's (1977) original peaceful coexistence where grids are derived from metrical trees (later prosodic constituents) has finally become unchallenged mainstream under the influence of Hayes (1984), Selkirk (1986), Nespor & Vogel (1986) and others.

Relations with Lexical Phonology and the metrical grid 363

The only arboreal categories that Selkirk admits are therefore the syllable (which mediates between segments and the grid) and the intonational phrase, which is understood as a "sense unit". That is, intonational phrases are "units of information structure" and hence have a purely semantic definition (Selkirk 1984:27f). Therefore, Selkirk concludes that "it is significant that the syntax of a sentence in no way determines its intonational structure" (Selkirk 1984:408). In sum, Selkirk (1984) proposes a version of Prosodic Phonology without prosodic constituency. In her system, the idea of the buffer remains unmodified, only is the buffer now incarnated by the metrical grid. As in all other versions of Prosodic Phonology, the translation of morpho-syntactic into phonological structure is done by mapping rules, whose output now is the metrical grid. The version of the grid that (Selkirk 1984:31ff) adopts, though, is particular in that it exists anyway and in absence of any mapping activity (while the Prosodic Hierarchy is a creature of mapping rules). This is because the grid, recall, is motivated independently from any issue related to the interface (by musical and rhythmic properties of speech). In this environment, Selkirk's (1984) mapping rules basically do two things: they either align syllables and specific grid marks, an idea that will be central twenty years later in OT (see §457), or they insert extra grid marks at some boundary.120 An example of the former is the universal rule of "Demi-Beat Alignment DBA" which operates on the lowest grid level and says "align just one demibeat with every syllable". The two "Basic Beat Rules" also contribute to mapping: they align syllables with beats on the second grid level on the grounds of their internal structure and their position in regard of a particular syntactic domain. For example, "align a syllable of compositional type x with a beat", or "align a syllable in position y with a beat" (Selkirk 1984:54ff). Still another activity of mapping is the insertion of new grid marks at level 0 ("syntactic silent grid positions"); this is called "Silent Demibeat Addition SDA" (Selkirk 1984:305ff) and actually expresses Selkirk's (1972) old idea that sandhi phenomena are more likely to occur at high speech tempo. Once grid marks are aligned and inserted, proper phonological rules (those sensitive to syntax as well as those sensitive to morphology) make reference to grid structure, and to nothing else. 120

Selkirk (1986:376) will admit that the latter mechanism tacitly reintroduces unbeloved SPE-type boundaries through the back door.

364 Chap 10: Prosodic Phonology, on the representational side 427

8.2.2. Selkirk (1986): the grid is born from mapping on the grounds of prosodic constituency Two years later, Selkirk (1986) redefines the nature of the buffer. That is, she takes a more conciliated position that allows for a peaceful coexistence of prosodic constituency and the metrical grid. Selkirk admits the existence of three layers of prosodic constituency that correspond exactly to the three levels of syntactic X-bar structure: the phonological word (X0), the phonological phrase (X'') and an intermediate layer that she terms the "small phonological phrase" (X') (see §§395f). In this landscape, the metrical grid exists as a parallel phonological structure that is created by mapping on the basis of prosodic constituency (Selkirk 1986:376). The system thus acknowledges two distinct mapping mechanisms that are serially ordered: prosodic constituency is the output of regular mapping rules (whose input is morpho-syntactic structure); the grid is then created by a second set of mapping rules, whose input is the prosodic constituency. The function of the grid, then, is threefold according to Selkirk (1986:375f): it represents prominence relations (strong vs. weak), rhythmic patterning and sandhi effects (Silent Demibeat Addition, see §426).

428

8.2.3. Alignment on Nespor & Vogel's peaceful (and modular) coexistence Selkirk's (1986) move in fact is an alignment on Nespor & Vogel's (1986) position, who write on the first page of their book that SPE was wrong because it did not recognise the fundamentally fragmented nature of phonology. (173) "While such a model [SPE] is appealing for its simplicity and has, indeed, yielded many interesting results, it is our contention that this view of phonology is fundamentally inadequate. That is, on the basis of developments in phonological theory over the past decade, it seems that the phonological component cannot be considered a homogeneous system, but rather must be seen as a set of interacting subsystems, each governed by its own principles, such as the theories of the metrical grid, lexical phonology, autosegmental phonology, and prosodic phonology." Nespor & Vogel (1986:1)

It is to be noted that the argument, which relies on the idea that phonology is manifold, quite obviously owes to the modular architecture of grammar that was promoted by contemporary syntactic theory. The idea in

Relations with Lexical Phonology and the metrical grid 365

Government and Binding (Chomsky 1981), was that not only linguistic disciplines such as syntax and phonology are modules: they also subdivide into autonomous sub-modules. Syntax thus accommodated X-bar theory, theta theory, barrier theory, government theory, case theory, binding theory and control theory (see §628). In this spirit, Nespor & Vogel (1986) talk about sub-systems (the word modularity is not used, and no reference to Fodor 1983 is made). The implicit reference to modularity, however, is quite remote from the one that is discussed in §§413,586: it seems to reduce to the recognition of several sub-components within a linguistic discipline. How separate modular entities are identified and how they communicate are questions that are not on Nespor & Vogel's agenda. In any event, the result of this phonological manifoldness is peaceful cohabitation of the Prosodic Hierarchy and the metrical grid. This perspective was already argued for by Hayes (1984); it is further supported by Selkirk (1986), Nespor & Vogel (1989) and Nespor (1987, 1988, 1990). In this environment, the only interface with morpho-syntax is the Prosodic Hierarchy, which subsequently is the input for the construction of the grid. As for the first mapping from morpho-syntax, the second mapping from prosodic constituency to the grid may create non-isomorphic structure. This occurs for example in the case of so-called stress clash when some repair is needed (thirtéen vs. thírteen men). Nespor (1990) sums up the general picture as follows. (174) "Both prosodic structure and the metrical grid are significant levels of representation: prosodic structure mediates between syntax and the prosodic component of postlexical phonology, and the grid mediates between prosodic phonology and the phonology of rhythm. External sandhi rules (as well as internal ones, for that matter) apply on a hierarchy organized into constituents, while rhythmic rules apply on a hierarchy that contains only a sequence of more or less prominent periodicities. According to this view, the interface between syntax and phonology is limited to prosodic phonology; in the case of rhythmic phonology one can hardly speak of reference to syntax at all." Nespor (1990:244)

This peaceful coexistence appears to be consensual in phonology since the late 80s, and to the extent that the grid has survived into more recent constraint-based environments, is still practised.

366 Chap 10: Prosodic Phonology, on the representational side 429

8.3. Lexical Phonology I: conflict with Prosodic Phonology

430

8.3.1. No concurrence above the word level The relation between Prosodic Phonology and Lexical Phonology is clear as far as the level above the word is concerned: only the former can describe the influence of syntax on phonology. Lexical Phonology is a possible player only below the word level where morphology interacts with phonology. This is because Lexical Phonology holds that syntactic activity is postlexical and phonological processes "automatic" at this level. The tools of the theory thus cannot differentiate between syntactic configurations that do vs. those that do not provoke some phonological effect. This line of division is consensual and clearly stated by Nespor & Vogel (1986:30) (see also Selkirk 1986:402 note 6). In other words, the scope of Lexical Phonology is a proper subset of the scope of Prosodic Phonology, which however nothing prevents from also being active below the word level. Therefore it remains to be seen what the division of labour below and at the word level looks like.

431

8.3.2. Hayes (1989 [1984]): Prosodic Phonology above, Lexical Phonology below the word level One option is to say that there is no division of labour at all. That is, Lexical Phonology and Prosodic Phonology have complementary scope: the former operates only below the word level (hence manages morphology), while the latter is only competent for conditioning factors at and above the word level (hence takes care of syntax). This is the position of Hayes (1989 [1984]). (175) "The Prosodic Hierarchy should be construed solely as a theory of syntactic juncture, with word-internal juncture handled within the Theory of Lexical Phonology." Hayes (1989 [1984]:207)

As a consequence, Hayes' (1989 [1984]) version of the Prosodic Hierarchy excludes syllables and feet, the two layers below the word level: it consists only of the five uppermost layers (the phonological word, the clitic group, the phonological phrase, the intonational phrase and the phonological utterance). Significantly, this division between morphology and syntax coincides with the red line that separates categories of the buffer (i.e. that

Relations with Lexical Phonology and the metrical grid 367

carry morpho-syntactic information and are top-down constructions) from phonological categories (the syllable and feet, i.e. which carry no morphosyntactic information and are bottom-up constructions) (see §401). 432

8.3.3. Selkirk (1984): Lexical Phonology is redundant and has to go A question that Hayes does not address is why Lexical Phonology is needed in the first place when Prosodic Phonology can do its job. This issue, economy, is Selkirk's (1984:412ff) central argument: if the Prosodic Hierarchy is the only candidate for managing the syntactic influence on phonology, and if its scope can be extended to the level below the word, why should there be an extra theory that handles the bearing of morphology on phonology?121 The following quote is from the conclusion of Selkirk's 1984 book. (176) "We see no theoretical or empirical advantage to the Siegel-Allen-Kiparsky theory of word formation as a characterization of the syntax of words. We have argued instead for characterizing word structure exactly as sentence structure is characterized." Selkirk (1984:415)

Selkirk (1984) thus holds that the instructions which generate words and sentences are the same (while they are potentially distinct on the stratal perspective). Also, in her view the morpho-syntactic tree matters below as much as at and above the word level (while it is entirely irrelevant on the stratal count). The concern for a "single engine" that builds morphological and syntactic structure alike heralds Distributed Morphology, whose central claim is that morphology is but lower syntax (see §533, and specifically §540 on Selkirk's work). According to Selkirk (1984), Lexical Phonology has to go because it is redundant: regular representational devices – the syntax of words – can do the job. Interestingly, her argument has got nothing to do with Prosodic Phonology itself: nothing hinges on any property or on any tenet of this 121

Interestingly, the reverse move is also possible: if there is competition below the word level and Lexical Phonology wins, economy calls for the elimination of Prosodic Phonology at and above the word level. Hence, as pointed out by Kleinhenz (1998:22), there is a natural alliance between Lexical Phonology and direct syntax approaches (see §407): the latter denies the existence of prosodic constituency in the postlexical area by making phonology directly refer to syntactic categories.

368 Chap 10: Prosodic Phonology, on the representational side theory. Selkirk could have made the same point in an environment where prosodic constituents – or the grid for that matter – are unheard of.122 Therefore the difference between Selkirk (1984) and Nespor & Vogel's (1986) mainstream position that promotes peaceful coexistence does not concern Prosodic Phonology. It is simply about a broader view on the interface, about a view beyond the purely representational transmission of information for which Prosodic Phonology was designed. While Nespor & Vogel do not have any opinion on how the procedural side of the interface should work (and hence can live with whatever comes the way), Selkirk defends a precise position.

433

8.3.4. Inkelas (1990): Prosodic Phonology with the empty shell of Lexical Phonology Inkelas (1990) may also be counted as support for the position that Lexical Phonology has to go because it is superfluous. On her analysis, nothing is left of Lexical Phonology, even though she formally leaves the empty shell in place. Inkelas' (1990:37ff) leading idea is to eliminate syllables and feet from the Prosodic Hierarchy (see also Zec 1988, Fitzpatrick-Cole 1996:306f). 123 One good reason for this move was already discussed in §401: syllables and feet owe nothing to mapping rules; their existence and shape is exclusively motivated by domestic phonological phenomena (they are bottom-up constructions). Since the Prosodic Hierarchy is about transmitting extra-phonological information to phonology, syllable and feet are not concerned. 122

123

The only echo of Selkirk's raid on Lexical Phonology that I have come across is Mohanan (1982:97ff), who was working on a manuscript of Selkirk (1984). Selkirk (1984) and Inkelas (1990, on which more in the following section) set aside, Mohanan is also the only text that I am aware of where the conflict between Lexical Phonology and Prosodic Phonology is made explicit. Mohanan ends up proposing a modification of Selkirk's arboreal solution that allows for the incorporation of lexical strata. Hannahs (1995:61ff) argues along similar lines. According to him, the introduction of prosodic constituency into the lexicon allows us to do away with the Lexicon altogether. Unlike Inkelas, however, Hannahs (1995:66f) takes this result to be particular to French: it depends on the language-specific situation and may not be generalised. In Italian for example, he argues, the regular stratal analysis of Lexical Phonology is compelling.

Relations with Lexical Phonology and the metrical grid 369

Another argument developed by Inkelas (1990:40ff) builds on mismatches between syllable/foot structure and prosodic structure. Inkelas presents evidence to the end that feet may need to straddle the boundary of phonological words, thus violating the basic principle of prosodic architecture, the Strict Layer Hypothesis (§383). That is in her example, feet are not exhaustively contained in words. Up to this point, nothing has been said regarding the relationship between Prosodic Phonology and Lexical Phonology. If it is true that syllables and feet are not prosodic constituents, a waterproof division of labour à la Hayes (1989 [1984]) seems to be possible: Prosodic Phonology manages the interface with syntax, while Lexical Phonology takes care of the interaction with morphology. Inkelas (1990:28ff) dismisses this perspective on the grounds of Indirect Reference: according to her, Lexical Phonology makes direct reference to morphological categories and is therefore incompatible with Prosodic Phonology. (177) "In its standard formulations, the theory of Lexical Phonology […] is incompatible with the prosodic hierarchy theory. As part of its crucial claim that phonological rules can apply at each step of word formation, Lexical Phonology has assumed that these rules look directly at morphological structure, applying within strings of morphemes supplied to them by the morphology. However, this latter assumption is clearly incompatible with the Indirect Reference Hypothesis." Inkelas (1990:29f)

In order to put Lexical Phonology in conformity with Indirect Reference, Inkelas argues, prosodic constituency must extend to the area below the word. Like syntactic information at and above the word level, morphological information must be translated before Lexical Phonology can make use of it. The carriers of morphological information are part of the Prosodic Hierarchy, but syllables and feet are out of business. Inkelas (1990:33f) therefore introduces a new set of prosodic constituents which, just like their higher level peers, are created by mapping rules (two other mechanisms, irrelevant here, may also bear on their shape). She claims that there are as many layers below the word as there are lexical levels in Lexical Phonology (but does not name them). In the end, Selkirk's (1984) and Inkelas' (1990) positions are quite superposable, even as far as representational detail is concerned (compare the trees in Selkirk 1984:77 and Inkelas 1990:34f, the former are further discussed in §540). Inkelas' understanding, however, is that her conclusions do not harm Lexical Phonology (she calls her model Prosodic Lexical Pho-

370 Chap 10: Prosodic Phonology, on the representational side nology) since her purpose precisely is to restore Lexical Phonology in the legal frame of Indirect Reference. The critical question, then, is whether Inkelas' model uses strata in order to do morpheme grouping, or whether the regular morpho-syntactic tree does this job. Her implementation clearly follows the latter option, which means that her system has nothing in common with Lexical Phonology (whose heart are strata, see §432).

434

8.3.5. Lexical Phonology does not violate Indirect Reference The purpose of the preceding section was to present a contribution to the debate regarding the relationship between Prosodic Phonology and Lexical Phonology. Whether Inkelas (1990) is right or not is a different issue. I believe she is not, for the premise of her reasoning is wrong: Lexical Phonology does not violate Indirect Reference. Since this question is more generally relevant, some discussion is provided below. Inkelas (1990) believes that Lexical Phonology does not comply with Indirect Reference because of interactionism (§146): the interleaving of morphological rules and phonological activity prompts a situation where phonological rules apply only to a certain chunk of the word. According to Inkelas, this makes phonology sensitive to a given morphological piece, i.e. one that is identified on morphological, rather than on phonological grounds. This is all true, but Inkelas confuses "phonology applies to X" with "phonology has chosen X". Indirect Reference is only violated in the latter case, i.e. when the structural description of a phonological rule somehow mentions the morphological identity of the string to which it applies. This is never the case in Lexical Phonology, where the stratal architecture on the contrary has eliminated almost all reference to extra-phonological information (see §§165f). To an extent unrivalled by other interface theories, Lexical Phonology thereby affords purely phonological rules in the sense that they do not mention any (translated or untranslated) morpho-syntactic information. Nothing is wrong with phonology interpreting only substrings of words: this is how cyclic derivation works. When a substring is computed, phonology does not know anything about its morphological status, not any more than when a whole word is interpreted. Phonology works on whatever is submitted for interpretation, and as long as it does not participate in the selection of the chunks at hand, there is no direct reference. All implemen-

Relations with Lexical Phonology and the metrical grid 371

tations of cyclic derivation respect the purely computational role of phonology: chunk-submission is decided elsewhere in SPE, Lexical Phonology, Selkirk (1984), Kaye (1995) or Distributed Morphology. 435

8.4. Lexical Phonology II: peaceful coexistence (soft version)

436

8.4.1. How labour is supposed to be divided: "direct reference to morphological structure" Contrasting with Hayes' (1989 [1984]) division of labour and Selkirk's elimination of Lexical Phonology (or Inkelas' transformation of Lexical Phonology into prosodic constituency), a third option is peaceful coexistence of Lexical Phonology and Prosodic Phonology. This is Nespor & Vogel's (1982:226, 1986:18f, 27ff) take which, like in other areas of Prosodic Phonology, has become mainstream. Peaceful coexistence means that certain phonological processes which have a morphological conditioning are managed by the tools of Lexical Phonology, while others make reference to prosodic constituency. The rationale according to which the labour should be divided is explained by Nespor & Vogel (1986:18).124 (178) "There are, of course, also phonological processes that must make direct reference to morphological structure and/or specific morphological elements in the formulation of their environments. Such rules must clearly be ordered before the rules that apply in a strictly phonological domain, since, as we have mentioned, at this point the morphological structure is no longer available. Since such morpho-phonological rules are different from purely phonological rules that are the subject of the present work, we will not discuss further how this type of interaction between morphology and phonology must be handled. We will assume here that morpho-phonological processes are accounted for by a different type of mechanism, such as the one proposed recently within the framework of lexical phonology." Nespor & Vogel (1986:18)

The critical contrast is thus between purely phonological rules and those that make "reference to morphological structure and/or specific mor-

124

This position is reiterated by Nespor & Ralli (1996:361f). Nespor & Vogel (1982) on the other hand do not address the issue.

372 Chap 10: Prosodic Phonology, on the representational side phological elements". The Prosodic Hierarchy, of course, counts as a purely phonological item.

437

8.4.2. There is no natural division of rules into "purely phonological" vs. "morpho-phonological" There are two worries with this definition: for one thing, it is not true that rules in Lexical Phonology make direct reference to morphological structure and/or specific morphological elements. This seems to have been a common belief in the mid 80s: it also grounds Inkelas' (1990) idea that Lexical Phonology violates Indirect Reference (§433). It was shown in §434 that this is not the case. Rather the contrary is true: Lexical Phonology has eliminated (almost) all reference to morphological diacritics and boundaries (§§163,213). The second worry is that there are no grounds for deciding whether or not a given rule makes "reference to morphological structure and/or specific morphological elements". There is no natural division of that kind among phonological rules. Nespor & Vogel (1986) do not mention that on the count of Prosodic Phonology, mapping rules decide whether or not (and how) morpho-syntactic structure is transformed into "purely phonological" prosodic structure. For any phonologically relevant morphological division, then, who decides whether there is a mapping rule that translates it into prosodic constituency? The mapping mechanism is poorly understood (§386) and entirely unconstrained (§393): anything and its reverse can be a mapping rule. Hence whether or not some morphological division is reincarnated in terms of prosodic constituency depends on the taste of the analyst – there is certainly nothing in the phenomena themselves that predestines them for translation (or makes them recalcitrant). Another way to put things is to ask what would be a phonologically relevant morphological division that is not translated into prosodic constituency. And why should there be a prohibition of translation in the first place? Nespor & Vogel (1986) seem to take a landscape for granted where these questions have been answered, hence where rules fall into two natural classes: one that makes reference to translated morphological properties (which are thus managed by Prosodic Phonology), another that makes reference to non-translated morphological properties (which are dealt with by Lexical Phonology). Such a division does not exist, and the only way to introduce it would be by an arbitrary decision of the analyst.

Relations with Lexical Phonology and the metrical grid 373 438

8.4.3. Two examples Let us consider two examples from Nespor & Vogel (1986:28f) for the sake of illustration. In English, the place of the prefix-final nasal assimilates to the following stop in i[m]-possible (compare with i[n]-audible), but not in u[n]-predicatable. Nespor & Vogel argue that (179) "the formulation of Nasal Assimilation must take into account information about the morphological structure of the words in question. This type of rule is, therefore, different in nature from the strictly phonological rules we will consider in subsequent chapters of this book." Nespor & Vogel (1986:28)

They thus try to sell a decision that was made by the analyst for a natural property of the rule in question. That is, Nespor & Vogel do not mention the reason why the level 1 vs. level 2 distinction which is at stake (see §§94,167 where nasal assimilation was discussed at greater length) is not represented in prosodic structure: because somebody has decided that mapping rules are unable to divide affixes into two classes, one sharing a prosodic constituent with the root, the other ending up in a constituent of its own. Nothing, however, withstands this to be done: mapping rules would just need to be tailored in that sense. In actual fact, this option is chosen by Rubach & Booij (1984:12ff) and Booij (1992:53): the prefix un-, is made a phonological word of its own. A purely phonological rule, then, can assimilate nasals to following obstruents within phonological words. The prefix in- will be affected since it is merged with the prosodic word of the root ([in-possible]PW), while un-predictable offers no PW-internal nasalobstruent sequence and hence escapes assimilation ([un]PW[predictable]PW]. Nasal Assimilation is thus a good witness of the fact that the "purely phonological" treatment of an alternation is a matter of taste of the analyst, rather than a natural property of the phenomenon itself. Another example quoted by Nespor & Vogel (1986:29) is categorysensitive stress placement in English: récord, pérvert, éxtract (nouns) vs. recórd, pervért, extráct (verbs) and the like. Nespor & Vogel declare this kind of phenomenon out of scope for Prosodic Phonology visibly because they cannot imagine that mapping rules may be sensitive to major categories. Nothing, however, prevents mapping rules from taking into account morpho-syntactic categories. If the analyst decided to write a category-

374 Chap 10: Prosodic Phonology, on the representational side sensitive mapping rule, the alteration at hand would become "purely phonological" and thus fall into the scope of Prosodic Phonology.125 In sum, there does not appear to be any morphological division that mapping rules could not translate into prosodic constituency. Hence peaceful coexistence is built on sand: two mechanisms can hardly coexist if nobody knows how the labour is divided. 439

8.4.4. Peaceful coexistence supposes split mapping Peaceful coexistence also bears on the position of mapping in the overall derivational architecture. Nespor & Vogel (1986:18f) point out that mapping cannot simply apply after the Lexicon of Lexical Phonology since it relies on morphological divisions which, according to Lexical Phonology, are erased at the end of each stratum (bracket erasure, see §167). That is, morphological structure is no longer available postlexically – the no look-back mechanism in Lexical Phonology that embodies the insight of Praguian segregation (§174). The solution that Nespor & Vogel (1986:19) favour divides mapping rules into two sets: those that work on morphological structure, and those whose input is syntactic structure. The former apply before Lexical Phonology, while the latter intervene postlexically. Ten years later, Nespor & Ralli (1996:361f) confirm this split mapping mechanism. What this means is that Praguian segregation (§153) is catered into Prosodic Phonology, and that this separate treatment of morphology and syntax is a direct consequence of peaceful cohabitation. Certainly an unwarranted consequence since one goal of Prosodic Phonology originally was to provide a uniform screen (the Prosodic Hierarchy) for the reference to all phonologically relevant morpho-syntactic information.

440

8.5. Lexical Phonology III: peaceful coexistence (radical version) Finally, a fourth way of looking at the relationship between Lexical Phonology and Prosodic Phonology is a radicalised version of Nespor & Vo125

Independently of the issue discussed here, category-sensitive stress is an obvious and well-known problem child for the generalisation discussed in §398 according to which phonology may react on morpho-syntactic structure, but not on labels. Further discussion is provided in §752.

Relations with Lexical Phonology and the metrical grid 375

gel's (1986) peaceful cohabitation: below the word level, phonological rules make random reference to either morphological structure or prosodic constituency, and there is no principle that defines which phenomena use one or the other strategy. Lexical Phonology and Prosodic Phonology are thus both candidates for managing everything that is going on below the word level; whether one or the other comes to bear is random and unimportant. This approach is argued for by Rubach & Booij (1984) as well as by Booij (1988, 1992), Booij & Lieber (1993), Szpyra (1989:183ff), Vogel (1991) and Hannahs (1995:2); its modern heirs are Stratal OT and DOT (see §§477,483). These authors and Nespor & Vogel (1986) march in opposite directions in order to set up peaceful coexistence: while the latter come from Prosodic Phonology and try to integrate the mechanisms of Lexical Phonology, the former root in Lexical Phonology and attempt to incorporate the Prosodic Hierarchy into their system. Following this track, the literature quoted always aims to show that prosodic constituency, including the phonological word, must be available in the Lexicon. Examples are Rubach & Booij (1984), Booij (1988, 1992), Inkelas (1990), Vogel (1991) and Hannahs (1995). At the same time, thus, these authors take Lexical Phonology for granted, to the effect that the question whether Prosodic Phonology could take over the functions of Lexical Phonology is not a subject of debate. In this approach, the difference between Lexical Phonology and Prosodic Phonology becomes a question of perspective: both theories cover the same empirical field, but look at it from the representational or the procedural point of view. (180) "Lexical Phonology concerns the derivational aspect of phonological theory, and Prosodic Phonology concerns the representational aspect of phonology." (Booij 1988:63).

Also, Rubach & Booij (1984) and Rubach (1984:221ff) use prosodic constituency in order to account for so-called bracketing paradoxes (see §243). The structure of un-happi-er should be [[un [happy]] er] since it means "more unhappy", rather than "not more happy": un- has scope over the rest of the word. Phonologically, however, the structure should be [un [[happi] er]] since the synthetic comparative -er attaches only to stems with maximally two syllables (big - bigger, happy - happier); adjectives with more syllables have an analytic comparative (beautiful -

376 Chap 10: Prosodic Phonology, on the representational side *beautifuller/more beautiful). In order to get the synthetic un-happi-er, thus, -er should be concatenated before un- is added. Rubach & Booij (1984) propose that in most cases phonological structure is a function of morphological structure, but sometimes both do not coincide. That is, the phonological constituent structure is the Prosodic Hierarchy, which according to non-isomorphism may not replicate morphosyntactic divisions. Unhappier, then, is a case of readjusted mapping: the structure that is used by morpho-syntax is the one required by the scope of un-, while prosodic structure is looked at by the phonological mechanism that counts syllables. That is, the prosodic phrasing of unhappier isolates the prefix: [un]m[happy]m is the output of mapping (subscript "m" is Rubach & Booij's prosodic constituent "mot", the classical prosodicwWord). The syllable-counting mechanism then applies to the mot, and hence allows for the attachment of the synthetic -er. Without this solution, the paradox is a serious problem for Lexical Phonology that Rubach & Booij (1984) assume: lexical strata imply that phonology applies after every step in the word formation. Mismatches are thus predicted not to exist. On the analysis presented, however, the "outsourcing" of allomorph selection to the Prosodic Hierarchy saves Lexical Phonology. Therefore Rubach & Booij (1984:24) conclude their article by saying that "the theory of prosodic phonology also functions as a wellmotivated 'protective belt' for the theory of Lexical Phonology." Finally, Booij (1988, 1992) takes the fact that cyclic, hence lexical rules, refer to syllable- or foot structure as a proof that prosodic constituency, which includes these categories, is present in the lexicon. On the other hand, lexical rules make also reference to morphological structure. According to Booij (1988:72), this is witnessed by rules that refer directly to morpho-syntactic labels such as "the final syllable is extrametrical in adjectives" (a rule proposed by Hayes 1982). On this count, Indirect Reference is thus nothing that marshals interface activity: direct syntax cohabitates peacefully with prosodic constituency and the derivational system of Lexical Phonology. In other words, everything that interface theory has produced since SPE is scrambled. The cocktail, then, may well be made of ingredients that are incompatible with one another or do the same job. Manifoldness allows for a larger empirical coverage and hence appears to be an advantage.

Relations with Lexical Phonology and the metrical grid 377 441

8.6. Conclusion At least two conclusions may be drawn from the foregoing pages. There is competition between Prosodic Phonology and Lexical Phonology at and below the word level, and peaceful coexistence does not work in the way imagined by Nespor & Vogel (1986). No property of morphologically conditioned phonological phenomena allows for a classification into clients of Prosodic Phonology and clients of Lexical Phonology. No waterproof division of labour between the two theories may thus be elaborated, which means that Nespor & Vogel's (1986) perspective is not workable. For a given phenomenon, analyses using either the technology of Lexical Phonology or Prosodic Phonology are possible, and the analyst may well be unable to decide which one fares better or is more appropriate. We are thus left with the radical option proposed by the Lexical Phonology quarters: anything goes. A given phenomenon may randomly have a stratal or a prosodic analysis. This is certainly not how one would expect a competition to be solved in science. It is common to say that in our current understanding we are unable to tell two competitors apart. But granting them equal rights without willingness to show that either (or both) need(s) to be abandoned is a questionable position. We will see in §483 that the modern heirs of Lexical Phonology, Stratal OT and DOT, continue this strand: both stratal and prosodic technology is available, and there is no division of labour (even though Stratal OT attempts to restrict the morphological impact on phonology, see §490). Selkirk's (1984) option (and, to a certain extent, Inkelas' 1990) appears to be more consistent: she acknowledges that two solutions compete for the same set of phenomena and draws the conclusion that one has to go. Calling on Occam's Razor, she argues that Lexical Phonology is redundant since it needs extra technology (strata) that come for free when cyclic derivation is practised on the grounds of regular morpho-syntactic structure. She also makes the point that the empirical scope of Lexical Phonology (morphology) is a proper subset of the scope of Prosodic Phonology, which covers both the interaction with morphology and syntax.

378 Chap 10: Prosodic Phonology, on the representational side 442

9. Prosodic Morphology

443

9.1. Representational and non-representational incarnations Before closing this chapter, due mention needs to be made of Prosodic Morphology. Not so much because this theory is concerned with the interface (it is not), but because it was very influential and eventually opened out into Optimality Theory. Indeed, two central mechanisms of OT, correspondence and alignment, originate in Prosodic Morphology, as McCarthy & Prince (2001:vii) explain in the introduction to the ROA version of the original 1986 manuscript (McCarthy & Prince 1986, of which there is also a 1996 version: McCarthy & Prince 1996 [1986]). Prosodic Morphology has known two periods, representational and anti-representational. We will only be concerned with the former in this section, which expressed relevant generalisations by means of the units of the Prosodic Hierarchy (up to the word level, i.e. moras, syllables, feet and eventually the prosodic word). After the advent of OT, here as elsewhere representational units could not be considered linguistic primes anymore for the only decision-making entity is the constraint chamber: generalisations must be expressed in terms of processes, not of objects (see Scheer 2010a). The constraint-based version of Prosodic Morphology is called Generalized Template Theory. The rich body of literature of the representational period before templates per se were abandoned includes McCarthy & Prince (1988, 1990) and many others; the OT-inspired dissolution of templates into constraint interaction was engaged in McCarthy & Prince (1994, 1995a,b) and is currently developed in ongoing work. Downing (2006) provides a comprehensive and informed overview of both periods and shows how they are interconnected.

444

9.2. Representational period: prosodic morphemes are units of the Prosodic Hierarchy

445

9.2.1. A typical example: reduplication In its representational incarnation, Prosodic Morphology uses the autosegmental structure of the Prosodic Hierarchy in order to analyse purely phonological patterns that are related to morphological activity and call for a phonological generalisation. A good example, typical and foundational for

Prosodic Morphology 379

Prosodic Morphology, is reduplication: the cross-linguistic record shows that reduplicated items are never arbitrary in number and sequence. That is, the reduplicative morpheme picks segments from the base according to a non-arbitrary mechanism where the size and the linear order of reduplicated items is predetermined. Descriptions of reduplication, then, necessarily mention the number and linear order of consonants and vowels: "take the first CV pair of the base and concatenate it to the left of the base" would be a typical instruction. The goal of Prosodic Morphology is to capture the fact that size (and linearity) restrictions on the output of morphological operations are not random: it is not the case that any arbitrary string of segments may be the required output shape of reduplication (or any other relevant process). Rather, the mechanics of constant morpheme shape conform to recurrent patterns, and these patterns, Prosodic Morphology contends, are surface manifestations of the lower units of the Prosodic Hierarchy (up to the prosodic word). In other words, the real identity of reduplicative morphemes is not segmental (although some segmental properties may be pre-specified): their lexical identity is a syllable or a foot, which are then "filled" by segmental material from the base and impose their inherent restrictions on size and linearity (no arbitrary strings).

446

9.2.2. Another typical example: (Semitic) templatic morphology Another typical area of competence of Prosodic Morphology is Semitic templatic morphology. In Classical Arabic for example, so-called measure I forms of verbs conform to the template CVCVC- (e.g. katab- "to write"), while measure II forms follow CVCCVC- (kattab-, gemination of the medial consonant). Since a specific meaning is associated to measures (I is unmarked, II is intensive/iterative, and there are a number of measures), they must be considered morphemes. Therefore the bare information regarding the sequences of Cs and Vs must be somehow stored in the lexicon independently of the segmental content, but together with their meaning, i.e. "unmarked", "intensive/iterative" etc. McCarthy (1979) transposes the traditional analysis of Arabic grammarians into the emerging autosegmental frame, but leaves it at an amorphous list of templates that are made of sequences of Cs and Vs (to which melody is attached). McCarthy & Prince (1996 [1986]:2ff) argue that this misses the generalisation that these CV-strings are not arbitrary. Rather, they conform to well-formedness conditions of units of the Prosodic Hier-

380 Chap 10: Prosodic Phonology, on the representational side archy: Arabic for example does not have any template of the kind CVCCCVC-. This is, McCarthy & Prince argue, because templates imitate the prosodic structure of the language at hand: in our case, Arabic syllabification disallows triconsonantal clusters. Therefore the constant shape of measure I, II etc. morphemes must be recorded in terms of prosodic units, syllables in our case, rather than segments or Cs and Vs. 447

9.2.3. Moras, syllables and feet do not carry any morpho-syntactic information This line of reasoning was extended to all instances where languages show morphologically defined constructions whose output has a constant shape. So-called prosodic morphemes capture relevant generalisations: they are made of units of the Prosodic Hierarchy which impose size and shape restrictions, thereby producing the canonical forms observed. Aside of reduplication and templatic morphology, other phenomena that are typically treated in terms of prosodic morphemes are minimal word constraints, infixation and fixed segmentism. In any event, we are only talking about phenomena and representational items that concern chunk sizes at and below the word level: Prosodic Morphology is about (prosodic) morphemes. Interestingly, the aforementioned (§401) front line between mapping-created, top-down constructed units (the prosodic word and above) and truly phonological units which are the result of a bottom-up construction and owe nothing to mapping also surfaces in the discussion of Prosodic Morphology: while the prosodic word was explicitly included into the list of prosodic items that can make prosodic morphemes by McCarthy & Prince (1996 [1986]:6), its presence was called into question later on (see the discussion in Downing 2006:8f). In its representational skin, Prosodic Morphology may be characterized as an application of the autosegmental idea (which was still fresh in the mid-80s) to the phenomenon of constant shape and canonical forms. The "prosodic" aspect is more a matter of the terminology chosen than a claim regarding the Prosodic Hierarchy: all autosegmental categories above the skeleton were organized in the Prosodic Hierarchy, hence anybody who uses units of this hierarchy will come across the word "prosodic". Prosodic Morphology as it stood then was thus unrelated to the subject matter of the present book, i.e. the transmission or encoding of morpho-syntactic information to/in phonology: relevant units – moras, syllables, feet – do not carry any such information.

Prosodic Morphology 381 448

9.3. Generalized Template Theory: morphology marshals phonology The take of Generalized Template Theory is significantly different. Constant shape is not interpreted anymore as a consequence of phonological units whose genesis owes nothing to morpho-syntactic information. The units in question – moras, syllables, feet – are still the same, but their genesis is now impacted by morpho-syntax: they are correlated – aligned is the technical term – with three types of morphological units, the affix, the root and the stem. This means that the parsing of lower level units into higher level units – for example the syllabification of segments – obeys regular phonological rules, but also depends on instructions regarding the edges of affixes, stems and roots. In other words, the units of the Prosodic Hierarchy below the word level are now subjected to mapping as well: they cease to be purely bottom-up constructions. In order to see how alignment constraints transmit morpho-syntactic information to phonology, let us consider a simple example from German (Kleinhenz 1998:39f). The word auf-essen "to eat up" is made of a stem (essen) and a prefix (auf). It is realised with an epenthetic glottal stop auf-ʔessen. The glottal stop is non-contrastive: it is always filled in wordinitially in case the word begins with a vowel underlyingly. According to Kleinhenz' analysis, the pronunciation at hand is the result of the ranking ONSET >> ALIGN (Stem, L, PrWd, L) >> DEP. The alignment constraint demands that the beginning of the stem coincides with the beginning of a prosodic word. Regular mapping would express the same requirement by the fact of assigning two separate prosodic words to the prefix and to the stem. Now ONSET demands the presence of a consonant in the first syllable of essen – hence the epenthetic glottal stop. The concurrent solution that syllabifies the prefix-final consonant into the onset of the stem fails because of ALIGN: in au.fessen the stem and the prosodic word would be misaligned, and this is fatal since ALIGN outranks the epenthesis-hostile DEP. In a language with the reverse ranking, au.fessen without glottal stop would be optimal. Alignment constraints thus inject morphological information into syllabification and footing, which in the earlier version of Prosodic Morphology parsed the string on the basis of phonological information alone. Downing (2006) describes the old ambition (constant shape) and the new way to go about it in Generalized Template Theory as under (181) below.

382 Chap 10: Prosodic Phonology, on the representational side (181) "The central proposal of the Generalized Template Theory (GTT) of prosodic morpheme shapes is that prosodic morphemes have a restricted repertoire of prosodic shapes because they draw on the canonical shapes of a restricted repertoire of morphological categories. These canonical shapes follow from general theoretical principles correlating particular morphological categories (Stem, Root, Affix) with particular prosodic constituents and from a constraint grammar defining the canonical shapes as unmarked." Downing (2006:35)

The introduction of this system leads over to the chapter on Optimality Theory where constraint-based mapping is generalised: as there are no rules there can be no mapping rules either. But of course there is mapping, i.e. the transmission of morpho-syntactic information to phonology, and its encoding in from of the Prosodic Hierarchy. The ALIGN (and WRAP) constraint families are central to this way of creating the Prosodic Hierarchy. The take of OT regarding the interface in general and constraint-based mapping in particular is introduced in the following chapter.

449

10. Conclusion: translation yes, buffer no Looking back at what has been done on the representational side of the interface since SPE, it appears that Prosodic Phonology has inherited the central concepts of Chomsky & Halle, which occasionally may have been (re)named: Indirect Reference, the existence of a Translator's Office, nonisomorphism, mapping, the Black Box (readjustment rules). Prosodic Phonology is to be credited for weaving a consistent theory based on these ingredients. Namely the institutionalisation (and exceptionlessness) of Indirect Reference and a Translator's Office that is located in modular no man's land is a landmark in the evolution of interface architecture. Even though this was done for the wrong reason (non-isomorphism), it is the faithful embodiment of intermodular communication in Fodorian modular architecture (on which more in the Interlude §586). Strangely enough, though, the modular argument (which is much akin to structuralist Level Independence, see §415) does not appear to be afforded in the Prosodic Phonology literature. Valuable progress has also been made on the empirical side: Prosodic Phonology has produced a large number of detailed studies regarding morpho-syntactic influence on phonology. The intensive study of the mapping mechanism, although ultimately still inconclusive, has greatly broadened our knowledge of how morpho-syntactic information seeps into phonology.

Conclusion: translation yes, buffer no 383

Finally, an internal and a global concern need to be mentioned. The former is about the hybrid character of the Prosodic Hierarchy, which encompasses layers that store morpho-syntactic information (the buffer) as much as constituents which have got nothing to do with extra-phonological information. The former are (top-down) constructed by mapping rules, while the latter (syllable and feet) are bottom-up constructions and exist in domestic phonology in absence of any issue related to morpho-syntactic information (§401). This hybridity is certainly not recommendable. An interface theory ought to provide for a phonological representation of phonologically relevant extra-phonological information, and for nothing else. Making the same formal device endorse domestic phonological structure as much as the output of the Translator's Office is a misconception of the interface. This issue was pointed out most sharply by Inkelas (1990) (§433). The global concern is about the body of Prosodic Phonology itself: the Prosodic Hierarchy is made of domains, and it is a diacritic. The traditional interface currency, boundaries, was evacuated in favour of domains with no good reason and hardly any discussion, in any event in absence of contrastive argumentation. The diacritic argument levelled against boundaries was made in the erroneous belief that the alternative, domains, is nondiacritic. The discussion of Prosodic Phonology is consensual in at least one point: whatever the way of encoding interface information, diacritics do not qualify. If in an autosegmental guise, the Prosodic Hierarchy is a diacritic as much as SPE-type boundaries – hence both have to go (Scheer 2008a). The key to the problem, then, is the fact that the diacritic character of boundaries is but one aspect of their identity. It has constantly been confused with the property that really sets boundaries apart from domains: local vs. non-local action. The real question which is still pending at the end of this discussion, thus, is whether higher-level intervention in phonology should be local or non-local (more on this issue in §706). Whatever the answer, a non-diacritic implementation is mandatory.

450

Chapter 11 Optimality Theory

451

1. Setting the scene

452

1.1. OT makes no contribution on the representational side Optimality Theory is a theory of constraint interaction, hence of computation – not of representation. There are no genuine OT representations: the theory makes no claim regarding objects; hence using this or that type of representation is the responsibility of the analyst and subject to debate. This state of affairs is remarkably different from the situation that prevails in all other frameworks which we have come across thus far: theories propose a particular representational vocabulary, and this vocabulary is part of their identity. Hence the basic phonological currency in structuralist times were phonemes; SPE then installed segments (or distinctive features), before autosegmental structure became the regular representational currency in the early 80s. It was pointed out in §72 that each stage of the evolution of phonological representations comes with its own interface theory – or rather, whatever the current representational currency is also the output of translation. Hence morpho-syntactic information was carried by (juncture) phonemes in structuralist times, by [-segment] segments (boundaries) in SPE, and by autosegmental prosodic constituency since the 80s. Given that OT does not make any genuine contribution on the representational side, there is no additional step to be expected on this path. OT uses the representational stock of the 80s, which in the realm of the interface means that the Prosodic Hierarchy is taken over. It is simply adapted to the constraint-based environment.

453

1.2. Representations marshalled by constraints Another aspect of OT is that representations are not granted any sovereign arbitral award. The only thing in OT that determines the (relative) grammaticality of an item is constraint interaction. That is, an eventual arbitral award issued by a representation (such as line crossing for example) may well be outranked by some other constraints. Therefore the rule of the game is quite different when compared to theories that are not constraint-based:

386 Chap 11: Optimality Theory when a representation is ill-formed, this may or may not mean that it is out – representations as such do not mean anything (Scheer 2010a). On the other hand, representations are also the result of constraint interaction: the output orientation of OT, embodied for example in Richness of the Base, does not allow for specific structure in underlying forms. Hence structure must be emergent, rather than given. It is therefore a product of constraint interaction, rather than primitive (e.g. Itô et al. 2004). In sum, the almightiness of constraints in OT also subordinates representations, which originate in the constraint chamber and are marshalled by it.126 Like other representational objects from the 80s, prosodic constituency is imported into the OT environment without modification: the familiar categories such as the phonological word, the phonological phrase etc. are taken over. The way they stock and unload morpho-syntactic information, though, has been adapted. That is, prosodic constituency is not created by mapping rules anymore, and phonological rules do not make reference to it. Instead of rules, both mapping and reference are assured by a set of specialised constraints (§457): the ALIGN and WRAP families express the "desire" of a certain matching between units of the Prosodic Hierarchy and morpho-syntactic structure. These constraints, as violable as all others, are in competition with purely phonological constraints in the same ranking. The result, then, determines the kind of morpho-syntactic influence that phonology experiences.

454

1.3. Anti-cyclicity and the serial offspring of Lexical Phonology Strictly parallel computation is the heart of OT.127 In a parallel environment, derivational mechanisms are outlawed. This is why OT has produced 126

127

The balance of structure (representations) and process (computation: constraints) is the prism that St. Anderson (1985) uses in order to look at the evolution of phonology. In Vol.1:§305 (also Scheer 2003b), this issue is discussed in the light of OT, where computation is king. The factual elimination of structure as a player in OT is fully assumed and actually welcomed by the mainstream literature. In the outlook of his textbook on OT, Kager (1999:420f) for example writes that "a second major development that is likely to continue is a reduction of the role of representations in favour of constraint interaction." Although there is also a version where computation is serial, albeit in a way that is different from extrinsic rule ordering (see Scheer forth): Prince & Smolensky's (1993) Harmonic Serialism and its revival by McCarthy (2007).

Setting the scene 387

a body of anti-cyclicity literature, i.e. where it is argued that OT is incompatible with cyclic spell-out. Non-serial alternatives are therefore developed: in the same way as for the closely related issue of opacity, several cyclicity killers have been proposed (which appear to be competing, §464). This has also led to new ways of implementing morpheme-specific phonologies: unlike on the stratal count where strings necessarily run through later strata and experience the associated stratum-specific computation, a parallel alternative subjects strings only to the computational system that they are specified for. Co-phonologies and indexed constraints represent this architecture, where morpheme-specific mini-grammars act in parallel, rather than being serially ordered (§478). Since the stratal architecture is derivational (strata are serially ordered), Stratal OT and DOT, the modern continuators of Lexical Phonology, were unwarranted in the original parallel environment of OT (§483). The question, though, is whether the parallel ambition of OT applies only to phonology proper or extends to interface operations (§464). While the answer is not obvious a priori, the practice of OT at least in the 90s clearly imposed parallel computation also at the interface. This ties in with another issue: current practice in OT is in overt conflict with basic generative principles, a fact that is hardly reflected in the literature. That is, modular contours are blurred if existent at all (§§469,523), inside-out interpretation is abandoned together with cyclicity (§468), and the procedural aspect of the interface seems to be denied altogether (thus dispensing with Interface Dualism). Relevant in this context is also an alternative to morpheme-specific phonologies that is entertained in OT: so-called interface constraints make direct reference to "designated" morpho-syntactic categories and therefore overtly violate the principle of Indirect Reference (§406); direct syntax approaches are also common practice elsewhere in OT (ALIGN and WRAP). The faulty position of OT in regard of modularity is discussed at greater length in §523. We have seen in previous chapters, though, that OT is not the only modularity offender; a comparison with other theories is provided in §702. Finally, van Oostendorp's Coloured Containment (§503) and Orgun's Sign-Based Morphology (§512, an application of HSPG to the interface) The book on the derivational residue in OT that Ben Hermans and Marc van Oostendorp have edited (Hermans & van Oostendorp (eds.) 1999) and especially their introduction to the volume provide an (early) overview of the relationship of OT with serialism.

388 Chap 11: Optimality Theory are presented. These approaches share a monostratal view on the interface where translation (mapping) does not exist and morpho-syntactic information is constantly available during phonological computation (direct syntax again).

455

2. Adaptation of Prosodic Phonology to the constraint-based environment

456

2.1. Introduction The constraint-based environment either allows or requires adaptations of the familiar Prosodic Phonology model to the general OT architecture. The disentangling of Selkirk's Strict Layer Hypothesis (SLH, §383), which was monolithic in the 80s, falls into the former category, while anti-cyclicity and rule-based mapping are instances of the latter. The following pages discuss the adaptation of rule-based mapping (§457) and the Strict Layer Hypothesis (§461) (anti-cyclicity is considered in §464 below). Finally, we also look at how OT treats the mapping puzzle – a resident question in the study of the interface (§463).

457

2.2. Constraint-based instead of rule-based mapping

458

2.2.1. Mapping understood as the coincidence of constituent edges: ALIGN The Prosodic Hierarchy represents the (mis)match of morpho-syntactic constituency and phonologically relevant chunks of the linear string. The operation that relates both, mapping, may thus be described as the (mis)alignment of the edges of morpho-syntactic constituents with the edges of prosodic constituents. As far as I can see, this implementation of mapping has been introduced by Selkirk (1984:52ff), who builds prosodic structure (the metrical grid in her case) by alignment rules (e.g. Basic Beat Rules, Demi-Beat Alignment DBA, see §426). Constituent margins then become the centre of interest two years later in Selkirk's (1986) edge-based mapping (§386). Finally, McCarthy & Prince (1993) generalise alignment to more empirical situations and make it the central tool for the interface with morpho-syntax (see Itô & Mester 1999a, McCarthy & Prince 2001:vii, Peperkamp 1995:227ff for a historical overview).

Adaptation of Prosodic Phonology to the constraint-based environment 389

Today ALIGN is a constraint family with a uniform template: the left or right edge of a given unit coincides with the left or right edge of another unit. The units in question may be phonological, morphological or syntactic, and both units involved in an alignment constraint may belong to the same area (phonology, morphology, syntax, see for example Yip 1998).

459

2.2.2. Parallel mapping: translation and reference to prosodic constituency are conflated The following example from German, repeated from §448 for convenience, provides illustration of how alignment works (Kleinhenz 1998:39f). The word auf-essen "to eat up" is made of a stem (essen) and a prefix (auf). It is realised with an epenthetic glottal stop auf-ʔessen. The glottal stop is noncontrastive: it is always filled in word-initially in case the word begins with a vowel underlyingly. According to Kleinhenz' analysis, the pronunciation at hand is the result of the ranking ONSET >> ALIGN (Stem, L, PrWd, L) >> DEP. The alignment constraint demands that the beginning of the stem coincides with a prosodic word; regular mapping would express the same requirement by the fact of assigning two separate prosodic words to the prefix and to the stem. Now ONSET demands the presence of a consonant in the first syllable of essen – hence the epenthetic glottal stop. The concurrent solution that syllabifies the prefix-final consonant into the onset of the stem fails because of ALIGN: in au.fessen the stem and the prosodic word would be misaligned, and this is fatal since ALIGN outranks the epenthesis-hostile DEP. In a language with the reverse ranking, au.fessen without glottal stop would be optimal. Regular rule-based mapping would describe this variation by a parameter on the assignment of the prosodic word, which may or may not encompass prefixes. That is, the domain of glottal stop insertion is the prosodic word: glottal stops are inserted at their left edge. In German, the prefix makes a prosodic word of its own; the prefix-final consonant cannot join essen because there is no syllabification across prosodic word boundaries (in German). Essen will therefore be subject to glottal stop insertion. In the hypothetical language where au.fessen is encountered, the prefix and the stem cohabitate within a single phonological word: this allows the prefix-final consonant to join essen, and no glottal stop will be inserted.

390 Chap 11: Optimality Theory As may be seen, OT produces the winner without mentioning any particular process and thus without any process (or rule) making reference to the Prosodic Hierarchy. The effect is achieved by the interspersing of alignment with regular phonological constraints that are not involved in issues related to the interface (ONSET and DEP). This system significantly contrasts with rule-based mapping: it conflates a two-step procedure into one single constraint. On the regular account, prosodic constituency is first created by mapping; phonological rules then make reference to the prosodic structure. These steps are procedurally ordered, and domestic phonological rules refer to the carrier of morphosyntactic information in the description of their environment, as much as they may refer to truly phonological conditions. Neither of these mechanisms survives in constraint-based mapping: there is no procedural difference between the translation of morphosyntactic into prosodic constituency and the reference of phonology to this structure. Also, the carriers of phonological computation, constraints instead of rules, do not mention prosodic constituency anymore: ONSET, DEP or other constraints do not make reference to the Prosodic Hierarchy. Prosodic structure appears only in one constraint family, ALIGN (and WRAP, see below), which is specialised into importing morpho-syntactic information. This looks like a simplification of the mapping mechanism, which of course roots in the anti-derivational philosophy of OT: a two-step procedure whereby mapping precedes reference to prosodic constituency is incompatible with parallel computation. We will see in §526 below, though, that this evolution, which looks like an advance, creates serious trouble in a modular perspective: it gives up the idea that mapping is done in modular no-man's land (§381).

460

2.2.3. Parametric variation: interaction of ALIGN and WRAP A second constraint family that is specialised in mapping has been introduced by Truckenbrodt (1995, 1999): WRAP militates for having all items that are dominated by a morpho-syntactic constituent included in the same prosodic constituent. For example, WRAP (DP, Phon Phrase) is satisfied only if all material of a DP is mapped into one single phonological phrase. WRAP and ALIGN create tension: opposite demands are issued. While the former militates for the invisibility of morpho-syntactic divisions on the phonological side (phonology follows purely domestic rule within a given

Adaptation of Prosodic Phonology to the constraint-based environment 391

prosodic constituent), the latter makes morpho-syntactic divisions phonologically salient. WRAP is therefore said to be cohesional, whereas ALIGN has a demarcative function. Another aspect of the introduction of WRAP regards the mapping typology and the expressive power of constraint-based mapping. That is, the number of different prosodic phrasings that can be generated by the interleaving of WRAP and ALIGN is much higher than what the simple ordering of ALIGN with non-interface constraints produces. On the backdrop of the inflational evolution of mapping rules (§§388f), Selkirk (2000:231f) argues that this is an advantage. She celebrates the new factorial possibilities offered by OT in general and the antagonistic WRAP and ALIGN families in particular. As in all other areas of grammar, cross-linguistic variation is covered by the factorial typology of universal constraints, i.e. the various incarnations of ALIGN and WRAP in the case of mapping. And as elsewhere in OT, generative power is considered to be a good thing. Whether the grammar produces large-scale overgeneration is not a prime concern. In the particular case of mapping at least, it does not appear to arouse any debate in the literature. 461

2.3. The Strict Layer Hypothesis made less strict The Strict Layer Hypothesis (SLH) imposes formal restrictions on prosodic arborescence (see §383): a constituent of layer n must be exhaustively contained within a constituent of the immediately higher layer n+1, and can only exhaustively contain constituents of the immediately lower layer n-1. Hence there can be no nested constituents, nor can any association line bypass a layer. Responding to evidence and criticisms that challenge the monolithic character of the SLH (e.g. Inkelas 1990, Itô & Mester 2003 [1992], Ladd 1986, 1996), Selkirk (1996:189ff) recognises that it makes sense to factor out four more primitive component constraints which can be manipulated independently. Table (182) below (repeated from §383) shows the four pieces into which Selkirk proposes to cut the SLH. Selkirk still holds that (182a,b) are universal principles (hence undominated constraints in OT), but admits that (182b,c) may be violated in some cases, which she stresses are limited and local. The demotion of the SLH to a partially violable set of heterogeneous constraints has been largely adopted in subsequent work (e.g. Vogel 1999, Mazzola 1999).

392 Chap 11: Optimality Theory (182) The Strict Layer Hypothesis cut into pieces a. Layeredness a node of given layer cannot dominate a node of any higher layer (i.e. a syllable cannot dominate a foot) b. Headedness each node of layer n must dominate at least one unit of layer n-1. c. Exhaustivity association lines may not bypass any layer: no association of two units that belong to non-adjacent layers is allowed. d. Nonrecursivity nested structures are prohibited: no node may dominate a node of the same label.

462

2.4. Phase-based mapping More recently, constraint-based mapping in OT has been impacted by and adapted to syntatic Phase Theory (§304). Kratzer & Selkirk (2007) introduce the idea that chunks which are defined by the syntax as phases also grossly correspond to constituents of the Prosodic Hierarchy. Specifically, they propose that "the highest phrase within the spellout domain is spelled out as a prosodic major phrase" (Kratzer & Selkirk 2007:106, emphasis in original). We are thus still in the environment that was defined by Selkirk (1986) where three prosodic constituents are recognised (§394): the major phrase (called prosodic phrase then, and corresponding to X'' in an X-bar structure), the prosodic word (corresponding to X°), and an intermediate item called minor phrase (small phonological phrase then, corresponding to X'). On Kratzer & Selkirk's (conservative: §773) assumption that only CP and vP are phases, the idea is thus that spell-out domains, i.e. CPs and vPs, correspond to major phrases on the phonological side. This is supposed to be a universal equivalence and the result of mapping; language-specific variation in prosodic phrasing is then achieved not by the syntax-phonology mapping as before, but by purely phonological "prosodic markedness constraints, which operate to produce surface prosodic structures that are more nearly phonologically ideal" (Kratzer & Selkirk 2007:126). This is a significant departure from a Prosodic Phonology essential: mapping becomes universal and phase-driven, and the great amount of language-specific variation in prosodic phrasing is achieved in the phonology by a purely phonological mechanism.

Adaptation of Prosodic Phonology to the constraint-based environment 393

Another issue is the fact that there are more layers of the Prosodic Hierarchy than just the major phrase: what happens with them if mapping reduces to the equivalence "phase = major phrase". In a note, Kratzer & Selkirk (2007:125) hint at the possibility that the "prosodic word could be understood as the spellout of lexical (not functional) heads, while intonational phrase could be the spellout of 'comma phrase'." Other work along the idea that phases form prosodic islands includes Piggott & Newell (2006), Dobashi (2003), Ishihara (2007) and Kahnemuyipour (2009) (Elordieta 2008:274ff provides a survey). Elsewhere, Phase Theory has impacted mapping in different ways. Cheng & Downing (2007) for example propose alignment constraints that take the phase as an argument: ALIGN-L(PHASE, INTP) "align the left edge of a phase with the left edge of an Intonational Phrase."

463

2.5. Mapping: focus on new factors, but the puzzle is the same as before Recall from §§111,386 that the mapping puzzle is resident in the study of the interface, and a long-standing source of frustration. OT has not contributed any more to a solution than previous theories. Rather, the OT literature moved away from the study of morpho-syntactic influence on phonological processes in order to concentrate on mapping-conditioning factors that had not been taken into account before, such as information structure and eurhythmy (i.e. the strive towards prosodic constituents of comparable length). Selkirk (2000) opened this direction, which led to a renewed interest in mapping. Unlike in her direction-giving 1986 article (see §394), the purpose is not to find a means of channelling the inflational variation of phonologically relevant morpho-syntactic configurations. Rather, focus is laid on the different factors that may condition the construction of the Prosodic Hierarchy. Each type of influence, then, is associated to a particular constraint family. Selkirk distinguishes five constraint types. 1) ALIGN and WRAP are the core of the mapping mechanism, a family that she calls interface constraints.128 2) Domination constraints determine the geometric properties of the output of mapping. While the arboreal configuration of prosodic structure was defined in terms of the Strict Layer Hypothesis in the 80s, it is 128

Note that Selkirk's interface constraints have nothing in common with the constraints of the same name that are discussed in §494.

394 Chap 11: Optimality Theory now described by four independent constraints (see §460). 3) Purely phonological constraints translate pressure concerning the size of prosodic constituents or their stress properties. 4) Another constraint family takes into account information structure, i.e. properties such as focus and topic that are relevant for intonation (Selkirk 2007, 2008). 5) Finally, a set of prosodic constraints regulates eurhythmy that was evidenced by Ghini (1993), Sandalo & Truckenbrodt (2002) and others (Truckenbrodt 2007:451ff provides an overview): languages may prefer prosodic constituents of equal length. Selkirk (2000) argues that OT is well suited for expressing the variable pressure that can impact the translational process: the action of different forces can be implemented by scrambling the respective constraints in the same constraint hierarchy. Factorial typology, then, produces the surface variation observed. As in other areas, factorial typology probably overgenerates quite a bit, an issue that does not appear to arouse discussion. In sum, thus, a shift of focus has occurred when comparing the OT literature on mapping with the situation in the 80s. The problem then was to make sense of the seemingly anarchic and inflational cross-linguistic picture: why is it that boundaries X and Y, but not Z produce the same phonological effect? What do X and Y have in common that is not shared by Z? What does a general system look like that may describe all mapping situations with a few principles and a number of parameters? These questions are no longer discussed: given the wild and largely untamable variation, phonologists appear to have given up on this issue. Instead, the amorphous lists of variable phrasings are decomposed into different isolatable factors which may describe the result in terms of constraint interaction.

464

3. Cyclic derivation and its relation with phonology

465

3.1. Anti-cyclicity in OT and elsewhere: cutting off deep generative roots

466

3.1.1. Cyclic spell-out is a bone of contention: OT vs. generative grammar Since Chomsky et al. (1956:75), cyclic derivation lies at the heart of generative interface theory (§§80,100): it was implemented by all theories prior to OT. At least in its classical incarnation, however, OT must reject cyclic derivation because of its general anti-derivational orientation: computation, of whatever nature, must be parallel. In other words, there is no place for Interface Dualism (§82) in OT (see also §686).

Cyclic derivation and its relation with phonology 395

The intrinsically derivational character of cyclic derivation, and hence its absolute ban in OT, is made explicit for example by Kager (1999). (183) "Proponents of serial theory have argued for derivational levels which do not coincide with the input, nor with the output. An important argument for intermediate derivational levels was based on phonological properties carried over from morphologically simplex forms to complex forms. Such transderivational transfer was modelled as the transformational cycle (Chomsky and Halle 1968), a mode of rule application in morphologically complex words in which rules apply in an 'inside-out' fashion, from smaller to larger morphological domains. Cyclic rule application is intrinsically derivational, as it implies intermediate levels between the input and output at which phonological generalizations are captured." Kager (1999:277, emphasis in original)

Kager's measure for attributing the label "derivational" is the existence of intermediate forms. This position reflects the holistic view that considers a linguistic derivation as a whole without differentiating between inner- and inter-modular computation: for Kager there is no difference between phonological computation and interface activity.

467

3.1.2. Anti-cyclic voices from other quarters Orgun (1999:248ff) provides a well-documented survey of objections that have been raised against cyclicity. Beyond OT, cyclicity has been under fire from Declarative Phonology (Cole & Coleman 1992) because of its derivational character. Charges against interactionism have been led from the quarters of "Cognitive" Grammar129 and connectionism (Goldsmith 1993, Lakoff 1993, Karttunen 1993, all in the same book), as well as for reasons of cognitive and computational plausibility (Sproat 1992). Alternatives to unbeloved cyclicity that have been developed in OT are discussed below: parallel mini-grammars (co-phonologies, indexed constraints, §478), interface constraints (§494), analogy (which is called Output-Output faithfulness in OT, §497) and Orgun's HPSG-based perspec129

I use quotation marks in order to refer to the framework that was founded by Ronald Langacker (1987) because the label chosen suggests that this theory has a copyright on cognitive aspects of grammar, and that anything which is nonLangackerian must be non-cognitive. What about calling a particular theory of, say, biology, "scientific biology"? More on "Cognitive" Grammar in §§591,596.

396 Chap 11: Optimality Theory tive (§512). Also, §475 discusses the (inclusive) relationship between opacity killers and cyclicity killers.

468

3.1.3. The stake is high: inside-out interpretation Rejecting cyclic derivation en bloc has far-reaching consequences since inside-out interpretation, a central insight of generative thinking, will have to go as well. Recall from §161 that inside-out interpretation is shared by all versions of generative interface theory: it is implemented by interactionist as much as by non-interactionist architectures. Inside-out interpretation embodies the empirical insight that phonological computation mirrors morpho-syntactic structure (see §672, Bermúdez-Otero forth b). It will be difficult to throw this generalisation over board. Van Oostendorp (2007:143) for example points out that allowing for more embedded structure to be interpreted after less embedded structure creates anti-cyclicity, a pattern that appears to be absent from the record of natural language.

469

3.2. OT could respect modular contours – but it does not

470

3.2.1. Cyclic derivation and phonological computation are entirely independent as long as phonology is phonology proper What OT really (and tacitly) does when rejecting cyclic spell-out because of its derivational character is to subject the entire grammar together with its interfaces to the non-derivational ambition. The question is whether this ambition requires such a move. We will see that in fact it does not: the deliberate non-distinction between phonology and its interface with morphosyntax does not follow from any principle of OT; rather, it is a choice that was made by OT practitioners – a choice that is not made explicit, and whose origins lie in the misty and largely unreflected relationship of OT with modularity. Orgun (1999) points out that phonology proper is not impacted at all by whatever spell-out procedure is in place. He writes that OT and other approaches which reject cyclicity

Cyclic derivation and its relation with phonology 397 (184) "have taken it for granted that cyclic phonology, like rule ordering, is derivational and that this is sufficient reason to look for alternatives to cyclicity. […] In this paper, I reject the presupposition underlying these approaches, contending that there is an important distinction between rule ordering and phonology-morphology interleaving." Orgun (1999:250)

He then concludes that (185) "whether or not there is a derivational residue in phonology is entirely a question for phonological theory proper. Phonology-morphology interleaving is not a source of derivationalism." Orgun (1999:251)

The distinction between phonology and its interface with morphosyntax ought to be self-evident in the generative paradigm, whose backbone is a modular architecture. In the modular perspective of the inverted T (§86), phonology and the interface mechanism that transports pieces of the linear string into PF (and LF) are not the same thing. That is, phonological computation and the piece-transporting computation that is done upon spell-out are distinct. Cyclic spell-out is a particular take on how pieces of the morpho-syntactic structure are submitted to phonological computation: in small chunks and from inside out. It does not interfere with phonological computation at all. Looked at from the vantage point of OT, the anti-derivational ambition concerns only EVAL, i.e. the stage of the derivation where a piece of the linear string is singled out for phonological interpretation and transferred to PF. How this piece is selected, how big it is and what morphosyntactic status it has is entirely irrelevant for constraint interaction, and in fact invisible. Following the regular scenario, thus, cyclic spell-out identifies chunks and dips them into the phonological bath (where a chemical reaction transforms the raw material). Therefore spell-out is entirely orthogonal to phonological computation itself – unless modular contours are blurred. That is, chunk-submission and phonological computation may not be distinct if phonology as such is not properly delineated from other grammatical activities – or, in terms of OT, if CON contains a scrambled heterogeneous set of constraints that combine morphological (or even syntactic), phonological and interface-related instructions (more on this in §§526f).

398 Chap 11: Optimality Theory 471

3.2.2. Modularity is not an issue for classical incarnations of OT Classical incarnations of OT follow the all-in-one perspective (see §523), even though as far as I can see there is no principled reason that roots in the foundations of the theory. All-in-one means that morphological and phonological constraints may happily be interspersed, that the body of a given constraint may refer to both phonological and morphological information, that so-called interface constraints (§494) are allowed to revive the direct syntax approach in overt (and unreflected) violation of Indirect Reference (§406). Most importantly, though, we have seen in §455 that constraintbased mapping transports mapping, into the phonology: mapping (i.e. the translation of morpho-syntactic into prosodic structure) and reference to its result is done by the same constraint, ALIGN (and WRAP), which is located in EVAL and ranked among other, fully phonological constraints. Doing mapping in the phonology is incompatible with modularity (recall from §381 that mapping must be done in modular no-man's land), and so is the permanent reference of ALIGN (and WRAP) from inside of EVAL to morphosyntactic categories. The question is thus how far the parallel ambition of OT goes: is it confined to phonology, or does it encompass interface management, the transportation of chunks of the morpho-syntactic structure into phonology, or even morphology or syntax as a whole? In terms of OT, this question comes down to the issue of how many domain specific CONS there are: do we have a phonological CON that is distinct from a morphological CON and a CON that manages spell-out? And if so, how is the membership of individual constraints in a specific CON (or in multiple CONS) decided? As far as I can see, this kind of question is not debated in the OT literature. Rather, in its current incarnations OT behaves as if modularity were not an issue. The rare reflections regarding this question that I have come across in the OT literature are discussed in §§527f below (e.g. Burzio's 2007 explicit dismissal of modularity and Russell's 1999 proposal to have one single constraint ranking that encompasses scrambled phonological, morphological and syntactic constraints). At some point, OT will have to make up its mind regarding modularity, rather than leaving the issue in wafting mist: either modular contours are respected, in which case constraint-based mapping (i.e. ALIGN and WRAP), interface constraints and a number of other currently entertained tools have to go; or OT is declared non-modular, in which case its status as a generative theory is called into question. This is

Cyclic derivation and its relation with phonology 399

especially true in the current syntactic environment where the interface, that is sharp modular contours, play a central role (see §304). Of course, OT is not a monolithic whole, and every analyst has to make up his or her mind regarding modularity and the exact span of the parallel ambition. However, constraints such as ALIGN are so deeply rooted in (all versions of) OT that it is hard to see how they could be simply plugged out without this producing serious shock waves over the entire theory. On the other hand, anti-cyclicity does not follow from any core property of the theory as far as I can see: OT could as well respect a strictly modular environment where the parallel ambition only concerns phonological computation – hence where derivational interface management would be perfectly sound. In conclusion, the allergy that OT has developed against cyclicity has got nothing to do with OT itself – it is a consequence of an independent choice which scrambles phonology and interface management (eventually even other grammatical components) into one single constraint ranking. This choice needs to be placed in the broader context of modularity, which is systematically violated by current incarnations of OT.

472

3.2.3. Cyclic derivation, but morphology and phonology merged in the same constraint chamber (Wolf 2008) Finally, adhering to cyclic derivation does not mean that modularity is respected. This is shown by an approach called Optimal Interleaving (OI) that is developed by Wolf (2008). Couched in McCarthy's (2007) OT-CC (Harmonic Serialism), the goal of OI is to implement the insight of Lexical Phonology that phonology and morphology are interleaved. Therefore cyclic (inside-out) derivation is adopted, but the kind of interleaving familiar from interactionism where two computational systems ship pieces of growing size back and forth is made much more intimate: morphology and phonology are computed in the same constraint hierarchy where phonological and morphological constraints are interspersed, and lexical insertion takes place in the phonology (or rather: in the computational system that encompasses phonology and morphology). As a result, a constraint hierarchy where phonological and morphological constraints cohabitate evaluates candidate chains such as under (186) below where the cyclic derivation proceeds by progressively filling in

400 Chap 11: Optimality Theory morphological structure with phonological material (from the root outwards). (186) Wolf (2008:166)

The derivation is thus certainly cyclic, but the merger of morphology and phonology in the same computational system is a harsh violation of modularity (more on everything-is-one approaches and the scrambling trope of OT in §§528f).

473

4. Morpheme-specific mini-grammars in OT

474

4.1. Introduction The following pages review the lines of attack that OT has developed in order to replace cyclicity, whose derivational character is held to be disqualifying. Some of the surrogate devices have explicitly been conceived as cyclicity killers, others have an independent existence. The presentation aims to cover all approaches to the question how extra-phonological information is processed in phonology that OT has produced. It proceeds in two steps: first direct heirs of Lexical Phonology are examined, that is approaches which implement morpheme-specific mini-phonologies. Other views on the interface are then discussed in a second step (§493).

475

4.2. Opacity killers, cyclicity killers and their relationship Over the past fifteen years or so, OT has been struggling with opacity, which is traditionally analysed in terms of extrinsically ordered rules.130 Various parallel solutions have been proposed (opacity killers), but none has proven successful: Output-Output (OO) correspondence (e.g. Benua 1995, 1997), sympathy (various versions, e.g. McCarthy 1999, 2003b), Comparative Markedness (McCarthy 2003a), targeted constraints (Wilson 2000, 2001), enriched inputs (Sprouse 1997) and a couple of others. 130

What follows is a mere description of how the OT scene dealt with opacity. It does not contribute anything to the debate regarding the typology of opacity, which is a generic term for a number of empirical situations that may be quite different.

Morpheme-specific mini-grammars in OT 401

On the other hand, the idea that holistic parallelism is wrong is gaining ground. According to this view, phonology proper is strictly parallel, but its interface with morpho-syntax is derivational. This is the position of the direct heirs of Lexical Phonology, i.e. Stratal OT and DOT. More recently, phonological computation itself has been declared derivational by McCarthy's (2007) revival of Harmonic Serialism (Prince & Smolensky 1993); interestingly, it was McCarthy who had engaged OT into antiderivationalism in the first place. As a matter of fact, all versions of OT with derivational elements have been designed under the pressure of opacity – and at the same time do the job of cyclicity killers. As we will see, however, the reverse is not true for cyclicity killers (or what is taken as such), which have got nothing to say about opacity. That a particular view on opacity impacts cyclic spell-out makes sense: opaque derivations always involve the concatenation of some morpheme. This is what Green (2007:169) points out: purely phonological opacity does not exist. Opacity is thus necessarily related to the representation of morpho-syntactic information in phonology, which in turn transits through the phonological cycle. Therefore a way to go about opacity is to work on how morpho-syntactic information impacts phonology.

476

4.3. Morpheme-specific mini-phonologies: parallel vs. reranked incarnations On the following pages, the four currently entertained approaches that implement morpheme-specific mini-grammars are discussed: co-phonologies (§478), indexed constraints (§482), DOT and Stratal OT (§483) (relevant references are provided below). The two latter are considered to be direct continuators of Lexical Phonology both in spirit and regarding their derivational character, while the two former are not thought of in these terms. I argue that Stratal OT and DOT are derivational in their selfunderstanding and in the perception of the field, but that this derivational aura owes more to the historical reference (Lexical Phonology) than to a real need that is imposed by data or theory, at least as far as affix classbased phenomena are concerned. The critical property which opposes Stratal OT/DOT on the one hand and co-phonologies/indexed constraints on the other is reranking. On the analysis of the former, mini-grammars apply in serial order, just like strata in classical Lexical Phonology: first a string is interpreted at stratum 1/by

402 Chap 11: Optimality Theory mini-grammar 1, then it is assessed by stratum 2/mini-grammar 2. The two mini-grammars do not exist independently at any given point in time; rather, there is only one single constraint chamber whose content is rearranged once stratum 1 computation is completed. This operation is called reranking (of constraints in EVAL); it is reranking, and nothing else, that makes DOT and Stratal OT derivational. On the other hand, co-phonologies and indexed constraints provide for permanently distinct mini-grammars that co-exist in parallel. Table (187) below depicts the two options at hand. (187) parallel vs. serially ordered mini-grammars a. co-phonologies, indexed constraints b. Stratal OT, DOT string 1

string 2

all strings

phono 1

phono 2

phono 1

phonology

reranking phono 2

phonology

The difference is thus chronological (or derivational). Several minigrammars coexist permanently under (187a): phonology is made of a set of distinct subregularities, each of which forms a natural and consistent system. By contrast under (187b), there is only one constraint chamber at any given point in time: the contrast between distinct mini-grammars is achieved by the reranking of constraints, an operation that takes place in the midst of phonological activity. Another important difference is the fact that under (187b) strings must run through all successive mini-phonologies on their way to the surface. That is, a string which is created at the first stratum will also experience the computation of the second stratum. By contrast under (187a),

Parallel mini-grammars 403

strings that are specified for computation by mini-phonology 1 will never meet mini-phonology 2 (and vice-versa). On the pages below, implementations of the two scenarios are discussed. 477

5. Parallel mini-grammars

478

5.1. Co-phonologies: two constraint rankings

479

5.1.1. Lexical material selects a specific computational system Co-phonologies are the closest match of (187a): they really implement two parallel, permanent and waterproof mini-grammars. Affixes, but also entire words, may be lexically specified for selecting this or that co-phonology (of which of course there may be more than two). In the case of entire words, co-phonologies mimic the traditional view on loanword phonology: each word is lexically specified as a member of a certain class such as "native", "foreign", "latinate", "learned" and the like. A specific co-phonology with a particular constraint ranking is associated to each lexical class. Whenever a word is engaged in a derivation, then, its diacritic marking selects the corresponding co-phonology. This management of course is much akin of the diacritic marking of rules that was common practice in SPE: "rule X applies only to words that are marked as [+latinate]". Co-phonologies are used by, among others, Itô & Mester (1995), Orgun (1996a), Inkelas (1996, 1998, 1999), Orgun & Inkelas (2002), Anttila (2002).

480

5.1.2. How affix class-based phenomena could be analysed In the case of affixes, the mini-grammar for which the affix is marked is "called" in order to assess the string that the affix heads. A string such as [root-affix 1] will thus be computed by mini-phonology 1, while miniphonology 2 will be called for the assessment of [root-affix 2]. Unfortunately, I was unable to find relevant literature that shows how (English) affix class-based phenomena are analysed with co-phonologies. The following is thus my personal extrapolation of how this could be done. The coverage is only partial: I will not push the speculation too far. It is also worth mentioning that the literature on co-phonologies that I am famil-

404 Chap 11: Optimality Theory iar with does not mention morpho-syntactic structure. This bit of the discussion below is also mine, for the sake of comparison with previously reviewed theories. With these reservations borne in mind, table (188) below shows how more complex strings could be handled by co-phonologies. Examples are taken from English affix class-based stress assignment. (188) parallel morpheme-specific mini-phonologies a. univérs-al-ness b. govern-mént-al γ class 2

phon 2 β

class 1

phon 1 α

x

γ

phon 1 root

class 1

phon 1 β

class 2

phon 2 α

x

phon 1 root

In each case, the material that is dominated by the node that the affix projects is computed by the mini-phonology that the affix selects for. In the English example, the rule that distributes penultimate stress is present in class 1 phonology, but absent from class 2 phonology. Hence [univers-al] and [govern-ment-al], the two strings that are headed by a class 1 affix, receive penultimate stress: univérs-al and govern-mént-al. Since the stress rule is absent from class 2 phonology, stress remains unshifted with respect to the next lower node when class 2 nodes are interpreted: the result is univérs-al-ness and góvern-ment, respectively.

481

5.1.3. Are co-phonologies a version of Halle & Vergnaud (1987a)? In order for roots in isolation to receive stress, they also need to be somehow spelled out. This can be done as under (188) by hard-wiring a specification for all roots to be assessed by class 1 phonology. This replicates the logic of Halle & Vergnaud's (1987a) system where, recall, roots are always interpretational units (§280). As for Halle & Vergnaud, Kaye's (1995) proviso according to which the word is always an interpretational unit must be

Parallel mini-grammars 405

dismissed by co-phonologies: did class 1 phonology systematically apply at the word level, *univers-ál-ness would be produced. The analogy with Halle & Vergnaud is also confirmed by the selection of the affix class that triggers stress assignment: recall from §281 that unlike in Kaye's (1995) system where class 2 affixes are interpretationtriggering, class 1 affixes play this role with Halle & Vergnaud. This is also the case under (188). These three choices (roots are interpretational units, words are not, class 1 affixes trigger spell-out) may thus suggest a convergence which, however, is only apparent. For co-phonologies also carry out interpretational activity at class 2 nodes: the stress rule just happens not to be part of the processes that belong to class 2 phonology. By contrast, Halle & Vergnaud (1987a) oppose nodes that trigger interpretation (cyclic) to nodes that do not (non-cyclic). This is precisely where the fundamental front line runs between theories that implement morpheme-specific phonologies (such as co-phonologies) and theories that work with just one computational system (such as Halle & Vergnaud's). Continuing the speculation about how affix class-based phenomena could be analysed with co-phonologies is idle. The solution that was sketched could also be said to hit far from the mark since it obviously supposes inside-out interpretation, i.e. cyclic spell-out: it must first be decided that stress does not shift in góvern-ment before it can be decided that it shifts one vowel right in govern-mént-al. Co-phonologies, however, are supposed to be a cyclicity-killer. 482

5.2. Indexed constraints: two grammars in the same constraint ranking Another way of implementing (187a) is to make all string-specific constraints cohabitate in the same constraint ranking. In this perspective, lexical items are specified for computation by mini-grammar A or minigrammar B as before. However, constraints come in two versions, A and B (FAITH for example may divide into FAITHA and FAITHB). Hence a lexical item that is specified for computation by mini-grammar A will only be assessed by the subset of A constraints. Another subset, B constraints, computes only B items. Every string-specific constraint thus exists twice, in an A- and a Bguise. Different versions of the same constraint, then, have different rankings in the same constraint chamber, both with respect to each other and regarding string-unspecific constraints (which are not doubled). In Pater's

406 Chap 11: Optimality Theory (2009:125) description, "a single constraint can be multiply instantiated in a constraint hierarchy, and each instantiation may be indexed to apply to a particular set of lexical items. These indexed constraints are universal markedness and faithfulness constraints, whose application is relativized to a set of lexical items." Itô & Mester (1999b) for example use indexed constraints in order to describe Japanese loan words, which according to them fall into four lexical classes, depending on the degree of assimilation: native words, established loans, assimilated foreign words and unassimilated foreign words. Their analysis then establishes FAITH1, FAITH2, FAITH3 and FAITH4 in the same constraint ranking, where each version of FAITH is responsible for one of the four classes. Indexed constraints may be viewed as a modern implementation of diacritics that SPE added to rules in order to restrict their application to a specific subset of morphemes. They have been introduced by Prince & Smolensky (1993) for alignment, and were later generalised to other constraints (both faithfulness and markedness) by, among others, Itô & Mester (1999b, 2001) and Pater (2000). There is ongoing debate between defenders of co-phonologies and supporters of indexed constraints (e.g. Pater 2009). The difference is not easy to transform into contrasting predictions: different phonological minigrammars are implemented as different rankings in one case, in one single ranking in the other. Or, from the morphological point of view, morphemes select either constraint rankings or the index of a constraint. Table (189) below depicts this contrast. (189) morpheme-specific mini-phonologies: parallel implementations a. co-phonologies b. indexed constraints /…X1…X2…/ /…Xx…Xy…/

phono 1

phono 2

constraint 1x constraint 4y constraint 2 constraint 3x constraint 1y constraint 4x constraint 3y

phono 1

phono 2

The following section considers the serial competitor of parallel morpheme-specific mini-phonologies.

Reranked mini-grammars: Stratal OT, DOT 407 483

6. Reranked mini-grammars: Stratal OT, DOT

484

6.1. Serial vs. parallel solutions

485

6.1.1. Parallel OT couched in the stratal architecture of Lexical Phonology DOT (Derivational Optimality Theory, Rubach 1997, 2000b, 2003, Booij 1997) and Stratal OT (Kiparsky 2000, Bermúdez-Otero forth a) are constraint-based versions of stratal Lexical Phonology (although their contrast with Lexical Phonology does not reduce to this difference). They replicate the stratal architecture according to (187b): a number of lexical strata (and a postlexical phonology) are serially ordered and afford distinct phonological computations. At each level, candidates are assessed by a classical constraint ranking in entirely parallel fashion, but constraints may be reranked between levels. Kiparsky (2000) outlines the research programme of Stratal OT in the following terms. (190) "LPM-OT's goal [LPM stands for Lexical Phonology and Morphology] is to reduce cyclicity to I/O faithfulness, and opacity to interlevel constraint masking. Thus, if α is the constraint system of some domain (say, stems) and β the constraint system of a larger domain (word level or postlexical) the β's markedness constraints can render α opaque. These are the only sources of cyclic effects and opacity: there are no O/O constraints, no paradigm uniformity constraints, and no sympathy constraints. The intrinsic seriality of LPM-OT provides a handle on opaque and cyclic constraint interactions without retreating to the unconstrained ordering theory of preOT days." Kiparsky (2000:352, emphasis in original)

Kiparsky's positioning is very clear: he proposes to do parallel OT, but within the original stratal architecture of Lexical Phonology. That is, the overall derivation is serial, but within each stratum interpretation is strictly parallel. 486

6.1.2. The critical contrast with parallel implementations is not discussed Stratal OT is thus serial in its self-understanding as much as in the perception in the field. Given that anti-derivationalism is the headstone of classical OT, this is not a minor issue. The trouble is that, as far as I can see, the derivationalism of Stratal OT/DOT is taken for granted without demonstra-

408 Chap 11: Optimality Theory tion of the fact that this solution is any different from the parallel implementation of morpheme-specific mini-grammars under (187a). The whole derivational issue hinges on reranking, and on nothing else. Given the anti-derivational bias of OT, the default assumption is that parallel solutions are correct. Therefore, what one expects defenders of Stratal OT and DOT to show is that reranking is crucial, i.e. that approaches which lack this operation are not viable. Comparative discussion of serial and parallel implementations of morpheme-specific phonologies, however, is not easy to come by (see the following section). One contrast between stratal and parallel implementations of morpheme-specific phonologies has already been mentioned in §476. On the stratal account, strings have to run through all successive strata on their way to the surface. That is, roots are computed by all mini-grammars in the course of a derivation, and strings that are affixed at stratum X also experience the stratum-specific computation at stratum X+1 (and at all subsequent strata). By contrast in a system with parallel mini-grammars which are not related by any derivational means, every string only experiences the specific computation that it selects for: strings that are made of a root and a class 1 affix are computed by class 1 phonology, but never meet class 2 phonology, and vice-versa. I have not come across any argument in the literature that is based on this contrast in underapplication.

487

6.1.3. Indexed constraints, but not co-phonologies, are perceived as a competitor Kiparsky (2000) and Bermúdez-Otero & McMahon (2006:402ff) offer comparative discussion with a non-derivational solution for cyclicity- and opacity effects (Output-Output constraints, see §497), but do not address non-derivational instantiations of morpheme-specific mini-grammars. On the other hand, Bermúdez-Otero (forth a:§2.4.5.1) argues that since the tools of Stratal OT allow for the analysis of morpheme-specific phonology, constraints do not need to be indexed.

Reranked mini-grammars: Stratal OT, DOT 409 (191) "Classical LPM expresses domain limitations indirectly: the formulation of a phonological rule need not mention the grammatical categories within which the rule applies or does not apply; rather, rules are assigned to levels and apply whenever a construction triggers a cycle of the appropriate level. The same is true in Stratal OT: specific hierarchizations of CON are assigned to particular levels; the level-n hierarchy applies only in a level-n domain. […] As a result, individual constraints need not be relativized to particular morphological or syntactic constructions: constraint indexing is prohibited." Bermúdez-Otero (forth a:§2.4.5.1)

Once the stratal model is chosen, indexed constraints are thus superfluous. This is certainly correct, but Bermúdez-Otero does not explain why this choice should be made in the first place. Unlike indexed constraints, co-phonologies do not appear to be perceived as a competitor by Stratal OT. Bermúdez-Otero (forth a:§4.6.1) indeed uses the term co-phonology as a synonym for domains (domains in the sense of Lexical Phonology, i.e. strata): "in Spanish, when a verb stem […] constitutes a phonological domain, it selects a stem-level cophonology that constructs an ω without foot structure." Also, Bermúdez-Otero & McMahon (2006:404ff) accept Anttila's (2002) co-phonology-based account of particular stress-shifting suffixes in English. The critical contrast, then, reranking, does not seem to make much of a difference, and in any event remains unreflected. In the DOT literature, the situation is much the same: Rubach (1997, 2000a,b, 2003) presents cases that call for a treatment by morphemespecific mini-phonologies, but as far as I can see takes for granted that these follow the two-step logic. Alternative accounts based on parallel mini-grammars are not considered. What we end up with, then, are well-documented arguments in favour of the existence of distinct mini-grammars; but demonstration is lacking that these must entertain a serial relationship. 488

6.2. Stratal OT is more than just an OTed version of Lexical Phonology

489

6.2.1. Lexical Phonology anew: unhorsing the SPE heritage Its roots notwithstanding, Stratal OT is more than just an OTed version of classical stratal Lexical Phonology. For example, Bermúdez-Otero (forth a:§1.2.1) dismisses the SPE-legacy that (in his words) crippled Lexical Phonology: underspecification, Kiparsky's SCC-K and structure preserva-

410 Chap 11: Optimality Theory tion (see §214). Also, Bermúdez-Otero (forth a:58ff) remains agnostic regarding interactionism. By contrast, Stratal OT is firmly committed to Praguian segregation (Bermúdez-Otero forth a:66ff): together with the two morpheme-specific mini-phonologies, grammar then contains (at least) three distinct computational systems (i.e. rankings of CON): the stem-level ranking, the wordlevel ranking and the phrase-level ranking.

490

6.2.2. Inflational access to morpho-syntactic information is a concern in Stratal OT DOT and Stratal OT follow the general practice in OT that was described in §455: the Prosodic Hierarchy is taken over from the 80s, but translation is now done by constraint-based mapping. That is, reference to prosodic constituents is made through the constraint families ALIGN and WRAP. The relationship between the procedural tools inherited from Lexical Phonology and the representational devices coming from Prosodic Phonology is as unclear in OT as it was between the original theories (see §423). That is, functional overlap and hence competition (rather than the peaceful coexistence that was tacitly promoted in the 80s) between procedural and representational means of talking to phonology are an issue: OT has a general trope for the proliferation of tools that do the same job (opacity killers, cyclicity killers etc.). Bermúdez-Otero (forth a) is among the few references that I have come across which explicitly identify the unrestricted access of phonology to morpho-syntactic information as a problem (Orgun & Inkelas 2002:116 is another example). He describes the current situation in OT as follows. (192) "Restricting the phonology's access to morphological and syntactic structure is a long-standing goal of generative linguistics. In nonstratal OT, however, this project seems to have been more or less abandoned. […]In the absence of cyclicity, all the grammatical structure of a linguistic expression must be simultaneously accessible to the phonology. In such circumstances, the task of limiting the uses to which the phonology can put this grammatical information devolves upon the theory of CON. So far, however, there has been no concerted effort to define the set of constraints that refer to nonphonological information: the situation is, in effect, a near free-for-all. " Bermúdez-Otero (forth a:43)

Reranked mini-grammars: Stratal OT, DOT 411 491

6.2.3. The solution is the same as before: Indirect Reference and peaceful coexistence In reaction to this, Bermúdez-Otero (forth a:45) calls on the classical principle of Indirect Reference (§§377,406), which cuts down the blooming diversity (also including interface constraints, see §494). (193) "Indirect Reference Hypothesis (strong version) Phonological constraints do not refer to morphological or syntactic information, modulo constraints on the alignment of prosodic categories with grammatical constituents. This results in a massive simplification of CON with respect to nonstratal OT: we retain IO-faithfulness and hybrid alignment constraints, but dispense with OO-correspondence, indexed constraints, MORPH-REAL, and the like." Bermúdez-Otero (forth a:45, emphasis in original)

Representational communication with phonology is thus disallowed altogether, except through the ALIGN channel. But even here BermúdezOtero attempts to marshal unrestricted constraint-based mapping (see §455): he assumes only a restricted set of alignment constraints which has "very limited access to non-phonological categories" (Bermúdez-Otero 1999:112). What this means in practice, though, remains to be seen. Bermúdez-Otero's (forth a) conclusion on the representational issue, then, calls for a phonology that is freed from representational reference to morpho-syntactic information. (194) "(2,76) No Symbols of Grammatical Affiliation Phonological representations do not contain symbols that refer to morphological or syntactic structure. Under principle (2,76), a phonological representation p can be globally associated with a grammatical unit g; […] The content of p, however, must be exclusively phonological: the primitives of phonological representation do not include symbols of morphological or syntactic affiliation. Accordingly, (2,76) banishes not only the juncture phonemes of the Structuralist tradition and the boundary symbols of SPE (+, =, #), but also the morphological brackets of LPM and any such conceivable device. Notably, (2,76) prohibits the morphological or syntactic indexation of features, segments, or any other internal constituent of phonological representations." BermúdezOtero (forth a:45)

412 Chap 11: Optimality Theory His practice, however, does not stand up to this promise. BermúdezOtero's principle (2,76) cuts away all kinds of diacritic rank growth, except the Prosodic Hierarchy and direct reference to morpho-syntactic categories in ALIGN constraints (as explicitly mentioned in (193)). That is, constituents of the Prosodic Hierarchy are as much symbols of grammatical affiliation as were juncture phonemes, hashmarks and brackets. All of these devices comply with the principle of Indirect Reference, but all of them are diacritics whose only purpose is to store morpho-syntactic information (§402). In the end, Stratal OT thus appears to continuate the aforementioned peaceful coexistence of Lexical Phonology and Prosodic Phonology: there does not appear to be a principled way of deciding whether a given phonological phenomenon that is sensitive to morpho-syntactic information is the result of stratal (i.e. derivational) or representational activity.

492

6.3. How different can mini-grammars be? The interface literature in OT either continuates morpheme-specific phonologies, the legacy of Lexical Phonology, or develops approaches that are incompatible with modularity. The former were reviewed above, the discussion of the latter is to come. Before pursuing, though, let us look at an argument that has been levelled against morpheme-specific miniphonologies, in fact against all types of distinct computational systems. Golston (1996) and Orgun (1996a) discuss the "cophonology proliferation problem" (Orgun 1996a:114). In Golston's (1996) view, multiple mini-grammars do not qualify because in principle they allow for a languages to be co-defined by very distant and completely different computational systems: one mini-grammar could feature, say, a three-vowel system such as the one that is found in Arabic, while another could allow for a flourishing vowel system of the Swedish type. This, Golston argues, is not a situation that is ever found in natural language: the phonologies of different strata of a given language are always intimately akin; the range of variation found is not remotely comparable to what the cross-linguistic picture offers. Golston (1996:192) therefore pleads "against analyses which treat multiple levels in terms of multiple grammars.[…] The phonology of a language is always uniform, stable as a single set of constraints." Orgun (1996a:114ff) points out the same problem, but unlike Golston tries to constrain the degree to which co-phonologies in the same language can differ from one another. Both Golston and Orgun only talk about co-

Other cyclicity killers 413

phonologies, but the logic of their argument is against the existence of mini-grammars as such, whatever their implementation. Benua (1997:90ff) and McCarthy (1999, 2007:42ff) indeed extend the critique to the option under (187b), which is represented by Stratal OT/DOT. The perspective here is slightly different, though. While Stratal OT/DOT also implements morpheme-specific mini-grammars, these are not waterproof. That is, they entertain a derivational relationship through the reranking operation. The mini-grammars of (187a) on the other hand do not "see" each other at all. It can therefore be argued by defenders of serially ordered miniphonologies that their distance is a function of the reranking operation: the more distant the two mini-grammars, the more reranking has taken place, i.e. the more constraints have been "displaced". This is indeed the line of attack of Ricardo Bermúdez-Otero, who argues that the limits of interstratal reranking are dynamically imposed by acquisition (iterative stratum construction, Bermúdez-Otero 2003) and the rules of diachronic evolution (Bermúdez-Otero 1999:98ff).131 Reranked mini-grammars and the constraints that apply among them via the reranking operation therefore appear to be the modern version of the classical concern for a certain unity of grammar: recall from §151 that domain assignment in Lexical Phonology already embodied the idea that there is just one set of phonological rules, whose members are assigned to this or that stratum, or to postlexical phonology. That is, identical rules that occur in several locations of the grammar are manifestations of the same regularity, rather than different rules that happen to have the same formulation.

493

7. Other cyclicity killers

494

7.1. Interface constraints

495

7.1.1. Back to direct syntax Another way of making the interface non-cyclic that is entertained in the OT literature is to simply abandon Indirect Reference. So-called interface constraints (Anttila 2002) and work such as Pak (2008) and Samuels 131

It was mentioned in §487 that according to Bermúdez-Otero & McMahon (2006), co-phonologies are not incompatible with derivationally ordered multiple mini-grammars: a given stratum in Stratal OT may use a number of cophonologies.

414 Chap 11: Optimality Theory (2009a:284ff) revive direct syntax approaches that were entertained in the 70s and early 80s (see §407). Recall that since the mid 80s, the need for translation has been recognised: Indirect Reference was the guiding light in generative interface theory. This has led to the architecture of Prosodic Phonology where a Translator's Office that stands in modular no-man's land converts morpho-syntactic into phonological structure (§§381,406). All this is thrown over board by interface constraints and, as far as I can see, this is done without discussion. Also, nothing is said regarding the articulation of interface constraints with prosodic constituency and continuators of Lexical Phonology: as in other areas, OT allows for a toolbox that contains a number of instruments which may do the same job, and which may be conflicting. 496

7.1.2. Back to SPE-type morphological diacritics The formulation of interface constraints is purely phonological, but it is supplemented with a diacritic. Diacritics in co-phonologies and indexed constraints determine the mini-phonology to which the constraint belongs; with interface constraints they identify the particular morphological category that the constraint applies to (and in absence of which it is ineffective). Interface constraints thus make reference to designated morphosyntactic categories. They may be illustrated by Kager's (2000:146f) treatment of class 1 vs. class 2 affixes in Dutch. On his analysis, the contrast is ultimately encoded arboreally: following Selkirk (1984:77), class 1 affixes are attached lower in the tree than class 2 affixes. The exact position of affixes in the tree, however, is governed by constraints; one of them prevents class 1 affixes (such as -eer, -iteit) from attaching to higher positions: "NONRECSTEM: No Stem (affixed by -eer, -iteit etc.) immediately dominates a Stem." This type of constraint thus needs to mention every affix (or affix class for that matter) in its body. In SPE, rules were commonly supplemented with a statement in prose or a diacritic in subscript that specifies to which morpho-syntactic category it applies (e.g. the compound rule or the nuclear stress rule, see §95). Interface constraints are thus the faithful reincarnation of this practice in a constraint-based environment. Interface constraints such as FAITH-root and FAITH-affix have been introduced by McCarthy & Prince (1995a); further relevant literature includes Smith (1999), Borowsky (2000) and Alderete (2001). Van Oosten-

Other cyclicity killers 415

dorp (2007:125ff) and Anttila (2002) discuss the spectrum of interface constraints. Anttila (2002:2) also provides an overview of the particular morphological categories to which interface constraints have made reference in the literature. Typical are more general categories such as roots, affixes, nouns, verbs, lexical vs. functional morphemes and affix classes, but the possibility of reference to individual morphemes is also entertained (e.g. Raffelsiefen 1996:207f, Hammond 1995, Russell 1999). Finally, reference to designated morphological categories is an option that is open to all kinds of constraints: faithfulness, markedness and alignment. 497

7.2. Analogy (Output-Output correspondence)

498

7.2.1. OO correspondence was designed as a cyclicity killer OO correspondence expresses the traditional concept of analogy in a constraint-based environment. That is, instead of describing a faithfulness relation between an input and an output, OO constraints impose a faithfulness relation between two independent outputs: a word falls prey to the influence of another word. Output-Output (OO) correspondence has been introduced by Benua (1995, 1997) and further developed by, among others, Kenstowicz (1996), Burzio (2000a,b, 2001), McCarthy (2005). It has rapidly become a standard tool in OT. Interestingly, Benua's original goal was to do away with the phonological cycle and lexical strata – because of their derivational flavour. (195) "This theory obviates the traditional analysis that deviant phonology in complex words is the product of cyclic derivation. Given transderivational relations, cyclic effects are produced by constraint interaction in nonprocedural Optimality Theory. […] In this transderivational theory, phonology is sensitive to morphology because phonological faithfulness relations hold over paradigmatically-related words. There are no cycles or levels of derivation. Complex words, like simplex words, are derived in a parallel grammar, without any intermediate stages." Benua (1997:vi-vii)

This still appears to be the basic motivation for and merit of OO correspondence: in the editorial note that heads the reprint of parts of Benua (1997), John McCarthy stresses the historical dimension of the idea.

416 Chap 11: Optimality Theory (196) "The phonological cycle is one of the central insights of Chomsky and Halle (1968). This insight was developed further in the theory of Lexical Phonology (Kiparsky 1982[a,b]; Mohanan 1986). The idea is that phonological rules first apply to simple words and then apply again to morphologically complex words derived from them. The cycle and like conditions were worked out within a rule-based framework, and so it is natural to re-examine these ideas within OT. The proposal in this chapter is that effects formerly attributed to the cycle or lexical strata should be analyzed in terms of faithfulness constraints – not the familiar input-output faithfulness, but rather faithfulness between different output forms or related words." McCarthy (ed.) (2004:419)

Hence OO correspondence counts as a non-derivational means of eliminating cyclicity.

499

7.2.2. OO is not a cyclicity killer by itself When it comes to the concrete implementation of OO correspondence, however, it appears that the mere ability of faithfulness constraints to make reference to words does not guarantee a cycle-free analysis. In the same way as waterproof mini-grammars and interface constraints, OO correspondence must use diacritics in order handle cyclic effects. An example is Benua's (1997) treatment of the English stem-final [-m] - [-mn] alternation (da[mn]-ation class 1 vs. damn-ing [m] class 2, damn [m], see §167). Benua splits OO constraints into two types, for which affixes are subcategorised. (197) "OO1-Identity constraints evaluate class 1 paradigms […] and OO2-Identity constraints rule class 2 paradigms. Both sets of OO-Identity constraints are ranked in the English hierarchy of markedness and IO-Faith constraints. When analogous OO1-Identity and OO2-Identity constraints have different rank in the grammar, the two classes of affixed words pattern differently." Benua (1997:163ff)

That is, Benua uses morpheme-specific mini-phonologies, indexed constraints on her take: her constraints are doubled and supplemented with an index, which decides to which affix class they apply. Therefore OO correspondence may not be said to be a cyclicity killer per se: it is not true that it does away with the unwarranted cycle. A cyclesurrogate (indexed constraints in our example) is still needed on top of the OO mechanism. Only "regular" cyclicity killers such as co-phonologies or

Other cyclicity killers 417

indexed constraints are direct competitors of the cycle that can do all of its labour. Finally, Bermúdez-Otero (forth b) argues on the empirical side that OO correspondence cannot cover all patterns that are classically managed by cyclic derivation. He also shows that it overgenerates: OO correspondence can describe patterns that are not granted for by what BermúdezOtero calls the Russian Doll Theorem (interaction of opacity and cyclicity).

500

7.3. Enriched representations (van Oostendorp) Finally, the OT literature offers another strategy for the treatment of cyclic phenomena: in a series of articles, van Oostendorp (1999, 2002, 2004) argues that at least some of the effects at hand, namely in Dutch, have a purely phonological solution.132 This perspective, however, supposes a sufficiently rich system of phonological representations. Regarding what is traditionally assumed to be class 1 vs. class 2 suffixes in Dutch (Booij 1977), van Oostendorp's basic observation is that the former are all vowel-initial and monosyllabic, while the latter are all consonant-initial and plurisyllabic. Only one suffix, -achtig (class 2) disobeys this generalisation. Class membership can thus be predicted on purely phonological grounds (-achtig in the end will be shown to be /-ʔachtig/, i.e. consonant-initial). On the other hand, van Oostendorp discusses the well-known asymmetry between prefixes and suffixes: the latter are "dependent" (for example, they regularly resyllabify across their boundary), while the former are "autonomous" (they are more reluctant to do that). Van Oostendorp proposes to directly encode this observation in a constraint, which he calls Morphological Syllable Integrity. The motor of the various events that are observed at prefix boundaries, then, is the tension between a general tendency to have morpheme boundaries coincide with boundaries of prosodic constituents (PR≈LEX) and the wish to have "good" syllables, i.e. which possess an onset.

132

A condensed version of this body of work appears in van Oostendorp (2006a:47ff).

418 Chap 11: Optimality Theory 501

8. More direct syntax: representational continuity between morphology and phonology

502

8.1. Introduction Finally, it is worth presenting two sister approaches to the phonologymorphology interface that practise direct syntax (§407) in a separate section. Both share the idea that untranslated morpho-syntactic information (structure and labels) is completely and permanently availability during phonological computation, and both claim a strong representational, i.e. anti-derivational position. Van Oostendorp's Coloured Containment is prepared to let phonological constraints make reference to the fact that phonological objects (such as segments and features) do or do not belong to the same morpheme, and that they were or were not present underlyingly. A more radical type of representational continuity is argued for by Orgun's (1996a) Sign-Based Morphology, which is an application of HPSG to phonology. On this count, there is no difference between morpho-syntax and phonology at all: everything is monostratal (a property also claimed by van Oostendorp 2006a:5, 2007:144f), which means that there is just one global representation that encompasses all relevant morpho-syntactic, semantic and phonological information. Proposing a representational analysis for affix class-based and related phenomena is outstanding in itself: since SPE this has been the kernel competence of procedural tools. Another consequence of global representations of course is the abandon of modular divisions. These issues are discussed below.

503

8.2. Coloured Containment (van Oostendorp 2006a)

504

8.2.1. Faithfulness between morphology and phonology Van Oostendorp (2006a,b, 2007) introduces a perspective on the interface which, as far as I can see, is unprecedented in the study of the morphologyphonology interface – and which is only possible in OT. The idea is to use faithfulness for the comparison of morphological and phonological structure. Van Oostendorp thus takes advantage of a mechanism that lies at the heart of OT and which is used in various incarnations elsewhere: InputOutput, Base-Reduplicant and Output-Output (although in van Oosten-

More direct syntax: continuity between morphology and phonology 419

dorp's Containment-based environment the two latter are not expressible and hence done away with). Coloured Containment thus reverts to direct syntax: instead of replicating the technology of the 80s (strata, prosodic constituency) in a constraint-based environment (which as we have seen is current practice in OT), van Oostendorp builds an interface theory on the grounds of a genuine OT concept. In this sense, Coloured Containment is certainly the most OT of all interface theories: faithfulness is tested at the interface.

505

8.2.2. Representational identification of the old and the new Prince & Smolensky (1993) have introduced the idea of faithfulness and its first implementation, Containment. McCarthy & Prince (1995a) have abandoned this view on faithfulness to the benefit of Correspondence Theory. The driving force of van Oostendorp's work is the idea that Containment is the correct theory of faithfulness, despite the problems that it faces and the fact that Correspondence has become largely consensual in OT. Containment requires that every element of the phonological input representation be contained in the output. This immediately prompts the question how anything could be deleted at all. Van Oostendorp argues that autosegmental representations offer a natural solution: an item that is present in the input may be delinked in the output, to the effect that it is still present (thus satisfying Containment) but phonologically inert and inaudible. Table (198) below (from van Oostendorp 2007:124) shows a relevant situation where "deleted" material is floating. (198) /takp/ → [takpi] μ t

a

k

p

i

φ

On the assumption of a process whereby the first member of the cluster in the underlying /takp/ remains unpronounced, (198) shows the output representation: the /k/ is still present (hence Containment is satisfied) – it is just not parsed by phonology (which is represented by the φ structure). The

420 Chap 11: Optimality Theory consonant is thus floating, which according to regular autosegmental conventions means that it does not make it to the surface. The reverse process, epenthesis, also seems to escape a description in terms of Containment: what is the input-output relation of an item that is absent from the input when the input must be contained in the output? Under (198), though, the epenthetic character of the last vowel may be identified by the fact that it is not affiliated to morphology (which is represented by the μ structure). This is because all phonological material that is present in the input necessarily comes from lexical insertion and therefore belongs to a morpheme. In sum, items that float in regard of morphological structure are epenthetic, while those that float with respect to phonological structure are deleted. This is how van Oostendorp distinguishes the old and the new, a distinction that was already discussed in §307.

506

8.2.3. Information transmission: representational vs. procedural An interesting question to be asked is what exactly makes van Oostendorp's way to import morpho-syntactic information into phonology different from the regular mechanism, i.e. mapping as practised in Prosodic Phonology. The answer translation: both morpho-syntactic and phonological trees are static in van Oostendorp's system – there is no computational action that transforms one into another. Van Oostendorp thus rightfully claims a representational approach to the interface, i.e. one where (derivational) translation/mapping is not needed. This non-derivational mantra commits van Oostendorp to direct syntax, and also to the absence of readjustment (§91, called non-isomorphism in Prosodic Phonology, §416), which is only possible when a computational system carries out mapping. We will see in §511 below, however, that van Oostendorp in fact grants for regular mapping-created prosodic constituency, thereby building a hybrid system where all channels are open (while direct syntax and prosodic constituency ought to be competitors (see §416).

507

8.2.4. Independent representations marshal GEN and afford to be coloured The critical instrument of van Oostendorp's analysis under (198) are autosegmental representations: the old and the new are defined by the (non-)association to some (phonological or morphological) constituent. It is

More direct syntax: continuity between morphology and phonology 421

also clear that only sovereign representations in the sense of §453 will do. That is, representations need to have absolute validity: they must not be under the spell of the constraint chamber. According to van Oostendorp, this is guaranteed by Consistency of Exponence. This notion has been introduced by McCarthy & Prince (1993, 1994); it prohibits changes in the exponence of a phonologically specified morpheme. That is, the relationship between a morpheme and the phonological material that represents this morpheme in its lexical specification are one-to-one. No material created by phonological processes can have a morphological affiliation, and no morpheme can acquire a phonological representative that was absent from the lexicon. Van Oostendorp represents the players of this rigid correspondence by colours: the material of a given morpheme (both its morphological and phonological representatives) has a uniform colour, which is distinct from the colour of other morphemes. New material, i.e. which results from phonological processes, is colourless. An example of how a representation that distinguishes old (coloured) and new (colourless) material looks like appears under (201) below. Consistency of Exponence effects an important restriction on GEN, which is not all-powerful anymore (McCarthy & Prince 1993, 1994 are explicit on this fact): the colour-based formulation of van Oostendorp (2006a:105) states that "Gen cannot change the morphological colour of any phonological element."

508

8.2.5. Faithfulness between morphological and phonological structure This representational system allows van Oostendorp (2006a:106, 2007:125f) to formulate constraints that all have one thing in common: they impose restrictions on phonological processing on the grounds of the morphological identity of the items involved (see also van Oostendorp 2005). Table (199) below mentions three examples. (199) Coloured Containment constraints use morphological structure in order to restrict phonological action a. PARSE-Φ(α) [against deletion] the morphological element α must be incorporated into the phonological structure.

422 Chap 11: Optimality Theory (199) Coloured Containment constraints use morphological structure in order to restrict phonological action b. PARSE-Μ(α) the phonological element α must be incorporated into the morphological structure. c. ALTERNATION if an association line links two elements of colour α, the line should also have colour α.

The availability of morphological information in phonology also allows for the formulation of what van Oostendorp calls recoverability constraints. Recoverability, in both directions, has a general definition and may incarnate as individual constraints. Examples appear under (200) below. (200) recoverability constraints a. Morphological recoverability (MR) [a constraint family] Phonological structure mirrors morphological structure as closely as possible. 1. Integrity Old vowels may not contain new material. 2. Morphological syllable integrity All segments in a syllable should be in the same morphological domain as that syllable. The morphological domain of a segment S is the smallest morphological word in which S occurs. b. Phonological recoverability Every morpheme in the input should be represented in the phonological output.

An area where recoverability constraints are useful is opacity: quite plausibly, opacity disappears (though perhaps not in all cases) if surface representations offer permanent and unalterable access to the underlying form. This is the case for Coloured Containment representations (van Oostendorp 2006a:44ff). Consider for example the process /takp/ → [tapi] and its surface representation under (198). Assume that the epenthesis of the final vowel was caused by the presence of the /kp/ cluster: in the hypothetical language at hand, final clusters trigger the epenthesis of a vowel, and a subsequent process deletes obstruents in codas. Hence the existence of the -i is opaque in the phonetic form [tapi] – but not when we have access to the full-blown representation under (198) where the epenthesistriggering cluster is recoverable.

More direct syntax: continuity between morphology and phonology 423 509

8.2.6. Coloured Containment applied to derived environment effects Another area that suits the coloured distinction between old and new material are derived environment effects. These describe a situation where a phonological process applies across morpheme boundaries, while strings without boundaries remain unmodified (see §177). That is, the process is triggered in presence of new material: the phonological representatives of a foreign morpheme are new with respect to the old material of the original morpheme. By contrast, the same process does not apply if the agent and the patient belong to the same morpheme, i.e. when both are "old". Van Oostendorp argues that Coloured Containment is well suited for the analysis of derived environment effects because colours show derived environments.133 Van Oostendorp (2006a:92ff, 2007:135ff) reviews a number of classical cases and concludes that derived environment effects always involve spreading of some melodic prime (2006a:94, 2007:137). The architecture of Coloured Containment allows for the formulation of the constraint (199c) ALTERNATION which bears on the colour of association lines: if an association line links two elements of colour α, the line should also have colour α. That is, original pieces of the same morpheme must be linked in the underlying form. All other combinations are licit: an association line that is created by phonological processing (i.e. a line representing spreading) is absent underlyingly and therefore colourless. It may thus relate 1) material that belongs to different morphemes, 2) material that represents a morpheme and new, i.e. colourless material. Colourless material may either be epenthetic or the result of a phonological process such as spreading. Consider the following prototypical example from Korean which is discussed by van Oostendorp. In this language, voiceless dental stops are affricated before -i, provided that both segments belong to different morphemes. Hence /hæ tot-i/ "sunrise NOM" comes out as [hæ dodÉʒi], while the underived /mati/ "knot" remains [madi]. The Coloured Containment 133

Derived environment effects traditionally encompass phonologically and morphologically derived environments (§179). An environment counts as derived in the former sense iff at least one phonological process has contributed to its creation. Van Oostendorp (2006a:92ff, 2007:135ff) considers morphologically derived environments, which are of primary interest here. He discusses the case of phonologically derived environments in van Oostendorp (2006a:96f, 2007:140f) (more on the Polish case that organises the debate in this area in §517). Note that van Oostendorp (2007) is a piece of van Oostendorp (2006a), with identical structure and often identical wording.

424 Chap 11: Optimality Theory representation under (201) below shows why spreading cannot go into effect in the latter case (Greek letters in subscript indicate contrasting colours, "c" is shorthand for colourless). (201) Korean affrication a. /hæ tot-i/ "sunrise NOM" tα

oα dÉʒα

c

iβ α

b. /mati/ "knot" *mα aα dÉʒα c

[high] β

iα α

[high] α

As may be seen, ALTERNATION prohibits the existence of the slanted line under (201b) (the colourless spreading line relates two items of the same colour), but not under (201a) where the colourless spreading line associates an α-colour element with a β-colour item. Van Oostendorp (2006a:94ff) also analyses Turkish disharmonic roots along these lines. In the Turkish root-controlled harmony system, the observation is that root vowels impose their backness and roundness to suffix vowels as well as to epenthetic vowels, while roots with disharmonic lexical material are not harmonised. This is shown under (202) below. (202) Turkish disharmonic roots a. suffixes are harmonised NOMsg GENsg NOMpl ip ip-in ip-ler kɨz kɨz-ɨn kɨz-lar yüz yüz-ün yüz-ler b. epenthetic vowels are harmonised careful form colloquial form pranga fetters pɨranga kruvazör kuruvazör cruiser c. roots may be disharmonic vali governor kitap book bobin spool

rope girl face

Van Oostendorp argues that the process at hand is a derived environment effect because both classes of "new" items – suffixal vowels and vowels that break up an illicit cluster – are subject to harmony, while "old" vowels, i.e. those that represent the same morpheme underlyingly, remain unaffected. This is precisely what is predicted by ALTERNATION: colourless

More direct syntax: continuity between morphology and phonology 425

lines may relate old root vowels and new epenthetic vowels or suffixal vowels that bear a different colour; they may not, however, associate old vowels of the same colour. 510

8.2.7. Anti-Lexical Phonology: the interface representationalised Interestingly, thus, van Oostendorp's analysis of derived environment effects is entirely representational: at no point does any procedural element intervene. Once the morphologically informed Coloured Containment representation is available, the computation is strictly parallel. Therefore, van Oostendorp (2006a:5, 2007:144f) insists on the monostratal character of his system: only the surface representation is relevant: input representations or eventual intermediate steps are not needed. This of course is an interesting result for a theory such as OT that has a strong anti-derivational ambition. Derived environment effects are typical cyclic effects. As far as I can see, they have never been analysed by purely representational means. Van Oostendorp's approach may thus be understood as the opposite of Lexical Phonology: instead of proceduralising typical representational phenomena (§213), it representationalises typical procedural phenomena. Whether this move can be extended to other cyclic phenomena remains to be seen.

511

8.2.8. Mapping: a second means of talking to the phonology A number of questions are prompted by the architecture of Coloured Containment. One regards mapping: what exactly does phonological structure in a Coloured Containment representation (abbreviated as φ under (198)) encompass? In case it accommodates prosodic constituency, how has this structure been created? Roughly speaking, the answer is that the classical Prosodic Hierarchy is in place as before, and that the mapping mechanism which is at its origin works in the same way as elsewhere in OT. That is, the extra technology of Coloured Containment comes in addition to the standard interface mechanism. The detail is a little more complicated, though. Van Oostendorp departs from directionality-based mapping, which was introduced by Selkirk (1986) and became consensual in OT in the guise of the ALIGN family (see §§394,457). Van Oostendorp (2006a:50) argues that "the notions 'left' and 'right' are relevant only for phonology, not so much for morphology or syntax, in which other notions, such as hierarchy and embedding, play a role."

426 Chap 11: Optimality Theory Instead, he reverts back to Prince & Smolensky's (1993) constraint PR≈LX, which requires that morpheme boundaries coincide with the boundaries of prosodic constituents (but treats right and left boundaries on a par). Nevertheless, ALIGN is still present in the grammar: instead of translating morpho-syntactic into prosodic structure, though, it relates only prosodic constituents among themselves (van Oostendorp 2006a:49). ALIGN thus performs a kind of secondary mapping: primary mapping is assured by PR≈LX and imports morpho-syntactic information into the phonology. Its output, prosodic constituency, is then further aligned. This twofold mapping activity somehow interacts with the genuine technology of van Oostendorp's system, i.e. Coloured Containment constraints (199) and Recoverability constraints (200). These compare morphological and phonological structure, but the latter already includes morphosyntactic information that was imported by PR≈LX into prosodic constituency. The existence of two parallel ways for morpho-syntax to bear on phonology by representational means needs some clarification, namely regarding the issue of an eventual competition (or division of labour). As far as I can see, Coloured Containment is the only interface theory that allows for two distinct representational means of talking to the phonology.

512

8.3. Sign-Based Morphology (Orgun 1996a)

513

8.3.1. Monostratal HPSG-style representations where syntactic, semantic and phonological information is scrambled Orgun (1996a, 1998, 1999) (also Orgun & Inkelas 2002) proposes a model called Sign-Based Morphology (SBM), which is an application of the declarative monostratal approach to the interface. The declarative perspective is represented in phonology by Declarative Phonology (Scobbie et al. 1996, Coleman 2005), and in syntax by HPSG (Pollard & Sag 1994).134 134

Monostratalism is a broad architectural issue; it divides approaches before decisions regarding particular phonological theories can be made. It is therefore not obvious to accommodate the discussion of SBM in the chapter on OT. Orgun (1999:252) for example argues that the cyclic issue concerns the organisation of the interface alone and hence is independent of whatever phonological theory runs below that. Even though SBM is thus formally agnostic regarding the correct phonological theory, Orgun's work has been discussed on an OT backdrop, and Orgun couches his analyses in OT.

More direct syntax: continuity between morphology and phonology 427

Declarative models share the anti-derivational take with OT. Unlike OT, however, Orgun argues that no anti-derivational patches (cyclicitykillers) are needed for cyclic phenomena in SBM since the general monostratal architecture does a non-derivational job at the interface for free. According to him, cyclic derivation is a source of serialism only on the unwarranted (but standard) assumption that 1) terminal nodes of the morpho-syntactic tree are the only information-bearing elements and 2) concatenation is different from interpretation. By contrast, Orgun holds that rather than just terminal nodes, every node of the morpho-syntactic tree is considered a linguistic sign and thus fully specified for sound, meaning and morpho-syntactic properties. In SBM, every node is thus subject to phonological constraints: its properties are the result of a computation over the specifications of its daughters. In this perspective, then, "cyclic phonological effects follow as an automatic consequence of local tree wellformedness" or, in other words, "in declarative fashion from static constituent structure configurations" (both quotes Orgun 1996a:51). In order to see how SBM works, first consider under (203) below the regular arboreal representation of the Latin word dix-erā-mus "say pastperf-1pl", as opposed to its representation in SBM (example from Orgun 1996b:12). (203) terminal-based vs. sign-based structure of Latin dix-erā-mus a. terminal-based b. SBM representation β

syn ζ (#10,#7) = #13 sem ξ (#11,#8) = #14 phon φ (12,#9) = #15 /dixerāmus/

α dix erā mus

syn ζ (#1,#4) = #10 sem ξ (#2,#5) = #11 phon φ (#3,#6) = #12 /dixerā/ syn #1 sem #2 phon #3 /dix/ syn #1 sem #2 phon #3 /dix/

syn #4 sem #5 phon #6 /erā/

syn #7 sem #8 phon #9/mus/

428 Chap 11: Optimality Theory In SBM, every node is fully specified for phonological, syntactic and semantic information. Terminal nodes inherit their material from the lexicon (#1 to #9), while the content of a non-terminal node is the result of a computation over the material of its daughters at the three levels mentioned (#10 to #15). For instance, ξ(#2,#5) applies semantic computation to the string that consists of concatenated #2 and #5; the result is the semantic value #11. The construction of the morpho-syntactic tree is thus concomitant with the computation of phonological and semantic properties of the string, which necessarily follows the constituent structure from the most embedded to the topmost level. Each node, then, must be well-formed according to the constraints defined by each of the three components, i.e. syntax, semantics and phonology.

514

8.3.2. Cyclic effects in SBM Orgun (1996a:19ff, 1998, 1999:260ff) repeatedly uses the following example from Turkish in order to show how SBM handles cyclic effects. He reports that Turkish imposes a minimal word constraint on affixed forms, which must be at least bisyllabic (while unaffixed items may be monosyllabic). This is shown by the word for the musical note "do", which is doo and fine by itself. When the possessive marker -m is added, though, the result of simple concatenation is ungrammatical: *doo-m "my do". One could expect this to be solved by the suffixation of another item, which makes the overall construction bisyllabic: *doo-m-u "my do ACC". The presence of the accusative marker -u, however, does not produce a grammatical result either. According to Orgun, *doo-m-u, which satisfies the bisyllabic constraint on the surface, is ungrammatical because the constraint was violated at a previous derivational stage when the accusative marker was not present yet. Or rather, in the non-derivational perspective of SBM, the constituent dominating doo and -m in the overall structure [[[doo]m]u] is as much exposed to the bisyllabic constraint as the word-level constituent that dominates all three morphemes. That is, "every node in a given constituent structure is subject to phonological constraints" (Orgun 1996a:19), and the illformedness of any node at any of the three levels (syntactic, semantic, phonological) makes the entire construction ill-formed. The ground rule that all constituents must be well-formed regarding syntactic, semantic and phonological constraints imposes a ternary struc-

More direct syntax: continuity between morphology and phonology 429

ture for cases of non-cyclic phonology, i.e. where generalisations are surface-true. Orgun (1996a:23ff) reports that the root je "to eat" may stand alone, but as doo-m before is agrammatical when the passive suffix -n is added: *je-n. This time, however, the suffixation of another suffix is a valid repair: je-n-ir "to eat passive imperfective" is well-formed. This, Orgun argues, is due to constituent structure. The word at hand identifies as a flat ternary, rather than as a nested structure: the three morphemes are dominated by the same word-level constituent – hence there is no node that violates the bisyllabic restriction. 515

8.3.3. There is no interface if all is one and the same thing The phenomena that Orgun discusses are not typically cyclic. It remains to be seen what an SBM analysis of the classical data sets (such as affix classes) looks like. But let us assume that all cyclic phenomena can be handled in the way that was shown in the previous section. The questions, then, lie on the conceptual side: SBM is a representative of the monostratal perspective that is known from Declarative Phonology and HPSG in syntax. On the count of this approach, all is one and the same thing: computation is done in the representation, concatenation and interpretation are the same thing, and no difference is made between syntactic, semantic and phonological computation, which all apply at the same time to the same object. Hence SBM is an instantiation of the monostratal idea according to which representations should incorporate as much information as possible and thereby provide a correct post hoc picture of what has happened: the same representation contains lexical ingredients, computational mechanisms, the full derivational history with all intermediate steps and the final result. In short, an accurate description of the overall situation. The question is whether linguistics is about a post hoc description of the events, or about their explanation. Also, it may be asked whether the monostratal perspective is but a notational variant of the regular system that looks at things from the linguist's perspective once the battle is over. Evidence against the perspective of a notational variant is the ternary structure that Orgun is forced to propose in order to analyse non-cyclic morphology in Turkish (see the previous section). On the regular modular approach where concatenation and interpretation are distinct, no need for ternary morpho-syntactic structure arises. Orgun thus has to live with ternary structure (actually with n-ary constituents: a non-cyclic item may be

430 Chap 11: Optimality Theory made of n morphemes, which will all have to be sisters). This is certainly not an easy programme to defend on the morpho-syntactic side. Orgun's Sign-Based Morphology is not just an alternative approach to the morphology-phonology interaction that may be plugged in in place of some other interface theory. It supposes a peculiar view on the architecture of grammar that is at variance with the baseline of structuralist and generative thinking where morpho-syntax and phonology (as well as semantics) are different entities (modules). By contrast on monostratal assumptions, everything is one and the same thing. It therefore does not really make sense to talk about any interface at all.

516

9. Derived environment effects in OT

517

9.1. Introduction The OT literature has produced a few original proposals, i.e. which use genuine tools of OT, in order to come to grips with derived environment effects. Other approaches to the phenomenon that are entertained in OT are mere adaptations of previous analyses to the constraint-based vocabulary (adding to confusion, they may run under different labels, though). Van Oostendorp's (2006a, 2007) Coloured Containment and McCarthy's (2003a) Comparative Markedness fall into the former category: they attempt to achieve the relevant distinction between the root and affixbearing strings by contrasting "old" and "new" items: either by doubling each constraint (McCarthy), or by making constraints directly sensitive to morphological affiliation (van Oostendorp). Łubowicz (2002) also uses a genuine OT tool, Constraint Conjunction. Analogy (Burzio 2000a) has also been brought to bear, as much as interface constraints (Root Faithfulness, Anttila 2009) and a simple rephrasing of the observation that roots are immune in terms of a constraint (Cho's 2009 FAITH-LEX). The latter is also a revival of Kiparsky's old attempt at deriving derived environment effects from the Elsewhere Condition. Finally, Yu (2000) implements a specific mini-phonology for roots (a root cophonology).

Derived environment effects in OT 431 518

9.2. Coloured Containment and Constraint Conjunction (Łubowicz) We have already seen one way to handle derived environment effects in OT: van Oostendorp's (2006a, 2007) Coloured Containment (§509). Van Oostendorp (2006a:98ff, 2007:141ff) also provides a review of three other OT-based approaches to derived environment effects, i.e. Constraint Conjunction, Comparative Markedness and Root Faithfulness. Let us first look at the former. Discussion here entirely revolves around classical and oft-quoted data from Polish. These have originally been introduced by Rubach (1984:120f), whose analysis is adapted to OT by Łubowicz (2002, 2004, 2005). The case at hand is peculiar insofar as it concerns a phonologically derived environment (rather than a derived environment that is created by morphological concatenation, see §179). That is, a process goes into effect when it applies to an input that is itself the result of a phonological operation, but does not apply to items that were previously unaffected by phonological computation (i.e. to phonologically underived items). In Polish, the palatalised version of /g/ is [ʒ]: róg [ruk] "horn" roż-ek [rɔʒɛk] "id., diminutive", skrag-a [skarga] "complaint" - skarż-yć [skarʒɨtÉɕ] "to complain". Critical for the argument is the following SPEstyle assumption that Rubach (1984:110ff) has anchored in the analysis of Polish: /g/ goes through /dÉʒ/ on its way to [ʒ], even though /dÉʒ/ never appears on the surface. That is, there are two independent rules, palatalisation and (context-free) deaffrication, which apply in this order: the former first turns /skarg-ɨtÉɕ/ into /skardÉʒ-ɨtÉɕ/, which then serves as the input into the latter; the result is [skarʒɨtÉɕ]. Unlike /dÉʒ/ that is the result of the application of palatalisation, underlying /dÉʒ/ does not undergo deaffrication: it surfaces as such. This is shown by the word brydż [brɨdÉʒ] "bridge", whose diminutive form is brydż-ek [brɨdÉʒɛk] (rather than *[brɨʒɛk]). It needs to be stressed that the existence of any derived environment issue related to these data hinges on the assumption of the intermediate stage /dÉʒ/. On an analysis where /g/ directly palatalises to [ʒ], there is no deaffrication process and hence no derived environment effect at all. Recall that the intermediate item /dÉʒ/ never appears on the surface: there is no independent evidence for it, and deaffrication is context-free. In Rubach's orignal analysis, underlying /dÉʒ/ remains unaffected because the application of the deaffrication rule is restricted to derived envi-

432 Chap 11: Optimality Theory ronments. Łubowicz' analysis relies on the conjunction of a markedness constraint which militates against voiceless dental affricates, *dÉʒ, and a faithfulness constraint which requires coronal segments to remain coronal, IDENT-[CORONAL]. The result of the conjunction, [*dÉʒ & IDENT[CORONAL]]segment, assigns a violation mark to candidate segments which are dÉʒ and unfaithful to their coronal origin at the same time. This is the case of intermediate /dÉʒ/ in /skardÉʒ-ɨtÉɕ/ (from /skarg-ɨtÉɕ/), but not of underlying /dÉʒ/ in [brɨdÉʒɛk] (from /brɨdÉʒɛk/). Van Oostendorp (2006a:98f, 2007:142f) raises two objections against this analysis. For one thing, the formal tool of Constraint Conjunction is unrestricted, which means that derived environment effects related to the conjunction of any two constraints should be observed. This is obviously not the case since there is a non-arbitrary relationship between dÉʒ and deaffrication (rather than, say, dÉʒ and ONSET). The second objection points out that Łubowicz' analysis is made for phonologically derived environments: it has got nothing to say about morphological divisions. It is therefore difficult to see how it could be extended to regular morphologically derived environment effects. Van Oostendorp shows that the limitation to the "segment" as the domain of application of Łubowicz' conjunction [*dÉʒ & IDENT-[CORONAL]]segment is a stipulation that does not follow from anything; monster languages are described if the portion of the linear string to which the conjunction applies is varied. Łubowicz' (2005) attempt to restrict the domain of application to the segment requires McCarthy's (2003a) Comparative Markedness, on which more below. Van Oostendorp argues that derived environment effects, whether concerning phonologically or morphologically derived environments, are a unitary phenomenon and hence call for a uniform analysis. What all derived environment effects have in common is the fact of distinguishing old from new material: old either in terms of morphological concatenation or in terms of phonological derivation (epenthesis, palatalisation etc.). This distinction, van Oostendorp argues, is correctly expressed by his colours. 519

9.3. Comparative Markedness (McCarthy) and analogy (Burzio) Like van Oostendorp's Coloured Containment, McCarthy's (2003a) Comparative Markedness distinguishes between old and new items – not features, segments or morphemes, though, but constraints. That is, every

Derived environment effects in OT 433

markedness constraint divides into a version that refers only to old violations, and another one that counts only new violations. For example, *dÉʒNew will apply only to "new" dÉʒ, that is to those that were not underlyingly present. By contrast, *dÉʒOld will concern underlying items, but not those that are produced by GEN. High ranking *dÉʒNew therefore achieves the derived environment effect observed in Polish. An argument against using Comparative Markedness for derived environment effects is that the reverse situation, i.e. high ranking *dÉʒOld, produces "non-derived environment effects", that is in our case the deaffrication of underlying, but not of derived dÉʒ. This pattern, however, is absent from the record: only underived environments block the application of rules. Also, Comparative Markedness only distinguishes old and new constraints – no statement is made regarding morphological divisions. It is therefore unable to distinguish between old and new morphemes; the extension of the analysis to morphologically derived environments is thus impossible without additional machinery. The extra machinery that Comparative Markedness uses in order to get to grips with morphologically derived environments is analogy, i.e. Output-Output faithfulness in terms of OT. Recall from §497 that OO-constraints have also been used as a cyclicity killer. Old and new morphemes are then distinguished by Output-OutputNew constraints, which are only satisfied if the violation was absent in the underlying form. Van Oostendorp (2006a:99, 2007:143) points out that this mechanism, just like for phonologically derived environments, predicts the existence of unattested monster languages. This is because Output-OutputNew implies the existence of Output-OutputOld, which counts violations only if there is an underlying form in which they are already present. This allows for constructing anticyclicity, i.e. a pattern where a process applies to mono-morphemic items, but is blocked when a morpheme boundary is present in the string. This would be an underived environment effect, something that is just as unattested as blocking in phonologically derived environments. Finally, OO-constraints have also been used alone in order to account for derived environment effects: according to Burzio (2000a,b), their multifunctionality also covers this phenomenon.

434 Chap 11: Optimality Theory 520

9.4. Direct syntax (interface constraints): Root Faithfulness (Anttila) Of course, the easiest way of making a difference between underived roots and a morphologically complex string is to make direct reference to the root. We are thus back to a version of interface constraints, which – in overt violation of Indirect Reference – make direct reference to designated morpho-syntactic categories (see §494). This is what Anttila (2009) promotes: he makes faithfulness constraints selectively applicable to roots only, which of course produces the desired effect. If FAITH-ROOT outranks the constraint that is responsible for the process at hand, modified candidates will be optimal everywhere but in roots. In other words, Root Faithfulness simply translates the descriptive observation into a morphological diacritic which is placed on constraints and restricts their application to a subset of morpho-syntactic categories.

521

9.5. Revival of the Elsewhere Condition (Cho, Iverson) Anttila's (2009) analysis of derived environment effects appeared in the same book as Cho's (2009), a Festschrift for Paul Kiparsky. Iverson (2004) is a development of Cho's contribution, which circulated as a manuscript (dated 2002 by Iverson). Cho (2009) argues for a revival of Kiparsky's (1982a,b) attempt to interpret derived environment effects in terms of the Elsewhere Condition. Recall from §191 that this requires to make each lexical entry a rule whose output is identical to its input (lexical identity rules). Since no rule can be more specific than lexical identity rules which simply rewrite their input, all rules that try to modify lexical entries in isolation will be failed by the Elsewhere Condition which in case of competition gives precedence to the more specific rule. Rules that try to modify plurimorphemic strings, however, will not have to compete with lexical identity rules since there is no lexical entry corresponding to the morphologically complex item. In the OT adaptation of Cho (2009), the prohibition for isolated lexical items to be modified by phonological computation is called Lexical Faithfulness. It is encoded as such in a top-ranked constraint, FAITH-LEX: "output candidates must not be distinct from corresponding lexical entries". GEN may thus produce all kinds of variation of a mono-morphemic item: all modified candidates will incur a fatal violation of FAITH-LEX, to the effect that the completely faithful candidate always wins. As in Kiparsky's original version, morphologically complex candidates whose underlying

OT is a strong modularity-offender: violations are in-built 435

form was modified will not be eliminated by FAITH-LEX since there is no corresponding lexical entry that they need to be faithful to. This analysis simply translates the observation that mono-morphemic items do not change into the prose of a constraint, which is made undominated. The same questions then arise as in the 80s: it is not true that monomorphemic items remain unaltered all through. They are affected by "lowlevel", "phonetic" or "automatic" processes: the structure-building vs. structure-changing distinction may be relevant, which means that underspecification becomes an issue as well. This has all been discussed in the 80s (see §§192ff), and Iverson (2004) tries to implement the insights gained together with his own in the FAITH-LEX frame defined by Cho (2009).

522

9.6. Co-phonologies (Yu) Finally, Yu (2000) puts co-phonologies to use in order to account for derived environment effects: a specific mini-grammar is devoted to roots. These therefore experience a phonological computation that is different from strings that contain affixes. In Yu's (2000:122) terms, a root-specific phonology φroot is thus opposed to an affix-specific phonology φaffix.

523

10. OT is a strong modularity-offender: violations are in-built

524

10.1. Modularity yes or no – this is the question, however rarely addressed in the OT literature The OT literature is not very wordy regarding architectural issues. Kager's (1999) overview of the theory for example offers a chapter on acquisition and OT syntax, but relations of phonology with morpho-syntax are only mentioned in a sub-chapter on alignment (and the modular issue is entirely absent as far as I can see). The existence of modules implies a derivational relationship between them, which in (Chomskian) linguistics comes down to the inverted T (§86): the input to phonology (and semantics) is somehow the output of syntax. At best, then, the modular architecture (on which more in the Interlude §586) appears to be a background assumption in OT that most of the time remains implicit and does not really impact particular analyses.

436 Chap 11: Optimality Theory Russell (1997) describes the interface landscape in OT as follows. (204) "It is often tacitly assumed that there is a morphology-like component which chooses the right underlying representations and ships them off to GEN in the phonological component, complete with handy morphological annotations like 'Prefix' or 'Stem', but little effort has been spent on figuring out what this component is or how it works. In fact, while there is a growing body of work in OT syntax and OT phonology, there are still few clear ideas about how they relate to each other. Is there a classical serial relationship between the two, with an OT syntax first calculating the optimal syntactic representation, which then serves as the input to an OT phonology (perhaps stopping off at an OT morphology component in the middle)? Or is there some larger, integrated grammar, where EVAL chooses all at once the best overall combination of a phonological, a syntactic and a semantic representation?" Russell (1997:129)

Two years later, he confirms that the classical modular architecture with a serial order among modules is the regular backdrop of work in OT. (205) "Most work in OT seems to have implicitly adopted this assembly-line view of the overall architecture of language. While individual modules (specifically phonology and syntax) are argued to function non-derivationally, the relationship between modules is usually assumed to be linear and directional. Each module has an input and an optimal output – the inputs come from somewhere, and the outputs go somewhere for further processing." Russell (1999:6)

Burzio (2007), another item of the sparse body of literature that discusses the architecture of grammar in OT, does not merely mourn the derivational legacy of generative thinking, or leave it at a timid suggestion for a fully scrambled constraint chamber that encompasses phonetics, phonology, morphology, syntax and semantics. Addressing modularity explicitly, Burzio declares that this view of the architecture is falsified and needs to be overcome.

OT is a strong modularity-offender: violations are in-built 437 (206) "The mainstream generative tradition had postulated discrete modules that feed one another in a cascading arrangement: Morphology would feed Phonology which would then feed Phonetics. This hypothesis makes grossly incorrect predictions about the range of possible interactions. It predicts that Phonology could not be driven by Phonetics except perhaps indirectly via evolutionary effects that weed out phonetically ill-suited phonologies, and it predicts that Phonology may not have any effect on Morphology. The incorrectness of the first prediction has been forcefully underscored by a very productive line of work of recent years aimed to show how perceptual cues and perceptual distances are behind phenomena that have been the traditional bread and butter of phonological work. See, e.g. Hayes et al. (2004). The present article addresses the incorrectness of the second prediction, by considering syncretism – an eminently morphological phenomenon, which is nonetheless controlled by phonological factors in certain cases." Burzio (2007:1)

This statement is commendable since it clarifies the situation: in this perspective modularity-violating OT practice is no longer in contradiction with the architectural superstructure. On the other hand, though, it requires a revision of the general generative philosophy, which is deeply anchored in the modular approach. If Burzio is right, the generative approach to language as we know it for some fifty years is fundamentally wrong, and so is current syntax where clear-cut modular contours are the heart of Phase Theory and indeed of the interface-driven minimalist programme. The issue of whether modularity is right or wrong is not addressed on the pages below. The Interlude §586 is devoted to the presentation of modularity in Cognitive Science and the generative enterprise. The following sections merely identify in which way exactly OT systematically or occasionally (i.e. depending on the tools chosen by the analyst) violates modularity. Also, the connectionist part of its genetic endowment that inveigles to modularity violation is discussed (§529). In sum, OT is a strong modularity offender, but we have seen in previous chapters that OT is not alone in this respect in generative quarters (SPE in §99, direct syntax approaches in §§137,407). A summary of generative modularity offenders is provided in §702 below. In completion of the discussion in §469, three critical issues regarding the positioning of OT with respect to modularity are thus discussed below: ALIGN and WRAP make permanent reference to morpho-syntactic categories, the formulation of constraints often conflates phonological and morphological statements, and mapping is done in the phonology (i.e. ALIGN and WRAP are interspersed with regular phonological constraints), rather than in modular no man's land.

438 Chap 11: Optimality Theory 525

10.2. Direct Syntax in OT is regular and uncontradicted The debate between direct syntax approaches and Indirect Reference that lied at the heart of the interface debate in the 80s has been reviewed in §406. The position that is consensual in generative thinking since then is that reference to morpho-syntactic information is necessarily indirect. Also, direct syntax and Indirect Reference are in competition: there is no way to make both perspectives coexist. If direct reference to morpho-syntactic information is permitted, Prosodic Phonology is superfluous and has to go. Against this backdrop, the observation is that the OT literature makes regular use of direct reference to morpho-syntactic categories, typically without addressing the issue raised by Indirect Reference. Also, direct syntax elements and prosodic constituency appear to regularly coexist without this causing indisposition. Hence the Prosodic Hierarchy is inherited from the 80s, but made an empty shell. Pak (2008:60ff) and Samuels (2009a:284ff) are rare occasions where the existence of the Prosodic Hierarchy in a direct syntax environment is (rightfully) called into question (see also §580). Three instances of direct syntax in OT have been discussed: interface constraints (§494) are the most overt representatives, but van Oostendorp's Coloured Containment (§503) also falls into this category. The most frequent use of direct reference to morpho-syntactic items, however, is made by ALIGN and WRAP. This instance of direct syntax is important since the constraints at hand are used by all versions of OT as far as I can see. Note that ALIGN and WRAP not only make reference to morphosyntactic (tree) structure: they also constantly refer to morpho-syntactic labels (the stem, the DP etc.). Kager (1999) for example summarises the activity of ALIGN as follows. (207) "The categories 'Cat1' and 'Cat2' range over the alphabets of grammatical and phonological categories, for example: (68) GrammCat: {Word, Stem, Root, Affix, …} ProsCat: {PrWd, Foot, Syllable, Mora, …} These categories can also be filled by specific morphemes in the grammars of individual languages." Kager (1999:118f, emphasis in original)

This is also true for Stratal OT which, recall from §488, fosters a particular interest in constraining the access of phonology to morpho-syntactic information. Bermúdez-Otero (1999:112) says that alignment constraints should have a "very limited access to non-phonological categories". The issue, however, is categorical, not gradient: either there is or there is no

OT is a strong modularity-offender: violations are in-built 439

direct reference to morpho-syntactic categories. Violating Indirect Reference just a little bit has exactly the same effect as violating it a whole lot: modularity and the Prosodic Hierarchy are thrown over board. 526

10.3. Parallel mapping puts the Translator's Office in the phonology Another effect of constraint-based mapping has been discussed in §458: translation and reference to the output of translation are conflated. In Prosodic Phonology, translation is done by mapping rules; phonological processes then make reference to its output, the Prosodic Hierarchy. These reference-making processes are necessarily different from the mapping activity, which they must serially follow. This is also required by the modular architecture: mapping rules have access to both morpho-syntactic and phonological structure since they convert one into the other. Modules are domain specific and hence cannot parse the categories of other modules (§611). Therefore, translation cannot be done in either morpho-syntax or phonology: the Translator's Office must stand in modular no man's land (see §413). The effect of parallel mapping, however, is that translation is done in the phonology: ALIGN and WRAP are interspersed with regular phonological constraints in the same constraint ranking. This situation is incompatible with a modular architecture.

527

10.4. Scrambling: morpho-phonological contours are blurred The parallel ambition of OT fosters a tendency to scramble the computation of information that belongs to different domains: morphological, phonological, phonetic and perhaps even syntactic and semantic constraints cohabitate in the same constraint chamber. It was shown in §469 that this everything-is-one perspective appears to be the default assumption in classical OT. The undifferentiated application of the anti-derivational philosophy has created this trope which the theory itself does not enforce: the piece-transporting mechanism (cyclic spell-out) could be perfectly derivational if the parallel ambition were restricted to phonology proper (the phonological module). Kager (2000) promotes quite the opposite view: rather than considering the scrambling of phonological and morphological constraints a prob-

440 Chap 11: Optimality Theory lem, he argues that this morpho-phonological intimacy is an advantage: the more modular contours are blurred, the better a theory fares. (208) "Phonological and morphological constraints are ranked together in a single hierarchy. One might argue that parallelism is the counterpart of the 'interleaving' of morphological and phonological rules in the derivational model of Lexical Phonology. However, parallel Correspondence Theory predicts a broader kind of sensitivity of morphology to phonology than is possible under interleaving Lexical Phonology. While interleaving restricts phonological sensitivity of affixation to properties that are present in the stem 'before' the affixation, the parallel model allows for sensitivity to the full range of output properties of the base-plus-affix combination." Kager (2000:123)

Here again, the total-scrambling view is exposed without discussion of the fact that it conflicts with the basic generative interface architecture and modularity. Another issue is the body of constraints itself: since formulations are in prose and entirely unrestricted, they may well, and actually do, contain both phonological and morphological instructions. Hence even if there were a ban on interspersing phonological and morphological constraints, modular contours would still be blurred since it is sometimes impossible to decide whether a constraint is morphological or phonological in kind. Yip (1998) is explicit on this indecision. (209) "These results make it hard to identify a clear dividing line between morphology and phonology. What is more, they go much further to blur the distinction than does the interleaving of phonology and morphology found in lexical phonology. In lexical phonology, each component has its own character: the entities are different, and the rules are different. In Optimality Theory, this is not necessarily the case. Alignment is the most striking example. Alignment appears to play a role in pure morphology, in pure phonology, and at the interface." Yip (1998:219)

But of course, even if phonological and morphological constraints could be properly distinguished, this would not guarantee a modular segregation: with explicit reference to Natural Generative Phonology, Green (2007) does away with constraints whose body is ambiguous, but crucially maintains scrambling of the clean phonological and morphological result.

OT is a strong modularity-offender: violations are in-built 441 (210) "The role of phonology [reduces] to those processes that can be analyzed solely in terms of universal markedness constraints and their interaction with faithfulness constraints, while placing the responsibility for processes that make reference to language-specific properties in the domain of morphology, by using language-specific morphological constraints that are ranked in the same hierarchy with faithfulness constraints und universal markedness constraints." Green (2007:171)

The only voice that I have come across which pleads for a clear distinction between phonology and morphology as well as for a restricted (rather than an unlimited) access to morphological information is Bermúdez-Otero's (forth a) Stratal OT (§488). But even here, parallel mapping introduces massive direct reference to morpho-syntactic categories, in violation of modularity.

528

10.5. Radical scrambling: one single constraint ranking for morpho-syntax, semantics and phonology Russell (1999) argues for a radical version of the scrambling perspective which, though more articulated, joins monostratal HPSG practice where all is one and the same thing anyway (see §512): language is made of one single constraint chamber where morpho-syntactic, semantic and phonological constraints are interleaved (a slightly less radical variant is Wolf's 2008 system where phonology and morphology, but not syntax, are merged into the same constraint hierarchy, see §472). Russell (1997:129) entertains this view as a mere option (see the quote in §524), but adopts it two years later. (211) "An OT grammar evaluates all sub-representations (e.g., phonology, syntax, semantics) in parallel. There is no serial derivation between modules such that, for example, syntax is the 'input' to morphology or phonology. […] MOT [Russell's model] rejects the assembly-line view of how subrepresentations are related to each other. It takes seriously the claim that the job of a grammar is not to construct a representation to order (or even to choose a representation based on some input), but simply to look at a complete linguistic representation and judge whether it is a legal or illegal representation of the language." Russell (1999:5f)

The serial relationship between modules (the assembly-line view), Russell argues, is unwarranted in a parallel perspective. However, he does not argue against modularity as such: on his analysis modular status is granted to morpho-syntax, semantics and phonology because they "deal

442 Chap 11: Optimality Theory with qualitatively different kinds of formal stuff" (Russell 1999:15). As a consequence, they enjoy different representations. Rather than being subject to three distinct computations, however, these are then evaluated by grammatical activity in parallel and at the same time. That is, an "OT grammar can impose interface constraints [in the sense of §494] on which phonological, syntactic, and semantic representations can co-occur with each other" (Russell 1999:5). This view is incompatible with Fodorian modularity for sure: modules are computational units. It does not make sense to dissociate representations, which contain the domain specific vocabulary that is computed, from the computation itself. There is no way for distinct representations to live in different modules while being subject to a uniform computation. 529

10.6. Two souls are dwelling in OT: generative and connectionist Regular practice in OT has reached an unprecedented degree of scrambling between phonological and extra-phonological information. As it stands, the result is incompatible with a modular architecture. Modularity, however, is the spine of generative interface theory. This issue is largely unreflected in the OT literature. As far as I can see, no intrinsic property of OT requires to go down the non-modular road, even if the scrambling of information from all sides appears to be a typical trope in OT. On the other hand, it is also true that OT would have to undergo a thorough revision of deeply rooted elements such as ALIGN, should it be made compatible with modularity. The reason for the OT trope to blur contours of functional units may well be the orientation of the theory towards constraint-satisfaction. Before OT was born, constraint satisfaction and parallel (rather than serial) computation was identified as a core property of connectionism, the theory of the cognitive system that competes with modularity (see §588). David Rumelhart, a major figure of the connectionist approach, provides the following explanation of how computation works in connectionism. (212) "In addition to the fact that connectionist systems are capable of exploiting parallelism in computation and mimicking brain-style computation, connectionist systems are important because they provide good solutions to a number of very difficult computational problems that seem to arise often in models of cognition. In particular they are good at solving constraintsatisfaction problems. […]

OT is a strong modularity-offender: violations are in-built 443 Many cognitive-science problems are usefully conceptualized as constraintsatisfaction problems in which a solution is given through the satisfaction of a very large number of mutually interacting constraints. The problem is to devise a computational algorithm that is capable of efficiently implementing such a system. Connectionist systems are ideal for implementing such a constraint-satisfaction system." Rumelhart (1989:142)

Based on such a description, an independent observer may conclude that OT is a connectionist theory. This impression will be reinforced by another contrast between connectionism and modularity: the former promotes an all-purpose computation where computational units can work with any input and are not specialised for any particular task, while the latter devises specific and unexchangeable computational units that only understand a particular vocabulary and are only able to carry out a particular task (§597). Connectionism is indeed a significant piece of OT's genetic endowment: Paul Smolensky comes from Cognitive Science and was at the forefront of the connectionist scene when the theory individuated from Artificial Intelligence in the mid 80s (Smolensky 1987, 1988a,b). Smolensky (2003:385f) himself explains this development, and the connectionist foundations of OT are laid out at length in Smolensky & Legendre (2006). Obviously, however, OT is also a symbolic theory: connectionist parallel distributed computation works on symbolic units, typically segments. This is in contradiction with connectionism, which is content-free, i.e. rejects symbolic representations (see §593). Smolensky & Legendre (2006) precisely set out to bridge this gap. This line of attack is reminiscent of the early days, when connectionism and traditional symbolic theories were understood as complementary, the former being an interesting way of implementing the latter at a lower level that is closer to neurology (the sub-symbolic level). Connectionism only grew into an irreconcilable competitor of the standard symbolic approach as the theory unfolded (see §599). Surely many OT practitioners live in the belief that OT is a straight generative theory – which it is, but only on the one hand. On the other hand, connectionism was inserted into its genetic code before OT was born, and silently pushes the theory to conflict with modularity and generative essentials ever since. The pre-natal bridge to connectionism is not usually made explicit in the OT literature, and the same goes for its connectionist behaviour (see Scheer 2010a, forth). In a connectionist environment, the all-is-one trope of OT makes perfect sense and is actually welcomed. In a modular perspective, it is a constant source of conflict. The conclusion, then, is as before: OT needs to

444 Chap 11: Optimality Theory make up its mind where it stands, and to face the consequences of this decision.

530

11. Conclusion OT does not make any representational contribution to the question of how extra-phonological information is processed in phonology because it does not develop genuine phonological vocabulary. Rather, the Prosodic Hierarchy is adapted to the constraint-based environment in the same way as other representational tools of the 80s are. On the procedural side, the antiderivational stance of classical OT has prompted anti-cyclicity, an attitude that in fact rejects the existence of any procedural aspect in interface management. This means that Interface Dualism is abandoned together with cyclic derivation (inside-out interpretation), two hallmarks of generative thinking. The global application of the anti-derivational requirement to grammar without differentiating between phonology and the interface (or even morphology and syntax) is another serious issue (§469). No property of OT forces it to go down this road (the anti-derivational ambition concerns phonological computation, not the interface), and Stratal OT/DOT indeed promote an architecture where relations with morpho-syntax are serial, while phonological computation proper is strictly parallel. As elsewhere in OT, competing and incompatible solutions have been proposed for the same problem, cyclicity in our case. There are not as many cyclicity-killers as there are opacity-killers, but the phenomenon is the same: distinct and eventually incompatible proposals coexist without this arousing much comparative discussion (§484). This is true namely for modern heirs of Lexical Phonology, i.e. solutions that accommodate morpheme-specific mini-grammars, which may be parallel (co-phonologies, indexed constraints) or serially ordered (Stratal OT, DOT). A genuine contribution to interface theory is the way in which OT implements mapping, i.e. the translation of morpho-syntactic structure into prosodic constituency. Constraint-based mapping follows from the unavailability of ordered (mapping) rules. It introduces a new logic whereby the translational process and the phonological reference to its result are merged: ALIGN and WRAP do all the job at once (§457). While this may appear to be a technical advantage at first, it turns out to be part of the serious issue that OT has with modularity: mapping is done in the phonology (rather than in modular no-man's land).

Conclusion 445

A strength of OT is certainly the description of a large array of typological variation with a small set of interacting constraints. This aspect of the theory comes to bear in the area of mapping: the formerly monolithic Strict Layer Hypothesis has been split into various smaller and violable units. Also, the study of mapping-conditioning factors that were not in the focus of previous research has been promoted (eurhythmy, information structure) (§461). The mapping puzzle as such, however, remains untapped: as in previous theories, we still have no handle on why this or that group of morpho-syntactic boundaries do or do not provoke a phonological effect. Finally, OT has a serious issue with the generative architecture of grammar because of its misty and largely unreflected relationship with modularity. OT appears to have a natural tendency to scramble conditioning factors of various ontological and linguistic origins in the same constraint chamber. It seems to me that this tendency is not just the result of a particular practice that could as well be different; rather, it is a trope that roots in output orientation, competition and parallelism, in short: in the connectionist endowment of OT. OT is by far not the only modularity offender in generative quarters (a summary is provided in §702), but its offending character is systematic and deeply rooted (Scheer 2010a, forth).

531

Chapter 12 Distributed Morphology

532

1. Introduction The centre of gravity of Distributed Morphology revolves around antilexicalism, late insertion, the non-difference of morphology and syntax and the distribution of list-type information over three distinct lexica. Relevant references include Halle & Marantz (1993), Halle (1997a), Marantz (1997a), Embick & Noyer (2007), Embick (forth); Harley & Noyer (1999) provide a well informed overview. The communication of morpho-syntax with phonology is not a primary concern of Distributed Morphology; therefore relatively little work has been done that is directly relevant for phonology. As we will see below, Distributed Morphology has an issue with Lexical Phonology, namely with the Lexicon. This is consistent with the fact that it rejects morpheme-specific mini-phonologies – not explicitly, but constantly in practice. That is, Distributed Morphology follows the classical take of SPE whose modern incarnations are Halle & Vergnaud (1987a) and Kaye (1995): the same computational system assesses strings of whatever morphological composition.135 On the other hand, Distributed Morphology subscribes to Chomsky's derivation by phase (§304), which means that it is interactionist – like Lexical Phonology and Kaye (1995), but unlike SPE and Halle & Vergnaud (1987a). The roadmap is as follows. The issue that Distributed Morphology has with Lexical Phonology is introduced in §533 together with the central tenets of the theory. The analysis of opacity in Distributed Morphology is then discussed in §541: the classical view where it is the result of computation (selective spell-out and Phase Impenetrability) is opposed to the DM alternative that is based on a contrast in arboreal structure (direct merge of the root with a morpheme). Other relevant aspects are the free distribution of phonological and semantic opacity (§564), the eventual independence of PF and LF phases (asymmetric spell-out, §567) and the difference between predictable (párenthood) and unpredictable (cómparable) PF opacity (§568). The possible existence of lexicalist analyses in the anti-lexicalist 135

Recall from §§234,283 that chunk-specific phonologies are a separate question.

448 Chap 12: Distributed Morphology environment of Distributed Morphology is discussed in §569, and finally the modularity-offending device PF movement is considered (§574).

533

2. Setting the scene: Distributed Morphology vs. Lexical Phonology

534

2.1. From Halle & Vergnaud (1987a) to Distributed Morphology: against the two-place approach Halle & Vergnaud (1987a) may be interpreted as a first attempt to unhorse Lexical Phonology which, however, was too dominant in the 80s in order for a frontal attack to be successful. Halle & Vergnaud therefore chose a reformist, rather than a revolutionary strategy (Halle 1997b:302 talks about "the Halle & Vergnaud 1987a version of lexical phonology", see §218). What they do, however, is to eliminate the two headstones of Lexical Phonology, morpheme-specific mini-phonologies and interactionism (§212). While Distributed Morphology (and thus Morris Halle) subscribes to derivation by phase and hence to interactionism (§304), Halle & Vergnaud's rejection of morpheme-specific mini-phonologies is carried over. Recall from §303 that this is consistent: derivation by phase is the modern incarnation of Halle & Vergnaud's selective spell-out (§225), which does the same job as morpheme-specific mini-phonologies; having both devices in the same theory would be redundant (§303). The modern charge against Lexical Phonology that completes this programme is led by Marantz (1997a, 2001). The bone of contention is the Lexicon: Lexical Phonology concatenates items in two distinct and procedurally separated locations, and it computes the interpretation of sound and meaning twice: once in the Lexicon whose output are words, another time after syntax, which produces sentences. This is what Marantz (2000, 2001) calls the two-place approach, which is depicted under (213) below. Contrasting with this scenario, Distributed Morphology allows for only one location where objects are glued together (morpho-syntax); also, concatenation is done by a single device, Merge, which operates over morphemes and words alike. Marantz (2000, 2001) refers to this alternative as the single engine hypothesis.

Setting the scene: Distributed Morphology vs. Lexical Phonology 449

(213) architecture of grammar in Lexical Phonology basic Lexicon content and output: underived roots and affixes

Lexicon concatenation of building blocks coming from the basic Lexicon. Output: words

syntax concatenation of building blocks coming from the Lexicon. Output: sentences

computation computation of of sound meaning

computation computation of sound of meaning

535

2.2. The single engine approach is agnostic with respect to the phonological interpretation of the string (morpheme- and chunk-specific parsing) Marantz' slogans "No escape from syntax" and "Syntactic hierarchical structure all the way down" illustrate the claim that there is only one concatenative engine. It is important to sort out what exactly the single engine approach is incompatible with. As far as I can see, only the existence of two separate concatenative systems for the construction of words and sentences conflicts with the tenets of Distributed Morphology. Phonological interpretation is not concerned at all: Distributed Morphology rejects multiple computational systems in morpho-syntax, but has got nothing to say about multiple computational systems in phonology. That is, from the vantage point of DM, PF may or may not accommodate morpheme-specific and/or chunkspecific mini-phonologies (§234). These may indeed be specified to apply to different morphemes (level 1 vs. level 2) or chunks (word-level, Praguian segregation) independently of whether the string was pieced together by one or several engines.

450 Chap 12: Distributed Morphology It is only the independent commitment of Distributed Morphology to derivation by phase (and hence to Phase Impenetrability) that rules out the morpheme-specific perspective (§303). 536

2.3. The general architecture of Distributed Morphology Let us now look at the general architecture of Distributed Morphology. The classical inverted T model that grounds generative thinking since the 60s is repeated below for convenience (from §86). (214)

the inverted T model lexicon morpho-syntax (concatenation of building blocks)

PF (computation of sound)

LF (computation of meaning)

Distributed Morphology endorses this scenario but, as was reported, conflates syntax and morphology. The theory also elaborates on the status and function of the lower part of the syntactic tree: terminal features may be (re)arranged by specific (morphological?) processes such as fission, fusion and impoverishment. These take place between syntax and lexical insertion, i.e. before feature bundles that qualify for morphemic status are replaced by phonologically interpretable material. Late Insertion embodies the insight that syntax manipulates only abstract morpho-syntactico-semantic features that are void of phonological content. This perspective departs from the earlier Government & Binding architecture where items with full phonological specification were manipulated in syntax ("sealed suitcases"). This is where the name of the theory comes from: rather than having a unique lexicon where all information is grouped in the entries of one single list, non-predictable properties are distributed over three distinct lists: 1) the input to the syntax (morpho-syntactico-semantic features), 2) vocabulary items (the input to phonology) which consist only of phonological material and replace morpho-syntactico-semantic features upon lexical

Setting the scene: Distributed Morphology vs. Lexical Phonology 451

insertion, and 3) the so-called encyclopaedia where vocabulary items are associated to meaning, and idioms and non-linguistic knowledge stored.136 The implementation of Late Insertion supposes a phonological lexicon (the vocabulary, consisting of vocabulary items) that is underspecified in comparison to earlier practice: vocabulary items do not supply morphosyntactic features; they only provide phonological material as well as relevant information regarding the context in which this material may be inserted. The general architecture of Distributed Morphology is thus like under (215) below (adapted from Marantz 1997a:204). (215)

general architecture of Distributed Morphology List 1 (morpho-syntactic and semantic features)

Syntax (concatenation of building blocks: Merge and Move)

Morphology (fission, fusion, impoverishment) List 2: vocabulary

Late Insertion

PF

LF

List 3: encyclopaedia

This organisation is discussed in the following section.

136

In Distributed Morphology, the word "idiom" does not just mean "an expression whose meaning cannot be deduced from the meaning of its components"; rather, it refers to any sound-meaning association. Hence "dog" (meaning "animal with four legs that barks etc.") is as much an idiom as "kick the bucket" (meaning "to die") (but grammatical morphemes are not if their meaning follows from morpho-syntactic features).

452 Chap 12: Distributed Morphology 537

2.4. The unity of syntax and morphology under debate

538

2.4.1. Specific morphological operations: does DM live up to its ambition? According to Distributed Morphology, there is only one location where items can be glued together, and one set of principles that controls the gluing mechanism: "Merge and Move". Table (215), though, depicts syntax and morphology in two distinct boxes. This expresses the doubt that Distributed Morphology as it stands lives up to its ambition: operations such as fission, fusion or impoverishment have been introduced as a consequence of anti-lexicalism and Late Insertion, but are unknown in syntax. A related issue is post-syntactic movement (PF movement, Embick & Noyer 2001, Embick & Noyer 2007:318ff), which also considerably weakens the theory: it splits Merge into a syntactic (before vocabulary insertion) and a morphological (after vocabulary insertion) version, which do not obey the same locality properties. This for sure is washing out the ambition of a unified concatenative device (more on PF movement in §574 below). In sum, crediting Distributed Morphology with a unified computation for syntax and morphology is not an easy thing to do. Williams (2007) for example raises the question whether Distributed Morphology is really that much different from classical lexicalist frameworks. He concludes that the difference lies in word, rather than in fact.

539

2.4.2. Voices that argue for the traditional stance: syntax ≠ morphology Beyond the question whether Distributed Morphology stands up to its unitary ambition, the claim that morphology is just the syntax of morphemes is rejected by a large body of literature, which grants some degree of autonomy to morphology. Among the approaches that defend an autonomous morphology are Aronoff (1976), Booij (1977 et passim), St. Anderson (1982, 1992), Ackema & Neeleman (2004, 2007) and the work around the Lexical Integrity Hypothesis that roots in Lexical Phonology (Mediterranean Morphology Meetings, e.g. Booij et al. 2007). Julien (2007) and Lieber & Scalise (2007) provide a survey of the debate regarding the relationship of morphology and syntax. The more specific question of (anti-)lexicalism is discussed for example in Roeper (2005), Newmeyer (2005) and Li (2005) (see also §569).

Setting the scene: Distributed Morphology vs. Lexical Phonology 453

Distributed Morphology takes a decidedly anti-lexicalist position that resembles a lot the "abstract" practice in the 60s, and opposes the lexicalist period of the 70s and 80s (which is namely endorsed by GB syntax, but also by Chomsky 1995a). These broad issues of course are not the primary focus of the book, and no genuine contribution in this area is to be expected. However, they directly impact the way in which morpho-syntax can bear on phonology – namely PF movement is an issue in this context (see §574).

540

2.5. Earlier versions of "No escape from syntax": Selkirk (1984) and Inkelas (1990) A word about the history of the unity of syntax and morphology before moving on. (Rather) on the phonological side, the question whether syntactic and morphological representations and/or computation are distinct or the same is (at least) as old as Lexical Phonology. When Lexical Phonology made its first steps in the early 80s, Selkirk (1984) levelled the unificational argument against this theory (§4320). Just like Distributed Morphology ten years later, it is the number of locations where concatenation is done, as well as the number of means by which it is achieved that lie at the heart of her critique. Selkirk (1984) refuses the division of concatenative activity into a syntactic (postlexical in Lexical Phonology) and a morphological (lexical) sector: she argues that the same mechanism should control the spell-out of both morphological and syntactic structure, the former being only the lower area of the syntactic tree. Selkirk calls this the "syntax-first" model. (216) "The general model we propose, for both words and phrases, is that a fully elaborated syntactic constituent structure (of the word as of the phrase) is mapped, in cyclic fashion, into a complete phonological representation. […] It is claimed that for words, as for phrases, a set of context-free rewriting rules provides the foundations of that description. This is a 'syntax-first' model, where the organization of the grammar into components imposes an ordering of constituent structure formation rules before rules constructing or modifying phonological representation." Selkirk (1984:82) "According to Kiparsky and his predecessors, sentence structure and word structure must be generated in radically different fashion. […] We see no theoretical or empirical advantage to the Siegel-Allen-Kiparsky theory of word formation as a characterization of the syntax of words. We have ar-

454 Chap 12: Distributed Morphology gued instead for characterizing word structure exactly as sentence structure is characterized – by a set of context-free rewriting rules. This means of course that the cycle cannot be eliminated in principle in words, for rewriting rules cannot be ordered and thus cannot be interspread with rules of phonological interpretation. Rather, they generate full-fledged embedded structures, which are available for interpretation only once they have been generated. […] [T]he desired unity in the theory of grammar can be restored. Words and sentences can be generated by the same sorts of rules and can be phonologically interpreted by the same general principles of the syntax-phonology mapping." Selkirk (1984:415, actually the very last sentences of the book)

Table (217) below shows the representations that Selkirk (1984:77) proposes for affix class-based phenomena. The contrast between affix classes is encoded in terms of their respective height in the morphosyntactic tree (root-level affixes are class 1, word-level affixes class 2 items). This is precisely the solution favoured by Distributed Morphology (see §546). (217) Selkirk (1984:77): representation of class 1 vs. class 2 affixes a. bare root b. root + class 1 affix c. root + class 2 affix Word | Root

nation

Word | Root Root | nation-

Word

Af | al

Word | Root | nation-

Af

hood

At that time, the competing Lexical Phonology scenario was stratal: rather than by contrasting morpho-syntactic representations, affix classbased effects were achieved by morpheme-specific mini-grammars which carry out computation at different strata (§150). Later on in the 80s when Prosodic Phonology entered the scene, the Praguian segregation of morphology and syntax that Lexical Phonology implements also became an issue (see §429). Everybody agrees that the Prosodic Hierarchy is the only interface mechanism for units above the word level, that is where words are concatenated (see §§158,787f). At and below the word level, however, there are two candidates for interface management: Lexical Phonology and the units of the Prosodic Hierarchy below the word level (the syllable and feet, eventually moras).

Direct merge and opacity 455

While mainstream Prosodic Phonology (starting with Nespor & Vogel 1986) has tried to sweep the competition under the rug (§435), Inkelas (1990) has drawn attention to the issue. She suggests to extend prosodic constituency to the Lexicon, a move that makes Lexical Phonology an empty shell (§433).

541

3. Direct merge and opacity

542

3.1. Introduction: DM also looks at interpretative effects at LF Let us now look at more narrowly phonological issues and their management in Distributed Morphology. Due to its primary focus on morphosyntax, the body of work regarding phonological phenomena is rather scarce. The discussion below is based mainly on Marvin (2002, 2008), Newell (2005a, 2008), Piggott & Newell (2006) and Lowenstamm (2008). What Distributed Morphology has on top of "phonological" theories of the interface is that it cares for semantic interpretation: while SPE, Lexical Phonology, Halle & Vergnaud (1987a), Kaye (1995) or Prosodic Phonology only try to account for the phonological behaviour of morphologically complex items, Distributed Morphology also aims to cover their semantic behaviour. This is relevant and cannot be circumvented since, as we will see, concomitant and non-arbitrary effects are sometimes encountered on both interpretative sides. An interesting property of Distributed Morphology is that it appears to provide for two distinct origins of phonological opacity: direct merge (i.e. to a root) and Phase Impenetrability. However, only the former is held to produce semantic opacity. As far as I can see, this puzzle is not made explicit in the DM literature. Below it is tackled first by exposing direct merge, the opaque virtue that Distributed Morphology attaches to it and phenomena that fall into this category (§543). In a second step, I examine how Distributed Morphology analyses classical affix class-based phenomena (§557). Both approaches are then crossed: the goal is to gain a complete picture of how opacity and affix class-based phenomena are (or could be) treated in Distributed Morphology in general, and in the frame set by Marantz (2001, 2007) in particular. The general guideline of Marantz' approach is that opacity is the result of structure, rather than of computation.

456 Chap 12: Distributed Morphology 543

3.2. In DM, opacity is due to direct merge

544

3.2.1. Direct merge to a root produces opacity, merge to an xP guarantees transparency Distributed Morphology follows the idea that roots do not possess category information per se. Rather, category membership is the result of the merger of a root with a category-forming affix. This affix may be phonologically void (Marantz 2001). Category-forming affixes come in three flavours: those that, merged with roots, produce verbs (i.e. a vP), those that produce nouns (i.e. an nP), and those that produce adjectives (i.e. an aP). On this analysis, a noun such as dog which lacks overt morphological structure is a root that has been merged with an empty nominalising affix. This is shown under (218a). (218) category-forming affixes, direct and indirect merge a. bare root b. direct merge c. indirect merge aP nP n | ø

aP √ | dog

a | affix

a | affix √ | root

vP v | ø

√ | root

In case a phonologically contentful affix is attached directly to the root as a sister as under (218b) (or rather, strictly speaking, below the first xP), the result may be opaque, i.e. idiosyncratic (idiomatic) and unpredictable (non-compositional). This is, as Marantz (2001:6f) explains, because lexical items have idiosyncratic and unpredictable properties which may be taken into account when they are merged with another piece: the head in question sees the root, which must satisfy its selectional requirements. Opacity, then, is a consequence of the idiosyncrasy of the root, which impacts the negotiation of sound and meaning upon interpretation. In this perspective, the idiosyncrasy of morphologically simplex and morphologically complex items have the same source, i.e. the negotiation of the idiosyncrasy of a root. The only difference is that in the former case (e.g. dog) the root is merged with a functional head (which is phonologi-

Direct merge and opacity 457

cally and semantically empty), while in the latter the head contributes phonological and semantic information. As a result, dog and cat are as much idiomatic expressions as phrasal idioms (e.g. take a break) (Marantz 1997a:206ff). By contrast, (219) "heads attaching outside a little x [like under (218c)] take as complements a structure in which the root meaning (and pronunciation) has already been negotiated. […]Structurally, when a head attaches outside of little x, it sees the features of x locally, not the features, properties, or identity of the root merged with x. So its selectional properties are satisfied by the features of x, drawn from the universal syntactic feature set, not the properties of the root, which are idiosyncratic to the language and to the individual speaker." Marantz (2001:7)

Opacity is thus a property of "lower" attachment to a root before the first negotiation of sound and meaning, i.e. below the first xP. by contrast, a structure where a head attaches "higher up", i.e. outside of an xP, is necessarily transparent because it negotiates with its sister, which is itself an item that has already been spelled out: the idiosyncrasy of the root cannot bite. The reader will have noticed that this scenario supposes that every xP is independently spelled out. This is indeed a basic assumption that is made by Distributed Morphology – an assumption that is far from being selfevident (more on this shortly in §§552,557).

545

3.2.2. An example of direct-merge-created PF and LF opacity: cómparable vs. compárable What Distributed Morphology takes to be the effect of direct vs. indirect merge may be illustrated by the "minimal pair" cómparable vs. compárable (Marvin 2002:75). The former word is both phonologically and semantically opaque, while the latter is transparent on both sides: cómpar-able means "roughly the same" and is stressed on the initial vowel. Were its meaning compositional, it would come out as "X+able", i.e. "to be able to be compared". Also, it does not have regular penultimate stress. By contrast, compár-able really means "to be able to be compared", and has transparent penultimate stress. If non-compositionality is the consequence of direct merge, the contrast at hand identifies as under (220) below.

458 Chap 12: Distributed Morphology (220) representation of (semantic) opacity in DM a. cómparable: opaque b. compárable: transparent aP aP a | -able

a | -able √ | compare

vP v | ø

√ | compare

The following section shows how the opaque effect of direct merge is motivated (beyond the root-visibility that was mentioned in §544). 546

3.2.3. The origin of the idea that direct merge causes opacity: idiosyncrasy is a property of "lower" items in the tree The idea that direct merge is a source of opacity on both interpretative sides is a ground rule in Distributed Morphology. As far as I can see, it is a spinoff of the classical evidence showing that word formation based on roots has different properties when compared to word formation based on existing words. The existence of a linguistically significant and cross-linguistically resident contrast of this kind is undisputed. It is at the origin of all the traditional distinctions that Distributed Morphology rejects: concatenation of morphemes vs. concatenation of words, inflectional vs. derivational morphology, morphology vs. syntax. Marantz (2000, 2001) therefore attempts to find a solution within the single-engine landscape of Distributed Morphology. 137 One thing that di137

Marantz (2000, 2001) appear to be the origin of the idea that direct merge causes idiosyncrasy. Marantz (1997a) already promoted the idea that "lower" items produce idiosyncrasy, while "higher" merge is transparent, but had not yet identified the distinction "below vs. above the first xP": "I haven't yet figured out anything like the complete theory of locality for special meanings" (Marantz 1997a:208). The trouble with Marantz (2000, 2001) is that they are conference handouts. They have been the basis for subsequent (student) work in Distributed Morphology and are constantly referred to in the literature, but they are difficult to

Direct merge and opacity 459

vides root-based and word-based complexes is that the former produce idiosyncrasies (both phonological and semantic, i.e. non-compositionality) and paradigmatic gaps, while the latter do not: they tend to be regular and productive. Common to all traditional two-place approaches is that idiosyncratic root-based formations are somehow "lower" than regular word-based constructions: the root is a smaller, hence more embedded object in comparison to the word. Marantz (2000, 2001, 2007:199ff) endorses this insight, but attempts to provide an implementation in a single engine environment: rather than by the locus of the computation (i.e. the lexicon/morphology vs. syntax), he argues that the difference between the two categories is defined by the relative position of an affix with respect to category-distributing labels (i.e. xPs). Marantz (2007) therefore opposes inner and outer word formation, the former producing words from roots, the latter from existing words. Table (221) below reproduces Marantz' (2007:201) illustration. (221) inner vs. outer morphology outer

y

x inner x

root

Formally, then, Marantz (2007) defines inner morphology as everything that occurs below the first xP, while everything above the first xP is outer morphology; it includes all category changing and derivational morphology. The former (may) produce(s) opacity and paradigmatic gaps, while the latter is phonologically and semantically transparent. Thus the contrast between the opaque cómparable under (220a), as opposed to the transparent compárable under (220b). Marantz (2007) concludes that

access. Whenever possible, the present discussion is therefore based on Arad (2003), Marvin (2008) and Marantz (2007). The latter appears to be the written version of Marantz (2001).

460 Chap 12: Distributed Morphology (222) "the first phase above the root within a word shows properties often associated with lexical word formation, while higher phases triggered, e.g., by category-changing little x heads, behave in accordance with generalizations about syntactic word formation." Marantz (2007:219)

The take-home message, then, is not only the identification of opacity as a phenomenon that is located in the area below the first xP. It is also this: while merge above the first xP guarantees transparency, merge below the first xP does not automatically imply opacity – it only opens the possibility for an opaque interpretation. In other words, an opaque item must have a structure where an affix was merged below the first xP, while a transparent item does not give any indication regarding its structure (it may be the result of direct as much as of indirect merge). The difference between affixes which are merged below vs. above the first xP has become the basis for further work in Distributed Morphology: empirical contrasts of various kinds have been interpreted in these terms, regarding opaque effects at both LF and PF. Work on the LF side includes Arad (2003) and Embick & Marantz (2008), while Marvin (2002, 2008) follows this track at PF. 547

3.3. Marvin's (2002) DM-analysis of condensation vs. compensation

548

3.3.1. SPE's classical take: two different suffixes Marvin (2002) applies Marantz' system to phonological consequences of morphological structure. Specifically, Marvin (2002:66ff) seeks an explanation for words that appear to contradict the famous contrast between condensation and compensation which runs through the phonological literature since Chomsky & Halle (1968:39, 370) (see §97). Let us start by recalling the facts and the original SPE analysis. The noteworthy thing about the pair of words at hand is that the antepenultimate vowel may be reduced to schwa in comp[ə]nsation, but not in cond[ɛ]nsation, which seems to be identical at first sight. SPE relates this fact to contrasting morphological makeups: -ation is just one suffix in condens-ation, but it divides into -ate and -ion in compens-at-ion. This is witnessed by the fact that the verb to condense exists, while *to compense does not: the verb related to compensation is to compensate. Hence the derivations √condense → condens-ation, against √compens → compens-ate → compens-at-ion.

Direct merge and opacity 461

The vowel reduction facts, then, follow from stress: the irreducible vowel in cond[ɛ]nsation was stressed at a previous derivational stage (condénse); its main stress is converted into secondary stress by the suffixation of -ation, but continues to protect its host from being reduced. By contrast, the second vowel in compensation has never borne primary stress at any derivational stage: this word starts out as to cómpens-ate, and the suffix -ion produces compens-át-ion. Being unprotected by stress, the vowel may reduce, and comp[ə]nsátion is produced. 549

3.3.2. The SPE analysis is empirically flawed: Halle & Kenstowicz (1991) admit a lexical conditioning The trouble is that there are words which do allow for vowel reduction despite the fact that the vowel in question bore stress on an earlier cycle. Examples are quite common: consúlt - cons[ə]ltation, consérve cons[ə]rvation, transpórt - transp[ə]rtation, transfórm - transf[ə]rmation, confírm - conf[ə]rmation, infórm - inf[ə]rmation and so on. These thus have the same derivational history as condensátion (cf. condénse), but behave as if they belonged to the comp[ə]nsátion (cf. cómpensate) pattern. This messy situation was only touched on in SPE: Chomsky & Halle (1968:161) mention transfórm - transf[ə]rmation and presént pres[ə]ntation as true exceptions, and leave it at that. 138 Halle & Kenstowicz (1991:460f, 490f) take a closer look at these exceptions; they conclude that the contrast at hand is unpredictable and hence an idiosyncratic property of words. That is, whether a word whose pretonic vowel bore stress at an earlier derivational stage allows for the reduction of this vowel (transpórt - transp[ə]rtation) or not (condénse - cond[ɛ]nsation) is somehow recorded in lexicon. Halle & Kenstowicz' (1991) solution for this lexical recording relies on a device called Stress Copy, which has been introduced by Halle & Vergnaud (1987a:104ff) (see §231). Stress Copy stores stress information of previous cycles in a parallel structure (the metrical grid) before Stress Erasure takes place (Halle & Kenstowicz 1991:490f offer further discussion). On these grounds, Halle & Kenstowicz (1991:460f, 490f) hold up the 138

Curiously enough, though, Chomsky & Halle (1968:116, 161) record information (which admits vowel reduction) as an instantiation of the compensation pattern, i.e. where the second vowel has never received stress in the derivational history. This is obviously wrong: to infórm exists.

462 Chap 12: Distributed Morphology basic SPE analysis whereby the primary stress of a former cycle protects vowels against reduction. That is, Stress Copy stores primary stress of previous cycles, but applies only on the basis of a lexical diacritic that is present in condensátion, but absent from transportátion.

550

3.3.3. *Transport-ate: intermediate derivational forms that happen not to exist as words Marvin (2002:66f) argues for a different scenario. She points out that the contrast between (the well-behaving) cond[ɛ]nsation and (the ill-behaving) transp[ə]rtation is one of opacity: the derivational history of the former may be reconstructed from the surface, while the fact that the reduced vowel of the latter bore stress on a former cycle is hidden away by the reduced transp[ə]rtation. Holding up the exceptionlessness of the original stress-based analysis of SPE, Marvin argues that the reduced vowel in transp[ə]rtation has never been stressed. That is, transportation is not derived from the verb to transpórt. What is it then derived from? The anti-lexicalist mantra of Distributed Morphology allows for an answer that further atomises -ation: transportation is derived from *to tránsport-ate, a form that is not an actual English word. That is, -ation is always composed of -ate and -ion irrespectively of whether the intermediate item [root+ate] happens to exist as an independent word or not (Marvin 2002:65f): √compense → cómpens-ate → compens-át-ion as much as √condense, √transport → *cóndens-ate, *tránsport-ate → condens-át-ion, transport-át-ion.

551

3.3.4. Direct (transportation) vs. indirect (condensation) merge of -ate On this morphological backdrop, the headstone of Marvin's analysis is that stress is only assigned upon phonological computation: roots as such (√condense, √transport) do not bear any stress at all. The corresponding verbs to condénse, to transpórt are derived by the affixation of a categoryforming little v, and stress assignment then concerns the entire vP. This is shown under (223a) below. Critically, then, the root √transport has not experienced stress assignment before affixation of -ate occurs that produces the intermediate form *tránsport-ate: stress assignment would have yielded transpórt, which – counter to fact – would have protected the last vowel against later

Direct merge and opacity 463

reduction. The key question is thus when stress assignment occurs – that is, when exactly the structure is spelled out. Following Marantz (2001), Marvin's (2002) take is that every xP is a phase head (see (223a)).139 The requirement for the intermediate *tránsport-ate not to have gone through transpórt thus comes down to the fact that transport has never been alone in an xP. This, in turn, can only mean that -ate is its sister in *tránsport-ate, as under (223b). The contrast between cond[ɛ]ns-át-ion and transp[ə]rt-át-ion, then, is due to the position of -ate in the tree: in transparent cond[ɛ]ns-át-ion, it attaches outside of the root xP, while in the opaque transp[ə]rt-át-ion it is merged as the sister of the root. This is shown under (223b,c) below. (223) condensation vs. transportation: contrast based on direct vs. indirect merge of -ate a. to transpórt b. transp[ə]rt-át-ion c. cond[ɛ]ns-át-ion nP nP vP v | ø

PF √ | transport

n | -ion

PF vP

v | -ate

n | -ion PF

√ | transport

PF vP

v | -ate

PF vP

v | ø

PF

√ | condense

Marvin (2002) thus offers an analysis which acknowledges the lexically arbitrary character of the contrast between cond[ɛ]ns-át-ion and transp[ə]rt-át-ion, but offers an encoding thereof which is more elaborate than Halle & Kenstowicz' (1991) lexical diacritic. Such a diacritic must still somehow distinguish condensation and transportation, but its effect in grammar is now more precise: the latter root is specified for taking on -ate as a sister, while the former first receives its category from the merger with little v. 139

Following Marvin (2002), this assumption is also endorsed in subsequent work by Barragan & Newell (2003), Newell (2005b), Piggott & Newell (2006) and Piggott (2007).

464 Chap 12: Distributed Morphology 552

3.3.5. All xPs are phase heads, direct merge is the anti-lexicalist way of encoding lexicalist observations This analysis rests on two premises: the idea that all xPs trigger spell-out (on which more shortly), and "abstract" anti-lexicalism that tolerates intermediate unattested forms. It was already mentioned in §539 that unlike the lexicalist 70s and GB-syntax, Distributed Morphology subscribes to the (almost unbounded) anti-lexicalism of the 60s. This frame being set, it accounts for opaque forms by direct merge: directly merged affixes produce items that are one spell-out short as compared to indirectly affixed roots. This lack of spell-out produces idiosyncrasy, i.e. properties of the whole that cannot be predicted from its pieces. It is quite obvious that direct merge is the equivalent of lexicalised forms in an anti-lexicalist environment: in a lexicalist perspective, opacity would be recorded in the lexicon as a property of the whole.

553

3.3.6. Direct merge may, but does not need to produce opacity For the time being, only phonological opacity has been considered. Marvin (2002:68f) bolsters her analysis by the observation that some of the items which are phonologically opaque are also semantically non-compositional. For instance, this is the case of inf[ə]rmátion, which affords reduction and lacks the intermediate form *ínformate, hence falls into the category of transp[ə]rtátion (*tránsportate). Information, however, is not just the nominalised form of the verb to inform: while his relaxation of the conditions can be drawn from he relaxed the conditions, *his information of my friend about the lecture is not a good counterpart of he informed my friend about the lecture. Marvin thus argues that information is semantically and phonologically opaque for the same reason: the direct merge of -ate as under (223b). Crucially, though, this is only a possibility: direct merge may, but does not need to produce opacity (see §546). Alongside with semantically opaque items such as information, there are words like ad[ə]ration (*ádorate) which instantiate the same pattern but relate to the corresponding verb (to adóre) in the same way as the transparent condensátion does to condénse (see (223c)). That direct merge not always produces opacity is also confirmed on the phonological side: comp[ə]nsátion, one half of the original "minimal pair" is derivationally transparent since unlike for transportátion and in-

Direct merge and opacity 465

formátion, its intermediate form to cómpensate exists. Therefore, the possible reduction of the vowel, which has never received stress (there is no verb *to compénse), may be reconstructed from the derivational history. Yet like all other items with a reducible vowel, comp[ə]nsátion must instantiate the direct merge structure under (223b). 554

3.4. Phase Impenetrability à la carte?

555

3.4.1. The strong DM claim that all xPs are spelled out imposes processspecific Phase Impenetrability Distributed Morphology in general and Marvin (2002) in particular subscribe to derivation by phase, and to Phase Impenetrability. Marantz' radical take on spell-out according to which each and every xP (i.e. vP, nP, aP) is independently interpreted (Marantz 2007, Embick & Marantz 2008:6, Embick forth) imposes a very strong freezing effect on strings, much stronger than what is encountered on regular assumptions regarding spell-out, both on the syntactic and on the phonological side. That is, in case only a subset of nodes is spelled out (this is the practice in current syntax), Phase Impenetrability effects are only expected at these particular points in the derivation (which in Chomsky's original take reduce to vP and CP, maybe DP; more on this in §773). If below the word all xPs are spelled out, Phase Impenetrability imposes much stronger freezing effects than what is known from above the word level. Note that the spell-out of all xPs is a strong claim, but not quite as radical as spell-out-as-you-merge that is entertained in syntax by Epstein et al. (1998:46ff) (see §776). That is, there are nodes below the word level which are non-categorising and hence do not fall into the xP set: depending on the analysis, functional heads such as ASP, VoiceP, DegP or AspP appear in the literature (e.g. Embick 2003:148, Embick & Marantz 2008, Alexiadou et al. 2009, Alexiadou & Grimshaw forth). On Marantz' take, then, and more generally in Distributed Morphology, only categorisers (nP, vP, aP) are phase heads: functional heads are not spelled out. This being said, spelling out all xPs is strong enough a claim to drive Marvin (2002) into trouble, which arises on the occasion of English primary stress. That is, the most basic derivations involving stress-shifting class 1 suffixes are indeed supposed not to exist: stress shifts two times on the way from órigin over orígin-al1 to origin-ál1-ity1, but Phase Impenetrability should freeze it on the root-initial vowel since class 1 affixes are

466 Chap 12: Distributed Morphology regular and thus attach to an xP of their own. The spell-out of the root node should thus prohibit any further manipulation of stress in later derivational steps. Marvin (2002:56ff) therefore concludes that primary (but not secondary) stress is an exception to Phase Impenetrability, which does not apply to this particular phenomenon. What that means is that Phase Impenetrability is demoted to an optional condition on spell-out: it applies à la carte to some processes, but not to others. Process-specificity considerably weakens Phase Impenetrability. Recall, however, that process-specific no look-back was already discussed in §§241,302 in an entirely different context, but also based on evidence from English stress. It will also be on the agenda in §§780,796, and a summary is provided in §823.

556

3.4.2. We know that stress is a strange guy anyway Given the eventual need to allow for a process-specific weakening of Phase Impenetrability, it would certainly be nice to find a rationale that divides processes in PIC-obeying and (at least potentially) PIC-violating phenomena. Research along these lines has not really begun as far as I can see, but the track is worth pursuing. In our example, if it can be shown that stress has a certain property that predestines (or forces) it to violate Phase Impenetrability, the predictive character of the device is restored in an interesting, process-specific way. This is what Marvin (2002:56ff) tries to do: she points out that primary stress also violates Phase Impenetrability at a much higher level, i.e. between words. So-called stress clash (also called the Rhythm Rule, Liberman & Prince 1977), describes how the concatenation of words that produces adjacent word-stresses provokes stress shift: the classical example is thirtéen vs. thírteen mén. Stress clash is provoked by a syntactic operation (the concatenation of two words). This operation must thus have access to word stress, which it is able to read and to modify in spite of the fact that the word has been spelled out long before, i.e. at least at the word level. Another thread that runs through Part I also hints at the fact that something is wrong with stress. Recall from §192 that it was stress that violates the SCC-K (that is, the generalisation regarding derived environments). The exceptional character of stress is also noticed by Szpyra 1987:171, note 3).

Direct merge and opacity 467

The violation of the SCC-K by stress (roots can be stressed although they are a underived) pushed phonologists into the distinction between structure-building and structure-changing processes (§§193f, also note 53). The SCC-K, a no look-back device just like Phase Impenetrability, was supposed to apply to the latter (say, a palatalisation), but not to the former. On regular assumptions, stress is absent from the lexicon in languages where words are not specified for stress (such as English). Foot structure, the incarnation of stress, is then built during the phonological derivation, which is the sense in which stress placement is a structure-building operation. It was also mentioned in §237 that Halle & Vergnaud (1987a:86f) report that languages where stress is lexical (such as Vedic and Russian) – and hence stress-shift structure-changing, rather than structure-building – do observe Phase Impenetrability (the SCC-K). Finally, also recall from §315 that in Kaye's system where Phase Impenetrability is responsible for the contrast between párent - parént-al1 vs. párent-hood2, stress violates Phase Impenetrability in just one sub-pattern, i.e. the one that also violates Siegel's (1974) affix ordering (govern-mént2-al1). Based on this record, Rubach (1984:222) ventures to make the generalisation that stress disregards Phase Impenetrability (or the SCC-K) because it is a prosodic (i.e. a suprasegmental) process. Support for this generalisation comes from Bantu, where processes involving tone are able to look back into previously spelled out units (Marlo 2008). More discussion is provided in Part II (§§780,814). In any event, the behaviour of stress appears to be self-contradictory when looked at through the prism of no look-back devices: it does not care for Phase Impenetrability when class 1 affixes shift stress (origin-ál1-ity1, govern-mént2-al1), i.e. whether preceded by a root, a class 1 or a class 2 affix, see §561), but does produce Phase Impenetrability effects upon the interpretation of class 2 affixes (párent-hood2). 557

3.5. All xPs are spelled out vs. selective spell-out plus the PIC

558

3.5.1. Affix class-based phenomena in DM Let us now see how Distributed Morphology goes about regular affix classbased phenomena. The literature does not appear to offer any explicit analysis (which is quite strange). In its absence, the minimal frame is set by

468 Chap 12: Distributed Morphology the commitment of Distributed Morphology to Chomskian phase theory (§304). The consequences are listed under (224) below. (224)

Distributed Morphology subscribes to derivation by phase; this means that a. morpheme-specific mini-phonologies must be dismissed b. selective spell-out is required c. Phase Impenetrability is required

DM subscribes to Phase Impenetrability (see §555), and it was already mentioned that following SPE, Halle & Vergnaud (1987a) and Kaye (1995), the theory uses only one computational system for strings of whatever morphological quality. Selective spell-out, however, is problematic: we have seen in §547 that the baseline of Marantz' system is to require spell-out at every xP. It was shown in §555 that this does not amount to Epstein et al.'s (1998) spell-out-as-you-merge (§776) where literally every node is a phase head since functional heads (VoiceP etc.) are not spelled out. Distributed Morphology can thus claim selective spell-out (not all nodes are interpreted), but the kind of selectiveness is quite different from what we know both from phonology (where phasehood is a lexical property of pieces: class 1 vs. class 2) and syntax (where phasehood is independent from the distinction between functional vs. non-functional heads). Recall that in phonology, it is precisely the selectiveness of the interpretation-triggering nodes (class 1 vs. class 2) that allow Halle & Vergnaud (1987a) to treat affix classes with only one computational system (§225). As will be shown below, it is the peculiar kind of selective spell-out that DM imposes which makes the analysis of affix class-based phenomena impossible in this theory. 559

3.5.2. Underapplication cannot be done when all xPs are spelled out A mechanical consequence of Marantz' system that has only one computational system but spells out every xP is its inability to produce underapplication. Recall that this is what affix class-based phenomena are about, and that underapplication can either be achieved by morpheme-specific miniphonologies (§150) or by selective spell-out (Halle & Vergnaud, see §228), which in Kaye's system is supplemented with Phase Impenetrability (see §279). But even in Lexical Phonology and subsequent implementations of morpheme-specific mini-phonologies, interpretation does not occur at every morpheme boundary, which is about what "spell-out at every xP" amounts

Direct merge and opacity 469

to. The closest match of Marantz' system is SPE where all nodes are indeed cyclic boundaries, that is subject to interpretation (see §103). Distributed Morphology is thus facing a serious problem: either Marantz' requirement that every xP is spelled out is wrong, or affix class-based phenomena cannot be done. Note that the spell-out of every xP is not only a bewildering claim when looking at phonological effects. It also departs from current syntactic phase theory. Chomsky's (2000a) original take is that there are only two points in the derivation of a sentence, vP and CP (maybe DP), where spellout occurs. Since then, the evolution is to grant phasehood to more and more nodes that dominate smaller and smaller chunks (§775 describes this atomisation). As far as I can see, though, there is no parallel in syntax to the rationale that is based on the distinction between functional heads (which do not provoke spell-out) and non-functional heads (xPs, which are interpretation-triggering) that is practised in Distributed Morphology. We are thus left with another property of Distributed Morphology which adds to the doubts that the theory stands up to its ambition to unify morphology and syntax (§538): the distribution of phasehood below and above the word level is not the same. This looks much like two different worlds – exactly what Distributed Morphology has set out to show to be wrong. In order to be able to do affix class-based opacity effects, and to be in harmony with phase theory (which requires selective spell-out and a critical role played by Phase Impenetrability), it is argued below that a correct theory must implement selective spell-out also below the word level. That is, the idea that all xPs are phase heads is on the wrong track. But, as we will see, the direct merge analysis may do a good job in a specific sub-area of opacity effects, i.e. where the kind of opacity that is encountered is not predictable.

560

3.5.3. Opposite ways to go: distinct representations (DM) vs. distinct computation (Kaye, Halle & Vergnaud) Distributed Morphology looks at LF and PF effects through the prism of opacity. In this perspective, párent-hood is opaque because its stress pattern cannot be reconstructed from the surface, while parént-al has surface-true, hence transparent penultimate stress. Recall from §§272f that this is also Kaye's way of approaching affix class-based phenomena.

470 Chap 12: Distributed Morphology As a matter of fact, the two systems produce very different and incompatible analyses: while according to DM the root and the affix of párent-hood must be sisters (because they produce opacity at PF), they must not be sisters on Kaye's analysis. This is shown under (225) below. Note that in Halle & Vergnaud's (1987a) system, -hood cannot be the sister of parent either: the relevant structure is like Kaye's under (225b), except that the higher nP (i.e. the word level) is not spelled out. (225) opacity-based analyses of affix class-based phenomena opaque párent-hood c. transparent parént-al a. DM b. Kaye (1995) (both DM and Kaye) nP PF aP PF nP n | hood

PF √ | parent

n | hood

nP n | ø

PF √ | parent

n | al

nP n | ø

√ | parent

All three approaches agree on the structure of transparent parént-al, which is as under (225c). Critical for Kaye and Halle & Vergnaud is that (225b) and (225c), whose arboreal structure is identical, are not spelled out in the same way: while on Kaye's account both nPs are spelled out in the former (on Halle & Vergnaud's: only the lower nP), the lower nP must not be an interpretational unit in the latter (Halle & Vergnaud: the higher nP must not be spelled out). This is what selective spell-out is about. In conclusion, then, it appears that the two analyses are opposed according to the fundamental dichotomy along which this book is organised: representation vs. procedural computation. This is shown under (226) below. (226) DM (structure) vs. selective spell-out (computation) structure a. DM variable: affixes are sisters of the root vs. sit in their own xP b. Kaye, constant: Halle & Verevery affix sits in its own ganud xP

computation constant: spell-out at every xP variable: only a subset of xPs is spelled out

Direct merge and opacity 471

One general argument in favour of (226b) has already been made: Distributed Morphology is committed to selective spell-out (and hence to Phase Impenetrability) via phase theory, to which it subscribes. Below I take advantage of the fact that the two competing solutions have not been developed on the grounds of the same set of data: the question is explored whether the respective phenomena can be analysed in terms of the other approach. That is, whether one theory can be reduced to the other.

561

3.5.4. A direct merge analysis for affix class-based phenomena is not viable Let us first have a closer look at the consequences of analysing the parent parental pattern in terms of direct merge. There is good reason to believe that this perspective cannot be correct. For one thing, it can hardly be accidental that the opaque stress pattern of párent-hood is the same as the one that is produced by regular stress assignment when párent is pronounced in isolation. Direct merge is only said to be a source of opacity – what this opacity will look like remains unspecified. What we see, however, is not just any opaque pattern: precisely the one that is produced by a two-step spell-out appears. The difference between predictable (párent-hood) and unpredictable (cómpar-able) opacity is discussed in §§562,568 below: while the direct merge analysis is not suited for the former, it may be a way to go about the latter. Another strong argument against the direct merge analysis comes from the fact that there are cases where other affixes may intervene between the root and opacity-producing class 2 affixes (such as -hood). In góvern-ment2-hood2, and univérs-al1-ness2, -hood and -ness behave as expected, i.e. they do not shift stress (*governmént-hood, *universál-ness) and hence create an opaque non-penultimate pattern. However, due to the presence of the intervening -ment- and -al-, they cannot be the sister of the root. Therefore the opacity that they are responsible for cannot be due to direct merge. The conclusion, then, is as follows: if there is only one affix -hood, and if the opaque effect has always the same cause, -hood is never the sister of the root, not any more in párent-hood than in góvern-ment-hood.

472 Chap 12: Distributed Morphology 562

3.5.5. Analysing cómparable vs. compárable with selective spell-out Let us now see whether, conversely, the contrast between compárable and cómparable may be analysed in terms of selective spell-out. Table (227) below shows how opacity is produced procedurally on the backdrop of identical structure (and two distinct -able suffixes, one class 1, the other class 2). (227) analysis of cómparable vs. compárable with selective spell-out and Phase Impenetrability a. compár-able1 b. cómpar-able2 aP PF aP PF n | able1

vP v | ø

√ | compare

n | able2

vP v | ø

PF √ | compare

Like all other class 2 affixes, -able2 triggers the spell-out of its sister under (227b) (in Kaye's system), but -able1 under (227a) does not (in addition, both words are spelled out at the word level). The pronunciation and meaning of the root is thus negotiated separately under (227b), but not under (227a). Phase Impenetrability will then prevent both PF and LF properties to be reassessed when the word as a whole is interpreted. What this analysis does not tell us, however, is why the root under (227b) receives stress on the first vowel, i.e. cómpare: regular stress assignment would have produced compáre, as is witnessed by the verb to compáre. The same goes for the non-compositional meaning: nothing tells us why the spell-out in isolation of compare produces the meaning that is observed, rather than some other non-transparent meaning. The difference with respect to PF opacity, though is obvious: while LF opacity is due to some lexical property of the root, stress assignment should be the result of regular computation. The fact that it is not may point towards a lexical recording of stress for this kind of root – an option that deserves serious consideration.

Direct merge and opacity 473 563

3.5.6. Conclusion In sum, Marantz' (2001, 2007) idea that opacity is due to a specific structural configuration (direct merge) faces serious trouble: it is unable to organise underapplication, i.e. to account for affix class-based phenomena. Marantz' (2001, 2007) system also makes a prediction to the end that no additional affix can be merged between the root of an opaque form and the affix that causes opacity. That is, items such as [root+A+B] where B induces opacity are not supposed to exist since opacity must stem from the fact that B is merged under the xP of the root. This prediction is counterfactual at PF (in góvern-ment2-hood2, univérs-al1-ness2); whether it is also wrong for LF opacity remains to be seen. Beyond empirical issues, Marantz leaves traditional grounds where opacity is the result of a distinction in procedural activity (selective spellout of identical structure and Phase Impenetrability). On his analysis, opacity roots in a structural distinction (direct vs. indirect merge) that uses a uniform computational mechanism (spell-out at every xP). This take installs a mechanism below the word level that is at variance with current syntactic phase theory where spell-out is selective (even among categorisers), and where Phase Impenetrability accounts for opacity. On the other hand, the perspective that was developed in phonology (by Halle & Vergnaud 1987a and Kaye 1995) allows for a uniform analysis of phenomena below and above the word level, which are based on selective spell-out and Phase Impenetrability. What remains unexplained, however, is the irregular PF opacity of cómparable, which contrasts with the predictable opacity of párent-hood. In order to approach this issue, the pages below examine in which way PF and LF opacity can be crossed.

564

3.6. Articulation of semantic and phonological opacity

565

3.6.1. All logical possibilities occur A question that has been addressed on various occasions on the foregoing pages is the concomitance of semantic and phonological opacity. We have seen that they may indeed go hand in hand (cómparable vs. compárable, §544). There are also cases, however, where PF opacity is produced in absence of LF opacity: párenthood instantiates this pattern. And of course, transparency on both sides may also occur: this is the case of paréntal.

474 Chap 12: Distributed Morphology Table (228) below shows the four logically possible configurations. (228)

possible distribution of phonological and semantic opacity phonological semantic example compare with opacity opacity a. + + cómpar-able compár-able b. + – párent-hood párent c. – + twink-ling twinkel-ing d. – – parént-al párent

(228a,b,d) have been discussed. Let us thus look at the missing configuration (228c). The example comes from SPE (Chomsky & Halle 1968:86) and is discussed by Marvin (2002:36ff). Twinkeling [twɪŋkəlɪŋ] (which may also be pronounced with a syllabic liquid, in which case there is no schwa: [twɪŋklÿɪŋ]) means "the act of twinkling" and is phonologically irregular since it bears a schwa in an open syllable (or, alternatively, because a syllabic liquid appears before a vowel). By contrast, twinkling [twɪŋklɪŋ] means "a short moment" and does not bear any schwa (or syllabic liquid). On the SPE analysis that Marvin (2002:34) adopts (see also Piggott & Newell 2006:8ff, Newell 2008:20f), the verb in isolation twinkle [twɪŋkəl] possesses a schwa as the result of regular schwa epenthesis into word-final obstruent-liquid clusters: the underlying form is /twɪŋkl/ (or, on the alternative pronunciation, has a syllabic liquid due to a rule that produces liquid syllabicity in the same context). Hence the presence of the schwa (or of a syllabic liquid) in twinkeling [twɪŋkəlɪŋ] "the act of twinkling" is explained if the phase structure [[twinkl] ing] is assumed: the root is spelled out in isolation before -ing is merged. Schwa is thus regularly inserted into the final obstruent-liquid cluster (or its liquid made syllabic) upon the interpretation of /twinkl/. On later passes through phonology, Phase Impenetrability prevents epenthesis (or liquid syllabicity) to be undone. Twinkeling [twɪŋkəlɪŋ] is thus in instance of (228b): like párenthood, it shows PF, but no LF opacity. On the other hand, twinkling [twɪŋklɪŋ] has no schwa (or syllabic liquid) and means "a short moment". Having no schwa in an open syllable (or a non-syllabic liquid before a vowel) is a phonologically transparent behaviour, which however is accompanied by semantic opacity: "a short moment" is nothing that can be deduced compositionally from to twinkle and -ing. Twinkling thus illustrates the missing pattern (228c).

Direct merge and opacity 475 566

3.6.2. Neither the PIC nor direct merge can account for all patterns Given the selective spell-out perspective, phonological transparency enforces the structure [twinkl-ing] (one single phase): were the root spelled out by itself, schwa would be inserted (or the liquid would be syllabic). The question is thus what the origin of LF opacity could be in (228c) cases. Phase Impenetrability does not qualify because there is no inner phase to be preserved. Looking at Marantz' (2001, 2007) theory, another candidate is direct merge: LF opacity could be produced by the sisterhood of the root and the affix – recall from §553 that direct merge may, but does not need to produce opacity. Hence there is nothing wrong with the two pieces of the transparent twinkling [twɪŋklɪŋ] "a short moment" being sisters and producing LF, but no PF opacity. If LF and PF opacity as a result of root-affix sisterhood is freely distributed, the reverse situation, i.e. where direct merge produces phonological, but no semantic opacity, is also expected to exist. The pattern at hand, (228b), however, is already covered by Phase Impenetrability, and an analysis of párenthood along direct merge appears to be impossible (see §561). What would be needed, then, is a (228b) case analogous to transparent twinkling "a short moment", i.e. where a Phase Impenetrability analysis is ruled out on independent grounds. It thus seems that neither selective spell-out nor direct merge is able to cover the four-way pattern under (228) alone: Phase Impenetrability cannot explain the semantic opacity of (228c) twinkling [twɪŋklɪŋ] "a short moment", while direct merge is unable to account for (228b) párenthood (and more generally for regular affix class-based phenomena).

567

3.6.3. Independent LF and PF phases? The conclusion of the previous section is only relevant on the tacit assumption that LF and PF phases are always concomitant: when a given node is spelled out, its content is sent to and interpreted at both LF and PF. That is, "asymmetric" spell-out whereby the content of a node is sent to one interpretative module, but not to the other, is not an option. It is obvious that phase theory would be significantly weakened if a given node could be independently spelled out at LF and PF. Chomsky (2004) is explicit on this.

476 Chap 12: Distributed Morphology (229) "Assume that all three components are cyclic. […] In the worst case, the three cycles are independent; the best case is that there is a single cycle only. Assume that to be true. Then Φ [the phonological component] and Σ [the semantic component] apply to units constructed by NS [narrow syntax], and the three components of the derivation of proceed cyclically in parallel. L [language] contains operations that transfer each unit to Φ and Σ. In the best case, these apply at the same stage of the cycle. […] In this conception there is no LF: rather, the computation maps LA [lexical array] to piece-by-piece cyclically." Chomsky (2004:107)

Responding to empirical pressure of various kinds, though, independent LF and PF spell-out is proposed or considered by, among others, Marušič (2005), Marušič & Žaucer (2006), Felser (2004), Matushansky (2005), den Dikken (2007), Megerdoomian (2003) and Caha & Scheer (2008). There is no doubt that simultaneous phases are to be preferred. The question is whether they resist empirical pressure. In case they do not, phase theory loses a good deal of its appeal and predictive power, but the challenge raised by the free distribution of LF and PF opacity under (228) evaporates. Asymmetric spell-out would allow for [twinkl-ing]PF to be spelled out at PF with no semantic implications, while LF could be said to interpret the structure [[twinkl]LF ing]LF with no consequences at PF. The global spell-out would thus follow the structure [[twinkl]LF ing]LF,PF. 568

3.7. Conclusion: predictable vs. unpredictable opacity and associated analyses It was pointed out in §562 that the predictability of the stress pattern of párent-hood is good reason not to believe in a direct merge scenario for this kind of construction. While direct merge merely says "the item is irregular/opaque" without giving any indication which way opacity goes, the cyclic analysis makes a prediction as to where exactly irregular stress falls (i.e., on the vowel that receives main stress on the inner cycle: [[párent] hood]). The selective spell-out analysis is therefore unable to account for the opacity of cómpar-able: [compare] alone would receive stress on the first vowel where it is interpreted in isolation, and hence come out as compár-able. It thus seems that there are two ways of being irregular/opaque: one that is predicted by Phase Impenetrability on the grounds of regular phono-

Anti-lexicalism is an orientation that allows for lexicalist analyses 477

logical computation that is done on an inner cycle, and another one where the pronunciation is entirely unpredictable. Interestingly, LF opacity does not appear to provide for the same distinction: LF opacity is always unpredictable and idiosyncratic since it follows from lexical properties of the root, rather than from computation. The two kinds of opacity are paralleled with two mutually exclusive analyses, none of which, however, can account for all patterns of freely distributed LF and PF opacity (§565). One possibility is thus that selective spell-out is the correct analysis for predictable opacity (párent-hood), while unpredictable opacity (cómpar-able) is due to direct merge.

569

4. Anti-lexicalism is an orientation that allows for lexicalist analyses

570

4.1. The phonological side of the (anti-)lexicalist issue The headstones of Distributed Morphology are the single engine approach, late insertion, the distribution of list-type information over three distinct lexica and anti-lexicalism. Ongoing debate regarding the latter issue from the morpho-syntactic point of view was already touched on in §§539,552. While the 60s and SPE were anti-lexicalist, Chomsky (1970) engaged the field into the lexicalist alternative, which was dominant in the times of GB. In the 90s, work in minimalist syntax encompasses both lexicalist and strongly anti-lexicalist approaches; DM stands on the latter side. Anti-lexicalim is also a resident issue in phonology: in the 70s, it lied at the heart of the abstractness debate (see §126, Vol.1:§305), which has remained inconclusive. The question is whether there is any phonological computation involved in certain morphologically complex words at all. Or rather, in what the linguist is able to identify as a morphologically complex item. The classical example is the pair electri[k] - electri[s]-ity (see §126, note 28): the linguist identifies two morphemes – but does grammar make the same analysis? Electricity as a whole could as well be a single lexical entry, just like dog and table. In this case, there is no phonological activity at all when electricity is uttered: no underlying /k/ is palatalised by the following front vowel; the [s] is already /s/ at the underlying level. There is no concatenative, i.e. morphological activity either: electricity is made of one single piece, rather than of two pieces. The question is thus whether the pieces that a linguist is able to identify are really the pieces that are used in grammatical computation. Another way to look at the same problem is along the trade-off between the size of

478 Chap 12: Distributed Morphology the lexicon and the amount of phonological (and morphological computation): the anti-lexicalist stance is that the smaller the lexicon, the better the theory. That is, the more words are cut into pieces, the smaller the number of lexical entries, and the bigger the number of phonological processes that are needed in order to derive the correct surface form. This is what Jonathan Kaye calls the central dogma, which produces decidedly outlandish results when systematically applied: Lightner (1978:18f, 1981, 1985) derives sweet and hedonistic, queen and gynaecology, thirst and torrid, eye and ocular and so forth from the same underlying form, which supposes modern English speakers who perform Grimm's Law, Verner's Law and the pre-Greek s > h shift (see §126, Vol.1:§333). This is absurd, and nobody today will endorse such a scenario. Another argument against the everything-is-phonology attitude of SPE is the nature of phonological processes that the phonology needs to endorse in order to bring underlying forms to the surface: many are "unnatural" (in a sense that is hard to define) and transform phonology into an unrestricted pool of operations where anything can be transformed into anything. Finally, the following critique has been raised at least since Kiparsky (1968-73): anti-lexicalist practice produces a blooming rule apparatus which they need to suppose in fact describes the events of diachronic evolution that have occurred hundreds or thousands of years ago. There is no good reason to assume that a present-day native applies rules that were active centuries ago, or that these rules remain unchanged over such a long period of time. On a diachronic backdrop (Vennemann's 1976 [1971] "in the name of phonological realism"), the most radical opposition against the everythingis-phonology attitude of SPE has grown into Natural Generative Phonology (see §§127f). Government Phonology also ranges at the far end of the antianti-lexicalist scale (see §276). To date the debate is inconclusive in the sense that nobody is able to provide a set of formal criteria (what was called the evaluation measure or evaluation metrics in the 70s, e.g. Kiparsky 1974) that affords to decide, for a given case, whether an item that looks morphologically complex is really considered as such by the grammatical system. All modern theories have somehow swung into a midfield position between the radical anti-lexicalism of SPE and the extreme position of Natural Generative Phonology. One may conclude that it is not good or bad per se to have a big or a small lexicon, or to do little or a lot of computation. Arguments must be made on a case-by-case basis. The goal of the discussion below is to determine the position of Distributed Morphology in this spectrum. The anti-lexicalist a priori is obvi-

Anti-lexicalism is an orientation that allows for lexicalist analyses 479

ous, but it remains to be seen how far DM is prepared to go, and which are the phonological consequences of this position. Recall that Kaye's (1995) take on the issue, i.e. lexical access, was discussed at length in §351. 571

4.2. A third player: allomorphy, suppletion Two words that are felt as being related (phonetically, semantically, diachronically or paradigmatically) may thus represent one single or two independent lexical entries (vocabulary items in the terminology of Distributed Morphology). In the latter case, items which the linguist identifies as morphologically complex may also be just one single vocabulary item. This situation is depicted under (230) below. (230)

nP n | ø

nP √ | dog

n | ø

√ | electricity

Aside from these possibilities, there is a third option: allomorphy. On this count, electricity represents two pieces, which however are not concatenated by the regular syntactic gluing operation (Merge). Rather, the selection of the stem electri[k] or electri[s] is made according to the grammatical context, here "in presence of -ity, choose electri[s]". The same mechanism controls the distribution of other allomorphs such as go and went, good and better and so forth (suppletion).140 In DM, allomorphy/suppletion is a case of Readjustment (see Embick forth, the term is directly borrowed from SPE). In the diagram under (215) which shows the general architecture of Distributed Morphology, Readjustment is done upon Late Insertion, i.e. when vocabulary items compete for replacing morpho-syntactic features of terminal nodes. The overall situation is thus as under (231) below.141 140

141

Lowenstamm (2008) argues that the number of true cases of allomorphy and suppletion can be cut down when a more fine-grained analysis is able to show that different exponents express different (rather than the same) terminals (multiple exponence). For the sake of completeness, analogy needs to be added to the list. This way of relating two words is irrelevant for the present purpose.

480 Chap 12: Distributed Morphology (231)

possible treatments for words that are phonetically, semantically, diachronically or paradigmatically related a. phonology one single lexical entry, the difference is the result of phonological computation b. allomorphy, suppletion one single lexical entry with several allomorphs. A specific suppletive mechanism (Readjustment) decides which allomorph is chosen in specific grammatical contexts. No phonological computation is involved at all. c. two independent lexical entries one single lexical entry, no concatenative (morphological, suppletive) or phonological activity.

Let us now see how these three options are considered by practitioners of Distributed Morphology. 572

4.3. Distributed Morphology relies on the usual "phonological similarity" While Distributed Morphology has a clear bias in favour of (231a), it allows for the implementation of all options: as far as I can see, no claim is made as to what kind of phenomenon can, should or must instantiate which mechanism. If this is true, then Distributed Morphology behaves like all other theories since the 80s: decisions need to be made on a case-by-case basis, and they often enough follow the intuition of the analyst. That is, notions such as "plausible phonological relatedness" may be critical. Talking about two manifestations of the Dutch nominal plural, Harley & Noyer (1999) make the following statement. (232) "Since -en and -s are not plausibly related phonologically, they must constitute two Vocabulary Items in competition. Morphophonological allomorphy occurs where a single Vocabulary Item has various phonologically similar underlying forms, but where the similarity is not such that phonology can be directly responsible for the variation. For example, destroy and destructrepresent stem allomorphs of a single Vocabulary Item; the latter allomorph occurs in the nominalization context. DM hypothesizes that in such cases there is a single basic allomorph, and the others are derived from it by a rule of Readjustment. The Readjustment in this case replaces the rime of the final syllable of destroy with -uct. (Alternatively such allomorphs might both be listed in the Vocabulary and be related by 'morpholexical relations' in the sense of Lieber 1981)." Harley & Noyer (1999:5, emph. in original)

Anti-lexicalism is an orientation that allows for lexicalist analyses 481

Needless to say, "phonological similarity" is nothing that may be quantified or evaluated by some formal criterion. It leaves us with just our intuitive judgement. In an ideal world, one might expect a theory of suppletion to be able to tell us what possible suppletive activity is, and what is not. Marantz (1997b) has tried to advance in this direction. But significantly, Harley & Noyer, for the case of destroy - destruction, remain undecided between the suppletive solution (which they first advertise) and one that relies on two independent lexical entries (which they put into brackets).

573

4.4. There is no semantic effect if cómparable and compárable are made of two independent Vocabulary Items In which way is this debate relevant for the issues that have been discussed thus far? If a phonological or an allomorphic computation cannot be taken for granted, a pair such as cómparable - compárable could as well represent two separate lexical entries. In this case, there is no semantic effect due to computation or contrasting arboreal structure at all since there is no computation and no complex structure of any kind in the first place. On this take, the LF difference observed is not any different from the one that exists between dog and cat: LF opacity is always unpredictable (unlike PF opacity, see §568). In this lexicalist perspective, the opaque item, cómparable, was lexicalised at some point; its meaning and stress pattern have then started to drift away from the original regular and compositional interpretation. This option is depicted under (233) below. (233) two independent Vocabulary Items a. cómpar-able b. compár-able "roughly the same" "to be able to be compared" aP aP a | ø

n | able √ | cómparable

vP v | ø

√ | compare

482 Chap 12: Distributed Morphology Alternatively, the pair could also be the result of allomorphic activity, a scenario which supposes that stress is part of the lexical properties of the allomorphs: cómpare would be selected by class 2 -able, while compáre would be adjoined to class 1 -able. This is shown under (234a). Finally, the fully computational solution (along Marantz' direct merge analysis) is shown under (234b). (234) allomorphic and phonological scenarios for cómparable and compárable a. allomorphy b. fully computational solution root + class 1 affix root + class 2 affix aP aP a | ø

aP √ | X

X = cómpare with class 1 -able, compáre with class 2 -able.

a | able1

√ | compare

"roughly the same"

a | able2

vP x | ø

√ | compare

"to be able to be compared"

In case there are two independent lexical entries, or if the allomorphic scenario is correct, the entire discussion regarding opacity and direct merge (§541) may turn out to be vain: it makes only sense if it is established on independent grounds, and beforehand, that the opaque items in question are really the result of computation. It thus looks like we are not any more advanced regarding this issue than generative linguists were in the 70s: there is simply no way to measure whether a given word is fully lexicalised or not, whether it is the result of allomorphic or of regular morphological activity. The overt anti-lexicalist orientation that Distributed Morphology promotes offensively does not make the situation any different: anti-lexicalism airs a general sympathy for everything-is-computation; it is unable, however, to make principled decisions: allomorphic/suppletive and lexicalist solutions are always possible.

PF movement 483 574

5. PF movement

575

5.1. Syntactic motivation Distributed Morphology has also introduced PF movement into the interface debate, a tool that has no precedent as far as I can see. The idea is that morpho-syntactically defined items can continue to move along morphosyntactic structure after vocabulary insertion. There is thus a point in the derivation where the morpho-syntactic tree is still available, but terminal elements are phonological, rather than syntactic. Movement, then, is as much feature-driven as in syntax – only are the features phonological. That is, displacement of morpho-syntactically defined chunks can be triggered by their phonological properties. Motivation for PF movement comes from both syntactic and phonological quarters. On the former side, a number of phenomena resist regular analysis: they do not appear to have any syntactic or semantic motivation, and they may violate otherwise well established generalisations. Typical cases in point are extrapositions, which Chomsky (2001, 2005) argues are not due to core syntactic movement operations; rather, they take place in the phonological component. Some examples appear under (235) below (from Göbbel 2007). (235) extraposition a. [A new book __] appeared last year about Turner. b. He sold [a painting __] at Sotheby's by Turner. c. Okay, you saw a picture yesterday, but by just whom did you see a picture yesterday OF?

Displaced constituents (e.g. by scrambling or topicalisation) are normally "frozen" for further extraction, but (235c) shows that extraposed constituents can be further extracted from. Göbbel (2007) argues that PF movement offers a solution: the preposition is dislocated after whmovement has applied. The two reasons mentioned (absence of syntactic motivation, violation of syntactic regularities) appear to be the basis of PF movement on the syntactic side. Relevant literature that follows this line of attack includes Chomsky (2001:21ff), Embick & Noyer (2001), Sauerland (1998) and Sauerland & Elbourne (2002). PF movement may thus be considered an instrument of the minimalist programme: classically syntactic labour is outsourced in order to be understood as an interface condition. (Narrow)

484 Chap 12: Distributed Morphology syntax is cleaned up at the expense of a big, manifold (and dirty) interface/PF (see §727). On the other hand, there are also voices that argue against a PFmotivation (or PF-management) in the analysis of particular displacement phenomena that are PF-outsourced in the literature mentioned. Julien (2006) for example shows that the displacement operations that Chomsky (2001) takes to be phonological in nature may be treated in syntax proper.

576

5.2. Phonological motivation

577

5.2.1. Cases where phonology impacts syntax On the phonological side, PF movement builds on phenomena that challenge phonology-free syntax. Recall from §412 that the principle which holds that phonology never influences morpho-syntactic operations (Zwicky & Pullum 1986a,b) probably needs to be restricted to melody, i.e. to phonological properties that occur below the skeleton. Properties above the skeleton such as intonation, focus, minimal word constraints and the like appear to be able to bear on the workings of morpho-syntax. For example, on Truckenbrodt's (1995) analysis, extraposition of an intonational phrase is blocked by an intervening intonational phrase in cases such as under (236) below. (236) extraposition blocked by an intervening intonational phrase *A book delighted Mary by Chomsky. (A bóok __i)φ (delighted Máry)φ by Chomskyi.

Following Truckenbrodt, locality conditions similar to the ones known from syntax also hold on the phonological side, that is when constituents are defined in terms of the Prosodic Hierarchy. This is the kind of evidence which suggests that syntactic movement may be sensitive to phonological constraints.142 Its treatment requires the 142

Another typical representative of this category are displacement phenomena that appear to be sensitive to the "heaviness" of the displaced object, which is commonly calculated according to the amount of phonological material that it is made of (a long DP is heavier than a short DP). Heavy NP shift is the typical example. While the grammaticality of ?He sold at Sotheby's (a painting) is dubious, increasing the size of the right-displaced item significantly improves the result: He sold at Sotheby's (a painting by Turner). Selkirk (2001) proposes a

PF movement 485

simultaneous availability of morpho-syntactic structure and phonological content. There must thus be a derivational stage where spell-out has already taken place (i.e. with vocabulary items inserted), but where morphosyntactic structure is still available. This ties in with the syntactic motivation for PF movement that was discussed: displacement in the kind of intermundia described does not obey the same set of constraints as in a purely syntactic environment. This is killing two birds with one stone: displacements that are embarrassing for syntactic theory are expelled to PF, and this is also the derivational stage where phonologically sensitive movement operations take place.

578

5.2.2. Phonology-internal triggers: Piggott & Newell on Ojibwa The cases discussed in the previous section represent classical instances of phonologically conditioned syntactic phenomena: the phonological conditioning is located at or above the skeleton. In contrast to this pattern, Piggott & Newell (2006) and Newell (2008:108) propose a PF movement analysis in absence of any syntactic phenomenon: the trigger is phonological, and the responding PF movement is syntactically undetectable. Also, the phonological phenomenon at hand is of melodic nature: an anti-hiatus reaction. The PF movement proposed hinges on an independent piece of analysis regarding the interpretation of hiatus avoidance in Ojibwa (Eastern Algonquian). Table (237) below shows that this language sometimes does, but at other times does not react when a hiatus comes up.

phonological solution that is driven by the principle of Weight Increase: a sequence of prosodic constituents is ordered such that their weight (i.e. their length) progressively increases (i.e. from left to right). The "phonological heaviness" of morpho-syntactic constituents and their alleged impact on syntax was discussed in §420. Being "heavy" is an observational category which, since Nespor & Vogel (1986), is too quickly and too easily taken as the critical property for the phenomenon observed: longer constituents are not only longer ("heavier") – they also have more (morpho-syntactic) structure. The phenomena observed may thus as well be due to this greater complexity, i.e. something that has nothing to do with phonology. This issue is discussed at greater length in §421.

486 Chap 12: Distributed Morphology (237) (non-)reaction of Ojibwa against hiatus a. /namee- im/ → nameem sturgeon poss b. /gii- aagam- osee/ → giiaagamosee past-snowshoe-walk c. /ni- aagam-osee/ → nidaagamosee 1sg- snowshoe-walk

"sturgeon (possessive)" "he walked in snowshoes" "I walk in snowshoes"

Under (237a), an anti-hiatus reaction causes the loss of a vowel; under (237b), however, the hiatus survives. On the analysis of Piggott & Newell (2006), the difference lies in phase structure: in the former case the hiatus occurs within the same phase ([namee-im]nP), while in the latter it straddles a phase boundary ([gii [aagam-osee]vP]CP). No reaction is recorded in the latter case because the vP-initial vowel was already spelled out in the inner vP phase: Phase Impenetrability prevents its reassessment when the outer cycle is interpreted. Based on this analysis, the hiatus should also survive under (237c), which however it does not. Piggott & Newell show that there is syntactic evidence to the end that both personal and tense markers occur outside of the vP/nP: they are spelled out at the CP phase. Hence the hiatus should appear on the surface in [ni [aagam-osee]vP]CP as much as in [gii [aagam-osee]vP]CP. Another issue is the way in which the hiatus is resolved: while the regular means appears to be vowel deletion, an epenthetic consonant is inserted under (237c). In this situation, Piggott & Newell (2006:23) (also Newell 2008:130f) observe that whether or not the hiatus survives in identical syntactic configurations depends on a phonological property of the item in the outer phase: morphemes with a long vowel appear in hiatus, while elements with a short vowel provoke consonant epenthesis. That is, short ni- "1sg" and ga- "future indefinite" trigger epenthesis, while long gii- "past" and wii- "future non-indefinite" do not react on a following vowel. On Piggott & Newell's (2006:21) analysis, this is due to the fact that degenerate feet (i.e. which are made of only one syllable) are only allowed at the right edge of a phase. Everywhere else, the size of the phonological material that is computed at its insertion site must be big enough to form a regular bisyllabic foot. This is indeed the case for the bimoraic gii- in [gii [aagam-osee]vP]CP. 143 Monomoraic ni- in [ni [aagam-osee]vP]CP, how143

It is unclear to me on the occasion of which phonological computation the size of gii-/ni-, and hence their viability as a bimoraic foot, is evaluated: the vP phase of [ni [a-gam-osee]vP]CP is first spelled out, and at the CP phase the pho-

PF movement 487

ever, can only build a degenerate mono-moraic foot, which is prohibited in non-phase-final position. This is the trigger for PF movement: in order to save the derivation, ni- moves down into the vP phase (Piggott & Newell call this movement cliticisation), where it can build on other phonological material in order to be correctly footed. The resulting structure, then, is [__i [(nii-a)-(gam-o)(see)]vP]CP, which hosts an illegal phase-internal hiatus. Regarding the two distinct hiatus-avoiding strategies, Piggott & Newell (2006:27ff) argue that they mirror the derivational contrast at hand. That is, derivations are constantly monitored for phonological wellformedness: strings are checked for constraint violations upon domain specific, and also after PF movement has taken place. Ojibwa then uses different hiatus-killers depending on whether the violation occurs upon lexical insertion (vowel deletion as under (237a)) or after PF movement (d-insertion as under (237c)). While this is certainly a correct description of the facts on the assumption of Piggott & Newell's analysis, it does not explain why there should be different hiatus-avoiding strategies, and why they are distributed in the way they are. Be that as it may, the goal of this presentation is not to evaluate whether a particular analysis fares well or not. What was to be illustrated is the fact that PF movement may also be used for purely phonology-internal purposes: the trigger is phonological, the effect observed is phonological, and the movement is undetectable by morpho-syntactic indicators.

579

5.2.3. Phonologically motivated allomorphy: Lowenstamm on French PF movement is also a central tool of Jean Lowenstamm's recent work that combines phonological representations with the sub.word architecture of Distributed Morphology (Lowenstamm 2008, also Rucart 2006, Lampitelli forth a,b). One example is Lowenstamm's (2008:110ff) analysis of gender in French. French features a puzzling pattern that looks like a phonologically driven allomorphy at first sight: while possessive pronouns regularly agree in gender with following consonant-initial nouns (see (238a)), only the

nological material of both cycles should be interpreted together (i.e. [ni a-gam-osee]CP). Ni- should therefore be able to rely on the material of the inner cycle for foot construction without any PF movement. The authors probably suppose that cross-phase foot construction is ruled out by Phase Impenetrability (but they are not explicit on this).

488 Chap 12: Distributed Morphology masculine mon/ton/son (first, second, third person, respectively) appear before vowel-initial nouns, even if these are feminine (see (238b)). (238) agreement of possessive pronouns in French a. Je vois maF voitureF Je vois monM copainM I see my car I see my friend b. Je vois mon? armoireF Je vois monM aquariumM I see my wardrobe I see my aquarium

The behaviour of the possessive is obviously related to the fact that French elides determiner-final vowels in hiatus position, i.e. when they occur before a vowel-initial word: the definite article is la (fem.) in la moto "the motorbike" and le (masc.) in le train "the train", but l' in l'armoire "the wardrobe" (fem.) and l'aquarium "the aquarium" (masc.). The question therefore is why /saF armoireF/ does not appear as s'armoire, just like /laF armoireF/ produces l'armoire. Given this situation, the classical analysis of monM armoireF simply relies on suppletion: for some unknown reason, the masculine form is chosen when the noun is feminine and vowel-initial. Lowenstamm (2008) proposes a different analysis. He first decomposes determiners: the lateral in la [la] (fem.) and le [lə] (masc.) is the exponent of definiteness, while the -a and zero represent feminine and masculine gender, respectively. 144 The same goes for the possessives ma/mon (first person), ta/ton (second person) and sa/son (third person): m-, t- and s- represent possessiveness, while -a and zero are the gender markers as before.145 The contrast between l'armoire (from /laF armoireF/) and son/*s'armoire (from /saF armoireF/), then, is the result of the different position of definiteness and possessiveness in the morpho-syntactic tree. This is shown under (239) below (from Lowenstamm 2008:113f).

144 145

Lowenstamm assumes that the schwa which appears on the surface in the masculine article is the result of default epenthesis (hence schwa is in fact a zero). Lowenstamm (2008) is not explicit on the status of masculine -on, which could also be the exponent of masculine gender for possessives. Since Lowenstamm promotes an analysis where masculine gender is uniformly zero in French, however, -on in m-on appears to be epenthetic, even when the following noun is masculine. Its epenthetic status with vowel-initial feminine nouns is the heart of his analysis, to be exposed below.

PF movement 489 (239) Lowenstamm (2008): l'armoire vs. *s'armoire a. /laF armoireF/ → l'armoire b. /saF armoireF/ → *s'armoire, son [sɔ˜ n] armoire DP s

D' ɔ˜

FP

insertion of default ç̃ F'

DP D

l

nP nP

n | a

√ | armoire

3

n' n | a

√ | armoire

On Lowenstamm's analysis, the feminine gender -a is always generated in little n and falls prey to regular elision before armoire in both (239a) and (239b). This is all that happens to /laF armoireF/, and the result is l'armoire. With the possessive comes more structure, though, namely a functional projection FP that Lowenstamm does not elaborate on. He argues that the possessive, third person s- in our case, is generated in Spec,nP and raises to Spec,DP ("3" stands for third person). Crucially, then, "the person features of the possessive in Spec,DP cannot be licensed by an empty head" (Lowenstamm 2008:114). This is why some phonological material must be inserted into the head of DP. According to Lowenstamm, the default vowel for this job is [ɔ˜ ] in French, a claim that he tries to substantiate independently.146 He does not explain, however, why an empty head is able to license the definiteness features under (239a), which also sit in Spec,DP (/laF armoireF/ → l'armoire). One concludes that according to his analysis,

146

Which is certainly not an easy thing to do. Lowenstamm mentions that [ɔ˜ ] is the surface subject in impersonal constructions such as on [ɔ˜ ] a soutenu que… "one has claimed that…". But he does not explain how the division of labour is organised between the epenthetic schwa that on his analysis fills in zero masculine gender (/l-zero/ → [lə]), and the epenthetic [ç̃]. Schwa is the vowel that is traditionally assumed to be unmarked and eventually epenthetic in French.

490 Chap 12: Distributed Morphology Spec,DP can be licensed by a phonologically empty head in case it hosts definiteness, but not if it accommodates person features. Lowenstamm's analysis combines morpho-syntactic structure with phonological terminals, which move along the morpho-syntactic tree. This is exactly what PF movement is about: morpho-syntactic structure is still available, but domain specific has already taken place, to the effect that terminals are phonological in nature. We have seen in §578 that these phonological terminals can trigger movement along the tree (Piggott & Newell's hiatus-induced movement). In this context, Lowenstamm's analysis has two interesting properties: phonological objects move along the tree for morpho-syntactic reasons, and morpho-syntactic well-formedness is determined by phonological content. The former is illustrated for example by ma voiture "my car" where the exponent of feminine gender, -a, raises to Spec,DP in the equivalent structure of (239b) in order to join the exponent of person features. Crucially, though, this only happens once the survival of the -a has been negotiated with the following noun: -a only raises if it does not fall prey to elision. In case it does as under (239b), the epenthetic [ɔ˜ ] is needed precisely in order to fill in the gap of the eliminated -a. We are therefore sure that the movement of consonants and vowels along the tree is not just metaphorical: phonological computation has already taken place when the gender exponent -a starts to move. Conversely, morpho-syntactic well-formedness may be determined phonologically in Lowenstamm's analysis, whose critical distinction is between definiteness features that can, and person features that cannot be licensed by a phonologically empty head of DP. Lowenstamm's analysis shows how powerful PF movement is: morpho-syntactic and phonological information are available simultaneously and without any specified point in the derivation where one is converted into the other (domain specific).147

147

Lowenstamm (2008) does not mention the fact that his analysis is an instantiation of PF movement: the presence of phonological terminals in a morphosyntactic tree goes without discussion.

PF movement 491 580

5.3. Overgeneration, direct syntax, modularity and the creation of two distinct computational systems Facing the merits that are pointed out in the literature quoted, PF movement is certainly exposed to a number of objections. For one thing, it dramatically increases the generative power of the the theory. Syntax weaves a strict corset of well-formedness conditions; if now displacement is possible in complete absence of syntactic control, one might want to ask what the whole struggle against overgeneration was worth in the first place. Unloading unwarranted phenomena into PF or an ill-defined intermundia makes syntax minimalistically clean, but leaves other members of the family dirty (see §727). If displacement can be freely triggered by phonological properties, one would expect a large set of PF-movement phenomena, certainly larger a set than what may be ascribed to this mechanism. This includes a number of outlandish "sandhi" phenomena: hiatus avoidance at word breaks for example could be solved by moving one word away. This kind of behaviour, however, is nothing that can be found in natural language. Also, the construction of a post-syntactic space where a computation takes place that is different in kind from syntactic computation undermines the basic claim of Distributed Morphology according to which there is no difference between syntax and morphology. PF movement certainly feeds the doubts that were voiced in §537 regarding the question whether DM lives up to its ambition. The following statement is very clear. (240) "We develop a theory of movement operations that occur after the syntactic derivation, in the PF component, within the framework of Distributed Morphology. The theory is an extension of what was called Morphological Merger in Marantz 1984 and subsequent work. A primary result is that the locality properties of a Merger operation are determined by the stage in the derivation at which the operation takes place: specifically, Merger that takes place before domain specific, on hierarchical structures, differs from Merger that takes place post-domain specific/linearization." Embick & Noyer (2001:555, emphasis in original)

We are thus left with two sets of locality properties for Merge according to whether the operation occurs in or after syntax. Also, Marantz' (1984) Morphological Merger is now called PF movement, which takes place after syntax. This is advertised as a major result by authors who promote the DM approach and hence "reject the view that morphology is insulated from syntax" (Embick & Noyer 2001:592).

492 Chap 12: Distributed Morphology Finally, PF movement begs the question regarding modularity: what kind of module carries out computation that has access to both morphosyntactic structure and phonological material? A module that corresponds to this derivational intermundia cannot exist if domain specificity defines modular contours (see §§610,640): syntactic features and phonological material are ontologically different and thus uninterpretable in the sister module. It cannot be argued that PF movement takes place in what phonologists call phonology, i.e. the phonological module (whose relationship with PF is in need of definition, see §726): here only phonological vocabulary can be processed, but PF movement supposes the presence of the full morpho-syntactic tree structure, including node labels. This again leads to a violation of domain specificity: the phonological computational system cannot parse or compute morpho-syntactic labels or trees. Since the terminals are made of phonological material when PF movement takes place, another bewildering fact about the trees at hand is that their nodes and labels are the projection of nothing (see §715): how could the morpho-syntactic tree "survive" in absence of its terminal features? On standard assumptions, a tree is nothing but a projection of properties of the terminal elements (see Chomsky's inclusiveness, §648). PF movement is thus a strong modularity offender. The modularity-offending character of PF as such and its relationship with true phonology (i.e. the thing that phonologists call phonology) is further discussed in §§731,738). The modularity-offending character of PF movement is also confirmed on another count: Distributed Morphology has a natural sympathy for direct syntax (§407), another modularity-violating device. PF movement is a strong form of direct syntax since on top of untranslated structure and labels, it promotes phonology-triggered displacement. Interestingly, though, and unlike what is done in OT where direct syntax is also commonplace (§525) but happily cohabitates with prosodic constituency, Pak (2008:42ff, in a DM environment) and Samuels (2009a:284ff) rightfully argue that if direct syntax is correct, the Prosodic Hierarchy is redundant and needs to be done away with altogether. Recall from (§407) that this is indeed a compelling move: prosodic constituency is useless if direct reference to morpho-syntactic structure and labels can be made (and in absence of non-isomorphism, which was shown in §416 to be an artefact of the domain-based perspective, rather than a fact about language).

Conclusion 493 581

6. Conclusion

582

6.1. Direct merge vs. selective spell-out: a geometric, rather than a computational solution Distributed Morphology is neither a theory of phonology nor a theory of the interface. The matters that are of interest for the interface-inclined phonologist are therefore only secondary in the DM literature, and the body of work at hand is fairly restricted. Distributed Morphology follows Chomsky's minimalist perspective where much labour is unloaded from syntax in direction of the interface(s). Phase theory, the central instrument of this approach, is largely debated in syntactic quarters, but discussion usually ends when it comes to its lower end: PF is a black box for syntacticians, who hardly go beyond "and then PF is in charge", or "this is solved at PF" (see §727). The phonologically relevant work in Distributed Morphology goes into greater detail. The direction was given by Marantz (2001, 2007): in response to the robust observation that "lower" items produce irregularity (while "higher" pieces are regular), he introduces the idea that direct merge is responsible for opacity, both phonological and semantic. This allows him to maintain just one arboreal structure and one computational system (i.e. one engine), against the traditional split between a morphological and a syntactic hierarchy with associated distinct computational systems. As a result, non-transparency at PF and LF is a geometric property (being a sister of the root may create opacity), rather than a consequence of selective spell-out. As far as I can see, this is an entirely novel take: opacity in general and affix class-based phenomena in particular have always been ascribed to a procedural (computational) mechanism on the grounds of identical geometric structure. Distributed Morphology now does the reverse: opacity is due to a static geometric contrast (direct vs. indirect merge), which is accompanied by an invariable computation (spell-out at every xP) (§560). This analysis where spell-out and Phase Impenetrability play no role is applied in a small body of literature whose main representatives (regarding phonology) are the Ph.D of Tatjana Marvin (2002) and the work by Glyne Piggott and Heather Newell at McGill (Barragan & Newell 2003, Newell 2005a,b, 2008, Piggott & Newell 2006, Piggott 2007).

494 Chap 12: Distributed Morphology 583

6.2. Spell-out at every xP stands in the way A shared assumption of the work in DM is the idea that all categorising heads, i.e. nP, vP, aP, are phase heads. The definition of phasehood is a debated issue and a field of rapid evolution in the syntactic literature (see §771). It is certainly true that there is a trend towards the atomisation of phasehood: interpretational units are smaller and smaller, i.e. more and more nodes acquire this status. Spelling out every categorising, i.e. nonfunctional node, however, is way beyond the current syntactic horizon. It was shown in §557 that spell-out at every xP disables Distributed Morphology to account for affix class-based phenomena: underapplication cannot be expressed. On the other hand, it was shown in §566 that selective spell-out is unable to handle cases of the twinkling type where LF, but not PF interpretation is opaque. How PF and LF opacity can combine was studied in §564: free distribution prevails. When all four cases are examined, it thus appears that both direct merge and selective spell-out are unable to account for one pattern. This opens the possibility for both solutions being complementary according to the phenomenon considered. What makes direct merge and selective spell-out incompatible is the additional claim that all xPs are phase heads. Nothing withstands the simultaneous existence of a kind of opacity that is due to direct merge with another kind that originates in selective spell-out if spell-out at every xP is abandoned. Spell-out at every xP is also at variance with syntactic phase theory, to which Distributed Morphology is committed: derivation by phase supposes selective spell-out, and phase heads are not defined according to whether or not a node is functional head. Phase theory also relies on Phase Impenetrability, a mechanism that plays no role in the direct merge analysis of opacity (but which is used in Kaye's 1995 version of selective spell-out).

584

6.3. Unifying ambitions and specific tools for morphology It was mentioned on a number of occasions that Distributed Morphology hardly stands up to its ambition, i.e. the unification of morphology and syntax. Fission, Fusion and Impoverishment are specific morphological devices that are unknown in syntax, and spell-out at every xP is incompatible with current syntactic practice. PF movement also lines up here: beyond the severe overgeneration that it allows for and the blurred modular contours which ensue, it intro-

Conclusion 495

duces a specific computational system whose locality conditions are different from the ones that are known from syntactic computation. PF movement takes place in a misty intermundia, PF, which is neither syntax nor phonology, but a little bit of both (more on the modularity-offending hermaphrodicity of PF in §726).

585

6.4. Two specificities: LF and no proposal regarding representational communication When compared to other theories of the interface, Distributed Morphology has two specific characteristics that stem from its morpho-syntactic orientation: attention is paid to LF phenomena, and no statement is made regarding representational communication with phonology. We have seen that Distributed Morphology considers semantic as much as phonological effects of morpho-syntactic structure and spell-out. The former were discussed since the earliest generative literature on morphology (Siegel 1974, Allen 1978, 1980, Aronoff 1976, Pesetsky 1979), but are by and large absent from SPE and further literature on the interface: neither Lexical Phonology nor Prosodic Phonology look much at LF properties in order to explain the behaviour of phonology at the interface. The fact that there is a non-arbitrary relationship between effects on both sides at least in some cases is beyond doubt. Interface theory is therefore called to be able to derive LF and PF effects from the same morphosyntactic structure. The existence of simultaneous LF and PF effects is also support for the classical (syntactico-centristic) perspective of the inverted T. Finally, it is to be noted that Distributed Morphology makes no contribution to the representational communication with phonology: no general position is taken whether any objects that represent morpho-syntactic information are inserted into phonological representations, and if so, what they might look like (boundaries, prosodic constituents etc.). Only Pak (2008:42ff) and Samuels (2009a:284ff) call into question the very existence of prosodic constituency, which according to them should be done away with in favour of direct reference to morpho-syntactic structure and labels (§580).

586

587

Interlude Modularity Chapter 1 Introduction: the relative absence of modularity in interface thinking

On various occasions of the historical survey in Part I, modularity was referred to, and a number of arguments were made in its name. A striking fact is that the interface literature itself does not seem to take modularity into account: it is hard to find any explicit reference, and even harder to find modularity-based reasonings.148 This is unexpected, to put it mildly, since modularity is part and parcel of the generative approach to language: it is never missing in introductory classes and introductions to the general architecture of grammar in textbooks. Quite opposite of its supposed overarching importance, however, is its influence on theory design. In comparison to straight theories of syntax, phonology and other subdisciplines, this of course is a much more pressing issue for interface theories. The absence of modularity from interface thinking and interface design is certainly one reason for the current situation in OT: modularity is supposed to be in place (OT is a generative theory), but it is systematically violated without this arousing much discussion (§§469,523). We have seen that OT is not the only modularity offender (SPE and Distributed Morphology are other cases in point), though. A summary as well as further discussion and classification of generative modularity offenders is provided in §702 below. A telling example of the absence of modularity in interface thinking was reported in §407: the head stone of Prosodic Phonology, the principle of Indirect Reference, is a direct expression of modularity – but it was introduced without any reference to this concept. Rather, non-isomorphism was invoked – an empirical argument that turns out to be immaterial when 148

There are number of noticeable (and rather recent) exceptions, including namely the work by Charles Reiss and Eric Raimy: Reiss (2000), Hale & Reiss (2008:105ff), Isac & Reiss (2008), Reiss (2008), Cairns & Reiss (2009), Reiss forth, Raimy (2003), Idsardi & Raimy (forth). Unfortunately this work has quite little visibility in current interface thinking of mainstream theories.

498 Modularity Chap 1: The relative absence of modularity in interface thinking carriers of morpho-syntactic information are local, rather than domainbased (§416). In sum, what happened is that Prosodic Phonology did exactly the right thing – Indirect Reference – for the wrong reason (nonisomorphism). Although modularity was contemporary (Fodor's foundational book appeared in 1983), the simple argument that direct reference to untranslated morpho-syntactic objects is ruled out by the modular architecture was never made as far as I can see. An introduction to modularity as such that exposes the modular view of how the cognitive system works is therefore not superfluous. Linguists, especially phonologists, may or may not be familiar with the cognitive foundations of modularity, what it requires, and what it rules out. Also, modularity has already done critical labour, and will do still more in this book, where it is used as a referee (§36). It goes without saying that the pages below do not aim at a fullyfledged introduction to modularity: the specialised Cognitive Science and psychological literature that is mentioned throughout does a much better job. The same goes for connectionism, the competing theory of cognitive organisation. This Interlude is only meant as a short armamentarium in modularity for phonologists (and more generally linguists). Isac & Reiss' (2008) and Boeckx's (2010) recent (text)books on language and cognition cover a number of issues that are discussed below: they provide a broad introduction to language and linguistics from the Chomskian (and, in the case of the latter, specifically biolinguistic) point of view, and argue on the backdrop of Cognitive Science (without however engaging into discussion with connectionism: modularity is taken for granted).

588

Chapter 2 Modularity and connectionism, mind and brain

589

1. Monism vs. dualism, symbolic vs. non-symbolic representations

590

1.1. Levels of representation in the standard cognitive model Under the header of what today is called the cognitive revolution (e.g. Gardner 1985), the modular approach to cognition was put on the agenda in the 50s and 60s as an alternative to (psychological) behaviourism and parts of (linguistic) structuralism. Rather than describing the stimuli and the responses of an organism, focus was put on the actual cognitive processes that take place when speech is produced and processed (a black box in behaviourism). Rather than describing a linguistic system without location in space and time, the cognitive operations that it supposes became the centre of interest. This call for cognitive realism is essentially what Chomsky's (1959) critique of Skinner's book is about. Generative linguistics were leading in the introduction of the new cognitive conception then, and today language remains a central issue. Critical for modularity and generative linguistics is the difference between mind and brain, which is akin to the distinction between competence and performance. Although the mind of course has a neural implementation, it may be studied independently of the neuro-biological reality. In fact, trying to get hold of language by looking at its neuronal reality alone is quite unlikely to produce significant insight. On the other hand, models of the mind are constrained by the limitations of what is neurally possible and plausible. The best understanding of language may therefore be expected from a dialectic exchange between the study of mind and the study of brain, bottom-up as much as top-down.149 The debate between the classical cognitive model on the one hand and connectionism on the other is about 25 years old; the following discussion only ambitions to provide a brief summary of some basic aspects. More detail is available for example in Dinsmore (1992). Laks (1996) and Pylyshyn & Lepore (eds.) (1999) offer informed overviews; more specialised literature includes Newell (1980), Fodor & Pylyshyn (1988), 149

Simon & Kaplan (1989:7f) and Pylyshyn (1989a:60ff) elaborate on the standard notion of levels of representation in Cognitive Science.

500 Modularity Chap 2: Modularity and connectionism, mind and brain Smolensky (1988a, 1991), Fodor & McLaughlin (1990), Harnad (1990), and other references that are mentioned as we go along. Finally, Fodor (1985) provides a helpful overview of the different schools of thought in Cognitive Science.

591

1.2. A language of thought: symbolic vs. anti-symbolic views of cognition Theories of the mind necessarily use representations and symbols (such as trees, DPs, nuclei etc.), which are supposed to get as close as possible to the units that are manipulated by the mind. This is the "language of thought", a notion that was introduced by Fodor (1975) and has been debated since then. (241) "There has always been opposition to the view that we have symbol structures in our heads. The idea that the brain thinks by writing symbols and reading them sounds absurd to many. It suggests to some people that we have been influenced too much by the way current electronic computers work. The basic source of uneasiness seems to come from the fact that we do not have the subjective experience that we are manipulating symbols. But subjective experience has been a notoriously misleading source of evidence for what goes on in the mind. Research in human information processing reveals countless processes that clearly must be occurring (for example, parsing, inference) of which we have little or no subjective awareness." Pylyshyn (1989a:61)

The rejection of symbolic representations has condensed into the theory of connectionism in the 80s (Rumelhart et al. 1986). Connectionism challenges the standard cognitive model on the grounds of the mind-brain distinction, which is denied (typically, though not by all representatives of connectionism, as we will see): only the brain is relevant since only the neural level is decision-making. Therefore anything that goes beyond the study of the brain and its modelling (in terms of artificial neural networks) is misleading and unhelpful. That is, computation should be brain-style (rather than a machine-style), as Rumelhart (1989:134f) puts it: the basic calculating units must be (eventually artificial) neurons, and the items processed are numbers, rather than symbolic representations. Connectionism (or at least some versions thereof) is thus reductionist in kind: it does not accept the mind-brain dichotomy. Symbolic and representational systems describe things that have no neural basis and hence are

Monism vs. dualism, symbolic vs. non-symbolic representations 501

pure speculation (e.g. Churchland 1993, Chomsky 1995b exposes the rationalist refutation of reductionism). Philosophically speaking, connectionism is monistic (the cognitive system is made of the brain and of nothing else), while the classical cognitive model is dualistic (the mind and the brain exist and are both relevant). Also, connectionism is a typical incarnation of empiricism which relegates any reasoning that is not data-based into the realm of unwarranted speculation. By contrast, the dualistic mind-brain approach falls into the tradition of rationalism/mentalism. Interestingly, the charge against the classical cognitive model is thus led in the name of cognitive realism, just as was the introduction of the cognitive model in the 50s-60s: cognitive realism then, neural realism now. The linguistic implementation of the empiricist-connectionist approach is called "Cognitive" Grammar (see §596 for references and further discussion). The name of this framework is a good illustration of the issue at hand: it is deliberately chosen in order to warrant a copyright on the word "cognitive". Defenders of this model try to establish that their theory is the only cognitive theory about language – pure propaganda.150 But they are deadly serious about stamping generative linguistics as non-cognitive. The following quote is form a textbook that introduces to "Cognitive" Grammar. (242) "While most linguists, nowadays, would no doubt agree that linguistics is a cognitive discipline […], there have been important approaches within linguistics which have denied, or simply ignored, the discipline's cognitive dimension. Among these we can identify the formalist and the behaviourist approaches. (A cynic might say that quite a lot of modern linguistics is actually to be located within the formalist approach, with appeals to cognitive aspects being little more than lip-service to a modern fashion.) […] Work by Chomsky and his sympathizers, as well as various offshoots of Chomsky's theories, […] are very much formalist in orientation." Taylor (2002:6f)

In this vein, Taylor (2002:6) talks about Chomskian linguistics as "cognitive linguistics", i.e. in quotation marks. Saying that one of the founders of modern Cognitive Science denies or ignores cognitive issues in language is coming on a little strong. This kind of statement illustrates the empiricist-connectionist line of attack, though (which is sometimes expressed with more subtlety).

150

Which is the reason why I refer to the framework as "Cognitive" Grammar in this book.

502 Modularity Chap 2: Modularity and connectionism, mind and brain 592

1.3. What would adult science look like without symbols? Looking at what appears to be one of the central debates of Cognitive Science from the outside, the question arises whether any other science, especially any adult science, could afford such a discussion. What is physics, what is chemistry, what is biology about? Understanding how things work, or trying to produce a photograph of the natural object under study? Scientific discoveries have always been made in symbolic terms (I am not sure whether there is any exception at all): objects are thought, described and drawn before instruments provide evidence for their existence (think of molecules, atoms, the double helix, electrons, protons or whatever is your favourite). And scientists write down mathematical formulae in order to describe processes. The point is that in all adult sciences, the only scientific reality is a representation of the real-world reality, in terms of a drawing and/or in terms of a formula. The relationship between both is often non-trivial, and it typically takes a lot of effort in order to be able to go from one to another: this is what engineering is about. Engineers construct machines on the grounds of a scientific insight. The engineering effort is typically undertaken long after the death of the scientist on whose discovery it is based. Engineers are the negotiators between the scientific reality and the realworld reality. Take the atom: the well-known representation under (243a) is the only scientific reality that counts. No doubt the picture under (243a) is idealised and only remotely resembles a real atom that would be visualised by some image-giving system (microscopes today can see individual atoms, but they cannot yet look into them). Or consider what it takes to understand how a computer works, i.e. its basic building block, the transistor: do we need to know about (243b), or do we need to inspect the makeup of the kind of physical item under (243c)? Science is about gaining insight, not about engineering. No physicist would ever challenge the physical reality of (243a); optical confirmation or any other instrumental evidence is of course welcome – but this is secondary with respect to the reality of (243a) that owes its legitimacy to its explanatory virtue. In particle physics, highly sophisticated machines, particle accelerators, are busy providing experimental evidence (though indirect: nobody has ever seen a quark on a photograph) for all the particles that are predicted by theoreticians whose only equipment is a sheet of paper and a pencil. And the particles predicted were found one by one.

Monism vs. dualism, symbolic vs. non-symbolic representations 503 (243) physical and scientific reality a. an atom: scientific reality

b. a transistor: symbolic

c. a transistor: physical

Of course representations such as (243a) may turn out to be imprecise, incomplete or, according to the degree of incompleteness, simply wrong. But they have been established as the scientific reality, the only reality that counts in science, on grounds that have got nothing to do with any attempt to mimic its real-world properties: electrons turn around protons and neutrons because there is evidence to this end – remote effects whose inspection has led to (243a) as the best hypothesis. And it was by the same mechanism – observation of remote effects that make only sense if a certain structure is assumed – that it was discovered that protons are made of still smaller units, quarks. Will anybody argue against thirteendimensional string theory because strings are only a symbolic metaphor for some real "reality"? Another point of interest is that in the history of (adult) science, progress has always been made when something was understood – not when it could be implemented. Einstein did not do any real-world experiment in order to build relativity theory – it was enough for him to think. Like it is often the case in adult science, the core of his discovery is a symbolic mathematical formula, E = mc2. Long after Einstein died, relativity (among

504 Modularity Chap 2: Modularity and connectionism, mind and brain other things) eventually allowed humans to walk on the moon, and to engineer machines that allow a precise location on the planet (GPS). It was mentioned earlier that perhaps with a few exceptions, objects of the real world have always been thought and drawn on a piece of paper before they were eventually visualised by advanced instrumentation: cells, molecules, the double helix, quarks, strings, H2O and so forth.151 It is therefore difficult to understand why regular scientific standards seem to be questioned just when the object of study happens to be the mind/brain. Prohibiting to draw pictures and to talk about things in terms of symbols because the real world does not have any of those would not cross any physicist's mind. On the other hand, it is obvious and undisputed (in physics, chemistry and Cognitive Science) that whatever symbolic hypothesis is made, it must be compatible with what we know about the real world. That is, a biological theory that dismisses the cell can be rejected out of hand as much as a cognitive theory that supposes infinite storage capacity. The relationship between symbolic scientific reality and real-world objects – their "implementation" – is intricate and dialectic, but in any case non-arbitrary: hypotheses must be implementable, but it takes a lot in order to show that a symbolic system could not possibly be the correct representation of a natural object. The history of science offers countless examples where some hypothesis was judged outright impossible, sometimes for centuries, before being found to be real (think of the history of astronomy). Scientific hypotheses are abandoned because they are shown to be wrong, not because they are made of symbols or non-symbols. It is hard to see why symbolic representations in cognitive matters – a syntactic tree for example – should have any different heuristic, epistemological or scientific value than (243a). In both cases, the negotiation with the real world is dialectic and complicated, and the distance with the symbolic representation may be more or less important. Since the relationship between representation and "reality" is dialectic, experimentation in both cases will provide valuable evidence for the eventual amendment of the 151

In phonology, Morris Halle has used this line of argumentation a long time ago in order to justify the reality of the phoneme: "Helmholtz postulated that electric current is a flow of discrete particles without having isolated or even having much hope of isolating one of these particles. The status of the phoneme in linguistics is, therefore, analogous to that of electrons in physics, and since we do not regard the latter as fictional, there is little reason for applying this term to phonemes. They are every bit as real as any other theoretical entity in science" Halle (1964:325).

Connectionism and its representatives in linguistics 505

representation, just as much as the experiments are designed on the grounds of hypotheses whose basis is the representation. In particle physics, particle accelerators provide experimental evidence, while in Cognitive Science neurobiology (image-giving systems) plays this role. In both cases, representational and experimental activity must of course work hand in hand. And again, there is no question about that in adult science: it would be utterly absurd to doubt that this collaboration is necessary. Non-adult science is not only non-adult because it is younger and not (yet) confirmed by massive experimental evidence and engineering success – it is also childish because it does not behave like adults: reducing activity to either representational or neurobiological endeavour seriously reduces the chances of a serious scientific prospect. The dialectic relationship between the functions of the mind and their (specialised and localisable) neural existence is further discussed in §614.

593

2. Connectionism and its representatives in linguistics

594

2.1. The symbolic front line and its roots in Cognitive Science The symbolic front line introduced above is worked out in a special issue of Cognition on Connectionism and Symbol Systems in 1988 (number 28, edited by Steven Pinker and Jacques Mehler, also published as Pinker & Mehler (eds.) 1988). As the editors explain in the introduction, this issue is more or less the answer of standard Cognitive Science to the connectionist challenge was raised by Rumelhart et al. (1986). Especially the article by Fodor & Pylyshyn (1988) works out the symbolic vs. anti-symbolic line of division. While Dinsmore (1992) provides a book-length contrastive overview of the two approaches, Newell (1989) is a non-comparative introduction to symbolic models of cognition (and their parallels with computers). The issue is discussed from a wholesale perspective by Pylyshyn (1999). But of course connectionism does not reduce to the symbolic issue. The quote of Fodor & Pylyshyn below locates the connectionist research programme in the context of traditional Cognitive Science.

506 Modularity Chap 2: Modularity and connectionism, mind and brain (244)

"Connectionism really does represent an approach that is quite different from that of the Classical cognitive science that it seeks to replace. Classical models of the mind were derived from the structure of Turing and Von Neumann machines. They are not, of course, committed to the details of these machines as exemplified in Turing's original formulation or in typical commercial computers; only to the basic idea that the kind of computing that is relevant to understanding cognition involves operations on symbols. […] In contrast, Connectionists propose to design systems that can exhibit intelligent behaviour without storing, retrieving, or otherwise operating on structured symbolic expressions." Fodor & Pylyshyn (1988:4f)

Below some basic information on how the connectionist model works is provided.

595

2.2. How neural networks work The connectionist approach to the cognitive system is based on the assumption that there are only two significant items, neurons and synapses. The former are computational units, which are fed by the latter: synapses transport information among neurons and decide how much of it gets through the pipe: this is the weight of a neuronal connection, which decides on the activation value of the target neuron. The quote below provides a summary of how a neural network works. (245) "Connectionist systems are networks consisting of very large numbers of simple but highly interconnected 'units'. Certain assumptions are generally made both about the units and the connections: Each unit is assumed to receive real-valued activity (either excitatory or inhibitory or both) along its input lines. Typically the units do little more than sum this activity and change their state as a function (usually a threshold function) of its sum. Each connection is allowed to modulate the activity it transmits as a function of an intrinsic (but modifiable) property called 'weight'. Hence the activity on an input line is typically some non-linear function of the state of activity of its sources. The behavior of the network as a whole is a function of the initial state of activation of the units and of the weights on its connections, which serve as its only form of memory." Fodor & Pylyshyn (1988:5)

There are many surveys and introductions to how connectionist systems work, e.g. Rumelhart (1989), Stillings et al. (1995:63ff), Thagard (2005:111ff); Smolensky (2003) and Smolensky & Legendre (2006) provide a linguistically oriented overview. The following section takes a closer

Connectionism and its representatives in linguistics 507

look at one particular aspect of the connectionist approach, the non-status of stored information and the outgrowths of this idea in linguistics.

596

2.3. No distinction between storage and computation (the rule/list fallacy) Beyond the symbolic issue, the references are clear from the quote under (245): regular Cognitive Science follows the Turing/von Neumann idea that computation is done by short-term/working memory in a procedural way (step by step) on the grounds of a storage device – long-term memory. The dissociation of actual action and independently stored instructions that govern this action is the essence of the Universal Turing Machine where a "head" performs action on the grounds of instructions that are found on a "tape". This simple architecture was improved by John von Neumann, who introduced a distinction between two kinds of storage systems: the one that contains the actual instructions for action (Relative Access Memory, which today would be called the programme/software), and another one that is just a data-storage device where things can be stored that are not needed for current action, and from which they can be retrieved when necessary. To date this is the basic architecture of computers, and also of Artificial Intelligence (AI) (e.g. Haugeland 1989:133ff, see also §603). On the cognitive side, the memory that stores instructions for action (the programme/software) is the essence of the relevant module, the datastoring device is long-term memory and the computational space where actual action is performed is short-term memory. There is neuropsycholgical evidence that declarative (long-term memory) and procedural (instructions for action) knowledge are distinct: they can be dissociated by brain damage (Stillings et al. 1995:62, 312ff, Squire 1987). In sum, computation and storage are crucially independent: the former builds on the latter. That is, a process transforms a pre-existing object. Connectionism denies the distinction between computation and storage: the only "storage" that it provides for is the online value of activation levels. In practice, the "experience" of a neural network – the equivalent notion of memory – is acquired when the patterns of connectivity change: neurons may develop new connections (synapses), may lose old connections, or modify the strength (weight) of existing connections (the two former are often viewed as a special case of the latter). The computational units themselves have no variable behaviour that contributes to the properties of the

508 Modularity Chap 2: Modularity and connectionism, mind and brain whole, which are exclusively determined by the connective network (see Stillings et al. 1995:114ff on connectionist models of memory). All linguistic theories since Antiquity of course rely on the assumption that there is a lexicon which exists independently of grammatical activity. Grammar transforms lexically stored objects into actual speech. The linguistic mirror of the connectionist non-separation of storage and computation is so-called "Cognitive" Grammar, which was founded by Ronald Langacker (1987) (see Taylor 2002). Langacker (1987 Vol.1:42) talks about the "rule/list fallacy". The phonological off-spring of this line of thought is represented by exemplar- and usage-based approaches in general, and by Joan Bybee in particular. The following quote is explicit on that. (246) "Perhaps the most fundamental difference between the model to be explored here and structuralist or generativist models is the rejection of the notion that material contained in rules does not also appear in the lexicon and vice versa. […] Linguistic regularities are not expressed as cognitive entities or operations that are independent of the forms to which they apply, but rather as schemas or organisational patterns that emerge from the way that forms are associated with one another in a vast complex network of phonological, semantic, and sequential relations." Bybee (2001:20f).

We are thus light-years away from anything that could be reconciled with generative thinking, actually with any linguistic thinking at all at least since the 19th century.

597

2.4. All-purpose parallel vs. specialised step-by-step computation Another important connectionist headline is Parallel Distributed Processing, which contrasts with the classical Turing/von Neumann assumption that computation is serial: the output of one computation is the input to another. On the connectionist count, several computations take place simultaneously, like in the brain. Rumelhart (1989) explains why computation cannot be step-by-step when it is carried out by real brains: this would require too much time.

Connectionism and its representatives in linguistics 509 (247) "The operations in our models then can best be characterized as 'neurally inspired.' How does the replacement of the computer metaphor with the brain metaphor as model of mind affect our thinking? This change in orientation leads us to a number of considerations that further inform and constrain our model-building efforts. Perhaps the most crucial of these is time. Neurons are remarkably slow relative to components in modem computers. Neurons operate in the time scale of milliseconds, whereas computer components operate in the time scale of nanoseconds – a factor of 106 faster. This means that human processes that take on the order of a second or less can involve only a hundred or so time steps. Because most of the processes we have studied – perception, memory retrieval, speech processing, sentence comprehension, and the like – take about a second or so, it makes sense to impose what Feldman (1985) calls the '100-step program' constraint. That is, we seek explanations for these mental phenomena that do not require more than about a hundred elementary sequential operations. Given that the processes we seek to characterize are often quite complex and may involve consideration of large numbers of simultaneous constraints, our algorithms must involve considerable parallelism. Thus although a serial computer could be created out of the kinds of components represented by our units, such an implementation would surely violate the 100-step program constraint for any but the simplest processes." Rumelhart (1989:135, emphasis in original)

Another aspect of connectionist computation is that the units which carry out computation – neurons, or clusters thereof – are not specialised for a particular computational task, or for a particular input material. Rather, neurons are an all-purpose computational unit that is able to perform any computation on the grounds of any type of information submitted. This is why connectionist computation is called distributed. A corollary of distributed computation is the claim that computation is opportunistic and does not need any specialisation of its support units, the neurons: computation is colourless. We will see below that the modular approach works with the exact reverse assumption: there are stable, genetically endowed, content-sensitive computational units that are devised for a very narrow and specific function, which can only work with a particular type of input vocabulary, and can do nothing else than what they are designed for. In a historical perspective, Marshall (2001:510) comes up with a quote by Charlton Bastian from the late 19th century that is surprisingly modern in anticipating the debate between modularity and connectionism.

510 Modularity Chap 2: Modularity and connectionism, mind and brain (248) "The fundamental question of the existence, or not, of real 'localizations' of function (after some fashion) in the brain must be kept altogether apart from another secondary question, which though usually not so much attended to, is no less real and worthy of our separate attention. It is this: Whether, in the event of 'localization' being a reality, the several mental operations or faculties are dependent (a) upon separate areas of brain-substance, or (b) whether the 'localization' is one characterized by mere distinctness of cells and fibres which, however, so far as position is concerned, may be interblended with others having different functions. Have we, in fact, to do with topographically separate areas of brain-tissue or merely with distinct cell and fibre mechanisms existing in a more or less diffuse and mutually interblended manner?" Bastian (1880, emphasis in original)

Significantly, Bastian's book is entitled "The brain as an organ of mind" (more on this relationship in §627). 598

2.5. What it all comes down to: connectionist computation is content-free The properties of the connectionist architecture mentioned conspire to the assertion that the mind does not know what it is doing when computation takes place: computation is only general-purpose, that is non-specialised for any task or function; it works without reference to any symbolic code, which would make the operations specific to a particular domain or content since symbols are symbols of something, and may be opposed to symbols of a different kind. General-purpose parallel computation cannot rely on memory either because memory, again, would be the memory of something, that is specific to a particular content. We will see below (§604) that modularity takes the opposite position on every issue mentioned, and that this follows from the conception that computation is computation of something: it is specific to a domain and to a function, that is content-sensitive and content-imparting (Cosmides & Tooby 1992a focus on these notions as a key difference between the modular and the connectionist approaches).

599

3. Conclusion: peaceful coexistence at first, but not for long In sum, there are two competing conceptions of how the mind/brain works: the cognitive system is either made of interconnected all-purpose units (neurons), or a network of specialised and unexchangeable units (modules).

Conclusion: peaceful coexistence at first, but not for long 511

The connectionist line of attack is to challenge the standard cognitive model because of its unwarranted analogy with microelectronic computers (that carry out specialised and serial computation and work with short- and long-term memory): nothing entitles to conclude that the mind/brain should work like the most sophisticated man-created machine that is currently available. This is challenging the symbolic model on the very grounds on which it emerged during the cognitive revolution of the 50s-60s: psychological realism. Still more real than symbolic units are neurons, and in their self-understanding, connectionist models are neurally inspired and develop a brain-style (rather than a machine-style) computation (Rumelhart 1989:134). Interestingly, the connectionist and the standard cognitive approaches were not understood as irreconcilable competitors at first: in the early literature when connectionism individuated from the symbolic cognitive mainstream, it was viewed as an interesting complement which offers a biologically and neurally grounded implementation of the higher level symbolic system. Smolensky (1987) for example is explicit on this. (249) "In this paper I present a view of the connectionist approach that implies that the level of analysis at which uniform formal principles of cognition can be found is the subsymbolic level, intermediate between the neural and symbolic levels. Notions such as logical inference, sequential firing of production rules, spreading activation between conceptual units, mental categories, and frames or schemata turn out to provide approximate descriptions of the coarse-grained behaviour of connectionist systems. The implication is that symbol-level structures provide only approximate accounts of cognition, useful for description but not necessarily for constructing detailed formal models." Smolensky (1987:95)

Connectionism was thus a question of levels: following the work by Marr (1982) on vision, the symbolic level could stand unchallenged as long as connectionist neural networks were understood as an intermediate level between the (neural) biology of the brain and the functional view of the symbolic level. Dinsmore (1992) and Macdonald & Macdonald (1995) have edited books that revolve entirely around the possibility of understanding the classical theory of the mind and connectionist mimicking of neurons and synapses as two ways of looking at the same object which are equally legitimate: rather than constructing competing views of the mind/brain, they are complementary and both necessary for the understanding of the mind/brain, just as zoology and biology are for the understanding

512 Modularity Chap 2: Modularity and connectionism, mind and brain of living species. 152 The book about modularity and language edited by Garfield (ed.) (1987) also falls in this early period where a synthesis was still an option. Fodor & Pylyshyn (1988), from the camp opposite to Smolensky's, also arrive at this conclusion. (250) "Treat Connectionism as an implementation theory. We have no principled objection to this view (though there are, as Connectionists are discovering, technical reasons why networks are often an awkward way to implement Classical machines). This option would entail rewriting quite a lot of the polemical material in the Connectionist literature, as well as redescribing what the networks are doing as operating on symbol structures, rather than spreading activation among semantically interpreted nodes. Moreover, this revision of policy is sure to lose the movement a lot of fans. As we have pointed out, many people have been attracted to the Connectionist approach cause of its promise to (a) do away with the symbol level of analysis, and (b) elevate neuroscience to the position of providing evidence that bears directly on issues of cognition. If Connectionism is considered simply as a theory of how cognition is neurally implemented, it may constrain cognitive models no more than theories in biophysics, biochemistry, or, for that matter, quantum mechanics do. Al1 of these theories are also concerned with processes that implement cognition, and all of them are likely to postulate structures that are quite different from cognitive architecture. The point is that 'implements' is transitive, and it goes all the way down." Fodor & Pylyshyn (1988:67f, emphasis in original)

One senses, though, that the peaceful coexistence will not last long on either side: the battle is engaged for pocketing the biggest possible piece 152

One track followed in the "bridging" literature, i.e. which tries to make classical Cognitive Science and connectionism peacefully cohabitate, is to take artificial neural networks that are designed according to connectionist principles, and to add content-labels to the neurons, which are also arranged graphically in order to mimic the image provided by the classical theory. In linguistics, for example, Stevenson (1999) draws regular syntactic trees whose nodes are interpreted as neurons with the relevant linguistic labels, and whose branches are synapses. This of course is violating all principles on all sides: connectionist neural networks cannot have any content, and branches of syntactic trees represent domination relationships: they have got nothing to do with activation levels. In the same spirit, Plaut (2003) relates sub-components of grammar by a neural network: phonology is characterised as a neural entity that entertains relationships (in terms of activation levels) with other neural entities representing semantics, acoustics and articulation.

Conclusion: peaceful coexistence at first, but not for long 513

of the cake: how much of cognitive activity is symbolic, and how much is connectionist? Where exactly are decisions made? Smolensky's (1987) conclusion also goes this way. (251) "The heterogeneous assortment of high-level mental structures that have been embraced in this paper suggests that the symbolic level lacks formal unity. This is just what one expects of approximate higher-level descriptions, which, capturing different aspects of global properties, can have quite different characters. The unity which underlies cognition is to be found not at the symbolic level, but rather at the subsymbolic level, where a few principles in a single formal framework lead to a rich variety of global behaviours." Smolensky (1987:108)

The front lines have grown more rigid since the 80s, and as far as I can see there is not much left of a peaceful level-specific coexistence of symbolic and connectionist models. They are globally competing theories of cognitive activity, even if this or that connectionist model may not completely exclude the existence of a symbolic residue. It was already mentioned in §529 that Smolensky & Legendre's (2006) version of OT, Harmonic Grammar, pursues the peaceful coexistence programme in linguistics: parallel distributed computation (constraint ranking) operates over traditional symbolic items (typically segments). The main connectionist import into linguistics is parallel computation, which is the headstone of (all versions of) OT. The question whether this import into a rationalist theory of language (generative grammar) can succeed without serving as a Trojan Horse for other empiricist properties of connectionist thinking is discussed in Scheer (2010a:205ff). Namely the D of PDP (Parallel Distributed Processing) is modularity-offending, or gears OT towards the dissolution of modular contours, since it promotes a scrambling trope where all types of computation (phonetic, phonological, morphological and even syntactic for some) are carried out in one single constraint chamber (see §529). On the other hand, the modern version of the classical symbolic model is Jerry Fodor's modularity, which is introduced on the following pages.

600

601

Chapter 3 The modular architecture of the mind: where it comes from 1. The brain as a set of functional units: F-J Gall's early 19th century phrenology The idea that certain brain areas have localised and specific functions goes back to the inventor of phrenology, Austrian physician Franz-Joseph Gall (1758-1828), who also first proposed that the brain is the (only) organ of the mind (e.g. emotions are not located in the heart, but in the brain). (252)

Franz-Joseph Gall (1758-1828)

Phrenology holds that the mind decomposes into a number of individual mental faculties which are localised in a specific area of the brain and correlate with precise areas of the overlying skull bone. Hence in the 19th century when phrenology was popular especially in the Anglo-Saxon world, phrenologists divided the human skull into areas that represent faculties such as combativeness, wit, hope, willpower, cheerfulness or number (mathematics): maps were drawn which identified certain areas of the skull with specific faculties. Table (253) below reproduces two relevant figures.

516 Modularity Chap 3: Origins of the modular architecture of the mind (253)

American Phrenological Journal 1848

Webster's Academic dictionary 1895

Phrenology and its history are described in greater detail for example by Sabbatini (1997) and van Wyhe (2004). Marshall (2001) situates the idea that cognitive functions are independent (in mind and brain) in the history of philosophy, namely in Ancient Greek thinking. Boeckx (2010:154ff) and Nicolas (2007) discusses phrenology in the context of modern functional anatomy (the localisation and cartography of cognitive functions in the brain, see §614).

602

2. Independent faculties, their correlation with size and the skull bone The basic observation that led to the decomposition of the mind/brain into a set of basic faculties was that certain individuals are good at doing some mental activity (e.g. counting) but bad at some other (e.g. memorising), and that the distribution of acuity concerning various faculties over humans cannot be predicted. Given this inter-individual independence of faculties, then, the conclusion is that they must be independent units – mentally and in the brain.

Faculty psychology married with computation theory 517

Based on this observation, Gall promoted the idea that there is a correlation between the acuity of a given faculty and its physiological size – and that in addition the latter corresponds to a predictable zone on the skull bone. That is, the greater the acuity for a given faculty, the bigger the corresponding area of the brain, whose size is directly proportional to a corresponding zone of the skull ("the skull fits the brain like a glove fits the hand"). Hence somebody with a large number area will be good at mathematics, but somebody with a small moral area will be of little morality etc. Predictably enough, this alleged skull-to-mental-faculty relation was misused for racist purposes and the apology of this or that ideology or belief. This kind of instrumentalisation included the superiority of the white race in colonial 19th century, but also "domestic" issues: on the grounds of phrenology, the Irish were argued to be close to the Cro-Magnon man and thus to have links with the "Africinoid" races (van Wyhe no year). This was the more tempting as the mental faculties that phrenologists focused on concerned higher cognitive functions such as personality and character, rather than lower functions such as perceptual systems etc. For these reasons, phrenology has been largely discredited in the 20th century. Gall and 19th century phrenologists of course were wrong in asserting that the size of the area of the brain that accommodates a mental faculty correlates with the performance of this faculty in any way. And it is also not true, of course, that mental faculties correlate in any way with areas of the skull bone. Gall's central tenet, however, has received massive empirical support in modern times: today it is an established cognitive and neurobiological fact that all areas of the brain do not perform all tasks: some are specialised in doing this, others in doing that – mental and neural structure has a functional architecture. The modern incarnation of this idea was formulated in terms of faculty psychology by Jerry Fodor (1983 et passim) before neuroimaging systems that reveal gross functional areas in the brain were available.

603

3. Faculty psychology married with computation theory (von Neumann - Turing) In Fodor's work, the idea that the human cognitive system is composed of several functionally and computationally autonomous sub-systems is married with the model of computation that was developed in the 40s, and which is the basis of Artificial Intelligence, Chomskian linguistics, the

518 Modularity Chap 3: Origins of the modular architecture of the mind standard model of Cognitive Science and various strands of mathematics, logic and the Humanities (see Gardner 1985 for a overview). Also, the technology of modern micro-computers is based on this work (hardware and software alike: structured modular programming), whose most prominent figures are the British mathematician Alan Turing (1912-1954) and the Hungarian-American mathematician John von Neumann (1903-1957). Overview literature of what is commonly called the von Neumann-Turing model and its application to Cognitive Science includes Herken (1995), Clapin (2002), Pylyshyn (1984, 1989a,b), Haugeland 1989:133ff); an introduction from the linguistic perspective is provided by Boeckx (2010:33ff). An early argument in favour of a computationally based modular conception was made by the founder of Computational Neuroscience, David Marr, on the grounds of vision (Marr 1982). Central to this theory of computation are the following claims: computation is based on distinct short-term (working) and long-term memory (this is the essence of the Universal Turing/von Neumann Machine, see §596), it is serial (i.e. step-by-step, rather than parallel), it is based on a symbolic, pre-determined and machine-specific language, and it is organised in functional units that are devised to solve a specific problem – the modules. The symbolic issue was already discussed in §589. Fodor's modular organisation of the mind/brain thus unites the two strands, computation theory and faculty psychology (which roots in phrenology and related traditions in psychology, see Posner 1981).

604

605

Chapter 4 The modular architecture of the mind: how it works

1. Higher and lower cognitive functions, modules and the central system Given Gall's idea that the mind is a set of functional sub-systems, the question arises what exactly counts as a module: how many different faculties are there, how coarse-grained and of what type are they, what kind of evidence can be brought to bear in order to identify them and how can they be delineated? Regarding the problem of functional taxonomy, Gall himself already argued against very broad abilities (whose operations may apply to different domains) such as intellect, acuity, volition, attention, judgement or memory (Fodor 1983 calls these horizontal faculties). Just like instinct (of birds to sing etc.), these abilities do not have a specific neurological localisation in Fodor's model. Rather, they emerge from the conjugation of more fine-grained abilities (which Fodor 1983 calls vertical faculties) such as vision, audition or number processing. A range of this kind of problemsolving entities (which are known as lower cognitive functions in psychology) are thus the construction workers of higher cognitive abilities such as moral and social judgement, which Fodor (1983) calls the central system. Table (254) below depicts the relationship between Fodor's central system and modules (how modules communicate with other modules, and with the central system, is a central issue that is discussed in §650 and also in Vol.2). (254) Fodor (1983): modules and the central system that they inform number sense

module Y

vision

central system

audition

module X

language

520 Modularity Chap 4: The modular architecture of the mind, how it works The central system is (or rather: the central systems are) informed by the work that is done by modules, but it is not a module itself. Namely, higher cognitive abilities that are the result of the central system lack the two main characteristics that define modules: they are not domain specific, and they are not (informationally) encapsulated (more on these notions shortly). Also, they try "to make sense" of the information that is submitted to them and hence may be goal oriented. Unlike central systems, modules are "dummy" and non-teleological: they have do decisional latitude, do not make or evaluate hypotheses and hence do not try to achieve any goal: they are simple computational systems which calculate a predictable output on the grounds of a given input ("input systems" which are stimulus-driven). They provide evidence that the central system needs in order to manage hypotheses, but are entirely insensitive to whatever the central system may "ask" them to do. Modules do their job fast, well, they are very reliable, and they are mandatory: humans cannot decide to switch them off. For example, visual stimulus always ends up as a three dimensional picture, language is always processed as such and not as noise, and subjects cannot help identifying what kind of surface their fingers are running over. Prime examples of lower cognitive functions that qualify as modules have already been mentioned: audition, vision, number sense. At least the two former are no doubt genetically endowed. Being innate is thus another property of modules. Fodor (1983:44) grants modular status to "the perceptual faculties plus language" – an interesting definition. Following the Fodorian track, general introductions of the modular approach to the mind include Stillings et al. (1995:16ff), Segal (1996), Cattell (2006) and Samuels et al.(1999:85ff). Following Marr (1982), vision is certainly the best studied cognitive faculty which indeed provides pervasive evidence for a modular architecture of the mind/brain (e.g. the papers on vision in Garfield (ed.) 1987:325ff, Stillings et al. 1995:461ff).

606

2. How much of the mind is modular?

607

2.1. Peripheral vs. massive modularity: is there a non-modular core? Fodor (1983) is pessimistic about our ability to understand how central systems work: he assumes that they are resistant to scientific theorising and ultimately to human understanding because they cannot be appraised

How much of the mind is modular? 521

through the modular prism: "the more global […] a cognitive process is, the less anybody understands it" (Fodor 1983:107). A different line of thought expands the modular architecture to central systems as well. Pinker (1997) and Plotkin (1998) are the most prominent figures of this direction: according to them, all mental processes are computations. Smith (2002, 2003) also questions the strict separation between modules and non-modular central systems, and Smith & Tsimpli (1995:164ff, 1999) are optimistic regarding our chances to understand how central systems work: they craft the notion of quasi-modules, which they believe higher cognitive functions are produced by. The volume edited by Hirschfeld & Gelman (eds.) (1994) also contains a number of papers that argue for the domain specificity of higher cognitive functions such as social categories, cultural representations and emotions (domain specificity is a central property of modules, see §611 below). Following the same track, Higginbotham (1987:129f) argues that language is a central system and modular. Sperber (1994, 2001) also promotes the modular character of central systems: according to his massive modularity, the brain is modular through and through. Fodor (1987:27) calls this the "modularity thesis gone mad": he has always held the view that not all cognitive functions are modular in nature. Fodor (1987) for example is a defence of this position. The article opens like this: "There are, it seems to me, two interesting ideas about modularity. The first is the idea that some of our cognitive faculties are modular. The second is the idea that some of our cognitive faculties are not." More recently, Fodor (2000) is a book entirely devoted to the question whether all or only part of the cognitive system is based on a modular architecture. The book is an exegesis and a refutation of Pinker's and Plotkin's "New Synthesis Psychology" (which Fodor calls rationalist psychology, see also Fodor 1998). Gerrans (2002) provides an informed overview of the debate regarding the articulation of modules with central systems. 608

2.2. Is the central system impenetrable for human intelligence? What really is behind this debate is (against a possible prima facie impression) a categorical, rather than a gradual distinction – one that has deep philosophical roots and far-reaching consequences. That is, the modular paradigm falls into two opposing camps, one holding up Descartes' position that the mind, or at least some of it (the central system in Fodor's terms), is beyond what can be understood by human intelligence and will always

522 Modularity Chap 4: The modular architecture of the mind, how it works remain an impenetrable mystery (the soul is of course lurking behind the mind of Descarte's mind-body dichotomy); by contrast, the other camp makes no difference between lower and higher cognitive functions, which are both the result of modular activity. We have seen that the former view is defended by Fodor (1998, 2000), but also by Chomsky in linguistics (e.g. Chomsky 1984:6f, 23f, Chomsky 1995b:2f, chapter 4 of Chomsky 1975 is called "Problems and mysteries in the study of human language"). Fodor's and Chomsky's position blocks any inquiry into how the mind really works (all of the mind for Descartes, just a subset of it, the central system, for Fodor/Chomsky) before it has even started: don't try to find out how it works, you will fail anyway. This has direct consequences for the dialogue with the implementational level (see §614): only a subset of the mind may be mapped onto neuro-biology – the central system is not based on any neuro-biological activity, or at least will humans never be able to understand what the relationship is.

609

2.3. Is the mind (are modules) the result of Darwinian adaptation? The latter position, i.e. where all cognitive functions are in principle accessible to human intelligence and must ultimately be able to be mapped onto neurobiology, is what Fodor calls rationalist psychology. In other quarters, it is called evolutionary psychology in recognition of the fact that it is intimately interwoven with the Darwinian perspective. Pinker (1997) and Plotkin (1998) hold that the mind, like the brain and all other properties of living beings, is the result of an adaptive evolution which was marshalled by selectional pressure over millions of years. Obviously, if all is the result of environment-driven adaptation, no part of the mind can stand aside. Which means, viewed from the other camp, that Fodor and Chomsky must deny the idea that all of the mind is the result of Darwinian selection. This is precisely what they do in the biolinguistic programme: the controversy between Hauser et al. (2002) (also Fitch et al. 2005) and Pinker & Jackendoff (2005a,b) is about this issue. Hauser et al. (2002) argue that the FLN (Faculty of Language in the Narrow sense), i.e. what really makes language distinct and unique (with respect to other cognitive functions), boils down to recursion (of morphosyntax) and the ability to talk to interpretational systems (phonology and semantics), that is to Merge and Phase. They also hold that the FLN is the only property of language that could not possibly be the result of an (adap-

Core modular properties 523

tive) evolution based on an animal ancestor: the FLN is given (more on this debate in §633). This claim lies at the heart of the biolinguistic programme (where phonology and semantics for example are not specifically human, see §639) and is further developed with specific attention for phonology by Samuels (2009a,b). On the other hand, the general viewpoint of evolutionary psychology on the mind is exposed by Cosmides & Tooby (1992a, 1994), Barkow et al. (1992). Samuels et al. (1999) offer a valuable digest of the debate between peripheral (Fodor/Chomsky) and massive (evolutionary psychology) modularity on the backdrop of the opposition between what they call Chomskian and Darwinian modules. Even though based on a non-evolutionary perspective, Sperber (1994, 2001), Smith (2002, 2003) and Smith & Tsimpli (1995:164ff, 1999) go along with the Darwinian party.

610

3. Core modular properties

611

3.1. Domain specificity Modules are computational units that are devised for just one highly specific task. Therefore the symbolic vocabulary that they work with is as specific as their task: the input, its transformation by computation and the output are written in a specific vocabulary. A module can only understand its own vocabulary: whatever information is submitted that is not written in the specific symbol code of the module is uninterpretable: it is treated as noise and simply ignored. For example, the visual module can only take visual stimulus as an input. It will ignore any auditive or other alien information. Arguments for domain specificity come from various fields, including neuropsychology, computational theory and cognitive evolution (Gerrans 2002:261 provides an overview, see in particular Cosmides & Tooby 1992a). Hirschfeld & Gelman (eds.) (1994) provide an overview of domain specificity and the kind of domains that can be isolated (which include higher cognitive functions such as social categories, culture-specific representations and emotions); Fodor (2000:58ff) discusses the various ways in which domain specificity has been used. Domain specificity will be put to use in §640 in order to identify the grammar-internal modular architecture.

524 Modularity Chap 4: The modular architecture of the mind, how it works 612

3.2. Informational encapsulation Modules are also (informationally) encapsulated, which means that during the computation performed, they do not need and cannot take into account anything that was not present in the input. That is, once the input is defined and computation has begun, nothing can alter the course of events, and the output is produced in complete disregard of any module-external information such as high-level expectations, beliefs (coming from the central system), memory, inference and attention or results of other modules.153 Conversely, modules are unable to communicate any intermediate result of their work: transmission to other modules or to the central system is only possible once the computation is completed. In sum, modules are autistic (Fodor 2000:62ff, Gerrans 2002 and Smith & Tsimpli 1995:30f provide a concise introduction to encapsulation). The effect (and hence existence) of encapsulation is typically shown on the grounds of optical illusions. Under (255) below appear a number of well-known cases, which all demonstrate that humans are "fooled" by their visual system even if they know beforehand that what they "see" is not true: there is no way to willingly marshal vision according to prior knowledge of the central system, to some desire or presupposition. Vision does whatever it does without asking any other cognitive system, and even against the will of the subject: no other cognitive system, modular or central, can "break into" vision in order to change its course once computation has begun.154

153

154

This of course does not withstand the existence of networks of modules or of "loops" whereby the result achieved by a given module serves as the input of several other modules and eventually, enriched with additional information, is pulled several times through the same module. The reason and genesis of the illusions are secondary for the argument. Also note that the effect is the same for all humans (who are subject to the illusion: some are not), i.e. perfectly independent of culture, language, age, social parameters and so forth.

Core modular properties 525 (255)

Zöllner: all long lines are parallel

Müller-Lyer lines are equal in length

Poggendorff the lower line on the righthand side of the rectangle is the same (straight) line as the one on the lefthand side

Hering vertical lines are parallel

Encapsulation has been challenged on the grounds of the possibly non-encapsulated communication between the central system and modules whereby the former affects ongoing modular computation. Arguments to this end have been made on the connectionist side (e.g. Elman 1994), but

526 Modularity Chap 4: The modular architecture of the mind, how it works also in the quarters of developmental psychology (Karmiloff-Smith 1998). Ongoing debate is reviewed by Gerrans (2002), who argues in favour of encapsulation. The syntactic application of encapsulation is Chomsky's (1995a:228) inclusiveness, on which more in §648. Encapsulation will also play a role in §621 when the modular status of grammar and its subsystems is discussed.

613

3.3. Summary: how to identify a module A module is thus a hard-wired and genetically determined computational unit that builds on a fixed and localisable neural structure; it is domain specific (i.e. content-based), autonomous, automatic, mandatory, stimulusdriven and insensitive to central cognitive goals. Segal (1996:145) provides an informed and concise overview of the modular idea in its various incarnations. His list of core properties contains nine items, which are shown under (256) below. (256) core properties of cognitive (Fodorian) modules according to Segal (1996:145) a. domain specificity b. informational encapsulation c. obligatory filtering d. fast speed e. shallow outputs f. limited inaccessibility g. characteristic ontogeny h. dedicated neural architecture i. characteristic patterns of breakdown

Crucially for linguistics (as we will see below), a module is designed for a special purpose and can only work with the specific vocabulary associated – all the rest is noise: modules "solve a very restricted class of problems, and the information it can use to solve them with is proprietary" (Fodor 1998). Now recall Chomsky & Halle's (1968) description of the phonological rule system that was already quoted in §84: it is quite surprising an anticipation of what Fodorian modules will look like 15 years later.

Specialised neurons and neural localisation of cognitive functions 527 (257) "The rules of the grammar operate in a mechanical fashion; one may think of them as instructions that might be given to a mindless robot, incapable of exercising any judgment or imagination in their application. Any ambiguity or inexplicitness in the statement of rules must in principle be eliminated, since the receiver of the instructions is assumed to be incapable of using intelligence to fill in gaps or to correct errors." Chomsky & Halle (1968:60)

Given these core modular properties, a question is how modules are practically delineated within the host of cognitive functions. The typical answer is domain specificity: a computation that builds on heterogeneous primitive units cannot be done in one and the same module. As we will see below (§622), there is serious debate in linguistics regarding which entities (sub-disciplines) exactly are identical or distinct computational systems (ongoing controversy namely concerns morphology and syntax, see §537). In this situation, the guiding light will be to look at which kind of vocabulary is processed on each side, and whether it is the same. In case it is not, the two entities cannot be incarnations of the same module. §618 introduces yet another way of detecting modules, (double) dissociation, which may be called external in comparison to the internal handle that is offered by domain specificity. While the latter requires only the inspection of linguistic properties (the vocabulary used), (double) dissociation requires the examination of speakers that experience significant cognitive and/or brain damage. Overview literature regarding the general properties of modules includes Segal (1996), Pinker (1997), Plotkin (1998), Sperber (2001), Gerrans (2002), Jackendoff (2002:218ff), Smith (2002, 2003) and Fodor (2000). Cosmides & Tooby (1992b:93ff) provide a historical overview of the modular idea from the psychologist's perspective.

614

4. Specialised neurons and neural localisation of cognitive functions

615

4.1. Mind-brain relationship The modular approach to the mind/brain has a number of implications in philosophy, neurobiology, psychology, linguistics and other areas of knowledge. Chomsky (2002:45ff) provides a historically oriented overview, arguing that Cognitive Science, where the mind-brain dichotomy is still puzzling, may learn a lot from the history of adult sciences like physics and chemistry, which had to face similar problems a while ago.

528 Modularity Chap 4: The modular architecture of the mind, how it works Below two testing grounds that have been prominently explored in the literature are given closer attention: on the one hand, the prediction that functional specialisation of the mind implies the existence of specialised and localisable suites of neurons in the brain. On the other hand, the prediction that if there are functionally specialised modules and neurons, cognitive and/or brain damage must be able to "plug them out" without this affecting other faculties (§618). Let us first take a closer look at the former. It was mentioned in §592 that the relationship between symbolic representations and real-world objects, in physics, chemistry and biology as much as in Cognitive Science, is intricate and dialectic, but in any case non-arbitrary. It is an established neurobiological fact that all neurons do not do everything: suites of neurons may be specialised (although not necessarily to do one single thing). For example, that the Broca and Wernicke areas of the brain are specifically related to the processing of language is known since the 19th century (Broca 1861, see e.g. Loevenbruck et al. 2005 for a modern investigation) and has even found its way into popular science. Posner (2001) and Nicolas (2007) offer a historical survey of the efforts that are made in order to localise cognitive functions in the brain (see also Boeckx 2010:149ff regarding language). The quote from Bastian's (1880) book "The brain as an organ of mind" that is reported in §597 is quite instructive in this respect.

616

4.2. Functional anatomy: the existence of specialised and localisable (suites of) neurons is undisputed Gerrans (2002:259) points out that the symbolic issue which was discussed in §589 is entirely independent of the functional specialisation that is predicted by modularity: whether or not computation relies on symbols and hence is content-sensitive is orthogonal to the question whether the mind/brain is made of functional units. Neurobiologically speaking, the existence of functionally specialised suites of neurons provides evidence for modularity no matter what it actually is that the neurons in question process. The modern study of vision for example has produced undisputed evidence that specialised neurons exist and are localisable in the brain.

Specialised neurons and neural localisation of cognitive functions 529 (258) "Functional specialisation occurs within the neural systems on which vision depends. The visual cortex contains individual and suites of neurons specialised for detecting orientation, disparity, wavelength, velocity, direction and size. (Marr, 1982). This neuroanatomical organisation reflects the functional organisation of the visual module for good reason. Rather than involve the same region in more than one task, 'Regional specialisation, on the contrary means that columns of cells can he connected with neighbours that have related functions.' (Shallice, 1988, p. 19)." Gerrans (2002:263)

Gerrans (2002:263) goes on to point out that "clearly however, there is no one-to-one mapping from cognitive function to unique neural location" – it was mentioned earlier that a complicated and dialectic relationship is certainly what is expected. In our past and present understanding where symbolic representations and functional anatomy are still light-years away from being able to be matched, the classical strategy is to approach the problem by thinking of it in terms of levels. Since linguistic representations are much too coarse grained, and neuobiological functioning way too fine grained, intermediate levels may help to do justice to both and make our understanding of the overall situation progress. This is what connectionism set out to do in the late 80s when it was considered to be an implementational level of symbolic representations, but still ranging above neurobiology: recall from §599 that this is Smolensky's (1987) position, which he carried over into his version of OT (Smolensky & Legendre 2006). The work by David Poeppel, often in collaboration with linguists, also follows this track: here the relevant level of abstraction is the circuit, a brain structure that computes simple operations. A circuit is simple enough not to exceed what can be measured neurobiologically, but at the same time complex enough to allow for the representation of a simple linguistic process. This research programme is laid out in Poeppel & Hickok (2004), Embick & Poeppel (2005) and Poeppel & Embick (2005); relevant work includes Hickok (2004, 2007), Poeppel (2008). Boeckx (2010:158ff) echoes this line of thought and inserts it into a broader (also historical) perspective. The plasticity of the brain is another factor: suites of neurons that support a given cognitive function may "migrate" within the brain due to damage, hemispherectomy, growth of a tumour and the like. Locations of cognitive functions in the brain are thus approximative anyway (no two brains are identical), and only valid for subjects with normal development. Finally, nothing requires a cognitive function to be supported by a suite of neurons

530 Modularity Chap 4: The modular architecture of the mind, how it works where all neurons involved are adjacent: a function may be spread over several locations in the brain.

617

4.3. Some literature The section of Brain and Biology of Dupoux (ed.) (2001) offers a number of overview articles that report on how functional brain-imaging techniques (PET, fMRI, ERPs) 155 interact with classical psychology and Cognitive Science, eventually achieving the localisation of a number of cognitive functions in the brain. Posner (2001) is about language and attention; Newport et al. (2001) show how neuroscience can enlighten questions of language acquisition and the notion of critical periods (and vice-versa); Peretz (2001) uses brain imaging in order to approach auditory and emotional processing in relation with music. The section on the neurological basis of language in Banich & Mack (eds.) (2003) offers similar perspectives. In the area of phonology, the book edited by Durand & Laks (2002) contains three articles that discuss the relationship between cognitive functions and their neural implementation in the brain, namely from the modern neuro-imaging point of view. Démonet et al. (2002) hunt down anatomical and temporal indexes of the neural activities that underlie language perception (especially phonological and lexical-semantic processes); they also discuss the evolution from aphasia-based to neuro-imaging-based techniques that are and were used in the study of the physiology of cognition. Schwartz et al. (2002) discuss the relationship of perception, action control and phonology. They argue that empiricist approaches based on self-organising statistical and exemplar-based models are unsuited since they lack constraints on regulation and control, which according to the authors can only come from a perception-action link at the intersection of phonetics and phonology. Abry et al. (2002) draw arguments from aphasia, especially from a babbling-like type (which tan-tan [tãtã], the bequeathed production of Broca's patient, could illustrate), in order to localise the control of CVrecurring utterances in the non-lateral left hemisphere. These bibliographical indications are not systematic: the literature on the topic is booming, and I am not a specialist of the field. The purpose is 155

PET = Positron Emission Tomography, fMRI = functional Magnetic Resonance Imaging, ERPs = Event-Related Potentials.

Modules can be plugged out without affecting other faculties 531

just to show that there is something like Cognitive Neuroscience, a new discipline that has emerged in the recent past, which tries to achieve "the synthesis of mind and brain", as Posner (2001) puts it.

618

5. Modules can be plugged out without affecting other faculties

619

5.1. Double dissociation Let us now consider the other interesting prediction made by modular theory: so-called double dissociation is based on the functional character of modules; it opens a fairly precise empirical testing ground on the basis of cognitive and/or brain damage. If different functions are managed by different modules with no overlap (i.e. a given function is computed by one and only one module), then any module should be able to be plugged out, causing the loss of the function that it is responsible for, but leaving the rest of the system without damage. If on the other hand there are no modules and all functions are intertwined (this is the connectionist take), the damage of a particular area should impact a variety of other areas and functions. Hence showing that two cognitive functions are entirely independent is providing support for the modular architecture. The demonstration of functional independence requires so-called double dissociation (e.g. Smith 2003): given two abilities, different subjects may lose one while retaining the other and vice versa. A trivial case is blindness and deafness: some people are blind but not deaf, others are deaf but not blind. Hence it is reasonable to assume that vision and audition do neither use the same hardware (brain) nor the same software (mind).

620

5.2. Documented cases: face recognition, number sense Examples of (double) dissociation are frequently reported in the pathological literature (e.g. Karmiloff-Smith et al. 1995). One case in point is discussed by Smith (1998:9): "prosopagnosia is the sad condition which afflicted Oliver Sacks' eponymous subject. 'The man who mistook his wife for a hat'. As a result of a cerebral lesion involving the visual system he became unable to recognise faces, even though he could still identify people from their voice or smell or touch, and his conceptual abilities were unaffected." This is evidence to the end that vision is not just one undiffer-

532 Modularity Chap 4: The modular architecture of the mind, how it works entiated cognitive function. Rather, it falls into several computational systems, one of which is specialised in face recognition (also, shape and colour appear to be determined independently). On the other hand, Moscovitch et al. (1997) document the symmetric dissociation: they study a case where object vision is impaired, but face recognition is normal. Another case in point is the number sense, which appears to actually involve two separate modules: one that computes small numbers up to four or five with high precision and very rapidly (paucal); the same module also roughly guesses bigger numbers (8 items are less than 20, 20 items are about 20, not 60) (approximate). A different module manages so-called verbal counting: it operates over natural numbers and performs mathematical calculus (the four basic operations). Based on evidence from subjects with various cerebral lesions, Dehaene (1997) reports that the paucal and the approximative abilities are always associated: if a subject is impeded in one area, the other will also be affected. However, verbal counting and paucal/approximate counting are doubly dissociated: subjects who cannot cope with one may have undamaged abilities regarding the other. Dissociation is also documented for implicit and explicit knowledge by Reber & Squire (1998). Besides pathological evidence that relies on damage of the mind and/or brain, the (double) dissociation of cognitive functions may also be demonstrated on the grounds of data from development: Hermer & Spelke (1996) for example study spatial reorientation in this perspective.

621

5.3. Double dissociation of language Regarding language, Neil Smith and Ianthi-Maria Tsimpli have documented the case of Christopher over a relatively long period (Smith & Tsimpli 1991, 1995, 1999, also Smith 1998, 2002, 2003). Christopher has severe cognitive deficits (he cannot look after himself, has trouble to find his way around, poor hand-to-eye coordination etc.), but an extraordinary talent for the acquisition and use of language. That is, Christopher is fluent in some 15 or 20 languages, in which he can construct morphologically and syntactically well-formed sentences after minimal exposure. His enhanced talent for language and languages falls into the same category as other cases of the so-called savant syndrome: brain-damaged savants that are otherwise mentally handicapped and typically autistic are documented with unbelievable skills for calendrical calculation (ability to tell instantly on which day of the week any date in past, present or future centuries falls) or

Modules can be plugged out without affecting other faculties 533

various artistic talents (e.g. musicians who can play complex passages after a single hearing). Smith & Tsimpli (1995) explain the relevance of Christopher's case (and other similar cases) for establishing language as a module. (259) "Although no one else has been reported as displaying the multi-lingual prowess that Christopher does, these cases illustrate the same dissociation between linguistic and general cognitive abilities as is exhibited by such individuals as Laura […], by Williams Syndrome children […], by 'chatterbox' children […], and by hyperlexics […], all of whom have great linguistic ability in the presence of severe cognitive deficits. Examples in the opposite direction – cases of people with impaired language in the presence of normal intellectual ability – are provided by some deaf people, some aphasics, and by those suffering from SLI (Specific Language Impairment), where brain damage (in some cases genetically caused) occasions a language deficit independently of the rest of the cognitive profile […]. The existence of these varied conditions provides a classical example of double dissociation: language can be impaired in someone of otherwise normal intelligence, and – more surprisingly – someone with intelligence impaired by brain damage may none the less have normal, or even enhanced, linguistic ability." Smith & Tsimpli (1995:3, emphasis in original)

The basic argument for the (double) dissociation of language is thus the fact that in the case of Christopher the language faculty may work well (actually better than normal) in absence of control by "general intelligence": Smith & Tsimpli (1991:325) write that "it is clear that his talent exists in the absence of the normal 'general intelligence' one might expect to find associated with multi-lingualism." They also draw on encapsulation of Christopher's language performance, which – as predicted by Fodorian modularity – is independent from general purpose considerations or global goals: the general goal to produce non-nonsensical translations is not able to "break into" Christopher's language performance while translating.

534 Modularity Chap 4: The modular architecture of the mind, how it works (260) "In the present context informational encapsulation would mean that Christopher's linguistic ability was independent of his general cognition and could operate in the absence of 'central' control. His method of translating makes this extremely plausible. When asked to translate, he starts instantly and proceeds word by word rather like an automaton. If he is asked to slow down and mull over the meaning of the whole passage in an attempt to improve his performance, he shows visible signs of distress and professes himself incapable of doing any such thing. Moreover his equanimity at producing nonsensical translations indicates either that he is incapable of discerning such nonsense, or that his linguistic (morpho-syntactic) system operates in divorce from any semantic or pragmatic control." Smith & Tsimpli (1991:325)

This is evidence to the end that language as such is a module: it appears to be doubly dissociated. That is, it may be "plugged in or out" without affecting the (non-)deficient rest of cognitive abilities. Williams Syndrome also provides strong evidence to this end (see Tager-Flusberg & Sullivan 2000). Williams Syndrome "is a rare genetic disorder resulting in certain characteristic facial features and physical problems as well as a unique and particularly striking cognitive profile. Subjects are retarded, with an average IQ of around 50. They are also particularly impaired with respect to arithmetical and visual-spatial abilities. However they exhibit an unusually high level of linguistic ability, with a particular penchant for sophisticated and unusual vocabulary items" (Segal 1996:154, who offers a concise general introduction to Williams Syndrome). This kind of evidence that is typically used for the (double) dissociation of language form other cognitive functions has also been given other interpretations: Bates (1994) argues for the innateness and the localization of language in the brain, but against its domain specificity. The argument is based on the fact that the dissociations observed are never one hundred percent waterproof: linguistic deficits are accompanied by minor nonlinguistic impairment as well, and non-linguistic developmental disorders such as Williams Syndrome have also some impact on language. In any event, modularity and Chomsky's conception of a language organ (§627, e.g. Chomsky 1995b, Chomsky 2000b:106ff) predict that language as such is a plug-in unit in the concert of cognitive functions. The question, then, is whether it may be decomposed into yet smaller units. The modular structure of language itself is hinted at by the quote under (260), where morpho-syntax appears to be estranged from semantic and pragmatic control. This is what the following pages are about.

622

Chapter 5 Modularity of and in language, related systems

623

1. Modularity in the early days of generative grammar: 50s-60s

624

1.1. A spearhead of the cognitive revolution of the 50s in language Language has always played a prominent role in the development of Cognitive Science: it was a prime candidate for the implementation of the von Neumann-Turing programme (see §603, Gardner 1985:182ff) that really started to penetrate modern science in the 50s (e.g. Gardner 1985:28ff). Noam Chomsky and generative linguistics were the spearhead of the computational conception in the realm of language (e.g. Cosmides & Tooby 1992b:93ff, Chomsky 1993b). In 1972, computer scientists Allen Newell and Herbert Simon recall the 50s and the inception of Cognitive Science. (261) "Within the last dozen years a general change in scientific outlook has occurred, consonant with the point of view represented here. One can date the change roughly from 1956: in psychology, by the appearance of Bruner, Goodnow, and Austin's Study of Thinking and George Miller's 'The magical number seven'; in linguistics, by Noam Chomsky's 'Three models of language'; and in computer science, by our own paper on the Logic Theory Machine." (Newell & Simon 1972:4, emphasis in original)

Also, Chomsky has always considered that the study of language is undissociable from the study of mind: cognitive realism is a founding statement of the generative enterprise – since Chomsky (1959) it constitutes the fraction line with (certain types of) structuralism (in linguistics) and behaviourism (in psychology and learning theory). In this context, modularity is a necessary ingredient of the generative enterprise, both regarding language in the concert of other cognitive functions and its internal organisation. The former area may be illustrated by the following quote from Chomsky (1975) (among a host of others, e.g. Chomsky 1972, 1980, 1984, 1988, 1993b etc., Higginbotham 1987 and Hirschfeld & Gelman 1994:5ff provide historical discussion).

536 Modularity Chap 5: Modularity of and in language, related systems (262) "[T]he position we are now considering postulates that this faculty [the language faculty] does exist, with a physical realization yet to be discovered, and places it within the system of mental faculties in a fixed way. Some may regard this picture as overly complex, but the idea that the system of cognitive structures must be far more simple than the little finger does not have very much to recommend it. The place of the language faculty within cognitive capacity is a matter for discovery, not stipulation. The same is true for the place of grammar within the system of acquired cognitive structures. My own, quite tentative, belief is that there is an autonomous system of formal grammar, determined in principle by the language faculty and its component UG. This formal grammar generates abstract structures that are associated with "logical forms" (in a sense of this term to which I will return) by further principles of grammar. But beyond this, it may well be impossible to distinguish sharply between linguistic and nonlinguistic components of knowledge and belief. Thus an actual language may result only from the interaction of several mental faculties, one being the faculty of language. There may be no concrete specimens of which we can say, these are solely the product of the language faculty; and no specific acts that result solely from the exercise of linguistic functions." Chomsky (1975:43)

Another point of interest is that language has always been considered a prime candidate for modularity – more than other cognitive systems – in the debate regarding which faculties exactly are modular, and which ones are not, i.e. result from the activity of Fodorian central systems (see the question "how much of the mind is modular?" in §606). Smith & Tsimpli (1995:30) for example distinguish between perceptual and cognitive systems, where the former identify as "the sensorium plus language", while the latter are Fodor's central systems (fixation of belief, thought, storing knowledge). On this view, language is on a par with vision, audition, taste, smell and the sense of touch. The intimate relationship of language and modular theory is also reflected by the fact that Fodor's (1983) seminal book has emerged from a class on cognitive theory that Fodor co-taught with Chomsky in fall 1980.

625

1.2. LSLT: language is made of modules (levels), a concatenation algebra and interfaces The internal organisation of language in terms of distinct computational and functional systems that are specialised in morpho-syntax, phonology and semantics is made explicit in one of the earliest generative documents:

Modularity in the early days of generative grammar: 50s-60s 537

Chomsky's Logical Structure of Linguistic Theory, published only in 1975, is based on a 1955-56 manuscript (one chapter of which was eventually outsourced to make Chomsky's Ph.D thesis). In the structuralist environment of the time, Chomsky holds that the basic units that language is made of are levels (see §61 on structuralist Level Independence). (263) "[T]he theory of linguistic structure [is], essentially, the abstract study of 'levels of representation'. […] A linguistic level is a system L in which we construct unidimensional representations of utterances. Thus a level has a certain fixed and finite 'alphabet' of elements, which we will call its 'primes.' Given two primes of L we can form a new element of L by an operation called 'concatenation,' symbolized by the arch ^. Thus if a and b are (not necessarily distinct) primes of L, we can form a^b and b^a as new elements of L. Concatenation is essentially the process of spelling, where primes are taken as letters. Given the element a^b and the prime c, we can form a new element (a^b)^c." Chomsky (195556:105)

Levels are thus computational systems, which operate each on a specific vocabulary and produce suites of vocabulary items associated with a hierarchical structure that reflects their concatenative history. This is precisely the description of a Fodorian module: a domain specific computational unit that works only on its own proprietary vocabulary (§611). And the quote also describes what is known today as the basic minimalist engine, Merge. Viewed from this perspective, phrase structure rules were an ephemeral interlude (see also Lasnik & Lohndal 2009:46 on the 50s-roots on Merge). LSLT is at variance in an interesting way with the inverted T model that emerges in the 60s and represents the generative architecture of grammar up to the present day (§86). Indeed, LSLT does not make any difference between concatenative and interpretative systems: all levels carry out concatenation. In phonology, Chomsky holds that phonemes are concatenated. (264) "[T]he level of phonemes for English has among its primes the symbols p, i, n, which can be concatenated to form the string p^i^n, which is the Pmmarker of 'pin'." Chomsky (1955-56:66)

The number and nature of the levels that Chomsky assumes also witnesses the structuralist environment.

538 Modularity Chap 5: Modularity of and in language, related systems (265) "We will find it necessary to distinguish at least the following levels for linguistic description: phonemes (Pm), morphemes (M), words (W), syntactic categories (C), phrase structure (P), and transformations (T). The grammar must indicate the structure of each utterance on each of these levels." Chomsky (1955-56:66)

Finally, the existence of distinct computational systems that produce hierarchised suites of distinct vocabulary items requires interface mechanisms for inter-level communication. This need for translation, also a perfectly modern and modular concern, is made explicit by Chomsky. (266) "A linguistic level is not determined completely by the statement that it is a concatenation algebra. We must also specify its relations to other levels (i.e., the conditions of compatibility between levels)." Chomsky (1955-56:106)

In sum, all basic ingredients of Fodorian modularity are already present in one of the earliest generative documents: LSLT defines a number of modules that language is made of (many more than what the inverted T will recognise, though); these modules are input-output systems and have a "concatenation algebra", which in modern terms means that they are domain specific and carry out Merge (all modules, not just morpho-syntax, though have the concatenative privilege); finally, LSLT modules have the ability to talk to other modules, Phase in modern vocabulary. 626

1.3. Modularity on its way: from LSLT to Aspects and SPE Newmeyer (1986:172f, 198f) provides some elements of how the modular conception that was laid out in LSLT was progressively introduced and worked out in generative grammar. He mentions Chomsky & Miller (1963) as an early source for the explicit statement that syntax and phonology are distinct computational (input-output) systems. (267) "We regard grammar as having two fundamental components, a syntactic component of the kind we have already described and a phonological component to which we now briefly turn our attention. […] The phonological component embodies those processes that determine the phonetic shape of an utterance, given the morphemic content and general syntactic structure of this utterance. […] As distinct from the syntactic component, it plays no part in the formulation of of new utterances but merely assigns to them a phonetic shape." Chomsky & Miller (1963:306f, emphasis in original)

Modularity implies biology and innateness: the language organ 539 "The phonological component can be thought of as an input-output device that accepts a terminal string with a labelled bracketing and codes it as a phonetic representation." Chomsky & Miller (1963:308)

The inverted T model was formally introduced in Aspects (Chomsky 1965:15ff) (see §§84,86); it fixes the existence of (at least) three independent computational systems (morpho-syntax, phonology, semantics) and to date represents the bottom line of the generative architecture of grammar. The quote from SPE under (257) (§613) also shows that the generative conception of the internal structure of grammar was already modular in the 60s, even though of course this particular vocabulary item was not used. Since 1965, Chomsky has been constantly explicit on the modular character of language (e.g. Chomsky (1972, 1975, 1984). An early source is also Chomsky (1965 [2006]), a chapter included in later editions of Chomsky (1972), but which Chomsky explains in the preface to the second edition was actually written in 1965. In this text, the units of the inverted T model are referred to as components (syntactic, phonological, semantic). The further development of modularity within the inverted T model in the 80s is discussed in §628 below.

627

2. Modularity implies biology and innateness: the language organ A consequence of the view that language is a module is its genetic determinacy: recall that modules, among other things, have the property of being genetically endowed (§613). This is where Chomsky's biological conception of language – known under the header of the language organ and more recently the biolinguistic programme (§§609,639) – comes from. On this view, the neural existence of the language module and the genetic endowment for its inception in the growth of young humans gives rise to an organ just like the liver, the heart or other parts of the human body that are specialised in some particular task: cleaning or pumping of blood etc. The only peculiarity of the language organ, then, is to be localised in the brain, rather than elsewhere in the body.156 156

In psycholinguistic quarters that were a priori Chomsky-friendly, the biological conception of language was anything but popular in the 80s: people refused even to think about an eventual neural correlate of cognitive functions. Dehaene et al. (2001) report on this pre-brain imaging period with the follow-

540 Modularity Chap 5: Modularity of and in language, related systems In this perspective, linguistics is "that part of psychology that focuses its attention on one specific cognitive domain and one faculty of mind, the language faculty" (Chomsky 1980:4). Therefore, "we may regard the language capacity virtually as we would a physical organ of the body and can investigate the principles of its organization, functioning, and development in the individual and in the species" (Chomsky 1980:185) (also e.g. Chomsky 1975:11: "the idea of regarding the growth of language as analogous to the development of a bodily organ is thus quite natural and plausible. It is fair to ask why the empiricist belief to the contrary has had such appeal to the modern temper"). The modern offspring of this genuinely generative tradition is Chomsky's biolinguistic program (e.g. Hauser et al. 2002, Chomsky 2005, on which more in §639; see Jenkins 2000 for an overview). Together with UG, the language organ is probably the best-known property of generative grammar outside of its own quarters. It has become a buzz-word in popular scientific texts and neighbouring disciplines, foremost philosophy and psychology where its validity is challenged and provokes much discussion. This debate goes far beyond the scope of the present book. Relevant literature from both sides includes Stich (1972), Katz (1984), Devitt & Sterelny (1989), Kasher (1991), Fodor (1981), Chomsky (2002); Wauquier (2005:175ff) provides an informed overview.

628

3. Grammar itself is made of modules: GB-subtheories and their (questionable) status as cognitive modules

629

3.1. The inverted T is the baseline since the 60s If language is one piece of the modular architecture of mind, the question arises whether there is only one single computational unit that carries out all grammatical calculation, or whether there are several linguistic modules. In turn, if language is made of distinct computational systems, the question is how many linguistic modules there are, and how exactly they are delineated. The early generative literature on language-internal modularity was already discussed in §623: the bottom line that is condensed in Aspects ing quote from Jacques Mehler, which sums up Mehler et al. (1984): "For all I know, language perception might be going on in the brain. but my research would not be affected if it was found to be occurring in the left pinky." Dehaene et al. (2001) and the section on Brain and Biology of Dupoux (ed.) (2001) that they introduce then show how things have changed today.

Grammar itself is made of modules: are GB-subtheories cognitive modules? 541

(Chomsky 1965:15ff) grants modular status in terms of a an independent and domain specific computational system to the three components of the inverted T model: morpho-syntax, phonology and semantics. These represent the core of "internal" linguistic activity and are related to nonlinguistic cognitive activity by at least a conceptual device (which matches real-world objects and concepts with linguistics items) and pragmatics. Or, in other words, the interplay of the three "internal" components is called grammar, while their exchange with grammar-external cognitive activity produces language (see Newmeyer 1986:172ff for a historical description and the state of the art in early GB).

630

3.2. GB-subtheories are presented as modules, but insulated from the cognitive context Modularity was embodied in the inverted T, but as such not much of an issue until the 80s when the principles and parameters framework was introduced. In his Pisa lectures, Chomsky (1981) divides syntax into six autonomous subsystems (bounding theory, government theory, theta theory, binding theory, case theory, control theory), which he refers to as modules. The modular structure of GB is the major innovation of the new theory. Surprisingly enough, though, Chomsky studies language-internal (or rather: syntax-internal) modularity in absence of any reference to the broader claim that the entire human cognitive system is modular in kind. This is to be appraised in the context of Fodor (1983), the major reference of the modular theory of mind that was in the making when Chomsky wrote, and also of the fact that Fodor's book grew out of lecture notes of a class on contemporary cognitive theory that Jerry Fodor co-taught with Noam Chomsky in fall 1980 (see the acknowledgements in Fodor 1983). While language plays an important role in Fodor (1983), the idea that the language faculty and/or its sub-components are just specific cases of cognitive modules that are embedded in a broader modular architecture is absent from Chomsky (1981). Chomsky does not mention the fact that modules, whether they compute grammatical or other functions, have a number of properties (domain specificity, encapsulation, dissociation) which allow us to delineate their contours. The result is not really in line with the history of generative grammar and Chomsky's personal contribution to Cognitive Science: when generative linguists think of grammar-internal modularity, what they typically have in mind is an exclusively linguistic horizon (more on that in §636).

542 Modularity Chap 5: Modularity of and in language, related systems One may be inclined to believe that this is a good deal of the reason why (real cognitive) modularity has played virtually no role within generative grammar, even when it would have been decisive (while it has always been an argument for the definition of grammar as opposed to other cognitive functions). I have come across two obvious cases where the absence of the modular argument is really surprising, looked at from the outside: interactionism and direct syntax (see §§222,414 and the summary in §36).

631

3.3. Chomsky (1981): subcomponents (inverted T) vs. subsystems (theta theory etc.) In the first chapter of the Pisa lectures entitled "Outline of the theory of core grammar", Chomsky (1981) first exposes the classical position according to which the three extremities of the inverted T are modules. (268) "The theory of UG must therefore specify the properties of (at least) three systems of representation - S-structure, PF, LF - and of three systems of rules: the rules of the syntactic component generating S-structures, the rules of the PF-component mapping S-structures to PF, and the rules of the LFcomponent mapping S-structure to LF. Each expression of the language determined by the grammar is assigned representations at these three levels, among others." Chomsky (1981:4)

He then introduces the distinction between subcomponents of the rule system of grammar, and subsystems of principles. The relevant quote is provided under (269) below. Subcomponents are the members of the inverted T, i.e. the default candidates for modular status, while subsystems are the subtheories of GB, which the book sets out to present. Chomsky (1982:4ff) is analogous, the only noteworthy difference being the fact that X-bar theory is added to the six subsystems that are mentioned under (269). The characterisation of morpho-syntax (and semantics) as a network of interacting sub-devices is the central innovation of GB, and this is why the focus of Chomsky (1981) of course is on these subsystems. They are the new perspective, and this is what the book is all about. Also, they exist within the old architecture of grammar, the inverted T, which remains untouched by the new perspective.

Grammar itself is made of modules: are GB-subtheories cognitive modules? 543 (269) "UG consists of interacting subsystems, which can be considered from various points of view. From one point of view, these are the various subcomponents of the rule system of grammar. From another point of view, which has become increasingly important in recent years, we can isolate subsystems of principles. I will assume that the subcomponents of the rule system are the following: (1)

(i) lexicon (ii) syntax (a) categorial component (b) transformational component (iii) PF-component (iv) LF-component

[…] The subsystems of principles include the following: (2)

(i) bounding theory (ii) government theory (iii) T-theory (iv) binding theory (v) Case theory (vi) control theory" Chomsky (1981:5)

632

3.4. Are GB-subsystems cognitive modules? In the quotes above, Chomsky does not use the word "module" in order to refer to GB-subtheories, which he calls subsystems. These are opposed to subcomponents, i.e. the three computational systems of the inverted T. The question is whether these GB-subtheories ought to be regarded as modules in the Fodorian cognitive sense. Implicit in the quote under (269) is that they are not granted this status: subcomponents are distinct rule systems, while subsystems are only distinct sets of principles. A module, however, is defined as a computational system. Whether GB-subtheories are inputoutput systems that carry out a computation is certainly debatable, but I was unable to find any relevant discussion in the literature, both contemporary of the 80s and more recent overview-oriented (e.g. Lasnik & Lohndal 2009, Freidin & Lasnik forth, Newmeyer 1986:198ff), precisely (one senses) because linguistics were (and still are to a certain extent) insulated from the cognitive macrostructure in linguistic quarters.

544 Modularity Chap 5: Modularity of and in language, related systems The question to be asked is thus whether GB-subtheories act as mere filters that define well-formedness, or whether they carry out actual computation, i.e. modify an input. It seems to me that well-formedness filters and computational systems are not the same thing, but the question certainly deserves to be debated.157 Applying the other diagnostics for cognitive modules that are available appear to be inconclusive, at least from my point of view: GBsubtheories could perhaps be argued to be informationally encapsulated, but again the fact that they referee well-formedness would seems to require that they intervene at different stages of the syntactic construction. Domain specificity is also debatable: at first sight it looks like syntax as a whole uses the same vocabulary: person, gender, number etc. But these are the lexical ingredients, i.e. the input to the syntactic module as defined by the inverted T. GB-subtheories apply during the syntactic derivation, and their input, if any, will not be made of lexical material. Hence it could be argued that the case theory module works only with a specific vocabulary that is dedicated to case, the government module only calculates locality and so forth. This view is taken by Hornstein (2009:7, note 13) and is also entertained by a modern heir of the GB architecture, Edwin Williams' (2003) Representation Theory. Williams explains that (270) "several different aspects of clausal structure are characterized as separate 'sublanguages' (to anticipate: Theta Structure (TS), Case Structure (CS), Surface Structure (SS), Quantification Structure (QS), Focus Structure (FS). Then the syntax of a sentence will be a collection of structures, one […] from each of these sub-languages and a set of shape-conserving mappings among them." Williams (2003:2)

This perspective may be argued to face a problem because of its nested modular structure (see also the quote in §636): it supposes that there is a "big" syntactic derivation (the syntactic module of the inverted T) which somehow accommodates "small" sub-modules (the GB-subtheories) that concur to the overall result. It appears that this can only work in violation of informational encapsulation of the "big" syntactic module: its computation should be complete before any result can be sent to other (sub)modules. The same objection may be levelled against derivation by phase, though: morpho-syntactic computation sends partial results to PF and LF before it has reached its end, i.e. the matrix CP. This may be solved 157

See Scheer (forth) on the computational question and how computation is conceived of in phonology and syntax.

Grammar itself is made of modules: are GB-subtheories cognitive modules? 545

by considering that the "big" syntactic derivation is not performed by a module in one single go, but that the same module treats pieces of a sentence in several computations. Again I am not aware of any discussion in the (linguistic) literature on the compatibility of multiple spell-out with encapsulation, but Williams will be able to hook on the argument that will be made by defenders of derivation by phase, whatever that will be: the situations are parallel. Finally, the diagnostic based on double dissociation is probably not workable for GB-subtheories since it will be difficult to come by relevant pathologies (the evidence available is discussed in §642 below).

633

3.5. Biolinguistics: an evolutionary argument against language-internal modularity (Hornstein 2009) Biolinguistics looks at language from the biological and evolutionary perspective and thereby continues the earlier strand of Chomsky's language organ (§§609,627,633). Based on Hauser, Chomsky & Fitch (2002), the idea is that the appearance of language in the evolution of the species sets a restrictive frame that imposes certain properties upon grammar. Contrary to earlier generative thinking where much of what was found to be universal was put into UG, i.e. held to be genetically endowed and specific to language, UG is by and large emptied. Chomsky (2005) identifies three factors in language design: 1) UG (i.e. genetically endowed properties that are specific to language), 2) experience and 3) more general cognitive capacities that are not specific to language or even to the species. The shift, then, is from UG to the third factor: language relies on mechanisms that are much less specific to language than what was believed in earlier generative theories. Hauser et al. (2002) suggest that UG could actually reduce to recursion (Merge) and the ability to communicate with the interfaces (Phase): this is the Faculty of Language in the Narrow sense (FLN) (Pinker & Jackendoff 2005a,b oppose this view, see §609). On this backdrop, Hornstein (2009:4ff) levels an objection against language-internal modularity that is based on evolutionary timelines. As was mentioned, Hornstein (2009:7, note 13) believes that GB-subtheories do qualify for modular status. The Language Faculty as such, however, he argues, must not be cut into further subsystems because modular complexity implies genetic endowment and a relatively slow adaptational evolution along the Darwinian lines of environment-driven selection. The appearance

546 Modularity Chap 5: Modularity of and in language, related systems of language, however, was much too rapid in order to fit into an adaptive scenario. (271) "A common assumption is that language arose in humans in roughly the last 50,000 - 100,000 years. This is very rapid in evolutionary terms. It suggests the following picture: FL [Faculty of Language] is the product of (at most) one (or two) evolutionary innovations which, when combined with the cognitive resources available before the changes that led to language, delivers FL. This picture, in turn, prompts the following research program: to describe the pre-linguistic cognitive structures that yield UG's distinctive properties when combined with the one (or two) specifically linguistic features of FL. The next three chapters try to outline a version of this general conception. The approach, I believe, commits hostages to a specific conception of FL. It does not have a high degree of internal modularity. The reason for this is that modular theories of UG suppose that FL is intricately structured. It has many distinct components that interact in complex ways. On the assumption that complexity requires natural selection and that natural selection requires time to work its magic (and lots of it: say on the order of (at least) millions of years), the rapid rise of language in humans does not allow for this kind of complexity to develop. This suggests that the highly modular structure of GB style theories should be reconsidered." Hornstein (2009:4f, emphasis in original)

Hence we are back to the debate exposed in §606 between the two camps within the modular approach: following Descartes, Chomsky and Fodor hold that some properties of the cognitive system (Fodor's central systems) lie beyond what can be understood by human intelligence and will always remain an impenetrable mystery; by contrast, evolutionary psychology (Pinker, Plotkin, also Sperber and Smith) believe that all of the mind is modular and accessible to human understanding. A correlate of these contrasting positions is the Darwinian issue: in Fodor's and Chomsky's perspective, central systems are not the result of adaptive evolution along the laws of natural selection, while Pinker & Co hold that all properties of the mind/brain are modular and hence the result of adaptive evolution. Hornstein's argument thus supports the Fodorian/Chomskian view, and it is interesting to observe that Hornstein (2009:5, note 9) calls on literature from evolutionary psychology (Pinker 1997, Cosmides & Tooby 1992a) in order to back his critical assumption that complexity requires natural selection. His alternative, i.e. the emergence of the Language Faculty as a (by-)product of one or two evolutionary innovations based on pre-

GB modules and their perception in non-linguistic quarters 547

human cognitive capacities, is the basic idea of Hauser et al. (2002) and the biolinguistic programme. In this view, the FLN is not a product of selective adaptation, while the FLB (the Faculty of Language in the Broad sense) is shared with animals and results from natural selection. Finally, it should be added that the entire discussion is only about (morpho-)syntax: in the biolinguistic perspective, phonology and semantics belong to the animal-based FLB and hence do not really belong to grammar, i.e. to what is language-specific in the human cognitive system. PF and LF are thus supposed to have been present before the one or two innovations that produced Merge and Phase, and therefore to be available to and mastered (or masterable) by (certain) present-day animals (more on this in §639). This also explains why Hornstein in his discussion of what exactly counts as a module does not even mention the classical inverted T: his biolinguistically shaped horizon ends before PF and LF are in sight. The conclusion that one draws is that the inverted T still exists and that the three endpoints are still Fodorian modules – only are PF and LF not located in grammar anymore.

634

4. GB modules and their perception in non-linguistic quarters

635

4.1. Chomsky (1981) calls GB-subtheories modules without comment It was mentioned in §631 that at the outset of the Pisa lectures Chomsky (1981) distinguishes between subcomponents (the endpoints of the inverted T) and subsystems (GB-subtheories). Later in the book, however, he gives up on this distinction and systematically talks about modules when referring to the six GB-subtheories (also Chomsky 1982:29). (272) "The full range of properties of some construction may often result from interaction of several components, its apparent complexity reducible to simple principles of separate subsystems. This modular character of grammar will be repeatedly illustrated as we proceed." Chomsky (1981:7) "Note that the distribution of the empty categories, the differences among them and their similarities to and differences from overt elements are determined through the interaction of quite simple principles that belong to several different subtheories, in accordance with the modular approach to grammar that we are pursuing throughout." Chomsky (1981:72)

548 Modularity Chap 5: Modularity of and in language, related systems "This dissociation of properties is what we would expect on our modular assumptions - that is, on the assumption that such processes as 'passive' are composed of more fundamental abstract features, such as the elements of Case theory, T-theory, etc." Chomsky (1981:126) "Note that the full range of properties of PRO and trace discussed in §2.4, as well as the partially similar properties of overt anaphors, are determined by the interaction of four theories: the theory of bounding (for trace), the theory of control (for PRO), the theory of Case (for elements with phonetic content, including overt anaphors), and the theory of binding (for all NPs). The latter two theories are developed within the theory of government. Again, we see the highly modular character of the theory of grammar, with the basic subsystems of principles discussed in chapter 1 serving as quite simple and fundamental abstract components that interact to yield a complex range of properties." Chomsky (1981:192)

After having introduced each individual GB-subtheory in chapter 2 (which is called "Subsystems of core grammar"), the conclusion of this chapter confirms the modular character of GB-subtheories. (273) "The system that is emerging is highly modular, in the sense that the full complexity of observed phenomena is traced to the interaction of partially independent subtheories" Chomsky (1981:135)

Finally, the modular interpretation of GB-subtheories is also repeated on the last page of the book: "[t]he system that has been developed is highly modular" (Chomsky 1981:344). All this is done without mention or discussion of the Cognitive Science background, and without indicating whether the GB modules are considered to be cognitive modules. As we will see below, the discussion of the previous section has never been led, and everybody – in generative quarters and elsewhere – takes for granted that GB modules are cognitive modules in the Fodorian sense because Chomsky has used this word. 636

4.2. Perception of GB-modules in non-linguistic quarters: puzzlement While in generative quarters the word module was continued to be used without any particular reference to the broader organization of the cognitive system, observers from other fields, and especially from Cognitive Science of course, took for granted that the modules of the linguists are cognitive modules in the Fodorian sense. This led to some puzzlement and confusion.

GB modules and their perception in non-linguistic quarters 549

In the introduction to the volume on domain specificity of cognitive functions ("toward a topography of mind"), that they edit, Hirschfeld & Gelman (1994) talk about the situation in GB from the position of the external observer. (274) "Chomsky and other maintain that these findings provide compelling evidence for the claim that the mind is modular, comprising a number of distinct (though interacting) systems (the language faculty, the visual system, a module for face recognition), each of which is characterized by its own structural principles. […] Chomsky, however, has also suggested that the mind is modular in a somewhat different way. […] This, in other more technical writings, Chomsky has described 'modules of grammar' (e.g., the lexicon, syntax, bounding theory, government theory, case theory, etc.) (1988:135). Here the notion of modularity appears to be tied to specific subcomponents or subsystems of the language faculty rather than to the modular uniqueness of the language faculty itself. The grammar, in the traditional sense, is located at the intersection of these distinct modules. It is not clear whether these two notions of modularity are to be distinguished, and if so how to interpret the relationship between them. One possibility is that modules are nested, that is, the language faculty is a separate module that in turn consists of distinct component operations or modules. Another interpretation – supported indirectly by the fact that Chomsky speaks of the language faculty as a module to nonlinguists but speaks of the language faculty as consisting of modules to linguists – is that the mind is, strictly speaking, modular with respect only to these second-level component modules. The language faculty itself would accordingly be a more vaguely defined construct resulting from the operation of these modules, but one that in itself is not modular in the sense of being defined in terms of a distinct set of principles." Hirschfeld & Gelman (1994:8, emphasis iin original)

Looked at from the inside, Hermon (1985) is a typical representative of the approach to modularity that was (and is) widespread in generative quarters. Hermon's goal is to demonstrate that the interplay of the subtheories of the newly established GB theory are well suited to account for parametric variation, and this is the only reason why the book is called Syntactic Modularity. There is no reference to any extra-linguistic work regarding modularity (Fodor 1983 is absent from the reference section), and the reader does not learn anything about cognitive modules, how they work, how they are defined, detected etc. Similar cases are Farmer (1984) and Nespor & Vogel (1986) (see §428).

550 Modularity Chap 5: Modularity of and in language, related systems 637

5. Minimalism and biolinguistics do away with GB modules

638

5.1. Minimalism: GB-subtheories have to go In their textbook on minimalism, Hornstein et al. (2005:11ff) evaluate the validity of GB-subtheories in the light of minimalist requirements and, following the GB-tradition, refer to them as modules. The items of the inverted T, i.e. morpho-syntax, phonology and semantics, are called levels (Hornstein et al. 2005:9 and elsewhere). The word "modularity", however, only appears when language as a whole is considered in the concert of other cognitive functions (Hornstein et al. 2005:3, note 2). A core goal of the minimalist programme is to reduce the GBsubtheory zoo to what is strictly required for conceptual reasons and by interface conditions. Grammar is perfect unless deviated from perfection by interface requirements: minimalist design produces "a theory of language that takes a linguistic expression to be nothing other than a formal object that satisfies the interface conditions in the optimal way" (Chomsky 1995a:171). In this minimalist frame, Hornstein's (2009) argument against GB-modules that is based on the third-factor perspective of biolinguistics was already discussed in §633. This does not mean, however, that the insight of GB is wrong: Hornstein (2009:6) believes that the generalisations made by GB are "roughly empirically correct". However, they are in need of further interpretation: minimalist work in general and his book in particular is about how to have the labour of GB-subtheories done by different, non-modular means, and to derive the empirical generalisations of GB by more general principles; these are ideally of the third factor kind, i.e. unspecific to language.

639

5.2. Grammar reduces to morpho-syntax: PF and LF are neither languagenor species-specific It was already mentioned in §633 that interestingly enough, PF and LF are not even mentioned when Hornstein (2009) and others talk about the faculty of language in a biolinguistic perspective. This is because according to Hauser et al. (2002), PF and LF do not belong to the FLN (Faculty of Language in the Narrow sense). Rather, they are a piece of the FLB (Faculty of Language in the Broad sense), which humans share with (certain) animals. Unlike the FLN which did not have enough time to emerge by Darwinian means, the FLB came into being through adaptive evolution that occurred

Identifying linguistic modules 551

under selective pressure during the common evolution of certain animals and the ancestors of homo sapiens long before the critical hardwaremodification occurred that made emerge FLN. Filling in this scenario on the phonological side, Samuels (2009a,b) tries to show that phonology is entirely a third factor mechanism, i.e. that there is nothing language- or species-specific to human phonology: (certain) animals are perfectly equipped to do human phonology. In sum, language reduces to morpho-syntax, which is made of one single Fodorian module, i.e. morpho-syntax. This perspective reassesses the delineation of grammar (phonology and semantics stand aside), and hence makes the inverted T appear in a different light. The new limits of grammar, however, do not change anything in the relationship between the three actors of the inverted T: PF and LF may not be specifically linguistic modules anymore, but they are still modules. Therefore whatever the relationship of morpho-syntax with them, it must follow the general rules of intermodular communication: the way morpho-syntax talks to PF and LF is not any different from the way it talks to other cognitive modules such as audition or vision.

640

6. Identifying linguistic modules

641

6.1. How to identify grammar-internal modules Given the GB-interlude and the minimalist and eventually biolinguistic perspective, one may say that generative grammar is back to where it started in the 60s: we are left with the inverted T. That is, there are three relevant modules, morpho-syntax, PF and LF, of which the latter two lie outside of grammar if one wants to follow biolinguistics. At least two more systems are relevant and interact with these: pragmatics and a conceptual device. Below evidence is gathered that allows us to evaluate this working hypothesis. Methods for identifying cognitive modules are as before, one internal (domain specificity, §§610f), the other external (double dissociation, §618). As far as I can see, there is only little evidence available from the latter source, which will be reviewed in the following section. The literature that builds on domain specificity in order to tease apart the number and nature of language-relevant modules is not substantial either.

552 Modularity Chap 5: Modularity of and in language, related systems 642

6.2. Dissociation: Pragmatics, Lexicon vs. morpho-syntax Dissociation arguments come either from stable synchronic states where cognitive functions are selectively impaired (i.e. from subjects with cognitive and/or brain damage), or from acquisition, where different cognitive functions are dissociated in their development (see §620). When zooming into grammar (understood as including phonology and semantics) from the larger perspective of cognitive functions, the most coarse-grained differentiation is between the two standardly assumed peripheral systems that relate grammar to other functions, and grammar itself. The two systems in question are pragmatics and a conceptual device. Evidence for the independence of the pragmatic and the grammatical systems was already discussed in §621: Christopher, the savant studied by Smith & Tsimpli (1991 et passim), appears to be unable to make pragmatic pressure (to produce a sound translation) influence his linguistic performance. Also, Chien & Wexler (1990) provide evidence from acquisition for the dissociation of binding (a grammatical principle) and pragmatics. In an early study on the dissociation of language-related cognitive functions based on pathological data, Curtiss (1981) concludes that while morpho-syntax is insulated from other cognitive functions, the development of lexical and relational semantic knowledge hinges on broader conceptual abilities. (275) "[D]ata from case studies of children show […] clear dissociations between language and nonlanguage cognitive abilities. The implications of such data are discussed. The major implications appear to be that lexical and relational semantic abilities are deeply linked to broader conceptual development but morphological and syntactic abilities are not. The development of a normal linguistic system, however, one in which grammar is systematically related to meaning, requires concurrent and concomitant linguistic and nonlinguistic cognitive development." Curtiss (1981:15)

Finally, Newmeyer (2006:241f) proposes a kind of double dissociation argument without recurring to cognitive and/or brain damage. In order to show that syntax and semantics are independent computational systems, he demonstrates that a particular syntactic structure does not select for semantic values, and conversely that a particular semantic or discourse-based construct may map onto various syntactic structures. Newmeyer (2006) thus defends a strict modular segregation of syntax and semantics. He shows that the seeds of blurred modular contours between these two items of the inverted T were sown by Chomsky (1981),

Identifying linguistic modules 553

who introduced the idea that thematic roles are directly relevant for the statement of syntactic generalisations. Since then and especially in the minimalist environment, the bonds between syntactic position and semantic interpretation have been strengthened. On this backdrop, Newmeyer (2006) proposes evidence from an analysis of English negation that militates against a conflation of syntactic and semantic features. Higginbotham (1987) also argues for the autonomy of syntax and semantics in a modular perspective, which is tightly correlated to the reading of Fodor (1983). He reviews the classical linguistics-internal arguments that Chomsky has made since the 50s in order to establish the mutual independence of syntax and semantics. The following section discusses the reverse attitude, i.e. which validates the convergence of syntax and semantics on the grounds of domain specificity.

643

6.3. Domain specificity (Starke): morpho-syntax-semantics vs. phonology In unpublished work,158 Michal Starke argues that morphology, syntax and semantics are just one module because they use the same vocabulary: number, person, animacy, quantification, aspect and so forth are categories that are used, understood and processed by syntax as much as by morphology and semantics.159 Much unlike phonology, where number, person and the like are unknown: phonology does not use or process these categories. Conversely, morphology, syntax or semantics neither process or are sensitive to genuinely phonological concepts such as labiality, stopness and the like. On Starke's count, then, phonology (as much as pragmatics and the conceptual device) works with specific vocabulary and is thus a module distinct from morpho-syntax-semantics. Discussing the detail of the evidence that Starke relies on would lead too far afield here (a published version will hopefully be available at some point). Let us merely note the structure of his argument, which is along the lines of domain specificity. 158

159

Starke's work has been presented at various conferences and at the Central European Summer School in Generative Grammar (EGG) in 2002 (Novi Sad) and 2006 (Olomouc). Of course semantics is to be understood as "grammatical" semantics, i.e. the system that assigns an interpretation to morpho-syntactic structure. The meaning of lexical items and the relation with the conceptual world are entirely different issues.

554 Modularity Chap 5: Modularity of and in language, related systems The result is a broad distinction of two macro-modules, phonology and morpho-syntax-semantics, which are supplemented by (at least) two modules that mediate between grammar and other cognitive functions (pragmatics and the conceptual device).

644

6.4. Domain specificity (Jackendoff, Chomsky): phonology is distinct Jackendoff's (1987, 1992, 1997) modular theory, Representational Modularity (which Jakendoff 2002:218ff prefers to call Structure-Constrained Modularity today), also points out the obvious ontological gap between phonology and other linguistic devices, which is greater than the distance between any other two linguistic candidate disciplines. (276) "The overall idea is that the mind/brain encodes information in some finite number of distinct representational formats or 'languages of the mind.' Each of these 'languages' is a formal system with its own proprietary set of primitives and principles of combination, so that it defines an infinite set of expressions along familiar generative lines. For each of these formats, there is a module of mind/brain responsible for it. For example, phonological structure and syntactic structure are distinct representational formats, with distinct and only partly commensurate primitives and principles of combination. Representational Modularity therefore posits that the architecture of the mind/brain devotes separate modules to these two encodings. Each of these modules is domain specific. […] The generative grammar for each 'language of the mind,' then, is a formal description of the repertoire of structures available to the corresponding representational module." Jackendoff (1997:41)

Chomsky (2000a) makes the same point. (277) "The phonological component is generally assumed to be isolated in even stronger respects: there are true phonological features that are visible only to the phonological component and form a separate subsystem of FL [the Faculty of Language], with its own special properties." Chomsky (2000a:118, emphasis in original)

Domain specificity within grammar thus identifies what appears to be the deepest fraction line, which separates phonology on the one hand and all other classical disciplines (syntax, morphology and semantics) on the other.

Identifying linguistic modules 555

Jackendoff ends up with three modules that are involved in the management of grammar: phonology, syntax and the conceptual device. He calls modules processors and distinguishes between integrative and interface processors (see Vol.2). The latter translate the output of the former into vocabulary items that can be understood by other "true" modules, i.e. phonology, syntax and the conceptual device in our case. Intermodular communication is discussed in §649 below (and at greater length in Vol.2).

645

6.5. Phonology-free syntax That phonology is ontologically distinct also shines through more familiar linguistic work that does not think in modular categories or look at domain specificity: the distinctness of phonology is the core message of phonologyfree syntax, which was discussed in §412 (also §253). In Zwicky & Pullum's (1986a,b) strong original version, phonologyfree syntax holds that syntax and morphology are deaf for any phonological information: there is no morpho-syntactic process that has a phonological conditioning. For example, there is no syntactic movement on record that is triggered only if, say, the candidate begins with a labial. A weaker version (see §412) distinguishes between melody (i.e. phonological objects located below the skeleton) on the one hand and syllabic as well as prosodic properties on the other (e.g. supra-skeletal structure). The inability of melody to bear on morpho-syntax appears to be a correct generalisation, while syllabic and prosodic properties such as intonation, minimal word constraints and other counting operations are found to condition morphological and syntactic processes. The fact that the weaker version is probably correct is actually more interesting from the modular perspective: it means that the basic phonological vocabulary is unintelligible for morpho-syntax, but that projections thereof may be perceived. Which thus nicely confirms that domain specificity is about vocabulary, rather than about the output of a computation that is based on this vocabulary (§651 elaborates on this).

646

6.6. Late Insertion is the segregation of phonological and other vocabulary The ontological separation between phonology and morpho-syntax is also central in Distributed Morphology: while up to GB morpho-syntactic computation was done on the basis of complete lexical information that in-

556 Modularity Chap 5: Modularity of and in language, related systems cluded syntactic, morphological and semantic features as much as phonological material (sealed suitcases), Late Insertion is the idea that phonological material is absent from morpho-syntactic computation (see §536). Only morpho-syntactic information is available at the beginning of a derivation; phonological material (vocabulary items) is only inserted after the completion of the morpho-syntactic derivation. 647

6.7. Phonology vs. phonetics Although this book does not consider the relationship of phonology with phonetics, i.e. the (eventual) lower limit of phonology, it is worth pointing out that domain specificity is also used in the large body of literature that debates this issue in order to insulate both areas: this is what Hale & Reiss (2008:118) do. Kingston (2007) provides a good overview of the positions that are taken, and especially of the debate whether phonology and phonetics are distinct modules or instances of the same computational system.

648

7. Encapsulation is called inclusiveness in syntax Informational encapsulation is a core property of modules (§610): modules produce an output on the grounds of a domain specific input, and there can be no communication with anything beyond the module (i.e. possible sources of additional information) during the computation. It is worthwhile to be mention that Chomsky's (1995a:228) inclusiveness is the syntactic formulation of encapsulation: syntactic structure must be exclusively based on information that is present in the input; no element may be added in the course of a syntactic derivation.

649

Chapter 6 How modules communicate

650

1. Intermodular communication requires translation Let us now turn to intermodular communication. A direct consequence of the fact that different modules speak different languages (of the mind) is their inability to understand each other. Modules can only parse objects that belong to their own language, i.e. which are part of the domain specific vocabulary that they are designed to process. This is what Jackendoff explains in the quote below. (278) "'Mixed' representation[s] should be impossible. Rather, phonological, syntactic and conceptual representations should be strictly segregated, but coordinated through correspondence rules that constitute the interfaces." Jackendoff (1997:87ff)

Applied to the phonological module, this means that phonology could not react on any untranslated input from the morpho-syntactic module. This is precisely the principle of Indirect Reference that was introduced by Prosodic Phonology (see §§377,406): phonology can only take into account morpho-syntactic information that was previously translated into phonological vocabulary. The whole architecture of Prosodic Phonology is shaped according to Indirect Reference: a Translator's Office mediates between morpho-syntax and phonology. That is, the morpho-syntactic output is mapped onto prosodic constituency, which is the input to phonology. The basic idea of intermodular communication that materialises in the architecture of Prosodic Phonology is thus the following: in order for two modules to talk to each other, there must be a mediating instance which understands the vocabulary of both the input and the output module and translates information from one into the other. Untranslated information is noise and will be ignored by the receiving module. The idea that morpho-syntactic information must be translated before phonology can use it has always been present in phonological theory since the 19th century. A summary of how translation was practised since structuralist times is provided in §692 below.

558 Modularity Chap 6: How modules communicate 651

2. Translation of what?

652

2.1. Modular computation: vocabulary (input) vs. structure (output) When talking about modules, an important distinction is between vocabulary (content) and structure. The structure of a module is the result of its computation: based on an input that is made of domain specific vocabulary, computation builds structure. In syntax for instance, the morpho-syntactic tree is the projection of morpho-syntactic features; in phonology, syllable structure is the projection of segmental properties. While vocabulary is necessarily domain specific, structure does not have to be. That is, different modules may produce the same type of structure, i.e. which has identical properties (one important aspect of which is hierarchical organisation). This means that structure is predestined for being shipped to other modules at the end of a computation: in case both vocabulary items (terminals) and structure are shipped off, the receiving module will be unable to make sense of the former, but may be able to interpret the latter, which is domain-unspecific. This is a somewhat tentative hypothesis that depends a lot on how intermodular translation actually works. Closer inspection of this issue is only provided in Vol.2. In anticipation of the discussion, the regular view is that translation is done by a specialised translation module, i.e. by some computation (the Translator's Office in Prosodic Phonology, Jackendoff's interface processors). An alternative is translation through a lexical access: like in a dictionary, each item of the input vocabulary is matched with an item of the output vocabulary. That is, translation does not involve any computation (this is what Michal Starke argues for).

653

2.2. Is structure, but not vocabulary, translated? In principle, translation could translate vocabulary items and structure alike. There is some indication, though, that modules may be sensitive to structure, but not to vocabulary. This empirical generalisation was already mentioned in §398: there is at least a strong trend for phonology to be sensitive to morpho-syntactic structure, i.e. geometric properties of the tree, while node labels are by and large ignored. The same is true in the opposite direction: phonology-free syntax (see §645, originally discussed in §412) is in fact melody-free syntax: the basic vocabulary items of phonology are the objects that occur below the

Translation is selective, and the choice of translated pieces is arbitrary 559

skeleton, i.e. melodic primes such as labiality, stopness and so on. On the other hand, syllable structure and other properties of supra-skeletal phonological representations are the result of phonological computation that is based on this basic vocabulary. If it is true that structure may be translated, while vocabulary remains untranslated, melody is predicted to be unable to bear on morpho-syntax – and this is indeed what we observe. By contrast, supra-skeletal phonological structure may be read by other modules. This is indeed the result produced by the literature that set out to challenge phonology-free syntax: all cases where phonology influences morpho-syntax appear to concern phonological properties that are located above the skeleton (§412, see also §662). While the empirical situation is unambiguous in the direction from phonology to syntax (cases where morpho-syntax reacts on labiality or the like are not on record), it is not exactly clear-cut in the opposite direction: cases such as the well-known stress-determining difference between nouns and verbs in English (récord vs. recórd etc.) stand in the way. Categorysensitive phonology is further discussed in §752. Finally, vocabulary may not only be excluded as an input to translation – maybe it does not qualify as its output either. This is suggested by the discussion in §661 below: melody (i.e. phonological vocabulary) is not only invisible for morpho-syntax – it is also unheard of as a carrier of morpho-syntactic information in phonology. In sum, then, vocabulary would be excluded from translation altogether: only structure qualifies as its input and output.

654

3. Translation is selective, and the choice of translated pieces is arbitrary Another pervasive property of intermodular communication appears to be the fact that translation is never complete. That is, only a subset of the structure of the sending module is made available to the receiving module through translation. Also, it appears that the pieces which are chosen for transmission cannot be predicted. Ray Jackendoff's work regularly draws attention to the underfeeding of the receiving module.

560 Modularity Chap 6: How modules communicate (279) "Correspondence rules perform complex negotiations between two partly incompatible spaces of distinctions, in which only certain parts of each are 'visible' to the other." Jackendoff (1997:221) "The overall architecture of grammar consists of a collection of generative components G1, …, Gn that create/ license structures S1, …, Sn, plus a set of interfaces Ijk that constrain the relation between structures of type Sj and structures of type Sk. […] Typically, an interface Ijk does not 'see' all of either Sj or Sk; it attends only to certain aspects of them." Jackendoff (2002:123)

The fractional character of translation in intermodular communication is further discussed in Vol.2, where illustration from various cognitive functions is provided.

655

4. Outlook: intermodular translation is the focus of Vol.2 How exactly the respective translating mechanisms work has received little or no attention at first: the distribution of juncture phonemes was stated in prose (if anything), and SPE had a universal algorithm that distributed hashmarks according to hierarchical morpho-syntactic structure (§90). Prosodic Phonology paid much more attention to the labour that translation requires, which was done by a special kind of rules, i.e. mapping rules (§388). In OT, mapping is constraint-based (instead of rule-based, §457), and in Jackendoff's (1997 et passim) general landscape, modules communicate by means of so-called correspondence rules, which in his more recent work appear as Interface Processors (see Vol.2). Of course, translation only concerns the representational aspect of the interface: cyclic chunk-submission runs independently. The genuine contribution of Vol.2 to the interface discussion lies on the representational side. Two issues are examined in detail. On the one hand, the question is asked whether it is reasonable (or even workable) to have two distinct means of piecing together phonological material (representational objects) that serves as the input to the phonological module: vocabulary (lexical) insertion on the one hand (which concerns morphemic information, origin: the lexicon), and "added" non-morphemic information that comes in through translation (juncture phonemes, boundaries, prosodic constituency, origin: output of the Translator's Office). It will be argued that the answer is no: this is not reasonable. All phonological material that phonological com-

Outlook: intermodular translation is the focus of Vol.2 561

putation uses originates in the lexicon (One Channel Translation, an idea of Michal Starke). On the other hand, the diacritic issue is examined. In the history of post-war phonology, the output of translation have always been the units of the current theory: juncture phonemes in structuralist times, ([-segment]) segments in SPE, prosodic constituency in the autosegmental 80s. The trouble is that all of these are diacritics: a juncture phoneme is obviously not a phoneme, a boundary is evidently not a segment, and prosodic constituency, unlike all other tree structure (in phonology as much as in morpho-syntax) is a projection of nothing, i.e. not a bottom-up construction (more on this in §715). Direct Interface (Scheer 2008a, 2009a,c, to be introduced at greater length in Vol.2) holds that the output of translation must be genuinely phonological objects, that is items which belong to the vocabulary that the phonological module works with, and which exists independently of any interface issue (i.e. of extra-phonological factors).

Part II Lessons from interface theories

656

Chapter 1 A guide to the interface jungle

The following pages take stock of the historical survey of Part I, as well as of the Interlude. The goal is to filter out important issues and ideas that are relevant beyond the theoretical context in which they were produced, and whose impact is constant over time as theories come and go. The result ought to provide a thematic access to interface questions, something like a blown-up index: the reader can look up this or that issue, follow this or that idea, and will be directed to all relevant discussion that was provided in Part I (and in the Interlude) under the header of a particular theory and a specific period. Interestingly, it will turn out that the relevant headers of chapters and sections are typically not the ones that one comes across in the literature: the distilling process of the historical mill produces focal points of interest that are not usually used in order to talk about the interface. Examples are inside-out interpretation (i.e. cyclic derivation), selective spell-out, local vs. non-local representational intervention in phonology, privativity and the word-spell-out mystery, to mention just a few (see also §843). Given this pool of issues, ideas and questions, our guide through the interface jungle is organised along the following coordinate system: the basic distinction between representational and procedural aspects of the interface that is constant in this book on the one hand, the classification of questions according to whether they are settled or open (that is, whether or not they still discriminate among present-day theories) on the other. Hence there is a chapter on issues that are settled (on both procedural and representational sides: chapter 3), a chapter on open questions regarding procedural issues (chapter 6), a chapter on open questions that are not specifically related to either procedural or representational questions (chapter 5), and finally a chapter on open questions on the representational side (chapter 4). This chapter has ended up not being called like that because the three issues that it is made of – translation, the diacritic issue and the (non-)locality of intervention – are settled in verb, but (unfortunately) not

564 Chap 1: A guide to the interface jungle in fact, let alone in current practice. This general structure is completed by a chapter on empirical (i.e. pre-theoretical) generalisations (chapter 2), which is what we start out with below. The purpose of Part II is certainly to provide a structured and thematic access to issues, ideas and solutions that roughly 70 years of interface thinking have produced. As was mentioned in the introduction to the book, however, this is but a means to an end: what Part II is really after is to allow the reader to learn from the history of the interface; in the best case, the synopsis will teach us how we may, should or must not go about the interface in further theory design: the vanishing point is a correct interface theory. Like the rest of the book, the following pages are thus written in the interest of the future, rather than as a guide through the past: which errors have been made that we may want to avoid? Which principles emerge that must not be played fast and loose with? Have theories really stood up to the standards that they were advertising? Finally, it is worth cautioning that the narrative attitude which has prevailed thus far will stand back behind a less neutral prose: rather than merely reporting on what this and that theory says, proposals, attitudes and mechanisms are evaluated. As we go along, two measures will be used that build on grounds which lie beyond classical phonology-based reasoning: modularity and what I call intermodular argumentation. Both referee competing phonological theories on the grounds of extra-phonological systems: syntactic evidence is brought to bear in the latter, properties of cognitive computation and architecture in the former case.

657

Chapter 2 Empirical generalisations

658

1. Introduction This chapter lists empirical generalisations regarding the interface of morpho-syntax and phonology that appear to hold independently of particular theoretical orientations. They are thus different from issues that are consensual today (but eventually were not in the past) and involve a specific view on how the interface works. Inside-out interpretation (§672) or the need for translation (§692) for example fall into this category. Relevant empirical generalisations may or may not be discussed in the literature (and hence in Part I), and they may or may not be identified as such. Those that the reader has already come across are treated in the following chapter on issues that are settled. The present chapter only discusses cases which for some reason are not debated in the literature as far as I can see. Under (280) below appear the four relevant empirical generalisations that I can think of. (280) empirical generalisations regarding the interface a. morpho-syntax has no bearing on the content of phonological computation: the application of phonological computation can be altered under morpho-syntactic influence, but not its properties b. morpho-syntax and melody are incommunicado c. there are no boundaries inside morphemes, hence morpho-syntax has no bearing on morpheme-internal phonological objects d. there is no specific phonetic correlate of morpho-syntactic information

(280c) and (280d) have already been discussed in Part I; the issues at hand are summed up in the following chapter (§667 and §668, respectively). (280b) has been touched on in §412 (phonology-free syntax), but not as an independent issue. Together with (280a) (which was not covered in Part I at all), it is discussed below in the present chapter.

566 Chap 2: Empirical generalisations 659

2. Morpho-syntax has no bearing on the content of phonological computation An important generalisation regarding the interface is that whatever morpho-syntax does to phonology, it is unable to create, to suppress or to modify phonological processes. That is, phonological computation exists independently of morpho-syntactic computation and of any procedural or representational agent thereof. It is not phonological instructions themselves that are impacted by extra-phonological information: it is their application that may be altered by morpho-syntax. On the procedural side, of course, there is no influence on the content of phonological computation: procedural influence on phonology is achieved by the successive submission of growing chunks of the whole to an invariable phonological computation. This may produce opacity effects but leaves the phonological computation alone with the vocabulary items that are the result of the spell-out of morphemes. In the case of representational communication, some carriers of morpho-syntactic information that are not part of the morphemic endowment (juncture phonemes, hashmarks, prosodic constituency) have been added to the input to phonological computation. These carriers may then modify the result of the phonological computation (as compared with a situation where they are absent). The computational system itself, however, i.e. the set of instructions (the rule inventory and ordering, the constraint inventory and ranking etc.), remains unaltered. Empirically speaking, the effect of this division of labour is that nobody has ever seen a phonological process that is active in a certain morpho-syntactic environment (or in mono-morphemic strings), but not in others. It does happen frequently, of course, that processes are blocked (or triggered) in a specific morpho-syntactic environment, and this may have procedural as much as representational reasons (e.g. affix class-based phenomena, see §§163,166). This does not mean, however, that the content of the phonological computation was modified: its application has just been suspended. Of course, this does not withstand the existence of distinct computational systems, which may either be morpheme-specific (level 1 vs. level 2 in Lexical Phonology, §148) or chunk-specific (lexical vs. postlexical in Lexical Phonology, §234). What was said in the previous paragraph is only true for phenomena that fall into the competence of a given computational system. Or, put differently, the properties of phonological computation themselves remain unmodified, whatever morpho-syntactic influence

Morpho-syntax has no bearing on the content of phonological computation 567

comes to bear, also in theories that accommodate several computational systems. As far as I can see, the inalterability of phonological computation in the face of morpho-syntactic information is entirely consensual across all theories and protagonists over the 70 years of interface literature that have been covered. That is, all theories have attempted to reduce morphosyntactic influence to procedural and/or representational influence along the lines described. Although it is a logical possibility, nobody has seriously thought of modifying the phonological computational system according to morpho-syntactic contexts. One could imagine for example a rule k → tÉʃ / __i,e (or its equivalent constraint) that is changed into k → tÉs / __i,e at a certain suffix boundary. Another way to alter phonological computation would be to add a rule (or a constraint) when some morpho-syntactic division is run through the phonology. Nobody has ever attempted to do that: morpho-syntax influences the application of pre-existing phonological computation (by blocking or triggering processes), but it does not create or modify it. The fact that the definition of the properties of phonological computation is an exclusively phonological matter on which morpho-syntax has no bearing follows from modularity (see §622): morpho-syntax (or morphology and syntax) and phonology are distinct computational systems that work on distinct vocabulary items (domain specificity). The only way for modules to talk to each other is through the exchange of their respective outputs. The output of modules, however, is structure, not computation. Therefore the respective computational systems that modules are made of live in complete autarky and could not possibly bear on one another. The generalisation at hand is probably is too obvious to arouse specific concern in the literature: I have not come across any text that makes it explicit. The fact that it follows from modularity is certainly a good point for this theory of cognitive organisation; in turn, it bolsters the idea that morpho-syntax and phonology are distinct computational systems (modules).

568 Chap 2: Empirical generalisations 660

3. Morpho-syntax and melody are incommunicado

661

3.1. Morpho-syntax can neither read melody nor bear on it

662

3.1.1. Phonology-free syntax is in fact melody-free syntax Recall from the discussion of phonology-free syntax in §412 (also §§653,253) that phenomena where melodic properties (i.e. phonological specifications that are made below the skeleton) condition a morphological or a syntactic process appear to be absent from the record (this was actually Zwicky & Pullum's 1986a,b original observation). By contrast, it seems that phonological properties located above the skeleton can bear on morpho-syntax. Hence something like "verbs that begin with a labial are raising verbs" is unheard of.

663

3.1.2. Carriers of morpho-syntactic information do not include melody There is good reason to believe that communication in the opposite direction, i.e. from morpho-syntax to phonology, is restricted in the same way: carriers of morpho-syntactic information that are inserted into phonology through the representational channel always land at (juncture phonemes, SPE-type hashmarks) or above (prosodic constituency) the skeleton; they do not include melody.160 Hence processes whereby, say, raising verbs palatalise, but non-raising verbs do not, are unheard of. On the other hand, phonotactic and suprasegmental effects in phonology typically depend on morpho-syntactic information. Extrasyllabicity, extraprosodicity, the restriction of word-initial clusters to obstruentsonorant sequences, the tolerance of heavy clusters at the right edge (e.g. English sixths [sɪksTs]) are all examples where a non-melodic effect depends on a calculus regarding string edges, which is a morphological information. Analogous melodic effects such as, say, "all velars palatalise word-initially", are not on record.

160

The only case that I am aware of where morpho-syntactic information was really proposed to have a melodic incarnation is Lass' (1971) analysis where the word boundary identifies as [-voice]. This early attempt at translating morphosyntactic information into something else than diacritic boundaries is discussed in §135.

Morpho-syntax and melody are incommunicado 569

The (ontological) gap between items that occur below and above the skeleton should then allow for the latter to be the output of translation. Beyond structural units such as syllabic space and prosodic constituency, an interesting candidate is tone: traditionally represented above the skeleton and counted as suprasegmental because of its lateral mobility, tone seems to have a melodic identity (H, L). Whether tone is a melodic or a suprasegmental item (or both) is an old question. Interesting evidence comes from an analysis of Yoruba cyclic accentuation: Manfredi (forth, section 2.2) shows that an H is distributed by each phase. That is, an otherwise inexplicable high tone with no lexical origin appears in every spell-out domain (CP, TP, DP on Manfredi's analysis): phases distribute an H. This points to the ability for tone to be the output of translation and places it in the suprasegmental category of items that do not live below the skeleton: tone is not melodic. Finally, the exclusion of melody from the interface also ties in with the argument that Pyle (1972) has raised against SPE-type boundaries. Recall from §136 that he observed that selling # for a ([-segment]) segment must be wrong because rules may transform segments into other segments, but # never becomes [p] or [a] (or vice-versa): # → a / X__Y is unheard of. This is an expression of the generalisation at hand in the terminology of the 70s: morpho-syntactic information (the #) can never be transformed into melody (segments such as [a]).

664

3.2. (Floating) morphemes without linear realisation originate in the lexicon Of course the ultimate effect of morpho-syntactic bearing on phonology is always either phonotactic (e.g. the restriction of word-initial consonantal sequences to obstruent-sonorant clusters) or melodic. The question is whether there are cases where a melodic effect is direct, i.e. not mediated by some structure at or above the skeleton. Prima facie candidates for non-mediated influence are cases where a morpheme has no surface manifestation that could be identified in the linear string: it merely modifies material that belongs to other morphemes. Umlaut, or any floating morpheme in autosegmental terms for that matter, fall into this category. One may be tempted to conclude, then, that the boundary of the morpheme at hand materialises as the triggering melodic item in question. Such an analysis appears to be outlandish, though, and as far as I can see has never been proposed.

570 Chap 2: Empirical generalisations Let us take a closer look at umlaut. In German for example, umlaut may mark the plural of a noun. In some cases, it appears together with a suffix (Haus - Häus-er "house sg, pl"), but in others it is the only marker. The plural of Mutter [mʊtɐ] "mother" is Mütter [mʏtɐ] for example: it is only marked by the palatalising effect on the root vowel. In the massive body of literature regarding this phenomenon, as far as I can see nobody has argued for an analysis whereby the plural boundary incarnates as a palatal agent, which then causes fronting of the back root vowel. All phonological analyses (there are also analyses that rely on allomorphy or other non-phonological mechanisms) hold that the lexical identity of the plural morpheme is a palatal piece of melody that is regularly linearised after the root. In linear SPE-type analyses, the morpheme is an -i (its historical identity), which fronts the root vowel before being deleted by a subsequent rule. In autosegmental environments, the lexical identity of the plural morpheme is a floating palatal prime, which parachutes on the root vowel and is never seen as such because it has no syllabic constituent that it can associate to. In any event, the lexicon is found to be the origin of the palatal element: no phonologist has thought of it in terms of the output of the translation of morpho-syntactic structure.

665

3.3. Conclusion: vocabulary excluded from translation altogether? Given the lexical origin of the melody of floating morphemes, there is good reason to exclude the area below the skeleton from the possible target zone of representational communication: whatever objects are the output of translation, they can only land at or above the skeleton. As far as I can see, if tacitly, the non-role of melody at the interface is undisputed: all phonological carriers of morpho-syntactic information in phonology that interface theories have used over the past 70 years are nonmelodic (i.e. are inserted at or above the skeleton): juncture phonemes, SPE-type hashmarks and prosodic constituency. When conjoining melody-free syntax and the fact that the output of translation cannot be melodic items, the conclusion makes sense: melody and morpho-syntax are entirely incommunicado. This appears to be a robust generalisation. In modular terms, this ties in with the discussion in §651 where the tentative generalisation was made that only structure, not vocabulary items, are subject to translation in intermodular communication (see §752 on the

Morpho-syntax and melody are incommunicado 571

particular case of phonologically (ir)relevant morpho-syntactic vocabulary). In our case indeed, phonological vocabulary – melody – is excluded from translation into morpho-syntax. A new aspect of intermodular communication is added, though: the generalisation concerning the reverse direction suggests that vocabulary (melody) is also unable to be the output of translation. This would mean that vocabulary does not participate in intermodular communication at all: neither as input nor as output. Even though, recall from the discussion in §653, nothing in principle withstands the translation of vocabulary (or its being the output of translation). Even if in a modular environment it makes sense to ignore those objects that the receiving module will not be able to parse anyway if untranslated, the two-way incommunicado between melody and morpho-syntax is in need of explanation. As far as I can see, no theory of either the cognitive or the linguistic system knows why, say, labiality is excluded from the translational process in direction of morpho-syntax, and why morpho-syntactic information can be translated into phonological vocabulary, but not into melody.

666

Chapter 3 Issues that are settled

667

1. There are no boundaries inside morphemes Two issues that were at the forefront of structuralist discussion appear to be settled: there are no boundaries inside morphemes, and there is no phonetic correlate of morpho-syntactic divisions. Level Independence (§61) is a headstone of structuralist theory; in the perspective of a discovery procedure where linguistic representations are constructed bottom-up from phonetics over phonemes and morphemes to sentences, look-ahead is prohibited. Therefore there can be no morphological information in phonology. Even though the principle was largely admitted, structuralist practice was manifold (§§64-66), and morphosyntactic information was reintroduced through the back door in the phonological guise of juncture phonemes. If Level Independence was to be held up, then, the coincidence of juncture phonemes and morpho-syntactic divisions had to be accidental. This opened (or rather: pushed) the door for the disastrous practice of referring to ghost-junctures in the midst of morphemes: anything and its reverse could be ascribed to them (§69). Chomsky et al. (1956) expunged this kind of juncture abuse. In order to do so, Level Independence first needed to be unhorsed: not only is there nothing wrong with the presence of morpho-syntactic information in phonology, but this kind of information is actually necessary and regular (§76). Morpho-syntactic control over the distribution of juncture was then restored: boundaries – the name that was now given to juncture phonemes – can only occur in place of a morpho-syntactic division (§78). This necessary egress from the denial of morpho-syntactic bearing on phonology, and from juncture abuse, was made at the expense of the introduction of diacritics (§696): the diacritic representation of morphosyntactic information (by # and the like at that time, later by the Prosodic Hierarchy) runs through the entire history of interface theory up to the present day. I take it to be the major plague on the representational side of the interface – the more so since at times it hides behind anti-diacritic lip service (see §698). In any event, the morpho-syntactic control over the distribution of boundaries (or any other phonological representative of morpho-syntactic information) is a headstone of generative identity in the realm of the inter-

574 Chap 3: Issues that are settled face. It was taken over in SPE and since then was never called into question: there are no boundaries in the middle of morphemes, and morphosyntax has no bearing on morpheme-internal phonology. There is massive empirical support to the end that morpho-syntax is toothless regarding morpheme-internal phonology: phonological effects of extra-phonological intervention are only encountered at morpho-syntactic breaks. Every phonologist has come across cases where the kind of phonology which is observed at morpheme edges is different from the phonology that applies morpheme-internally. This appears to be a robust crosslinguistic generalisation; it is observed by, among others, Rubach & Booij (1990), Piggott (1999) and Broselow (2003). Finally, it is important to note that the edge-interior asymmetry is not asymmetric in just any way: phonology is always regular, normal, unmarked, clean, well-behaved and so forth in the middle of morphemes, while edges typically introduce irregularities and exceptional patterns that cause theories to devise edge-specific mechanisms such as extrasyllabicity (or extra-metricality, extra-prosodicity for that matter).161

668

2. There is no phonetic correlate of morpho-syntactic information

669

2.1. Phonetic correlate of morpho-syntactic breaks: definition Another issue that roots in structuralist thinking but has a generative offspring in Natural Generative Phonology is the question whether morphosyntactic information, i.e. juncture or boundaries, have a direct phonetic expression. In the structuralist perspective, morpho-syntactic divisions are represented by juncture phonemes. Juncture phonemes are phonemes and therefore should have a phonetic correlate just like all other phonemes (§70). For his "open juncture" phoneme, Moulton (1947) for example proposes allophonic variation between a "pause" and "zero". It should be clear that what this discussion is about is not the influence that juncture phonemes or morpho-syntactic divisions have on 161

Another generalisation, which does not directly bear on the present discussion, is the fact that both edges are deviant, but not in the same way: left-edge and right-edge specificities are not the same. For example, the former introduces the typical TR-only restriction on clusters that is known from Indo-European languages, while the latter often allows for heavy clusters that are unheard of morpheme-internally (see Rubach & Booij 1990, Vol.1:§377). This issue is further discussed in Vol.2.

There is no phonetic correlate of morpho-syntactic information 575

neighbouring segments or phonological processes. This influence of course exists – it is the reason why the issue of morpho-syntactic bearing on phonology arises in the first place. Rather, a putative phonetic correlate of juncture is a surface manifestation of the object that conveys morpho-syntactic information itself. That is, we are talking about a phonetic exponent that is stable and occurs independently of whether and how juncture bears on phonological processes. Finally, it is to be noticed that talking about a phonetic exponent of a morpho-syntactic division makes only sense if morpho-syntactic intervention in phonology is local, i.e. materialises as a representational object that is inserted into the linear string. Juncture phonemes and SPE-style boundaries are local carriers of morpho-syntactic information, while domain-based prosodic constituency is not (the string is not enriched by any objects that can be identified in linear terms). The distinction between local and nonlocal representational intervention is further discussed in §706. 670

2.2. Generative diacritics make no noise, except in Natural Generative Phonology The question of a phonetic correlate of morpho-syntactic divisions has been handed over to generative theory, where it appears in form of the featural content of boundaries. In SPE, boundaries are regular segments which, like all other segments, are defined by features, one of which specifies that they are [-segment] (while real segments such as /p/ and /a/ are [+segment]) (§87). This is the generative way to make sure that boundaries have no phonetic manifestation: implicit in the [-segment] specification is that [-segment] segments are not pronounced. Note that just like in structuralism, the representational objects that carry morpho-syntactic information into phonology are variants of the phonological vocabulary that is used by the theory: phonemes then, segments now (and autosegmental domains later on in the autosegmental 80s, see §692). That morpho-syntactic information does not behave like phonemes or segments was shown early on (Pyle 1972, see §131), but phonologists then put a lot of effort into ignoring this fact. Until the early 80s, where it was used in order to replace one diacritic (boundaries) by another (prosodic constituency), claiming that this move does away with unacceptable diacritics (§369). Since then and up to the present day, phonology lives in the belief that ugly SPE-diacritics have been successfully eliminated, and that modern autosegmental theory is clean: in rule-based as much as in con-

576 Chap 3: Issues that are settled straint-based environments, the Prosodic Hierarchy stands unchallenged (or almost unchallenged: Pak 2008, Samuels 2009a, §580). What has really happened is that the linear diacritic that bore morpho-syntactic information was replaced by an autosegmental diacritic. This issue is further discussed in §694 – it is closely related to the question whether carriers of morphosyntactic information should be local (linear) or not (non-linear) (see §706). A generative revival of the idea that morpho-syntactic divisions have a phonetic correlate has occurred in the 70s in Natural Generative Phonology, which not only on this occasion walks in the footprints of structuralism. The basic idea was that "true" phonology can only process phonetic information: any regularity that needs to make reference to morphosyntactic information is a morpho-phonological, not a phonological rule (Prules vs. MP-rules) (§§127f). Unfortunately, there are countless cases where rules qualify as true P-rules (they are completely automatic, exceptionless and use only phonetic information), except that they are sensitive to a boundary. If boundaries need to be taken into account (something that even natural phonologists did not deny), then, they had to have a phonetic existence. Just as in structuralist times, this road led nowhere. Since Natural Generative Phonology, the absence of phonetic correlates of morphosyntactic information is not called into question anymore. It is safe to consider this issue settled.

671

3. Affix ordering is wrong Dorothy Siegel's (1974) Ph.D has introduced the generalisation known as affix ordering: class 1 and class 2 affixes may combine in all possible ways, except for [class 2 - class 1] sequences, which do not occur. This finding had an important impact on the development of interface theory: it paved the way for Lexical Phonology. SPE was aware of the existence of English affix classes (see §92), but ignored that they were also morphologically relevant. Siegel's affix ordering showed that the same empirical object – affix classes – has phonological as well as morphological effects. What was needed, thus, was a theory that accounts for both types of effects with the same mechanism. This conspiracy (alongside with the abstractness debate and Praguian segregation, see §154) produced Lexical Phonology: the stratal architecture could kill two birds with one stone (see §146).

Interpretation is inside-out, grammar is interactionist, brackets are out 577

Quite ironically, then, the initial spark of Lexical Phonology has turned out to be simply wrong. English features numerous words where a class 2 suffix occurs closer to the stem than a class 1 suffix (e.g. patent-abil2-ity1, organ-izat2-ion1, develop-ment2-al1, govern-ment2-al1); relevant discussion was reported in §243. Logically enough, the bankruptcy of affix ordering has then been used as an argument against Lexical Phonology and its stratal architecture by Halle & Vergnaud (Halle & Kenstowicz 1991:459, Halle et al. 1991:142). Finally, it may be worthwhile to have a second thought about the anglo-centristic character of affix classes and affix ordering: the literature (including this book) ruminates the same English examples for 30 years. Despite reasonable effort to evidence affix classes in other languages (see §143), though, they remain quite marginal a phenomenon crosslinguistically speaking.162 The case of affix ordering is worse: as far as I can see, no equivalent has ever been found elsewhere. We are thus left with a historical situation where the major theory of procedural interface management has grown on what appears to be a highly isolated pattern – in fact that unique a pattern that it does not exist at all: the empirical generalisation is wrong. This does not mean, of course, that Lexical Phonology is wrong as well: it is not infrequent in science that correct conclusions are drawn on wrong premises. The episode of affix ordering nonetheless leaves a displeasing aftertaste of anglo-centricity.

672

4. Interpretation is inside-out, grammar is interactionist, brackets are unnecessary relics

673

4.1. Inside-out interpretation and cyclic derivation

674

4.1.1. Two ways of organising inside-out interpretation: brackets vs. interactionism One of the oldest insights of generative grammar is that morphosyntactically complex strings are parsed from the root outwards upon interpretation, i.e. from the most to the least embedded chunk. Inside-out interpretation has been introduced by Chomsky et al. (1956) (see §80) and is 162

The desire to find affix classes elsewhere may sometimes have prompted spurious results: Dutch was known as an affix class-language for many years, but this was probably an erroneous belief: van Oostendorp (2004) shows that the alleged suffix classes in fact have a phonological definition (see §500).

578 Chap 3: Issues that are settled hard-wired in generative thinking at least since SPE, where it appears as the Transformational Cycle (later called the phonological cycle) (see §101). As far as I can see, if often tacitly, inside-out interpretation is accepted and practised by all versions of generative grammar: in syntactic, semantic and phonological quarters alike.163 There are two ways of organising inside-out interpretation: either brackets are inserted into the linear string that is then shipped to LF/PF and the two interpretational devices are able to "read" them in order to make their way through the pieces. Or grammar is interactionist, which means that chunks are spelled out one by one under morpho-syntactic control, i.e. moving up the tree, and come back from interpretation after having been assessed by LF and PF. On this count, concatenation and interpretation is interspersed (§146), and brackets are done away with. (281) two ways of organising inside-out interpretation a. inside-out interpretation assured b. inside-out interpretation assured by brackets by interactionism β Y

phon α

x

β

phon

Y

X spell-out

α x

SYNTAX

phon phon X spell-out

PHONOLOGY phonology receives and computes: [[X] Y] instruction: inside-out computation 163

phonology receives and computes: 1. X 2. XY no particular instruction

The only doubt that one could have is about parallel implementations of morpheme-specific phonologies which are practised in OT (co-phonologies and indexed constraints, see §477): here interpretation of pieces is not necessarily bound to morpho-syntactic structure. The relevant literature is not explicit about the issue, but it may be conjectured that even here interpretation proceeds from the most embedded to the least embedded chunk (see §481).

Interpretation is inside-out, grammar is interactionist, brackets are out 579

Before looking at which theory implements which solution, let us make the difference explicit. Table (281) above contrasts the two options. In the example, every node is spelled out, which is the state of affairs in SPE (see §103). How exactly the spell-out mechanism works is a separate issue that is discussed in the section on selective spell-out (§763). The following section fleshes out the last line of the table, i.e. the additional instruction that PF needs under (281a).

675

4.1.2. Brackets require a PF parsing device and make a procedural insight representational (and diacritic) A difference between (281a) and (281b) is that the former requires an additional instruction: it can only work if phonology is advised to find out which is the most embedded bracketed interpretational chunk of the string, and then to apply computation successively to less embedded bracketed chunks. This also means that phonology must be able to "read" brackets. That is, phonology must in fact be made of two separate devices, one identifying interpretational units in the bracketed string and keeping in mind which chunk is next, the other applying actual phonological computation to the sub-string that is currently delineated by the parsing device. As far as I can see, this consequence of (281a) is not made explicit in the literature. SPE takes brackets and their existence in phonology – augmented with morpho-syntactic labels such as NP etc. – for granted without discussion (see §95). It seems to be self-evident for Chomsky & Halle (1968) that phonology can read and interpret brackets. The only right that brackets are denied is to be referred to by phonological rules: these can make reference to translated (boundaries), but not to untranslated (brackets) morpho-syntactic structure (§98) (reference to morpho-syntactic content, i.e. the labels, is allowed all through). What this means is that PF is not exactly what it is meant to be in generative thinking, i.e. an interpretational device: it is more than that since it must also be able to carry out chunk-recognition, a job that is typically attributed to the spell-out mechanism which interprets morpho-syntactic structure, and which has certainly got nothing to do with the computation/interpretation of a string. In other words, (281a) forces phonology into replicating the labour of spell-out: chunk-recognition (i.e. the identification of interpretational units) is done twice, once in morpho-syntax, another time in phonology. This is certainly not something that (281a) can score with.

580 Chap 3: Issues that are settled On the other hand, no extra machinery or any specific instruction is needed under (281b): phonology applies phonological computation/interpretation to whatever string is submitted; there is no need for any chunk-recognition device in PF. Unlike (281a), (281b) is thus a good match of the regular generative description that makes PF a purely interpretational system. We will see in §797, however, that the interactionist option (281b) also requires an additional device, i.e. a memory-keeper: chunks that are interpreted must somehow be stored and pieced together at the end of the derivation (i.e. when spell-out arrives at the matrix CP) before they are pronounced. The contrast between (281a) and (281b) also appears when looking at brackets themselves: brackets are diacritic objects that the linear string which phonology computes is enriched with. Under (281b), the string is "clean" in the sense that it does not contain any diacritics in addition to the items that are taken into account by the phonological computation. Today there is a broad agreement (to which practice, however, does not really live up) that diacritics in phonology is not a good thing to have. §692 discusses this issue at length. Finally, it is to be noted that brackets transform a procedural insight into a representational issue. What inside-out interpretation expresses is the fact that first a chunk that has a given (embedded) morpho-syntactic status is interpreted, before another chunk (larger and less embedded) is processed. In the perspective of (281a), this parsing issue is "stored" in the representations that are sent to phonology by the means of brackets, a representational object; these are then converted back into procedural information by the parsing device that PF must accommodate. One may doubt that this wrapping of procedural information into a representational guise for shipping purposes, which requires encoding and decoding on both sides of the pipe, is really helpful. Especially in presence of an alternative that takes inside-out interpretation for what it is, a procedural insight, and does not need any additional machinery (PF parsing device, brackets).

676

4.2. The line-up of interface theories in regard of interactionism

677

4.2.1. Revolution (Lexical Phonology) and counter-revolution (Halle & Vergnaud) Let us now look at how generative theories of the interface have positioned themselves in regard of the two options under (281).

Interpretation is inside-out, grammar is interactionist, brackets are out 581

(281a) is the classical world-view that was consensual in traditional grammar and classical generative thinking (SPE on the phonological, Aspects up to GB on the syntactic side): it was taken for granted that morphosyntax and phonology (semantics) are strictly ordered. That is, morphosyntactic concatenation is entirely completed before phonological and semantic interpretation begins. "All concatenation before all interpretation" was long held to be part and parcel of the inverted T model. It was mentioned in §86 that this statement is an additional and entirely independent plug-in: the organisation of the grammatical architecture in terms of one concatenative and two interpretational devices makes no statement about the question whether or not concatenation and interpretation are interleaved. In the early 80s, interactionism as under (281b) barged into this settled and undisputed landscape. Interestingly, it was introduced by Lexical Phonology for practical, rather than for principled reasons, and in complete absence of any architectural motivation: interactionism allowed to kill two birds with one stone (morphological and phonological effects of affix classes, see §§146,671). From the orthodox generative vantage point, interactionism was a revolution – a revolution in phonology, though, which remained confined in its original duchy: as far as I can see, interactionism did not irradiate into syntax or semantics at all in the 80s, where business continued as if phonologists were not out of their mind. The response was therefore only phonological: Halle & Vergnaud (1987a) led the counter-revolution under the explicit banner of anti-interactionism in order to restore the authority of the principle "all concatenation before all interpretation" (§222).

678

4.2.2. Phonology and the interface are not the same thing Kaye (1995) takes an original position in this debate: he practices cherrypicking. That is, while Lexical Phonology was declining in the early 90s and phonology had become harshly anti-derivational over night with all new theories hooking on this trend (OT, Declarative Phonology, Government Phonology), he holds up interactionism (§275). On the other hand, Kaye (1995) implements selective spell-out and interpretation-triggering affixes, i.e. ideas that originate in Halle & Vergnaud's (1987a) antiinteractionist work. Mentioning Government Phonology in the list of anti-derivational theories may appear as a contradiction with Kaye's interactionism – but in

582 Chap 3: Issues that are settled fact it is not: Kaye was the first in the debate on derivationalism, I think, to carefully distinguish between phonological computation proper (his φ-function) and the mechanism that ships the result of the morpho-syntactic derivation to phonology (his concat-function) (§§267,271). While the former is strictly non-derivational in Kaye's view (there are no ordered rules), the latter is perfectly derivational, and there is no contradiction: phonological computation and the interface are two different things – from a modular point of view (§§622,649) there is no alternative anyway. This issue is also central for OT and the anti-cyclic attitude that is adopted by classical representatives of this theory. It was reported in §464 that the anti-cyclicity literature that dominates (or today perhaps: has dominated) the interface discussion in OT does not distinguish phonological computation and chunk submission (§469). Without discussion, antiderivationalism is taken to be a property of grammar as a whole, including the interface, and not just of the computational systems (such as phonology) that it is made of. This undifferentiated view of grammar certainly has a holding in the blurred modular contours, if any, that OT lives with. Or, in other words, the connectionist pieces that OT carries in its genetic endowment (see §529) contribute to this kind of globalising view on grammar that leads to the systematic violation of modularity (§523) and the nondistinction between phonology and its interface. Finally, note that the anti-cyclic position of OT goes way beyond simple anti-interactionism: it rejects in fact all versions of cyclic derivation, that is of inside-out interpretation. It is the cyclic nature of the parsing that is at stake – whether this is implemented by brackets or interactionism is irrelevant. Therefore OT has an issue with a very deep layer of generative thinking (see §465). Or rather, those versions of OT that are anti-cyclic: recall that just like Kaye (1995), Stratal OT and DOT, the modern heirs of Lexical Phonology, implement an interactionist architecture where strata are serially ordered, but phonological computation itself is strictly parallel (§483).

679

4.2.3. Derivation by phase: when generative grammar became interactionist By the end of the 90s, then, the generative mainstream – i.e. syntax – becomes interactionist as well: as a direct consequence of the minimalist programme, Epstein et al. (1998:46ff), Uriagereka (1999) and Chomsky (2000a et passim) make interactionism – which is known as derivation by phase in syntactic quarters – an absolutely central concept of syntactic the-

Interpretation is inside-out, grammar is interactionist, brackets are out 583

ory, and of the general architecture of grammar as a whole. This move is described in §304. Like before, though, morpho-syntax and phonology appear to be perfectly waterproof: just like non-interactionist syntactic theory of the 80s took no notice of the contemporary interactionist debate in phonology, present-day anti-interactionist (or even anti-cyclic) phonology, as far as I can see, shows no reaction to this major change in the interface landscape at all. Either anti-interactionism (or anti-cyclicity) continues to be promoted without mentioning that this overtly conflicts with current syntactic theory, or silent conversions take place: Morris Halle, the leader of the antiinteractionist movement of the 80s, is a prominent figure of interactionist Distributed Morphology today. Regarding the issue of morpheme-specific mini-grammars, §828 below argues that modern phase theory has direct consequences for phonological theory: intermodular argumentation is now possible, and this is all to the good (see §841, also Scheer 2008c, 2009b, 2010b). Finally, it is useful to look at phase theory and interactionism in the light of the pendular movement between procedural and representational orientations of (generative) grammar. In minimalist times, major tenets of syntactic theory in the 80s such as lexicalism and representationalism are called into question and charged at least in some quarters. Concomitantly, much of the spirit of the 60s is restored (generative semantics §721, antilexicalism to a certain extent, see §§539,569). Also, phase theory (including Phase Impenetrability) in some areas is a direct continuator (or rather: competitor) of representational tools of the 80. An obvious case in point is the formulation of Barriers (and bounding nodes) in terms of Phase Impenetrability (e.g. Rizzi 2005, Frascarelli 2006:2f, Boeckx & Grohmann 2007).

680

4.3. Only interactionism makes inside-out interpretation compatible with modularity

681

4.3.1. Modularity referees in favour of interactionism Brackets as under (281a) store morpho-syntactic structure in the linear string that is the input to phonological (and semantic) computation. They are incompatible with modularity since they represent untranslated morphosyntactic information in phonology: recall from §651 that intermodular (representational) communication requires translation. On top of that, they

584 Chap 3: Issues that are settled are diacritics in phonology, and diacritics are unwarranted anyway (more on that in §692). Now interactionism does away with brackets: it is the alternative organisation of inside-out interpretation. From the point of view of modularity and the diacritic issue, this is a highly beneficial effect. Or, in other words, the only way to make inside-out interpretation compatible with modularity is to have an interactionist architecture. This was pointed out on various occasions in the book (§§161,170,305).

682

4.3.2. Nobody used modularity in the 80s Modularity was thus relevant for Lexical Phonology and during the 80s when the debate with Halle & Vergnaud (1987a) and the anti-interactionist counter-revolution was led: it could have refereed the quarrel (in favour of interactionism). Although the development of modularity was exactly contemporary with Lexical Phonology (Pesetsky 1979, Kiparsky 1982a,b on the one hand, Fodor 1983 on the other), the argument from modularity has never been made as far as I can see. This ties in with the observation that was made in §414: on the representational side, the major interface theory of the 80s was born from the dispute with direct syntax approaches that allow for free reference to morpho-syntactic structure and content (labels). Against this practice, Prosodic Phonology has introduced the principle of Indirect Reference which requires that phonology can only make reference to translated morphosyntactic information – exactly what modularity imposes. But here again, although contemporary (Selkirk 1984, Nespor & Vogel 1986 on the one hand, Fodor 1983 on the other), modularity was never mentioned. Instead, the notion of non-isomorphism was invoked (which was held to be a fact about language, but which on closer inspection turns out to be a fact about the domain-based bias of Prosodic Phonology, see §416). It is certainly true that modularity was always a core property of the generative architecture of grammar (see §623) – but it is also true that it remained an intro-class topic without much consequence on actual linguistic analysis in the 80s (see §630). In the discussion of Phase Theory since the late 90s, this has not changed substantially as far as I can see: modularity as an operative and refereeing principle is more or less absent from the debate. In any event, it seems reasonable to conclude that the interpretation of morpho-syntactic structure is inside-out, and that grammar is interaction-

Interface Dualism 585

ist. Brackets are an unwarranted relic of times when linguists had understood that interpretation is from the root outwards, but had not considered the possibility that concatenation and interpretation are interleaved.

683

5. Interface Dualism

684

5.1. If tacitly, (almost) all theories implement Interface Dualism The prism that I use in this book in order to look at the interface is the distinction between a procedural and a representational channel of communication that morpho-syntax uses in order to talk to phonology. Interface Dualism is the claim that both channels are needed and real: an interface theory that denies the existence of either one, or which does not use either, is on the wrong track. Interface Dualism may appear to be trivial: all theories of the interface that we have come across in the historical survey in one way or another implement both procedural and representational communication. Well, almost all (see the two following sections). Also, the two major interface theories that still today dominate the landscape have been selectively focusing on either channel: Lexical Phonology has carried out an extreme proceduralisation of the interface (see §213), while Prosodic Phonology is an exclusively representational theory. The two theories were contemporary, but in spite of a significant and potentially conflictual intersection (see §429) there was no real dialogue. The official and diplomatic statement simply claimed peaceful coexistence (see §423). Interface Dualism as understood in this book not only requires interface theories to accept the existence of a procedural and a representational channel; it also means that a sound interface theory needs to think of how the two channels interact. That is, peaceful coexistence ("the others do whatever they want, this is none of our business") is simply a wrong statement: there is competition between procedural and representational solutions (see §§320,429). The two issues at hand are discussed in two sections below, one regarding the division of labour between procedural and representational mechanisms (§748, a question that remains open in this book), the other inquiring on an important source of competition, i.e. the alleged absence of phonological effects of cyclic spell-out above the word level (§786). In any event, the claim of this book is that whatever its specific orientations, a correct interface theory needs to provide for a procedural and a

586 Chap 3: Issues that are settled representational way of talking to phonology. I have taken the liberty to place Interface Dualism in the chapter regarding issues that are settled. It was mentioned that it is largely undisputed: if tacitly, (almost) all interface theories provide for representational and procedural communication. Except two cases that were mentioned in Part I: Kaye (1995) and orthodox OT. These are examined below. 685

5.2. Representational communication needed (contra Kaye 1995) Kaye (1995) explicitly refuses representational communication: he holds that the only way for morpho-syntax to bear on phonology is through cyclic spell-out (see §276). Boundary information concerns edges of morphemes and words, which phonological processes make regular reference to. Procedural mechanisms are unable to take on this function by definition since they cannot refer to specific locations in a string. In fact Kaye's system does allow for reference to edges, but only to phase edges (or domain edges in his terminology). For example, /mn/ is reduced to [m] in domain-final position (damn, damn-ing vs. damn-ation, see §321). Kaye argues that this is different from regular representational communication since no specific representational object is inserted into the linear string. The edge of a domain is simply defined by the fact that nothing precedes or nothing follows. However, there is evidence to the end that this is not good enough: edges produce stable phonological effects across languages (see Vol.1:§87, Ségéral & Scheer 2008, Scheer 2009a,c). For example, if the beginning of the word produces a particular effect in a given language, this effect will not be arbitrary: word-initial consonants are observed to be strong (rather than weak), word-initial consonant clusters are restricted to obstruentsonorant sequences (rather than to sonorant-obstruent etc.), and first vowels of words are prevented from alternating with zero. Hence there is a "conspiracy" of the beginning of the word, and the fact of being domain-initial does not produce any indication whether consonants should be strong or weak etc. Not any more by the way than diacritics such as # or PrW: anything and its reverse can happen in the vicinity of domain edges, # and PrW. This pattern is called the Direct Effect in Vol.2 and Scheer 2009a,c because what is needed is a non-diacritic carrier of morpho-syntactic information: this is argued to be an empty CV unit which, unlike diacritic

Interface Dualism 587

sleepers (that only impact phonology when a rule or a constraint makes reference to them) will provoke a phonological effect no matter what. The two word edges and the way in which they bear on phonology are discussed at greater length in Vol.2.

686

5.3. Procedural communication needed (contra orthodox OT) On the other end of the spectrum, the anti-derivational foundations of OT have produced the orthodox position of this theory where the cycle and any form of cyclic derivation is rejected out of hand (§§465,475). This, however, is not a compelling position for OT: rather than following from OT's commitment to parallelism, it follows from the nonperception of the internal structure of grammar. That is, OT applies the antiderivational request holistically to the grammar as such without even considering the possibility that it could apply to grammatical sub-systems such as phonology, but not to the system that relates these sub-systems, i.e. the interface (see §469). Nothing withstands a version of OT where modular contours are sharp, i.e. where phonological computation is distinct from morpho-syntactic and semantic computation, and where intermodular communication is still an independent shipping-system. This view is indeed defended by Stratal OT and DOT (see §483): phonological computation is perfectly parallel, but there are multiple constraint rankings that are cyclically fed by an independent shipping mechanism. It was suggested in §529 that OT's trope for scrambling everything is due to the connectionist half of its genetic endowment. Connectionism materialises namely as parallel computation, and works against the generative half of OT's genetic code, which favours modularity. The result is a situation where modularity is acknowledged, but constantly violated (§523). Particularly heavy weighs the fact that the rejection of cyclicity also throws inside-out interpretation over board, one of the deepest generative insights (§§468,672). The fact that morpho-syntactic structure is interpreted from inside out – nobody seriously doubts that it is – certainly constitutes a strong argument in favour of a cyclic shipping mechanism: intermodular communication is serial. In the same way, concomitant phonological and semantic effects of a given morpho-syntactic structure (cómparable vs. compárable, see §§545,564) provide strong empirical evidence for the existence of a cyclic spell-out mechanism. Finally, the fact that cyclic spell-out is a critical property of current syntactic theory since Chomsky's (2000a et passim) derivation by phase

588 Chap 3: Issues that are settled (§304) adds to the trouble that phonologists who reject procedural chunksubmission are in: they will have to claim that modern syntax – actually the entire generative architecture of grammar – is flat out wrong. Against this backdrop, it needs to be noted that OT is much less antiderivational today than it was in the 90s, when orthodox antiderivationalism was coined. It was mentioned that Stratal OT and DOT carefully distinguish communication at the interface(s) from intra-modular, i.e. phonological computation: the latter is parallel, while the former is derivational. But more recently McCarthy (2007) has given up on parallelism altogether, including within phonological computation itself: he revives Harmonic Serialism (under the label of OT-CC), a derivational option entertained by Prince & Smolensky (1993) where the result of parallel constraint evaluation loops back into GEN until no harmonic improvement can be achieved anymore.

687

Chapter 4 Modularity, translation, the diacritic issue and local vs. domain-based intervention: settled in verb, but not in fact

688

1. Introduction

689

1.1. Three questions: two are settled, one has never been discussed This chapter is about two questions that have been discussed at some point in the history of interface theory, but today are taken to be solved. It is also about an issue which, although being critical for the way morpho-syntax talks to phonology through the representational channel, has never been debated as far as I can see. The two former questions are modularity (and its major consequence, translation) on the one hand, and the diacritic issue on the other hand. The latter is the question how exactly carriers of (non-morphemic) morpho-syntactic information intervene in phonology: locally (i.e. as a piece in the linear string that is located between two morphemes) or not (i.e. in form of autosegmental domains that cannot be localised in the linear string). The former follow the local philosophy of Trubetzkoy's Grenzsignale, juncture phonemes and SPE-style boundaries, while the latter are represented by prosodic constituency. This contrast was identified in §3660, where it was also pointed out that the transition from linear SPE-type boundaries to non-linear domainbased carriers was not identified as such: as far as I can see, nobody has ever evaluated whether morpho-syntactic intervention in phonology should be local or non-local, what the consequences and the predictions are, what kind of evidence pleads in favour or disfavour of either view and so on (see §365 where the transition from boundaries to domains is described, and especially the summary in §376).

690

1.2. Modularity and diacritics are only settled in verb The two questions that have been discussed in the literature – modularity/translation and the diacritic issue – could be accommodated in either the

590 Chap 4: Issues settled in verb, but not in fact chapter on settled or on open questions. They appear to be settled because everybody agrees that modularity is part and parcel of the generative architecture of grammar – but then modularity is violated by a number of generative theories (and namely by OT, §§469,523) in a number of different ways. Everybody also agrees that diacritics ought to be banned from phonology – but then phonology happily lives with autosegmental diacritics in lieu of linear SPE-type boundaries since the early 80s (§399). The two issues are thus settled in verb, but not in fact, and this may be the reason why nobody talks about them: OT hardly discusses architectural properties (§524), and the diacritic character of the Prosodic Hierarchy is not even remotely hinted at. That is, he field lives in the belief that the diacritic issue was solved when ugly SPE-style boundaries were replaced by beautiful and "truly phonological" (sic, see §405) prosodic constituency.

691

1.3. The three questions are the backbone of the representational channel Together with the locality-of-intervention question that is in need of discussion, modularity/translation and diacritics are the backbone of what representational communication with phonology is about. One could actually say that the representational side of the interface reduces to these three questions. Vol.2 is about these issues, and they actually made me write the book in the first place: the theory I am working in, CVCV, has got something to say regarding 1) the non-diacritic output of translation (CV units), 2) the locality of morpho-syntactic intervention and, by way of consequence, 3) the existence and properties of the translational process. The historical survey in Part I may thus be viewed as a prelude to the contribution of Vol.2, and the present chapter is designed to explain that there is an issue at all: the (non-)respect of modularity, translation and the No Diacritics! requirement may appear to be an old hat. They are not.

692

2. Translation in structuralist and generative interface theory

693

2.1. Interface design was done in absence of a modular/cognitive background – but translation has always been practised On modular assumptions, translation is obligatory: whatever information morpho-syntax wants to be carried by an item that phonology can make reference to, this item must be the output of a translational process (§650).

Translation in structuralist and generative interface theory 591

This cognitive foundation was entirely absent from interface theory until very recently, and today still is in some quarters. Structuralists for sure were not working with a cognitive background. Generative linguistics and its interface architecture surely implement cognitive/modular ideas: Noam Chomsky actively participated in the computational movement that, following the Turing - von Neumann programme, founded Cognitive Science in the 50s (§§603,623). This notwithstanding, as far as I can see there is no identifiable and explicit footprint of modular ideas when it comes to translation beyond a general sympathy for modularity. We have seen indeed that generative phonology has violated modularity/translation on a number of occasions (including SPE, see §95, more on this in §702 below), and with the exception of the work by Charles Reiss, Eric Raimy and colleagues (see §587, note 148), I am not aware of any explicit mention of modularity in the older (or even more recent) literature that would have been used in order to impose translation. What is more, we have come across two cases where modularity would have been a decisive referee in ongoing debate, but was not used in the argumentation: Indirect Reference (§414) and interactionism (§680). What I want to say is that in spite of what appears to be a complete absence of modular background in the operational design of interface theories, translation has always been present. Modularity, then, appears to be some kind of cognitive post-hoc confirmation of what linguists have always done anyway – to which degree and with what kind of exceptions is discussed below. 694

2.2. The birth and variable incarnation of diacritics

695

2.2.1. Juncture phonemes and SPE-type boundaries: diacritic translation and various degrees of camouflage §61 (and more generally the chapter on structuralism, i.e. §59) has described how structuralism tried to make the carrier of morpho-syntactic information a truly phonological object that has got nothing to do with morphology. Juncture phonemes were the result of the descriptivism-rooted requirement of Level Independence: the bottom-up discovery procedure did not allow phonology to contain any morpho-syntactic information. Much effort was put into the camouflage of the extra-phonological identity of morpho-syntactic information, which is of course necessary for

592 Chap 4: Issues settled in verb, but not in fact phonological description: as indicated by their name, juncture phonemes were supposed to be phonemes, that is truly phonological objects. SPE basically does the same thing, only that the phonological currency has changed in the meantime: phonology is now made of segments (rather than of phonemes), which means that boundaries are [-segment] segments (§87). The generative camouflage is not really less outlandish than the one of its structuralist predecessor: # is not any more a segment than it is a phoneme. The unwarranted consequences of its alleged segmental status have been made explicit early on, namely by Pyle (1972) (see §136). The real difference with the structuralist strategy is the admitted hybridity of the carrier of morpho-syntactic information. Instead of denying the morpho-syntactic origin of boundaries, the generative perspective organises its identity as a morpho-syntactic agent in phonology (see §§90f). Boundaries are thus supposed to be truly phonological units (i.e. segments) and carriers of morpho-syntactic information at the same time. This masquerade, however, was entirely transparent right from the start: unlike in structuralist theory where the phonemic status was given real credit (see §701 below), hardly anybody took the [-segment] camouflage seriously. The naked # was taken for what it really is in all phonological quarters: a unit whose only purpose is to store and to release morphosyntactic information.

696

2.2.2. The abandon of Level Independence makes boundaries diacritics A direct consequence of this situation is the emergence of the diacritic issue: if nobody believes that boundaries are segments, they must be nonphonological objects – diacritics (this notion was defined in §405; it is further discussed in Vol.2). The legalisation of morpho-syntactic information in phonology has thus prompted the diacritic issue – which is here to stay. The question what arbitrarily chosen symbols have got to do with phonology, or why phonology, but not other linguistic modules, should have them, is a major concern in interface theory since SPE and until the introduction of prosodic constituency in the early 80s, which is (wrongly) believed to have solved the problem.

Translation in structuralist and generative interface theory 593 697

2.2.3. Since structuralism, the output of translation has always been a diacritic At least since the late 70s, enough discomfort with diacritic boundaries was accumulated (see §131) to make the disqualification of diacritics broadly consensual. This, however, does not mean that a better solution was in sight. A radical alternative was to do away with boundaries and to replace them by nothing, i.e. to give up on translation altogether (direct syntax); §702 below discusses this option where phonological rules make direct reference to morpho-syntactic categories. The mainstream, though, replaced linear diacritics (boundaries) by autosegmental diacritics (the Prosodic Hierarchy). We have seen in §§373,402 that, paradoxically enough, the argument which was used in order to promote this move was precisely the claim that diacritics do not qualify Since structuralist times and up to the present day, then, the output of translation has always been a diacritic, and the degree of awareness of this fact among the protagonists of the respective theories is variable: high for SPE-type boundaries, choked but sensible for juncture phonemes, completely absent for prosodic constituency.

698

2.2.4. No Diacritics ! – no diacritics ? Vol.2 takes up the diacritic issue: truly non-diacritic carriers of morphosyntactic information are argued for. These can only be recruited among the basic phonological vocabulary that exists anyway and in absence of any issue related to the interface. In the present context where the Prosodic Hierarchy stands (almost) unchallenged, this may appear to be preaching to the converted: the received chronicle has booked diacritics as a solved problem since the early 80s when ugly diacritic boundaries were replaced by "truly phonological" (sic, see §405) prosodic constituents. This move was described in §365 and is further discussed in §711 below. The point is that the converted were converted because they had a strong desire to leave the old faith, rather than because the new religion was intrinsically appealing: it did not take a lot in the late 70s to convince boundary-frustrated phonologists (see §131) that an alternative solution offers relief.

594 Chap 4: Issues settled in verb, but not in fact 699

2.3. Modularity and translation were invented by structuralism

700

2.3.1. Non-cognitive modularity: Level Independence enforces translation The dismissal of Level Independence and hence the recognition that morpho-syntactic information plays a role in phonology is typically quoted when it comes to explain the difference between structuralist and generative theory (e.g. St. Anderson 1985:313ff, Durand 2006:2266, Aronoff 2003). This is certainly correct – but it spots light only on one side of the coin. Disqualifying Level Independence as a description-based relic of naive structuralist times is nearsighted. For Level Independence also expresses the idea that morpho-syntax and phonology are incommunicado, that they are two distinct ontological entities – exactly the insight of modularity. Structuralist practice was then to circumvent the prohibition to use morphological information in phonology by its translation into a truly phonological object – a (juncture) phoneme. This gross camouflage may be sniggered at from hindsight. But, as was argued in §§72,415, this would fall short of structuralist thinking. Willingly or unwillingly, consciously or not, structuralists were the first linguists to translate morpho-syntactic into phonological vocabulary. And it was Level Independence that forced them to do so. Level Independence and ensuing translation may therefore be considered the birth of modular thinking in linguistics, albeit on grounds that have got nothing to do with any cognitive perspective.

701

2.3.2. Translation affords the assessment of phonological theories according to their behaviour at the interface Interestingly enough, structuralist translation was far more serious than subsequent translational attempts. For the output identity – a phoneme – was taken seriously: as all other phonemes, juncture phonemes had to enjoy free distribution (and hence independence from morpho-syntactic divisions, see §§66f), and they had to have a phonetic correlate (§70). That this led to absurd results (juncture abuse, see §69) is all to the good: it falsifies a phonological theory that is based on this particular kind of translation. This opens the possibility for phonological theories to be assessed (or falsified) by their behaviour at the interface – if only the arbitral award of

Translation in structuralist and generative interface theory 595

the interface is taken seriously. That is, when the specific vocabulary of a given phonological theory makes outlandish predictions, if translation is not given up on and if diacritics are not a possible output of translation, it may be concluded that the vocabulary (phonemes, segments, autosegmental trees) is not appropriate: the phonological theory needs to be modified (see §138). For example, had the absurd consequences of SPE-type boundaries that were pointed out by Pyle (1972) and Rotenberg (1978) (§131) been taken seriously, phonologists would have been forced to change the then current phonological theory. That is, the outlandish behaviour of # at the interface would have enforced the conclusion that segments, of which # was supposed to be a sub-species, are not the adequate interface currency. In other words: there must be something else than just segments in phonology: autosegmental representations.

702

2.4. Generative modularity offenders: reference to untranslated morphosyntactic information

703

2.4.1. Translation was not a standard in generative theory until the mid 80s When talking about translation in SPE, the diacritic character of boundaries is only a minor problem when compared to the systematic offense that is made to modularity. It was shown in §95 that SPE freely allows for the presence of untranslated morpho-syntactic information in phonology: both morpho-syntactic structure and its labels (NP, VP etc.) are constantly available in untranslated guise. The former, however, are only there to run inside-out interpretation: phonological rules cannot refer to brackets (§97) SPE thus provides for both translated (boundaries) and untranslated (morpho-syntactic labels) reference to morpho-syntactic information. This being said, there was no clear rationale for a division of labour: should a given phonological rule make reference to morpho-syntactic information rather in translated or in untranslated form? In the late 70s, the post-SPE practice led to a general frustration regarding the diacritic issue (§131). In recognition of the problems that piled up, some voices called for the complete elimination of translated information (boundaries) in favour of direct reference to morpho-syntactic categories. This strand condensed in the 80s under the label of direct syntax approaches and became the major rival of Prosodic Phonology (§407), whose founding statement is Indirect Reference (§§378,406). This principle – the

596 Chap 4: Issues settled in verb, but not in fact ban of any untranslated reference to morpho-syntactic information – has become the baseline of generative interface theory on the representational side. 704

2.4.2. Turning back the wheel: weak and strong modularity offenders in (more or less) recent development Hence it is only since the advent of Prosodic Phonology in the 80s that generative theory meets the standards that were set by structuralist Level Independence, and later on by modularity. In this sense, structuralism was far more generative than SPE, and it took generative phonology twenty years to catch up with the modernity of Level Independence. More recently, however, the clear modular waters of Indirect Reference are muddied again by a number of approaches that allow phonology to make reference to untranslated morpho-syntactic information. These fall into two kinds: those that work on a modular basis where morpho-syntax and phonology are clearly distinct ontological spaces and computational systems but allow for direct reference to morpho-syntactic information (weak offenders), and those where the existence of distinct computational systems is either unclear or overtly denied (strong offenders). Weak modularity offenders include SPE, the direct syntax approaches of the 70s and 80s and more recently van Oostendorp's Coloured Containment (§503). Strong modularity offenders are OT as a whole (§§469,523) and Distributed Morphology. In DM, PF movement is incompatible with modularity because it supposes that a computation simultaneously accesses morpho-syntactic and phonological vocabulary (§§574,580), and current analyses also interleave PF operations such as linearisation with phonological rules (§739). In OT and DM the existence and/or the contours of distinct modules is unclear. They are denied altogether in Sign-Based Morphology, an outgrowth of HPSG (§512). Recall that the OT-trope to scramble all linguistic facts into one single constraint ranking is probably not unrelated to its connectionist roots (§529): connectionism is an all-purpose computational theory of everything that makes content-unspecificity a programmatic claim (see §§597f). Also, PF as conceived of in current minimalist syntax is a strong modularity offender: in PF syntactic-looking operations are carried out because they displease in narrow syntax (clean syntax, dirty PF). On the consensual assumption that vocabulary insertion occurs upon spell-out of narrow syntax, PF (or the subset of PF that excludes what phonologists call

Translation in structuralist and generative interface theory 597

phonology) is an ill-defined intermundia where computation needs to simultaneously access morpho-syntactic and phonological vocabulary – a violation of domain specificity (§§738,747). Finally, the diacritic issue also violates modularity, if in a much weaker sense: all generative (and non-generative) theories to date propose diacritics as an output of translation (juncture phonemes, SPE-type boundaries, prosodic constituency). This is incompatible with domain specificity (§611): like all other modules, phonology can only parse and interpret its own, proprietary vocabulary. 705

2.4.3. A note on Structural Analogy Dependency Phonology promotes the notion of Structural Analogy, e.g. J. Anderson (1985, 1986, 1987, 1992), Anderson & Ewen (1987:283ff), Durand (1995). In Hulst's (2000:209) words, "grammar recapitulates, rather than proliferates, structures and principles". Government Phonology, a sister theory, has adopted this idea: a number of devices that are (or were) developed by Government Phonology are directly motivated by syntactic theory and phenomena. Most prominent among those are government, licensing and the projection principle (Kaye et al. 1990).164 The question is whether Structural Analogy violates modularity, or at least favours a bias towards modularity violation: it appears to make phonology and morpho-syntax similar, rather than distinct. It seems to me that the issue cannot be decided ex cathedra: instruments that root in Structural Analogy do not violate modularity per se. Modularity violation is about denying the existence of distinct computational systems, or about blurring their contours. It is also about having vocabulary manipulated in a computational system that is not domain specific to this system. None of these is an automatic or necessary consequence of Structural Analogy. For example, Dependency Phonology and Government Phonology have never claimed that phonology and syntax share a common vocabulary over which computation is performed. Or government for instance (beyond the fact that today it is not the same device anymore in Government Phonology, and has disappeared from current syntax) does not violate modularity in any way that I can see. This also goes for a recent application of Rela164

Other instances are so-called prosodic government (which applies c-command to phonology, Lowenstamm & Kaye 1986:115, Lowenstamm 1989) and a phonological version of the Minimality Condition (Charette 1989, Harris 1994:169f, Kaye et al. 1990:224f).

598 Chap 4: Issues settled in verb, but not in fact tivized Minimality (Rizzi 1990) to phonology (Brun-Trigaud & Scheer 2010): modularity is unaffected if locality conditions on syntactic and phonological processes are similar or identical.

706

3. Local vs. non-local carriers of morpho-syntactic information

707

3.1. Local boundaries vs. non-local domain-based intervention It was pointed out in §3660 that the major cultural break on the representational side of the interface since the 19th century has occurred when Prosodic Phonology replaced local boundaries by non-local domains in the early 80s. That is, phonologists have always thought of carriers of morphosyntactic information in terms of items that are inserted into the linear string, i.e. which are immediately preceded and followed by some phonological symbol (that represents morphemic information). This is true for neogrammarian, structuralist and generative thinking: Trubetzkoy's Grenzsignale, juncture phonemes and SPE-style boundaries represent this conception. In the early 80s, however, a previously unconsidered non-local alternative, prosodic constituency, took over. Since then and up to the present day, the domain-based nature of translated morpho-syntactic information stands unchallenged (§365 traces back this evolution in greater detail). The locality of intervention is well incarnated by the traditional notion of sandhi: sandhi phenomena occur at the break of two morphemes (internal) or words (external), and are triggered by this division. Following the SPE-prototype, let us call boundary any local carrier of morpho-syntactic information, as opposed to domain, which is a nonlocal means of carrying morpho-syntactic information into phonology. Table (282) below contrasts the two options; in both cases, carriers of extraphonological information appear as ¥. Morpho-syntactic units (morphemes or words) appear as a linearised sequence of pieces. (282) local vs. non-local intervention in phonology a. local: b. non-local: ¥ is inserted at morpho-syntactic ¥ dominates a number of pieces divisions ¥ [piece 1] ¥ [piece 2] ¥ [piece 3]

[piece 1] [piece 2] [piece 3]

Local vs. non-local carriers of morpho-syntactic information 599

The domain-based option under (282b) is what has become broadly consensual in generative phonology since the introduction of the Prosodic Hierarchy in the early 80s. The difference with respect to the traditional local alternative under (282a) is that a number of pieces of the linear string are spanned by a domain. That is, labelled clusters are created: an individual piece belongs to a domain (a ¥). By contrast under (282a), a piece cannot belong to a boundary (to a ¥). On the other hand, a boundary has a linear location: it follows some piece, and precedes some other piece. It does not make sense to talk about domains that intervene between two pieces: domains are made of pieces, but they are not defined by a linear precedence relation with the items that they dominate. 708

3.2. Notational variants and real differences

709

3.2.1. Domain-based can be translated into local reference and vice-versa In the original conception of Prosodic Phonology, there were three ways for a phonological rule to make reference to prosodic domains (Selkirk 1980a:111f, Nespor & Vogel 1986:77). Table (283) below repeats them from §384 for convenience. (283) rule types in Prosodic Phonology a. domain span rule b. domain juncture rule c. domain limit rule

Domain span rules apply in a certain segmental environment iff this environment is contained within a specific domain. Domain juncture rules apply when some part of the segmental string needs to be adjacent to a boundary that separates two items of the same prosodic unit. Finally, domain limit rules apply when certain segmental conditions are met, but only at the beginning or at the end of a given unit of the prosodic constituency. Leaving the latter aside (which has not experienced a substantial empirical echo), it is clear that domain-based intervention is actually a superset of local intervention. That is, domain juncture rules can reproduce the local effect, but in addition domain span rules can make reference to something that local boundary-based intervention appears to be unable to express: a phonological effect that applies to several morpho-syntactic pieces in a row.

600 Chap 4: Issues settled in verb, but not in fact This impression is wrong: just as much as domains can express local reference, local boundaries can express generalisations that apply to a set of neighbouring pieces. We have actually come across an exemplary case in §419 where the domain-based bias of Prosodic Phonology led to a wrong conclusion: non-isomorphism is a fact about the prism that Prosodic Phonology uses in order to look at the evidence, not about the linguistic fact itself. The example discussed since SPE is about a cat, a rat and cheese. It is repeated under (284) below from §418 for convenience. (284) a. This is [the cat that caught [the rat that stole [the cheese]]] b. [This is the cat] [that caught the rat] [that stole the cheese]

The major syntactic divisions of the sentence under (284a) do not coincide with its intonational structure under (284b). Hence, goes the argument since SPE, whatever drives phonology to decide that intonation is as under (284b), it is not the output of the syntactic module: there is no node in the syntactic tree that uniquely dominates every intonational span of the sentence. This discrepancy between (284a) and (284b) is what Prosodic Phonology (and the generative mainstream since then) calls nonisomorphism. When we look at the same facts through the glasses of local boundaries (instead of through the domain-based prism), though, there is no nonisomorphism at all: intonational chunks simply begin with every CP. Hence it is enough to say that the intonation-building mechanism starts a new unit every time it hits a CP-boundary. Intonational and syntactic structure are thus perfectly isomorphic, and no specific autosegmental constituency on the phonological side is needed in order to express the relevant generalisation. 710

3.2.2. The difference is conceptual, not empirical Abstracting away from this specific example, the following is true: any generalisation that involves a uniform behaviour of several neighbouring pieces of the linear string, and which is intuitively formulated in terms of a domain, may be translated into local terms. What it takes to do that is to insert a particular boundary (a ¥ in the terms of (282)) between those pieces that belong to the domain at hand (and eventually before the first and after the last piece), and nowhere else. A process that makes reference to ¥ will

Local vs. non-local carriers of morpho-syntactic information 601

then target exactly the those morpho-syntactic breaks that are described by the original domain. Note that it does not matter whether the piece-transitions that need to receive a boundary are a homogeneous (or natural) class on the morphosyntactic side. The question which class of morpho-syntactic divisions provokes a given phonological effect is what I call the mapping puzzle (§753). Whatever the output of translation, boundaries or domains, it needs to be created through some mapping process, and domains do not represent morpho-syntactic structure any "more directly" than boundaries. It is thus hard to imagine an empirical situation that can be accounted for by local, but not by domain-based intervention, or vice-versa. The conclusion, then, is that the difference is conceptual, rather than empirical. 711

3.3. The local baby and the diacritic bathwater Below the local and the non-local options are compared regarding their conceptual properties. Before this is done, however, it may be worthwhile to recall that the issue has never been discussed. What was discussed instead is something else: boundaries were outlawed because they are diacritics (§373), against domains which were held to be "truly phonological" objects (see §405). It did not occur to anybody to even ask the question whether domains are diacritics, or to prove that they are not. Prosodic constituency was a non-diacritic self-runner. The fact is that by all standards they are as much a diacritic as SPE-type boundaries (see §402). Beyond the diacritic argument that was made, it is hard to find any other anti-boundary argument in the Prosodic Phonology literature (see §369): boundaries were not really an issue. They were considered oldfashioned, and they were discredited anyway by a decade of frustrating practice (§131). The real reasons for the fact that the entire field subscribed to the overnight take-over of prosodic constituency may thus well be a frustration-born "anything but boundaries" on the one hand, and the simple application of the new autosegmental technology to the interface on the other hand. If segmental, syllabic and other structure is autosegmental, so must be the carrier of morpho-syntactic information. The fact that the Prosodic Hierarchy is a child of autosegmentalism (rather than of an empirical or a conceptual issue with boundaries) was shown in §368. In sum, then, what happened was a double fault: boundaries were dismissed because they were diacritics, but just in order to be replaced by another diacritic; and the local vs. non-local question was decided without

602 Chap 4: Issues settled in verb, but not in fact having been discussed: the local option was thrown over board together with (diacritic) boundaries. Or, in other words, the local baby was thrown out with the diacritic bathwater. Both properties, however, are logically independent: a local boundary may or may not be diacritic, and so may be a non-local domain (at least in principle, more on this shortly). This does not help us to decide whether morpho-syntactic intervention in phonology should be local or not – but at least locality may now be re-considered as a relevant question.

712

3.4. There can be non-diacritic boundaries, but what would a non-diacritic domain look like?

713

3.4.1. Non-diacritic boundaries (can) exist In order to decide whether local or non-local intervention is correct, let us thus ask the question what a non-diacritic boundary would look like, and what a non-diacritic domain could be. Table (285) below repeats the definition of what counts as a diacritic that was used in §405. (285) definition of the term "diacritic" a diacritic is a non-native object in module X: it is only used when information from outside of X is processed. It is absent from events that do not appeal to extra-Xal information.

Hashmarks and constituents of the Prosodic Hierarchy (i.e. omegas) meet the conditions: they are non-phonological intruders in the phonological world which do labour only when phonology appeals to extraphonological information, and which are systematically absent from phonological processes that do not use extra-phonological information (such as ordinary palatalisations for example). A non-diacritic is thus an object that exists in phonology anyway, even in absence of any appeal to extra-phonological information. Lass' (1971) proposal that a boundary materialises as the feature [-voice] in phonology (see §135) therefore satisfies the non-diacritic requirement: [-voice] exists in phonological processes that have got nothing to do with extraphonological information. We know that in fact melodic primes are not good candidates for the output of translation because melody and morphosyntax are incommunicado (see §663). Lowenstamm (1999) has introduced another type of carrier of morpho-syntactic information: syllabic space. In his theory (Government Pho-

Local vs. non-local carriers of morpho-syntactic information 603

nology in general, CVCV in particular, see Lowenstamm 1996, Scheer 2004a), the minimal syllabic item is an empty CV unit, and it is an empty CV that marks the beginning of the word. Three cross-linguistically stable effects of the initial CV have been identified: they are briefly described in §685 (and at greater length in Vol.1:§§87,402, Ségéral & Scheer 2008, Scheer 2009a,c). In subsequent literature, interface phenomena that are unrelated to the left edge of the word have also been analysed in terms of syllabic space: boundary information is found to be carried by CV units in the analysis of the negative in Kabyle Berber (Bendjaballah 2001), a verbal marker in Chleuh Berber (Lahrouchi 2001) or tense in German strong verbs (Bendjaballah & Haiden 2003a,b). An empty CV unit also carries morphosyntactic information in Guerssel & Lowenstamm's (1990) analysis of Classical Arabic (what they call the derivational syllable) and Pagliano's (2003) analysis of intrusive consonants in French. These cases are discussed at length in Vol.2. Their enumeration at this stage of the discussion is only meant to show that there is an alternative to diacritic boundaries: syllabic space is certainly not a diacritic since it is a necessary ingredient of phonology even in absence of extra-phonological factors. At the same time, it is local, rather than domain-based: just like SPE-type boundaries, syllabic space is necessarily inserted into the linear string at morpho-syntactic divisions. Note that the particular incarnation of syllabic space is irrelevant for the argument: as x-slots, regular syllabic constituents (onset etc.) or the CV units of Government Phonology and CVCV. If the question is thus whether local boundaries can be non-diacritic, the answer is yes: counter to what early Prosodic Phonology wanted to make phonologists believe, nothing withstands the existence of nondiacritic boundaries. One solution is worked out in Government Phonology, but there are certainly other possibilities in other theories. 714

3.4.2. Top-down constructions are diacritic by definition (prosodic word and higher) Let us now try to conceive of non-diacritic domains. A non-diacritic domain would have to exist in phonology independently of any issue related to extra-phonological information. Clearly, this excludes all higher layers of the Prosodic Hierarchy. Recall from §401 that it is a recognised and admitted fact in Prosodic Phonology that prosodic constituents fall into two cate-

604 Chap 4: Issues settled in verb, but not in fact gories: those that are top-down, and those that are bottom-up constructions. All higher constituents, i.e. from the prosodic word on, represent the former type, which has the additional characteristic that no phonological property contributes to its construction: prosodic words and the like come into being through translation, and through translation only (see §421). They exist in order to transmit morpho-syntactic information, and for no other reason. On the other hand, prosodic constituents below the word level (i.e. feet, syllables and eventually moras) are bottom-up constructions: they are projections of genuinely phonological vocabulary (ultimately of melodic primes).165 Also, the computation that produces them is purely phonological, i.e. in no way influenced by extra-phonological information. In short, the existence of syllables and feet (eventually of moras) is entirely independent of any extra-phonological information: if there were no interface, syllables and feet would still exist, while prosodic words and higher constituents would not. The lower bottom-up constructed layers of the Prosodic Hierarchy thus appear to be sound candidates for non-diacritic domains. Whether this is indeed the case is discussed in §716. By contrast, the higher layers from the prosodic word on are diacritic by definition: they serve no other purpose than storing extra-phonological information.

715

3.4.3. Higher layers of the Prosodic Hierarchy are the projection of nothing Another remarkable property of the higher layers of the Prosodic Hierarchy is the fact that they are the projection of nothing. Linguistics is (perhaps not exclusively, but certainly a lot) about hierarchical structure, which is most commonly represented in form of trees. As far as I can see, a property that is shared by all hierarchical structure in 165

It may seem at first sight that this is not true for Government Phonology, where lexical entries are fully syllabified and hence no syllabification algorithm builds syllable structure during phonological computation. This impression is mistaken. The presence of syllable structure in the lexicon does not imply that syllable structure is not a projection of melodic primes. For one thing, it must somehow get into the lexicon, and hence children must transform a linear phonetic signal into a syllabified lexical entry. The same is true for adults when they lexicalise new lexical material (loans, acronyms etc.). Hence just like in other theories there is a syllabification algorithm in Government Phonology – which however is active upon lexicalisation, rather than during regular phonological computation.

Local vs. non-local carriers of morpho-syntactic information 605

linguistics is that it is the hierarchical structure of something. That is, syntactic trees and phonological structure such as syllables are the projection of terminal elements, which originate in the lexicon, i.e. syntactic features and segments (i.e. clusters of melodic primes), respectively. This is what modular computation is all about: on the basis of (domain specific) vocabulary items, computation creates structure (see §652). Remarkably enough, though, prosodic constituents above the word level are the projection of nothing. They dominate items, and ultimately terminals, but these have no bearing on the structure that dominates them at all. Quite the opposite is true: the properties of terminals are determined by the higher constituent structure, which is imposed from the outside, i.e. as the result of translation. As far as I can see, the exclusively top-down (or, in modular terms, outside-in) character of the higher part of the Prosodic Hierarchy is unprecedented in linguistics. The question may therefore be asked whether it is appropriate to talk about domains at all; and whether this kind of fake domain has any conceptual status in linguistics. As far as I can see, the fact that prosodic domains do not fulfil the basic requirement for hierarchical structure, i.e. projection, is not discussed in the literature.

716

3.4.4. Projections created by phonological computation cannot be the output of translation Let us now turn to bottom-up constructed domains such as syllables and feet, which are non-diacritic and hence seem to be legitimate carriers of morpho-syntactic information. They are faced with another problem, though: if they are exclusively phonological, i.e. if no extra-phonological property contributes to their construction, how could they ever carry morpho-syntactic information? The property that makes them non-diacritic also disqualifies them from being the output of translation. This may also be looked at from a more formal perspective: syllables and feet (eventually moras) cannot be carriers of morpho-syntactic information because they are the result of phonological computation. Like all other domains (except, precisely, the higher layers of prosodic constituency), syllables and feet (eventually moras) are projections of basic vocabulary: syllables (and moras) are a function of segments, while feet are built on syllables. In modular terminology, syllables and feet (moras) represent the structure of the phonological module, while segments are its vocabulary.

606 Chap 4: Issues settled in verb, but not in fact Both are related by phonological computation: the latter is its input, the former is (a piece of) its output. Carriers of morpho-syntactic information, however, are necessarily created outside of the phonology, and by a means that is independent of phonological computation. Syllables and feet (moras), however, are entirely determined by the properties of their terminals. Therefore they do not qualify as the output of translation.

717

3.5. Conclusion: possible carriers reduce to syllabic space Let us now recapitulate the constraints that lie on possible carriers of morpho-syntactic information. Domains are out of business altogether: they are either diacritics (the prosodic word and higher layers of prosodic constituency), or the result of phonological computation. This leaves no room for any kind of domain: all domains fall into one or the other category. By contrast, boundaries are not in the same trouble: if chosen among the objects that are present in phonology anyway, i.e. in absence of issues related to extra-phonological information, they pass the diacritic filter. What they have to face, though, is the independent and robust generalisation according to which melody cannot be the output of translation (§663). That is, only objects at and above the skeleton can participate in translation. The window for possible carriers of morpho-syntactic information thus shrinks quite dramatically: we are looking for objects 1) that are not the result of any (phonological) computation and 2) which occur at or above the skeleton. I argue in Vol.2 that the broad direction that is defined by these constraints is correct: what is inserted into phonology as a carrier of morpho-syntactic information is syllabic space. Depending on the theory, this may be skeletal slots or syllabic constituents. We have briefly opened a window on an implementation that is based on empty CV units. This perspective is discussed at greater length in Vol.2 (see also Vol.1:§§87,402, Ségéral & Scheer 2008, Scheer 2009a,c).

718

4. Conclusion This chapter was designed to gather those issues that are relevant for Direct Interface, and which therefore lead over to Vol.2 (see also Scheer 2008a, 2009a,c). Essentially, the three questions that make the core of the representational side of the interface have been discussed: 1) modularity and

Conclusion 607

translation, 2) the diacritic issue, 3) the (non-)locality of intervention. The former is about the existence of translation, the second about its output, while the latter determines how this output is integrated and processed by phonology. Results (and hence preparation of Direct Interface) are the following: 1) there was and still today is a diacritic issue (§698); 2) domains do not qualify as carriers of morpho-syntactic information (neither prosodic nor any other constituency) (§712); 3) morpho-syntactic intervention must be local (because non-local intervention is domain-based and domains do not qualify) (§711); 4) the window for the nature and insertion of carriers of morpho-syntactic information is extremely shrunk: only items of the domain specific phonological vocabulary qualify, but melody is out of business (§717). The robot portrait of non-diacritic boundaries thus points to skeletal or syllabic space: this is what morpho-syntactic intervention is made of, and this is where it lands.

719

Chapter 5 Open questions (general)

720

1. Grammatical architecture: alternatives to the inverted T

721

1.1. Generative semantics All through Part I, it was assumed, more or less tacitly, that the correct architecture of grammar, at least concerning generative thinking, is the syntactico-centristic inverted T (see §86): one central device concatenates pieces, which are then sent to two interpretative devices that assure the interface with the extra-linguistic world: LF with meaning, PF with sound. This section is just to point out that within generative quarters the inverted T architecture has (had) at least two competitors: generative semantics and parallel modules. For people who have not lived through the late 60s and early 70s, the former appears to be some kind of mythical period of early generative times that one reads about in books on the history of (generative) linguistics (e.g. Newmeyer 1986:81ff, 1996:101ff, R. Harris 1993:101ff): in the late 60s, a handful of hippies led by John Robert (Háj) Ross, George Lakoff, James McCawley and Paul Postal have taken syntax to feature traces of semantic planning and concluded that semantics feeds syntax, rather than the reverse. The story of revolutionary generative semantics usually ends with the victory of the counter-revolution, which was led by Chomsky with no mercy and succeeded to expunge the semantico-centristic alternative root and branch by the mid 70s (Katz & Bever 1974 elaborate on the revolutionary - counter-revolutionary interpretation). It is often said today, however, that recent minimalist analyses take up a number of insights from generative semantics (such as the idea that to kill in fact decomposes into cause to die). Overview literature includes R. Harris (1993), which was already mentioned. Going into greater detail would lead too far afield in the frame of the present book, and I am not exactly competent in this area. The only thing that this section aims at is to mention that alternatives to the inverted T exist, that they exist within the generative paradigm, and that generative semantics is one of them.

610 Chap 5: Open questions (general) 722

1.2. Parallel modules

723

1.2.1. Against syntactico-centrism (Jackendoff) A more recent alternative to the inverted T is the idea that morpho-syntax, semantics and phonology work in parallel, rather than in a specific serial order. This approach has been introduced and is promoted by Ray Jackendoff; it is known as Representational Modularity (Jackendoff 1987, 1994, 1997), and more recently under the label of Structure-Constrained Modularity (Jackendoff 2002) (see Vol.2 for further detail regarding the difference). Ackema & Neeleman's (2004, 2007) work for example follows this parallel line of attack. The idea is that each module – minimally syntax-morphology, semantics, phonology – retrieves the information that is relevant for its task from the lexicon. On these grounds, morpho-syntactic, semantic and phonological representations are constructed in parallel and independently from one another. So-called correspondence rules (or interface processors today), Jackendoff's way to do translation, assure the communication among modules: they are responsible for the exchange of information, which occurs upon need and may be operated at any constructional stage. Jackendoff's parallel alternative is motivated by the rejection of the typical generative syntactico-centristic conception of grammar (Jackendoff 1997:15ff): syntax is not the central construction worker that reduces semantics and phonology to interpretative devices. Rather, all modules are equal-righted: they enjoy generative abilities and build their own structure. Whether this involves the concatenation of pieces or not is a secondary question.166 It is worthwhile going into a little more detail here for two reasons: for one thing, Jackendoff has done relevant work on intermodular communication, which means that we will meet him again in Vol.2, where his computational conception of translation is discussed at length. The other reason is that modular parallelism is entertained in the area of interest of the book, the interface of phonology and morpho-syntax, in work by Sabrina Bendjaballah and Martin Haiden. This is what we turn to in the following section. 166

Carvalho (2006) also argues against the purely interpretative status of phonology, albeit from the structuralist point of view. Based on productive verbal paradigms in Portuguese, he contends that "morphological markedness uses stress placement and phonological markedness in order to build a five-degree scale of the complexity of inflected forms" (Carvalho 2006:157).

Grammatical architecture: alternatives to the inverted T 611 724

1.2.2. Designated portions of the skeleton project morpho-syntactic features (Bendjaballah and Haiden) Central in Bendjaballah and Haiden's work is the template (e.g. Bendjaballah & Haiden (2002, 2003a,c, 2007). Their idea is to generalise an insight that Guerssel & Lowenstamm (1990) gained on the basis of Semitic. That is, morphological operations do not take place just anywhere in templates; rather, they apply to designated pieces thereof. These specific portions of the template must somehow be specified and lexically recorded: they are stable in a given language. An open question is whether they are really part of the template, i.e. inert in regular circumstances and "activated" upon specific morphological action; or whether they are a regular morpheme that is absent from the lexical recording of the template and joins in when the morpheme(s) that they represent are actuated. In this case, the only peculiarity of the designated portions – typically a CV unit – is their status as an infix (rather than as a prefix/suffix). While Semitic-style non-concatenative systems provide for the possibility of infixation, breaking up a root is not a standard process in languages whose morphology is known to be purely affixal. Bendjaballah & Haiden thus make the hypothesis that all languages may accommodate skeletal portions that are designated for being the theatre of specific morphological activity. Their analysis of German ablaut is a case in point (Bendjaballah & Haiden 2003b). In this perspective, the skeleton acquires the role of a central market place that mediates between phonological melody and morphological activity: it filters the morphemic value of melodic items. A more specific assumption is that only melody which is associated to some skeletal position can have morphemic value (i.e., floating pieces of melody cannot). That is, melody acquires a morphemic status only when it is associated to designated pieces of the skeleton, which project morphosyntactic features. Phonological alternations are therefore morphosyntactically irrelevant unless they produce a modification of the interpretation of feature-projecting skeletal chunks. Bendjaballah and Haiden's approach is akin to the work that was mentioned in the section on LF movement (§579). Whether, like Lowenstamm's (2008), it requires simultaneous presence of phonological material and morpho-syntactic structure is a question that needs to be discussed in relation with the homomorphous character of parallel modules that Haiden and Bendjaballah assume (e.g. Haiden 2005:166ff).

612 Chap 5: Open questions (general) One aspect of the notion of modular homomorphematicity is the rejection of the morpho-syntactic privilege for concatenation that divides the inverted T into a concatenative and two interpretative devices. According to the parallel alternative, all modules have concatenative ability, which means that "phonology has independent access to the structure-building algorithm of the computational system" (Haiden 2005:166). Homomorphous modules, then, need to coordinate their concatenative activity and exchange information in order to keep the parallel constructions in line with each other. 725

1.3. Conclusion: the baseline of the generative paradigm is modularity Further discussion would lead too far afield in the frame of this book, which only reviews at any significant detail the subset of generative approaches that follow the inverted T. This restriction – which however covers more than the lion share of the landscape – was implicit during Part I and is now made explicit. If one cannot speak of any competition with generative semantics anymore, the question whether morpho-syntax is the central piece of grammatical architecture or not remains pending. The take-home message that emerges from this brief excursion is a clearer definition of what it means to be generative. It does not necessarily mean that morpho-syntax precedes LF/PF: it does not (or rather: did not) in generative semantics. And it does not necessarily mean that only morphosyntax is able to operate concatenation (Merge): all modules concatenate in Jackendoff's parallel perspective. In sum, not all generative approaches to the architecture of grammar are syntactico-centristic. What being generative does mean, however, is that there are modules: all options discussed are based on distinct computational systems for morpho-syntax, phonology and semantics. This is but a confirmation of the deep modular roots that generative thinking has acquired since the 50s and Chomsky's participation in the computational movement that implemented the ideas of Turing and von Neumann (see §603). The question of generative modularity offenders (of which the list was drawn in §702), then, remains pending.

PF – a strange hermaphrodite animal 613 726

2. PF – a strange hermaphrodite animal

727

2.1. Clean syntax, dirty phonology/PF?

728

2.1.1. Minimalism shrinks syntax A pervasive effect of the minimalist programme is that a whole lot of phenomena and mechanisms that were held to be syntactic in earlier practice are unloaded from syntax. As indicated by its name, the minimalist logic is indeed (286) "to examine every device (principle, idea, etc.) that is employed in characterizing languages to determine to what extent it can be eliminated in favor of a principled account in terms of general conditions of computational efficiency and the interface condition that the [language] organ has to satisfy for it to function at all." Chomsky (2004:106)

This is to say that anything that syntax is supposed to do beyond the concatenation of pieces must have an extra-syntactic reason, and the two reasons that Chomsky acknowledges are interface conditions ("bare output conditions") and computational efficiency. We have come across the latter in §306: derivation by phase is motivated by the suspicion that the computation of an entire sentence is too complex given the limitations of active memory. Sentences are thus computed in successive smaller pieces. In Chomsky's biolinguistic approach, language is "perfect" just like other natural systems are (in reference to Galileo's "nature is perfect", see Chomsky 2004:105, also Chomsky 1995a:221, Jackendoff 1997:19). Pursuing the goal of a "perfect" FLN (faculty of language in the narrow sense, Hauser et al. 2002, see §633), the minimalist programme therefore ambitions to reduce the FLN to just those properties that cannot be derived from computational efficiency and constraints imposed by the two interfaces. This also implies that LF and PF do not count as really linguistic (§639): they are imperfect and lie outside of the FLN, which reduces to narrow syntax. The elimination of conceptually unnecessary devices from syntax does not mean that they are simply beamed out of sight into some extragrammatical space, though. It means that they are poured into the interfaces, i.e. into what is called LF and PF.

614 Chap 5: Open questions (general) 729

2.1.2. Minimalism pumps up PF The minimalist transfer of activity from syntax to the interfaces loads PF much more than LF. Under this charge, the status of PF changes quite dramatically. While in the classical inverted T that generative grammar has lived with since the 60s (§86) PF was more or less coextensive with phonology, i.e. the phonological computational system, it is now pumped up with a whole lot of operations and items that have got nothing to do with what phonologists call phonology. In the minimalist landscape, PF has become "phonology plus something", and as we will see below nobody really knows what either the whole or the "something" is. Syntacticians are quick and happy to dump all kinds of things into the PF-dustbin, but do not care much whether the dustbin is suited to receive the parcel, or whether the parcel can be handled by those who know the dustbin from the inside. That is, PF is often treated as a black box: syntacticians do not look into what actually happens to the parcel when it is treated "at PF". The only thing that matters is that "PF" somehow manages to do things that syntacticians do not want to do in clinically pure syntax.

730

2.1.3. Dumping into the PF dustbin and hoping that it is big enough A typical case of syntax-to-PF outsourcing is clitic placement, which is traditionally managed in syntax. Bošković (2001) and Franks & Bošković (2001) for example argue for a PF-based solution: clitics in Bulgarian, Macedonian and Serbo-Croatian are generated in several locations by the syntax, and then PF "chooses" the copy that is actually pronounced. Elaborating on Chomsky's (1993a) copy and delete where only the highest copy is pronounced and all others are deleted "at PF", Franks & Bošković (2001) propose an analysis whereby (287) "a chain is pronounced in the head position, with lower copies deleted in PF, unless pronunciation in the head position would give rise to a PF violation. If the violation can be avoided by pronouncing a lower copy of the chain, then the lower copy is pronounced instead of the head of the chain." Franks & Bošković (2001:176)

One wonders, then, what a relevant PF violation looks like, and whether it has anything to do with phonology. The authors explain the conditioning factor as follows.

PF – a strange hermaphrodite animal 615 (288) "This is the well-known Tobler-Mussafia (TM) effect. It results from a phonological difference between the relevant Bg [Bulgarian] and Mac [Macedonian] clitics: in Bg these elements are strictly enclitic, whereas in Mac they are not. When no nonverbal lexical material that could support the clitics occurs in front of the clitics, the verb must precede the clitics in order to provide phonological support for them. This 'repair' strategy of pronouncing the verb to the left of the clitics occurs in Bg, but not Mac." Franks & Bošković (2001:175)

The PF violation that makes the pronunciation of the highest copy illegal thus falls into two components: 1) the fact that in the particular language at hand an item is an enclitic, and 2) the fact that in this language there is no "repair" strategy which places a word before the enclitic. The result is an enclitic without host, which leads to ungrammaticality. Looking at this analysis, one wonders about the use of the word "phonological": Franks & Bošković say that the effect is based on a "phonological difference", but there is nothing that a phonologist would call phonological, or that any phonological theory could manage. Clitichood is certainly not anything that is defined in phonology, or that is manipulable by phonological computation, and repairs that manipulate entire morphemes are certainly not the result of phonological computation either.

731

2.1.4. The syntacticians' phonology is not what phonologists call phonology The preceding section has shown that there is quite some confusion around the "phonological" causes of what happens "at PF", and the example quoted is not isolated. Bošković' (2001) book is called "On the nature of the syntax-phonology interface. Cliticization and related phenomena", but the interested phonologist will not find anything that he would call phonological or due to phonological computation in the phenomena or analyses presented. "Deletion at PF" is also popular with pieces that are not involved in a chain, and which may be much bigger than simple clitics: the analysis of ellipsis and one specific form of this phenomenon, sluicing, may delete words, phrases or sometimes entire CPs (e.g. Merchant 2001, Fox & Lasnik 2003). The standard analysis today is that ellipsis reduces to nonpronunciation at PF (Gengel 2009). It was mentioned that the triggering conditions of Franks & Bošković' (2001) analysis are called phonological but have got nothing to do with phonology. The same is true for the action that is taken at PF: pho-

616 Chap 5: Open questions (general) nologists may perhaps delete features or segments, but certainly not words, VPs or CPs. No phonological theory is suited for the manipulation of this kind of object, which phonologists look at like an ant looks at a jumbo jet. Under the minimalist pressure, PF-sinking of syntactic problems has become a natural and unquestioned move over the past decade. Interestingly, though, this trend is by and large confined in syntactic quarters: syntacticians talk about syntactic phenomena and may use the word "phonology" in relation with PF, but nothing that phonologists would call phonology is involved. In other words, the PF world that takes on more and more importance among syntacticians is quite waterproof with respect to phonology as conceived of by phonologists.167 The following section documents the ambient confusion around the new minimalism-born PF.

732

2.2. Confusion and mistiness: what does PF mean, what does it contain? When I first heard of PF as a student, I thought that "Phonological Form" was meant, and was quite surprised when I learned later on that PF actually stands for "Phonetic Form". This is at least the way Chomsky (1995a:2) spells out PF. The confusion lingers on in my mind, and I still need to make sure I think of phonetics when hearing and talking about PF. Maybe I am not the only one who gets confused by the meaning of the P. Chomsky for example does not make much difference between phonology and phonetics. In the quote under (289) for instance, phonology is missing, and phonetics appears instead. (289) "We may think of the language, then, as a finitely specified generative procedure (function) that enumerates an infinite set of SDs [structural descriptions]. Each SD, in turn, specifies the full array of phonetic, semantic, and syntactic properties of a particular linguistic expression." Chomsky (1995a:14f)

167

As far as I can see, the only phenomenon that sometimes comes up in PF-based analyses of syntacticians, and which could qualify for phonological status, is intonation (sentence stress). This is not surprising since intonation is the single one phonological phenomenon that syntax is classically concerned with. It is not so sure, however, that intonation is the result of any phonological computation at all (more on this in §806).

PF – a strange hermaphrodite animal 617

In the quote below, one senses that Chomsky actually means the parametric variation of phonology, rather than of phonetics: parameter settings are a typical property of the former, but not really of the latter area. (290) "At the PF level, properties of the language can be readily observed and variation is possible within the fixed repertoire of phonetic properties and the invariant principles of universal phonetics." Chomsky (1995a:27)

Finally, the confusion is overt in the following quote: if there is any distinction between phonology and phonetics at all, syllable structure is a property of the former. (291) "The PF representation π is a string of phonetic primes with syllabic and intonational structure indicated, derived by a computation from σ." Chomsky (1995a:35)

The three quotes mentioned are all from the first chapter of The Minimalist Program, i.e. actually from a 1993 paper that Chomsky coauthored with Howard Lasnik. In later chapters of the book, Chomsky does make a difference between phonology and phonetics, as shown by the quote below. (292) "Similarly, the phonological component contains no rules to express special cases of general properties of universal phonetics or of phonetic representations." Chomsky (1995a:152)

He now talks about a phonological component, and this component seems to be different in kind from phonetics. One may thus incline to conclude that phonetics is not part of PF. At the outset of chapter four, one reads the following statement, which is puzzling in more than one respect. (293) "Notice that I am sweeping under the rug questions of considerable significance, notably, questions about what in the earlier Extended Standard Theory (EST) framework were called 'surface effects' on interpretation. These are manifold, involving topic-focus and theme-rheme structures, figure-ground properties, effects of adjacency and linearity, and many others. Prima facie, they seem to involve some additional level or levels internal to the phonological component, postmorphology but prephonetic, accessed at the interface along with PF (Phonetic Form) and LF (Logical Form)." Chomsky (1995a:220)

618 Chap 5: Open questions (general) It seems to be safe to consider that PF is what Chomsky calls a "level", and he again is explicit on the fact that the P means "phonetic". One also learns that there is a "phonological component" which is ordered after morphology but before phonetics. This component, then, appears to be a piece of PF. So far, so good – but how could this sub-division of PF then itself host additional internal levels? In a note to the last sentence under (293), Chomsky states that the phenomena at hand (which are supposed to be accommodated by the "additional level or levels internal to the phonological component") cannot be directly done at PF, but could perhaps be handled by "elements formed in the course of the mapping of syntactic objects to a PF representation." (294) "The PF level itself is too primitive and unstructured to serve this purpose, but elements formed in the course of the mapping of syntactic objects to a PF representation might qualify." Chomsky (1995a:379)

It is hard to imagine what kind of item could be at the same time external to PF but internal to the "phonological component", which itself is a piece of PF. In other writings of Chomsky's, one senses that PF and the "phonological component" are the same thing. (295) "There are some reasons to suspect that a substantial core of head raising processes […] may fall within the phonological component." Chomsky (2001:37) "'Stylistic' operations might fall within the phonological component." (Chomsky 2000a:144)

This selection of self-contradictory, puzzling and sometimes cryptic statements about PF could probably be pursued and extended to other authors. The purpose of their review is not to point out the inconsistency, but the fact that this inconsistency is not innocent: people talk about PF in these terms because PF is a misty and ill-defined intermundia. And people may get confused as I do about the P and the relationship of the strange PFanimal with what phonologists call phonology (and what Chomsky probably refers to as the phonological component).

PF – a strange hermaphrodite animal 619 733

2.3. Properties of PF: what kind of animals live in the intermundia?

734

2.3.1. Internal structure of PF One conclusion of the preceding discussion is that if anything, PF is a heterogeneous space. This is what is meant when the literature (namely the one related to Distributed Morphology) talks about "the PF branch", rather than just about PF. This opens up an entire workspace, indicating that there are many things going on "at PF", not just (the phonologists') phonology, and not just phonetics. It was mentioned in §538 that Distributed Morphology locates operations such as fission, fusion and impoverishment (which are unknown in syntax) in PF, and that PF movement obeys locality conditions that are distinct from the ones that control regular movement in (narrow) syntax (§580). Therefore Pak (2008:26) says that "[t]he PF branch […] is thus viewed as a highly articulated derivational component, which yields a number of intermediate structural representations before the final surface form is reached." This points to the manifoldness of the PF landscape: a lot of things are done, and a lot of different things are done. Like Chomsky and much of the literature, Pak does not make explicit whether she means a computational system in the sense of a Fodorian module when she talks about a "derivational component". The fact that there is a long way to go from the upper limit of PF that intersects with (narrow) syntax and its output is illustrated by the interesting term "surface PF" that Pak introduces in a footnote. (296) "The term PF is used to refer both to the derivation along the branch and to the surface form produced at the end of the branch; for clarity, I will use the term surface PF when this latter meaning is intended." Pak (2008:26, note 1, emphasis in original)

735

2.3.2. What happens "at PF" Let us now look at the kind of operations that syntacticians suppose are managed "at PF", i.e. those that do not concern regular phonology and, if included in PF, regular phonetics. The survey below of course does not ambition to be exhaustive: new things are constantly added, and as a phonologist my visibility is rather poor. This being said, the most typical thing that syntacticians want PF to do is certainly deletion (of copies, or involving ellipsis and sluicing). In the previous sections we have also come across

620 Chap 5: Open questions (general) other candidates for a management at PF: topic-focus, theme-rheme, figureground, linearity, head movement and stylistic operations. Richards (2004) adds an interesting hypothesis regarding the limitations of PF action. (297) "The assumption that PF cannot drive syntactic operations ties in with the more general thesis, presented in section 2.5.1.3, that PF can only operate with the structures that the syntax provides to it: in particular, PF cannot build extra structure, that is, create new positions not licensed during the syntactic part of the derivation." Richards (2004:12, note 2)

Hence PF movement for example can move phonological terminals around, but only to positions that are inherited by syntax. As was mentioned, Distributed Morphology is also concerned with the PF branch: this is where morphology-specific operations such as allomorphy, fission, fusion and impoverishment take place. Summarising Embick & Noyer (2001, 2007), Pak (2008) provides a list of operations that are assumed to take place at PF in order to derive "the surface form at the end of the branch [which] clearly has very different properties from the syntactic structure" (Pak 2008:26). (298) "PF operations (unordered list): a. Structural readjustments, a limited set of movement, rebracketing, and deletion/insertion operations whose surface effects are often recognized as 'syntax-morphology mismatches' b. Vocabulary insertion, which adds phonological content to function morphemes c. Linearization operations, which establish linear order between/across structures" Pak (2008:26, emphasis in original)

Under (298a), one senses that DM is aware of the fact that additional movement, deletion, insertion etc. is a source of dramatic overgeneration (see §580), and therefore prophylactically talks about "a limited set" of operations. As far as I can see, though, nobody knows in which way exactly the operations "at PF" are limited, let alone the reason for such limitations. Linearisation (and its interleaving with phonology) is further discussed in §741 below. Its occurrence at (the beginning of) PF is largely consensual also outside of DM quarters (except in Kayne's 1994 system).

PF – a strange hermaphrodite animal 621 736

2.4. Trying to make sense of PF from the modular point of view

737

2.4.1. PF is a cover term for a number of serially ordered computational systems Let us now try to make sense of PF and its internal structure from the modular point of view. This amounts to identifying individual computational systems that qualify for modular status, i.e. are domain specific. We have seen that various designations are used in the literature in order to refer to PF as a whole, or to its sub-systems: level, (phonological) component, module (Pak 2008:6, 29 calls the entire PF branch the "PF module"). Authors typically do not specify, though, whether they talk about computational systems, i.e. input-output systems, or just informally about a functional entity. Let us start with two points that are undisputed: 1) (narrow) syntax is a computational system with modular status of its own, and 2) the input to PF is created by the spell-out of (narrow) syntactic structure. Another nonnegotiable item is a phonology, i.e. a computational system with modular status that computes what phonologists call phonology. This system is domain specific because it works with its own proprietary vocabulary (labiality, occlusion etc.) which is distinct from other linguistic vocabulary. On traditional (generative) assumptions, this list is rounded off by a distinct computational system that is in charge of phonetics, and which is ordered after phonology. There is a fair amount of debate, especially in the OT literature, regarding the question whether phonetics and phonology are merged into one single computational system, i.e. one single constraint hierarchy (Kingston 2007 for example provides an informed overview). This is an expression of OT's scrambling trope (§§28,528). For the sake of exposition, below a distinct computational system appears for phonetics, but nothing hinges on that. We are thus left with a derivational sequence as under (299) below.

622 Chap 5: Open questions (general) (299) what PF is made of (narrow) syntax (computation A) spell-out vocabulary insertion, linearisation, PF movement, rebracketing, deletion, insertion, fission, fusion, impoverishment (computation B) phonology (computation C)

PF

phonetics (computation D)

PF thus consists of three independent and serially ordered computational systems. So far this view would be consistent with modularity. 738

2.4.2. The minimalism-born intermundia violates domain specificity It was already pointed out in §580, however, that there is trouble with computation B, i.e. the misty intermundia that was created by minimalism where all those syntactic-looking things are managed that are rejected from clean narrow syntax. Computation B is past vocabulary insertion; this means that phonological material is present and, according to PF movement, forms the terminal elements of the morpho-syntactic tree, which is also still available. As was shown in §580, however, this cannot be reconciled with domain specificity: computation B would have to access the morpho-syntactic labels of the tree, the tree geometrics and phonological vocabulary at the same time. Also, the tree labels would be the projection of nothing: on standard assumptions hierarchical structure is a projection of terminal elements. In a PF movement tree, however, phonological terminals would cohabitate with morpho-syntactic structure and labels: this does not make any sense. Computation B is thus a modular alien.

PF – a strange hermaphrodite animal 623

Another issue is linearisation: one wonders what it means to linearise upon spell-out from (narrow) syntax, but to leave the tree in place, and to allow for (PF) movement along its structure. A linearised string does not look like a tree, does it? Or does linearisation in fact take place at the end of computation B when PF movement has applied and a linear input to computation C, the phonologists' phonology, needs to be created? Linearisation is examined at greater length in §741 below. One thing is for sure, though: what phonologists call phonology requires a linear string as an input. Finally, one could think of the whole of PF, i.e. computations B, C and D, as a single macro-module which accommodates a number of submodules. This is visibly what Pak (2008:6, 29) means when she calls the entire PF branch a "PF module". Nested computation (or computational systems), however, should be excluded in a modular environment: it violates (informational) encapsulation (see §610, also §632). 739

2.4.3. Mixing phonology and the intermundia Minimalism-created intermundia, i.e. computation B, is already violating domain specificity by itself. There are proposals that go even further by mixing computation B with computation C, the real phonology. It was already mentioned in §580 that DM, through PF movement, promotes a radical form of direct syntax. It was also mentioned that Pak (2008:42ff, 60ff) sets out to do away with the Prosodic Hierarchy: she correctly argues that prosodic constituency is superfluous if phonology has direct access to morpho-syntactic information. In fact the core of her proposal is a much stronger violation of modularity: she argues that the action of computation C (phonology) is serially interleaved with computation B (the intermundia). (300) "I propose that phonological rules apply at various points in the PF derivation. Specifically, phonological rules are interleaved with different kinds of linearization procedures, which apply in PF in order to convert abstract hierarchical structures into fully linearized strings." Pak (2008:6, emphasis in original) "The hypothesis pursued in this dissertation is that phonological rules may also use Concatenation statements – as well as other kinds of linearization statements – as their domains. In other words, phonological rules are interleaved with linearization operations." Pak (2008:28, emphasis in original)

624 Chap 5: Open questions (general) DM thus turns out to be a strong modularity offender (see §704). Linearisation is further discussed in §741 below.

740

2.4.4. The internal structure of the intermundia: two distinct derivational stages, morphosyntax and morphophonology (Idsardi & Raimy forth) Idsardi & Raimy (forth) argue that what appears as the intermundia under (299) (computation B) in fact falls into two distinct and serially ordered systems as under (301). (301) internal structure of computation B according to Idsardi & Raimy (forth) computational properties system hier- adja- linear phon. directed archy cency order content graphs narrow syntax + – – – – (computation A) linearisation 1 immobilisation morphosyntax + + – – – (computation B1) linearisation 2 spell-out morphophonology – + – + + (computation B2) linearisation 3 serialisation phonology – + + + – (computation C) phonetics (computation D)

Their proposal is linearisation-driven: the two additional stages correspond to the idea that linearisation is a three-step operation: it falls into what they call immobilisation, spell-out and serialisation. Idsardi & Raimy consistently use the word "module" in order to refer to their systems, but they do not address the question which kind of proprietary vocabulary is used by each one of their computational systems. This is understandable since for them the need to further divide computa-

PF – a strange hermaphrodite animal 625

tion B stems from the idea that linearisation is progressive: each step in linearisation, then, must be a computational system by itself. It may be noted that the minimal merit of Idsardi & Raimy's system is that they individuate the "real" phonology – i.e. what phonologists call phonology – from the misty intermundia "PF": their phonology (computation C under (301)) is distinct from computation B (or B1, B2), and they are explicit on the fact that we are facing distinct modules. This may not be unrelated to the fact that (unlike authors from syntactic or DM quarters) they are phonologists and look at the misty "PF branch" from the phonologist's perspective. At variance with earlier writings of Raimy on directed graphs (Raimy 2000a,b, 2003, to be discussed in §746), Idsardi & Raimy (forth) call "phonology" (computation C under (301)) what appeared as phonetics before, i.e. the stage where directed graphs with loops (non-asymmetric representations as Raimy calls them) have been linearised to a strictly linear order of items without loops. Hence what Raimy called phonology (representations with loops) before now appears as morphophonology, and earlier phonetics (representations without loops) is now phonology. Finally, for some reason Idsardi & Raimy restrict the label "PF" to "narrow" phonology, i.e. computation C under (301). (302) "For the purposes of this paper, we will equate the phonological representation with PF and assume that there are further complex transformations in the phonetic modules which result in a representation interpretable at the SM [sensory-motor] interface." Idsardi & Raimy (forth:2)

This adds confusion to the overall landscape where PF is typically held to encompass everything after narrow syntax (the "PF branch"). Idsardi & Raimy's system is born from Raimy's (2000a,b) theory of (morpho-)phonology that is based on the analysis of reduplication (and infixation), and as was mentioned it is linearisation-driven. Their backdrop is thus significantly different from the one of Distributed Morphology, where the motivation for inquiring on the internal structure of the PFintermundia is the accommodation of operations such as allomorphy, fission, fusion and PF movement. It is therefore not really surprising that the results are quite different. For example, PF movement cannot be accommodated by Idsardi & Raimy's system: there is no single computational system under (301) that allows phonological content to cohabitate with hierarchical structure. The shift from computation B1 (morphosyntax) to computation B2 (morphophonology) eliminates hierarchical structure and introduces phono-

626 Chap 5: Open questions (general) logical content. In DM terminology, this is where vocabulary insertion takes place.

741

2.5. Linearisation

742

2.5.1. Introduction: no business of phonology A question that was not addressed at all in Part I but which is constantly mentioned on the preceding pages is the linear order of the phonological string, as opposed to the non-linear character of syntax. The reason why linearisation was not discussed in Part I is simply that it is no business of phonologists – but as was mentioned before, theories of the interface between morpho-syntax and phonology are made by phonologists (with the half-exception of Distributed Morphology). What everybody agrees upon is the fact that the input to phonological computation is a linear string.168 Or, in other words, phonological computation does not create linearity. This is why phonologists never talk about linearity and linearisation: they work on linear strings and thus take linearity for granted. Today everybody agrees that some mechanism is required which transforms the hierarchical morpho-syntactic structure (the tree) into a linear string. The question is where exactly this mechanism is accommodated, how it works and eventually into how many sub-components it falls. Roughly speaking, Kayne's (1994) LCA (Linear Correspondence Axiom) places linearisation in narrow syntax, while all other proposals (which typically build on the LCA) locate this operation in PF (where "in PF" opens a rather large array of possibilities, as we have seen in §733).

743

2.5.2. In minimalist times: no business of syntax either In syntax, linearity was long taken to follow from syntactic constituent structure: word order is a function of constituent order. This was the case until (and including) GB: phrase structure rules were parameterised and responsible for language-specific variation in word order. In a language where prepositions precede nouns, the rule was PP → P, NP, while in a 168

But even this notion needs to be further defined: we will see below (and have already seen in §740) that Raimy's (2003) and Idsardi & Raimy's (forth) system allows for different degrees of linearisation of the string.

PF – a strange hermaphrodite animal 627

language where the opposite order is observed, the rule PP → NP, P was instrumental. Hence the difference between right-branching languages such as English where heads precede their complements and left-branching languages like Japanese where the reverse order is found is managed by a parameter setting, the head parameter. In the minimalist perspective where trees are constructed by Merge and hence phrase structure rules are eliminated, a different means for deriving linear order needs to be found. Estranging syntax from linear order ties in with the observation that syntactic generalisations are about hierarchical organisation (command and dominance relations), not about linear order. Therefore Chomsky (1995a:334) concludes that syntax has got nothing to do with linearity, not any more than LF: linearity is only relevant for phonology. That is, it is imposed upon the linguistic system by external conditions of language use: linearity is the result of the constraints that follow from the transmission of human language whereby speech unfolds in a temporal sequence.

744

2.5.3. Both syntax-internal and syntax-external linearisation is minimalismcompatible Given that linear order is imposed upon language by a non-linguistic constraint, Chomsky (1995a:335) welcomes Kayne's (1994) antisymmetry in a minimalist perspective. Kayne's idea is to derive linear order in the syntax by leftward movement: at the end of all movement operations, the highest item is the leftmost in the linear order, and so on. On Kayne's take, this is true for all languages: SVO is the universal order underlyingly, and languages like Japanese with an overt SOV order are derived by leftward movement. Incorporating the mechanism that creates linear order into the syntax may at first appear contradictory with the minimalist insight that syntax and linear order are independent. Chomsky (1995a:335), however, argues that syntax-internal linearisation is a prototypical implementation of minimalist thinking whose headstone is to reduce syntactic representations and computation to "bare output conditions". Hence linearity, imposed from the outside by language use, marshals syntax: movement needs to be carried out in order to satisfy its requirements. Chomsky therefore writes that Kayne's perspective is "very much in the spirit of the Minimalist Program and consistent with the speculation that the essential character of CHL [the computational system of human

628 Chap 5: Open questions (general) language] is independent of the sensorimotor interface" (Chomsky 1995a:335). A few pages later, however, he places the LCA outside of (narrow) syntax, in PF: "we take the LCA to be a principle of the phonological component that applies to the output of Morphology" (Chomsky 1995a:340, note that Chomsky operates with the traditional split between (narrow) syntax and morphology). Another line of attack that follows Kayne's track is Fox & Pesetsky's (2004) analysis whereby successive cyclic movement (in narrow syntax) is derived from linearity requirements. That is, just like in Kayne's system and in line with the minimalist philosophy, syntax-internal movement is motivated by a necessity of a syntax-external cognitive system. The key idea of Fox & Pesetsky is that there is a mechanism that controls the result of linearisation at every phase and compares it with the linear order achieved at previous phases. In case there is a mismatch between former and current linear order, the derivation crashes. This is what Fox & Pesetsky call order preservation: linear order must be the same at every spell-out of every phase. For the sake of illustration, consider the example under (303) below (from Fox & Pesetsky 2004:5). (303) [To whom will he [__say [CP __ that Mary [VP __ gave the book __]]]]?

Did the movement skip Spec,VP and went directly to Spec,CP as indicated, and if VP is a phase head, Fox & Pesetsky argue, the derivation will crash at PF since the result of linearisation in the VP phase (where the displaced item remains in situ) is different from the linearised string after the CP phase (where nothing will be left in situ). By contrast, if movement goes through Spec,VP, the VP will be linearised with the displaced item in Spec,VP at the lower phase as well as at all other phases: order preservation is satisfied. Therefore, Fox & Pesetsky (2004:8) argue, "[a]n architecture of this sort will in general force successive cyclicity when movement crosses a Spell-out domain boundary." In sum, then, minimalist principles seem to be able to be used in order to locate linearisation in either (narrow) syntax or PF. Or rather, the competition between both options cannot be refereed on minimalist grounds.

PF – a strange hermaphrodite animal 629 745

2.5.4. Everybody but Kayne does linearisation "at PF" Work in Kayne's original perspective set aside, it appears that the field has by and large followed Chomsky's indication that linearisation occurs "at PF". It seems to me, though, that the question is treated without much argument in the literature: people acknowledge the existence of Kayne's LCA, but then do linearisation "at PF" without saying why it should or could not be done by movement in (narrow) syntax. 169 One argument is provided by Richards (2004): Kayne's LCA is based on c-command, i.e. on an asymmetric relationship between the items that are to be linearised. (304) "Linear Correspondence Axiom If α asymmetrically c-commands β then (the terminals dominated by) α precede(s) (the terminals dominated by) β" version of the LCA given by Richards (2004:11)

What about cases of mutual c-command, then? Richards argues that the minimalist Bare Phrase Structure, which eliminates trivial, i.e. vacuous or unary-branching projections, regularly produces this kind of structure, for example in [John [ate it]] where ate and it are sisters and c-command each other. Kayne's LCA fails when trying to linearise this kind of structure. Richards therefore concludes that (305) "the LCA cannot be a constraint on phrase-markers themselves, i.e. a property of Narrow Syntax, but must be a linearization strategy operative only after Spell-Out in the mapping of syntactic hierarchy onto phonotemporal order (cf. MP [Chomsky 1995a]: 340: 'We take the LCA to be a principle of the phonological component that applies to the output of Morphology'). This recasting of the LCA as a PF-mapping strategy (cf. the operation Linearize of Nunes 1999) conforms to the general principle that the 'horizontal' dimension of time and sequential ordering is relevant only in the phonological component, so that the syntactic component of CHL [the com169

E.g. Bobaljik (2002) and Embick & Noyer (2001, 2007), discussed below, who do linearisation at PF "by hypothesis" or "assume" that it occurs at PF: "[a]ssuming that linear order is not included in the syntactic representation, PF operations, because they are responsible for creating the interface level that mediates between syntax and the articulatory/perceptual systems, must at the very minimum be responsible for linearizing hierarchical structures" (Embick & Noyer 2007:293). Hence Kayne's syntax-internal option is dismissed without argument or discussion.

630 Chap 5: Open questions (general) putational system of human language] deals only in the 'vertical' dimension of hierarchical relations." Richards (2004:12, emphasis in original)

Richards (2004:23) then goes on to rehabilitate the old head parameter of GB in order to make it a parameter of the PF-located LCA. Based on Epstein et al.'s (1998:139ff) Precedence Resolution Principle, Richards (2004, chapters 2 and 4, 2007a) proposes a sisterhood-based linearisation strategy: if a c-command relation translates to precedence at PF, heads and complements provide contradictory instructions to PF since sisters c-command each other: the head must precede the complement, but the complement must also precede the head. Which precedence will be chosen, Epstein et al. and Richards argue, is a matter of which head-complement directionality is deleted. This is what Richards (2004, 2007a) calls desymmetrisation. While Esptein et al. (1998) apply this mechanism only to external Merge, Richards extends it to internal Merge, thereby promoting it to a parameter that governs the entire language. Parameterised desymmetrisation is then the old head parameter based on the LCA, but located in PF. Other syntax-based linearisation strategies also typically implement the head parameter in one way or another. Bobaljik (2002:216) for example says that "it should be clear that I am espousing a more or less traditional view of headedness parameters, for instance, that the German V' is headfinal while the English V' is head-initial; this is the information encoded in the precedence rules." Bobaljik's precedence rules are "a procedure that maps each node to an ordered pair: [X → Y] or [Y → X] (where the arrow is to be read as 'precedes')" (Bobaljik 2002:213). He then shows how a German SOV and an English SVO order can be derived from the same syntactic structure when reverse precedence rules are applied to certain nodes (the head parameter): German will have [I' VP → Infl] and [VP DP2 → V], while English accommodates [I' Infl → VP] and [VP V → DP2]. On Bobaljik's count, thus, each language possesses a language-specific set of precedence rules (that look much like good old phrase structure rules, except that there is only one item to the right of the arrow). In Bobaljik's system, this all happens upon spell-out from (narrow) syntax to PF, whereby the spell-out process falls into four distinct operations (which Bobaljik 2002:214, note 16 calls "components"): "a) assignment of precedence conditions to syntactic nodes [the precedence rules mentioned], b) chain reduction (= trace or copy deletion), c) conversion to linear string of X°s, d) vocabulary insertion" (Bobaljik 2002:214).

PF – a strange hermaphrodite animal 631

The same picture is found in DM quarters. Embick & Noyer (2001, 2007) and Embick (2007) hold that linearisation is concomitant with vocabulary insertion. (306) "By hypothesis, linear ordering is not a property of syntactic representations but is imposed at PF in virtue of the requirement that speech be instantiated in time (see Sproat 1985). It is therefore natural to assume that linear ordering is imposed on a phrase marker at the point in the derivation when phonological information is inserted, that is, at Vocabulary Insertion. (8) The Late Linearization Hypothesis The elements of a phrase marker are linearized at Vocabulary Insertion." Embick & Noyer (2001:562, emphasis in original)

Embick & Noyer work with a notational variant of Bobaljik's precedence rules which they call "Lin" (or the reverse: Bobaljik's precedence rules are a notational variant of Lin): (307) "linear order is a binary operator – represented by '*' – imposed by an operation Lin: Lin [X Y] → (X*Y) or (Y*X) This relationship is one of immediate (left-)adjacency; subsequent steps concatenate terminal nodes. Other types of conditions might be imposed by distinct linearization operations." (Embick & Noyer 2007:294).

Embick & Noyer thus cut linearisation into (at least) two distinct operations, Lin and Concatenate (see Pak 2008:26ff for a summary). The linearisation procedures reviewed share the idea that linearisation in fact falls into a number of different operations that are serially ordered: (at least) two with Embick & Noyer (2001, 2007), three with Idsardi & Raimy (forth) (immobilisation, spell-out and serialisation, §740), four with Bobaljik (2002) (precedence conditions, chain reduction, linearisation of X°s, vocabulary insertion).

746

2.5.5. Linearisation in phonology in order to derive phonetics (Raimy)? Raimy (2000a,b, 2003) proposes a view on linearisation that is quite different from what we have seen thus far: he looks at the issue through a phonological, rather than through a syntactic lens. Raimy's basic idea is that

632 Chap 5: Open questions (general) precedence relationships must be explicitly noted in all phonological representations. Linear order is implicit in the standard graphic notation: "kæt" (representing cat, the word used for the sake of illustration by Raimy) for example reads "k before æ before t", and the same is true for x-slots in an autosegmental representation. Raimy proposes to make this explicit and to note "# → k → æ → t → %" instead (where # indicates the beginning of the linear string, and % its end). He calls the result a directed graph. What is this good for if the graphic material added can be deduced from "kæt" and is therefore fully redundant? Raimy argues that there are cases where arrows actually represent morphological information: morphological computation is one that adds precedence relationships, i.e. arrows, to lexical material. The operation is trivial when prefixes and suffixes are concatenated, but involves the addition of "loops" within a root in case of reduplication and infixation. A process whereby "kæt" is entirely reduplicated, producing "kætkæt", thus boils down to the addition of an arrow that originates in the t and points to the k, as shown under (308) below. (308) linearisation of loop-including representations a. #→k→æ→t→% b. # → k → æ → t → k → æ → t → %

The loop under (308a) thus represents the morphological operation of reduplication. The resulting structure (which Raimy calls nonasymmetrical) is then spelled out into a strictly linear string (an asymmetrical structure) as under (308b): "[l]inearization eliminates non-asymmetrical precedence structures through repetition of segments which preserves the overall organization of a precedence structure while not causing problems of interpretation for the phonetics module" (Raimy 2003:133). The "problems of interpretation" that Raimy talks about are the loops: he assumes that phonology can live with loop-containing representations, but phonetics cannot – here a fully linearised string is needed. In sum, then, phonological representations are only partially linear (loops are non-linear elements); exactly analogous to the relationship between (narrow) syntax and PF, bare output conditions – the requirement of fully linearised strings at the phonetic level – provoke a spell-out. (309) "Analogous to the syntactic LCA (Kayne 1994), phonology contains a linearization process which ensures that representations have asymmetrical precedence structures and are thus interpretable at the phonetics interface.

PF – a strange hermaphrodite animal 633 Linearization in phonology repeats segments within a 'loop' in order to eliminate [non-]asymmetrical precedence relations." Raimy (2003:132, emphasis in original; obviously Raimy means to say non-asymmetrical in the last line, the absence of the non- is an error)

The operation that adds arrows to strings needs to "know" where to place the origin and the endpoint (anchor points). Adopting Raimy's directed graphs, Samuels (2009a:147ff) has studied the cross-linguistic variation of infixation in order to determine what kind of information the arrowplacing algorithm needs to be able to interpret. The list of anchor points that infixes look at in order to determine their landing site falls into two categories: edge-oriented and prominence-oriented. For the left edge for example, documented situations are "after the first consonant (or consonant cluster)", "after the first vowel", "after the first syllable" and "after the second consonant". Prominence-based attractors are stressed vowels, stressed syllables or stressed feet. An example discussed by Samuels (2009a:179ff) is Tzeltal (Maya) intransitivisation whereby a verb is made intransitive by infixing an -h- after the first vowel: compare puk "to divide among" with pu-h-k "to spread the word", and kuč "to carry" with ku-h-č "to endure". The representation of this operation with directed graphs is as under (310) below. (310)

h #→p→u→k→%

As was mentioned in §740, Raimy has changed the labels of his system since 2003: in Idsardi & Raimy (forth), he calls loop-containing representations morphophonology (instead of phonology), and representations from which loops have been eliminated phonology (instead of phonetics). This leaves the workings of his system unaffected, but does away with the idea that phonology can handle non-linear structure. That is, like all other approaches, the relabelled version of directed graphs holds that the input to "true" phonological computation, i.e. the one that phonologists call phonology and from which concatenation is absent, is strictly linear.

747

2.6. Conclusion: a minimalism-born monster PF is an ex nihilo creation of minimalism: until the mid-90s PF was coextensive with phonology, but since then adds an ill-defined intermundia be-

634 Chap 5: Open questions (general) tween (narrow) syntax and "true" phonology, i.e. what phonologists call phonology. Today one is forced to talk about "true" phonology or "what phonologists call phonology" because there is a great amount of confusion around the word phonology. Syntacticians want a whole lot of things to be "phonological" in order to get rid of them: clean syntax, dirty phonology. Most of the things that are called "phonological" in the syntactic literature where phenomena are dumped into the PF dustbin, though, are nothing that phonologists are familiar with or could handle. While the upper limit of PF is clearly defined (PF is created by spellout from (narrow) syntax), its lower limit and eventual internal structure are murky. It is consensual that phonological material is present as soon as PF structure is created: vocabulary insertion occurs at spell-out. What happens then, though, is left misty in the literature: a lot of operations that have got nothing to do with phonology, and then, at some point, "true" phonology. Idsardi & Raimy (forth) set aside (they are phonologists), I could not find a single explicit mention that what phonologists call phonology is a distinct computational system whose input must be a linear string that does not contain any morpho-syntactic arboreal structure anymore. When talking to syntacticians and asking the question where "true" phonology is in the wafting PF body, if any, one often hears that there must be a clearly distinct phonological system somewhere further down the road when all the pseudo-syntactic labour is done – but syntacticians are then unable to name a written source where this is made explicit. When one tries to make sense of PF in terms of domain-specific computational systems, a separate (and "true") phonology (plus eventually a phonetic system) can be identified, but an alien intermundia is left between spell-out and phonology (computation B under (299)). This intermundia cannot be a module because it mixes morpho-syntactic and phonological vocabulary. It also violates another fundamental principle of hierarchical structure: the nodes of the tree are projections of nothing, in any case not of the terminals, which are phonological. Finally, linearisation is also unclear: it seems to be consensual (Kayne 1994 set aside) that it occurs upon spell-out and vocabulary insertion, but the mechanisms that are available on the market (Richards 2004, 2007a set aside) are based on speculation, rather than on data and/or analysis. Also, they reproduce the traditional head parameter in one way or another: instead of being linearised by phrase structure rules like in the times of GB, sisters in the tree are now linearised by linearisation statements (called precedence rules etc.) in PF.

The balance of procedural and representational communication 635

The strangest animal in the PF zoo (which everybody seems to want to restrict, but nobody says what exactly may and what may not be done at PF) is certainly PF movement. One issue is linearisation, which is supposed to occur at spell-out, that is before PF movement. But what does it mean to "linearise" when the output of "linearisation" is a not a linear, but an arboreal structure along which items continue to be displaced? And how is a truly linear string created at the end of PF movement when "true" phonology is fed? After spell-out from (narrow) syntax, is there a second spell-out mechanism? All these questions, and foremost the massive violation of modularity by the intermundia computation B, raise the question what the whole PF business really buys us: a clean syntax, but a lot of trouble "at PF". Is it really worthwhile to move undesirable phenomena out of sight of syntacticians (and of phonologists) into an ill-defined buffer where established principles of linguistic analysis do not hold? Dumping them into PF is not analysing or solving them, and being allowed to analyse them with all kinds of ad hoc mechanisms that are unheard of elsewhere may turn out to be a Pyrrhic victory rather than anything else.

748

3. The balance of procedural and representational communication

749

3.1. Which channel for which phenomenon? Given an interface phenomenon, i.e. a phonological process that is conditioned by extra-phonological information, is there a way to know whether the effect is due procedural or representational communication with morpho-syntax? If so, what kind of diagnostics allows us to find out about the distribution of the two interface mechanisms? Is there a class of phenomena that can only be done procedurally, and another class which requires a representational solution? And are there phenomena that can have both procedural and representational analyses? These are questions that one is tempted to ask when Interface Dualism is taken seriously. Even though all theories of the interface provide for a procedural and a representational way to talk to phonology (with two exceptions, see §683), the question of the balance between procedural and representational communication has never really been debated as far as I can see, and workable answers have not been generated. One notable exception is on record, though: the discussion regarding the illegitimacy of Lexical Phonology in presence of representational solutions that build on

636 Chap 5: Open questions (general) prosodic constituency, which could be extended to the area below the word (see §429, Inkelas' 1990 prosodic constituency in the lexicon).

750

3.2. Prosodic Phonology tries to get away with peaceful coexistence and/or random distribution As was mentioned in §441, peaceful coexistence, the "official" attitude of Prosodic Phonology, is a non-answer that tries to sweep the issue under the rug. This being said, Nespor & Vogel (1986) are to be credited for at least having tried to define a set of phenomena that is taken care of by the procedural mechanism, as opposed to another type which is due to representational communication (see §435). According to them, the critical contrast is between "purely phonological" rules and those that make "reference to morphological structure and/or specific morphological elements". It was shown in §§437ff that this distinction is not operative: it is unable to delineate two distinct sets of interface phenomena. Another contemporary point of view is expressed by Rubach & Booij (1984 et passim), who simply give up on any attempt to answer the question (see §440): below the word level, interface phenomena are due to either procedural or representational devices, and there is no principle that defines which phenomena use which strategy. That is, Lexical Phonology and Prosodic Phonology are both candidates for managing everything that is going on below the word level; whether one or the other comes to bear is random and unimportant.

751

3.3. Random use of both channels is typical (and tacit), but it may be doubted that this is the right way to go In practice, random use of whatever channel suits the analysis is actually the way that basically all interface theories that we have come across behave; the only difference with respect to Rubach & Booij is that the random application of procedural and representational solutions remains unreflected (except in Stratal OT, see §§490f). This is true for the division of labour between the Transformational Cycle and boundaries in SPE as much as for the example that was mentioned in §130: the move from a representational to a procedural management of word boundaries when autosegmental syllable structure was introduced seems to have gone entirely unnoticed.

Morpho-syntactic structure may, but labels may not bear on phonology 637

A random division of labour between procedural and representational devices at the interface seems implausible, though. Over the past years while working on the book, I have tried to establish an inventory and a classification of interface phenomena in order to see whether generalisations emerge: is there a group that has only been analysed procedurally and that could not be done representationally? Is there a group with the opposite orientation? And perhaps one where procedural and representational analyses compete? A number of conference presentations have treated this question (Scheer 2005, 2006a,b, Newell & Scheer 2007), and the result was originally meant to be included here. Facing the complexity of the task, though, I have given up on the project, at least in the frame of this book (except for a broad direction, procedural first, which is briefly discussed in §320). I am convinced that there is a rationale according to which interface phenomena fall into a procedural and a representational group (and eventually one more where both mechanisms are applicable in principle). A clearly defined class of interface events that was discussed as such all through Part I are affix class-based phenomena. My best guess for these is that they are only procedural. That is, theories such as Lexical Phonology that recur to a representational crutch (brackets for level 2 rules, see §166) are on the wrong track. The evaluation of this claim requires further study. And also, the awareness of interface theories that the balance of procedural and representational solutions for interface phenomena is a relevant issue at all certainly needs to be sharpened. 752

4. Morpho-syntactic structure may, but content (labels) may not bear on phonology The idea that geometric properties of the morpho-syntactic tree may influence phonology, while morpho-syntactic features (the basic vocabulary of the module) and projections thereof (the node labels) may not, was first met in the discussion of Prosodic Phonology (§398). Observations that led to this hypothesis go back to Selkirk (1974) and Rotenberg (1978:111ff) and have been taken up in Prosodic Phonology (relevant literature is discussed in §398). The generalisation makes sense from the modular perspective since (domain specific, see §640) vocabulary of a given module cannot be parsed by other modules; on the other hand, structure is not handicapped in the same way (see §652 for the difference between vocabulary and structure).

638 Chap 5: Open questions (general) This has been pointed out in §653, where the hypothesis was considered that only structure, not vocabulary, can be translated in representational intermodular communication. Further evidence to this end (from the opposite direction, i.e. phonology → morpho-syntax) is discussed in §660: phonological melody is unable to bear on morpho-syntax (while phonological structure at or above the skeleton is). Also, (representational) carriers of morpho-syntactic information in phonology appear not to include melody. This points to a complete exclusion of vocabulary from translation, both as input and as output. The facts that phonological melody is unable to bear on morphosyntax and that carriers of morpho-syntactic information cannot land below the skeleton are undisputed empirical generalisations (§660). The empirical situation of the generalisation discussed in the present section, however, is much less clear-cut; this is why it is presented in the chapter on open questions. An oft-quoted prima facie counter-example is category-driven stress placement in English: the same items bear penultimate stress in case they incarnate as nouns (récord, pérvert, éxtract, rébel), but show final stress when they appear as verbs (recórd, pervért, extráct, rebél). I have come across a similar case in Czech, where vowel-final prefixes are long in case they attach to a noun, but short when they occur with a verb (Scheer 2001, 2003a:109ff, 2004c). Jennifer Smith has developed a specific interest in category-sensitive phonological phenomena: her work provides a collection of data sets that appear to stand against the generalisation at hand. Smith specifically focuses on what she calls noun faithfulness, i.e. her observation that nouns appear to show positional faithfulness effects, and to maintain phonological contrast that is not permitted in verbs. Relevant work includes Smith (1997, 1999, 2001). Interestingly, stress is again prominent in the cases of nounfaithfulness that Smith discusses. A possible way to get around this evidence is to think of the phonologically relevant lexical contrast as an effect of structure, rather than of vocabulary. That is, the contrast between nouns, verbs and adjectives may be encoded in terms of vocabulary at first, but is transformed into a structural difference by the morpho-syntactic derivation: nouns have got something more (or less) than verbs. What produces the phonological effect, then, is the contrasting structure, not the original vocabulary. This hypothesis should be testable on morpho-syntactic grounds: the additional structure should leave traces on this side of the coin as well.

The mapping puzzle 639

The generalisation that phonology can read structure, but not vocabulary, is thus certainly interesting and ties in with similar generalisations that all make sense from the modular point of view (the fact that melody and morpho-syntax are incommunicado). Unlike these, though, it faces a serious empirical challenge which requires further study.

753

5. The mapping puzzle

754

5.1. The issue(s), looked at by more or less helpless linguists The mapping puzzle has escorted the reader all through the book (the detail is recapitulated in §755 below). Rather than a particular theoretical device or a specific view on the interface, it concerns an empirical generalisation – or rather, the absence thereof. When looking at the inventory of morpho-syntactic configurations that do or do not produce a phonological effect across languages, one may expect some generalisations to emerge. After at least 25 years of serious research in the area that has produced a fair amount of cross-linguistic data and a reasonable number of case studies from genetically unrelated languages, the frustrating fact is that the harvest is very poor, in any case way below anything that is solid enough to ground a theory. This is one aspect of the mapping puzzle. Another, just as intriguing, concerns the properties of the conditioning patterns that are found. On many occasions, the set of morpho-syntactic divisions that trigger or inhibit a particular phonological effect do not appear to make sense: they are nothing that a syntactician would call a natural class. Why is it that this boundary produces the phonological effect at hand, but that boundary does not? Sometimes nothing really seems to unite the pool of triggering boundaries on the syntactic side. On the other hand, there may be non-triggering boundaries which from the syntactic point of view would be expected to line up with the triggering boundaries, but do not. Finally, it needs to be noted that the mapping puzzle concerns syntax, rather than morphology: the effects of specific sets of morphological boundaries (i.e. of boundaries below the word level) have been formalised in terms of affix classes. This is one of the basic insights of Lexical Phonology that was carried over to all subsequent theories: affixes may behave like this or like that because they are lexically specified for class membership. The trouble regarding boundary grouping at and above the word level is due to the fact that the analogue strategy, "word classes", is absurd:

640 Chap 5: Open questions (general) words are not recorded in the lexicon and hence cannot bear lexical specifications. All this being said, a scenario where natural language devises an arbitrary set of syntactic configurations to leave a phonological trace is fairly unlikely. It is also self-evident that phonology has got nothing to say about the issue: progress can only come from a better understanding of how syntax works. 755

5.2. The mapping puzzle since SPE The mapping puzzle was first mentioned in the chapter on the post-SPE period (§111). Since then it runs through Part I and has reduced linguists of all periods and all theoretical obediences to despair. Elisabeth Selkrik is certainly the one who has worked most on the issue: in her early work on Prosodic Phonology (Selkirk 1984) and in her theory of end-based mapping (Selkirk 1986, see §396) which evolved into alignment in OT (§458). Elisabeth Selkrik has also been the motor of the wave of empirical work on mapping that Prosodic Phonology has produced in the second half of the 80s and in the early 90s. More recently, she has participated in the adaptation of prosodic constituency to phase-based mapping (§462). Finally, she has inspired the shift of focus away from morpho-syntactic conditions to other factors that play a role in mapping, such as information structure and eurhythmy (§463). The theory that made the strongest contribution to the study of the mapping puzzle is Prosodic Phonology (see §§387,392f). When Prosodic Phonology was about to launch its empirical programme on mapping, Kaisse & Zwicky (1987) made the following statement in the introduction of a special issue of the journal Phonology. (311) "The study of postlexical rules sensitive to syntactic or prosodic structure is still in its infancy. Phonologists know comparatively little about the range of phenomena that can be encompassed by such rules, compared for instance with what is known about word-internal phonological processes or rules of syntax. At the moment theories must be advanced on the basis of data that are, from the language-internal point of view, rich and complex, but are also, from the cross-linguistic point of view, sparse and diverse." Kaisse & Zwicky (1987:4)

In spite of the progress that was made on both theoretical and empirical sides, Kaisse & Zwicky's description still makes a fairly good match of the situation today.

Privativity 641 756

6. Privativity

757

6.1. On the representational side: complete or privative translation of morpho-syntactic divisions?

758

6.1.1. All theories are non-privative since SPE Privativity is discussed on several occasions in Part I, where it is related to the representational channel, rather than to procedural communication with phonology. We will see, however, that it is a relevant concept also on the procedural side of the interface. But let us first look at privativity in the former area. It is an undisputed fact that only a small subset of the available morpho-syntactic information is actually relevant for phonology: most of it has no phonological effect at all. As far as I can see, this basic observation was first formulated by Chomsky et al. (1956), who conclude that phonology is underfed by translational activity (see §79): if only a small subset of morpho-syntactic information is used, the translational device has fed phonology only with this subset – phonologically irrelevant information has never been translated and is thus absent from phonology. In spite of this basic observation, the alternative non-privative view on the matter is implemented by SPE and, as far as I can see, all subsequent interface theories. That is, full morpho-syntactic information is shipped off to phonology regardless of whether it will be used or not. Phonology then appeals to whatever is relevant for its computation; finally, irrelevant information, which is present in form of SPE-type boundaries or prosodic constituency, is either erased by phonological action (as in SPE), or just sits in phonology and remains inert (as in Prosodic Phonology).

759

6.1.2. Unanimous non-privativity in individual theories The position of individual theories in regard of privativity is as follows. Structuralists such as Moulton (1947) use "+" in order to anchor the beginning and the end of each word in the phonemic transcription, irrespectively of whether it will have an effect or not (§64). The rigid mapping mechanism of SPE also sends all edges of major categories and higher projections thereof to the phonology without discriminating between those that will and those that will not be used (§90). That is, the "syntactic" # boundary restores full morpho-syntactic information in phonology: the syntactic dis-

642 Chap 5: Open questions (general) tance of two neighbours in the linear string is a direct function of the number of intervening #s. The input to phonological computation is thus a string made of lexical items and clusters of hashmarks. After the application of phonology, remaining hashmarks are erased by rule at the end of the derivation. In the 70s, the non-privative heritage of SPE has prompted the need to distinguish between those boundaries that are phonologically relevant and those that have no phonological function. The former have been called "phonological", the latter "morpho-syntactic" (see §132) – a misleading distinction since a "phonological boundary" is a contradiction in terms. Lexical Phonology is also on the non-privative side, at least those versions that use Mohanan-type brackets, which mark the beginning and the end of every morpheme in phonology (§169). Prosodic Phonology promotes non-privative translation as well, even if it is true that not all morpho-syntactic information is projected onto phonology (the Translator's Office makes readjustment decisions, see the discussion of non-isomorphism in §§380,416). Non-privativity in fact is a byproduct of the Strict Layer Hypothesis: all strings are exhaustively parsed at all prosodic levels independently of whether or not there is an associated phonological effect. That is, the full six-layered prosodic constituency is always constructed no matter whether there is evidence for particular divisions or not (see §§383,4000). Finally, models that implement direct syntax also stand on the nonprivative side: van Oostendorp's (2006a) Coloured Containment and Orgun's (1996a et passim) declarative sign-based morphology allow phonology to directly see and refer to all elements of morphological (and syntactic) structure (see §501). Bermúdez-Otero (forth a:21ff) discusses this approach in greater detail; he calls Orgun's view on the matter isomorphism (of morphological and phonological structure). The only representatives of privative translation that we have come across in Part I, then, are Chomsky et al. (1956) (§79). Noam Chomsky and Morris Halle reverted back to non-privativity twelve years later in SPE, a move that determined the attitude of the field up to the present day. It is therefore reasonable to think of their privative stance as a mere accident that was due to the rhetorical demand of the time: Chomsky et al.'s (1956) strategy was to introduce generative boundaries into the hostile structuralist environment by appealing to the well-reputed structuralist notion of economy: juncture exists in order to simplify phonemic transcription (§74).

Privativity 643 760

6.2. On the procedural side: selective vs. non-selective spell-out Privativity on the procedural side incarnates as selective spell-out, which is discussed at length in the chapter on open questions in the procedural area (§763). In short, the question is whether all nodes of the morpho-syntactic tree are spelled out, or only a subset thereof. The front line leaves Lexical Phonology (but not all brands thereof, see §769) on the non-privative side, against Halle & Vergnaud (1987a) and Kaye (1995) who are privative. On the face of it, Distributed Morphology is privative, but the privativity at hand is rigidly defined according to the type of projection: all categorising affixes (i.e. xPs) project phase heads, while functional projections never trigger interpretation (§583). It is also shown below that selective spell-out, which was introduced by Halle & Vergnaud (1987a), is the ground rule of modern phase theory: only a subset of nodes are phase heads in current syntactic theory. Anybody who subscribes to derivation by phase, and who believes that morphology and syntax are the same computational system, should therefore be bound to selective spell-out. The case of Lexical Phonology is different since phase theory cannot be positioned as a referee: if morphology and syntax are two distinct computational systems as is claimed by Lexical Phonology, there may be two distinct spell-out mechanisms. This means that the spell-out of morphemes could be non-selective even in presence of phase theory where the spell-out of words is selective (see §769).

761

6.3. Five arguments in favour of privativity Privativity is not a notion that ranks high among the issues that interface theories manipulate consciously. I have not come across any case in the literature where it is named and argued for or against. This is much unlike the situation in melodic representations where privativity is the major frontline that divides theories which use binary primes and theories that rely on monovalent/unary primes.170 The present section is devised to carry privativity into the spotlight of interface theory as a relevant design-property that contrasts competing approaches. Although the issue will not be settled, I believe that there is 170

Binary approaches are standard since Jakobson and SPE; the privative alternative has been introduced by Anderson & Jones (1974) and was implemented in Dependency Phonology (Anderson & Ewen 1987), Government Phonology (Kaye et al. 1985) and Particle Phonology (Schane 1984).

644 Chap 5: Open questions (general) good reason for granting a serious advantage to privative theories. Five points can be made. If the output of translation is non-diacritic, translation must be privative. This point is made in Vol.2 on the grounds of the fact that diacritic morpho-syntactic information is always accessed by a specific proviso in the formulation of processes (constraints or rules) of the kind "only within the prosodic word", or "if a hashmark precedes". A hashmark or a prosodic word alone are thus perfectly inert: they are "sleepers" that have no effect unless they are called on by a specific mention in a computational statement. By contrast, if non-diacritic, i.e. phonologically contentful objects are the output of translation, they have an immediate effect on phonological computation without needing to be referred to by computational statements. This is what I call the Direct Effect (see Vol.2, Scheer 2009a,c). In turn, this means that every single object that is inserted into phonological representations as the output of translation will have an effect: inserting a "sleeper" is impossible. Therefore, if carriers of morpho-syntactic information are to be non-diacritic, their distribution is necessarily privative. The second argument concerns the bare observation that only a small subset of morpho-syntactic information is phonologically relevant. This fact at least puts the burden of proof on those who wish to argue for nonprivative translation. Why should the grammatical system bother putting the translational mechanism to work in order to translate things that will never be used? And why should useless structure (boundaries, prosodic constituency) be built on the phonological side? The minimalist philosophy is certainly privative. That is, GB syntax in the 80s was full of useless structure: functional categories were built regardless of whether they would be relevant or not. The minimalist perspective is privative: phrase structure is only built when it serves a purpose in the derivation. Minimalist privativity is the result of a concern for extra-linguistic resources (§306) such as active memory, which non-privative structure wastes. Along the same lines but on the procedural side, Bermúdez-Otero (forth a:21ff) discusses the unwarranted empirical consequences of untempered proliferation of cycles. He concludes that selective spell-out is required for that reason. Current syntactic phase theory is an argument that has already been made: if this perspective is on the right track, and if there is only one spellout mechanism (see §763 on this question), the fact that derivation by phase relies on selective spell-out excludes systems where all nodes are interpreted.

Privativity 645

Finally, the general properties of intermodular communication that are discussed in §654 provide good evidence for privativity: in language as much as in other cognitive functions, the information that is made available to the receiving module through translation is only a small (and arbitrarily chosen) subset of what the sending module offers. In Vol.2, an additional argument will be made: local intervention in phonology (see §706) rules out non-privative translation of morphosyntactic information.

762

Chapter 6 Open questions (procedural)

763

1. Selective spell-out and interpretation-triggering affixes

764

1.1. Selective spell-out: what it is and how it works In SPE, interpretational units were defined by each morpheme (and of course word) boundary, except in case of successive affixes of the same major category, which belonged to the same cycle (see §103). Hence a word like theatricality is made of five morphemes that are distributed over three cycles: [[[theatr]N ic + al]A i + ty]N, that is three interpretational units. The idea that only a subset of morpho-syntactic divisions delineate interpretational units, and that it is a lexical property of affixes which decides whether or not such a unit is created, was introduced by Halle & Vergnaud (1987a) (see §§220f,225). Translated into the perspective of the morpho-syntactic tree and a spell-out operation, we are talking about selective spell-out of nodes and interpretation-triggering affixes: affixes are lexically specified (by some feature) for triggering interpretation or not. Upon concatenation, this feature is projected and identifies the dominating node. When the spell-out mechanism works through the tree, nodes that bear the spell-out feature – phase heads in modern syntactic terminology – are sent off to PF/LF. Or rather, which portion of the tree exactly is interpreted when spell-out hits a phase head is subject to debate. (312) selective spell-out: two options a. the phase head itself is spelled out (Halle & Vergnaud 1987a) phase head β tr. affix

PF/LF α

x

b. the sister of the interpretationtriggering affix is spelled out (Kaye 1995) phase head β tr. affix

root

α x

PF/LF root

648 Chap 6: Open questions (procedural) Two options are explored in the literature: either the phase head itself (Halle & Vergnaud 1987a) or the sister of the interpretation-triggering affix (Kaye 1995) is interpreted. This contrast is depicted under (312) above. In both cases, β is the projection of an affix that is lexically specified for triggering interpretation (a "triggering affix"). It is therefore a phase head. It is important to note that the rest of what happens, i.e. the interpretation of exactly which node is triggered, is a matter of the spell-out mechanism, not of any lexical property of affixes (or of two distinct types of phase heads). This is made explicit under (313) below. (313) how does the spell-out mechanism work? a. it sends the node that interpretation-triggering affixes project (i.e. the phase head) to PF/LF (Halle & Vergnaud 1987a) b. it sends the sister of interpretation-triggering affixes to PF/LF (Kaye 1995)

This means that the difference between Halle & Vergnaud (1987a) and Kaye (1995) has got nothing to do with selective spell-out itself: (314) below makes explicit what this notion implies – Halle & Vergnaud subscribe to all points as much as Kaye does. (314) selective spell-out: definition spell-out is selective iff a. there are nodes that are not spelled out (but will be phonologically realised). b. lexical pieces are specified for triggering or not triggering interpretation. c. this lexical specification is inherited by the node that the pieces project: the nodes in question are phase heads. d. the spell-out mechanism is sensitive to phase heads: only these nodes trigger the shipping of material to PF/LF.

The following section discusses the fact that Kaye's system appears to provide for what is known as the phase edge in current syntactic theory.

765

1.2. Phase edge and piece-driven phase

766

1.2.1. Spell out your sister: the phase edge in phonology Quite striking a parallel between Kaye's take on spell-out, (313b), and Chomsky's notion of phase edge is to be recorded (Scheer 2008c, 2009b).

Selective spell-out and interpretation-triggering affixes 649

Current phase theory holds that in case XP is a phase head, the spellout of XP only triggers the interpretation of the complement; the head and Spec,XP – the edge of the phase – are spelled out only at the next higher phase (Chomsky 2000a:108, see the quote from Chomsky 2001 under (128) in §306). Kaye's version of interpretation-triggering affixes and Chomsky's phase edge are contrasted under (315) below. (315) the phase edge in syntax and phonology a. Chomsky (2000a, 2001) b. Kaye (1995) phase phase head head XP Spec

β X'



tr. affix comp

PF/LF

α x

PF/LF root

Saying that Kaye's spell-out is the phonological version of the phase edge is of course imprecise: it is the result of phonological evidence, but it concerns morphology. That is, where Chomsky's mechanism spells out words and larger chunks, Kaye's system is about the spell-out of morphemes. What the parallel really is about, then, is syntax and morphology: chunks of whatever size seem to be spelled out by the same mechanism. That is, once it is determined that a given node is a phase head, the sister of its head is spelled out. In syntax, this is the complement (i.e. the sister of the head of the phase head-XP); in morphology, it is the sister of the interpretation-triggering affix, whose projection is the phase head. This is certainly a relevant contribution to the debate whether morphology and syntax are instances of the same computational system or not – an issue that runs through the the entire book (e.g. §533) and will crop up again shortly (§769). Also, the motivation for devising the spell-out mechanism to send the sister of the head of the phase head to PF/LF is the same with Chomsky and Kaye: modification-inhibiting no look-back (Phase Impenetrability). That is, only material in the complement of XP is frozen by Phase Impenetrability upon spell-out of XP; the edge is still accessible for extraction even when XP has been interpreted. Since material that is dominated by a phase head may still be active in further derivation, the phase edge offers

650 Chap 6: Open questions (procedural) an escape hatch where things can move to in order to escape the petrification of Phase Impenetrability. On the phonological side, the interpretation of the sister is necessary in order for it to be unable to be modified by later interpretation: the stress of [[párent] hood] cannot be modified on the outer cycle because [párent] was already interpreted independently. Related questions are the definition of phasehood and geometric properties of syntactic and morphological structure (the fact that (315b) does not appear to follow X-bar: there is no Spec). The latter debate would lead too far afield in this book: the status of X-bar structure and specifiers are subject to debate in current syntactic theory (e.g. Starke 2004). The former issue is discussed at greater length in §771: phasehood is also a topic of current syntactic debate. 767

1.2.2. Piece-driven vs. node-driven phase A last thing to be mentioned is a difference between syntactic and "phonological" (i.e. morphological) spell-out: in current syntactic phase theory, the question which nodes exactly are phase heads is at the forefront of the research agenda. What nobody doubts, however, is that phasehood is a property of nodes – this is what I call node-driven phase. That is, vP and CP are always phase heads (on Chomsky's 2000a count) independently of the lexical material that they contain. Of course they are also projections of terminals, but these terminals are grammatical, not lexical items. By contrast on the "phonological" side, the label of nodes plays no role at all in the definition of phasehood: following the notion of interpretation-triggering affixes, a node inherits phasehood, i.e. becomes a phase head, because the affix that it is the projection of bears a lexical feature that specifies is interpretation-triggering property. This is what I call piecedriven phase (Scheer 2008c). The opportunity for syntactic phase theory to consider piece-driven phase as a serious option is further discussed in §§781f (namely in the light of den Dikken's 2007 Phase Extension).

768

1.3. Systems with non-selective spell-out: Lexical Phonology (DM) The Lexical Phonology tradition practises non-selective spell-out: it subscribes to the idea that lexical pieces bear a specification that determines their spell-out; but unlike in selective spell-out, all affixes are armed with

Selective spell-out and interpretation-triggering affixes 651

this kind of feature. The spell-out mechanism is then sensitive to two (or several) specifications that a node may bear, and sends material to distinct computational systems at PF (the strata/levels). That is, there are not just phase heads, but there are several types of phase heads, level 1 and level 2. Or rather, it does not really make sense to talk about phase heads at all since there are no nodes that do not enjoy the interpretation-triggering status. In other words, what opposes selective and non-selective spell-out is privativity (§760): the former operates a yes-no distinction among nodes regarding their interpretation, while the latter implements a yes1-yes2 contrast: spell-out takes always place, but not in the same way.171 Another candidate for non-selective spell-out is Distributed Morphology, where all categorisers (nP, vP, aP) trigger interpretation, but functional heads do not (§555). While it is true that on the face of it spell-out is thus selective in DM, the kind of selection that is operated is quite different from what we know from syntax: rather than depending on the node (nodedriven phase), it depends on a class of node labels. The trouble that this system runs into when facing affix class-based phenomena and when compared to current syntactic phase theory was discussed in §583.

769

1.4. Spell-out is selective in syntax, but is this an argument? In order to assess the different competing options, let us first look at nonselective spell-out. At first sight, a strong argument against this orientation is the fact that syntax does not seem to work this way (the same argument will be made in §828 against morpheme-specific mini-phonologies): if current phase theory hits any close to the mark, spell-out is selective. That is, only a (small) subset of syntactic nodes are phase heads. It is true that the trend in syntax is towards smaller and smaller interpretational units, i.e. towards more and more phase heads that are lower and lower in the tree – but whether the vanishing point of this evolution, Spell-Out-as-you-Merge 171

The classical take of Lexical Phonology is to apply level-specific phonologies only at the end of strata, that is when two successive morphemes belong to different affix classes. This option is reminiscent of SPE (see §103) and empirically indistinguishable from a situation where literally every boundary triggers interpretation (which is what some versions of Lexical Phonology actually practise). In any event, Lexical Phonology does not distinguish between interpretation-triggering and interpretation-neutral affixes: all affixes trigger interpretation.

652 Chap 6: Open questions (procedural) (Epstein et al. 1998:46ff), can or will be reached is a question open to discussion (see §§775ff). As of the current state of affairs, a perspective where all nodes trigger spell-out is largely unsupported in syntax. A premise of this argument, though, is that there is only one single spell-out mechanism. Defenders of Lexical Phonology can argue that this premise is wrong. If syntax and morphology are two independent computational systems, it is reasonable (though not necessary) to assume that they will also have distinct spell-out mechanisms. These may then not work in the same way: syntactic spell-out (i.e. of chunks at and above the word level) may be selective as Chomsky says, while morphological spell-out (i.e. of chunks below the word level) spells out all nodes. This is exactly the position of Lexical Phonology indeed: a fundamental architectural property of classical Lexical Phonology is precisely that morphology and syntax are two distinct computational systems (see §152). Many (but not all172) modern heirs of Lexical Phonology follow this conception (see §539), which is precisely the reason why Distributed Morphology has an issue with (classical and modern versions of) Lexical Phonology (§533). We are thus back to the question whether morphology and syntax are distinct computational systems, or just one engine. It is only when this question is decided that the argument in favour of selective spell-out will turn out to bite or not (Scheer 2009b). In sum, it is hard to see how a decisive argument could be made in order to referee the competing perspectives of Lexical Phonology and selective spell-out. The empirical coverage of both approaches that is examined in §831, though, may indicate an advantage of the latter.

172

Bermúdez-Otero (forth a) is a modern heir of Lexical Phonology, but who does not necessarily subscribe to all design properties of classical Lexical Phonology (see §488). Bermúdez-Otero (forth a:21ff) compares Orgun's (1996a) nonprivative isomorphism (Orgun argues for direct syntax where phonology has full access to morpho-syntactic structure, see §512) with what he calls impoverishment. Impoverishment is privativity: it holds that "every domain [i.e. phase] is coextensive with some grammatical unit [i.e. node in the morpho-syntactic tree], but not all grammatical units create domains" (Bermúdez-Otero forth a:26). This shows that there is no principled incompatibility of selective spell-out and Lexical Phonology.

Selective spell-out and interpretation-triggering affixes 653 770

1.5. Conclusion: a unifying perspective In absence of decisive evidence regarding the single engine question, the unifying perspective of selective spell-out is certainly appealing and constitutes evidence in favour of selective spell-out in general, and in support of Kaye's version thereof in particular. On the empirical side, Kaye's system affords a coverage of the two fundamental patterns of affix class-based phenomena, the rule-blocking (level 1 rules) and the rule-triggering (level 2 rules) pattern (§310). Halle & Vergnaud on the other hand are unable to analyse the latter (see §250 and the comparison of empirical coverage in §831 below).173 Selective spell-out in general and Kaye's system in particular offer a converging perspective with syntax regarding three issues: selective spellout itself, the edge of the phase ("spell out your sister!") and modificationinhibiting no look-back (i.e. Phase Impenetrability, see §299). Recall from §279 that Kaye's system achieves underapplication by Phase Impenetrability. The unifying perspective is further discussed in §852 below. The latter parallel regarding Phase Impenetrability is a little different from the two former since it concerns an effect, rather than a property of the spell-out mechanism. Its independence from the single engine issue makes it a stronger argument: the presence of syntactic PIC effects is a good reason (a compelling reason for Chomsky, see the quote under (126) in §305) for predicting that phonological effects of the same kind will also be found.

173

It is true that Halle & Vergnaud can do anti-affix ordering items (govern-mént2-al1, §246), which appear to violate Phase Impenetrability and are therefore troublesome in Kaye's system (see §315). In fact, the same items are problematic for Halle & Vergnaud as well: they violate all no look-back devices, including the SCC-K to which Halle & Vergnaud subscribe. The problem is probably more general and has to do with stress: suprasegmental processes have long been observed to violate no look-back generalisations. This issue is discussed at greater length in §§554,780.

654 Chap 6: Open questions (procedural) 771

2. Phase theory

772

2.1. A rapidly growing field whose diversity can be confusing at times (not only for phonologists) Since Uriagereka (1999) and Chomsky (2000a, 2001), phase theory evolves at high speed, but today is still in an embryotic state and rather poorly understood (Boeckx & Grohmann 2007). Many different directions are explored, the relevant literature is growing rapidly, and the evolution is sometimes difficult to follow, especially for a phonologist. The overall picture may be quite confusing at times: everybody seems to have his/her private way of doing phases, many of which are incompatible. The common denominator reduces constantly, probably to not much these days, except that 1) things are sent off to LF/PF and 2) come back frozen (to various extents). Below I attempt to provide an overview of the directions that are taken, of their relation and (in)compatibility with Chomsky's original idea, and of their eventual meaning in or interpretation by phonological (phase) theory. Needless to say, all will be looked at through the prism of the phonological point of view, which means that almost all technical detail will be left out, to the effect that things which are important for syntacticians may be left unmentioned; by contrast, properties of phase theory that are not usually on the syntactic radar may be highlighted because they are relevant for phonology. Overview literature includes Richards (2004:57ff), Frascarelli (2006), Boeckx & Grohmann (2007), den Dikken (2007) and Grohmann (2007a,b).174 Finally, due mention needs to be made of the fact that there are also voices which believe that phase theory is not a good thing to have in linguistic theory. Scepticism is fed by various objections: Boeckx & Grohmann (2007) (see also Matushansky 2005) invoke principled reasons, the fact that too many problems arise, that too much confusion is created by the blooming diversity of approaches that is reported below, that the theory is riddled with internal inconsistencies (especially Matushansky 2005), and that the conceptual motivation is too poor. 174

The latter are the introductions to two successive special issues of Linguistic Analysis that Grohmann has edited. Under the general header of Dynamic Interfaces, these concentrate a number of relevant papers (one issue, 33.1-2, is about phases and derivations, the other, 33.3-4, inquires on spell-out and linearisation).

Phase theory 655 773

2.2. What counts as a phase?

774

2.2.1. Inner- and extra-syntactic criteria for distributing phasehood One question that anybody who accepts the frame of derivation by phase has to answer is what exactly counts as a phase. As far as I can see, common to all versions of phase theory is the fact that phasehood is a property of nodes. This is what I call node-driven phase, as opposed to piece-driven phase (see §767). In this perspective, the initial question is thus about the properties that a node must possess in order to be granted the privilege of a phase head. Once the criteria for phase head-eligibility defined, a further question is which nodes exactly satisfy them. While the former question is covered in the present section, the latter issue is discussed in the following sections. We will see that the election of a node as a phase head does not always follow from the application of the criteria that ought to govern the distribution of phasehood: typically, a node is grated the phase head privilege for practical reasons, i.e. because some data are argued to require spell-out at some stage of the derivation. Based on Chomsky (2000a:106) and other practice, den Dikken (2007:27ff) offers a comprehensive discussion of the properties that qualify a portion of the tree for being a phase. Chomsky tries to derive phasehood from extra-syntactic properties, which is certainly the strongest way to go about the question: ideally, syntax-internal considerations such as "this data set requires spell-out at node X" will be toothless, or rather, will be able to be evaluated on extra-syntactic grounds. Chomsky (2000a:106) defines a phase as a "natural syntactic object SO", that has a certain degree of independence both phonologically and semantically. Being independent means being prosodically isolable and movable on the phonological side, and being propositional on the semantic side, which means for example that phases are expected to be T-complete. A syntax-internal – thus weaker – control over phasehood that Chomsky proposes is the idea that phases are probes (i.e. loci of uninterpretable formal features), and more recently, that they "have an EPPposition as an escape hatch for movement and are, therefore, the smallest constructions that qualify for Spell-Out" (Chomsky 2004:124). As was mentioned earlier, the coronation of a node as a phase head does not always depend on these criteria in the actual practice that is found in the literature. The atomising evolution towards smaller and smaller

656 Chap 6: Open questions (procedural) phases that is described in the following section weakens (or rather, undermines) especially the extra-syntactic control over phasehood. Also, the literature has questioned Chomsky's criteria for phasehood, which are found to be non-operational (Legate 2003, Boeckx & Grohmann 2007). One argument that attacks the heart of the escape-hatch mechanism is the following.175 According to Chomsky, the item that should enjoy extra-syntactic independence is whatever is actually spelled out. Since, however, only the complement of vP and CP, rather than vP and CP themselves, is sent to LF/PF upon the spell-out of vP/CP, it should be the complement that enjoys interface independence. That is, complements of phases, not phases themselves, should be isolable units regarding sound and meaning. This, however, is obviously not the case. 775

2.2.2. The trend is towards atomisation Chomsky's original take on phasehood identifies CP and vP, maybe DP (Chomsky 2005:17f), as phase heads. The DP track has been followed, and also DP-internal phases are argued for (Svenonius 2004:265f, Matushansky 2005, Bošković 2005b). Den Dikken (2007:33ff) provides an overview of the DP issue, arguing that "there is good reason to believe that DP and CP are each other's counterparts in their respective domains (the noun phrase and the clause)" (p.33). TP is also under debate: while Chomsky (e.g. 2000a:106, 2004:124, 2008) is explicit on the fact that TP does not qualify as a phase head (because it is not propositional, see also Richards 2007b and the discussion in Abels 2003:62ff, 116ff), den Dikken (2007:29ff) points out that according to Chomsky's own criteria (see §774), this conclusion is not really obvious. Indeed, TP is assumed to act as a phase head in the literature, and also nodes below TP such as Voice° (Baltin 2007, Aelbrecht 2008) and AspP (Hinterhölzl 2006) are sometimes granted phasehood. Finally, Abels (2003:227ff) argues that PP may also (parametrically) be a phase head.176 A label-insensitive proposal that goes farther than the above quoted is Müller's (2004), who suggests that every maximal projection is a phase head. 175

176

The argument is made by Boeckx & Grohmann (2007:215), who report that it has three independent origins, Abels (2003), Grohmann (2003) and Epstein (2006). Samuels (2009a:248) also provides an overview of the syntactic phasehood literature.

Phase theory 657

Finally, it needs to be mentioned for the sake of completeness that – predictably enough – phasehood has also been proposed to be subject to cross-linguistic parameterisation: Gallego (2007, 2010) entertains a perspective where different nodes enjoy the phase head privilege in different languages (also Abels 2003:227ff for PP).

776

2.2.3. Is there a lower limit for phasehood? Is spell-out selective? Given the trend towards atomisation, the vanishing point of the evolution is that all nodes trigger interpretation or, in other words, that interpretation occurs upon every application of Merge. As a matter of fact, this view has actually been proposed in work by Samuel Epstein and others: Epstein et al. (1998:46ff), Epstein & Seely (2002:77ff, 2006:174ff) (see also §305). (316) "LF and PF necessarily evaluate linguistic entities at every point in the derivation. Thus, if X and Y are merged creating C, then C is necessarily input to both LF and PF." Epstein & Seely (2006:174, emphasis in original)

The question that the atomisation movement raises from the phonological perspective is not so much how many nodes are ennobled as phase heads, which nodes exactly are concerned and how high (or low) they are in the tree. Rather than about the granularity of phase heads, the question is whether there is a principled limit to atomisation. Or, in other words, whether there are any nodes left that must not trigger interpretation. If such nodes exist, spell-out is selective (in the sense of §§228,763). If no principled reason prevents atomisation from hitting the programmed vanishing point where Samuel Epstein waits for the rest of syntax, spell-out may as well turn out to be non-selective. In this case, the intermodular argument that was made in §769 is vacuous: if syntactic spell-out may or may not be selective, phonological spell-out may just as well be selective or not. It is only when syntax has a firm (selective) position that an argument in favour of selective spell-out can be made in phonology. What it takes for the intermodular argument to go through, then, is a clarification in syntax: can non-selective Spell-out-as-you-Merge be excluded or be reasonably disqualified? The following section discusses an argument to this end that is made in the syntactic literature.

658 Chap 6: Open questions (procedural) 777

2.2.4. Anti-locality of movement marshals atomisation One effect of phase theory which is always put to its credit is the fact that it provides an elegant and more general way of deriving the cyclic character of long distance movement: the labour that was done by Barriers (and bounding nodes, relativized minimality) before is now afforded by Phase Impenetrability (see §679, e.g. Rizzi 2005, Frascarelli 2006:2f). Counter this apparent convergence, Richards (2004:59f) argues on empirical grounds (defective intervention, i.e. cases where a single head can be valued by two distinct goal categories) that intervener-relativized locality is not reducible to a phase-based account. That is, the cycles of probes and goals do not necessarily coincide: while the former obey strict XP-by-XP cyclicity, the cycle of the latter must sometimes be larger. Boeckx & Grohmann (2007:211ff) also argue that the link between phase theory and successive-cyclic movement may actually turn out to challenge the former since a widely held view on the latter is that every maximal projection is a cycle (Boeckx & Grohmann mention a host of relevant literature). Under the PIC-analysis of successive-cyclic movement, this would mean that every maximal projection is also a phase (which is what Müller 2004 proposes) – thus further atomising phasehood. Boeckx & Grohmann argue that this leads to a contradiction since on Chomsky's count every phase induces a Phase Impenetrability effect. If all XPs are subject to Phase Impenetrability, however, "no extraction would be possible, as the complement of any phase would have to move to the edge of that phrase/phase, a movement step that would count as too local under any version of 'anti-locality'" (Boeckx & Grohmann 2007:212). Beyond the intended goal (which is to show that phase theory as such is in trouble), the point that Boeckx & Grohmann (2007) make is that antilocality (Grohmann 2003, Abels 2003:91ff) marshals the atomisation of phasehood. In the evolution that makes smaller and smaller chunks of the tree phase heads, there is a level where the phase edge (see §765 on this notion) will not be able to act as an escape hatch anymore for material that is trapped in the complement: anti-locality will prevent it from escaping. This purely syntactic argument – if it goes through – appears to be an important reference point in the ongoing discussion of phasehood: Chomsky's original distribution of phase heads was certainly too coarse, but the atomising evolution has also a lower limit, which is set by anti-locality constraints on movement. What anti-locality ultimately does, then, is to guarantee that spell-out will always be selective, no matter how fine- or coarse-grained phasehood

Phase theory 659

turns out to be. This result is relevant for phonology because it enables intermodular argumentation along the lines that were mentioned in the previous section: (non-)selective spell-out draws a red line between the two major conceptions of how procedural communication works that compete on the phonological side (see §763).

778

2.3. Extensions of the basic model

779

2.3.1. Asymmetric spell-out: independent access of LF and PF Beyond the central issue of phasehood, a number of proposals have been made how phase theory works. Most of them are plug-ins in the sense that they may or may not be subscribed to independently of each other, and also independently of how phasehood is sorted out. One question that has already been addressed in §567 is asymmetric spell-out, i.e. a scenario where LF and PF can be accessed independently. This perspective considerably weakens phase theory: "minimal pairs" such as cómparable vs. compárable where semantic and phonological opacity go hand in hand (the former is opaque on both sides, the latter on none, see §545) would be a mere accidental coincidence of LF and PF interpretation that occurs at the same node. §567 reproduces a quote of Chomsky's (2004:107) in this direction, and provides relevant literature. Nonetheless, there are also arguments in favour of asymmetric spellout (§567), and the question is open for further debate.

780

2.3.2. PIC à la carte: process-sensitive no look-back Another issue that we have come across, and which will be relevant below (§794), is what was called PIC à la carte in §554. No look-back devices are riddled with specific counter-examples for a long time, and these appear to not just involve any process. §556 has pointed out that stress is a notoriously bad guy who has violated the SCC-K (hence the derived environment generalisation) in the 80s (§192), and also does not care for Phase Impenetrability in Marvin's (2002) analysis (§555). On these grounds, it was suggested in §241 that Phase Impenetrability could apply à la carte, i.e. for a given phase constrain the application of some processes, but not of others. Certainly a relevant research question today is whether it is possible to identify a specific class of processes that violate no look-back, as op-

660 Chap 6: Open questions (procedural) posed to another class that produces no no-look back-effects. In the 80s, the candidate rationale was structure-building (able to violate no look-back) vs. structure-changing processes (unable to violate no look-back, see §§193f) whereby the prime candidates for the former were stress (construction of foot structure) and syllabification (construction of syllable structure) (see §293). Another track is the segmental vs. suprasegmental distinction (processes of the latter, but not of the former kind, are able to violate no lookback). That is, tone is found to violate Phase Impenetrability as much as stress does (Marlo 2008). What that would come down to is process-sensitive Phase Impenetrability. The reason why relevant phonological evidence is discussed here is that the same idea is also present in the syntactic literature: Bobaljik & Wurmbrand (2005:828f) and Bošković (2007) for example hold that Move, but not Agree, is subject to the PIC: agree relationships are notoriously long-distance and may reach into a lower frozen phase. 177 Grohmann (2007a:13) generalises the issue raised by Bošković's case study: does the PIC really apply to all narrow-syntax operations? Further study must show whether the line that separates the pool of phenomena that is and the pool of phenomena that is not impacted by Phase Impenetrability can be stated according to any rationale, i.e. whether the two sets can be described as "natural classes" of events, both syntactically and phonologically. Phonologists have certainly a serious advance in this field: even though the issue is not top ranked on the research agenda since the 80s (to say the least), they have thought about the question for a long time, and some cross-linguistic data are available. By contrast on the syntactic side, the eventuality of process-sensitive Phase Impenetrability is an entirely new (and rather exotic) perspective; as far as I can see syntacticians have not begun to try to find a rationale for properly separating violating and non-violating processes. This is certainly not unrelated to the fact that PIC à la carte is quite unminimalistic an idea: the PIC is an instrument of computational economy, and according to Chomsky economy conditions are always obeyed. PIC à la carte, then, amounts to economy à la carte. Note, however, that Chomsky (2004:107f) senses that "spell-out and forget" may be too strong 177

Another solution for the same pattern is also entertained in the literature: since in Romanian and Greek a long distance agree relationship (between a matrix verb and the embedded subject) reaches into a CP in violation of the PIC, Alexiadou (forth:15) hold that the CP layer (and hence the associated phase) does not exist in these languages.

Phase theory 661

a condition on phonological computation (see §306, more on this in §§796f).

781

2.3.3. Phase Extension: when phasehood depends on what the head is made of Since Chomsky's inception of derivation by phase, one of the few properties that appear to remain stable through the blooming evolution of the theory is the fact that phasehood is a rigid property of nodes. As den Dikken (2007) puts it, "once a phase, always a phase", and "not a phase at the outset, then never a phase". That is, phasehood is not syntactically manipulable: whatever happens in the course of a syntactic derivation has no influence on which node is a phase head, and which node is not. Den Dikken (2007) explores the opportunity of a more dynamic outlook on phasehood: nodes continue to be defined as phase heads upon construction, but this is only an option they take for being interpretationtriggering in the end. Whether or not pre-specified candidates really are the point of PF/LF access at the end of the day is determined by what happens in the syntactic derivation.178 That is, den Dikken proposes that the phasehood of a node which was acquired through initial node-based distribution may move together with the material that it dominates. This privilege, however, is restricted to heads and head movement. In practice, this means that a node which was not a phase head qua "initial distribution" may inherit this property from material that it has come to dominate through movement if this material was a phase-head (qua "initial distribution") in its base position. Since movement is only upwards, den Dikken calls his theory Phase Extension: its effect is that a syntactic derivation may cause interpretational units (i.e. the material that is sent to the interfaces as a block) to enlarge, but not to shrink. What is called "initial distribution" above is a not an obvious notion in den Dikken's theory. For example, CPs are deprived of the phase head privileges that Chomsky's system grants them per se; this does not mean, however, that they cannot be phase heads at the end of the day: depending 178

A related take where effective spell-out of a node depends on the properties of the syntactic derivation is convergence-based phasehood as proposed by Svenonius (2001) and Felser (2004). The idea is that phrases are sent to interpretation as soon as they are ready, i.e. when all of their unvalued features have been checked.

662 Chap 6: Open questions (procedural) on the syntactic derivation, they may become an interpretation-triggering node if they receive material that was a phase head in its old position. It is not worthwhile going into greater detail here, den Dikken explains at length how phasehood is distributed outside of inheritance through movement. What is important for the comparison with what I call piecedriven phase (which is practised in phonology) (see §§767,782) is profiled under (317) below. (317) Phase Extension – relevant properties for the comparison with phonology a. phasehood is defined by syntactic computation, rather than given. b. the phasehood of heads moves together with the head: if a phase head moves, it will also be a phase head in its new position. c. nodes may inherit phasehood from the material that they dominate; that is, the phasehood of a node may be a projection of its head.

Especially (317c) is useful for the comparison of how phasehood is defined in syntactic and phonological theory, as we will see in the following section.

782

2.3.4. Unification of piece-driven and node-driven phase by a lexical phasehood feature Den Dikken's (2007) Phase Extension may be interpreted as a first step towards a lexical definition of phasehood, i.e. where interpretation is triggered by a lexical property of pieces, a "phasehood feature", that is projected into morpho-syntactic structure. Recall that this is how all phonological effects of affix class-based phenomena work (piece-driven phase, see §§225,767). While the lexical origin of phasehood is thus undisputed below the word level (piece-driven phase), phasehood is thought of as a property of node labels in syntax (node-driven phase, see §§767,781). The assumption in syntax is that phase heads as such do not share any property that could be predicted from the syntactic derivation. Rather, the spell-out mechanism executes a hard-wired programme where it is specified that this node does, but that node does not trigger interpretation. The idea of a phasehood feature allows us to extend the phonological piece-driven phase to syntax: phase heads bear a phasehood feature that is projected by their head. The spell-out mechanism, then, is blind to labels: it only looks at the phasehood feature, spelling out nodes iff they have it.

No look-back (Phase Impenetrability) 663

In this perspective, nodes such as vP and CP have inherited their phasehood by a projection from a lexical item: vP is the result of the merger of v and VP, which means that little v is the carrier of the phasehood feature. If, say, TP is not a phase head (either universally or parametrically), there is no equivalent lexical "little t" that carries the phasehood feature. The parallel between an interpretation-triggering morpheme and little v is shown under (318) below. (318) piece- and node-driven phase unified: the head bears an interpretation-triggering (phasehood) feature a. "node-driven" phase b. "piece-driven" phase vP β Spec

v' v

class 2 VP

PF/LF

α x

PF/LF root

What is shared by all phase heads, then, is the fact that their head is a lexically recorded item that possesses a phasehood feature: little v under (318a), the class 2 affix under (318b). In both cases, the sister of the interpretation-triggering item is spelled out (the phase edge, see §766). In this perspective, the impression that phasehood is node-driven above, but piece-driven below the word level is a mirage that stems from the fact that affixes are "material" in the sense that they have a direct phonetic correlate, while items such as little v do not.

783

3. No look-back (Phase Impenetrability)

784

3.1. Limited scope of this section No look-back devices in general and Phase Impenetrability in particular have been discussed all through Part I, and they are also an important piece of the argumentation in Part II: they appear on various occasions in the section on selective spell-out (§763), in the general section on (syntactic) phase theory (§771) where Phase Impenetrability plays the role of a referee for the definition of phasehood, in the section on the word-spell-outmystery where the non-cyclicity of the interpretation of word sequences is ascribed to the absence of the PIC at this level (§794), in the section on

664 Chap 6: Open questions (procedural) chunk-specific interpretation where the PIC is made process-specific (§823, also §780), in the section on morpheme-specific mini-phonologies where the PIC is used as an argument (§828) and also in the final synopsis of convergent syntactic and phonological interface devices (§852). The present section therefore offers only a brief survey of the role that no look-back devices have played in previous (phonological) interface theories as discussed in Part I.

785

3.2. No look-back devices since Chomsky (1973) and their nonconflatability with derived environment effects Phase Impenetrability is the modern incarnation of the idea that computation has no or only limited access to "old" strings, i.e. to those that have already been subject to interpretation. No look-back devices have appeared in a number of guises since Chomsky (1973), who has first made cyclic derivation subject to a no look-back condition. The relevant history is described in §287 at some length, where it is shown that the exact wording of the no look-back condition matters: different formulations produce quite different empirical effects (§302). Noteworthy is the fact that the current version of no look-back, Phase Impenetrability, has emerged in phonology in the 80s (see also Riemsdijk's 1978 Head Constraint in §298) and was generalised by Kaye (1992, 1995) (§299). This is where the idea of modification-inhibiting no look-back entered the scene: the modern Phase Impenetrability Condition is an implementation of this specific way to organise no look-back. The PIC "freezes" previously computed strings, which are then unavailable for further computation and are "forgotten" (§306). Even if the phonological version(s) of modification-inhibiting no look-back need to be worked out (the phonological situation requires a weaker formulation, §302), the general idea is the same. Finally, it is to be mentioned that the ambition to merge derived environment effects and no look-back effects into one single device was entertained for a long time in the 80s, where it actually shaped a good deal of the research agenda at the interface. The origins and evolution of this project, which revolved around Kiparsky's SCC-K, are described in §177. Its failure is obvious at least since Kiparsky (1993) himself has declared the bankruptcy of his SCC-K (see §197, although a branch of Lexical Phonology, represented by Rubach & Booij 2003, continue on this track). BermúdezOtero (2008, forth a:§2.6.3) even holds that the SCC-K was the major

Why are there no phonological effects of the cyclic spell-out of words? 665

brake that prevented Lexical Phonology from evolving and contributed a good deal to its temporal discredit during the 90s (§214). It is shown in §182 that sensitivity to affix classes and to derived environments is logically independent: either may occur without the other.

786

4. Why are there no phonological effects of the cyclic spell-out of words?

787

4.1. The absence of cyclicity-induced external sandhi: a consensual fact that theories build on, but do not talk about Traditional terminology distinguishes between internal and external sandhi. The former refers to phonological effects of morpheme boundaries, while the latter describes phonological effects of word boundaries. This distinction comes in handy for the purpose of this section, which is about an absence, i.e. something that the literature does not talk about: the cyclic spellout of words. While the procedural management of morphemes has spilled a lot of ink (this is essentially what Lexical Phonology is about), I have not come across either a phenomenon that requires, or an analysis that proposes a procedural treatment of words or bigger chunks – save the notorious case of intonation, which is discussed at length in §800. The absence of procedural activity above the word level – or rather: of phonological traces thereof – is admitted as the correct empirical record in the field. The literature therefore offers only representational treatments of syntactically conditioned phonological effects. In practice, this means that all external sandhi phenomena are ascribed to some variation in prosodic constituency. The exclusive ambition of representational management at and above the word level is rarely made explicit, though. The only case that I am aware of was discussed in §§4320f: Selkirk (1984) and Inkelas (1990) observe that while prosodic constituency can cover the full spectrum of units (morphemes and words alike), Lexical Phonology is confined to the Lexicon, i.e. to morphemes. Since there is no place for two devices (procedural and representational) that do the same job below the word level, Inkelas (1990) argues, prosodic constituency should be extended to the Lexicon. Lexical Phonology, then, is an empty shell at best. The pages below describe the attitude of different interface theories with respect to the absence of phonological traces of the cyclic spell-out of words (§788). This phenomenon, which I call the word-spell-out-mystery,

666 Chap 6: Open questions (procedural) is quite extraordinary if spell-out of words is indeed cyclic: why should phonology refuse to react on piecemeal fire at and above the word, but produce traces thereof below the word level? While no answer to this question will be provided, §794 evaluates what it takes to implement the wordspell-out-mystery in a phase-based environment.

788

4.2. How interface theories behave: claims for and against cyclic spell-out of words, for and against its cyclic interpretation

789

4.2.1. The baseline position of SPE: everything is cyclic Given this empirical situation and its reception in the literature, let us briefly review the spectrum of positions that have been taken by phonological interface theories. Two things need to be carefully distinguished: on the one hand the submission of pieces to interpretational modules, which may or may not be cyclic; on the other hand, the phonological (or semantic) interpretation thereof (i.e. of whatever is submitted), which may also be cyclic or not. The former is a matter of the spell-out mechanism, while the latter may appear to be a decision of the phonological computational system: either I ignore that I receive a string in pieces and act only at the end when the last piece has come in, or I do my job any time a piece is submitted.179 The baseline position is represented by Chomsky et al. (1956) and SPE where cyclic derivation was introduced: both the spell-out and the phonological interpretation of word sequences is cyclic. The relevant quote from §102 is reproduced below. (319) "The principle of the transformational cycle […] appl[ies] to all surface structure whether internal or external to the word" Chomsky & Halle (1968:27).

790

4.2.2. Lexical Phonology: the interpretation of word sequences is not cyclic Lexical Phonology has a different take. It was mentioned in §153 that all versions of this theory implement Praguian segregation, i.e. two distinct 179

We will see in §§797f that it is reasonable to believe that whether interpretation is cyclic or not is as much a decision made by the spell-out mechanism as the decision whether the spell-out itself is cyclic.

Why are there no phonological effects of the cyclic spell-out of words? 667

chunk-specific phonologies that compute sequences of morphemes (word/lexical phonology) and sequences of words (sentence/postlexical phonology), respectively. The issue regarding chunk-specific phonologies is discussed in §811 below – it is entirely orthogonal to the cyclic question: both lexical and postlexical levels may or may not be declared cyclic. The term "cyclic rule" (which is still used today in a kind of lingua franca-understanding) is indicative of the front line that is set in Lexical Phonology: early versions of the theory assumed that all phonological rules which contribute to word-formation (i.e. which apply to sequences of morphemes, that is in the Lexicon) are cyclic, while all postlexical rules (i.e. those that apply to sequences of words) are non-cyclic. The cyclic condition on lexical rules was called into question later on (Rubach & Booij 1984 introduced lexical post-cyclic – i.e. non-cyclic – rules, see §194), but the necessarily non-cyclic character of postlexical rules stands unchallenged in all Lexical Phonology quarters up to the present day. 791

4.2.3. Lexical Phonology makes no claim about spell-out and installs noncyclic interpretation of word sequences without argument On the other hand, as far as I can see, Lexical Phonology makes no claim regarding the cyclic character of spell-out. The only thing that is central for this theory is the contrast between the cyclic interpretation of morpheme sequences (lexical phonology), against the non-cyclic interpretation of word sequences (postlexical phonology). The reasons for this fundamental distinction, however, are not made explicit as far as I can see. One may suppose that postlexical phonology is declared non-cyclic on the grounds of the observation that cyclicityinduced external sandhi is absent from the record. This, however, is no more than speculation since, as was reported in §158, Kiparsky (1982b) simply decrees that there is no cyclic derivation of words without argument or discussion. (320) "The former, the rules of lexical phonology, are intrinsically cyclic because they reapply after each step of word-formation at their morphological level. The latter, the rules of postlexical phonology, are intrinsically noncyclic." Kiparsky (1982b:131f, emphasis in original)

668 Chap 6: Open questions (procedural) 792

4.2.4. Halle & Vergnaud and Kaye: restoration of SPE – everything is cyclic Let us now turn to Halle & Vergnaud (1987a), who are committed to the SPE heritage, but also to Lexical Phonology and the lexicalist environment of the 80s, of which Praguian segregation is an expression (§216). The result is a system where both morphemes and words are subject to cyclic spell-out; the concatenative process, however, takes place in two rounds, one where words are created, another where sentences are built (wordinternal vs. word-sequence strata in Halle et al. 1991). Word- and sentenceconstruction is separated by a specific word-level phonology (see §238). This much for spell-out. Within this architecture, then, all phonological interpretation is cyclic, no matter whether the input are morphemes or words. That is, following their general orientation (§219), Halle & Vergnaud restore SPE: 1) there are no morpheme-specific phonologies; 2) there is a chunk-specific phonology at the word level (SPE's word-level rules, see §104, which Halle & Vergnaud call non-cyclic rules), but not at the postword level (i.e. contra Lexical Phonology, there is no distinction between a lexical and a postlexical phonology: both chunk-sizes are interpreted by the same computational system); 3) all phonological interpretation that sees cycles is cyclic; that is, the unique computational system that assesses morpheme- and word sequences is cyclic (§238), but word-level phonology is not because it cannot: it is punctual and hence does not see any cyclic boundaries. Kaye's (1995) position is the same as Halle & Vergnaud's as far as I can see: Kaye rejects morpheme-specific phonologies (§303), provides for a word-specific phonology (§283), but has morpheme- and word-sequences interpreted by the same computational system, which carries out cyclic interpretation.

793

4.2.5. Distributed Morphology is entirely agnostic in phonological matters Finally, Distributed Morphology is entirely agnostic in regard of the issue at hand, simply because it is not concerned with, and does not make any claim about, phonological interpretation (see §535). From the vantage point of DM, morpho-syntax cannot accommodate multiple computational systems, but PF may or may not accommodate morpheme-specific and/or chunk-specific mini-phonologies, whose interpretational action also may or may not be cyclic.

Why are there no phonological effects of the cyclic spell-out of words? 669 794

4.3. The word-spell-out mystery

795

4.3.1. Cyclic spell-out of words but no phonological traces? From a global perspective, the situation thus seems paradoxical: cyclic spell-out of words and larger chunks – derivation by phase in modern terminology – is a central piece of current syntactic thinking, but it looks like it has no phonological consequences. By contrast, the cyclic spell-out of morphemes is just as undisputed, but – as expected – leaves ample traces in the phonology. Having distinct chunk-specific phonologies that distinguish wordand sentence phonology as proposed by Lexical Phonology does not solve the problem: it merely records the contrast between the area that produces phonological effects (internal sandhi) and the one that does not (external sandhi). What it does not, however, is to correlate the phonological noneffect for chunks at and above the word level with the other end of the interactionist pipe: we want to know how it could be that the same input to (an) interpretational system(s) – the piecemeal submission of a string hacked into pieces of growing size – in one case produces (opacity) effects, but in another case leaves no trace at all.

796

4.3.2. Wrong data or an on/off switch for Phase Impenetrability There are only two ways that I can think of how this mystery can make sense: either the empirical generalisation is wrong (phonologists have not worked hard enough, if they have a closer look, they will find cyclicityinduced external sandhi), or interpretational systems are able to ignore their input conditions. The latter option means that a phonological system (or a semantic system for that matter) has a switch that decides whether "old" strings, i.e. those that have already undergone previous computation, are subject to a special treatment or not. Or rather, as we will see in §799, the switch at hand is borne by the spell-out-mechanism, not by the phonological computational system. In other words, word phonology would feature a no look-back device, while sentence phonology has no Phase Impenetrability Condition and hence treats all strings in the same way, old and new alike. A third option that is logically possible is certainly not a serious candidate and may be left unexplored: spell-out of words and larger chunks could be non-cyclic (while morphemes are submitted piecemeal to interpre-

670 Chap 6: Open questions (procedural) tation). This would mean that cyclic derivation in general and interactionist derivation by phase in particular are flat out wrong.

797

4.3.3. A solution: (chunk-specific) PIC à la carte Let us pursue the option according to which interpretational systems are parameterised for subjecting or not subjecting "old" strings to a special treatment. An interesting question is what "special treatment" actually means, and by whom it is organised. As far as I can see, the literature on no look-back devices in general, and on Phase Impenetrability in particular, does not really address this issue: no look-back is declared to be a property of the system, but what it actually takes for the system to implement the desired effect remains vague. It was mentioned in §306 that the (Chomsky's) whole motivation for hacking sentences into pieces and sending them to PF/LF piecemeal is to be able to achieve computational economy regarding active memory by imposing the Phase Impenetrability Condition on "old" pieces. The economy effect is achieved by allowing further computation to "forget" these "old" pieces and their internal structure. Also recall from §306 (see the quotes under (128) and (129)) that Chomsky (2001:12f, 2004:107f) is explicit on the fact that the economy of active memory concerns phonological as much as syntactic computation ("the phonological component too can 'forget' earlier stages of derivation"). Which means that Chomsky takes Phase Impenetrability to be a general condition on computational systems (at least in grammar). This is in line with the general minimalist philosophy where conditions on computational economy are always obeyed. Note, however, that Chomsky (2004:107f) senses that "spell-out and forget" may be too strong a condition on phonological computation (§306, quote (129)). This is what was observed in §302 anyway, and the alternative that we are currently exploring, PIC à la carte (see also §780), points in the same direction: a computational system may "choose" to benefit from computational economy or not. That is, a PIC condition lies on the phonological interpretation of sequences of morphemes (computational economy realised), while sequences of words remain unmarshalled by the PIC (computational economy waived). Note that this type of PIC à la carte is different from the one that we have already come across: the PIC à la carte discussed in §780 is process-specific, while the mechanism explored here is chunk-specific. In one case the type of process decides (for a given phase

Why are there no phonological effects of the cyclic spell-out of words? 671

boundary), while in the other the size of the chunk determines whether or not the PIC applies (chunks below the word yes, at and above the word no).

798

4.3.4. PIC à la carte that does not want to be named (Samuels 2009a) Samuels (2009a) has comet to the same conclusion, (chunk-specific) PIC à la carte, but refrains from admitting that the PIC is stalled above the word level. Her formulation is that it applies in two different ways, whereby the way to apply above the word level is not to apply. Samuels (2009a:248f) follows the core properties of the DM architecture and the DM definition of phase heads below the word level (categorisers have this status: n,v,a, see §555). From a minimalist and phase-based perspective, she holds that "spell-out domains are the only domains that phonology needs" (Samuels 2009a:249, emphasis in original), where spellout domains are phases (both below and above the word level). This leaves the question open what should be done with phases that phonology does not need (i.e. which do not leave any phonological footprint): this is exactly the issue that is raised by the word-spell-out-mystery. Samuels adheres to the rigid syntactic take that each and every phase is marshalled by a PIC condition (PIC violations are only apparent: Samuels 2009a:265ff). On this backdrop, she converts the lexical vs. postlexical distinction into a phase-based formulation whereby the PIC of postlexical rules is only enforced above the word level (Samuels (2009a:260ff). (321) "I maintain the distinction between these two rule types by arguing that all phonological rules obey the Phase Impenetrability Condition, but in one of two different ways. Lexical rules must obey the Phase Impenetrability Condition at both the morpheme level (phase heads n, a, etc.) and the clausal level (phase heads v, C, D, etc.); […] Post-lexical rules apply once a sequence of morphemes has been turned into an atomic unit and obey the Phase Impenetrability Condition only at the clausal level." Samuels (2009a:262)

Samuels thus follows Lexical Phonology in all points: 1) there are two distinct computational systems, lexical and postlexical, 2) processes (rules) are assigned to either (or to both), and 3) postlexical processes are blind to phase structure (i.e. above the word level) and ignore the PIC. Terminology and notational variation set aside, this is exactly the solution that is referred to as PIC à la carte in the previous section (and in §780).

672 Chap 6: Open questions (procedural) The basic insight is that in a phase-based environment the non-cyclic character of postlexical computation (i.e. the fact that there are no phonological footprints of phases above the word level) needs to be expressed by a chunk-specific suspension of the PIC.

799

4.3.5. Phase Impenetrability is a property of the spell-out mechanism, not of concatenative or interpretational systems Let us now consider the fact that computational economy needs to be somehow organised. Some kind of memory-keeper must keep track of which portion of the string was already subject to interpretation, and which portion is new (§307). Everything that we know about how modules work suggests that the modular computation itself is perfectly unable to do this job. All that modular computation does is performing a calculus on an input, and returning an output (modules are input-output devices, their computation is automatic, mandatory, "blind" and so on, see §613). Given this description, modules are perfectly unable to distinguish between those portions of the string submitted that are in need of computation, and those that are not: modules do not make decisions. This means that Phase Impenetrability is not a property of computational systems such as morpho-syntax, phonology (or semantics for that matter). What is it then a property of? The spell-out mechanism appears to be the only candidate. As a consequence, the system which is supposed by phase theory has three, rather than two individuatable units (as far as the derivation of sound is concerned): a concatenative system (morpho-syntax), a spell-out system and an interpretational system (phonology). While the former and the latter are modules, the status of the spellout mechanism is unclear: it reminds of the Translator's Office of Prosodic Phonology (see §377) and Jackendoff's (1997 et passim) interface processors (§§23,652), but does not share the basic property of these devices, i.e. the ability to understand the vocabulary and the structure of both the sending and the receiving modules. Rather, the spell-out mechanism reads the result of morpho-syntax and manages chunk-submission to phonology, which includes the distinction of "old" and "new" portions of the string.180 180

The status of the spell-out system in a modular architecture is a central and understudied question in interface theory. It is further discussed in Vol.2. Unpublished work by Michal Starke inquires on its properties (see also Caha 2009).

Why are there no phonological effects of the cyclic spell-out of words? 673

Table (322) below depicts what this system looks like based on an example where all nodes are phase heads, where (following regular assumptions regarding the phase edge, see §766) only the complement of the head of a phase head is actually sent to interpretation, and where the highest node is a CP, i.e. the end of the syntactic derivation. (322) no look-back managed by the spell-out mechanism morpho-syntax

action of the spellout mechanism

end of the derivation

restores the content of the memory: [Z Y X W]

…

Y γ

Z

phonology

1. adds X to memory 2. reads γ, sends Y β

X

1. stores [W] 2. reads β, sends X W

Y

α X

reads α, sends W W

When spell-out occurs at α, W is sent to phonology, where it is computed and sent back to morpho-syntax. It is then excluded ("forgotten") from further morpho-syntactic computation, but needs to be stored in order to be restituted at the end of the derivation. When spell-out treats β, only X is sent to phonology since, just like for morpho-syntactic computation, previously interpreted strings are excluded from phonological computation. This procedure works through the string until the end of the derivation, where the memory-keeper, that is the spell-out mechanism, restores its memory, which is the linear sequence of all interpreted items. This of course is only a rough schematic picture that does not come any close to what is actually going on. For one thing, the labour that the spell-out mechanism is supposed to do under (322) describes only a vanishingly small portion of its real action. Unpublished work by Michal Starke

674 Chap 6: Open questions (procedural) (see also Caha 2009) studies the properties of the spell-out mechanism in greater detail. Also, it is probably obvious for every phonologist that phonological computation needs to be able to see previously interpreted strings, even if it cannot modify them. Many phonological processes take material from more embedded morphemes into account (§307). Finally, as was mentioned in §302, "old strings are frozen" is too strong a condition in regard of the existence of phonological processes that apply across word boundaries. Therefore §780 (also §823 below) has concluded that the PIC applies to phonological processes à la carte: it must be specified for each process whether its application is marshalled by the PIC or not. From the minimalist perspective, the obligation for the spell-out mechanism (or some other device which is not on my radar) to act as a memory-keeper raises the question whether the trade-off between this extra burden for active memory and the computational economy that is realised by morpho-syntax and LF/PF is all that positive.

800

4.4. Intonation requires cyclic spell-out of words for sure – but this does not appear to concern phonological computation

801

4.4.1. Intonation is governed by syntactic structure Let us now examine intonation, the only case that I am aware of where the cyclic spell-out of words leaves phonological traces – if intonation is any phonological at all. At least since Bresnan (1971), it is established that intonation (also called sentence or phrasal stress) directly depends on syntactic structure. The topic is covered by a rich syntactic literature, including Berman & Szamosi (1972), Cinque (1993), Kahnemuyipour (2009) and Adger (2007). Bresnan's (1971) original data and analysis have been presented in §308; a characteristic example is repeated under (323) below for convenience (underscored words are intonationally prominent). (323)

syntax-sensitive intonation a. Helen left directions for George to follow. b. Helen left directions for George to follow.

(323a) means that Helen has left some directions that George should follow, while (323b) is an invitation for George to follow Helen. Since both sentences are phonologically identical but have contrasting syntactic struc-

Why are there no phonological effects of the cyclic spell-out of words? 675

ture, the different intonation must be a consequence of the latter: under (323a) follow is transitive and belongs to a relative clause whose head is directions, while under (323b) it is intransitive and complements directions. The causal relationship between syntax and phonology has also been thought of in the opposite direction: Szendrői (2001, 2003, 2004) argues that syntactic properties such as focus and climbing can be controlled by intonation. Intonation is thus a phenomenon that challenges Zwicky & Pullum's (1986a,b) original (strong) version of phonology-free syntax (see §§412,660 on this notion, which is probably more accurately referred to as melody-free syntax). In sum, there can be no doubt that cyclic spell-out of words and larger units impacts the calculus of intonation. In order for this fact to qualify as a prove that cyclic spell-out may also bear on phonology, however, it needs to be shown that the calculus of intonation is the result of phonological computation. Against the intuition that intonation is a phonological phenomenon, there may be good reason to believe that this is not the case. Also relevant is the fact that intonation is the typical phonological phenomenon which is argued to be recursive (or rather: to require recursive structure, which as we will see is not quite the same thing). A pervasive and largely undisputed empirical generalisation, however, is that recursion is the privilege morpho-syntax, the only concatenative device(s) in grammar (Hauser et al. 2002, Neeleman & van de Koot 2006, §§45f). This is reason enough to doubt that intonation, should it require recursive structure, is any phonological at all: its putative recursive nature disqualifies intonation as a phonological phenomenon. Whether or not intonation requires recursive structure will not be decided here. In case it does, however, it is shown that the bare existence of recursive structure in phonology does not mean that phonology itself is recursive as long as the structure in question is built by mechanisms that are foreign to phonological computation. We will see that this is indeed the case for all recursive prosodic structure that is argued for in the literature. We begin by considering this question.

802

4.4.2. Prosodic structure may be recursive – phonology is not Beyond the syntactic literature, intonation has been studied in the perspective of Janet Pierrehumbert where it is phonologically represented as tones (e.g. Pierrehumbert 1980, 2000, Beckman & Pierrehumbert 1986, 1988 and Ladd 2000 provide overviews). Also, this approach puts to use the tools of

676 Chap 6: Open questions (procedural) Prosodic Phonology (among many others, Liberman 1975, Ladd 1986, 1996, Gussenhoven 1992, 2004, Kratzer & Selkirk 2007, the articles in van Oostendorp & Horne (eds.) 2005 are specifically about the role of boundaries in intonation). As far as I can see, Ladd (1986) was the first voice to argue that intonation is recursive, in the sense that it requires recursive prosodic structure. Formally speaking, recursion is defined as a structure where a node is dominated by another node of the same kind. Ladd (1986) works with two prosodic constituents, the Major Phrase (MP, cf. Selkirk 1986, §394) and the Tone Group (TG). He aims at showing that intonation cannot be adequately described unless an MP may dominate other MPs, and a TG other TGs. Nested prosodic structure was ruled out by the original version of Selkirk's Strict Layer Hypothesis, and this is what Ladd (1986) stands up against. Under this pressure (among others), Selkirk (1996) abandons the ban on recursive prosodic structure: in the new constraint-based formulation, the non-recursion of prosodic structure is demoted to a violable constraint (§461). Languages and analysts are thus free to use nested prosodic structure. Since then, numerous analyses have taken advantage of this option (among many others, Booij 1996, Peperkamp 1997, Truckenbrodt 1999). 803

4.4.3. Phonology is not recursive: confusion between non-recursive phenomena and their eventual analysis with recursive constructions There are thus analyses that rely on recursive prosodic structure in order to account for intonation. They could be wrong, and the analytic tools that they use – prosodic constituency – could be unwarranted. It was argued in §§694,706 that prosodic constituents are diacritics as much as juncture phonemes and hash marks are, and that diacritics do not qualify for the output of translation. The alternative are carriers of morpho-syntactic information that belong to the phonological vocabulary, i.e. exist in phonological processes that do not appeal to any extra-phonological information. It was argued in §717 that only syllabic space qualifies as an output of translation. If this is correct, of course an alternative means for the analysis of intonation needs to be found that does not rely on prosodic constituency. This track is explored in §806: intonation may not be phonological in the first place, i.e. it may be entirely due to syntactic, rather than to phonological computation.

Why are there no phonological effects of the cyclic spell-out of words? 677

But beyond this debate where right and wrong is a matter of assessment, the whole reasoning according to which a particular type of analysis that is based on recursive structure demonstrates the existence of recursion in phonology is flawed. For this is confusing the existence of a phenomenon, and an analysis thereof. Recursion is a phenomenon of natural language, and as such has a pre-theoretical definition: relevant diagnostics are known. In syntax and morphology, it is defined as follows: you can keep repeating the same type of item indefinitely until grammar-external limits regarding memory etc. are reached. Relevant examples from syntax and morphology are shown under (324) below. (324)

recursion in syntax and morphology a. Peter thinks [that John says [that Amy believes [that…]]] b. Czech iterative -áv dĕlat "to do" dĕl-áv-at "to do repeatedly/often" dĕl-áv-áv-at "to do even more often" dĕl-áv-áv-áv-…-at "to do really really often" French re- prefixation (about the same in English) faire "to do" re-faire "to do again" re-re-faire "to do with two repetitions" re-re-re-faire "to do with three repetitions" re-re-re-re-…-faire "to do with n repetitions"

Recursive structure in natural language has the property of producing grammatically unbounded embedding: grammar happily generates and tolerates an infinite number of embedded clauses (or phrases), and in the case of recursive morphology, an infinite number of embedded morphemes. The limits on recursive structure in actual production are imposed by performance (factors such as memory), not by competence. That is, speakers will get confused upon the third of fourth level of embedding. Also, recursion is obviously a consequence of concatenation: there is no embedding without gluing pieces together. This is reflected in the operation Merge, which is the central (and only) recursion-creating device in current syntax. The simple fact that phonology does not concatenate anything but merely interprets fully concatenated strings shows that there could not be any recursion in phonology, at least not anything that satisfies morpho-syntactic standards.

678 Chap 6: Open questions (procedural) Empirically speaking, nothing that resembles the phenomena under (324) and their pre-theoretical description has ever been reported in phonology. This empirical situation is the reason why the absence of recursion is firmly established as a major property that sets phonology (and semantics) apart from morpho-syntax (e.g. Pinker & Jackendoff 2005a,b). It is entirely independent of eventual analyses that use recursive constructions: having some prosodic constituent such as the Prosodic Word ω and the "Prosodic Word Prime" ω' where the latter dominates the former does not make the phenomenon at hand recursive. The same goes for analyses of other phonological phenomena that use recursive constructions. Hulst (2010) has gathered a number of this kind of analyses regarding for example the internal structure of segments (melodic organisation). He argues that they document the existence of recursion in phonology. They do not: the only thing that they document is the existence of analyses that use recursive structure in order to account for non-recursive phenomena. The existence of recursion in phonology is established by a pre-theoretical and pre-analytical checklist, and nobody has ever found a phonological phenomenon that qualifies. 804

4.4.4. Prosodic structure is not created by phonological computation But let us return to intonation and consider an eventual situation where it turns out that the correct analysis of this phenomenon is based on recursive prosodic structure. Even in this case phonology would not be recursive. What it means for phonology to be recursive (if syntax and morphology define this notion) is to accommodate a recursive mechanism in phonological computation. On standard assumptions, though, prosodic constituency is not created by phonological computation: it is the result of a mapping mechanism that is located in modular no man's land (the Translator's Office, see §381).181 Mapping is an exclusively top-down operation that transforms relevant morpho-syntactic structure into phonologically legible items,182 and eventually introduces distortions that are not motivated by the morpho-syntactic input (non-isomorphism, see §380). 181 182

But recall that in OT (constraint-based) mapping is done in the phonology, see §526. This was shown in §§420f and §§714ff: mapping is uninformed of any phonological property, except of course for the syllable and the foot, which are bottom-up constructions. Prosodic Phonology acknowledges this difference (see §401).

Why are there no phonological effects of the cyclic spell-out of words? 679

We are thus left with a situation where recursive structure may exist in phonology, but with phonological computation being perfectly nonrecursive. 805

4.4.5. Phonology only interprets: Merge must be absent, hence the result of phonological computation is flat The absence of recursion in phonological (and semantic) computation is also central for the generative architecture of grammar. The inverted T model (§86) embodies the insight that there is only one concatenative system in grammar, morpho-syntax.183 Phonology and semantics are interpretational systems (modules) which do not glue any pieces together: they merely interpret a string whose size, composition and linearity is decided elsewhere. The fact that phonology and semantics do not concatenate anything is also a headstone of Hauser et al.'s (2002) argumentation, who hold that recursion (Merge) and the ability to talk to other modules in a specific way (phase) are the only properties of the human faculty of language that are specifically linguistic (§639). If phonology and semantics only interpret strings but are unable to build them, it follows that they must not have access to the building device, that is Merge in current syntactic theory. This is pointed out by Neeleman & van de Koot (2006), who provide detailed argumentation to the end that allowing for Merge (and hence for trees) in phonology and semantics wrongly predicts the existence of recursive structure, and hence of recursive phenomena in these modules.184 The (undisputed) empirical record (there are no recursive phenomena in phonology) as much as the headstone of the generative architecture of grammar (concatenation is the privilege of morpho-syntax) thus require the absence of Merge from phonological (and semantic) computation. Finally, it is worth pointing out that the research programme of Government Phonology in general, and of CVCV in particular, roots in the claim that syllable structure is better understood in terms of lateral relations among segments than by using traditional arboreal structure. This is where 183 184

An eventual sub-division between morphology and syntax notwithstanding. The present discussion is unaffected by this debate (see §537). Note that this is a strong argument against Jackendoff's (1997 et passim) parallel approach to modular structure where all modules are granted access to concatenation (§722).

680 Chap 6: Open questions (procedural) the title of Vol.1 comes from: a lateral theory of phonology. The lateral perspective, whose motivation is purely phonology-internal, thus allows for a non-contradictory implementation of the architectural requirement that phonology has no access to Merge, and that trees must not be the output of phonological computation: phonology-created structure is flat, and the absence of recursion is predicted (this argument was already made in the foreword to Vol.1, see §42).

806

4.4.6. Is intonation a phonological phenomenon at all? Let us now return to the initial question which, recall, was about the absence of phonological effects of the cyclic spell-out of words. We have seen that cyclic spell-out is a factor that influences intonation for sure. Whether through the mediation of the Prosodic Hierarchy or otherwise, intonation seems to militate against the generalisation that cyclic spell-out of words leaves no phonological traces. This, however, is only true if intonation is a phonological phenomenon in the first place. That is, according to our definition of phonology, if intonational structure is created by phonological computation. Cinque's (1993) and Wagner's (2005a) work shows that intonation is much more syntactically bound than the phonological literature may suggest. In actual fact, it raises the question whether intonation is any phonological at all: it may as well be syntax and nothing else. This is the conclusion of Cinque (1993), who derives the parameterisation of the the Nuclear Stress Rule (which is responsible for English intonation in SPE and much subsequent work, see §308) from purely syntactic factors. As a consequence, the Nuclear Stress Rule and phonological treatments thereof (such as Halle & Vergnaud's 1987a:263ff grid-based implementation that Cinque discusses specifically) are redundant and have to go: (325)

"if the effects of the Nuclear Stress Rule […] depend entirely n the direction in which depth of embedding develops, the the rules become redundant. They merely recapitulate what follows from purely syntactic parameters. Hence they should be eliminated." Cinque (1993:244)

This direction may appear counter-intuitive because intonation is somehow "made of sound", and hence should be treated in the phonology. Maybe – but we have seen that this does not mean that phonology builds the structure which ultimately gets to the surface in the guise of intonation. Phonology may well be responsible of nothing at all in the process that

Why are there no phonological effects of the cyclic spell-out of words? 681

decides where intonation falls: like elsewhere, its action is purely interpretative; in the case of intonation, the action of phonology could well reduce to making a particular portion of the string phonologically prominent, whereby the portion itself was designated by extra-phonological devices. A strong argument in favour of this perspective is the following fact: it appears that intonation may be calculated in complete absence of phonological material, that is before lexical (vocabulary) insertion and hence before phonological computation has a chance to do anything at all. Féry & Ishihara (2009) make this generalisation: knowing about the syntactic and the information structure of a sentence is enough in order to predict where intonation falls; the particular words and their phonological properties that will ultimately instantiate this structure are irrelevant. If it is true that intonation may be calculated in complete absence of phonological material, the logical conclusion is that phonological computation does not participate in the entire process: the prominence of this or that item is calculated in syntax alone; phonology (or phonetics) only provides a specific pronunciation for the portion that is made prominent. Finally, another argument for the independence of phonology and intonation is the fact that both appear to live in waterproof worlds: there does not seem to be any conditioning in either way. That is, sentence stress is not influenced by word stress (or vice-versa), intonation does not bear on segmental phenomena, nor is it impacted by labiality, occlusion etc., and so on. It may be concluded, then, that there is a purely syntactic alternative to the traditional phonological treatment of intonation. 807

4.5. Conclusion

808

4.5.1. If the generalisation is correct, Praguian segregation alone will not do At the end of this inquiry, we are left with a number of open questions, and some vague indications. The cyclic character of the spell-out of both morphemes and words is undisputed; why is it, then, that the literature abounds in phonological effects of the former, but does not seem to have met any phonological trace of the latter? One solution is that phonological effects of the cyclic spell-out of words have simply been overlooked: the generalisation is just wrong. Intonation, however, which appears to be a massive counter-example at first sight, turns out not to harm the word-spell-out-mystery (§800). On the con-

682 Chap 6: Open questions (procedural) trary, it rather adds grist to the mill of the mystery since it relies on recursive structure that phonology has no handle on and to which phonological computation does not contribute anything. In case the generalisation turns out to be correct, just jumping into Praguian segregation will not do: the existence of distinct chunk-specific phonologies, one for morpheme-, another for word sequences, is necessary, but not sufficient to explain what is going on. What we need on top of that is a means to make postlexical phonology ignore cyclic structure. For the zero hypothesis is certainly that if an interpretational system is exposed to piecemeal chunk submission, traces thereof will appear in its output. It is therefore certainly empirically adequate to decree that lexical phonology is, but postlexical phonology is not cyclic – this is the reaction of Lexical Phonology (§791). It does not tell us, however, how come and what it means that interpretational systems can "decide" to react on or to ignore piecemeal chunk submission. 809

4.5.2. Chunk-specific PIC à la carte One way to go about this question was explored in §794. "Reacting on cyclic structure" can mean two things for an interpretational system. One perspective is the solution of Lexical Phonology where the cyclic structure that (cyclic) lexical phonology encounters is interpreted by two distinct mini-phonologies according to the morpho-syntactic makeup of the string (level 1 vs. level 2, see §150). By contrast, the cyclic structure that phonology encounters at and above the word level is interpreted by only one computational system (postlexical phonology). Another way of encoding the reaction on piecemeal chunk submission is through the parameterisation of no look-back (i.e. of Phase Impenetrability). On the account of Kaye (1995), the phonological effect of the cyclic spell-out of morphemes is due to modification-inhibiting no lookback: while underapplication is due to morpheme-specific miniphonologies in Lexical Phonology, it is a consequence of Phase Impenetrability in Kaye's system (§279). In this perspective, if the phonological effect of the cyclic spell-out of morphemes is due to Phase Impenetrability, and if this effect is absent when it comes to the cyclic spell-out of words, the conclusion is that the PIC constrains the computation below, but not at and above the word level. This is what §§796f suggest: chunk-specific PIC à la carte. The next question, then, is what Phase Impenetrability is a property of. It appears that the

Chunk-specific phonologies 683

PIC is unlikely to be managed by concatenative/interpretational systems; rather, it is a property of the spell-out mechanism, which in addition to organising chunk-submission also decides to make "old" strings subject to a specific treatment (§799).

810

4.5.3. Outlook About all that was said on the pages above is more or less speculative. This is worth what it's worth, for the whole issue – the word-spell-out-mystery – appears to be unreflected in the literature. One reason is certainly the fact that the absence of cyclic impact on external sandhi is set in stone since Lexical Phonology and the correlated idea of a completely regular ("automatic") postlexical phonology. The study of a larger body of empirical material will have to show whether the current state of affairs is due to this theoretical bias (which, recall, was installed without argument), or whether natural language has really different ways of treating morphemes and words in phonological computation. Which is but another occasion to be thrown back to the question of the (non-)unity of morphology and syntax (§537).

811

5. Chunk-specific phonologies

812

5.1. Specific world-level phonology

813

5.1.1. Everybody has a word-specific phonology This and the following thematic block (§828) work through the question of multiple computational systems in phonology: chunk-specific phonologies are examined first, followed by morpheme-specific phonologies. SPE "officially" held that phonology is one single computational system, but in fact was already operating with a specific word-level phonology that was made distinct by representational means (thereby allowing to maintain the unity of a single set of ordered rules, see §104). This may be considered the (unconceded) birth of chunk-specific phonologies. As far as I can see, all subsequent theories, their disagreement elsewhere notwithstanding, provide for a specific word-level phonology in one way or another. The notion of chunk-specific phonology was introduced in §234, and a comparative tableau in §236 shows under which ban-

684 Chap 6: Open questions (procedural) ner they run in different theories. Phonological interpretation of word-sized chunks by a specific computation is referred to as lexical post-cyclic rules in Lexical Phonology (Rubach & Booij 1984, §194) and as non-cyclic rules in Halle & Vergnaud (1987a, §233). Although Kaye (1995) firmly and explicitly rejects multiple computational systems (phonological computation reduces to the φ-function, which is one is indivisible, see §267), he admits the existence of a specific word-level computation in practice (§§283,338). It is true, however, that I could not identify any formalised version of wordlevel phonology in OT-based interface theories (which typically implement morpheme-specific phonologies, either serially, §483, or in parallal fashion, §477).

814

5.1.2. Since SPE, word-specific phonologies are based on (English) stress A remarkable property of the idea of a word-specific phonology is that it has always been motivated by English stress. The original reasoning of Chomsky & Halle (1968) is exposed in §105 and has remained the typical motivation for word-specific phonologies up to the present day. Probably other phenomena from other languages have also been mobilised, but in any case the reader may be sure to meet English stress when it comes to explain word-specific phonologies, no matter what the theory. English stress requires that a specific computation applies exactly and exclusively to word-sized chunks: the rules in question must not apply to any chunk below the word level, nor to any piece that is bigger than a word. This is because other properties of the word are calculated according to where primary stress sits: vowel reduction (to schwa) applies to unstressed vowels (the detail is more complicated but irrelevant here, some illustration is provided in §§334,339). Since primary stress may be modified by stress-shifting (class 1) morphemes until the last concatenation, vowel reduction cannot apply before the last morpheme has joined in. On the other hand, vowels that have once been reduced to schwa cannot be "unreduced" by later computation of larger units: their reduction is set in stone. Intonation (sentence stress) for example manipulates the relative prominence of words as such, but is perfectly unable to modify the relative prominence of vowels (stress or vowel reduction) inside words. Vowel reduction must therefore apply exactly once in the derivation: at the word level, i.e. after the last morpheme has been concatenated and computed, and before any further computation that involves larger units.

Chunk-specific phonologies 685

The fact that English (primary) stress is a problem child needs to be taken into account, though. Recall from §780 that it has always messed up no look-back generalisations: it violated the SCC-K (i.e. could apply to non-derived environments, see §192), does not observe Phase Impenetrability in Marvin's (2002) analysis (§555) and also violates Kaye's modification-inhibiting no look-back in anti-affix-ordering strings (e.g. govern-mént-al, see §834). Given this criminal record of (English) stress, there may be reason to feel queasy about the fact that a prominent device of phonological interface theory, the existence of a specific computational system for word-sized pieces, historically roots in the analysis of English stress, and since SPE has been carried over to all major theories on the grounds of the same evidence. 815

5.2. Specific sentence-level phonology: Praguian segregation

816

5.2.1. Cyclic phonology of morphemes vs. non-cyclic phonology of words The other chunk-specific division of phonology into distinct computational systems that is argued for in the literature is called Praguian segregation in this book (see §153). It distinguishes between the interpretation of morpheme- and word sequences, which are held to be done by two distinct computational systems, lexical and postlexical phonology. Lexical Phonology has given an architectural status to this insight in placing word phonology (the Lexicon) before syntactic computation, while sentence phonology is only done when syntax has completed the concatenation of words (see §§152f). The idea of two distinct computational systems for the interpretation of morpheme- and word sequences was taken over by modern OT-based incarnations of Lexical Phonology (Stratal OT, DOT, see §483). On the other hand, this division is denied by SPE (§102), Halle & Vergnaud (1987a) (§238) and Kaye (1995) (§§283,301). Another effect of the Lexical Phonology tradition is the non-cyclic anchoring of sentence phonology (§158). While there is no a priori reason to believe that morpheme sequences are, but word sequences are not interpreted according to their cyclic structure, the non-cyclicity of postlexical phonology was set in stone by the idea that prevailed in early Lexical Phonology according to which cyclicity is an exclusive property of the Lexicon. This is what Kiparsky's adaptation of the Strict Cycle Condition (SCC-K) expresses (see §188).

686 Chap 6: Open questions (procedural) This question, together with more detail regarding the historical development of Praguian segregation, was discussed in §§787f. The empirical situation that the literature offers, which is undisputed though unspoken, appears to support the conclusion of Lexical Phonology: there do not seem to be any phonological effects of the cyclic spell-out of words. This is strange in itself, and was called the word-spell-out-mystery in §794. Its empirical validity, namely regarding intonation, was examined in §800. Given the theoretical bias that was introduced by Lexical Phonology (postlexical phonology must be non-cyclic), it therefore remains to be seen whether the absence of phonological effects of cyclic spell-out of words is a fact about this bias or really a fact about language.

817

5.2.2. Arguments for Praguian segregation Why should the computational system that interprets morpheme sequences be different from the computational system that assesses words? Because morpheme- and word-phonologies do not behave in the same way, is the answer of Lexical Phonology. Below this basic argument is examined along the lines of double dissociation, the diagnostic for the identification of distinct computational systems (distinct cognitive modules) that was discussed in §618. Another obvious argument in favour of Praguian segregation is the word-spell-out-mystery: if the generalisation is empirically correct, it is hard to see how a distinction between the interpretation of morphemes and words could be avoided.

818

5.3. Praguian segregation and double dissociation

819

5.3.1. Double dissociation applied to Praguian segregation Double dissociation (§618) is a method that provides evidence for the independence of two systems: if each can exist and work normally in absence of the other, there is reason to believe that they are distinct (rather than emanations of the same entity). Praguian segregation is the claim that the computational system which interprets morpheme sequences is different from the computational system that interprets word sequences. The opposite claim, i.e. that chunks

Chunk-specific phonologies 687

of all sizes are assessed by the same phonology, is made by SPE, Halle & Vergnaud (1987a) and Kaye (1995). Evidence in favour of the existence of two independent computational systems will thus be produced if it can be shown that phonological processes may be restricted to either morpheme- or word sequences, or apply across the board. Table (326) below makes the conditions of double dissociation explicit. (326) there is reason to believe that the interpretation of morpheme- and word sequences is done by two independent systems if there are phonological processes that a. apply to morpheme-, but not to word sequences and b. apply to word-, but not to morpheme sequences and c. apply to both morpheme- and word sequences

The three cases are examined one by one on the pages below. Relevant discussion regarding a chunk-specific PIC à la carte was already done in §241 and §§780,796f,809.

820

5.3.2. Processes that are restricted to morpheme sequences We have already come across (326a): English primary stress placement applies only below the word level: stress may be shifted upon the concatenation of (class 1) morphemes, but once the word level is reached the word is sealed: no further concatenation of the word with other pieces can trigger the reapplication of the rule that is responsible for primary stress placement. This is intuitively obvious: primary stress is also sometimes called word stress, and words have exactly one primary stress. Moving stress out of a word because of post-word concatenation would thus make a word stressless, which is impossible (for "real" words, but this distinction does not matter here). Also note that we are not talking about the question whether primary stress can be moved within the word upon post-word concatenation: it can if the process called stress clash (or the Rhythm rule) indeed moves the primary stress of a word that is too close to the primary stress of another word (thirtéen vs. thírteen mén, see §556). One interpretation of this phenomenon is that word stress is not completely sealed at the word level. Alternatively, stress clash may be interpreted as a redistribution of prominence among vowels that receive a certain degree of word stress (primary,

688 Chap 6: Open questions (procedural) secondary etc.), which however does not modify word stress itself. In any event, stress clash is not an instance of the stress placement rule that reapplies: this process is strictly bound by the limits of the word. Stress clash is a different phonological process that applies to word sequences. What would be awaited if the process of stress placement were active below and above the word level is a continuous move to the right as new words are concatenated, and a result where the string of words bears only one primary stress that appears close to the right margin of the sequence. This of course is not what we observe: the concatenation of fóllow and Jóhn is not treated like the merger of órigin and -al: main stress is not calculated anew, that is vowels that bore stress on earlier cycles are not destressed, and formerly unstressed vowels do not acquire stress. The game is not the same, of course, because fóllow and Jóhn bear already primary stress when they are concatenated; by contrast, órigin is merged with a stressless affix. This is true, but in order to derive the empirical contrast from that, a no look-back device is needed – one that seals word stress at the word level and makes it immune for any further access of the word stress placement rule. The solution in Lexical Phonology where Praguian segregation is applied is to make the process at hand present in lexical, but absent from postlexical phonology. On the other hand, systems that provide for only one computational system for all chunk sizes need to make the word an insuperable barrier for all processes of the (326a) pattern.

821

5.3.3. Processes that apply across the board Many phonological processes apply across word boundaries, i.e. instantiate (326c). Especially the Prosodic Phonology literature documents relevant patterns. English t-flapping is a classical case in point. The literature on t-flapping in General American, both descriptive and analytic, is abundant; it includes Kahn (1976), Kiparsky (1979), Mohanan (1982), Kaisse (1985:25ff) and Rubach (1996a). The data under (327) below are taken from Nespor & Vogel (1986:46f, 224ff) and Balogné-Bérces (2005).

Chunk-specific phonologies 689 (327)

t-flapping in General American185 a. word-internal /t/ [ɾ] city, atom b. word-final /t/ across word boundaries [ɾ] at issue a white owl invite Olivia at eleven just the other night a racoon was spotted in our neighbourhood

According to Nespor & Vogel (1986), flapping is entirely unbounded within the highest constituent of the Prosodic Hierarchy, the phonological utterance. That is, flapping applies in all syntactic environments alike provided the /t/ is word-final and intervocalic. Praguian segregation accounts for across-the-board processes by making the relevant rule part of both the lexical and the postlexical phonology of the language. Note that there is no way to predict whether a given rule will belong to either computational system, or to both. In the alternative perspective where only one computational system assesses chunks of whatever size, the across-the-board pattern is expected: "regular" phonological processes should be of this kind. All restrictions to a specific chunk size depart from the idea, inherent to this conception, that chunk size does not matter for the application of phonological rules (Scheer 2009a,c). 822

5.3.4. Processes that are restricted to word sequences Processes that follow pattern (326b), i.e. apply across word-, but not across morpheme boundaries, are the sentence-level version of derived environment effects (§177): like these, they apply only across a domain boundary, but are inert with the domain in question. And like derived environment effects, processes that follow this pattern are the nightmare case for systems that provide for only one interpretational system: instead of being inside-

185

Within words, flapping is inhibited if the following vowel bears stress: [ɾ] átom vs. *[ɾ] atómic. Also, only word-final /t/ is subject to flapping across word boundaries: word-initial /t/ does not react (*[ɾ] a tissue, which makes a minimal pair with [ɾ] at issue. These aspects of the phenomenon are irrelevant for the discussion.

690 Chap 6: Open questions (procedural) out, this pattern appears to parse the string from "outside in", i.e. in anticyclic fashion. Several instances of the pattern (326b) are reported in the literature, the most prominent being so-called Cracow voicing (to be discussed below). All cases that I could identify, though, concern voicing, and this is highly suspicious: the gordian knot is always a cross-word regressive influence of word-initial sonorants (and sometimes vowels) on preceding wordfinal obstruents (sometimes only on frcatives as in Catalan) whereby underlyingly voiceless obstruents are voiced. The same process, however, is not observed within morphemes or across morpheme boundaries. Cracow (Poznań) voicing is discussed by, among others, Bethin (1984) and Rubach (1996b), Catalan facts are described by Wheeler (1986) and BermúdezOtero (2006), and relevant data from West Flemish are documented by De Schutter & Taeldeman (1986). The fact that no phonological properties other than voicing seem to be involved casts doubt on the reality of the anti-cyclic pattern, and especially the transmission of voicing from sonorants and vowels to voiceless obstruents is suspicious (see e.g. Rice 1993 on active voicing of sonorants and vowels). Therefore the literature typically studies the question whether the effect is phonetic, rather than phonological (Michalski 2009:205ff, Strycharczuk 2010). Let us look at the best known case, Cracow voicing. The discussion below follows Rubach (1996b), who compares two varieties of Polish, Warsaw and Cracow. To a large extent, the two dialects show the same behaviour in regard of voice assimilation: both have final devoicing, and voice assimilation is consistent in obstruent clusters, across morpheme- as much as across word boundaries. What sets both dialects apart is their behaviour when an obstruent-final word is followed by a sonorant- or vowelinitial word. As is shown under (328) below, the Warsaw variety behaves as expected, i.e. devoices word-finally, without the word-final obstruent being subject to any influence from the following word. In Cracow, however, the word-final obstruent systematically voices when the following word begins with a sonorant or a vowel. (328)

Cracow-Poznań voicing a. across word boundaries /d # V/son/ [d # l] samochód Leona [d # o] samochód ojca /t # V/son/ [d # l] brat Leona [d # o] brat ojca

Leon's car father's car Leon's brother father's brother

Chunk-specific phonologies 691 (328)

Cracow-Poznań voicing b. across morpheme boundaries [t - o] brat-owa [t - a] brat-a [t - n] brat-ni c. morpheme-internally [ta] Katarzyna [tr] wiatru

brother's wife brother GENsg brotherly Cathrine wind GENsg

(328b) shows that the regressive assimilation does not go into effect when a voiceless obstruent is followed by a sonorant or a vowel across a morpheme boundary, let alone morpheme-internally as under (328c). In a system that acknowledges Praguian segregation, this pattern is simply accounted for by inscribing the rule at hand in the pool of postlexical, but not in the pool of lexical phonology. On the other hand, there is no way I can see in which theories that provide for only one interpretational system for all chunk sizes could account for this pattern. A no look-back restriction of the PIC-kind will not do: given inside-out interpretation, the process must not apply in earlier computation, but needs to go into effect when larger strings are assessed. No look-back devices are thus out of business, and one would need to introduce something like a "no look-ahead device", which is unheard of elsewhere. It was already mentioned that the problem looks strikingly similar to derived environment effects (for which nobody has a good solution, see §§207,516,837); this suggests that a the best solution, if any, will be one that is applicable to both phenomena, which differ only in the size of the chunks involved (morphemes in derived environment effects, words in (328)).

823

5.4. PIC à la carte

824

5.4.1. Multiple mini-phonologies and chunk-specific phonologies Let us now sum up what we know about distinct computational systems in phonology in general, and chunk-specific mini-phonologies in particular. The fact that a specific computational system for the word level is largely consensual was documented in §813. This means that there is no principled reason to reject the cohabitation of distinct computational systems in phonology, not any more than there is reason to doubt the existence of chunkspecific phonologies: if there is a particular interpretational system for the

692 Chap 6: Open questions (procedural) computation of the chunk size "word", there could also be one for other chunk sizes. The evidence for the computation of morpheme- and word sequences by distinct computational systems (Praguian segregation), however, is inconclusive: the empirical validity of the word-spell-out-mystery needs to be confirmed, and the evidence from double dissociation that was examined in §818 hinges on the interpretation of the Cracow voicing pattern as a real phonological (rather than phonetic) process. An entirely different question are distinct morpheme-specific computational systems, the basic tool of Lexical Phonology (level 1 vs. level 2 phonology). There may be no reason to reject multiple interpretational systems in phonology as such, and chunk-specific mini-phonologies in particular, but this does not mean that any additional mini-phonology is welcome. Morpheme-specific phonologies are discussed in §828 below.

825

5.4.2. Process-specific PIC, rather than "don't undo!" In §302, it was shown that for a variety of reasons Chomsky's "spell-out and forget" is too strong a condition in phonology. Two ways in which this rigid formulation of the PIC could be weakened were considered: processspecific PIC or "don't undo!" where only those properties of a previously interpreted string may not be modified that were created by phonological computation. Under "don't undo!", English stress is not negotiable beyond the word level because it was acquired by previous computation, but nothing prevents t-flapping from going into effect across word boundaries since word-final dentals were not subject to any previous processes. It was already pointed out in §307 that the notion "old" and "new" is not quite the same with the two solutions: under "don't undo!", an new item that needs to be guaranteed against further modification is not just any string that has was previously interpreted: it is only those particular properties of a string that were created by previous computation, i.e. which are absent from the lexicon. This seems to require some diacritic marking on top of the memory-keeping mechanism that is needed anyway for all no look-back devices (§799). This may be reason to believe that process-specific PIC is a less inconvenient way to distinguish the "old" and the "new". Finally, recall from §780 that process-sensitive PIC has also been proposed in syntax (Bošković 2007).

Chunk-specific phonologies 693 826

5.4.3. PIC à la carte: process- and chunk-specific The English facts discussed in §§820f show that in case a single computational system that applies to all chunk sizes is upheld (and hence Praguian segregation dismissed), the PIC must be process-specific: it marshals stress assignment, but not t-flapping. The process-specificity of (external) sandhi seems to be a hard empirical fact (that we have come across on a number of occasions in the book: §§193,241,302,780, Balogné-Bérces 2004, 2005). Process-specific PIC is not enough, though, to account for the English data. In addition, it needs to be somehow specified at which chunk size exactly stress assignment is marshalled by the PIC: at the word level (rather than, say, at every application of Merge). Since the PIC is associated to phases, this just means that the word is a phase in English – not quite spectacular an insight. We have seen in §809, however, that the word-spell-out-mystery, if real, requires that all syntactic phases, i.e. everything above the word level, be exempted from PIC-effects: the PIC is "switched off" for phonological computation beyond the word level (and no matter what process is involved). But even if the word-spell-out-mystery turns out to be spurious, t-flapping must somehow be allowed to ignore the PIC condition that should be associated with syntactic phases: the fact that it goes into effect shows that the vP phase for example does no harm to the process. This means that it is certainly important and useful to know what the phase structure in a language looks like, but that this does not tell us anything about its eventual phonological footprint: phases may or may not enforce Phase Impenetrability, i.e. every phase boundary may or may not be armed with a PIC. The PIC is thus both process- and chunk-specific: the system that enforces the PIC upon phonological computation looks at both parameters before making a decision. This is recapitulated under (329) below. Note that in case a process is declared to be unsubjected to the PIC at a given phase, this does not mean that it actually goes into effect at this phase: recall that in the present chapter we are only talking about procedural means to influence phonology. There are also representational ways of preventing a process from applying: classically, all external sandhi is managed by representational instruments (§787). Vol.2 discusses at length how non-diacritic carriers of morpho-syntactic information (CV units) can inhibit or trigger external sandhi.

694 Chap 6: Open questions (procedural) (329)

process- and chunk-specificity of the PIC for each phase (chunk-specificity), it is specified which processes (process-specificity) are subjected to a PIC condition. Examples CP vP DP word level 2 affixation level 1 affixation

word stress yes yes yes yes yes no

t-flapping no no no no no no

Finally, recall that parameterising the PIC may not appear to be a very minimalist thing to do at first sight: the PIC is an instrument of computational economy, and PIC à la carte amounts to economy à la carte. Computational economy conditions, however, are supposed to be always obeyed. This being said, Chomsky (2004:107f) senses that "spell-out and forget" may be too strong a condition on phonological computation (§306, quote (129)) – something that was observed in §302 anyway. Therefore PIC à la carte does not appear to be an outlandish mechanism in a minimalist environment (see §§780,797).

827

5.5. Conclusion Process-specificity and chunk-specificity seem to be undisputed facts of (external) sandhi phonology. Since the 80s, the standard way of managing both phenomena is Praguian segregation, a major insight of Lexical Phonology (which is carried over into all modern versions thereof, see §489). The preceding sections have evaluated the opportunity to replace the Praguian mechanism that is based on two distinct and chunk-specific computational systems (lexical and postlexical phonology) by a perspective that builds on phase theory and the PIC, and where only one computational system assesses all strings. Following this track, the instrument that encodes process-specificity and chunk-specificity is the parameterisation of Phase Impenetrability. In the system that builds on PIC à la carte, phonological interpretation is completely regular and inhibition-driven: by default it applies across all boundaries of all kinds unless a no look-back condition stands in the

Morpheme-specific mini-phonologies (level 1 - level 2) 695

way. The normal situation is thus regularity, and it takes specific grammatical action – adding a PIC condition – to produce irregularity. This is how regularly invisible structure is made visible to the phonology. Praguian segregation has certainly a number of things going for it: the fact that the word level is a recurrent barrier for phonological processes across languages, the word-spell-out-mystery and double dissociation. The two latter arguments, however, are inconclusive for the time being, and the question is left open in the end.

828

6. Morpheme-specific mini-phonologies (level 1 - level 2)

829

6.1. Cyclicity analysed by morpheme-specific mini-phonologies vs. by no look-back: a major front line in generative interface theory The idea that phonological interpretation could fall into two (or several) distinct computational systems according to the morphological properties of the string was introduced into interface thinking by Lexical Phonology, of which it is a (perhaps the) genetic footprint (see §212). Theories line up for or against morpheme-specific mini-phonologies in the following way: while SPE, Halle & Vergnaud (1987a), Kaye (1995) and Distributed Morphology reject the idea, it stands unchallenged in present-day OT-based phonology (in serial or parallel incarnations, see §§477,483). The alternative is an architecture where strings of whatever morphological composition are interpreted by the same computational system. Morpheme-specific mini-phonologies exist in order to account for affix class-based phenomena, that is cyclicity effects: what they do is to organise underapplication (opacity) (see §150). The alternative way to do the same labour is through selective spell-out (§228,, summary in §763) and a no look-back device (§279). It was shown in §303 that the two instruments – morpheme-specific mini-phonologies and no look-back devices – do the same job and are therefore competing devices: they are mutually exclusive. The line-up of theories and the use of no look-back cuts an aisle through the interface landscape, which it does not take long to interpret: a Chomskian strand that follows SPE and uses no look-back (which is a Chomskian idea, see §287) is opposed to an approach that has left the Chomskian track in the early 80s. A good measure of this contrast is the charge of orthodox generative quarters against Lexical Phonology that was led by Morris Halle in the 80s (Halle & Vergnaud 1987a, see §216).

696 Chap 6: Open questions (procedural) 830

6.2. Intermodular argumentation: if the PIC exists in syntax, it must be active in phonology as well The opposition between the Chomskian and the Lexical Phonology track is not merely a phonology-internal matter: it has also a syntactic branch. That is, the spine of the current minimalist architecture, phase theory, reactivates no look-back, a Chomskian idea (Chomsky 1973, see §287) that was unemployed in the 80s, which has incarnated into Phase Impenetrability today (§304). In other words, Lexical Phonology's morpheme-specific miniphonologies are not only opposed to the PIC-based alternative in phonology – they also have to face the fact that the PIC is a critical ingredient of current syntactic theory: no look-back is the instrument of active memory economy, which is the (extra-linguistic) reason why derivation by phase exists in the first place (§306). The following is thus a relevant question: could a situation be imagined where the PIC is a condition on morpho-syntactic, but not on phonological computation? That is, would it be consistent with minimalist assumptions that the economy of active memory forces syntax into a piecemeal derivation, while phonology does not care for using active memory? There is certainly reason to believe that this is not a workable perspective: active memory is costly, no matter whether it stores syntactic or phonological items; if the minimalist philosophy is on the right track, the phonological system is constrained by active memory economy just as much as the syntactic system. An overall picture where the PIC is a central element of syntactic workings, but entirely absent from phonological mechanisms is not viable. Unsurprisingly enough, this is Chomsky's take: "the computational burden is further reduced if the phonological component too can 'forget' earlier stages of derivation" (Chomsky 2001:12f, see §306 for a more complete version of this quote). Given these premises, the phonology-internal competition for the analysis of cyclicity between morpheme-specific mini-phonologies and the PIC-based system may be refereed: the latter is selected. Which, of course, also rules out a scenario where both solutions co-exist in the same grammar: no theory can afford accommodating two devices that do the same job (§303). The question just how much active memory economy is enforced, or should be enforced by the PIC, needs to be addressed, but is logically independent. Recall from §826 that PIC à la carte, i.e. the double parameterisa-

Empirical coverage 697

tion of the device (process- and chunk-specificity), intermits active memory economy under certain conditions under the pressure of pervasive phonological evidence: PIC à la carte is also computational economy à la carte. Following Structural Analogy (Dependency Phonology, see §705), a weaker version of the same argument can be made in absence of any reference to active memory economy. That is, even in case phonological theory owes nothing to syntactic theory, the following situation is encountered: a tool, no look-back, is present in syntax, and is also one of two competitors for the analysis of cyclicity in phonology. A unifying perspective thus beckons: syntactic and phonological phenomena could be explained by the same tool. Or, in other words, why should a phonology-specific tool be used when its competitor is needed in syntax anyway? And when we know independently that both systems are related by a pipe through derivation by phase?

831

7. Empirical coverage

832

7.1. Introduction: three competitors and the empirical record (affix classbased phenomena) What remains to be done is a comparison of the empirical coverage that the various theories reviewed can offer. In this book, affix class-based phenomena in English are chosen as a uniform and well-studied testing ground in order to compare competing theories of the procedural management of the interface (see §162). A by-product of this inquiry, as it were, is the discussion of derived environment effects, which has actually taken on quite honourable dimensions (§§177,516) and was a regular companion of the reader. Below derived environment effects are therefore included into the overview. The historical review of Part I has identified three relevant models: Lexical Phonology (past and present incarnations), Halle & Vergnaud (1987a) and Kaye (1995). The major front line runs between the former, where cyclic effects (underapplication) are accounted for by morphemespecific mini-phonologies, and the two latter, which do not provide for morpheme-specific computational systems in phonology (§828). Hand in hand with this choice (a necessary concomitance) goes the fact that Halle & Vergnaud's and Kaye's systems are based on selective spell-out, while Lexical Phonology spells out all nodes (§763).

698 Chap 6: Open questions (procedural) The difference between Halle & Vergnaud (1987a) and Kaye (1995), in turn, is that the latter uses a no look-back device (modification-inhibiting no look-back which is the ancestor of Chomsky's modern "freezing" PIC) in order to account for cyclic effects (underapplication), while the former do not. In §833 below these three theories are run against the empirical record, which divides into two major classes. These enjoy a pre-theoretical definition and therefore appear to deserve credit. Recall from §51 (and also §§163,166) that a rule-blocking pattern (level 1 rules in Lexical Phonology) may be distinguished from a rule-triggering pattern (level 2 rules in Lexical Phonology). In the former case, a process that goes into effect in absence of a particular morpho-syntactic division is inhibited, while in the latter case a process which otherwise would not occur is triggered by a specific morphosyntactic environment. Finally, §837 reviews the spectrum of analyses that have been proposed for derived environment effects.

833

7.2. Affix class-based phenomena

834

7.2.1. The rule-blocking pattern The rule-blocking pattern is unproblematic for the three theories: underapplication (to class 2 strings) is achieved by morpheme-specific minigrammars (Lexical Phonology, §150), selective spell-out alone (Halle & Vergnaud 1987a, §228) or in association with the PIC (Kaye 1995, §279). Two specific issues arise: strings that violate Siegel's (1974) affix ordering generalisation (i.e. where a class 2 affix occurs closer to the root than a class 1 affix, e.g. govern-ment2-al1, see §§142,243), and the independent spell-out of terminals (as opposed to nodes). The former cannot be covered by Lexical Phonology and by Kaye (1995), though not for the same reason: the stratal architecture is based on affix ordering and therefore structurally unable to do anti-affix-ordering strings;186 on the other hand, 186

Hence only serial implementations of Lexical Phonology that maintain the original stratal architecture (level ordering §150: first class 1 affixes are concatenated, then class 1 interpretation occurs, then class 2 affixes come in, then class 2 interpretation takes place, and no loop back is allowed) are concerned. OT-based parallel implementations of morpheme-specific mini-phonologies (see §477) have no trouble with anti-affix-ordering strings: the relevant miniphonology is simply "called", whatever the linear (and hierarchical) order of af-

Empirical coverage 699

Kaye's system cannot account for the items in question because they violate modification-inhibiting no look-back (§315). Significantly, though, this diagnostic is only based on one phenomenon, stress assignment, which is a notorious problem child (especially regarding the violation of no look-back devices, see §§556,780,814). Finally anti-affix-ordering strings are unproblematic for Halle & Vergnaud (1987a) (§246). The issue regarding the spell-out of terminals prior to their being merged only concerns Kaye (1995), whose system predicts that this must be possible (§§274,314). While spell-out only concerns nodes on standard assumptions, it was pointed out that the hypothesis of an independent interpretation of terminals is also currently entertained in syntax (see §316).

835

7.2.2. The rule-triggering pattern Bare stratal architecture combined with morpheme-specific miniphonologies cannot cover this pattern. Lexical Phonology therefore needs some additional device. What exactly such a device should look like was much debated in the 80s, but what was proposed always looked like patches, rather than like solutions. Some modern heirs of Lexical Phonology have simply given up: they outsource the pattern into suppletion (Bermúdez-Otero & McMahon 2006). The classical way to do the rule-triggering pattern in Lexical Phonology is due to Mohanan (1982, 1986), who recurs to a representational device (brackets) and a no look-back mechanism, bracket erasure: rules are made sensitive to brackets, which are erased at the end of each stratum (see §166). Given the procedural ambition of Lexical Phonology (§213), the reintroduction of a representational device through the back door was not very popular. Also, it was pointed out that no look-back and morphemespecific mini-phonologies are direct competitors – hence no theory can afford to accommodate both (see §§303,830). As far as I can see, brackets and bracket-sensitive rules have never been eliminated from the analysis of the rule-triggering pattern in Lexical Phonology: this is also true for attempts to find alternative means of encoding cyclicity when it became clear that Kiparsky's original take (according fixes. Bermúdez-Otero (forth a) follows this track. It was pointed out in §484 that the contrast between so-called stratal and parallel versions of morphemespecific mini-grammars is not as clear-cut as the literature may suggest: the former may well turn out to be a version of the latter, which means that the serial issue evaporates.

700 Chap 6: Open questions (procedural) to which cyclicity is a property of the Lexicon as such, see §188) was too strong. Non-cyclic islands in the Lexicon were proposed by Halle & Mohanan (1985) who make cyclicity a property of strata (i.e. strata may or may not obey the SCC-K, see §194) and by Rubach & Booij (1984 et passim) who maintain that all strata are cyclic, but introduce post-cyclic lexical rules, which apply after all strata, but still in the Lexicon (see also §194). All that was done around the SCC-K and the various areas where cyclicity could be enforced has not produced a handle on the rule-triggering pattern, which continues to require reference to brackets and some no lookback device. Which takes us back to the aforementioned incompatibility of the latter with morpheme-specific mini-phonologies. Finally, it was mentioned in §203 that Kiparsky's SCC-K is able to cover the rule-triggering pattern, an option that for some reason does not seem to have been explored. But the abandon of the SCC-K (§197) clears that option anyway. Given this inveterate problem child, modern representatives of the stratal architecture have simply given up on the rule-triggering pattern: Bermúdez-Otero & McMahon (2006:398) argue that it is not the result of any synchronic phonological activity. Rather, damn - damn-ing and damn-ation for example represent three independent lexical entries: /dQm/, /dQmn/ and /dQmneiʃn/. The phenomenon as such is thus outsourced to suppletive management. Soon after having proposed strata-specific cyclicity in Halle & Mohanan (1985), Morris Halle also gives up on the rule-triggering pattern: Halle & Vergnaud's (1987a) system is unable to account for the phenomenon, but unlike Bermúdez-Otero & McMahon (2006), Halle & Vergnaud simply do not talk about the issue: I could not find any mention of the ruletriggering pattern in the relevant literature (§250). Contrasting with the difficulty that Lexical Phonology and Halle & Vergnaud (1987a) experience, the rule-triggering pattern is unproblematic for Kaye's (1995) system (see §321). Logically, the PIC must be held responsible since its action is what sets Kaye's approach apart from the others. Finally, it is worth mentioning that Kaye's analysis of the ruletriggering pattern makes a prediction to the end that all processes of this kind must be subjected to a phonological condition on top of being sensitive to affix classes (see §325). In the case of English nasal cluster simplification, this is the fact of applying only in morpheme-final position (sign-ing vs. i[g]nore).

Empirical coverage 701 836

7.2.3. Is the rule-triggering pattern less real? It was already mentioned that the distinction between the rule-blocking and the rule-triggering pattern is pre-theoretical and exhausts the logical possibilities of how morpho-syntactic divisions can bear on phonological processes (see §51); it therefore deserves credit. The stratal architecture of classical Lexical Phonology echoes this division by making rule-blocking processes level 1 rules, against rule-triggering processes, which appear as level 2 rules. This is a remarkable result since the stratal makeup was not designed in order to reproduce the empirical contrast. There is thus reason to believe that it is empirically meaningful. Now the discussion has shown that theories are not well suited for the analysis of the rule-triggering pattern and therefore tend to make it less real, either relegating it to suppletion (Stratal OT), or sweeping it under the rug (Halle & Vergnaud 1987a). It was just pointed out, though, that there is little reason to believe that rule-triggering processes are less natural or less real than rule-blocking processes. Both represent logical possibilities of procedural intervention in phonology. It is therefore reasonable to acknowledge that Kaye's PIC-based system fares better empirically speaking, at least as far as English affix classbased phenomena are concerned. It uses the same technology for both relevant patterns (modification-inhibiting no look-back) and makes a number of predictions that have an impact in and beyond phonology (§§316,325).

837

7.3. Derived environment effects

838

7.3.1. A separate issue Let us now look at derived environment effects. Before examining specific analyses, it is important to recall that this phenomenon and affix classbased phenomena are two separate issues. Lexical Phonology in general and Paul Kiparsky in particular have put a lot of effort into showing that both phenomena can be reduced to a single mechanism, the SCC-K. All attempts in this direction failed, and it took quite some time to understand that this unifying ambition is on the wrong track (see §§197,785). It was shown in §182 that sensitivity to affix classes and derived environments is logically independent: one may occur without the other. This means that a theory may be successful in one area, but not in the other, without this having any significance beyond this very fact.

702 Chap 6: Open questions (procedural) 839

7.3.2. A blooming landscape of analyses Let us now consider individual lines of attack, starting with voices that argue for placing derived environment effects outside of grammar altogether. This idea originates in Anderson (1981) (§204); it is entertained in various brands by Kiparsky (1993), who tries to make the relevant distinction lexical (§196), and by Burzio (2000a,b), who attempts at making derived environment effects a case of analogy (§519, as so many other things).187 Solutions that propose a grammar-internal mechanism were discussed in relation with Lexical Phonology (§177, with a summary in §207), with Halle & Vergnaud's (1987a) (§237) and Kaye's (1995) system (§§284,300), as well as with OT (§§509,516). One group may be identified by the strategy to differentiate between "old" and "new" morphemes and/or computation. In the 80s (and in Lexical Phonology), this is the case of Mohanan's brackets (§201) and Kiparsky's SCC-K (§188); in the OT environment, van Oostendorp's (2006a) Coloured Containment (§§509,518), McCarthy's (2003a) Comparative Markedness (§519), and Cho's (2009) revival of Kiparsky's Elsewhere Condition (§521) hook up with this tradition. Another way to go about derived environment effects was opened by the idea of parallel (instead of serially ordered) morpheme-specific miniphonologies (see §477). Where others have to make acrobatic moves and add extra machinery, the parallel system just needs to set up a specific phonology for the root. This comes with no conceptual expense since it simply extends the use of the basic tool of the theory, morpheme-specific minigrammars. Yu (2000) has explored this track along the lines of cophonologies (§522). Finally, a third group of analyses may be criticized either for general reasons or from the vantage point of modularity. Łubowicz (2002) and Cho (2009) are hardly viable because of the general properties of their analyses. The former offers only a solution for phonologically derived environments that is not applicable to morphologically derived environments (the latter being of course the primary target in interface discussion). This was pointed out by van Oostendorp (2006a, 2007) (see §518). On the other hand, Cho (2009) simply writes a constraint which restates the observation that roots are faithful to their lexical makeup (FAITH-LEX, see §521). If we write a 187

This supposes that analogy – which is called Output-Output Faithfulness in OT – is considered a mechanism that lies beyond the limits of grammar proper.

Empirical coverage 703

specific constraint for every empirical phenomenon, we will no doubt successfully derive all existing patterns (as well as many others) – but the result will be closer to a sophisticated description in prose than to what a grammar looks like. Van Oostendorp's Coloured Containment (§§509,518) and Anttila's interface constraints (Root Faithfulness §520) violate modularity by virtue of making direct reference to morphological categories (direct syntax). Finally, Kula (2008) offers a representational solution for a number of well-known cases: the idea is to derive the different behaviour of derived and underived items from distinct lexical representations. Whether this approach may be generalised to all derived environment effects remains to be seen. 840

7.3.3. Anti-cyclic phenomena: derived environment effects have a big brother Derived environment effects describe a situation where computation applies to a morphologically complex chunk, but not to the pieces that it is made of. To the extent that these pieces are interpretational units (phase heads, which may or may not be the case), all theories that do not provide for morpheme-specific mini-phonologies are structurally unable to cover the pattern. This is because the smaller piece must be interpreted (because it is a phase head), but at the same time must not (because it is underived). These conflicting requests can only be satisfied by distinct interpretational systems. Now recall the interesting parallel at the word level that was described in §822: there are processes that apply across word boundaries, but do not go into effect across morpheme boundaries (or within morphemes). So-called Cracow voicing is the best-known representative of this class of processes (which all seem to be based on voicing). Their parallel with derived environment effects is obvious: in a three-chunk hierarchy made of the context 1) "within morphemes", 2) "across morpheme boundaries" and 3) "across word boundaries", Cracow voicing-type processes apply to 3), but not to 2) and 1), while derived environment effects apply to 2), but not to 1). As far as I can see, this parallel has gone unnoticed in the literature. One way to go about derived environment effects, then, would be to evaluate analyses with respect to their ability to also cover the big brother, i.e. sentence-level "outside-in" (or anti-cyclic) phenomena.

841

Conclusion Intermodular argumentation

842

1. Trying to get a handle on the interface

843

1.1. Looking at the interface through the prism of its history At the end of this survey, a summary is not really what is needed: the entire Part II is devised for this function, and the chapters and sections are meant to work out focal points of interest in interface design. It does not really make sense to go over them one by one, but it is certainly an interesting fact to be pointed out that the headlines of Part II are not exactly what a reader of the interface literature comes across: procedural vs. representational management of the interface, inside-out interpretation, selective spell-out, local vs. non-local representational intervention in phonology, privativity, the word-spell-out mystery, to mention just a few. There are two explanations for this discrepancy that I can think of. On the one hand, it is a result of the historical study: looking at an object over time does not produce the same image as a punctual examination. If this is the case, it was worth going through the pains of Part I and II (as well as through the Interlude). On the other hand, interface theories and phenomena have not been interpreted without direction: as was mentioned in the introduction, the book is conceived so as to look at the interface through two prisms, the difference between representational and procedural means of talking to phonology, and modularity. This may also make a difference.

844

1.2. Two more prisms used: Interface Dualism and modularity The procedural-representational prism mimics the filter that Stephen Anderson has applied to the history of phonology in the 20th century (St. Anderson 1985). Modularity is an extra-linguistic referee: following one of the deepest layers of generative thinking, I take for granted that the mind is organised along a modular architecture; and that language is modular in this sense. Modularity imposes a number of restrictions on how modular computation can look like, and on intermodular communication. Conclusions

706 Conclusion: Intermodular Argumentation have repeatedly been drawn on these grounds: interface theories have either proved compatible with modular requirements (e.g. Prosodic Phonology, see §410), or were found to offend modularity in various ways and for a number of reasons (see the summary in §702). Although modularity has grounded generative thinking since its earliest incarnations (see §§603,623 on Chomsky's participation in the computational perspective of post-war science), although a good deal of Fodor's (1983) formalisation of the modular idea was based on language and the book itself a result of a class co-taught with Chomsky, and even though modularity is part and parcel of every general introduction to generative linguistics, it is quite striking to see how small the footprint of modularity is in generative interface thinking. This is one thing that I have learned while working through the literature. There is reason to believe that modularity is a good compass for identifying the position of interface theories on the linguistic chess board. Those who do active interface design should be aware of what it takes to respect modular requirements, as well as of the consequences that that their violation has for the positioning of the theory in the landscape of cognitive science.

845

1.3. Three theory-external ways to get a handle on interface theory One goal of the book is thus certainly to make a historiographic contribution to structuralist and generative interface theory; that is, to make historical generalisations, to identify traps that theories may want to avoid, and to work out the basic ideas that circulate, often enough in various disguises that are not understood as such, and on the backdrop of the parasitic noise that theories and data may generate. One of the strands that runs through all of structuralist and generative interface thinking is actually modularity itself. Relevant evidence is bundled in §692: structuralists practised Level Independence, which conceives of morphology and phonology as two ontologically distinct worlds and enforces translation; Prosodic Phonology set up Indirect Reference and a Translator's Office, but visibly ignored that this architecture materialises the modular idea (§414); finally, Lexical Phonology introduced interactionism, that is modern derivation by phase, obviously without being aware of the fact that this architectural design is the only way to reconcile inside-out interpretation with modular requirements (§680).

Intermodular argumentation 707

The book is thus designed to study the history of interface thinking not just for the historiographic sake, but also in order to get a handle on what is right and wrong, how we can avoid to reinvent the wheel with every new theory, which mistakes have already been made and do not need to be repeated, which ideas seep through when theories come and go, and how we can apply what history has taught us to the present-day landscape. Vol.2 ought to be a result of this track. Therefore the book tries to get a handle on present-day interface theories; it attempts to gain insight into the properties that a correct interface theory must not have, and into those that it must have. One way is studying the history of the field. Another way is modularity, i.e. the recognition that language and its interfaces are constrained by the general properties of the cognitive system. A third way, finally, is one that has shined through on various occasions of the book: a pattern that I call intermodular argumentation. Bundling the different intermodular arguments that have been made is, I think, a good conclusion of Part II.

846

2. Intermodular argumentation

847

2.1. Intermodular potential of interactionist derivation by phase

848

2.1.1. Each end of the interactionist pipe may impact the other

The idea of intermodular argumentation is to exploit the potential of the interactionist architecture that multiple spell-out and derivation by phase (Epstein et al. 1998:46ff, Uriagereka 1999, Chomsky 2000a et passim, see §304) have introduced. The shipping back and forth of pieces between (morpho-)syntax and the PF/LF interfaces during the derivation of a sentence establishes a pipe between the concatenative and the interpretational devices that did not exist in GB or earlier versions of the inverted T-architecture. It creates a situation where syntactic theories and analyses may have direct consequences on the phonological side, and vice versa. Intermodular argumentation is interesting because it provides stronger evidence than what modular-internal reasoning can produce: it offers the maximal degree of independent assessment that linguists can expect without leaving their discipline (and arguments based on modularity go even beyond this limit).

708 Conclusion: Intermodular Argumentation 849

2.1.2. Morpho-syntax can referee competing phonological analyses and theories The structure of intermodular arguments is always the same; it may concern a particular analysis of a specific data set as much as theoretical devices. Only the latter is further discussed below: given that the same shipping device (the spell-out mechanism) relates morpho-syntax and phonology (or PF, as well as LF), its properties must be the same on both sides. That is, a situation where some device is critical on one side, but unheard of on the other, cannot be a correct description of the interface. In practice, what will be done below is only argumentation in one direction: current syntactic theory is used as a referee for competing phonological theories. The application of intermodular argumentation to particular analyses was only briefly illustrated in §320 on the occasion of the discussion of English nasal assimilation, to which the prefix-final nasal of un- is not subjected (while in- does assimilate). The argument is that the phonological facts observed must be a consequence of the different derivational history of un- and in- if – as is indeed the case – these show distinct behaviour in morpho-syntax. That is, a phonology-internal solution that makes no prediction as to what happened in the derivational history of morpho-syntax (as is proposed for example by Rubach & Booij 1984:11f and Vogel 1991), and which in fact can live with any morpho-syntactic analysis and its reverse, does not qualify. In this example, two competing phonological analyses (one representational, the other procedural) are refereed by the existence of a morph-syntactic contrast between the pieces involved.

850

2.1.3. Intermodular arguments can only be made through the procedural channel: translation is arbitrary It was also mentioned in §320 that intermodular argumentation is a privilege of the procedural channel: representational communication goes through translation, and the relation between the input and the output of translation is necessarily arbitrary (more on this in Vol.2). The conclusion that Newell & Scheer (2007) draw is that given competing and empirically equivalent representational and procedural analyses of the same phenomenon, choose the latter since it may make falsifiable predictions on the other end of the interactionist pipe (procedural first).

Intermodular argumentation 709 851

2.2. Conditions and limitations of intermodular argumentation Another question that is addressed below (§858) are the conditions that must be met for intermodular argumentation to be conclusive. We will see that – unfortunately – much of the intermodular refereeing potential hinges on a question that is subject to ongoing debate: whether morphology is just the lower part of syntax, or whether it is a computational system in its own right (see §539). Depending on this question is the number of spell-out mechanisms that exist in grammar which, as we will see, is critical for the comparison of phonological and syntactic effects of cyclic spell-out. If it turns out that morphology and syntax are two distinct computational systems, it could indeed be argued with some right that each system comes with its own spell-out mechanism. On the other hand, in case morphology and syntax are found to be emanations of one and the same computational system, there is of course no room for two distinct spell-out mechanisms. The reason why the number of spell-out mechanisms matters is the word-spell-out-mystery, which was discussed in §786: what we actually do when we compare phonological and syntactic effects of cyclic spell-out is to compare the spell-out of morphemes with the spell-out of words. That is, a phonological footprint seems to be produced by the cyclic spell-out of morphemes, but never of words. In order to be able to draw conclusions from the comparison of phonological and syntactic effects of cyclic spellout, we must first make sure that it is the same spell-out that we are looking at.

852

2.3. Six properties of spell-out that can be made intermodular arguments

853

2.3.1. Convergence of syntactic and phonological tools On a number of occasions in Part I and Part II, we have come across phonological devices that were found to be strikingly similar or identical to tools that are used in current syntactic theory. These parallels are significant because of their complete independence: they are made on entirely different sets of data, which have not even a remote resemblance in kind (phonology and syntax), and they are typically made by people who do not know that the same mechanism is used on the other side of the mirror. The devices at hand which converge, however, concern an object that is shared, in fact the pipe by which syntax and phonology are related: the spell-out mechanism.

710 Conclusion: Intermodular Argumentation This is too much convergence to suffer an interpretation where similarities are taken to be accidental. Table (330) below lists relevant items and locates them in the previous discussion. (330)

convergence between syntactic and phonological devices a. "freezing" no look-back: the PIC b. selective spell-out c. the phase edge: spell out your sister! d. interactionism e. piece-driven phase (vs. node-driven phase) f. spell-out of terminals (interpretation prior to merge)

§§287,783,828 §763 §766 §672 §§767,781f §316

Interestingly, it seems that all devices mentioned have been invented in phonology. All of them (save piece-driven phase) are part and parcel of syntactic theory today, but the phonological legacy is hardly mentioned in the literature.

854

2.3.2. Strong and weak version of the intermodular argument This convergence is remarkable as such and may certainly motivate attempts to unify the devices at hand: why should phonology have its private mechanism if an alternative exists that uses the same technology as syntactic theory? This reasoning follows Structural Analogy (Dependency Phonology, see §705), and it is the weak version of the intermodular argument that was made in §830 where morpheme-specific mini-phonologies and the PIC were compared as competitors for the analysis of cyclic effects; the argument was in favour of the latter because the PIC is needed in syntax anyway. There is also a strong version of the intermodular argument (§830): the premises are the same, but this time it is considered that convergence is not just an enjoyable thing to have; rather, the interactionist interface architecture is inconsistent if the spell-out mechanism needs to have some property on the syntactic side, which is either unheard of on the phonological side, or even incompatible with phonological theory (or the other way round). Therefore if we are sure that a spell-out-relevant device (such as the PIC) is critical and undisputed in syntax, it must also exist in phonology.

Intermodular argumentation 711 855

2.3.3. Import of phonology-based mechanisms into syntax In the opposite direction, there are two phonology-based mechanisms under (330) that are candidates for introducing phonological evidence into the syntactic debate: piece-driven phase and the spell-out of terminals. Some voices in the syntactic literature follow similar ideas: Phase Extension (§781) and counter-cyclic merger (late adjunction, see §316). These tracks, especially the possible interpretation of current syntactic node-driven phase in terms of phonological piece-driven phase (by means of a phasehood feature, §782), seem to be worth exploring in further study.

856

2.3.4. Four syntactic referees for phonological theories and the special status of interactionism Four out of the six convergent items under (330) are prime candidates for intermodular argumentation where syntax referees competing phonological theories (the phonological origin of the devices notwithstanding). There is no doubt that the PIC, selective spell-out, the phase edge and interactionism are absolutely critical ingredients of current syntactic phase theory. Following the strong version of the argument, they must therefore be active on the phonological side as well. The case of interactionism is peculiar insofar as being interactionist or not is not really discriminating phonological theories today (see §676). The strong anti-interactionist reaction of Halle & Vergnaud (1987a) (§222) against the introduction of the interactionist architecture by Lexical Phonology was in defence of generative orthodoxy, which at the time was still based on the idea that all concatenation is done before all interpretation (§305). Today, however, generative orthodoxy has changed allegiance: current syntactic theory is based on derivation by phase and hence has joined the interactionist party. Therefore, if interactionism was a strongly discriminating front line in phonology in the 80s, it is not anymore today. Only orthodox versions of OT that extend the anti-derivationalist ambition to the entire grammatical architecture (instead of restricting it to phonological computation) will reject interactionism (§678). This resistance is intimately related to the misty relationship that OT has with modularity: interactionism and the shipping of pieces can only exist when morpho-syntax and phonology are two distinct entities, something that OT is quite far from acknowledging or practising (see §523).

712 Conclusion: Intermodular Argumentation Interactionism is thus ill-suited for discriminating competing phonological theories, but certainly a knockdown argument that should make the anti-cyclicity fraction of OT reconsider their position, and OT as such its attitude regarding modularity.

857

2.3.5. Phonology must provide for a "freezing" PIC, selective spell-out and the phase edge Let us now look at the three remaining syntactic referees for phonological theories. If selective spell-out, the PIC and the phase edge are necessary properties of phase theory, phonological theories of the effects of cyclic spell-out must also have them.188 Recall from §763 that selective spell-out divides phonological theories in two camps, one where all nodes are spelled out (Lexical Phonology), another where spell-out is selective (Halle & Vergnaud 1987a, Kaye 1995). The former can thus be dismissed on intermodular grounds, while the latter qualifies. The PIC further filters phonological theories: Lexical Phonology and Halle & Vergnaud (1987a) do not use any no look-back device (for the analysis of affix class-based phenomena189), while modification-inhibiting no look-back is the mechanism that accounts for cyclic effects in Kaye's system (see §828). Syntactic theory thus selects Kaye (1995), which is the only theory to pass both filters. Finally, current phase theory requires that in case XP is a phase head, the spell-out of XP only triggers the interpretation of the complement; the head and its specifier – the edge of the phase – are spelled out only at the next higher phase (see §766). This requirement may be run against the record from competing phonological theories of cyclicity-induced effects: Lexical Phonology spells out all nodes, Halle & Vergnaud (1987a) spellout only the node that dominates interpretation-triggering affixes, while Kaye (1995) sends only the sister of interpretation-triggering affixes to interpretation (see §763). In an intermodular perspective, then, if the spell-

188

189

The intermodular argument regarding selective spell-out, the PIC and the phase edge are also made in three articles that are drawn from the book: Scheer (2008c, 2009b, 2010b). Of course, (classical versions of) Lexical Phonology feature(s) Kiparsky's (1982a,b) Strict Cycle Condition (SCC-K), which is a no look-back device. The SCC-K, however, is devised for derived environment effects – it plays no role in affix class-based phenomena (§237).

Intermodular argumentation 713

out mechanism spells out the complement of phase heads – their sister –, the latter is selected, while the two former are dismissed.

858

2.4. How many spell-out mechanisms are there in grammar?

859

2.4.1. The parallel is always between the spell-out of morphemes and words The headstone of the argumentation made in the previous section – which is tacitly assumed – is that the syntactic and the phonological effects of cyclic spell-out that we are talking about are the two ends of the same pipe. That is, they are produced by the same spell-out mechanism. Were they not, there would be no reason for "phonological" spell-out to mimic the properties of "syntactic" spell-out: the latter could, say, spell out the sister of the phase head, while the former spells out something else; the latter could also be selective and implement the PIC, while the former could ignore these devices. The existence of just one spell-out mechanism in the realm of the inverted T architecture is intuitive: a grammatical architecture that accommodates more than one parcel-shipping mechanism is not the first option that springs to one's mind. It is however an alternative that requires serious attention. Reconsider for example the phase edge in syntax. Saying that spelling out the sister of interpretation-triggering affixes is the phonological version of the phase edge is imprecise: it is certainly the result of phonological evidence, but it concerns morphology. That is, where Chomsky's mechanism spells out words and larger chunks, the phonological evidence is based on the spell-out of morphemes. And surprisingly enough (or not, depending on where one stands), this turns out to be the situation of all convergent devices: the phonological evidence for all items under (330) is based on the spell-out of morphemes, never on the spell-out of words. What the parallel really is about, then, is syntax and morphology: relevant chunks in both areas, morphemes and words, seem to be spelled out by the same mechanism – spell out your sister!

714 Conclusion: Intermodular Argumentation 860

2.4.2. The weak version of the argument can be made, but the strong version hinges on the unity of morphology and syntax This result may be considered a contribution to the ongoing debate whether morphology and syntax are instances of the same computational system or not (see §539). Leaving it at that is making the aforementioned weak version of the intermodular argument: the convergence of syntactic and phonological devices is probably not accidental and invites to think of a unified perspective. Whether or not the strong version of the intermodular argument can be made, however, hinges precisely on the debate whether morphology and syntax are one: if it turns out that morphology and syntax are two distinct computational systems, it could be argued with some reason that each system comes with its own spell-out mechanism.190 In this case, the strong intermodular argument cannot be made since the phonological evidence for, say, the phase edge concerns the spell-out of morphemes, whereas the syntactic evidence for the same device is based on the spell-out of words and larger pieces. If on the other hand it is found that morphology is just the lower part of syntactic structure, there can be only one spell-out mechanism, and the strong intermodular argument goes through. This objection to intermodular argumentation has already been made in §769 regarding selective spell-out: from the perspective of Lexical Phonology where morphology and syntax are two distinct computational systems, spell-out could well be selective in syntax, but not in morphology.

861

2.4.3. The word-spell-out-mystery strikes again We have thus reached a point of indecision: the prima facie arguments only hold if we can be sure that there is only one single spell-out mechanism, and this depends on the debate whether morphology and syntax are one. In this situation, an obvious thing to do is to try to circumvent this "technical" difficulty by simply looking at phonological effects of the cyclic spell-out of words (rather than of morphemes). But alas, as we have seen there are none: this is what the word-spell-out mystery is about (see

190

Note, however, that this is not a necessary conclusion: two distinct structures may as well be harvested by the same mechanism.

Intermodular argumentation 715

§786). All roads thus lead to the question whether morphology and syntax are one.

862

2.5. Conclusion Intermodular argumentation comes in a strong and a weak version. The latter follows Structural Analogy (§705) and can always been made since it simply points out that there is no good reason to favour a mechanism when a competing solution is based on a device that is needed in syntax (or in phonology) anyway. The strong version of the argument is eliminative in the sense that it disqualifies theories on one end of the interactionist pipe in presence of an alternative that is compatible with (or selected by) a requirement that is established on the other end. In order for the strong version to go through, however, it needs to be made sure that the spell-out mechanism which sends word sequences to PF is really the same as the shipping mechanism which submits morpheme sequences to phonological computation. The null hypothesis is certainly that all shipping business in the inverted T architecture is done by the same mechanism. But if it turns out that morphological and syntactic structure are distinct computational systems, a scenario may (but does not have to) be construed where two different structures come each with its own spellout mechanism. This means in turn that there is no guarantee for the properties of these mechanisms to be the same, for example regarding the PIC (present/absent), selective spell-out (yes/no) and the spell-out of the head's sister (yes/spell out something else). Concluding on the question whether morphology and syntax are instances of the same computational system is certainly appropriate for a book on the interface: it is one of the big questions in interface architecture; it will not be solved tomorrow, but it is a good thing to see that its prospect directly impacts another area, phonology. This intermodular conditioning is an effect of derivation by phase, which has opened new ways of evaluating interface theories as much as module-specific theories.

863

References References followed by the mention WEB are available at www.unice.fr/dsl/tobias.htm. Abels, Klaus 2003

Successive cyclicity, anti-locality, and adposition stranding. Ph.D dissertation, University of Connecticut. Abry, Christian, Muriel Stefanuto, Anne Vilain, and Rafael Laboissière 2002 What Can the Utterance 'Tan, Tan' of Broca's Patient Leborgne Tell Us about the Hypothesis of an Emergent 'Babble-Syllable' Downloaded by SMA? In Phonetics, Phonology and Cognition, Jacques Durand and Bernard Laks (eds.), 226-243. Oxford: Oxford University Press. Ackema, Peter, and Ad Neeleman 2004 Beyond Morphology. Interface Conditions on Word Formation. Oxford: Oxford University Press. Ackema, Peter, and Ad Neeleman 2007 Morphology ≠ Syntax. In The Oxford Handbook of Linguistic Interfaces, Gillian Ramchand and Charles Reiss (eds.), 325-352. Oxford: OUP. Adger, David 2007 Stress and Phasal Syntax. Linguistic Analysis 33: 238-266. Aelbrecht, Lobke 2008 Licensing ellipsis as Agree. Paper presented at the 27th West Coast Conference on Formal Linguistics, UCLA 16-18 May. Alderete, John 2001 Root-controlled accent in Cupeño. Natural Language & Linguistic Theory 19: 455-502. Alexiadou, Artemis, Elena Anagnostopoulou, Gianina Iordachioaia, and Mihaela Marchisi forth In support of Long Distance Agree. In Local modeling of nonlocal dependencies, Artemis Alexiadou, Gereon Müller and T. Kiss (eds.). Tübingen: Niemeyer. Alexiadou, Artemis, Elena Anagnostopoulou, and Florian Schäfer 2009 PP licensing in nominalizations. Proceedings of NELS 38: 39-52. Alexiadou, Artemis, and Jane Grimshaw forth Verbs, nouns and affixation. Allen, Margaret 1978 Morphological Investigations. Ph.D dissertation, University of Connecticut.

718 References Allen, Margaret 1980 Semantic and Phonological Consequences of Boundaries: A Morphological Analysis of Compounds. In Juncture, Mark Aronoff and Mary-Louise. Kean (eds.), 9-27. Saratoga: Anma Libri. Anderson, James 1965 The demarcative function. Lingua 13: 185-188. Anderson, John 1985 Structural analogy and Dependency Phonology. Acta Linguistica Hafniensia 19: 5-44. Anderson, John 1986 Structural analogy and case grammar. Lingua 70: 79-129. Anderson, John 1987 Structural Analogy and Dependency Phonology. In Explorations in Dependency Phonology, John Anderson and Jacques Durand (eds.), 15-48. Dordrecht: Foris. Anderson, John 1992 Linguistic representation. Structural analogy and stratification. Berlin & New York: Mouton de Gruyter. Anderson, John, and Colin Ewen 1987 Principles of Dependency Phonology. Cambridge: Cambridge University Press. Anderson, John, and Charles Jones 1974 Three theses concerning phonological representations. Journal of Linguistics 10: 1-26. Anderson, Stephen 1974 On the typology of phonological rules. In Papers from the parasession on Natural Phonology, A. Bruck, R. Fox and M. La Galy (eds.), 1-12. Chicago: Chicago Linguistic Society. Anderson, Stephen 1981 Why phonology Isn't "Natural". Linguistic Inquiry 12: 493-539. Anderson, Stephen 1982 Where's Morphology ? Linguistic Inquiry 13: 571-612. Anderson, Stephen 1985 Phonology in the Twentieth Century. Chicago: University of Chicago Press. Anderson, Stephen 1992 A-Morphous Morphology. Cambridge: Cambridge University Press. Anttila, Arto 2002 Morphologically conditioned phonological alternations. Natural Language and Linguistic Theory 20: 1-42.

References 719 Anttila, Arto 2009

Arad, M. 2003

Derived environment effects in colloquial Helsinki Finnish. In The Nature of the Word: Studies in Honor of Paul Kiparsky, Kristin Hanson and Sharon Inkelas (eds.), 433-460. Cambridge, MA.: MIT Press.

Locality Constraints on the Interpretation of Roots: The Case of Hebrew Denominal Verbs. Natural Language & Linguistic Theory 21: 737-778. Archangeli, Diana 1988 Aspects of underspecification theory. Phonology 5: 183-208. Aronoff, Mark 1976 Word Formation in Generative Grammar. Cambridge, Mass.: MIT Press. Aronoff, Mark 1980 The Treatment of Juncture in American Linguistics. In Juncture, Mark Aronoff and Mary-Louise Kean (eds.), 29-36. Saratoga: Anma Libri. Aronoff, Mark 2003 Juncture and Boundary. In International Encyclopedia of Linguistics, William Frawley (ed.), 43. Oxford: OUP. Aronoff, Mark, and S. Sridhar 1983 Morphological levels in English and Kannada, or Atarizing Reagan. In Chicago Linguistics Society 19, Papers from the Parasession on the interplay of Phonology, Morphology and Syntax, J. Richardson, M. Marks and A. Chukerman (eds.), 3-16. Chicago: Chicago Linguistics Society. Aronoff, Mark, and S. Sridhar 1987 Morphological levels in English and Kannada. In Rules and the Lexicon, Edmund Gussmann (ed.), 9-22. Lublin: Katolicki Universytet Lubelski. Balogné-Bérces, Katalin 2004 Connected speech phenomena in Strict CV phonology. The Even Yearbook 6: 1-10. Balogné-Bérces, Katalin 2005 Strict CV Phonology and the English Cross-Word Puzzle. Ph.D dissertation, Eötvös Loránd University, Budapest. Baltin, M. 2007 Deletion versus Pro-forms: a false dichotomy? Ms., New York University. Banich, M., and M. Mack (eds.) 2003 Mind, Brain and Language. Hillsdale: Erlbaum.

720 References Bao, Zhiming 1996

Local Tree Geometry and the Phonology-Syntax Interface. In Interfaces in Phonology, Ursula Kleinhenz (ed.), 27-45. Berlin: Akademie Verlag. Barkow, Jerome, Leda Cosmides, and John Tooby 1992 The Adapted Mind. Evolutionary Psychology and the Generation of Culture. Oxford: OUP. Barragan, L., and Heather Newell 2003 Cupeño morphology is(n't) inherently stressful. In Proceedings of the 32nd WECOL, B. Agbayani, V. Samiian and B.V. Tucker (eds.), 10-22. Fresno: Dept. of Linguistics, California State University. Basbøll, Hans 1975 Grammatical boundaries in phonology. Aripuc 9: 109-135. Basbøll, Hans 1978a Schwa, jonctures et syllabification dans les représentations phonologiques du français. Acta Linguistica Hafniensia 16: 147-182. Basbøll, Hans 1978b Boundaries and the ranking of rules in French phonology. In Etudes de phonologie française, Benoît de Cornulier and François Dell (eds.), 3-18. Marseille: Editions du CNRS. Basbøll, Hans 1981 On the function of boundaries in phonological rules. In Phonology in the 1980's, Didier Goyvaerts (ed.), 245-269. Ghent: StoryScientia. Basbøll, Hans 1986 Stød-sandhi. In Sandhi Phenomena in the Languages of Europe, Henning Andersen (ed.), 75-83. Berlin: Mouton de Gruyter. Bastian, Charlton 1880 The brain as an organ of mind. London: Kegan Paul. Bates, Elizabeth 1994 Modularity, domain specificity and the development of language. Discussions in Neuroscience 10: 136-149. Beckman, Mary, and Janet Pierrehumbert 1986 Intonational Structure in Japanese and English. Phonology Yearbook 3: 15-70. Beckman, Mary, and Janet Pierrehumbert 1988 Japanese Tone Structure. Cambridge, MA: MIT Press. Bendjaballah, Sabrina 2001 The negative preterite in Kabyle Berber. Folia Linguistica 34: 185-223.

References 721 Bendjaballah, Sabrina, and Martin Haiden 2002 The Strong Template in German. Paper presented at 4e Rencontres Internationales du GDR 1954 "Phonologies", Grenoble 68 June. Bendjaballah, Sabrina, and Martin Haiden 2003a Templatic Architecture. Recherches Linguistiques de Vincennes 32: 157-168. Bendjaballah, Sabrina, and Martin Haiden 2003b Meaningful vowels. In Proceedings of the XXVIII Incontro di Grammatica Generativa, Piero Bottari (ed.). Lecce: Congedo Editore. Bendjaballah, Sabrina, and Martin Haiden 2003c Templatic Inflection in German. Penn Working Papers in Linguistics 8: 29-41. Bendjaballah, Sabrina, and Martin Haiden 2005 Berber prepositions, templates and the PF interface. Paper presented at GLOW 28, Geneva 31 March - 2 April. Bendjaballah, Sabrina, and Martin Haiden 2007 A Typology of Emptiness in Templates. In Sounds of Silence: Empty Elements in Syntax and Phonology, Jutta Hartmann, Vera Hegedus and Henk van Riemsdijk (eds.), 21-57. Amsterdam: Elsevier North Holland. Bendjaballah, Sabrina, and Martin Haiden forth Berber Prepositions: Phonology and Syntax of Templates. Ms, University of Lille. Benua, Laura 1995 Identity effects in morphological truncation. University of Massachusetts Occasional Papers in Linguistics 18: 77-136. Benua, Laura 1997 Transderivational identity: phonological relations between words. Ph.D dissertation, University of Massachusetts at Amherst. Berman, A., and M. Szamosi 1972 Observations on sentential stress. Language 48: 304-325. Bermúdez-Otero, Ricardo 1999 Constraint interaction in language change: quantity in English and German. Ph.D dissertation, University of Manchester. Bermúdez-Otero, Ricardo 2003 The acquisition of phonological opacity. In Variation within Optimality Theory: Proceedings of the Stockholm Workshop on Variation within Optimality Theory, J. Spenader, J. Eriksson and A. Dahl (eds.), 25-36. Stockholm: Department of Linguistics, Stockholm University [longer version at ROA #593].

722 References Bermúdez-Otero, Ricardo 2006 Phonological domains and opacity effects: a new look at voicing and continuancy in Catalan. In the Workshop on Approaches to phonological opacity at GLOW. Barcelona. Bermúdez-Otero, Ricardo 2008 Evidence for Chung's generalization. Paper presented at 16th Manchester Phonology Meeting, Manchester 22-24 May. Bermúdez-Otero, Ricardo forth a Stratal Optimality Theory. Oxford: Oxford University Press. Bermúdez-Otero, Ricardo forth b Cyclicity. In The Blackwell companion to phonology, Marc van Oostendorp, Colin Ewen, Elizabeth Hume and Keren Rice (eds.). Malden, MA: Wiley-Blackwell. Bermúdez-Otero, Ricardo, and April McMahon 2006 English Phonology and Morphology. In The Handbook of English linguistics, Bas Aarts and April McMahon (eds.), 382-410. Oxford: Blackwell. Bertinetto, Pier Marco 1999 Boundary strength and linguistic ecology (Mostly exemplified on intervocalic /s/-voicing in Italian). Folia Linguistica 33: 267-286. Bethin, Christina 1984 Voicing assimilation in Polish. International Journal of Slavic Linguistics and Poetics 29: 17-32. Bickmore, Lee 1990 Branching nodes and prosodic categories. In The PhonologySyntax Connection, Sharon Inkelas and Draga Zec (eds.), 1-17. Chicago: University of Chicago Press. Blaho, Sylvia, Patrick Bye, and Martin Krämer (eds.) 2007 Freedom of analysis? Berlin: Mouton de Gruyter. Bloomfield, Leonard 1939 Menomini morphophonemics. Travaux du Cercle Linguistique de Prague 8: 105-115. Bobaljik, Jonathan 2002 A-Chains at the PF-Interface: Copies and ‘Covert’ Movement. Natural Language & Linguistic Theory 20: 197-267. Bobaljik, Jonathan, and Susi Wurmbrand 2005 The domain of agreement. Natural Language & Linguistic Theory 23: 809-865. Boeckx, Cedric 2010 Language in Cognition. Uncovering mental structures and the rules behind them. Oxford: Wiley-Blackwell. Boeckx, Cedric, and Kleanthes Grohmann 2007 Putting Phases in Perspective. Syntax 10: 204-222.

References 723 Boersma, Paul 1998 Functional phonology: formalizing the interactions between articulatory and perceptual drives. The Hague: Holland Academic Graphics. Boersma, Paul 2005 Some listener-oriented accounts of hache aspire in French. Ms, University of Amsterdam, ROA #730. Bolinger, Dwight, and Louis Gerstman 1957 Disjuncture as a cue to constructs. Word 13: 246-255. Booij, Geert 1977 Dutch Morphology. Lisse: de Ridder. Booij, Geert 1981 Rule ordering, rule application and the organization of grammars. In Phonologica 1980, Wolfgang Dressler, Oskar Pfeiffer and John Rennison (eds.), 45-56. Innsbruck: Institut für Sprachwissenschaft. Booij, Geert 1983 Principles and parameters in Prosodic Phonology. Linguistics 21: 249-280. Booij, Geert 1985a The Interaction of Phonology and Morphology in Prosodic Phonology. In Phono-morphology. Studies in the Interaction of Phonology and Morphology, Edmund Gussmann (ed.), 23-34. Lublin: Katolicki Universytet Lubelski. Booij, Geert 1985b Coordination reduction in complex words: a case for Prosodic Phonology. In Advances in nonlinear phonology, Harry van der Hulst and Norval Smith (eds.), 143-160. Dordrecht: Foris. Booij, Geert 1986 Two cases of external sandhi in French: enchaînement and liaison. In Sandhi Phenomena in the Languages of Europe, Henning Andersen (ed.), 93-103. Berlin: Mouton de Gruyter. Booij, Geert 1987 Lexical phonology and the organization of the morphological component. In Rules and the lexicon, Edmund Gussmann (ed.), 43-65. Lublin: Katolicki Universytet Lubelski. Booij, Geert 1988 On the relation between lexical and prosodic phonology. In Certamen Phonologicum, Pier Marco Bertinetto and Michele Loporcaro (eds.), 63-75. Torino: Rosenberg & Sellier. Booij, Geert 1992 Lexical Phonology and Prosodic Phonology. In Phonologica 1988, Wolfgang Dressler, Hans Luschützky, Oskar Pfeiffer and John Rennison (eds.), 49-62. Cambridge: CUP.

724 References Booij, Geert 1995 Booij, Geert 1996 Booij, Geert 1997

The Phonology of Dutch. Oxford: Clarendon Press. Cliticization as prosodic integration: The case of Dutch. The Linguistic Review 13: 219-242. Non-derivational phonology meets Lexical Phonology. In Derivations and Constraints in Phonology, Iggy Roca (ed.), 261288. Oxford: Oxford University Press.

Booij, Geert 2000 [1996] The phonology-morphology interface. In The First Glot International State-of-the-Article Book, Lisa Cheng and Rint Sybesma (eds.), 287-305. Berlin: Mouton de Gruyter. Booij, Geert, L. Ducceschi, Bernard Fradin, E. Guevara, Angela Ralli, and Sergio Scalise (eds.) 2007 Proceedings of the Fifth Mediterranean Morphology Meeting. Bologna: Università degli Studi di Bologna. Booij, Geert, and Rochelle Lieber 1993 On the simultaneity of morphological and prosodic structure. In Studies in Lexical Phonology, Sharon Hargus and Ellen Kaisse (eds.), 23-44. San Diego: Academic Press. Borowsky, Toni 1986 Topics in the lexical phonology of English. Ph.D dissertation, University of Massachusetts, Amherst. Borowsky, Toni 2000 Word-faithfulness and the direction of assimilation. The Linguistic Review 17: 1-28. Bošković, Željko 2001 On the nature of the syntax-phonology interface. Cliticization and related phenomena. Amsterdam: Elsevier. Bošković, Željko 2005a On null elements in syntax. Paper presented at Sounds of Silence, Tilburg 19-22 October. Bošković, Željko 2005b On the locality of left branch extraction & the structure of NP. Studia Linguistica 59: 1-45. Bošković, Željko 2007 Agree, Phases, and Intervention Effects. Linguistic Analysis 33: 54-96. Bresnan, Joan 1971 Sentence stress and syntactic transformations. Language 47: 257281.

References 725 Broca, Paul 1861

Remarques sur le siège de la faculté du langage articulé, suivies d'une observation d'aphémie (perte de la parole). Bulletin de la Société d'Anatomie de Paris 6: 330-357. Bromberger, Sylvain, and Morris Halle 1989 Why Phonology Is Different. Linguistic Inquiry 20: 51-70. Broselow, Ellen 2003 Marginal phonology: phonotactics at the edge. The Linguistic Review 20: 159-193. Brun-Trigaud, Guylaine, and Tobias Scheer 2010 Lenition in branching onsets in French and in ALF dialects. In Development of Language through the Lens of Formal Linguistics, Petr Karlík (ed.), 15-28. Munich: Lincom. Burzio, Luigi 2000a Cycles, non-derived environment blocking and correspondence. In Optimality Theory: Phonology, Syntax, and Acquisition, Joost Dekkers, Frank van der Leeuw and Jeroen van de Weijer (eds.), 47-87. Oxford: Oxford University Press. Burzio, Luigi 2000b Segmental contrast meets output-to-output faithfulness. The Linguistic Review 17: 368-384. Burzio, Luigi 2001 Zero Derivations. Linguistic Inquiry 32: 658-677. Burzio, Luigi 2007 Phonetically conditioned syncretism. In Selected proceedings of the 5th Décembrettes: Morphology in Toulouse, Fabio Montermini, Gilles Boyé and Nabil Hathout (eds.), 1-19. Somerville, MA: Cascadilla. Bybee, Joan 2001 Phonology and Language Use. Cambridge: Cambridge University Press. Caha, Pavel 2009 The Nanosyntax of Case. Ph.D dissertation, Universtiy of Tromsø. Caha, Pavel, and Tobias Scheer 2008 The Syntax and Phonology of Czech Templatic Morphology. In Annual Workshop on Formal Approaches to Slavic Linguistics. The Stony Brook Meeting 2007, Andrei Antoneko, John Bailyn and Christina Bethin (eds.), 68-83. Ann Arbor: Michigan Slavic Publications. WEB.

726 References Cairns, Charles, and Charles Reiss 2009 Architecture and Representations in Phonology. In Contemporary Views on Architecture and Representations in Phonology, Eric Raimy and Charles Cairns (eds.), 1-16. Cambridge, Mass., London: Bradford, MIT Press. Carvalho, Joaquim Brandão de 2002 De la syllabation en termes de contours CV. Habilitation Thesis, Ecole des Hautes Etudes en Sciences Sociales, Paris. WEB. Carvalho, Joaquim Brandão de 2006 Markedness gradient in the Portuguese verb: how morphology and phonology interact. In Prosody and Syntax, Ivan Fónagy, Yuji Kawaguchi and Tsunekazu Moriguchi (eds.), 157-174. Amsterdam: Benjamins. Cattell, Ray 2006 An Introduction to Mind, Consciousness and Language. London: Continuum. Charette, Monik 1989 The Minimality Condition in Phonology. Journal of Linguistics 25: 159-189. 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. WEB. Charette, Monik, and Asli Göksel 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. WEB. Chen, Matthew 1985 The syntax of phonology: Xiamen tone sandhi. Ms., University of California at San Diego. Chen, Matthew 1990 What must phonology know about syntax? In The PhonologySyntax Connection, Sharon Inkelas and Draga Zec (eds.), 19-46. Chicago: University of Chicago Press. Cheng, Lisa, and Laura Downing 2007 The prosody and syntax of Zulu relative clauses. SOAS Working Papers in Linguistics and Phonetics 15: 51-63. Chien, Yu-Chin, and Kenneth Wexler 1990 Children's Knowledge of Locality Conditions in Binding as Evidence for the Modularity of Syntax and Pragmatics. Language Acquisition 1: 225-295.

References 727 Cho, Young-Mee Yu 1990 Syntax and Phrasing in Korean. In The Phonology-Syntax Connection, Sharon Inkelas and Draga Zec (eds.), 47-62. Chicago: University of Chicago Press. Cho, Young-Mee Yu 2009 A historical perspective on nonderived environment blocking: the case of Korean palatalization. In The Nature of the Word: Studies in Honor of Paul Kiparsky, Kristin Hanson and Sharon Inkelas (eds.), 461-486. Cambridge, MA.: MIT Press. Chomsky, Noam 1955-56 The Logical Structure of Linguistic Theory. New York: Published 1975, Plenum. Chomsky, Noam 1959 Review of Skinner's Verbal Behavior. Language 35: 26-58. Chomsky, Noam 1965 Aspects of the Theory of Syntax. Cambridge, Mass.: MIT Press. Chomsky, Noam 1965 [2006] The formal Nature of Language. In Language and Mind, Noam Chomsky (ed.), 102-142. Cambridge: CUP. Chomsky, Noam 1970 Remarks on nominalisation. In Readings in English Transformational Grammar, R. Jacobs and P. Rosenbaum (eds.), 184-221. Walthamm, Mass.: Ginn. Chomsky, Noam 1972 Language and Mind. New York: Jovanovich. Chomsky, Noam 1973 Conditions on Transformations. In A Festschrift for Morris Halle, Stephen Anderson and Paul Kiparsky (eds.), 232-286. New York: Holt, Rinehart & Winston. Chomsky, Noam 1975 Reflections on Language. New York: Pantheon. Chomsky, Noam 1980 Rules and Representations. New York: Columbia University Press. Chomsky, Noam 1981 Lectures on Government and Binding. 7th edition 1993, Dordrecht: Foris. Chomsky, Noam 1982 Some concepts and consequences of the theory of Government and Binding. Cambridge, Mass.: MIT Press. Chomsky, Noam 1984 Modular Approaches to the Study of Mind. San Diego: San Diego University Press.

728 References Chomsky, Noam 1986 Barriers. Cambridge, Mass.: MIT Press. Chomsky, Noam 1988 Problems of Knowledge. Cambridge, Mass.: MIT Press. Chomsky, Noam 1993a A minimalist program for linguistic theory. In The view from Building 20. Essays in Linguistics in Honor of Sylvain Bromberger, Kenneth Hale and Samuel Keyser (eds.), 1-52. Cambridge, Mass.: MIT Press. Chomsky, Noam 1993b On the Nature, Use, and Acquisition of Language. In Readings in Philosophy and Cognitive Science, Alvin Goldman (ed.), 511-534. Cambridge, Mass.: MIT Press. Chomsky, Noam 1995a The Minimalist Program. Cambridge, MA: MIT Press. Chomsky, Noam 1995b Language and Nature. Mind 104: 1-61. Chomsky, Noam 2000a 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. Chomsky, Noam 2000b New Horizons in the Study of Language and Mind. Cambridge: Cambridge University Press. Chomsky, Noam 2001 Derivation by Phase. In Ken Hale: A Life in Language, Michael Kenstowicz (ed.), 1-52. Cambridge, Mass.: MIT Press. Chomsky, Noam 2002 On Nature and Language. Cambridge: Cambridge University Press. Chomsky, Noam 2004 Beyond explanatory adequacy. In Structures and Beyond. The cartography of syntactic structures, Volume 3, Adriana Belletti (ed.), 104-131. Oxford: Oxford University Press. Chomsky, Noam 2005 Three factors in language design. Linguistic Inquiry 36: 1-22. Chomsky, Noam 2008 On Phases. In Foundational Issues in Linguistic Theory, Robert Freidin, Carlos Otero and Maria-Luisa Zubizaretta (eds.), 133166. Cambridge, MA.: MIT Press. Chomsky, Noam, and Morris Halle 1968 The Sound Pattern of English. Cambridge, Mass.: MIT Press.

References 729 Chomsky, Noam, Morris Halle, and Fred Lukoff 1956 On Accent and Juncture in English. In For Roman Jakobson. Essays on the occasion of his sixtieth birthday, Morris Halle, Horace Lunt, Hugh McLean and Cornelis van Schooneveld (eds.), 65-80. The Hague: Mouton. Chomsky, Noam, and George Miller 1963 Introduction to the formal analysis of natural languages. In Handbook of Mathematical Psychology, Vol.2, P. Luce, R. Bush and E. Galanter (eds.), 269-322. New York: Wiley. Chung, Sandra 1983 Transderivational constraints in Chamorro phonology. Language 59: 35-66. Churchland, Patricia 1993 Can Neurobiology teach us anything about consciousness? Proceedings and Addresses of the American Philosophical Association 67.3: 23-40. Cinque, Guglielmo 1993 A null theory of phrasal and compound stress. Linguistic Inquiry 24: 239-297. Circle, Prague Linguistic 1931 Projet de terminologie phonologique standardisée. Travaux du Cercle Linguistique de Prague 4: 309-323. Clapin, Hugh 2002 Content and Cognitive Science. Language & Communication 22: 231-242. Clayton, Mary 1981 Word boundaries and sandhi rules in natural generative phonology. Language 57: 571-590. Clements, George N. 1978 Tone and syntax in Ewe. In Elements of stress, tone and intonation, Donna Jo Napoli (ed.), 21-99. Washington, D.C.: Georgetown University Press. Clements, George N. 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. Cobb, Margaret 1996 Clitics in Basque: a phonological approach. SOAS Working Papers in Linguistics and Phonetics 6: 26-39. Cobb, Margaret 1997 Conditions on Nuclear Expressions in Phonology. Ph.D dissertation, SOAS, London.

730 References Cole, Jennifer 1995

The Cylcle in Phonology. In The Handbook of Phonological Theory, John Goldsmith (ed.), 70-113. Oxford: Blackwell. Cole, Jennifer, and John Coleman 1992 No need for cyclicity within generative grammar. In CLS 28: Parasession of the Cycle in Linguistic Theory, J. Denton, G. Chan and C. Canakis (eds.), 36-50. Chicago: Chicago Linguistics Society. Coleman, John 1995 Declarative lexical phonology. In Frontiers in Phonology. Atoms, Structures, Derivations, Jacques Durand and Francis Katamba (eds.), 333-382. London & New York: Longman. Coleman, John 2005 Phonological Representations. Cambridge: Cambridge University Press. Condoravdi, Cleo 1990 Sandhi rules of Greek and Prosodic Phonology. In The Phonology-Syntax Connection, Sharon Inkelas and Draga Zec (eds.), 63-84. Chicago: University of Chicago Press. Cosmides, Leda, and John Tooby 1992a Cognitive adaptations for social exchange. In The adapted mind. Evolutionary psychology and the generation of culture, J. Barkow, Leda Cosmides and John Tooby (eds.), 163-228. Oxford: OUP. Cosmides, Leda, and John Tooby 1992b The Psychological Foundations of Culture. In The adapted mind. Evolutionary psychology and the generation of culture, J. Barkow, Leda Cosmides and John Tooby (eds.), 19-136. Oxford: OUP. Cosmides, Leda, and John Tooby 1994 Origins of domain specificity: the evolution of functional organization. In Mapping the Mind, L.A. Hirschfeld and S.A. Gelman (eds.), 85-116. Cambridge: CUP. Cowper, Elizabeth, and Keren Rice 1987 Are phonosyntactic rules necessary? Phonology 4: 185-194. Culicover, P., and Kenneth Wexler 1973 Three further applications of the freezing principle in English. Social Sciences Working Papers of the University of California at Irvine 48. Curtiss, S. 1981 Dissociations between language and cognition: cases and implications. Journal of Autism and Developmental Disorders 11: 15-30.

References 731 Cutler, Anne 1996

Prosody and the Word Boundary Problem. In Signal to Syntax: Bootstrapping from Speech to Grammar in Early Acquisition, James Morgan and Katherine Demuth (eds.), 87-99. Mahwah, NJ: Erlbaum. Cyran, Eugeniusz, and Edmund Gussmann 1998 Polish consonantal sequences: a phonological testing ground. In Structure and Interpretation, Eugeniusz Cyran (ed.), 127-138. Lublin: Pase. WEB. Cyran, Eugeniusz, and Edmund Gussmann 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. De Schutter, G., and J. Taeldeman 1986 Assimilatie van Stem in de Zuidelijke Nederlandse Dialekten. In Vruchten van z’n akker: opstellen van (oud-) medewerkers en oud-studenten voor Prof. V.F. Vanacker, M. Devos and J. Taeldeman (eds.), 91-133. Ghent: Seminarie voor Nederlandse Taalkunde. Dehaene, Stanislas 1997 The Number Sense. How the Mind Creates Mathematics. Oxford: Oxford University Press. Dehaene, Stanislas, Ghislaine Dehaene-Lambertz, and Laurent Cohen 2001 On Language, Biology and Reductionism. In Language, Brain and Cognitive Development. Essays in Honor of Jacques Mehler, Emmanuel Dupoux (ed.), 397-401. Cambridge, Mass.: MIT Press. Dell, François 1976 Schwa précédé d'un groupe obstruante-liquide. Recherches Linguistiques de Vincennes 4: 75-111. Dell, François 1986 L'accentuation dans les phrases en français. In Les représentations en phonologie, François Dell, Daniel Hirst and Jean-Roger Vergnaud (eds.), 65-122. Paris: Hermann. Démonet, Jean-François, Guillaume Thierry, and Jean-Luc Nespoulos 2002 Towards Imaging the neural Correlates of Language Functions. In Phonetics, Phonology and Cognition, Jacques Durand and Bernard Laks (eds.), 244-253. Oxford: Oxford University Press. den Dikken, Marcel 2007 Phase Extension: Contours of a theory of the role of head movement in phrasal extraction. Theoretical Linguistics 33: 1-41.

732 References 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. Devine, Andrew, and Laurence Stephens 1980 On the Phonological Definition of Boundaries. In Juncture, Mark Aronoff and Mary-Louise Kean (eds.), 57-78. Saratoga: Anma Libri. Devitt, Michael, and Kim Sterelny 1989 Linguistics: what's wrong with the 'right view'. In Philosophical Perspectives, Vol.3, J. Tomberlin (ed.), 497-531. Atascadero: Ridgeview. Dinsmore, John (ed.) 1992 The symbolic and connectionist paradigms: closing the gap. Hillsdale: Erlbaum. Dobashi, Yoshihito 2003 Phonological Phrasing and syntactic derivation. Ph.D thesis, Cornell University. 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 1978 The Syllable in Phonological and Prosodic Structure. In Syllables and Segments, Alan Bell and Joan Bybee Hooper (eds.), 25-43. Amsterdam: North Holland. 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. Downing, Laura 2006 Canonical Forms in Prosodic Morphology. Oxford: Oxford University Press. Dresher, Elan 1994 The prosodic basis of the Tiberian Hebrew system of accents. Language 70: 1-52. Dresher, Elan 1996 Introduction to Metrical and Prosodic Phonology. In Signal to Syntax: Bootstrapping from Speech to Grammar in Early Acquisition, James Morgan and Katherine Demuth (eds.), 41-54. Mahwah, NJ: Erlbaum.

References 733 Dressler, 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. Dressler, Wolfgang 1984 Explaining Natural Phonology. Phonology Yearbook 1: 29-51. Dupoux, Emmanuel (ed.) 2001 Language, Brain and Cognitive Development. Essays in Honor of Jacques Mehler. Cambridge, Mass.: MIT Press. Durand, Jacques 1995 Universalism in phonology: atoms, structures and derivations. In Frontiers in Phonology. Atoms, Structures, Derivations, Jacques Durand and Francis Katamba (eds.), 267-288. London & New York: Longman. Durand, Jacques 2006 La phonologie générative jusqu'en 1975. In History of the Language Sciences, vol.3 (HSK 18.3), Sylvain Auroux, E.F.K. Koerner, Hans-Josef Niederehe and Kees Versteegh (eds.), 22652270. Berlin: de Gruyter. Durand, Jacques, and Bernard Laks (eds.) 2002 Phonetics, Phonology and Cognition. Oxford: Oxford University Press. Dziubalska-Kołaczyk, Katarzyna 2001a Phonotactic constraints are preferences. In Constraints and Preferences, Katarzyna Dziubalska-Kołaczyk (ed.), 69-100. Berlin, New York: Mouton de Gruyter. Dziubalska-Kołaczyk, Katarzyna 2001b On the phonotactics of consonantal clusters. In Naturally! Linguistic studies in honour of Wolfgang Ulrich Dressler presented on the occasion of his 60th birthday, M. SchanerWolles, John Rennison and Friedrich Neubarth (eds.), 111-119. Torino: Rosenberg & Sellier. Dziubalska-Kołaczyk, Katarzyna 2002 Beats-and-binding phonology. Frankfurt am Main: Lang. Elman, Jeffrey 1994 Learning and development in neural networks: the importance of starting small. Cognition 48: 71-99. Elordieta, Gorka 1997 Morphosyntactic Feature Chains and Phonological Domains. Ph.D dissertation, University of Southern California. Elordieta, Gorka 2007 Segmental Phonology and Syntactic Structure. In The Oxford Handbook of Linguistic Interfaces, Gillian Ramchand and Charles Reiss (eds.), 125-177. Oxford: OUP.

734 References Elordieta, Gorka 2008 An overview of theories of the syntax-phonology interface. Journal of Basque Linguistics and Philology 42: 209-286. Embick, David 2003 Locality, Listedness, and Morphological Information. Studia Linguistica 57: 143-169. Embick, David 2007 Linearization and local dislocation: derivational mechanics and interactions. Linguistic Analysis 33: 303-336. Embick, David forth Localism versus Globalism in Morphology and Phonology. Ms., University of Pennsylvania. Embick, David, and Alec Marantz 2008 Architecture and Blocking. Linguistic Inquiry 39: 1-53. Embick, David, and Rolf Noyer 2001 Movement Operations after Syntax. Linguistic Inquiry 32: 555595. Embick, David, and Rolf Noyer 2007 Distributed Morphology and the Syntax - Morphology Interface. In The Oxford Handbook of Linguistic Interfaces, Gillian Ramchand and Charles Reiss (eds.), 289-324. Oxford: OUP. Embick, David, and David Poeppel 2005 Mapping syntax using imaging: prospects and problems for the study of neurolinguistic computation. In Encyclopedia of language and linguistics (2nd edition), Keith Brown (ed.). Oxford: Elsevier. Encrevé, Pierre 1988 La liaison avec et sans enchaînement: phonologie tridimensionnelle et usages du français. Paris: Seuil. Encrevé, Pierre 2000 The Old and the New: Some Remarks on Phonology and its History. Folia Linguistica 34: 57-85. This article is a translation of Encrevé, Pierre 1997. L'ancien et le nouveau. Quelques remarques sur la phonologie et son histoire. Langages 125: 100Epstein, Samuel 123. 1999 Un-principled syntax: the derivation of syntactic relations. In Working Minimalism, Samuel Epstein and Norbert Hornstein (eds.), 317-345. Cambridge, Mass.: MIT Press. Epstein, Samuel 2006 On I(nternalist)-Functional Explanation in Minimalism. Linguistic Analysis 33: 20-53. Epstein, Samuel, E.M. Groat, R. Kawashima, and H. Kitahara 1998 A Derivational Approach to Syntactic Relations. Oxford: OUP.

References 735 Epstein, Samuel, and T. Daniel Seely 2002 Rule Applications as Cycles in Level-Free Syntax. In Derivation and Explanation in the Minimalist Program, Samuel Epstein and T.Daniel Seely (eds.), 65-89. Oxford: Blackwell. Epstein, Samuel, and T. Daniel Seely 2006 Derivation in Minimalism. Cambridge: CUP. Fabb, Nigel 1988 English suffixation is constrained only be selectional restrictions. Natural Language and Linguistic Theory 6: 527-539. Farmer, Ann 1984 Modularity in Syntax: A Study of Japanese and English. Cambridge, Mass.: MIT Press. Feldman, J. A. 1985 Connectionist models and their applications: introduction. Cognitive Science 9: 1-2. Felser, Claudia 2004 Wh-copying, phases, and successive cyclicity. Lingua 114: 543574. Féry, Caroline, and Shinichiro Ishihara 2009 How Focus and Givenness Shapes Prosody. In Information Structure from Different Perspectives, Malte Zimmermann and Caroline Féry (eds.), 36-63. Oxford: OUP. Fitch, Tecumseh, Marc Hauser, and Noam Chomsky 2005 The evolution of the language faculty: Clarifications and implications. Cognition 97: 179-210. Fitzpatrick-Cole, Jennifer 1996 Reduplication meets the phonological phrase in Bengali. The Linguistic Review 13: 305-356. Fodor, Jerry 1975 The Language of Thought. New York: Crowell. Fodor, Jerry 1981 Some notes on what linguistics is about. In Readings in the Philosophy of Psychology, Ned Block (ed.), 197-207. Cambridge, Mass.: Harvard University Press. Fodor, Jerry 1983 The modularity of the mind. Cambridge, Mass.: MIT-Bradford. Fodor, Jerry 1985 Fodor's Guide to Mental Representation: The Intelligent Auntie's Vade-Mecum. Mind 94: 55-97. Fodor, Jerry 1987 Modules, Frames, Fridgeons, Sleeping Dogs, and the Music of the Spheres. In Modularity in knowledge representation and natural language understanding, Jay Garfield (ed.), 25-36. Cambridge, Mass.: MIT Press.

736 References Fodor, Jerry 1998 Fodor, Jerry 2000

The Trouble with Psychological Darwinism. London Review of Books 20.2: 11-13.

The mind doesn't work that way: The scope and limits of computational psychology. Cambridge, Mass.: MIT Press. Fodor, Jerry, and B. McLaughlin 1990 Connectionism and the problem of systematicity: why Smolensky's solution doesn't work. Cognition 35: 182-204. Fodor, Jerry, and Zenon Pylyshyn 1988 Connectionism and Cognitive Architecture: A Critical Analysis. Cognition 28: 3-71. Fox, Danny, and Howard Lasnik 2003 Successive-Cyclic Movement and Island Repair: The Difference between Sluicing and VP-Ellipsis. Linguistic Inquiry 34: 143-154. Fox, Danny, and David Pesetsky 2004 Cyclic linearization of syntactic structure. Theoretical Linguistics 31: 1-45. Franks, Steven, and Željko Bošković 2001 An argument for multiple spell-out. Linguistic Inquiry 32: 174183. Frascarelli, Mara 2006 Phases and Interpretation. In Phases of Interpretation, Mara Frascarelli (ed.), 1-14. Berlin: Mouton de Gruyter. Freidin, Robert, and Howard Lasnik forth Some roots of minimalism in generative grammar. In The Oxford Handbook of Linguistic Minimalism, Cedric Boeckx (ed.). Oxford: OUP. Gallego, Angel J. 2007 Phase Theory and Parametric Variation. Ph.D dissertation, Universitat Autònoma de Barcelona. Gallego, Angel J. 2010 Phase Theory. Amsterdam: Benjamins. Gardner, Howard 1985 The Mind's New Science: A History of the Cognitive Revolution. New York: Basic Books. Garfield, Jay (ed.) 1987 Modularity in Knowledge Representation and Natural Language Understanding. Cambridge, Mass.: MIT Press. Gengel, Kirsten 2009 Phases and Ellipsis. Linguistic Analysis 35: 21-42. Gerrans, Philip 2002 Modularity reconsidered. Language and Communication 22: 259268.

References 737 Ghini, Mirco 1993

Phonological Phrase formation in Italian: a new proposal. Toronto Working Papers in Linguistics 12: 41-77. Giegerich, Heinz 1999 Lexical Strata in English. Cambridge: Cambridge University Press. Göbbel, Edward 2007 Extraposition as PF Movement. To appear in Proceedings of WECOL 2006. Goldsmith, John 1990 Autsegmental and Metrical Phonology. Oxford: Blackwell. Goldsmith, John 1992 Local modelling in phonology. In Connectionism: Theory and practice, Steven Davis (ed.), 229-246. Oxford: OUP. Goldsmith, John 1993 Harmonic Phonology. In The last phonological rule, John Goldsmith (ed.), 21-60. Chicago: University of Chicago Press. Goldsmith, John 2008 Generative Phonology in the late 40s. Phonology 25: 37-59. Golston, Chris 1996 Prosodic constraints on roots, stems and words. In Interfaces in Phonology, Ursula Kleinhenz (ed.), 172-193. Berlin: Akademie Verlag. Green, Anthony 2007 Phonology Limited. Ms., University of Surrey Roehampton. [ROA#605]. Greenberg, Joseph 1970 On the 'language of observation' in linguistics. Working Papers on Language Universals (Stanford University) 4: 1-15. Grohmann, Kleanthes 2003 Prolific domains: On the anti-locality of movement dependencies. Amsterdam: Benjamins. Grohmann, Kleanthes 2007a Deriving Dynamic Interfaces. Linguistic Analysis 33: 3-19. Grohmann, Kleanthes 2007b Spelling out Dynamic Interfaces. Linguistic Analysis 33: 197-208. Guasti, Theresa, and Marina Nespor 1999 Is syntax Phonology-free? In Phrasal Phonology, René Kager and Wim Zonneveld (eds.), 73-97. Nijmegen: Nijmegen University Press. Guerssel, Mohand, and Jean Lowenstamm 1990 The derivational morphology of the Classical Arabic Verb. Ms., UQAM and University Paris 7.

738 References Gussenhoven, Carlos 1992 Intonational phrasing and the prosodic hierarchy. In Phonologica 1988, Wolfgang Dressler, Hans Luschützky, Oskar Pfeiffer and John Rennison (eds.), 89-99. Cambridge: Cambridge University Press. Gussenhoven, Carlos 2004 The phonology of tone and intonation. Cambridge: Cambridge University Press. Gussmann, Edmund 1998 Domains, relations, and the English agma. In Structure and Interpretation. Studies in Phonology, Eugeniusz Cyran (ed.), 101126. Lublin: Folium. WEB. Gussmann, Edmund 2002 Phonology: Analysis and Theory. Cambridge: Cambridge University Press. Gussmann, Edmund 2007 The Phonology of Polish. Oxford: Oxford University Press. 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. Gvozdanović, Jadranka 1986 Phonological domains. In Sandhi Phenomena in the Languages of Europe, Henning Andersen (ed.), 27-54. Berlin: Mouton de Gruyter. Haiden, Martin 2005 Concatenation and interpretation. In Organizing Grammar. Studies in Honor of Henk van Riemsdijk, Hans Broekhuis, Norbert Corver, Riny Huybregts, Ursula Kleinhenz and Jan Koster (eds.), 162-170. Berlin: Mouton de Gruyter. Hale, Kenneth, and Elisabeth Selkirk 1987 Government and tonal phrasing in Papago. Phonology 4: 151-183. Hale, Mark, and Charles Reiss 2008 The Phonological Enterprise. Oxford: OUP. Hall, R. A. Jr. 1946 Colloquial French Phonology. Studies in Linguistics 4: 70-90. Hall, R. A. Jr. 1948 Schismeme, open juncture, disjuncture. Studies in Linguistics 6: 21. Hall, Tracy 1992 Syllable Structure and Syllable-Related Processes in German. Tübingen: Niemeyer. Halle, Morris 1959 The Sound Pattern of Russian. The Hague: Mouton.

References 739 Halle, Morris 1964 Halle, Morris 1973a Halle, Morris 1973b Halle, Morris 1978

Halle, Morris 1986 Halle, Morris 1990 Halle, Morris 1997a Halle, Morris 1997b Halle, Morris 2005

On the Bases of Phonology. In Readings in the philosophy of language, Jerrold Katz and Jerry Fodor (eds.), 324-333. New York: Prentice-Hall. Stress rules in English: a new version. Linguistic Inquiry 5: 451464. The accentuation of Russian words. Language 49: 312-348. Formal vs. functional considerations in phonology. Studies in the Linguistic Sciences 8: 123-134. Reprinted in Brogyányi, Bela (ed.) 1979. Studies in diachronic, synchronic and typological linguistics: Festschrift for O. Szemerényi, 325-341. Amsterdam: Benjamins. On the Phonology-Morphology Interface. Ms, MIT. Respecting metrical structure. Natural Language and Linguistic Theory 8: 149-176. Distributed Morphology: Impoverishment and Fission. MIT Working Papers in Linguistics 30: 425-449. On stress and accent in Indo-European. Language 73: 275-313.

Palatalization/velar softening: what it is and what it tells us about the nature of language. Linguistic Inquiry 36: 23-41. Halle, Morris, James Harris, and Jean-Roger Vergnaud 1991 A reexamination of the stress erasure convention and Spanish stress. Linguistic Inquiry 22: 141-159. Halle, Morris, and Michael Kenstowicz 1991 The Free Element Condition and cyclic versus non-cyclic stress. Linguistic Inquiry 22: 457-501. Halle, Morris, and Alec Marantz 1993 Distributed Morphology and the Pieces of Inflection. In The view from Building 20. Essays in Linguistics in Honor of Sylvain Bromberger, Kenneth Hale and Samuel Keyser (eds.), 111-176. Cambridge, Mass.: MIT Press. Halle, Morris, and Ora Matushansky 2006 The Morphophonology of Russian Adjectival Inflection. Linguistic Inquiry 37: 351-404.

740 References Halle, Morris, and Karuvannur Mohanan 1985 Segmental Phonology of Modern English. Linguistic Inquiry 16: 57-116. Halle, Morris, and Andrew Nevins 2009 Rule Application in Phonology. In Contemporary Views on Architecture and Representations in Phonological Theory, Eric Raimy and Charles E. Cairns (eds.), 335-382. Cambridge, Mass.: MIT Press. Halle, Morris, and Jean-Roger Vergnaud 1987a An Essay on Stress. Cambridge, Mass.: MIT Press. Halle, Morris, and Jean-Roger Vergnaud 1987b Stress and the Cycle. Linguistic Inquiry 18: 45-84. Hamburger, H., and Kenneth Wexler 1975 A mathematical theory of learning transformational grammar. Journal of Mathematical Psychology 12: 137-177. Hammond, Michael 1995 There is no lexicon! Coyote Papers 10: 55-77. Hannahs, Stephen 1995 Prosodic Structure and French Morphophonology. Tübingen: Niemeyer. Hannahs, Stephen 1996 Phonological Structure and Soft Mutation in Welsh. In Interfaces in Phonology, Ursula Kleinhenz (ed.), 46-59. Berlin: Akademie Verlag. Hargus, Sharon 1993 Modeling the Phonology - Morphology Interface. In Studies in Lexical Phonology, Sharon Hargus and Ellen Kaisse (eds.), 45-74. New York: Academic Press. Hargus, Sharon, and Ellen Kaisse (eds.) 1993 Studies in Lexical Phonology. New York: Academic Press. Harley, Heidi, and Rolf Noyer 1999 Distributed Morphology. Glot 4: 3-9. Harnad, Steven 1990 The Symbol Grounding Problem. Physica D: Nonlinear Phenomena 42: 335-346. Harris, James 1993 Integrity of prosodic constituents and the domain of syllabification in Spanish and Catalan. In The view from Building 20. Essays in Linguistics in Honor of Sylvain Bromberger, Kenneth Hale and Samuel Keyser (eds.), 177-193. Cambridge, Harris, John Mass.: MIT Press. 1994 English sound structure. Oxford: Blackwell. WEB. Harris, Randy 1993 The Linguistics wars. Oxford: Oxford University Press.

References 741 Harris, Zellig 1951

Methods in Structural Linguistics. Edition 1960 entitled Structural Linguistics. Chicago & London: University of Chicago Press.

Haugeland, John 1989 Artificial Intelligence. The Very Idea. Cambridge, Mass.: MIT Press. Hauser, Marc, Noam Chomsky, and Tecumseh Fitch 2002 The faculty of language: what is it, who has it, and how did it evolve ? Science 298: 1569-1579. Hay, Jennifer 2002 From speech perception to morphology: affix ordering revisited. Language 78: 527-555. Hay, Jennifer, and Ingo Plag 2004 What constrains possible suffix combinations? On the interaction of grammatical and processing restrictions in derivational morphology. Natural Language & Linguistic Theory 22: 565-596. Hayes, Bruce 1982 Extrametricality and English stress. Linguistic Inquiry 13: 227276. Hayes, Bruce 1984 The phonology of rhythm in English. Linguistic Inquiry 15: 3374. Hayes, Bruce 1989 [1984] The Prosodic Hierarchy in Meter. In Manuscript circulated since 1984, published 1989 in Rhythm and Meter, Paul Kiparsky and G. Youmans (eds.), 201-260. Orlando, Florida: Academic Press. Hayes, Bruce 1990 Precompiled Phrasal Phonology. In The Phonology-Syntax Connection, Sharon Inkelas and Draga Zec (eds.), 85-108. Chicago: University of Chicago Press. Hayes, Bruce 1995a On what to teach undergraduates: some changing orthodoxies in phonological theory. In Linguistics in the Morning Calm 3, IkHwan Lee (ed.), 59-77. Seoul: Hanshin. Hayes, Bruce 1995b Metrical Stress Theory. Principles and Case Studies. Chicago, London: University of Chicago Press. Hayes, Bruce, Robert Kirchner, and Donca Steriade (eds.) 2004 Phonetically-Based Phonology. Cambridge: Cambridge University Press. Herken, R. (ed.) 1995 The Universal Turing Machine: A Half-Century Survey. Wien: Springer.

742 References Hermans, Ben, and Marc van Oostendorp (eds.) 1999 The Derivational Residue in Phonological Optimality Theory. Amsterdam: Benjamins. Hermer, Linda, and Elizabeth Spelke 1996 Modularity and development: the case of spatial reorientation. Cognition 61: 195-232. Hermon, Gabriella 1985 Syntactic Modularity. Dordrecht: Foris. Hickok, Gregory 2004 Dorsal and ventral streams: a framework for understanding aspects of the functional anatomy of language. Cognition 92: 67-99. Hickok, Gregory, and David Poeppel 2007 The cortical organization of speech perception. Nature Reviews Neuroscience 8: 393-402. Higginbotham, James 1987 The Autonomy of Syntax and Semantics. In Modularity in Knowledge Representation and Natural Language Understanding, Jay Garfield (ed.), 119-131. Cambridge, Mass.: MIT Press. Hill, Archibald 1954 Juncture and syllable division in Latin. Language 30: 439-447. Hinterhölzl, Roland 2006 The Phase Condition and cyclic Spell-out: Evidence from VPtopicalization. In Phases of Interpretation, Mara Frascarelli (ed.), 237-259. Berlin: de Gruyter. Hirschfeld, Lawrence, and Susan Gelman (eds.) 1994 Mapping the mind: Domain specificity in cognition and culture. Cambridge: CUP. Hirschfeld, Lawrence, and Susan Gelman 1994 Toward a topography of mind: an introduction to domain specificity. In Mapping the mind: Domain specificity in cognition and culture, Lawrence Hirschfeld and Susan Gelman (eds.), 3-35. Cambridge: CUP. Hoard, James 1966 Juncture and Syllable Structure in English. Phonetica 15: 96-109. Hoard, James 1973 On incorporating Grassmann's Law into Sanskrit phonology. Paper presented at the Western Conference on Linguistics, October 1973. Hockett, Charles 1942 A System of Descriptive Phonology. Language 18: 3-21. Hockett, Charles 1949 Two fundamental problems in phonemics. Studies in Linguistics 7: 29-51.

References 743 Hockett, Charles 1955 A Manual of Phonology. Baltimore: Waverly Press. Hockett, Charles 1958 A Course in Modern Linguistics. New York: Macmillan. Hooper, Joan 1975 The archi-segment in Natural Generative Phonology. Language 51: 536-560. Hooper, Joan 1976 An Introduction to Natural Generative Phonology. New York: Academic Press. Hornstein, Norbert 2009 A Theory of Syntax. Cambridge: CUP. Hornstein, Norbert, Jairo Nunes, and Kleanthes Grohmann 2005 Understanding Minimalism. Cambridge: CUP. Huddleston, Rodney 1972 The development of a non-process model in American structural linguistics. Lingua 30: 333-384. Hulst, Harry van der 2000 Modularity and Modality in Phonology. In Phonological Knowledge. Conceptual and Empirical Issues, Noel BurtonRoberts, Philip Carr and Gerard Docherty (eds.), 207-243. Oxford: Oxford University Press. Hulst, Harry van der 2010 A note on recursion in phonology. In Recursion and Human Language, Harry van der Hulst (ed.), 301-341. Berlin: de Gruyter. Hume, Elizabeth, and Keith Johnson (eds.) 2001 The Role of Speech Perception in Phonology. New York: Academic Press. Hurch, Bernhard, and Richard Rhodes (eds.) 1996 Natural Phonology: The State of the Art. Berlin: Mouton de Gruyter. Hyman, Larry 1975 Phonology: theory and analysis. New York: Holt, Rinehart & Winston. Hyman, Larry 1978 Word Demarcation. In Universals of Human Language, Vol 2, Joseph Greenberg (ed.), 443-470. Stanford: Stanford University Press. Hyman, Larry, Francis Katamba, and Livingstone Walusimbi 1987 Luganda and the strict layer hypothesis. Phonology Yearbook 4: 87-108.

744 References Hyman, Larry, and K. Kim 1973 On the non-status of morpheme boundaries in phonology. Paper presented at the Winter Meeting of the Linguistic Society of America, San Diego. Idsardi, William, and Eric Raimy forth Three types of linearization and the temporal aspects of speech. In Principles of linearization, Ian Roberts and M. T. Biberauer (eds.). Berlin: Mouton de Gruyter. WEB. Inkelas, Sharon 1990 Prosodic Constituency in the Lexicon. New York: Garland. Inkelas, Sharon 1993 Deriving Cyclicity. In Studies in Lexical Phonology, Sharon Hargus and Ellen Kaisse (eds.), 75-110. New York: Academic Inkelas, Sharon Press. 1996 Dominant affixes and the phonology-morphology interface. In Interfaces in Phonology, Ursula Kleinhenz (ed.), 128-154. Berlin: Akademie Verlag. Inkelas, Sharon 1998 The theoretical status of morphologically conditioned phonology: a case study of dominance effects. Yearbook of Morphology 1997: 121-155. Inkelas, Sharon 1999 Exceptional stress-attracting suffixes in Turkish: representations versus the grammar. In The Prosodiy-Morphology Interface, René Kager, Harry van der Hulst and Wim Zonneveld (eds.), 134-187. Cambridge: CUP. Inkelas, Sharon 2000 Phonotactic blocking through structural immunity. In Lexicon in Focus, B. Stiebels and D. Wunderlich (eds.), 7-40. Berlin: Akademie Verlag. Inkelas, Sharon, and Young-Mee Yu Cho 1993 Inalterability as prespecification. Language 69: 529-574. Inkelas, Sharon, and Cemil Orhan Orgun 1995 Level ordering and economy in the lexical phonology of Turkish. Language 71: 763-793. Inkelas, Sharon, and Draga Zec (eds.) 1990c The Phonology-Syntax Connection. Chicago & London: Chicago University Press. Inkelas, Sharon, and Draga Zec 1988 Serbo-Croatian pitch accent: the interactions of tone, stress and intonation. Language 64: 227-248.

References 745 Inkelas, Sharon, and Draga Zec 1990a Introduction. In The Phonology-Syntax Connection, Sharon Inkelas and Draga Zec (eds.), xiii-xv. Chicago: University of Chicago Press. Inkelas, Sharon, and Draga Zec 1990b Prosodically constrained syntax. In The Phonology-Syntax Connection, Sharon Inkelas and Draga Zec (eds.), 365-378. Chicago: Chicago University Press. Inkelas, Sharon, and Draga Zec 1995 Syntax-phonology Interface. In The Handbook of Phonological Theory, John Goldsmith (ed.), 535-549. Oxford: Blackwell. Isac, Daniela, and Charles Reiss 2008 I-Language. An Introduction to Linguistics as Cognitive Science. Oxford: OUP. Ishihara, Shinichiro 2007 Major phrase, focus intonation, multiple spell-out. The Linguistic Review 24: 137-167. Itô, Junko 1986 Syllable Theory in Prosodic Phonology. Ph.D dissertation, University of Massachusetts at Amherst. Itô, Junko, and Armin Mester 1995 Japanese Phonology. In The Handbook of Phonological Theory, John Goldsmith (ed.), 816-838. Oxford: Blackwell. Itô, Junko, and Armin Mester 1999a Realignment. In The prosody-morphology interface, René Kager, Harry van der Hulst and Wim Zonneveld (eds.), 188-217. Cambridge: Cambridge University Press. Itô, Junko, and Armin Mester 1999b The phonological lexicon. In The Handbook of Japanese Linguistics, Natsuko Tsujimura (ed.), 62-100. Oxford: Blackwell. Itô, Junko, and Armin Mester 2001 Covert Generalizations in Optimality Theory: the role of stratal faithfulness constraints. Studies in Phonetics, Phonology and Morphology 7: 273-299. Itô, Junko, and Armin Mester 2003 [1992] Weak Layering and Word Binarity. In Manuscript circulated since 1992, published 2003 in A New Century of Phonology and Phonological Theory. A Festschrift for Professor Shosuke Haraguchi on the Occasion of His Sixtieth Birthday, Takeru Honma, Masao Okazaki, Toshiyuki Tabata and Shin-ichi Tanaka (eds.), 26-65. Tokyo: Kaitakusha.

746 References Itô, Junko, Armin Mester, and Jaye Padgett 2004 Licensing and Underspecification in Optimality Theory. In Optimality Theory in Phonology, John McCarthy (ed.), 533-541. Oxford: Blackwell. Iverson, Gregory 2004 Deriving the Derived Environment Constraint in Nonderivational Phonology. Studies in Phonetics, Phonology and Morphology 11: 1-23. Iverson, Gregory, and Deirdre Wheeler 1988 Blocking and the Elsewhere Condition. In Theoretical Morphology: Approaches in Modern Linguistics, Michael Hammond and M. Noonan (eds.), 325-338. San Diego: Academic Jackendoff, Ray Press. 1969 Some rules of semantic interpretation for English. Ph.D dissertation, MIT. Jackendoff, Ray 1987 Consciousness and the computational mind. Cambridge, Mass.: MIT Press. Jackendoff, Ray 1992 Languages of the mind. Cambridge, Mass.: MIT Press. Jackendoff, Ray 1994 Patterns in the Mind. Language and human nature. New York: BasicBooks. Jackendoff, Ray 1997 The Architecture of the Language Faculty. Cambridge, Massachusetts: MIT Press. Jackendoff, Ray 2002 Foundations of Language. Brain, Meaning, Grammar, Evolution. Oxford: Oxford University Press. Jenkins, Lyle 2000 Biolinguistics. Exploring the Biology of Language. Cambridge: CUP. Joos, Martin (ed.) 1957 Readings in Linguistics: The development of descriptive linguistics in America since 1925. Washington D.C.: American Council of Learned Societies. Joos, Martin 1964 A chapter of semology in the English verb. In Report of the 15th Annual (1st International) Round Table Meeting on Linguistics and Language Studies, C. Stuart (ed.), 59-72. Washington, D. C.: Georgetown University Press. Julien, Marit 2006 On argument displacement in English and Scandinavian. Working Papers in Scandinavian Syntax 77: 1-69.

References 747 Julien, Marit 2007

On the Relation between Morphology and Syntax. In The Oxford Handbook of Linguistic Interfaces, Gillian Ramchand and Charles Reiss (eds.), 209-238. Oxford: OUP. Junković, Zvonimir 1980 Mot et unité accentuelle. Travaux du cercle linguistique de Nice 2: 117-126. Junković, Zvonimir 1990 Accentologie théorique et accentologie comparée. Travaux du cercle linguistique de Nice 12: 5-14. Kager, René 1999 Optimality Theory. Cambridge: Cambridge University Press. Kager, René 2000 Stem Stress and Peak Correspondence in Dutch. In Optimality Theory, Joost Dekkers, Frank van der Leeuw and Jeroen van de Weijer (eds.), 121-150. Oxford: Oxford University Press. Kahn, Daniel 1976 Syllable-based generalizations in English phonology. Ph.D dissertation, MIT. Published by Garland Press, New York 1980. Kahnemuyipour, Arsalan 2009 The Syntax of Sentential Stress. Oxford: OUP. Kaisse, Ellen 1983 The syntax of auxiliary reduction in English. Language 59: 93122. Kaisse, Ellen 1985 Connected Speech. The interaction of Syntax and Phonology. London, New York: Academic Press. Kaisse, Ellen 1990 Toward a Typology of Postlexical Rules. In The PhonologySyntax Connection, Sharon Inkelas and Draga Zec (eds.), 127143. Chicago: Chicago University Press. Kaisse, Ellen, and Sharon Hargus 1993 Introduction. In Studies in Lexical Phonology, Sharon Hargus and Ellen Kaisse (eds.), 1-19. New York: Academic Press. Kaisse, Ellen, and Patricia Shaw 1985 On the theory of Lexical Phonology. Phonology Yearbook 2: 130. Kaisse, Ellen, and Arnold Zwicky 1987 Introduction: syntactic influences on phonological rules. Phonology 4: 3-11. Kanerva, Jonni 1990 Focusing on Phonological Phrases in Chichewa. In The Phonology-Syntax connection, Sharon Inkelas and Draga Zec (eds.), 145-161. Chicago: Chicago University Press.

748 References Karmiloff-Smith, Annette 1998 Development itself is the key to understanding developmental disorders. Trends in Cognitive Sciences 2: 389-398. Karmiloff-Smith, Annette, Edward Klima, Ursula Bellugi, Julia Grant, and Simon Baron-Cohen 1995 Is There a Social Module? Language, Face Processing, and Theory of Mind in Individuals with Williams Syndrome. Journal of Cognitive Neuroscience 7: 196-208. Karttunen, Lauri 1993 Finite-State Constraints. In The last phonological rule, John Goldsmith (ed.), 173-194. Chicago: University of Chicago Press. Kasher, Asa (ed.) 1991 The Chomskyan Turn. Oxford: Blackwell. Katz, Jerrold 1984 An outline of a Platonist grammar. In Talking Minds: The Study of Language in Cognitive Science, Thomas G. Bever, John M. Carroll and Lance A. Miller (eds.), 17-48. Cambridge, Mass.: MIT Press. Katz, Jerrold, and Thomas Bever 1974 The fall and rise of empiricism. Bloomington: Indiana University Linguistics Club. WEB. Kaye, Jonathan 1989 Phonology. A cognitive view. Hillsdale: Erlbaum. WEB. Kaye, Jonathan 1990a 'Coda' licensing. Phonology 7: 301-330. WEB. Kaye, Jonathan 1990b What ever happened to dialect B ? In Grammar in Progress: GLOW Essays for Henk van Riemsdijk, Joan Mascaró and Marina Nespor (eds.), 259-263. Dordrecht: Foris. Kaye, Jonathan 1992a 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.), 141155. Cambridge: Cambridge University Press. WEB. Kaye, Jonathan 1992b Do you believe in magic? The story of s+C sequences. SOAS Working Papers in Linguistics and Phonetics 2: 293-313. Reprinted in A Festschrift for Edmund Gussmann, edited by Henryk Kardela & Bogdan Szymanek, 155-176. Lublin 1996: Lublin University Press. WEB.

References 749 Kaye, Jonathan 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. WEB. Kaye, Jonathan 2001 Working with licensing constraints. In Constraints and Preferences, Katarzyna Dziubalska-Kołaczyk (ed.), 251-268. Berlin & New York: Mouton de Gruyter. WEB. Kaye, Jonathan 2005 "GP, I'll have to put your flat feet on the ground". In Organizing Grammar. Studies in Honor of Henk van Riemsdijk, Hans Broekhuis, Norbert Corver, Riny Huybregts, Ursula Kleinhenz and Jan Koster (eds.), 283-288. Berlin: Mouton de Gruyter. Kaye, Jonathan 2008 Canadian Raising, eh? Ms. Kaye, Jonathan, Jean Lowenstamm, and Jean-Roger Vergnaud 1985 The internal structure of phonological representations: a theory of Charm and Government. Phonology Yearbook 2: 305-328. WEB. Kaye, Jonathan, Jean Lowenstamm, and Jean-Roger Vergnaud 1990 Constituent structure and government in phonology. Phonology 7: 193-231. WEB. Kaye, Jonathan, and Jean-Roger Vergnaud 1990 Phonology, morphology and the lexicon. Paper presented at GLOW, St. John's College, Cambridge University. Kayne, Richard S. 1994 The Antisymmetry of Syntax. Cambridge, Mass.: MIT Press. Kean, Mary-Louise 1971 Strict Cyclicity. Ms., MIT. Kean, Mary-Louise 1974 The Strict Cycle in Phonology. Linguistic Inquiry 5: 179-203. Kennedy, Chris 2001 Polar Opposition and the Ontology of ‘Degrees’. Linguistics and Philosophy 24: 33-70. Kenstowicz, Michael 1996 Base-identity and uniform exponence: alternatives to cyclicity. In Current Trends in Phonology. Models and Methods, Jacques Durand and Bernard Laks (eds.), 363-393. Salford: ESRI. Kenstowicz, Michael, and Charles Kisseberth 1977 Topics in Phonological Theory. New York: Academic Press. Kenstowicz, Michael, and Charles Kisseberth 1979 Generative Phonology. Description and Theory. San Diego: Academic Press.

750 References Kepke, John 1948 A comment on terminology. Studies in Linguistics 6: 20. Kidima, Lukowa 1990 Tone and Syntax in Kiyaka. In The Phonology-Syntax Connection, Sharon Inkelas and Draga Zec (eds.), 195-216. Chicago: Chicago University Press. Kingston, John 2007 The phonetics-phonology interface. In The Cambridge Handbook of Phonology, Paul De Lacy (ed.), 435-456. Cambridge: CUP. Kiparsky, Paul 1968-1973 How abstract is phonology? In Manuscript circulated since 1968 and published 1973 in: Three Dimensions of Linguistic Theory, Osamu Fujimura (ed.), 5-56. Tokyo: TEC. Kiparsky, Paul 1973a Abstractness, opacity and global rules. In Three Dimensions in Phonological Theory, Osamu Fujimura (ed.), 57-86. Tokyo: TEC. Kiparsky, Paul 1973b Productivity in Phonology. In Issues in Phonological Theory, Michael Kenstowicz and Charles Kisseberth (eds.), 169-176. The Hague: Mouton. Kiparsky, Paul 1973c 'Elsewhere' in Phonology. In A Festschrift for Morris Halle, Stephen Anderson and Paul Kiparsky (eds.), 93-106. New York: Holt, Rinehart & Winston. Kiparsky, Paul 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. Kiparsky, Paul 1979 Metrical Structure Assignment is Cyclic. Linguistic Inquiry 10: 421-441. Kiparsky, Paul 1982a Lexical morphology and phonology. In Linguistics in the morning calm, In-Seok Yang (ed.), 3-91. Seoul: Hanshin. Kiparsky, Paul 1982b 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. WEB. Kiparsky, Paul 1982c Explanation in Phonology. Dordrecht: Foris. Kiparsky, Paul 1985a Some Consequences of Lexical Phonology. Phonology Yearbook 2: 85-138.

References 751 Kiparsky, Paul 1985b On the lexical phonology of Icelandic. In Nordic Prosody 3, C. Elert, I. Johansson and E. Stangert (eds.), 135-164. Umeå: University of Umeå. Kiparsky, Paul 1993 Blocking in nonderived environments. In Studies in Lexical Phonology, Sharon Hargus and Ellen Kaisse (eds.), 277-313. San Diego: Academic Press. Kiparsky, Paul 2000 Opacity and cyclicity. The Linguistic Review 17: 351-365. Kisseberth, Charles, and Mohammad Imam Abasheikh 1974 Vowel Length in Chi-Mwi:ni -- a Case Study of the Role of Grammar in Phonology. In Papers from the parasession on Natural Phonology, Anthony Bruck, Robert Fox and Michael La Galy (eds.), 193-209. Chicago: Chicago Linguistic Society. Kleinhenz, Ursula (ed.) 1996 Interfaces in Phonology. Berlin: Akademie Verlag. Kleinhenz, Ursula 1998 On Words and Phrases in Phonology. A comparative study with focus on German. Ph.D thesis, University of Tübingen. Koerner, E.F.K. 2002 The Chomskyan 'Revolution' and its historiography. In Toward a History of American Linguistics, E.F.K. Koerner (ed.), 151-209. London: Routledge. Koerner, E.F.K. 2004 On "influence" in linguistic historiography: morphohonemics in American structuralism. In Essays in the History of Linguistics, E.F.K. Koerner (ed.), 65-101. London: Benjamins. Kratzer, Angelika, and Elisabeth Selkirk 2007 Phase theory and prosodic spellout: the case of verbs. The Linguistic Review 24: 93-135. Kula, Nancy 2008 Derived environment effects: a representational approach. Lingua 118: 1328-1343. Ladd, Robert 1986 Intonational phrasing: the case for recursive prosodic structure. Phonology 3: 311-340. Ladd, Robert 1996 Intonational Phonology. Cambridge: CUP. Ladd, Robert 2000 Bruce, Pierrehumbert, and the Elements of Intonational Phonology. In The interaction of constraints on prosodic phrasing. Prosody: Theory and Experiments, Merle Horne (ed.), 37-50. Dordrecht: Kluwer.

752 References Lahrouchi, Mohamed 2001 Aspects morpho-phonologiques de la dérivation verbale en berbère (parler chleuh d'Agadir). Ph.D dissertation, Université Paris 7. Lakoff, George 1993 Cognitive Phonology. In The last phonological rule, John Goldsmith (ed.), 117-145. Chicago: University of Chicago Press. Laks, Bernard 1996 Langage et cognition, l'approche connexionniste. Paris: Hermès. Laks, Bernard 2006 La phonologie générative naturelle et la phonologie naturelle. In History of the Language Sciences, vol.3 (HSK 18.3), Sylvain Auroux, E.F.K. Koerner, Hans-Josef Niederehe and Kees Versteegh (eds.), 2271-2280. Berlin: de Gruyter. Lampitelli, Nicola forth a The basic elements of a noun: morphosyntax of the Bosnian declensional system. Ms., Université Paris 7. Available at www.linguist.univ-parisdiderot.fr/~nlampitelli/drafts&papers.html. Lampitelli, Nicola forth b On the abstractness of the phonological component in a piecebased morphological theory. Ms., Université Paris 7. Available at www.linguist.univ-parisdiderot.fr/~nlampitelli/drafts&papers.html. Langacker, Ronald 1987 Foundations of Cognitive Grammar, 2 Vols. Stanford: Stanford University Press. Langendoen, Terence 1975 Finite state parsing of phrase structure languages and the status of readjustment rules in the grammar. Linguistic Inquiry 6: 533-554. Larson, G. 1992 Dynamic computational networks and the representation of phonological information. Ph.D dissertation, University of Lasnik, Howard Chicago. 1972 Analysis of negation in English. Ph.D dissertation, MIT. Lasnik, Howard, and Terje Lohndal 2009 Government-binding/principles and parameters theory. Wiley Interdisciplinary Reviews: Cognitive Science 1: 40-50. Lass, Roger 1971 Boundaries as obstruents: Old English voicing assimilation and universal strength hierarchies. Journal of Linguistics 7: 15-30. Lebeaux, David 1988 Language Acquisition and the Form of the Grammar. Ph.D dissertatioan, University of Massachusetts.

References 753 Legate, Julie Anne 2003 Some Interface Properties of the Phase. Linguistic Inquiry 34: 506-516. Legendre, Geraldine, Yoshiro Miyata, and Paul Smolensky 1990 Harmonic Grammar - a formal multi-level connectionist theory of linguistic well-formedness: theoretical foundations. In Proceedings of the twelfth annual conference of the Cognitive Science Society, the Cognitive Science Society (ed.), 388-395. Cambridge, MA.: Erlbaum. Lehiste, Ilse 1960 An Acoustic-Phonetic Study of Internal Open Juncture. Basel, New York: Karger (supplement to Phonetica 5). Lehiste, Ilse 1962 Acoustic Studies of Boundary Signals. In Proceedings of the Fourth International Congress of Phonetic Sciences, Antti Sovijärvi and Pentti Aalto (eds.), 178-187. The Hague: Mouton. Lehiste, Ilse 1965 Juncture. In Proceedings of the Fifth International Congress of Phonetic Sciences, Eberhard Zwirner and Wolfgang Bethge (eds.), 172-200. Basel: Karger. Leopold, Werner F. 1948 German CH. Language 24: 179-180. Li, Yafei 2005 X°: a Theory of the Morphology-syntax Interface. Cambridge, MA: MIT Press. 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. Lieber, Rochelle 1981 On the organization of the lexicon. Ph.D dissertation, MIT. Lieber, Rochelle, and Sergio Scalise 2007 The Lexical Integrity Hypothesis in a new Theoretical Universe. In Proceedings of the Fifth Mediteranean Morphology Meeting, Geert Booij, L. Ducceschi, Bernard Fradin, E. Guevara, Angela Ralli and Sergio Scalise (eds.), 1-24. Bologna: Università degli Studi di Bologna. Lightner, Theodore 1978 Generative Phonology. In A Survey of Linguistic Science, William Orr Dingwall (ed.), 1-32. 2nd edition Stamford, Connecticut: Greylock. Lightner, Theodore 1981 New explorations into Derivational Morphology. In Phonology in the 80's, Didier Goyvaerts (ed.), 93-99. Ghent: Story-Scientia.

754 References Lightner, Theodore 1985 On universal rules. Lingvisticae Investigationes 9: 419-420. Lobeck, Anne, and Ellen Kaisse 1984 On the domain of locality conditions. Proceedings of the West Coast Conference on Formal Linguistics 3: 170-178. Loevenbruck, H., Monica Baciu, C. Segebarth, and Christian Abry 2005 Broca's region under focus: An fMRI study of the production of deixis via syntactic extraction and prosodic focus. Journal of Neurolingustics, 18: 237-258. Lombardi, Linda 2001 Introduction. In Segmental phonology in Optimality Theory. Constraints and Representations, Linda Lombardi (ed.), 1-9. Cambridge: Cambridge University Press. Loporcaro, Michele 1999 Teoria fonologica e ricerca empirica sull'italiano e i suoi dialetti. In Fonologia e morfologia dell'italiano e dui dialetti d'Italia. Atti del 31° Congresso della Società di Linguistica Italiana, Paola Benincà, Alberto Mioni and Laura Vanelli (eds.), 117-151. Roma: Bulzoni. Lowenstamm, Jean 1989 Prosodic Government. Langues Orientales Anciennes, Philologie et Linguistique 2: 221-223. Lowenstamm, Jean 1996 CV as the only syllable type. In Current trends in Phonology. Models and Methods, Jacques Durand and Bernard Laks (eds.), 419-441. Salford, Manchester: ESRI. WEB. Lowenstamm, Jean 1999 The beginning of the word. In Phonologica 1996, John Rennison and Klaus Kühnhammer (eds.), 153-166. La Hague: Holland Academic Graphics. WEB. Lowenstamm, Jean 2008 On Little n, √, and Types of Nouns. In Sounds of Silence: Empty Elements in Syntax and Phonology, Jutta Hartmann, Veronika Hegedüs and Henk van Riemsdijk (eds.), 105-143. Amsterdam: Elsevier. 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. Łubowicz, Anna 2002 Derived environment effects in Optimality Theory. Lingua 112: 243-280.

References 755 Łubowicz, Anna 2004 Derived Environment Effects in Optimality Theory. In Optimality Theory in Phonology, John McCarthy (ed.), 523-532. Oxford: Blackwell. Łubowicz, Anna 2005 Locality of conjunction. In Proceedings of the 24th West Coast Conference on Formal Linguistics, John Alderete, Chung-hye Han and Alexei Kochetov (eds.), 254-262. Somerville, Mass.: Cascadilla. Macdonald, Cynthia, and Graham Macdonald 1995 Connectionism: Debates on Psychological Explanation. Oxford: Blackwell. Manfredi, Victor forth The referential prosody of bare arguments. Ms., Boston University. Available at Manzini, Rita http://people.bu.edu/manfredi/writing.html. 1983 Syntactic conditions on phonological rules. MIT Working Papers in Linguistics 5: 1-9. Marantz, Alec 1984 On the nature of grammatical relations. Cambridge, Mass.: MIT Press. Marantz, Alec 1997a No escape from syntax: don't try morphological analysis in the privacy of your own lexicon. University of Pennsylvania Working Papers in Linguistics 4.2: 201-225. Marantz, Alec 1997b Cat as a phrasal idiom: stem suppletion, or the arbitrariness of the sign. Paper presented at Paper presented at the University Paris 8. Marantz, Alec 2000 Roots. Paper presented at the Conference on Afro-Asiatic languages, Paris. Marantz, Alec 2001 Words. Handout from the 20th West Coast Conference on Formal Linguistics. Marantz, Alec 2007 Phases and words. In Phases in the theory of grammar, S.-H. Choe (ed.), 191-222. Seoul: Dong In. Marcus, Gary 1993 Negative evidence in language acquisition. Cognition 46: 53-85. Marlo, Michael 2008 Post-syntactic phonology: evidence from Bantu. Paper presented at the 5th North American Phonology Conference, Montreal 9-11 May.

756 References Marr, David 1982 Vision. San Francisco: Freeman. Marshall, John 2001 Cognition and Neuroscience: Where Were We? In Language, Brain and Cognitive Development. Essays in Honor of Jacques Mehler, Emmanuel Dupoux (ed.), 503-512. Cambridge, Mass.: MIT Press. Marušič, Franc 2005 On non-simultaneous phases. Ph.D. dissertation, SUNY, Stony Brook. Marušič, Franc, and Rok Žaucer 2006 On the intensional feel-like construction in Slovenian. Natural Language and Linguistic Theory 24: 1093-1159. Marvin, Tatjana 2002 Topics in the Stress and Syntax of Words. Ph.D dissertation, MIT. Marvin, Tatjana 2008 The interaction between stress, syntax and meaning in Slovenian priscianic formations. In Studies in Formal Slavic Linguistics, Franc Marušic and Rok Žaucer (eds.), 191-212. Frankfurt am Main: Peter Lang. Mascaró, Joan 1976 Catalan Phonology and the Phonological Cycle. Ph.D. dissertation. Matushansky, Ora 2005 Going through a phase. MIT Working Papers in Linguistics 49: 157-181. Mazzola, Michael 1999 On the independence of suprasegmental constituency. In Issues in phonological structure, S.J. Hannahs and Mike Davenport (eds.), 181-193. Amsterdam: Benjamins. McCarthy, John (ed.) 2004 Optimality Theory in Phonology. A Reader. Oxford: Blackwell. McCarthy, John 1979 Formal problems in Semitic phonology and morphology. Ph.D dissertation, MIT. PubliéeàNewYork1985:GarlandPress. McCarthy, John 1980 A note on the accentuation of Damascene Arabic. Studies in the Linguistic Sciences 10: 77-98. McCarthy, John 1999 Sympathy and phonological opacity. Phonology 16: 331-399. McCarthy, John 2003a Comparative Markedness. Theoretical Linguistics 29: 1-51.

References 757 McCarthy, John 2003b Sympathy, Cumulativity, and the Duke-of-York Gambit. In The Syllable in Optimality Theory, Caroline Féry and Ruben van de Vijver (eds.), 23-76. Cambridge: CUP. McCarthy, John 2005 Optimal Paradigms. In Paradigms in phonological theory, Laura Downing, Tracy Hall and Renate Raffelsiefen (eds.), 295-371. Oxford: Oxford University Press. McCarthy, John 2007 Hidden Generalizations: Phonological Opacity in Optimality Theory. London: Equinox. McCarthy, John, and Alan Prince 1986 Prosodic Morphology. Ms. McCarthy, John, and Alan Prince 1988 Quantitative transfer in reduplicative and templatic morphology. In Linguistics in the morning calm, Vol.2, Linguistic Society of Korea (ed.), 3-35. Seoul: Hanshin. McCarthy, John, and Alan Prince 1990 Foot and word in Prosodic Morphology: the Arabic broken plural. Natural Language and Linguistic Theory 8: 209-283. McCarthy, John, and Alan Prince 1993 Generalized Alignment. In Yearbook of Morphology 1993, Geert Booij and Jaap van Marle (eds.), 79-153. Dordrecht: Kluwer. Abridged version in McCarthy, John (ed.) 2004. Optimality Theory in Phonology, 451-463. Oxford: Blackwell. McCarthy, John, and Alan Prince 1994 The Emergence of the Unmarked: Optimality in Prosodic Morphology. In Proceedings of the North-East Linguistic Society 24, Merce Gonzalez (ed.), 333-379. Amherst: Graduate Linguistic Student Association. Downloadable at http://roa.rutgers.edu/ (#261). McCarthy, John, and Alan Prince 1995a Faithfulness and Reduplicative Identity. In Papers in Optimality Theory, Jill Beckman, Laura Walsh Dickey and Suzanne Urbanczyk (eds.), 249-384. Amherst: GLSA University of Massachusetts. McCarthy, John, and Alan Prince 1995b Prosodic Morphology. In The handbook of Phonological Theory, John Goldsmith (ed.), 318-366. Cambridge, Mass.: Blackwell. McCarthy, John, and Alan Prince 1996 Prosodic Morphology 1986. Ms., University of Massachusetts, Rutgers University.

758 References McCarthy, John, and Alan Prince 2001 Prosodic Morphology. Constraint Interaction and Satisfaction. Ms, ROA #482. McCawley, James 1968 The Phonological Component of a Grammar of Japanese. The Hague: Mouton. McMahon, April 2000 Lexical Phonology and the history of English. Cambridge: Cambridge University Press. McMahon, April 2007 Who's afraid of the vowel shift rule? Language Sciences 29: 341359. Megerdoomian, Karine 2003 Asymmetries in Form and Meaning: Surface Realization and the Interface Conditions. Paper presented at Approaching Asymmetry at the Interfaces, UQAM, Montreal. Mehler, Jacques, John Morton, and Peter Jusczyk 1984 On reducing language to biology. Cognitive Neuropsychology 1: 83-116. Merchant, Jason 2001 The Syntax of Silence: Sluicing, islands, and identity in ellipsis. Ph.D dissertation, University of California at Santa Cruz. Michalski, Grzegorz 2009 Phonology with interfaces: The morphophonology and postlexical phonology of English and Polish. Ph.D dissertation, University of Poznań. Miller, Philip, Geoffrey Pullum, and Arnold Zwicky 1997 The Principle of Phonology-Free Syntax: four apparent counterexamples in French. Journal of Linguistics 33: 67-90. Mohanan, Karuvannur 1982 Lexical Phonology. Ph.D dissertation, MIT. Mohanan, Karuvannur 1986 The Theory of Lexical Phonology. Dordrecht: Reidel. Mohanan, Karuvannur, and Tara Mohanan 1984 Lexical Phonology of the consonant system in Malayalam. Linguistic Inquiry 15: 575-602. Monachesi, Paola 1996 On the representation of Italian clitics. In Interfaces in Phonology, Ursula Kleinhenz (ed.), 83-101. Berlin: Akademie Verlag. Moravcsik, Edith 2000 Infixation. In Morphology. An international handbook on inflection and word-formation, Vol.1, Geert Booij (ed.), 545-552. Berlin: de Gruyter.

References 759 Moscovitch, M., G. Winocur, and M. Behrmann 1997 What Is Special about Face Recognition? Nineteen Experiments on a Person with Visual Object Agnosia and Dyslexia but Normal Face Recognition. Journal of Cognitive Neuroscience 9: 555-604. Moulton, William 1947 Juncture in Modern Standard German. Language 23: 212-226. Müller, Gereon 2004 Phase Impenetrability and Wh-Intervention. In Minimality Effects in Syntax, Arthur Stepanov, Gisbert Fanselow and Ralf Vogel (eds.), 289-325. Berlin: Mouton de Gruyter. Myers, Scott 1987 Vowel shortening in English. Natural Language and Linguistic Theory 5: 485-518. Napoli, Donna Jo, and Marina Nespor 1979 The syntax of word-initial consonant gemination in Italian. Language 55: 812-841. Neeleman, Ad, and Tanya Reinhart 1998 Scrambling and the PF-interface. In The Projection of Arguments: Lexical and Compositional Factors, Miriam Butt and Wilhelm Geuder (eds.), 309-353. Chicago: CSLI. Neeleman, Ad, and Hans van de Koot 2006 On syntactic and phonological representations. Lingua 116: 15241552. Neijt, Anneke 1985 Clitics in arboreal phonology. In Advances in nonlinear phonology, Harry van der Hulst and Norval Smith (eds.), 179-192. Dordrecht: Foris. Nespor, Marina 1985 The phonological word in Italian. In Advances in Nonlinear Phonology, Harry van der Hulst and Norval Smith (eds.), 193204. Dordrecht: Foris. Nespor, Marina 1986 The phonological word in Greek and Italian. In Sandhi Phenomena in the Languages of Europe, Henning Andersen (ed.), 65-74. Berlin: Mouton de Gruyter. Nespor, Marina 1987 Vowel degemination and fast speech rules. Phonology 4: 61-85. Nespor, Marina 1988 Aspects of the interaction between prosodic phonology and the phonology of rhythm. In Certamen Phonologicum, Pier Marco Bertinetto and Michele Loporcaro (eds.), 189-230. Torino: Rosenberg & Sellier.

760 References Nespor, Marina 1990 On the separation of prosodic and rhythmic phonology. In The Phonology Syntax Connection, Sharon Inkelas and Draga Zec (eds.), 243-258. Chicago: University of Chicago Press. Nespor, Marina 1999 Stress Domains. In Word Prosodic Systems in the Languages of Europe, Harry van der Hulst (ed.), 117-159. Berlin: Mouton de Gruyter. Nespor, Marina, Theresa Guasti, and Anne Christophe 1996 Selecting word order: the Rhythmic Activation Principle. In Interfaces in Phonology, Ursula Kleinhenz (ed.), 1-26. Berlin: Akademie Verlag. Nespor, Marina, and Angela Ralli 1996 Morphology-phonology interface: phonological domains in Greek compounds. The Linguistic Review 13: 357-382. Nespor, Marina, and Irene Vogel 1979 Clash Avoidance in Italian. Linguistic Inquiry 10: 467-482. Nespor, Marina, and Irene Vogel 1982 Prosodic domains of external sandhi rules. In The Structure of Phonological Representations, Part I, Harry van der Hulst and Norval Smith (eds.), 225-255. Dordrecht: Foris. Nespor, Marina, and Irene Vogel 1983 Prosodic structure above the word. In Prosody: Models and Measurements, Anne Cutler and Robert Ladd (eds.), 123-140. Berlin: Springer. Nespor, Marina, and Irene Vogel 1986 Prosodic Phonology. Dordrecht: Foris. Nespor, Marina, and Irene Vogel 1989 On clashes and lapses. Phonology 6: 69-116. Newell, Allen 1980 Physical Symbol Systems. Cognitive Science 4: 135-183. Newell, Allen, Paul Rosenblom, and John Laird 1989 Symbolic Architectures for Cognition. In Foundations of Cognitive Science, Michael Posner (ed.), 93-131. Cambridge, Mass.: MIT Press. Newell, Allen, and Herbert Simon 1972 Human Problem Solving. Englewood Cliffs: Prentice-Hall. Newell, Heather 2005a Bracketing paradoxes and particle verbs: a late adjunction analysis. In Proceedings of Console XIII, Sylvia Blaho, Luis Vicente and Erik Schoorlemmer (eds.), 249-272: Student Organization of Linguistics in Europe.

References 761 Newell, Heather 2005b The phonological phase. McGill Working Papers in Linguistics 19: 21-64. Newell, Heather 2008 Aspects of the morphology and phonology of phases. Ph.D dissertation, McGill University Montreal. Newell, Heather, and Tobias Scheer 2007 Procedural First. Paper presented at the 38th Poznań Linguistic Meeting, Poznań 13-16 September. Newman, Stanley 1946 On the stress system of English. Word 2: 171-187. Newmeyer, Frederick 1986 Linguistic theory in America. 2nd edition New York: Academic Press. Newmeyer, Frederick 1996 Generative linguistics: a historical perspective. London: Routledge. Newmeyer, Frederick 1998 Language Form and Language Function. Cambridge, Mass.: MIT Press. Newmeyer, Frederick 2005 Some Remarks on Roeper’s Remarks on Chomsky’s “Remarks". SKASE Journal of Theoretical Linguistics (www.skase.sk) 2: 2639. Newmeyer, Frederick 2006 Negation and modularity. In Drawing the boundaries of meaning: neo-gricean studies in pragmatics and semantics in honor of Laurence R. Horn, Betty Birner and Gregory Ward (eds.), 247268. Amsterdam: Benjamins. Newport, Elissa, Daphne Bavelier, and Helen Neville 2001 Critical Thinking about Critical Periods: Perspectives on a Critical Perion for Language Acquisition. In Language, Brain and Cognitive Development. Essays in Honor of Jacques Mehler, Emmanuel Dupoux (ed.), 481-502. Cambridge, Mass.: MIT Press. Nicolas, S. 2007 La localisation cérébrale des facultés mentales. Psychologie Française 52: 267-277. Nunes, Jairo 1999 Linearization of Chains and Phonetic Realization of Chain Links. In Working Minimalism, Samuel Epstein and Norbert Hornstein (eds.), 217-249. Cambridge, MA: MIT Press. Odden, David 1987 Kimatuumbi phrasal phonology. Phonology 4: 13-26.

762 References Odden, David 1990 Odden, David 1993 Oh, Mira 1995

Syntax, lexical rules and postlexical rules in Kimatuumbi. In The Phonology-Syntax Connection, Sharon Inkelas and Draga Zec (eds.), 259-277. Chicago: University of Chicago Press. Interaction between Modules in Lexical Phonology. In Studies in Lexical Phonology, Sharon Hargus and Ellen Kaisse (eds.), 111144. New York: Academic Press.

A prosodic analysis of nonderived-environment blocking. Journal of East Asian Linguistics 4: 261-279. Oostendorp, Marc van 1994 Affixation and integrity of syllable structure in Dutch. In Linguistics in the Netherlands 1994, Reineke Bok-Bennema and Crit Cremers (eds.), 151-162. Amsterdam: Benjamins. Oostendorp, Marc van 1999 Italian s-voicing and the structure of the phonological word. In Issues in phonological structure, S.J. Hannahs and Mike Davenport (eds.), 195-212. Amsterdam: Benjamins. Oostendorp, Marc van 2002 The phonological and morphological status of the Prosodic Word Adjunct. Linguistische Berichte, Sonderheft 11: 209-235. Oostendorp, Marc van 2004 Crossing morpheme boundaries in Dutch. Lingua 114: 1367-1400. Oostendorp, Marc van 2005 The Theory of Faithfulness. Ms., Meertens Instituut. Oostendorp, Marc van 2006a A theory of morphosyntactic colours. Ms., Meertens Instituut. Oostendorp, Marc van 2006b Transparent Morphology causes Phonological Opacity. Paper presented at GLOW 29, Barcelona 5-8 April. Oostendorp, Marc van 2007 Derived Environment Effects and Consistency of Exponence. In Freedom of Analysis?, Sylvia Blaho, Patrick Bye and Martin Krämer (eds.), 123-148. Berlin: Mouton deGruyter. Oostendorp, Marc van, and Merle Horne (eds.) 2005 Boundaries in intonational phonology. Special issue of Studia Linguistica 59, issues 2-3. Orgun, Cemil Orhan 1996a Sign-based morphology and phonology with special attention to Optimality Theory. Ph.D dissertation, University of California at Berkeley.

References 763 Orgun, Cemil Orhan 1996b Sign-Based Morphology: a declarative theory of phonologymorphology interleaving. Ms., University of San Diego. ROA #122 (abridged version published as Orgun 1999). Orgun, Cemil Orhan 1998 Cyclic and noncyclic phonological effects in a declarative grammar. In Yearbook of Morphology 1987, Geert Booij and Jaap van Marle (eds.), 179-218. Amsterdam: Kluwer. Orgun, Cemil Orhan 1999 Sign-Based Morphology: A Declarative Theory of PhonologyMorphology interleaving. In The Derivational Residue in Phonological Optimality Theory, Ben Hermans and Marc van Oostendorp (eds.), 247-268. Amsterdam: Benjamins. Orgun, Cemil Orhan, and Sharon Inkelas 2002 Reconsidering bracket erasure. In Yearbook of Morphology 2001, Geert Booij and Jaap van Marle (eds.), 115-146. Dordrecht & London: Kluwer. Orr, Carolyn 1962 Ecuador Quichua phonology. In Studies in Ecuadorian Indian languages I, B. Elson (ed.), 60-77. Norman, Oklahoma: SIL. Pagliano, Claudine 2003 L'épenthèse consonantique en français. Ce que la syntaxe, la sémantique et la morphologie peuvent faire à la phonologie. Ph.D dissertation, Université de Nice. Pak, Majorie 2008 The postsyntactic derivation and its phonological reflexes. Ph.D dissertation, University of Pennsylvania. Pater, Joe 2000 Nonuniformity in English stress: the role of ranked and lexically specific constraints. Phonology 17: 237-274. Pater, Joe 2009 Morpheme-specific phonology: constraint indexation and inconsistency resolution. In Phonological argumentation: essays on evidence and motivation, Steve Parker (ed.), 123-154. London: Equinox. Peperkamp, Sharon 1995 Prosodic constraints in the derivational morphology of Italian. Yearbook of Morphology 1994: 207-244. Peperkamp, Sharon 1996 On the prosodic representation of clitics. In Interfaces in Phonology, Ursula Kleinhenz (ed.), 102-127. Berlin: Akademie Verlag. Peperkamp, Sharon 1997 Prosodic Words. The Hague: Holland Academic Graphics.

764 References Peretz, Isabelle 2001 The Biological Foundations of Music. In Language, Brain and Cognitive Development. Essays in Honor of Jacques Mehler, Emmanuel Dupoux (ed.), 435-445. Cambridge, Mass.: MIT Press. Pesetsky, David 1979 Russian Morphology and Lexical Theory. Ms, MIT. Available at http://web.mit.edu/linguistics/www/pesetsky/russmorph.pdf. Pierrehumbert, Janet 1980 The phonology and phonetics of English intonation. Ph.D dissertation, MIT. Pierrehumbert, Janet 2000 Tonal Elements and their Alignment. In The interaction of constraints on prosodic phrasing. Prosody: Theory and Experiments, Merle Horne (ed.), 11-36. Dordrecht: Kluwer. Piggott, Glyne 1999 At the right edge of words. The Linguistic Review 16: 143–185. Piggott, Glyne 2007 Deriving word minimality by phase. Ms., McGill University. Piggott, Glyne, and Heather Newell 2006 Syllabification and the spell-out of phases in Ojibwa words. McGill Working Papers in Linguistics 20: 39-64. Pike, Kenneth 1947 Grammatical prerequisites to phonemic analysis. Word 3: 155172. Pike, Kenneth 1952 More on Grammatical Prerequisites. Word 8: 106-121. Pinker, Steven 1997 How the mind works. New York: Norton. Pinker, Steven, and Ray Jackendoff 2005a The faculty of language: what's special about it ? Cognition 95: 201-236. Pinker, Steven, and Ray Jackendoff 2005b The nature of the language faculty and its implications for the evolution of language (Reply to Fitch, Hauser and Chomsky). Cognition 97: 211-225. Pinker, Steven, and Jacques Mehler (eds.) 1988 Connections and Symbols. Cambridge, Mass.: MIT Press. Plag, Ingo, and Harald Baayen 2009 Suffix ordering and morphological processing. Language 85: 109152. Plaut, David 2003 Connectionist Modeling of Language: Examples and Implications. In Mind, Brain and Language, M. Banich and M. Mack (eds.), 143-167. Hillsdale: Erlbaum.

References 765 Ploch, Stefan 1996 Ploch, Stefan 2003

The Role of Parsing. SOAS Working Papers in Linguistics and Phonetics 6: 76-105. 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.), 73-116. Amsterdam & Philadelphia: Benjamins.

Plotkin, Henry 1998 Evolution in mind. Cambridge, Mass.: Harvard University Press. Poeppel, David, and David Embick 2005 Defining the relation between linguistics and neuroscience. In Twenty-first century psycholinguistics: Four cornerstones, Anne Cutler (ed.), 103-118. Mahwah, NY: Erlbaum. Poeppel, David, and Gregory Hickok 2004 Towards a new functional anatomy of language. Cognition 92: 112. Poeppel, David, William Idsardi, and Virginie van Wassenhove 2008 Speech perception at the interface of neurobiology and linguistics. Philosophical Transactions of the Royal Society 363: 1071-1086. Pollard, Carl, and Ivan Sag 1994 Head-Driven Phrase Structure Grammar. Chicago: CSLI & University of Chicago Press. Poser, William 1986 Diyari stress, metrical structure assignment and the nature of metrical representation. Proceedings of the West Coast Conference on Formal Linguistics 5: 178-191. Poser, William 1989 The metrical foot in Diyari. Phonology 6: 117-148. Poser, William 1990 Word-Internal Phrase Boundary in Japanese. In The PhonologySyntax Connection, Sharon Inkelas and Draga Zec (eds.), 279287. Chicago: University of Chicago Press. Posner, Michael 1981 Chronometric explorations of mind. Hillsdale: Erlbaum. Posner, Michael 2001 Cognitive Neuroscience: The Synthesis of Mind and Brain. In Language, Brain and Cognitive Development. Essays in Honor of Jacques Mehler, Emmanuel Dupoux (ed.), 403-416. Cambridge, Mass.: MIT Press. Potter, Simeon 1962 Syllabic Juncture. In Proceedings of the Fourth International Congress of Phonetic Sciences, Antti Sovijärvi and Pentti Aalto (eds.), 728-730. The Hague: Mouton.

766 References Prince, Alan 1983 Prince, Alan 1985

Relating to the Grid. Linguistic Inquiry 14: 19-100.

Improving Tree Theory. In Proceedings of the 11th Annual Meeting of the Berkeley Linguistics Society, M. Niepokuj, M. van Clay, V. Nikiforidou and D. Feder (eds.), 471-490. Berkeley: University of California, Berkeley. Prince, Alan, and Paul Smolensky 1993 Optimality Theory. Constraint Interaction in Generative Grammar. Ms, Rutgers University, University of Colorado (ROA version August 2002). Prunet, Jean-François 1986 Spreading and Locality Domains in Phonology. Ph.D dissertation, McGill University at Montreal. Prunet, Jean-François 1987 Liaison and Nasalization in French. In Studies in Romance Languages, Carol Neidle and Rafael Nuñez Cedeño (eds.), 225235. Dordrecht: Foris. Pullum, Geoffrey, and Arnold Zwicky 1988 The syntax-phonology interface. In Linguistics: the Cambridge survey. Vol. I, Frederick Newmeyer (ed.), 255-280. Cambridge: Cambridge University Press. Pyle, Charles 1972 On Eliminating BM's. In Papers from the eighth regional meeting of the Chicago Linguistic Society, Paul Peranteau, Judith Levi and Gloria Phares (eds.), 516-532. Chicago: Chicago Linguistic Society. Pylyshyn, Zenon 1984 Computation and Cognition. Cambridge, Mass.: MIT Press. Pylyshyn, Zenon 1989a On computation and cognition: Toward a foundation of cognitive science : A response to the reviews by A.K. Mackworth and M.J. Stefik. Artificial Intelligence 38: 248-251. Pylyshyn, Zenon 1989b Computing in Cognitive Science. In Foundations of Cognitive Science, William Posner (ed.), 51-91. Cambridge, Mass.: MIT Press. Pylyshyn, Zenon 1999 What's in your Mind? In What is Cognitive Science?, Zenon Pylyshyn and Ernest Lepore (eds.), 1-25. Oxford: Blackwell. Pylyshyn, Zenon, and Ernest Lepore (eds.) 1999 What is Cognitive Science? Oxford: Blackwell.

References 767 Raffelsiefen, Renate 1996 Gaps in Word Formation. In Interfaces in Phonology, Ursula Kleinhenz (ed.), 194-209. Berlin: Akademie Verlag. Raimy, Eric 2000a The phonology and morphology of reduplication. Berlin: Mouton de Gruyter. Raimy, Eric 2000b Remarks on backcopying. Linguistic Inquiry 31: 541-552. Raimy, Eric 2003 Asymmetry and linearization in phonology. In Asymmetry in Grammar, Vol.2, Anna Maria Di Sciullo (ed.), 129-146. Amsterdam: Benjamins. Reber, Paul, and Larry Squire 1998 Encapsulation of Implicit and Explicit Memory in Sequence Learning. Journal of Cognitive Neuroscience 10: 248-263. Reiss, Charles 2000 Optimality Theory from a cognitive science perspective. The Linguistic Review 17: 291-301. Reiss, Charles 2008 Modularity in the "sound" domain: implications for the purview of Universal Grammar. In The Oxford Handbook of Linguistic Interfaces, Gillian Ramchand and Charles Reiss (eds.), 53-79. Oxford: OUP. WEB. Reiss, Charles forth Intermodular explanation in cognitive science: An example from phonology. In Pylyshyn Papers, Don Dedrick and Lana Trick (eds.). Cambridge, Mass.: MIT Press. Revithiadou, Anthi 2006 Prosodic filters on syntax: an interface account of second position clitics. Lingua 116: 79-111. Rhodes, Richard 1974 Non-phonetic Environments in Natural Phonology. In Papers from the parasession on Natural Phonology, A. Bruck, R. Fox and M. La Galy (eds.), 285-296. Chicago: Chicago Linguistic Society. Rice, Keren 1987 On defining the intonational phrase: evidence from Slave. Phonology Yearbook 4: 37-59. Rice, Keren 1990 Predicting Rule Domains in the Phrasal Phonology. In The Phonology-Syntax Connection, Sharon Inkelas and Draga Zec (eds.), 289-312. Chicago: Chicago University Press.

768 References Rice, Keren 1993

A Reexamination of the Feature [Sonorant]: The Status of "Sonorant Obstruents". Language 69: 308-344.

Richards, Marc 2004 Object Shift and Scrambling in North and West Germanic: a Case Study in Symmetrical Syntax. Ph.D dissertation, University of Cambridge. Richards, Marc 2007a Dynamic Linearization and the Shape of Phases. Linguistic Analysis 33: 209-237. Richards, Marc 2007b On Feature Inheritance: An Argument from the Phase Impenetrability Condition. Linguistic Inquiry 38: 563-572. Riemsdijk, Henk van 1978 A Case Study in Syntactic Markedness: Peter de Ridder. Rizzi, Luigi 1990 Relativized Minimality. Linguistic Inquiry Monograph 16. Cambridge, Mass.: MIT Press. Rizzi, Luigi 2005 Phase theory and the privilege of the root. In Organizing Grammar. Studies in Honor of Henk van Riemsdijk, Hans Broekhuis, Norbert Corver, Riny Huybregts, Ursula Kleinhenz and Jan Koster (eds.), 529-537. Berlin: Mouton de Gruyter. Rizzi, Luigi, and Leonardo Savoia 1993 Conditions on /u/ propagation in Southern Italian dialects: a locality parameter for phonosyntactic processes. In Syntactic theory and the dialects of Italy, Adriana Belletti (ed.), 252-318. Torino: Rosenberg & Sellier. Roeper, Thomas 2005 Chomsky's Remarks and the Transformationalist Hypothesis. In Handbook of English Word-Formation, Rochelle Lieber and P. Stekauer (eds.), 125-146. Dordrecht: Kluwer. Romportl, Simeon 1984 Úloha rázu a jeho ekvivalentů při signalizování předĕlu. Slovo a slovesnost 45: 104-119. Rotenberg, Joel 1978 The Syntax of Phonology. Ph.D dissertation, MIT. Rubach, Jerzy 1981 Cyclic Phonology and Palatalization in Polish and English. Warsaw: Wydawnictwa Universytetu Warszawskiego. Rubach, Jerzy 1984 Cyclic and Lexical Phonology: The Structure of Polish. Dordrecht: Foris.

References 769 Rubach, Jerzy 1985 Rubach, Jerzy 1986 Rubach, Jerzy 1990 Rubach, Jerzy 1993 Rubach, Jerzy 1996a Rubach, Jerzy 1996b Rubach, Jerzy 1997 Rubach, Jerzy 2000a Rubach, Jerzy 2000b Rubach, Jerzy 2003

Lexical Phonology: lexical and postlexical derivations. Phonology Yearbook 2: 157-172. Abstract vowels in three dimensional phonology: the yers. The Linguistic Review 5: 247-280. WEB. Final Devoicing and Cyclic Syllabification in German. Linguistic Inquiry 21: 79-94. The Lexical Phonology of Slovak. Oxford: Clarendon Press. Shortening and ambisyllabicity in English. Phonology 13: 197237. Nonsyllabic analysis of voice assimilation in Polish. Linguistic Inquiry 27: 69-110. Extrasyllabic consonants in Polish: Derivational Optimality Theory. In Derivations and Constraints in Phonology, Iggy Roca (ed.), 551-581. Oxford: Clarendon. Glide and Glottal Stop Insertion in Slavic Languages: A DOT Analysis. Linguistic Inquiry 31: 271-317. Backness switch in Russian. Phonology 17: 39-64.

Polish palatalization in derivational optimality theory. Lingua 113: 197-237. Rubach, Jerzy, and Geert Booij 1984 Morphological and prosodic domains in Lexical Phonology. Phonology Yearbook 1: 1-27. Rubach, Jerzy, and Geert Booij 1987 Postcyclic versus postlexical rules in Lexical Phonology. Linguistic Inquiry 18: 1-44. Rubach, Jerzy, and Geert Booij 1990 Edge of constituent effects in Polish. Natural Language and Linguistic Theory 8: 427-463. Rubach, Jerzy, and Geert Booij 2003 Lexical Phonology - Overview. In International Encyclopedia of Linguistics, William Frawley (ed.), 443-447. Second edition Oxford: Oxford University Press.

770 References Rucart, Pierre 2006

Morphologie gabaritique et interface phonosyntaxique. Aspects de la morphologie verbale en afar. Ph.D dissertation, Université Paris 7. Rumelhart, David 1989 The Architecture of Mind: A Connectionist Approach. In Foundations of Cognitive Science, Michael Posner (ed.), 133-159. Cambridge, Mass.: MIT Press. Rumelhart, David E., James L. McClelland, and the PDP Research Group (eds.) 1986 Parallel Distributed Processing: Exploration in the MicroStructure of Cognition. 2 vols. Cambridge, Mass.: MIT Press. Russell, Kevin 1997 Optimality Theory and Morphology. In Optimality Theory. An overview, Diana Archangeli and D. Langendoen (eds.), 102-133. Oxford: Blackwell. Russell, Kevin 1999 MOT: sketch of an OT approach to morphology. Ms., University of Manitoba. Sabbatini, Renato 1997 Phrenology: the History of Brain Localization. Brain & Mind 1: [electronic resource: www.cerebromente.org.br/n01/index_i.htm]. Sag, Ivan 1974 The Grassmann's Law Ordering Pseudoparadox. Linguistic Inquiry 5: 591-607. Saltarelli, Mario 1970 A Phonology of Italian in a Generative Grammar. The Hague: Mouton. Samuels, Bridget 2009a The structure of phonological theory. Ph.D dissertation, Harvard University. Samuels, Bridget 2009b The third factor in phonology. Biolinguistics 3: 355-382. Samuels, Bridget forth The topology of infixation and reduplication. The Linguistic Review 27. Samuels, Richard, Stephen Stich, and Patrice Tremoulet 1999 Rethinking Rationality: From Bleak Implications to Darwinian Modules. In What is Cognitive Science?, Zenon Pylyshyn and Ernest Lepore (eds.), 74-120. Oxford: Blackwell. Sandalo, Filomena, and Hubert Truckenbrodt 2002 Some notes on phonological phrasing in Brazilian Portuguese. MIT Working Papers in Linguistics 42: 285-310.

References 771 Sauerland, Uli 1998

Scope reconstruction without reconstruction. In Proceedings of WCCFL 17, K. Shahin, S. Blake and E. Kim (eds.), 582-596. Stanford: Center for the Study of Language and Information. Sauerland, Uli, and Paul Elbourne 2002 Total Reconstruction, PF Movement, and Derivational Order. Linguistic Inquiry 33: 283-319. Schane, Sanford 1984 The fundamentals of particle phonology. Phonology Yearbook 1: 129-155. Scheer, Tobias 2001 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. WEB. Scheer, Tobias 2003a The Key to Czech Vowel Length: Templates. In Investigations into Formal Slavic Linguistics, Petr Kosta, Joanna Blaszczak, Jens Frasek, Ljudmila Geist and Marzena Żygis (eds.), 97-118. Frankfurt am Main: Lang. WEB. Scheer, Tobias 2003b Structure and Process: Computation is not King. Paper presented at the conference From representations to constraints and from constraints to representations, Toulouse 9-11 July 2003. Scheer, Tobias 2004a A Lateral Theory of Phonology. Vol.1: What is CVCV, and why should it be? Berlin: Mouton de Gruyter. Scheer, Tobias 2004b En quoi la phonologie est vraiment différente. Corpus 3: 5-84. WEB. Scheer, Tobias 2004c O samohláskové délce při derivaci v češtinĕ. In Čeština univerzália a specifika 5, Zdeňka Hladká and Petr Karlík (eds.), 224-239. Praha: Lidové noviny. WEB. Scheer, Tobias 2005 We need a translator's office, but the buffer has to go: Direct Interface. Paper presented at the 36th Poznań Linguistic Meeting, Poznań 22-24 April. Scheer, Tobias 2006a Interface Dualism. Paper presented at the 37th Poznań Linguistic Meeting, Poznań 20-23 April.

772 References Scheer, Tobias 2006b The balance of representation and computation at the Interface. Paper presented at the 14th Manchester Phonology Meeting, Manchester 25-27 May. Scheer, Tobias 2007 On the Status of Word-Initial Clusters in Slavic (And Elsewhere). In Annual Workshop on Formal Approaches to Slavic Linguistics. The Toronto Meeting 2006, Richard Compton, Magdalena Goledzinowska and Ulyana Savchenko (eds.), 346-364. Ann Arbor: Michigan Slavic Publications. WEB. Scheer, Tobias 2008a Why the Prosodic Hierarchy is a diacritic and why the Interface must be Direct. In Sounds of Silence: Empty Elements in Syntax and Phonology, Jutta Hartmann, Veronika Hegedüs and Henk van Riemsdijk (eds.), 145-192. Amsterdam: Elsevier. WEB. Scheer, Tobias 2008b Syllabic and Trapped Consonants in (Western) Slavic: the Same but yet Different. In Formal Description of Slavic Languages: The Fifth Conference, Leipzig 2003, Gerhild Zybatow, Luka Szucsich, Uwe Junghanns and Roland Meyer (eds.), 149-167. Frankfurt am Main: Lang. WEB. Scheer, Tobias 2008c Spell out your Sister! In Proceedings of the 27th West Coast Conference on Formal Linguistics, Natasha Abner and Jason Bishop (eds.), 379-387. Somerville: Cascadilla. WEB. Scheer, Tobias 2009a External sandhi: what the initial CV is initial of. Studi e Saggi Linguistici 47: 43-82. WEB. Scheer, Tobias 2009b Intermodular Argumentation and the Word-Spell-Out-Mystery. In Explorations of Phase Theory: Interpretation at the Interfaces, Kleanthes Grohmann (ed.), 23-65. Berlin: Mouton de Gruyter. WEB. Scheer, Tobias 2009c Representational and procedural sandhi killers: diagnostics, distribution, behaviour. In Czech in Formal Grammar, Mojmír Dočekal and Markéta Ziková (eds.), 155-174. München: Lincom. WEB. Scheer, Tobias 2009d Syllabic and trapped consonants in the light of branching onsets and licensing scales. In Studies in Formal Slavic Phonology, Morphology, Syntax, Semantics and Information Structure, Gerhild Zybatow, Uwe Junghanns, Denisa Lenertová and Petr Biskup (eds.), 411-426. Frankfurt am Main: Lang. WEB.

References 773 Scheer, Tobias 2010a What OT is, and what it is not. Review of The Cambridge Handbook of Phonology, ed. by Paul de Lacy. Journal of Linguistics 46: 193-218. WEB. Scheer, Tobias 2010b Intermodular argumentation: morpheme-specific phonologies are out of business in a phase-based architecture. In The Sound Pattern of Syntax, Nomi Shir and Lisa Rochman (eds.), 333-351. Oxford: OUP. Scheer, Tobias 2010c Review of Gussmann (2007) The Phonology of Polish. Studies in Polish Linguistics 5: 109-158. Scheer, Tobias forth Aspects of the development of generative phonology. In The Continuum Companion to Phonology, Bert Botma, Nancy Kula and Kuniya Nasukawa (eds.). London: Continuum. Scheer, Tobias, and Markéta Ziková forth The Coda Mirror v2. Acta Linguistica Hungarica. Schwartz, Jean-Luc, Christian Abry, Louis-Jean Boë, and Marie Cathiard 2002 Phonology in a Theory of Perception-for-Action-Control. In Phonetics, Phonology and Cognition, Jacques Durand and Bernard Laks (eds.), 254-280. Oxford: Oxford University Press. Scobbie, James 1991 Towards Declarative Phonology. Edinburgh Working Papers in Cognitive Science 7: 1-26. Scobbie, James, John Coleman, and Steven Bird 1996 Key aspects of declarative phonology. In Current Trends in Phonology: Models and Methods. Vol.2, Jacques Durand and Bernard Laks (eds.), 685-709. Salford, Manchester: ESRI. Segal, Gabriel 1996 The modularity of theory of mind. In Theories of Theories of Mind, P. Carruthers and P. Smith (eds.). Cambridge: CUP. Ségéral, Philippe, and Tobias Scheer 2001 La Coda-Miroir. Bulletin de la Société de Linguistique de Paris 96: 107-152. WEB. Ségéral, Philippe, and Tobias Scheer 2005 What lenition and fortition tells us about Gallo-Romance Muta cum Liquida. In Romance Languages and Linguistic Theory 2003, Twan Geerts, Ivo van Ginneken and Haike Jacobs (eds.), 235-267. Amsterdam: Benjamins. WEB.

774 References Ségéral, Philippe, and Tobias Scheer 2007 Le statut syllabique multiple des séquences muta cum liquida : l'exemple du gallo-roman. In Etudes sur le changement linguistique en français, Bernard Combettes, Christiane Marchello-Nizia and Sophie Prévost (eds.), 261-282. Nancy: Presses Universitaires de Nancy. WEB. Ségéral, Philippe, and Tobias Scheer 2008 The Coda Mirror, stress and positional parameters. In Lenition and Fortition, Joaquim Brandão de Carvalho, Tobias Scheer and Philippe Ségéral (eds.), 483-518. Berlin: Mouton de Gruyter. WEB. Seidl, Amanda 2001 Minimal Indirect Reference: a theory of the syntax-phonology interface. London: Routledge. Selkirk, Elisabeth 1972 The phrase phonology of English and French. Ph.D. dissertation MIT, published 1980 by Garland Press. Selkirk, Elisabeth 1974 French liaison and the X-bar notation. Linguistic Inquiry 5: 573590. Selkirk, Elisabeth 1978 The French foot: on the status of "mute" e. Studies in French Linguistics 1: 141-150. Selkirk, Elisabeth 1980a Prosodic Domains in Phonology: Sanskrit Revisited. In Juncture, Mark Aronoff and Mary-Louise Kean (eds.), 107-129. Saratoga: Anma Libri. Selkirk, Elisabeth 1980b The Role of Prosodic Categories in English Word Stress. Linguistic Inquiry 11: 563-605. Selkirk, Elisabeth 1981 On the nature of phonological representation. In The cognitive representation of speech, J. Anderson, J. Laver and T. Meyers (eds.), 379-388. Amsterdam: North Holland. Selkirk, Elisabeth 1981 [1978] On prosodic structure and its relation to syntactic structure. In Nordic Prosody II, Thorstein Fretheim (ed.), 111-140. Trondheim: TAPIR. Selkirk, Elisabeth 1984 Phonology and Syntax: The Relation between Sound and Structure. Cambridge, Mass.: MIT Press. Selkirk, Elisabeth 1986 On derived domains in sentence phonology. Phonology 3: 371405.

References 775 Selkirk, Elisabeth 1996 The prosodic structure of function words. In Signal to syntax: bootstrapping from syntax to grammar in early acquisition, James Morgan and Katherine Demuth (eds.), 187-213. Mahwah, NJ: Erlbaum. Selkirk, Elisabeth 2000 The interaction of constraints on prosodic phrasing. In Prosody: Theory and Experiments, Merle Horne (ed.), 231-261. Dordrecht: Kluwer. Selkirk, Elisabeth 2001 Phonological weight in sentence grammar: reexamining heavy noun phrase shift. Paper presented at the Workshop on prosody in processing, Utrecht 5-6 July. Selkirk, Elisabeth 2007 Focus prominence drives focus phrasing: Bengali intonation revisited. In Topic and focus: a cross-linguistic perspective, Chung-min Lee, Matthew Gordon and Daniel Büring (eds.), 217246. Dordrecht: Kluwer. Selkirk, Elisabeth 2008 Contrastive focus, givenness and the unmarked status of "discourse-new”. Acta Linguistica Hungarica 55: 331-346. Selkirk, Elisabeth, and Tong Shen 1990 Prosodic domains in Shanghai Chinese. In The Phonology-Syntax Connection, Sharon Inkelas and Draga Zec (eds.), 313-337. Chicago: University of Chicago Press. Selkirk, Elisabeth, and Koichi Tateishi 1988 Minor phrase formation in Japanese. Papers from the Annual Regional Meeting of the Chicago Linguistic Society 24: 316-336. Shallice, Tim 1988 From neuropsychology to mental structure. Cambridge: CUP. Siegel, Dorothy 1974 Topics in English Morphology. Ph.D. dissertation, MIT. Siegel, Dorothy 1980 Why there is no = Boundary. In Juncture, Mark Aronoff and Mary-Louise Kean (eds.), 131-134. Saratoga: Anma Libri. Simon, Herbert, and Craig Kaplan 1989 Foundations of Cognitive Science. In Foundations of Cognitive Science, Michael Posner (ed.), 1-47. Cambridge, Mass.: MIT Press. Smith, Jennifer 1997 Noun faithfulness: On the privileged behavior of nouns in phonology. Ms., University of Massachusetts, Amherst. ROA #242.

776 References Smith, Jennifer 1999 Noun faithfulness and accent in Fukuoka Japanese. In Proceedings of WCCFL 18, Sonya Bird, Andrew Carnie, Jason Haugen and Peter Norquest (eds.), 519-531. Somerville, MA: Cascadilla. Smith, Jennifer 2001 Lexical category and phonological contrast. In Papers in Experimental and Theoretical Linguistics 6: Workshop on the Lexicon in Phonetics and Phonology, Robert Kirchner, Joe Pater and Wolf Wikely (eds.), 61-72. Edmonton: University of Alberta. ROA #728. Smith, Neil 1998 Dissociations. Glot International 3.9: 9. Smith, Neil 2002 Modules, modals, maths and the mind. Glot International 6.8: 248-250. Smith, Neil 2003 Dissociation and modularity: Reflections on language and mind. In Mind, Brain and Language, M. Banich and M. Mack (eds.), 87111. Hillsdale: Erlbaum. Smith, Neil, and Ianthi-Maria Tsimpli 1991 Linguistic Modularity? A case study of a "savant" linguist. Lingua 84: 315-351. Smith, Neil, and Ianthi-Maria Tsimpli 1995 The mind of a savant. Language-learning and modularity. Oxford: Blackwell. Smith, Neil, and Ianthi-Maria Tsimpli 1999 Modules and quasi-modules: Language and Theory of Mind in a Polyglot Savant. Learning and Individual Differences 10: 193215. Smolensky, Paul 1987 Connectionist AI, symbolic AI, and the brain. Artificial Intelligence Review 1: 95-109. Smolensky, Paul 1988a On the proper treatment of connectionism. Brain and Behavioural Sciences 11: 1-74. Smolensky, Paul 1988b Putting Together Connectionism – again. Behavioral and Brain Sciences 11: 59-74. Smolensky, Paul 1991 Connectionism, constituency and the language of thought. In Meaning in Mind: Fodor and his critics, B. Loewer and G. Rey (eds.), 201-227. Oxford: Blackwell.

References 777 Smolensky, Paul 2003 Connectionism. In International Encyclopedia of Linguistics, William Frawley (ed.), 383-386. Oxford: OUP. Smolensky, Paul, and Geraldine Legendre 2006 The Harmonic Mind. From Neural Computation to OptimalityTheoretic Grammar, 2 Vols. Cambridge, Mass.: MIT Press. Sperber, Dan 1994 The modularity of thought and the epidemiology of representations. In Mapping the mind: Domain specificity in cognition and culture, L.A. Hirschfeld and S.A. Gelman (eds.), 39-67. Cambridge: CUP. Sperber, Dan 2001 In Defense of Massive Modularity. In Language, Brain and Cognitive Development, Emmanuel Dupoux (ed.), 47-57. Cambridge, Mass.: MIT Press. Sproat, Richard 1985 On deriving the lexicon. Ph.D dissertation, MIT. Sproat, Richard 1992 Morphology and Computation. Cambridge, MA: MIT Press. Sprouse, Ronald 1997 A case for enriched inputs. Ms., University of California at Berkeley. Squire, Larry 1987 Memory and brain. New York: OUP. Stampe, David 1972 How I Spent my Summer Vacation. Ph.D dissertation, University of Chicago. Reproduced by the Indiana Linguistics Club in 1979. Stanley, Richard 1969 The Phonology of the Navaho Verb. Ph.D dissertation, MIT. Stanley, Richard 1973 Boundaries in phonology. In A Festschrift for Morris Halle, Stephen Anderson and Paul Kiparsky (eds.), 185-206. New York: Holt, Rinehart & Winston. Starke, Michal 2004 On the inexistence of specifiers and the nature of heads. In Structures and Beyond. The Cartography of Syntactic Structures, Volume 3, Adriana Belletti (ed.), 252-268. Oxford: OUP. Stepanov, A. 2001 Late Adjunction and Minimalist Phrase Structure. Syntax 4: 94125. Steriade, Donca 1982 Greek Prosodies and the Nature of Syllabification. Ph.D dissertation, MIT.

778 References Steriade, Donca 1984 Glides and vowels in Romanian. In Proceedings of the 10th Annual Meeting of the Berkeley Linguistics Society, C. Brugman and M. Macaulay (eds.), 47-64. Berkeley: University of Berkeley. Steriade, Donca 1987 Redundant Values. In Papers from the 23rd Annual Regional Meeting of the Chicago Linguistic Society. Vol II, Parasession on Autosegmental and Metrical Phonology, Anna Bosch, Barbara Need and Eric Schiller (eds.), 339-362. Chicago: Chicago Linguistic Society. Steriade, Donca 1988 Greek Accent: A Case for Preserving Structure. Linguistic Inquiry 19: 271-314. Stevens, Alan 1980 Formative Boundary in Phonological Rules. In Juncture, Mark Aronoff and Mary-Louise Kean (eds.), 135-141. Saratoga: Anma Libri. Stevenson, Suzanne 1999 Bridging the Symbolic- Connectionist Gap in Language. In What is Cognitive Science, Ernest Lepore and Zenon Pylyshyn (eds.), 336-355. Oxford: Blackwell. Stich, Stephen 1972 Grammar, Psychology and Indeterminacy. Journal of Philosophy 69: 799-818. Stillings, Neil, Steven Weisler, Christopher Chase, Mark Feinstein, Jay Garfield, and Edwina Rissland 1995 Cognitive Science. An Introduction. Cambridge, Mass.: MIT Press. Stockwell, Robert, Donald Bowen, and I. Silva-Fuenzalida 1956 Spanish juncture and intonation. Language 32: 641-665. Strauss, Steven 1979 Against Boundary Distinctions in English Morphology. Linguistic Analysis 5: 387-419. Strycharczuk, Patrycja 2010 Phonetics, phonolgoy and Poznań /d/-voicing. In the Workshop on Positional phenomena in phonology and phonetics at GLOW 33. Wrocław. Svenonius, Peter 2001 On object shift, scrambling and the PIC. MIT Working Papers in Linguistics 39: 267-289. Svenonius, Peter 2004 On the edge. In Peripheries: syntactic edges and their effects, David Adger, Cécile de Cat and Georges Tsoulas (eds.), 259-287. Dordrecht: Kluwer.

References 779 Szendrői, Kriszta 2001 Focus and the Syntax-Phonology Interface. Ph.D dissertation, University College London. Szendrői, Kriszta 2003 A stress-based approach to the syntax of Hungarian focus. The Linguistic Review 20: 37-78. Szendrői, Kriszta 2004 A stress-based approach to climbing. In Verb clusters. A study of Hungarian, German and Dutch, Katalin É.Kiss and Henk van Riemsdijk (eds.), 205-233. Amsterdam: Benjamins. Szigetvári, Péter, and Tobias Scheer 2005 Unified representations for the syllable and stress. Phonology 22: 37-75. Szpyra, Jolanta 1987 Inputs to WFRs - phonological, intermediate or phonetic ? The case of verbs and deverbal nouns in Polish. In Rules and the Lexicon, Edmund Gussmann (ed.), 169-203. Lublin: Katolicki Universytet Lubelski. Szpyra, Jolanta 1989 The Phonology - Morphology Interface. London & New York: Routledge. Szymanek, Bogdan 1980 Phonological conditioning of word formation rules. Folia Linguistica 14: 413-425. Tager-Flusberg, Helen, and Kate Sullivan 2000 A componential view of theory of mind: evidence from Williams syndrome. Cognition 76: 59-89. Takahashi, Toyomi 1993 A farewell to constituency. UCL Working Papers in Linguistics 5: 375-410. Taylor, John 2002 Cognitive Grammar. Oxford: Oxford University Press. Thagard, Paul 2005 Mind. Introduction to Cognitive Science. Cambridge, Mass.: MIT Press. Trager, George 1962 Some Thoughts on 'Juncture'. Studies in Linguistics 16: 11-22. Trager, George 1972 Language and Languages. San Francisco: Chandler. Trager, George, and Bernard Bloch 1941 The syllabic phonemes of English. Language 17: 223-246. Trevian, Ives 2007 Stress-neutral endings in contemporary British English: an updated overview. Language Sciences 29: 426-450.

780 References Trnka, Bohumil 1940 Slovné a mezislovné signály v angličtinĕ, francouzštinĕ a češtinĕ. Listy Filologické 67: 223-232. Trubetzkoy, Nikolai Sergeyevich 1935 Anleitung zu phonologischen Beschreibungen. Brno: Edition du Cercle Linguistique de Prague. Trubetzkoy, Nikolai Sergeyevich 1936 Die phonologischen Grenzsignale. In Proceedings of the Second International Congress of Phonetic Sciences, Daniel Jones and D. B. Fry (eds.), 45-49. Cambridge: Cambridge University Press. Trubetzkoy, Nikolai Sergeyevich 1939 Grundzüge der Phonologie. 6th edition 1977, Göttingen: Vandenhoeck & Ruprecht. Truckenbrodt, Hubert 1995 Phonological phrases: their relation to syntax, focus and prominence. Ph.D dissertation, MIT. Truckenbrodt, Hubert 1999 On the Relation between Syntactic Phrases and Phonological Phrases. Linguistic Inquiry 30: 219-255. Truckenbrodt, Hubert 2007 The syntax-phonology interface. In The Cambridge Handbook of Phonology, Paul de Lacy (ed.), 435-456. Cambridge: CUP. Uriagereka, Juan 1999 Multiple spell-out. In Working Minimalism, Samuel Epstein and Norbert Hornstein (eds.), 251-282. Cambridge, Mass.: MIT Press. Vachek, Josef 1970 Some remarks on "juncture" in phonological analysis. In Proceedings of the sixth International Congress of Phonetic Sciences, Bohuslav Hála, Milan Romportl and Přemysl Janota (eds.), 963-965. Prague: Academia. van Wyhe, John 2004 Phrenology and the Origins of Victorian Scientific Naturalism. London: Ashgate. van Wyhe, John [no year] Phrenology: An Overview. The Victorian Web [www.victorianweb.org/science/phrenology/phrenologyov.html], June 2010. Vennemann, Theo 1974 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.

References 781 Vennemann, Theo 1976 [1971] Vowel alternations in English, German and Gothic: remarks on realism in phonology. In Linguistic and Literary Studies in Honor of Archibald A. Hill, vol. I: General and Theoretical Linguistics, Muhammad Ali Jazayery, Edgar C. Polomé and Werner Winter (eds.), 337-359. Lisse: de Ridder. Vogel, Irene 1982 La sillaba come unità fonologica. Bologna: Zanichelli. Vogel, Irene 1985 On constraining prosodic rules. In Advances in nonlinear phonology, Harry van der Hulst and Norval Smith (eds.), 217-233. Dordrecht: Foris. Vogel, Irene 1986 External sandhi rules operating between sentences. In Sandhi Phenomena in the Languages of Europe, Henning Andersen (ed.), 55-64. Berlin: Mouton de Gruyter. Vogel, Irene 1988 Prosodic constituents in Hungarian. In Certamen Phonologicum, Pier Marco Bertinetto and Michele Loporcaro (eds.), 231-250. Torino: Rosenberg & Sellier. Vogel, Irene 1990 The clitic group in Prosodic Phonology. In Grammar in Progress. Glow essays for Henk van Riemsdijk, Joan Mascaró and Marina Nespor (eds.), 447-454. Dordrecht: Foris. Vogel, Irene 1991 Level ordering in Italian Lexical Phonology? In Certamen Phonologicum II, Pier Marco Bertinetto, Michael Kenstowicz and Michele Loporcaro (eds.), 81-101. Torino: Sellier & Rosenberg. Vogel, Irene 1999 Subminimal constituents in prosodic phonology. In Issues in phonological structure, Stephen Hannahs and Mike Davenport (eds.), 249-267. Amsterdam: Benjamins. Vogel, Irene, and István Kenesei 1987 The interface between phonology and other components of grammar: the case of Hungarian. Phonology Yearbook 4: 243263. Vogel, Irene, and István Kenesei 1990 Syntax and semantics in phonology. In The Phonology-Syntax Connection, Sharon Inkelas and Draga Zec (eds.), 339-363. Chicago: University of Chicago Press. Wagner, Michael 2005a Prosody and Recursion. PhD. dissertation, MIT.

782 References Wagner, Michael 2005b Long-distance effects on prosody. Paper presented at CUNY sentence processing, University of Arizona 21 March - 2 April. Watson, Duane, and Edward Gibson 2004 The relationship between intonational phrasing and syntactic structure in language production. Language and Cognitive Processes 19: 713-755. Watson, Duane, and Edward Gibson 2005 Intonational phrasing and constituency in language production and comprehension. Studia Linguistica 59: 279-. Wauquier, Sophie 2005 Statut des représentations phonologiques en acquisition, traitement de la parole continue et dysphasie développementale. Habilitation thesis, EHESS Paris. Wells, Rulon S. 1947 Immediate Constituents. Language 23: 81-117. Wells, Rulon S. 1949 Automatic alternations. Language 25: 99-116. Welmers, W. E. 1947 Hints from morphology for phonemic analysis. Studies in Linguistics 5: 91-100. Wexler, Kenneth, and Peter Culicover 1980 Formal Principles of Language Acquisition. Cambridge, MA.: MIT Press. Wheeler, M. W. 1986 Catalan sandhi phenomena. In Sandhi phenomena in the languages of Europe, Henning Andersen (ed.), 475-488. Berlin: Mouton de Gruyter. Wilkinson, Robert W. 1974 Tense/Lax vowel harmony in Telugu: the influence of derived contrast on rule application. Linguistic Inquiry 5: 251-270. Williams, Edwin 1981 On the notions "Lexically Related" and "Head of a Word". Linguistic Inquiry 12: 245-274. Williams, Edwin 2003 Representation Theory. Cambridge, Mass.: MIT Press. Williams, Edwin 2007 Dumping Lexicalism. In The Oxford Handbook of Linguistic Interfaces, Gillian Ramchand and Charles Reiss (eds.), 353-381. Oxford: OUP. Wilson, Colin 2000 Targeted constraints: an approach to to contextual neutralization in Optimality Theory. Ph.D dissertation, Johns Hopkins University at Baltimore.

References 783 Wilson, Colin 2001

Consonant cluster neutralization and targeted constraints. Phonology 18: 147-197.

Wolf, Matthew 2008 Optimal Interleaving: Serial Phonology-Morphology Interaction in a Constraint-Based Model. Ph.D dissertation, University of Massachusetts. Yip, Moira 1998 Identity Avoidance in Phonology and Morphology. In Morphology and its Relation to Phonology and Syntax, Steven Lapointe, Diane Brentari and Patrick Farrell (eds.), 216-246. Stanford: CSLI Publications. Yu, Alan 2000 Stress assignment in Tohono O'odham. Phonology 17: 117-135. Zec, Draga 1988 Sonority constraints on prosodic structure. Ph.D dissertation, Stanford University. Zec, Draga 1993 Rule domains and phonological change. In Studies in Lexical Phonology, Sharon Hargus and Ellen Kaisse (eds.), 365-405. New York: Academic Press. Ziková, Markéta 2008 Alternace e-nula v současné češtině. Autosegmentální laterální analýza. Ph.D dissertation, Masarykova univerzita v Brně. Zuraw, Kie 2007 The Role of Phonetic Knowledge in Phonological Patterning: Corpus and Survey Evidence from Tagalog Infixation. Language 83: 277-316. Zwicky, Arnold, and Geoffrey Pullum 1986a The Principle of Phonology-free Syntax: introductory remarks. Ohio State University Working Papers in Linguistics 32: 63-91. Zwicky, Arnold, and Geoffrey Pullum 1986b Two spurious counterexamples to the Principle of Phonology-Free Syntax. Ohio State University Working Papers in Linguistics 32: 92-99.

864

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. – 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 "anti-cyclicity" of the entry "Optimality Theory" identifies as "→ Optimality Theory/ anti-cyclicity ". – 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 they occur: "note 258 (415)" refers to footnote 258 which occurs in §415. A absolute neutralisation 184 abstractness debate absolute neutralisation (SPE: nightingale) 184 and derived environment effects 184 and Lexical Phonology 144 conditions on the relationship between underlying and surface forms 184 electri[k] vs. electri[s]ity etc. 570 free ride (SPE: Pamela) 184 acquisition and lexical access (Kaye 1995) 358 activation value of a neuron neural network 595 active →memory addressing space lexical access (Kaye 1995) 350 adjuncts, morphological Kaye (1995): independent spell-out of terminal 318

affix class-based phenomena testing ground for theories of the interface: summary of the English fact 311 are independent from derived environment effect 838 organisation of →underapplication comparison Lexical Phonology, Halle & Vergnaud, Kay 303 analysis in Lexical Phonology: strata, level orderin 147, 150, 163, 166 analysis by Fabb (1988) and Hay (2002): selectional properties of affixes, perception, parsing 247 analysis by Halle & Vergnaud (1987a) 248 analysis by Kaye (1995) with no lookback 279, 310 representational (SPE) vs. procedural (Lexical Phonology) management 106, 162

786 Subject index affix class-based phenomena (continued) representational analysis by Selkirk (1984) 540 analysis with co-phonologies 480 Distributed Morphology 557 →underapplication and hence affix class-based phenomena cannot be done 559 direct merge analysis not viable 561 distinct representations (vs. distinct computation elsewhere) 560 empirical coverage of the three competing theories (overall summary) 832 coverage by competing theories (summary) rule-blocking pattern 834 rule-triggering pattern 835 the rule-triggering pattern is not less real 836 interpretation-triggering affixes spell out their own node (Halle & Vergnaud 1987a) vs. their sister (Kaye 1995) 282 Kaye (1995): class 2 does, class 1 does not trigger interpretation (reverse of Halle & Vergnaud 1987a) 281 affix classes SPE 92 representational vs. Lexical Phonology's procedural management 106, 162 analysis: →affix class-based phenomena cross-linguistic reality 143 Romance vs. Germanic heritage 143 bound stems 142 called into doubt by dual membership, optionality, overgeneration 245 incidence on stress placement 142 vs. derived environment effects: orthogonal 182 affix ordering Siegel (1974): foundations of Lexical Phonology 142 strata-based account in Lexical Phonology 147

affix ordering (continued) Kaye (1995) 315, 556 turns out to be wrong 244 is wrong (summary) 671 analysis of strings that violate affix ordering (govern-ment-al) (summary) 834 affix, category-forming roots do not possess category (Distributed Morphology) 544 affix-triggered interpretation →interpretation-triggering affixes birth of the idea in →Halle & Vergnaud (1987a) 220, 226 in Kaye (1995), interpretationtriggering is a lexical property of affixes 273 Kaye (1995): affixes that determine the spell-out of their own body vs. of their sister 274 alignment (OT) constraint-based mapping 457 ALIGN and WRAP, interaction of creates parametric variation 460 origins (a development of Selkirk's 1986 end-based mapping), definition 395, 458 also competent for units of the Prosodic Hierarchy below the word level (→Generalized Template Theory) 448 allomorphy, suppletion in Distributed Morphology 571 Alternation Condition introduction 185 exceptions: automatic, low level, "phonetic" processes 186 empirically valid? 186 instrument for the reduction of abstractness 129 also →Revised Alternation Condition analogy →Output-Output analytic vs. non-analytic (domains, morphology, affixes) →Kaye (1995) 272 anti-cyclicity →Optimality Theory/ anti-cyclicity

Subject index 787 anti-interactionism in →Halle & Vergnaud in →Optimality Theory/ anti-cyclicity anti-lexicalism of Distributed Morphology 550, 552, 539 vs. lexicalism in phonology 570 antisymmetry (Kayne 1994) and linearisation 744 arboreal structure →deforestation Artificial Intelligence 596, 603 artificial neural networks connectionism 591 asymmetric PF-LF spell-out? 567 atom (nuclear physics) 592 automatic processes →Alternation condition/ exceptions of 186 autonomous morphology →syntax ≠ morphology 539 autonomous word →word/ autonomous unit

B balance of procedural and representational analyses too hard: this book gives up on the ambition 6 summary 748 relationship Lexical Phonology - Prosodic Phonology 429, 435, 440 competing procedural vs. representational analyses: McCawley (1968) 114 →procedural first 320 bare output conditions 728 barriers atomisation of phasehood marshalled by the anti-locality of movement 777 cyclic movement in GB 298 and →bounding nodes: forerunners of Phase Theory 679 Basic Beat Rules Selkirk (1984) 426 behaviourism opposed by the cognitive revolution 590

biolinguistics language is perfect 728 and language organ 627 and modularity 609 grammar reduces to morpho-syntax, PF and LF are neither languagenor species-specific 37, 639 Language Faculty due to one or two evolutionary innovations 633 third factor explanations 638 black box →Prosodic Phonology/ black box bound stems diagnostic for affix classes 142 boundary general introduction of the word "boundary" by Halle (1959) 78 logically possible effects: triggering, blocking, no 50 are diacritic and local 366 argument against: they are a diacritic 373 bankruptcy of: demonstration by Pyle (1972) 136 wrong predictions: boundaries are not segments (Pyle 1972) 136 boundary contrast, minimal pairs 118 distinguished by variable phonological permeability 118 external evidence (Basbøll) 122 socio-linguistic function: the less formal the style, the smaller their influence 112 boundary detectors (→parsing cues, →Kaye 1995) 265 disjunction in English: wordfinally and before class 2 affixes (sign, sign-ing2 vs. signature1) 94 non-diacritic boundaries exist: empty CV units 713 summary translation into a diacritic 695 there are no boundaries inside morphemes 667 transition boundaries > prosodic constituency 711

788 Subject index boundary (continued) boundary strength absent in SPE 89 boundaries ranked: McCawley (1968) 114 diagnostic for boundary abuse 121 in post-SPE and general overview 117 Prosodic Phonology: attitude of Selkirk (1980a) 373 SPE, Selkirk (1972), clusters of # 112 with structuralist juncture: Chomsky et al. (1956) 75 prefix vs. suffix 52 SPE [-segment] segments, no phonetic correlate 88 different types: +, =, # 89 = boundary without posterity 89 # boundaries and brackets not redundant 97 boundary-sensitive rules 92 invisibility of boundaries 136 unrestricted: no guidelines for use in SPE 120 boundary erasure 90 Natural Generative Phonology have a phonetic correlate 127, 669f in the middle of morphemes 127 failure to eliminate word boundary from P-rules 128 post-SPE boundary mutation rules (readjustment rules) 112 # = [-voice] (Lass 1971) 135, 713 against boundaries (Rotenberg 1978) 137 are unwarranted diacritics, partly eliminated by autosegmentalism 130 are phonological aliens: diacritics (post-SPE) 132 boundary zoo in post-SPE times 117 boundary zoo, attempts to restrict it 119

boundary (continued) post-SPE (continued) consequence of autosegmentalism: proceduralisation of boundaries 130 define domains of rule application (McCawley 1968) 114 frustration with, therefore →direct syntax 132, 138 if not segments, what then? 134 only boundaries, no consideration of procedural influence 138 segmental status called into doubt (McCawley 1968) 134 Typology and strength in post-SPE 116 boundary abuse diagnostics 121 in structuralism 69 in Natural Generative Phonology 127 fighting boundary abuse in postSPE 119 boundary economy boundaries are last resort (Basbøll) 122 elimination of + 123 Lexical Phonology eliminated by procedural management 162 target for elimination in Lexical Phonology and Prosodic Phonology, but not for the same reasons 367 Prosodic Phonology (non-)discussion of in Prosodic Phonology 369, 376 are the non-autosegmental version of the Prosodic Hierarchy 403 are the primitive currency for mapping: prosodic domains are built on the basis of boundary grouping 397 arguments against, made by Prosodic Phonology 369 from boundaries to domains: a historical choice that has gone almost unnoticed 365

Subject index 789 boundary (continued) Prosodic Phonology (continued) replaced by autosegmental domains 361 share all formal properties with prosodic constituents 404 the local baby was thrown out with the diacritic bathwater 373 bounding nodes atomisation of phasehood marshalled by the anti-locality of movement 777 (or cyclic nodes): movement in the Extended Standard Theory 298 and Barriers: forerunners of Phase Theory 679 bracket erasure in SPE 102 in Lexical Phonology ≠ in SPE 176 and →Praguian segregation in Lexical Phonology 175 and no look-back effect in Lexical Phonology 174 in Lexical Phonology: needed for the →rule-triggering pattern 168 Lexical Phonology: internal structure of words invisible to syntax 157 brackets and bracket erasure SPE vs. Lexical Phonology 168, 171 bracket erasure relation of Mohanan's version with other no look-back devices 292 bracketing paradoxes Rubach & Booij (1984): solution based on prosodic constituency 440 analysis by Halle & Vergnaud (1987a) 246 anti-interactionist ammunition (Halle & Vergnaud 1987a) 243 cannot be done by Kaye (1995) 315 with English un- 319 brackets SPE: represent morpho-syntactic structure 96 status in SPE (summary) 675 SPE: brackets and # boundaries not redundant 97

brackets (continued) labelled brackets in SPE, labelsensitive rules 96 and interactionism 98 brackets vs. interactionism (summary) 673 bracket erasure: SPE vs. Lexical Phonology 168, 171 Lexical Phonology 165 elimination by interactionism 161 are boundaries through the back door 172 relation of Mohanan's bracket erasure with other no lookback devices 292 Mohanan: competitor of the SCCK 203 needed for the rule-triggering pattern 168 undo what was gained by interactionism 170 how to distinguish class 1 and class 2 boundaries 173 bracket-sensitive rules can do derived environment effects 202 bracket-sensitive rules (Mohanan) 168 reintroduction in Halle & Vergnaud (1987a) 219 in Kaye (1995): do not exist, only shorthand for the linear notation 275 brain mind and brain 588 localisation of cognitive functions in the brain 614 buffer →Prosodic Phonology/ buffer

C calendrical calculation double dissociation, savants 621 category roots do not possess category (Distributed Morphology) 544

790 Subject index category-sensitive phonology récord (noun) vs. recórd (verb) etc. 438 c-command in phonology 705 linearisation 745 Central Dogma (J. Kaye) all related forms are derived from the same UR 126 central system (Fodor) modularity 605f not the result of adaptive evolution 633 chunk-specific phonologies →computational systems, →no lookback devices/ PIC à la carte (chunk-specific), →Phase Impenetrability circuits Poeppel: levels intermediate between symbolic representations and neurobiology 616 clean syntax, dirty phonology unloading displeasing syntactic phenomena into PF 575, 580 summary 727 Clitic Group abandoned in Prosodic Phonology, note 96 (382) clitic placement outsourced into PF 730 cognitive functions higher (horizontal) and lower (vertical), how are they identified? 605 localisation in the brain 614 Cognitive Neuroscience synthesis of mind and brain 617 cognitive realism cognitive revolution 590, 624 cognitive resources →memory (active) morpheme identification: different paths to identification are more or less costly 357f cognitive revolution of the 50s-60s 590

Cognitive Science standard model 590, 603 symbolic vs. anti-symbolic views of the cognitive system 591 levels of representation 590 mind-brain relationship 615f modularity in 40 language of thought 591 generative linguistics, the spearhead of von Neumann-Turing computation in language 624 connectionist roots of OT 529 symbolic representations in OT 529 "Cognitive" Grammar linguistic incarnation of empiricist thought 591, 596 cohering vs. non-cohering affixes →affix classes, note 32 (142) Coloured Containment (Oostendorp) 503 monostratal = representational, antiderivational orientation 510 faithfulness between morphology and phonology 504 phonological action restricted by morphological affiliation 508 recoverability constraints 508 GEN marshalled by Consistency of Exponence 507 information transmission: representational (colours) vs. procedural (mapping) 506 old vs. new (strings, properties) 505f, 509 opacity 508 derived environment effects 509 prosodic constituency, double mapping 511 cómparable vs. compárable affix classes: dual membership 245 analysis in Distributed Morphology: direct vs. indirect merge 545 Comparative Markedness (McCarthy 2003a) used for derived environment effects in OT 519 compensation vs. condensation → condensation vs. compensation

Subject index 791 competence vs. performance 590 complexity, computational →computational complexity compositionality, non-compositionality LF opacity (Distributed Morphology) 545 computation in phonological theory 269 von Neumann-Turing model 603 Fodor's faculty psychology married with computation theory (Turingvon Neumann) 603 Parallel Distributed Processing (PDP) vs. serial 597 all-purpose (content-free) vs. specialised 597f modularity: computation is computation of something (domainspecificity) 598 connectionism vs. modularity 597f generative linguistics, the spearhead of von Neumann-Turing computation in language 624 intalterability of phonological computation predicted by modularity 659 morpho-syntactic information cannot modify the content of phonological computation (phonological instructions) 659 SPE definition of phonology, a forerunner of Fodorian modules 613 argument for parallel computation (connectionism): serial computation is not quick enough 597 in Government Phonology 267 computational complexity unwarranted in minimalism: motivation for derivation by phase 305 computational economy →economy (computational) Computational Neuroscience Marr (1982) 603

computational systems summary and typology 234-36 →chunk-specific phonologies SPE: cyclic vs. word-level rules 105 specific word-level phonology undisputed 824 for morphemes vs. words 22 Kaye (1995) recognises specific word-level rules 338 chunk-specific phonologies (summary) 811 specific word-level phonology 812 Praguian segregation (specific sentence-level phonology) 815 word-level phonology: role of English stress 814 chunk-specific vs. morpheme-specific phonologies in Lexical Phonology 154, 234 in Distributed Morphology 535 summary 257 →no look-back devices/ PIC à la carte (chunk-specific) morpheme-specific phonologies vs. no look-back: a major front line 829 vs. no look-back (PIC): refereed in favour of the PIC 830 in OT 473 in OT: parallel vs. reranked incarnations 476 and →selective spell-out are incompatible 303 vs. selective spell-out, summary 303 rejected by Distributed Morphology 534 PIC à la carte = computational economy à la carte 780, 797, 826 vs. Praguian segregation 818-27

792 Subject index computational systems (continued) syntax = or ≠ morphology? Selkirk (1984) vs. Lexical Phonology 540 Sproat (1985), Halle & Vergnaud (1987a) 223 Distributed Morphology 534f, 537, 539 SPE: one single computational system due to a representational definition of words and affix classes 107 Lexical Phonology level 1 vs. level 2 149-51 morphology vs. syntax 155 word- (lexical) vs. sentence (postlexical) phonology 154f Halle & Vergnaud (1987a) no distinct lexical vs. postlexical phonology 248 only one computational system 219 cognitive functions, double dissociation 618 definition of "phonology": is phonological what is the output of phonological computation 804, 806 different systems for syntax and the PF movement-intermundia undermine DM 580 just one spell-out mechanism (morphology and syntax are the same computational system), or two (they are not)? 858 the interface and phonology are distinct (but OT scrambles them) (summary) 678 concat function →Kaye (1995), akin to Merge 268 conceptual device module interacting with grammar 629 evidence from double dissociation 642 condensation vs. compensation SPE 97 Halle & Vergnaud (1987a), Stress Erasure, Stress Copy 231 →Distributed Morphology connection weight (artificial) →neural networks 591, 595

connectionism 588 vs. →modularity two competing theories of how the cognitive system works 588 early days: complementary (a question of levels), rather than incompatible 599 mind - brain relationship 588, 614 is empiricist, monistic (philosophy), reductionist in kind 591 Parallel Distributed Processing (PDP) vs. serial computation 597 denies symbolic representations and hence the mind 591 no distinction between storage and computation (rule-list fallacy) 596 all-purpose (content-free) vs. specialised computation 597f argument for parallel computation: serial computation is not quick enough 597 and memory 596 linguistic implementation of the nondistinction between computation and storage: "Cognitive" Grammar, usage-based theories (Bybee) 596 connectionist roots of →OT 529 contributes to blur the modular contours of OT 678 Consistency of Exponence 507 Constraint Conjunction used for derived environment effects in OT (Łubowicz) 518 constraint-based (vs. rule-based) mapping in OT 457 containment →Coloured Containment (Oostendorp) convergence-based phasehood Svenonius (2001), note 178 (781) co-phonologies lexical material selects a specific computational system 478 analysis of affix class-based phenomena 480 comparison with Halle & Vergnaud (1987) 481

Subject index 793 co-phonologies (continued) cophonology proliferation problem 492 morpheme-specific mini-phonologies in OT: parallel vs. reranked incarnations 476 copy and delete clitic placement outsourced into PF 730 correspondence rules Representational Modularity (Jackendoff) 723 counter-cyclic merger Kaye (1995): independent spell-out of terminals 318 cross-word phonology →sandhi (external) cycle →phase theory definition in SPE: by major categories 103 terminological confusion/pitfalls 236 cyclic (vs. non-cyclic) affixes Halle & Vergnaud (1987a), →affixtriggered interpretation, note 32 (142), 226f cyclic constituent Halle & Vergnaud (1987a) 227 cyclic derivation (inside-out) introduction 5 procedural (interactionism, phases) or representational (brackets)? 675 invention by Chomsky et al. (1956) 80 unprecedented in structuralism 80 evidence for 686 implementation of: brackets vs. interactionism (summary) 673 cyclic derivation (interface) ≠ phonological computation 470 intimate relationship with English stress 102 cyclic spell-out of words (external sandhi): no phonological traces: →word-spell-out-mystery

cyclic derivation (inside-out) (continued) anti-cyclic phenomena: apply to a morpho-syntactically complex chunk, but not to the pieces that it is made of: →derived environment effects (word level)/ Cracow voicing (sentence level) 822, 840 overview and implementation in SPE 102 definition of cycles in SPE: by major categories 103 Lexical Phonology expressed by strata 147 definition of cycles in 160 cyclic (=lexical) vs. non-cyclic (=postlexical) rules 158 of words: yes in Kaye (1995), who follows SPE and Halle & Vergnaud (1987a) 285 and Indirect Reference: phonology applies to chunks, but does not select them 434 anti-cyclicity in OT 28, 468 (and elsewhere): overview 465 summary 686 cyclic linearisation linearity-driven movement (Fox & Pesetsky 2004) 744 cyclic movement (successive) linearity-driven (Fox & Pesetsky 2004) 744 cyclic movement (syntax) early modification-inhibiting →no look-back 298 cyclic nodes cyclic movement in the Extended Standard Theory 298 cyclic rules in SPE 105 in Lexical Phonology: interactionist application 158 in Halle & Vergnaud (1987a), comparison with SPE 219, 233 split into structure-building vs. structure-changing rules 193

794 Subject index cyclicity killers (OT) 454, 467, 475 →co-phonologies →indexed constraints enriched representations (Oostendorp) 500

D Darwinian adaptation is the mind (are modules) the result of Darwinian adaptation? 609 Declarative Phonology phonological affiliate of →HPSG 27 computation in 269 Sign-Based Morphology 513 deforestation 42 and recursion 805 lateral structure predicts the absence of recursion 46 trees make wrong predictions: there is no recursion in phonology 45 arboreal and lateral structure are incompatible 44 demarcative function of phonology →Grenzsignale (Trubetzkoy) →Kaye (1995), parsing cues Demi-Beat Alignment DBA Selkirk (1984) 426 Dependency Phonology Structural Analogy 705 derivation by phase general presentation 304, 771 history how it was introduced in the late 90s 305 emergence of, favoured by the abandon of deep structure and a return to the derivationalism of the 70s 305 motivated by minimalist economy 305 (interactionism) early cases of: Bresnan (1971) 308 early modification-inhibiting no look-back: Head Constraint (Riemsdijk 1978) 298 Head Constraint (Riemsdijk 1978) vs. PIC: interpretation and hence PF played no role in the 70s 298

derivation by phase extreme version of: spell-out-as-youmerge (Epstein et al. 1998) 305 violation of encapsulation? 632 edge of the phase 298 → interactionism modern implementation of →cyclic derivation no reference to the fact that this idea was born in phonology: →interactionism 305 →selective spell-out: invented by Halle & Vergnaud (1987a) 220 initiated by Halle & Vergnaud (1987a) 228 Chomsky's "spell-out and forget" too strong for phonology: two candidates for a weaker version 302 the PIC was generalised by Kaye (1995) 359 derivational syllable Guerssel & Lowenstamm (1990) 713 derivational vs. inflectional morphology contrast rejected by Distributed Morphology 546 derived environment effects 177, 516 theories of (summary) 837 an important phenomenon that is poorly understood 208 morphologically vs. phonologically derived 179 phonologically derived environment effect: Finnish t → s / __i 180 morphologically derived environment effect: Polish palatalisation 181 empirically valid? 186 vs. affix classes: orthogonal 182 anti-cyclic phenomena: apply to a morpho-syntactically complex chunk, but not to the pieces that it is made of: →derived environment effects (word level)/ Cracow voicing (sentence level) 822, 840 are independent from affix class-based phenomena 838

Subject index 795 derived environment effects 177, 516 (continued) at the sentence level: apply across word-, but not across morpheme boundaries (Cracow voicing) 822 and the abstractness debate 183 free rides (SPE: Pamela) 184 solution 1: Strict Cycle Condition (SCC) 188 solution 2: lexical contrast (Kiparsky 1993) 196 solution 3: bracket-sensitive rules (Mohanan 1982) 201 solution 4: a non-linguistic phenomenon (Anderson 1981) 204 genesis of: acquisitional scenario (Anderson 1981) 206 a scenario for their genesis is not good enough 206 →Strict Cycle Condition (SCC) SCC-K and its violation only produced by 1) obligatory and 2) neutralising rules: why? 199 Kiparsky's Alternation Condition 185 Revised Alternation Condition (RAC) 187 Revised Alternation Condition falsified by Anderson's (1981) Fultonians 205 automatic, low level processes are postlexical 189, 192 automatic, low level, "phonetic" processes 186 derived environment created by structure-building rules? 193 derived environment sensitivity à la carte according to individual processes can be done by bracket-sensitive rules 202, 210 different lexical representations of identical segments (Kiparsky 1993) 198 Kiparsky's (1993) solution ≈ SPE 209

derived environment effects 177, 516 (continued) SCC-K and its violation (continued) SCC and Kiparsky's (1993) lexical option not viable 209 posterity of Kiparsky's solution based on lexical contrast 200 in Kaye (1995): a separate issue, no relation with no look-back 284 in OT 516 Comparative Markedness (McCarthy 2003a) 519 Constraint Conjunction (Łubowicz) 518 co-phonologies (Yu 2000) 522 revival of the Elsewhere Condition (Cho 2008) 521 Root Faithfulness (→direct syntax, Anttila) 520 in Coloured Containment (Oostendorp) 509 desymmetrisation →linearisation: Richards (2004, 2007) 745 developmental psychology →encapsulation (informational) 612 diacritics →Prosodic Hierarchy/ is a diacritic (from the PrWd upwards) →Prosodic Phonology/ domains are necessarily diacritic definition 405 do not qualify 44 birth and variable incarnation of (summary) 667, 694 diacritics sold as "true phonological objects" variable camouflage: juncture phonemes and boundaries (summary) 695 various degrees of awareness that items are diacritics 697 settled in verb, but not in practice: No Diacritics! 690 illusion of a diacritic-free phonology since the 80s 698

796 Subject index diacritics (continued) history born from the abandon of Level Independence 696 Trubetzkoyan Grenzsignale: immaterial, hence not diacritic 58 diacritic definition of boundaries due to the absence of phonemic status of juncture (Chomsky et al. 1956) 78 the status of boundaries as segments called into doubt (McCawley 1968) 134 critique by Rotenberg (1978) 137 boundaries are phonological aliens: diacritics (post-SPE) 132 partly eliminated by autosegmentalism 130 alternative translation into non-diacritic items, a structuralist invention 72 →direct syntax 132 →Direct Interface Direct Effect 685 Direct Interface overview 2, 5, 49 summary of issues relevant for 718 output of translation must be vocabulary items that are used in phonology in absence of extraphonological factors 655 diacritics make no predictions 58 translation into truly phonological objects: logical option not pursued in post-SPE 134 logical option, but explored only by Lass (1971) 134, 138 direct merge (to the root) source of opacity in →Distributed Morphology 543 vs. selective spell-out (summary) 582 is pointless in a lexicalist perspective: cómparable - compárable 573 direct syntax alternative to translation 702 alternative to diacritic boundaries 132

direct syntax (continued) and privativity 759 introduction of this option by Kenstowicz & Kisseberth 132 early and erratic proposals in post-SPE (Hyman, Pyle, Rotenberg) 123 out of frustration with boundaries 132, 138 way chosen when boundaries were found to be bankrupt (post-SPE) 137 vs. Prosodic Phonology & Indirect Reference 407 in OT 525 interface constraints in OT 494 Root Faithfulness in OT 520 in Sign-Based Morphology (Orgun) 512 in Coloured Containment (Oostendorp) 503 in Distributed Morphology 580 directed graphs (Raimy) linearisation in phonology? 746 loop-containing representations are called phonology in 2003, but "morphophonology" later on 740, 746 directionality of communication scope of the book: from morphosyntax to phonology 25 discovery procedure fieldwork-based method in structuralism 62 dissociation (double) →double dissociation Distributed Morphology 531 overall picture 532 summary properties and problems 581 a more complete theory of the interface: LF and PF effects 542 general architecture 536 internal structure of PF: the "PF branch" 734 late insertion 536 vocabulary (items) 536 no position taken regarding the representational communication with phonology 585

Subject index 797 Distributed Morphology 531 (continued) general architecture 536 (continued) agnostic with respect to multiple computational systems in phonology 535 DM is a strong modularity offender: phonology interleaved with non-phonological PF operations (linearisation) 739 idiom (phrasal), note 136 (536), 544 word-formation based on roots vs. on existing words (transport-at-ion derived from *transport-ate) 550 →PF movement allomorphy, suppletion 571 no rationale for deciding between allomorphy and phonological activity 572 vs. Lexical Phonology 533 anti-lexicalism 550, 569, 552 radically anti-lexicalist, but allomorphy/suppletion and lexicalist analyses are possible 573 rejects morpheme-specific miniphonologies 534 rejects the Lexicon of Lexical Phonology 534f rejects the two-place approach of Lexical Phonology 534f reproduces lexicalism with antilexicalist tools 552 resumes Halle & Vergnaud's (1987a) attack against Lexical Phonology 534 root-based vs. word-based morphology in a single engine perspective 546 single engine hypothesis: →morphology = syntax 534f, 537

Distributed Morphology 531 (continued) does not live up to its single engine ambition (syntax = morphology) 538, 584 different computational systems for syntax and the PF movement-intermundia undermine DM 580 different distribution of phasehood below and above the word 559 →derivation by phase subscribes to →interactionism 534 →Phase Impenetrability 555 is forced into PIC à la carte: the PIC does not apply to primary stress in English 555 problem with phasehood and the PIC: DM has a peculiar way to do selective spell-out (functional vs. non-functional heads) 558f two possible origins of phonological opacity: direct merge and →Phase Impenetrability 542 affix class-based phenomena distinct representations (vs. distinct computation elsewhere) 560 underapplication and hence affix class-based phenomena cannot be done 559 every xP is a phase head 544, 551f, 558f spell-out at every xP is much too strong a claim if Phase Impenetrability is applied 555 direct merge and opacity 541 tree-geometry is the origin of opacity 542, 544, 546 idiosyncrasy is a property of "lower" items in the tree 546 opacity = (direct) merge to a root, transparency = merge to an xP 544

798 Subject index Distributed Morphology 531 (continued) direct merge and opacity 541 (continued) direct merge may, but does not have to produce opacity 546, 553 condensation vs. compensation, SPE's classical take: two different suffixes 548 condensation vs. compensation, analysis by Halle & Kenstowicz (1991) 549 condensation vs. compensation, analysis by Marvin (2002) 550f condensation vs. compensation, existence of two groups of words with contrasting behaviour 549 in a lexicalist perspective there is no opacity in cómparable compárable at all 573 PF-LF opacity asymmetric PF-LF spell-out? 567 combinations of PF - LF opacity: neither the PIC nor direct merge can account for all patterns 566 concordance: cómparable vs. compárable 545 all combinations attested 565 maybe not the same thing 562 domain assignment Lexical Phonology: a given rule applies only at a given stratum 151 domain juncture-, domain limit-, domain span rules Prosodic Phonology 384, 709 domain specificity modularity 611 identifies modules 613 diagnostic for modular status of GBsubtheories 632 diagnostic for language-internal modular structure 643-46 in early generative grammar: LSLT (Chomsky 1955-56) 625 syntax, morphology, semantics vs. phonology 413

domain structure (→Kaye 1995) betrayed by parsing cues (consonant clusters) 337, 339 betrayed by vowel (non-)reduction 337 consequences of for stress and vowel reduction (English) 334 contrast between [[X] [Y]] and [[X] Y] 337 contrast between [[X] [Y]] and [X Y] 336f diachronic erosion of 339 dialectal and idiolectal variation due to variable domain structure 335 domain-based carriers vs. local carriers →local vs. domain-based carriers domain-final condition on rule application: Kaye (1995), rule-triggering pattern 322 reference to the domain-final location: no violation of modularity 323 domain-final empty nuclei in Government Phonology 329 from word-final to domain-final empty nuclei 331 detected by parsing cues 344 several in a row 332 are phase-initial and special for that reason 333 are subject to no look-back: licensed also on later cycles 332 Kaye's (1995) analysis of nasal assimilation (un- vs. in-) 314 Kaye's (1995) analysis of nasal cluster simplification 324 domains hierarchical structure, built on juncture (Chomsky et al. 1956) 75 of rule application, defined by boundaries (McCawley 1968) 114 domains (prosodic) →Prosodic Phonology DOT →Stratal OT 483 double dissociation 618 modularity vs. connectionism: different predictions 619 identifies modules 613 of language 621

Subject index 799 double dissociation 618 (continued) identification of language-related modules 642 face recognition, number sense 620 diagnostic for modular status of GBsubtheories 632 applied to Praguian segregation 818 dualism vs. monism 591

E economy phoneme economy 60, 65 generative ideas in a structuralist guise (Chomsky et al. 1956) 74f motivation for →derivation by phase 305 economy (computational) in minimalism: is always obeyed 797 PIC à la carte = computational economy à la carte 780, 797, 826, 830 edge of the phase →phase edge edge phonology distinct from morpheme-internal phonology 667 edge-based mapping →end-based mapping ellipsis, sluicing outsourced into PF (non-pronunciation "at PF") 731 Elsewhere Condition derivation of Kiparsky's Strict Cycle Condition (SCC) by 191 revival in OT (derived environments, Cho 2008) 521 empiricism vs. rationalism 591 Empty Category Principle (ECP) Government Phonology 330, 344 Kaye's (1995) analysis of nasal assimilation (un- vs. in-) 314 empty nuclei →domain-final empty nuclei detected by parsing cues 343 identified as domain-final by parsing cues 344 that are string-final upon computation (= domain-final empty nuclei) 329 word-final 330

encapsulation (informational) core property of modules 612 enhanced language skills of Christopher (Smith & Tsimpli) 621 is called inclusiveness in syntax (Chomsky) 648 diagnostic for modular status of GBsubtheories 632 violated by nested modules (derivation by phase, GB-subtheories)? 632 end-based mapping (Selkirk 1986) 394 puts to use the technology of the 80s: X-bar-based mapping 395 no advance regarding the mapping puzzle 392, 396 ERPs (Event-Related Potentials) localisation of cognitive functions in the brain 617 escape hatch (→phase edge) cyclic movement in the Extended Standard Theory 298 evaluation measure derived environment effects 198 evolutionary psychology modularity 609 Language Faculty due to adaptive evolution? 633 exceptionless rules Natural Generative Phonology: P-rules (vs. MP-rules) 127 in Lexical Phonology: postlexical rules 156 exemplar-based theories 596 Extension Condition Chomsky (1995) 319 external sandhi →sandhi (external) →word-spell-out-mystery extra-linguistic factors in language design comparison Head Constraint (Riemsdijk 1978) and PIC 298 motivation for derivation by phase 305 minimalism and biolinguistics (third factor explanations) 633, 638f applied to phonology (Samuels 2009a,b) 639 extrametricality in Latin stress assignment 295

800 Subject index extraposition motivation for PF movement 575

F face recognition double dissociation 620 Faculty of Language →Language Faculty faculty psychology (Fodor 1983) modularity is the heir of phrenology 602 married with computation theory (Turing-von Neumann) 603 faithfulness Containment vs. Correspondence 505 final →domain-final final empty nuclei →domain-final empty nuclei fission Distributed Morphology 536, 538 FLB (Faculty of Language in the Broad sense) animal-based, includes PF and LF 633 does not include PF and LF 639 modularity 609 not the result of adaptive evolution 633 fMRI (functional Magnetic Resonance Imaging) localisation of cognitive functions in the brain 617 foot structure early modification-inhibiting no lookback 295 Free Element Condition (FEC) early modification-inhibiting no lookback (stress) 295-97 is weak no look-back: processspecificity, restriction to structure that is absent from the lexicon 296 Chomsky's "spell-out and forget" too strong for phonology: two candidates for a weaker version 302 free ride SPE 184f Fultonians why phonology isn't natural (Anderson 1981) 205

functional anatomy localisation of cognitive functions in the brain 616 strategy: levels intermediate between symbolic representations and neurobiology: connectionism, Poeppel's circuits 616 functional theories of the interface →Grenzsignale (Trubetzkoy) →Kaye (1995) & overview 263f fusion Distributed Morphology 536, 538

G Gall, Franz-Joseph founder of phrenology, origins of modularity 601f GB-subtheories are cognitive modules? Violate encapsulation? 632 introduced without reference to the modularity of the cognitive system 630 GB subtheories and their perception as cognitive modules 628, 635 against language-internal modularity (Hornstein 2009) 633 generalisations (empirical) morpho-syntax and melody are incommunicado 660 Generalized Template Theory OTed version of Prosodic Morphology 448 Generative Semantics 12 alternative to the inverted T model 721 did not follow "all concatenation before all interpretation" 359 return of the 60s 679 generativism and structuralism continuity or rupture? 80 government (and licensing) 43 Government Phonology 705 interconstituent government 343 Kaye's (1995) analysis of nasal assimilation (un- vs. in-) 314

Subject index 801 Government Phonology also →Kaye (1995) Standard GP a hybrid arboreal-lateral model 43 Magic Licensing, note 82 (344) Interonset Government, note 82 (344) projection principle 705 prosodic government 705 computation in the φ-function 267 and Harmonic Serialism 269 strict CV: strings are computed rightto-left 333 →government licensing 705 →domain-final empty nuclei →empty nuclei Structural Analogy 705 Grenzsignale alert listener, help parsing 55 cradle of the functional perspective 56 positive vs. negative 57 are immaterial and not diacritics 58 are under morpho-syntactic control 58 and the →mapping puzzle: no predictions made 58 grid →metrical grid

H Halle & Vergnaud (1987a) non-interactionist version of Lexical Phonology 215 relevant literature 221 summary 254 anti-interactionism comparison with classical LP 217f, 235 directed against LP 219 arguments against interactionism 223 anti-interactionist strategy: attack affix ordering 244 stratal vs. non-interactionist architecture, two testing grounds 251

Halle & Vergnaud (1987a) (continued) general architecture partial picture 235 complete picture 238 selective spell-out, a groundbreaking idea 220, 256 heralded by Halle & Mohanan (1985) 233 Distinct pre- vs. post-word phonology yes, Praguian segregation no 248 multiple computational systems 234 Praguian segregation dismissed 219 takes morpho-syntactic trees seriously 226f the root is always an interpretational unit (a cycle) 230 both spell-out and interpretation are cyclic 792 restoration of SPE 217, 219 cyclic interpretation of all chunk sizes 238 dismissal of Praguian segregation 238 interpretational units 242 pre-word allomorphy = readjustment 235 summary: more akin with SPE than with LP 255 affix-triggered interpretation percolation of lexical properties of affixes to the morpho-syntactic tree 227 cyclic vs. non-cyclic affixes 226 cyclic vs. non-cyclic rules 219, 233 interpretational units 242 interpretation-triggering affixes 220 proceduralisation of the interface 235 elimination of boundaries 235 →SCC-K 237 comparison with Kaye (1995) overview 286 roots are, words are not interpretational units (reverse of Kaye 1995) 280

802 Subject index Halle & Vergnaud (1987a) (continued) comparison with Kaye (1995) (continued) underapplication to class 2 strings 279 what exactly is spelled out 226 empirical coverage rule-blocking & -triggering pattern 248 →affix class-based phenomena analysis of rule-blocking pattern (level 1 rules) 227f unable to do the rule-triggering pattern 250 analysis of bracketing paradoxes 246 analysis of stress metrical grid 237 Stress Copy 231 Stress Erasure Convention 230 structure-changing vs. structurebuilding rules 237 Harmonic Serialism (OT-CC), note 68 (269), 686 Optimal Interleaving (Wolf 2008): revival of interactionism with one single constraint hierarchy that merges morphology and phonology 472 Harmonic Grammar (Smolensky & Legendre 2006) peaceful coexistence of connectionist and symbolic levels 599 Head Constraint (Riemsdijk 1978) early modification-inhibiting no lookback 298 vs. PIC: interpretation and hence PF played no role in the 70s 298 head parameter linearisation 743 revived in PF-based linearisation strategies 745 helix (double, biology) 592 historiographic cherry-picking 9 looking at the interface through the prism of its history produces different results 843

history instrument for independent assessment 35, 843 homomorphemic parallel modules are (Bendjaballah & Haiden) 724 HPSG 27 Sign-Based Morphology (Orgun) 513 hundred-step programme (Feldman 1985) argument for parallel computation (connectionism): serial computation is not quick enough 597

I idiom (phrasal), idiom(atic) expression in →Distributed Morphology, note 136 (536), 544 impoverishment →Distributed Morphology 536, 538 inclusiveness application of (informational) →encapsulation to syntax (Chomsky 1995a) 612, 648 indexed constraints morpheme-specific mini-phonologies in OT: parallel vs. reranked incarnations 476, 482 Indirect Reference →Prosodic Phonology Prosodic Phonology, but roots in SPEreadjustment 385 in SPE (cat-rat-cheese) 84 a bad reason: non-isomorphism 416 a good reason for (and against direct syntax): modularity 410 and Level Independence 415 modular argument in favour of, absent from the literature 414 domain-specificity: syntax, morphology, semantics vs. phonology 413 goal for further study: to make carriers of morpho-syntactic information local AND non-diacritic 366 inflectional vs. derivational morphology contrast rejected by →Distributed Morphology 546

Subject index 803 initial CV non-diacritic →boundaries exist 44, 49, 713 innateness of modules 627 inner vs. outer morphology →Distributed Morphology (Marantz) 546 inside-out interpretation →cyclic derivation interactionism a new idea introduced by Lexical Phonology 146 historical overview in syntax and phonology 14 in interface theories since SPE: summary 305, 676 SPE: all concatenation before all interpretation 86 early cases of: Bresnan (1971) 308 Distributed Morphology subscribes: derivation by phase 534 when syntax became interactionist: derivation by phase 679 enforced by the minimalist concern for cognitive resources (active memory) 305 Optimal Interleaving (Wolf 2008): interactionism in OT-CC with one single constraint hierarchy that merges morphology and phonology 472 no contradiction with antiderivationalism (summary) 678 anti-interactionism (→Halle & Vergnaud 1987a) summary 223, 677 anti-interactionist ammunition: bracketing paradoxes 243 anti-interactionist strategy: attack affix ordering 244 stratal vs. non-interactionist architecture testing ground 1: lexical phonology sensitive to syntactic information? 252 testing ground 2: phonology-free syntax 253

interactionism (continued) anti-interactionism (OT) →anti-cyclicity in →OT (summary) 686 and brackets 98 brackets in Lexical Phonology undo what was gained by interactionism 170 brackets vs. interactionism (summary) 673 and modularity reconciles cyclic derivation and modularity 161, 275 makes inside-out interpretation compatible with modularity (summary) 681 intermodular argument against anti-cyclicity in orthodox OT 856 interactive Odden (1993), →interactionism 223 interface conditions in minimalism 305 interface constraints OT, revival of direct syntax 495 appeal to SPE-type morphological diacritics 496 Interface Dualism both procedural and representational communication is needed 6, 844 parallel with representation vs. computation in phonology (Anderson 1985) 8 (summary) 684 why a representational channel is needed 685 why a procedural channel is needed 686 no procedural communication in pregenerative times 80 idea introduced in SPE 83 post-SPE: only boundaries, no consideration of procedural influence 138 the general plot was set by the end of the 70s 138 dismissed by Kaye (1995): no representational channel 276

804 Subject index Interface Dualism (continued) random use of both channels is current practice in (almost) all theories, but probably not the right way to go 750f balance of procedural and representational analyses: which channel for which phenomenon? 320, 749 morpho-syntactic structure may, but content (labels) may not bear on phonology (?) 752 privativity (representational and procedural) 756 interface events pre-theoretical classification 50 interface processors Representational Modularity (Jackendoff) 723 intermodular communication overview 17 translation 649 intermodular argumentation summary 846 can only be made through the procedural channel 850 conditions of: we need to know whether morphology and syntax are one and the same or two distinct computational systems 851 potential of interactionist derivation by phase: syntax as a referee for competing phonological theories 847 procedural first 320 three referees: the PIC, selective spellout and the phase edge 857 convergence of 6 syntactic and phonological mechanisms 853f interactionism 856 spell-out is selective in syntax, hence it must be selective in phonology as well 769, 776 import of phonology-based mechanisms into syntax: piece-driven phase and the spell-out of terminals 855

intermodular argumentation (continued) just one spell-out mechanism (morphology and syntax are the same computational system), or two (they are not)? 858 morpheme-specific mini-phonologies vs. no look-back (PIC): refereed in favour of the PIC 830 intermundia →PF Interonset Government →Government Phonology, note 82 (344) interpretational units in SPE: delineated by →brackets (→cycles) 103 SPE vs. Lexical Phonology, →phase theory (example origin-al-ity) 160 representational (SPE) vs. procedural (Lexical Phonology) definition 161 in Lexical Phonology: defined by →strata 160 Halle & Vergnaud (1987a) 226, 242 the root is always one (Halle & Vergnaud 1987) 230 Kaye (1995): defined by selective spell-out 273 Kaye (1995): domains 271 Kaye (1995): words are, roots are not (reverse of Halle & Vergnaud 1987a) 280 interpretation-neutral affixes →affix-triggered interpretation 226, 272f interpretation-triggering affixes →affix-triggered interpretation idea introduced by Halle & Vergnaud (1987a) 220, 226 Kaye (1995) 272f Kaye (1995): affixes determine the spell-out of their sister 274 spell out their mother (Halle & Vergnaud 1987a) vs. their sister (Kaye 1995) 282

Subject index 805 interpretation-triggering affixes (continued) two options: spell-out of the sister or of the mother 764 Kaye (1995): class 2 does, class 1 does not trigger (reverse of Halle & Vergnaud 1987a) 281 intonation 800 confusion: recursion is a phenomenon whose existence is established by pre-theoretical properties, not by analyses that happen to use recursive constructions 803 intonation may require recursive structure, but this structure is not built by phonological computation 801f is the calculus of intonation done by phonological computation? 801 is not phonological: phonological computation does not contribute anything (Cinque 1993) 806 absence of mutual conditioning with phonology (segmental, word stress etc.) 806 can be predicted without knowing which lexical material will be inserted 806 tone-based analysis (Pierrehumbert) 802 inverted T model general introduction 86 syntactico-centrism: privilege of concatenation 25 defines the scope of the book 12, 25 logically independent from "all concatenation before all interpretation" 86, 224 restoration of "all concatenation before all interpretation" by Halle & Vergnaud (1987a) 223 interface theories beyond the inverted T 26 (generative) alternatives: generative semantics and Jackendoff's parallel modules 29, 720 contrast with LSLT (Chomsky 195556) where all modules have the concatenative privilege 625

inverted T model (continued) SPE vs. interactionist Lexical Phonology 223 summary interactionist vs. noninteractionist implementations 305 in GB: subcomponents (inverted T) vs. subsystems (GB modules) 631 restored in minimalism after the GB interlude 641f untouched by the new biolinguistic delineation of grammar 639

J junctural minimal pairs night rate vs. nitrate 63 German (Moulton) 64 juncture (structuralism) is multifunctional (simplification): serves phonological and morphosyntactic purposes (Chomsky et al. 1956) 77 abuse of juncture in the middle of morphemes: summary 667 morpho-syntactic control over juncture restored: Chomsky et al. (1956) 77f phonetic correlate? debate in structuralism 70 no: Chomsky et al. (1956), SPE 78, 88 no: summary 668 open juncture 64 revival of in →Natural Generative Phonology 127 juncture phoneme motivation: →Level Independence 60 diacritic →translation (summary) 695 phoneme economy 60, 65 environment for predicting allophones 60 different names of (transition, disjuncture, schismeme etc.) 71 suprasegmental phoneme? 70 abuse: in the middle of morphemes? "accidental" coincidence with morpho-syntactic divisions 65 yes: Hill (1954) 69, 667 no: Chomsky et al. (1956) 78

806 Subject index K Kaye (1995) (written) sources are sparse for Kaye's theory 260 is modern: modification-inhibiting (freezing) →no look-back, →phase edge 359 →Interface Dualism dismissed: procedural-only 276 phonetic hypothesis rejected, note 66 (263) vowel reduction in English: follows SPE, secondary stress protects 336 restoration of SPE (like Halle & Vergnaud 1987a) 359 summary comparison with Halle & Vergnaud (1987a) 286 Licensing Constraints 269 functional and perception-oriented approach to the interface 263f why does phonology exist? Faster, more effective and reliable communication 263 demarcative function of stress 265 parsing cues definition 263 theory-dependent (GP): detection of empty nuclei 342 theory-independent (clusters) 341 undetected morpheme breaks 345 concat- and φ-function 267 concat function akin to Merge 268 concat- and φ-function interleaved contrast between [[X] [Y]] and [[X] Y] 337 contrast between [[X] [Y]] and [X Y] 336f φ-function general properties 269f no multiple mini-grammars 270 interface-relevant properties of 270

Kaye (1995) (continued) φ-function (continued) morpho-phonology: most is not done by the φ-function, i.e. lies outside of phonology 276 selective spell-out defines interpretational units 273 Kaye's vs. Halle & Vergnaud's implementation 277 independent spell-out of terminals: a problem 317 multiple mini-grammars (morphemespecific and chunk specific) 283 morpheme-specific phonologies and selective spell-out are incompatible 303 recognises specific word-level rules 338 cyclic structure both spell-out and interpretation are cyclic 792 calls on Chomsky's (1973) →Strict Cyclicity, but applies something else 300 cyclic spell-out of words: yes, following SPE and Halle & Vergnaud (1987a) 285 domain structure general introduction 266 is the result of selective spell-out 273 logical possibilities with two morphemes 271 analytic vs. non-analytic domains/morphology/affixes 272 brackets do not exist, they are only shorthand for the linear notation 275 affixes may be lexically defined as domains 274 domains that are not created by spell-out 274 diachronic erosion of domain structure (compounds) 339

Subject index 807 Kaye (1995) (continued) domain structure (continued) dialectal and idiolectal variation due to variable domain structure 335 interpretational units 271 interpretation-triggering affixes spell out their sister (rather than their own node as in Halle & Vergnaud (1987a) 282 nasal assimilation requires independent spell-out of terminals 314 independent spell-out of terminals 274 independent spell-out of terminals: morphological adjuncts and counter-cyclic merger 318 words are, roots are not interpretational units (reverse of Halle & Vergnaud 1987a) 280 lexical entries lexical entries are ordered according to phonological structure 347 full addressing space is created, including "empty" slots 350 possible word: blick vs. lbick 349 role of phonological relatedness in lexical storage and look-up 354 structure of lexical entries: keep kept vs. table - house 352f suppletion: lexical structure of go went, look-up 354 lexical access identification by computation is less costly than look-up 355 and acquisition 358 lexical access: look-up vs. compute 351 lexical pointer (= via-rules in →Natural Generative Phonology) 353 morpheme identification by parsing cues (compute) vs. look-up in the lexicon 353

Kaye (1995) (continued) lexical access (continued) morpheme identification with hidden morphology (stepped) 356 morpheme identification: different paths to identification are more or less costly 357f role of phonological relatedness in lexical storage and look-up 354 affix classes: empirical coverage 310 rule-triggering pattern: cannot be done (but English nasal cluster simplification can be done) 322 rule-triggering pattern: the domain-final condition 322 rule-triggering pattern, domainfinal condition: no violation of modularity (unlike Mohanan's →brackets) 323 underapplication: achieved by no look-back, comparison with other models 279 →affix class-based phenomena class 2 does, class 1 does not trigger interpretation (reverse of Halle & Vergnaud 1987a) 281 cannot do class 2 - class 1 strings (bracketing paradoxes, govern-mént2-al1) 315, 556 correctly predicts the word-final class 2 disjunction 326 PIC: modification-inhibiting no lookback 299 how it works 301 Chomsky's "spell-out and forget" too strong for phonology: two candidates for a weaker version 302 derived environment effects are a separate issue, no relation with no look-back 284 robustness (=no look-back) 279

808 Subject index L labelled brackets SPE 95 labels vs. structure morpho-syntactic structure may, but content (labels) may not bear on phonology (?) 752 language and mind cognitive realism 624 Language Faculty →language organ linguistics is a subfield of psychology (Chomsky) 627 language-internal modularity →modularity, grammar-internal modules and their identification in GB 630 against (Hornstein 2009) 633 →nested modules violate (informational) →encapsulation 632 language of thought Fodor (1975) 591 language organ (Chomsky) modules are innate 627 double dissociation of language 621 →biolinguistics late adjunction Kaye (1995): independent spell-out of terminals 318 late insertion argument for a phonological module 646 in →Distributed Morphology 536 lateral structure (in phonology) →deforestation length of the string is prosodic phrasing sensitive to the length of the string? 421, note 142 (577) level also →strata level 1 rules →rule-blocking pattern →Lexical Phonology/ ruleblocking boundaries 164

level (continued) level 2 rules →rule-triggering pattern in English, Kaye's (1995) analysis based on the domain-final condition 322 Lexical Phonology, rule-triggering boundaries 168 level 1 vs. level 2 rules in Lexical Phonology 149-51 →affix classes, note 32 (142) Level Independence 667 survey 60 enforces translation, birth of modularity 700 early expression of (Fodorian) modularity 72 and Indirect Reference 415 induces →juncture phonemes (summary) 695f more and less strict application 62 orthodox position (Hockett, Joos, Trager) 67 more tolerant views (Pike, Vachek) 68 Z. Harris' ambiguous position 67 abolished by Chomsky et al. (1956) (and Halle 1959) 76 revival in →Natural Generative Phonology 127 level ordering Lexical Phonology: prohibition to loop back to previous →strata 147 levels of representation in Cognitive Science 590 lexical access (→Kaye 1995) morpheme identification by parsing cues 355 morpheme identification with hidden morphology (stepped) 356 morpheme identification: different paths to identification are more or less costly 357f organised by phonological structure 347 lexical entries (→Kaye 1995) structure of: keep - kept vs. table house 352f

Subject index 809 Lexical Faithfulness revival of the →Elsewhere Condition in OT (Cho 2008) 521 lexical identity rules →Elsewhere Condition, derived environment effects 191 in OT: Elsewhere Condition, derived environments 521 lexical insertion in Distributed Morphology 536 Lexical Integrity Hypothesis offspring of Lexical Phonology: morphology ≠ syntax 539 Lexical Phonology 139 two achievements: →interactionism and multiple mini-grammars 212 anti-interactionist version: →Halle & Vergnaud (1987a) 215 vs. Distributed Morphology 533 →affix class-based phenomena →derived environment effects 177 →Strict Cycle Condition (SCC) relations with Prosodic Phonology: →Prosodic Phonology/ relations with Lexical Phonology general architecture 145 general architecture summary 152 summary: the disastrous patches of the 80s (SCC-K, brackets) 214 →interactionism 146 lexical vs. postlexical phonology general 153 postlexical phonology: birth of the idea 154 morphologically conditioned vs. automatic application of rules 156 respective conditioning factors 157 why is →postlexical phonology non-cyclic? 156, 158 →cyclic (=lexical) vs. →noncyclic (=postlexical) rules 158 non-cyclic interpretation of word sequences imposed without argument or discussion 791

Lexical Phonology 139 (continued) lexical vs. postlexical phonology (continued) lexical →post-cyclic rules (Rubach & Booij 1984), →Strict Cycle Condition (SCC) 194 maximisation of procedural management 140 summary 213 procedural management of affix class-based phenomena 162 rule-blocking boundaries = →level 1 rules 164 elimination of rule blocking boundaries 165 trade-off: elimination of boundaries against domain assignment and brackets 165 hybrid representational - procedural theory because of the rule-triggering pattern 169 summary: the representational residue 214 alliance with Direct Syntax against Prosodic Phonology, note 121 (432) →derived environment effects and →affix classes: orthogonal 182 wrongly accused to violate →Indirect Reference (by Inkelas 1990) 433f →opacity 156f interpretational units: comparison SPE, Lexical Phonology, phase theory 160 multiple mini-grammars morpheme-specific (one for class 1 strings, another for class 2 strings) 148 one single pool of rules 151 →Lexical Phonology/ lexical vs. postlexical phonology strata (levels) 147 definition 147-52 selective rule application (domain assignment): a given rule applies only at a given stratum 149

810 Subject index Lexical Phonology 139 (continued) strata (levels) 147 (continued) domain assignment (Stratum Domain Hypothesis) 151 level 1 vs. level 2 rules 149-51 underapplication to class 2 strings 164 level ordering (prohibition to loop back into previous strata) 147 classical vs. Halle & Vergnaud 218 no look-back devices →Strict Cycle Condition (SCC) no look-back device: →bracket erasure 175 summary no look-back devices 214 bracket erasure: internal structure of words invisible to syntax 157 brackets & bracket erasure: needed for the rule-triggering pattern 168 lexical post-cyclic rules (Rubach & Booij 1984) →Strict Cycle Condition (SCC) 194 lexical vs. postlexical phonology →Lexical Phonology, lexical vs. postlexical phonology 153 lexicalism →anti-lexicalism initiated by Chomsky (1970), note 34 (144) in Lexical Phonology 155 Distributed Morphology 539 in phonology and syntax (of the 80s) 126 position of Kaye (1995) regarding cyclic spell-out of words 285 relations with no look-back: antidotes against overgeneration in the 70s 288 →Strict Cycle Condition (SCC) 192 Lexicon in Lexical Phonology: rejected by Distributed Morphology 534f

licensing →government →Government Phonology 705 Licensing Constraints Kaye (1995) 269 Linear Correspondence Axiom (LCA, Kayne 1994) 742, 744f linearisation 741 →PF the input to phonological computation is linear 742 both syntax-internal and syntaxexternal linearisation is minimalism-compatible 744 in →minimalism: no business of syntax 743 no business of phonology, therefore not discussed in syntactic quarters, but not in interface theories (which are made by phonologists) 742 everybody but Kayne does linearisation "at PF" 745 until GB: a function of constituent order, the head parameter 743 head parameter revived in PF-based linearisation strategies 745 if spell-out from narrow syntax linearises, in which way can the result be a PF-movement tree? 738 broad consensus: result of different serially ordered operations 745 Distributed Morphology concomitant with vocabulary insertion (DM) 745 Lin and Concatenate 745 Richards (2004, 2007) desymmetrisation 745 Epstein et al. (1998), Richards (2004, 2007) Precedence Resolution Principle 745 Bobaljik (2002) precedence rules 745 Idsardi & Raimy (forth) in three steps: phonology plus two distinct PF-modules 740

Subject index 811 linearisation 741 (continued) of phonological structure (Raimy)? 746 loanword phonology, note 54 (212) analysis with co-phonologies 479 local vs. domain-based carriers of morphosyntactic information definition and properties of the two world views 707 juncture phonemes & boundaries vs. prosodic constituency (summary) 706 transition boundaries > prosodic constituency (summary) 711 representational intervention in phonology: McCawley's (1968) early linear notation of domains 115 alternatives to prosodic constituency never discussed 689 translation of one into the other always possible 708 just notational variants? 115 arguments against domains 712-17 domains are necessarily diacritic prosodic word and higher prosodic constituents are the projection of nothing 715 top-down constructions are diacritic by definition 714 syllable and feet are entirely determined by their terminals 716 only local boundaries can be nondiacritic: syllabic space is the solution 717 localisation of cognitive functions 614 origins: phrenology (F-J Gall), correlation with areas on the skull bone 601f strategy: levels intermediate between symbolic representations and neurobiology: connectionism, Poeppel's circuits 616 locality conditions in phonology and syntax 705 long term (declarative) memory 596 look-ahead prohibited in structuralist discovery procedure 667

loop back to previous strata: prohibition in Lexical Phonology 147 low level processes →Alternation Condition/ exceptions of: automatic, low level, "phonetic" processes 186

M Magic Licensing Government Phonology, note 82 (344) mapping →Prosodic Phonology in SPE: general mapping algorithm 90 comparison SPE - Prosodic Phonology 385 split mapping enforced by Nespor & Vogel's (1986) peaceful coexistence with Lexical Phonology 439 phonology can see morpho-syntactic (tree) structure, but not its labels 398 in Prosodic Phonology (rule-based) 386 mapping and cross-linguistic variation: progressive parameterisation of mapping rules 388 variation regarding ω and φ 390f extended to units of the Prosodic Hierarchy below the word level: moras, syllables, feet 448 OT: constraint-based (parallel) mapping 457 effect achieved by interspersing ALIGN with regular phonological constraints 459 overgeneration not discussed 460, 463 variation: factorial typology of ALIGN and WRAP 460, 463 phase-based mapping (Kratzer & Selkirk 2007) 462 mapping puzzle general introduction 111 summary 753 and (functional) →Grenzsignale 57 post-SPE: alas, there are no natural classes of boundaries 111 Prosodic Phonology 387, 393, 396

812 Subject index mapping puzzle (continued) in Prosodic Phonology: the black box 380 OT: no advance, but focus on new factors 463, 530 melody incommunicado with morpho-syntax 660 privative representations 44 melody-free syntax rather than →phonology-free syntax 412 memory short/long term, working, declarative 596 active, short term limitations of: motivation for →derivation by phase 305 →minimalist perspective: concern for extra-linguistic factors and the →economy of cognitive resources 305 role in →lexical access (look-up) (Kaye 1995) 354 trade-off between active memory economy achieved by the →PIC and the need for a memory-keeper 799 memory keeper who decides to ignore "old" strings? (→spell-out mechanism) 21, 307, 799 mentalism (rationalism) vs. empiricism 591 Merge direct merge in →Distributed Morphology 543 early expression in LSLT (Chomsky 1955-56) 625f concat function, →Kaye (1995) 268 metrical grid Halle & Vergnaud (1987a) 237 Halle & Vergnaud (1987a), →Stress Copy 231 Halle & Kenstowicz (1991), compensation vs. condensation 549 metrical structure (foot structure) early modification-inhibiting no lookback: Free Element Condition 295

metrical trees (s/w) competition with the metrical grid, note 119 (424) mind and brain relationship 588, 590, 615f, note 156 (627) Minimal Indirect Reference Seidl (2001): hybrid model allowing for both direct syntax and prosodic constituency minimalism shrinks syntax 728 effects on →PF pumps up PF 729 PF is the minimalist dustbin 31 the minimalism-born intermundia (the part of PF that is not phonology) violates domain specificity 738 concern for cognitive resources (active →memory) enforces →interactionism (i.e. →derivation by phase) 305 does away with GB-modules 638 linearity is not a property of syntax 743 third factor explanations 638 Minimality Condition in phonology 705 mini-phonologies →computational systems modification-inhibiting no look-back →no look-back devices modularity 586 vs. →connectionism two competing theories of how the cognitive system works 588 bridging literature that aims at reconciliation 599 early days: complementary (a question of levels), rather than incompatible 599 →intermodular argumentation prism used in this book in order to look at the interface 844 core properties of modules 610 is dualistic and rationalist/mentalist (philosophy) 591 mind - brain relationship 588, 614

Subject index 813 modularity 586 (continued) offers control that is maximally independent from phonology 40 phonology-free syntax 645 localisation of cognitive functions in the brain 614 how are modules identified? →domain specificity, →encapsulation 610 →double dissociation 618 properties of modules: inputoutput, non-teleological etc. 605 (informational) encapsulation 612 nested modules: violation of encapsulation? 632 PF as a set of nested modules violates encapsulation 738 domain specificity 611 in Rotenberg (1978): phonology interprets phonological objects, not bananas 373 computation is computation of something 598 origins: phrenology (F-J Gall) 600ff modularity is the heir of phrenology (Fodor 1983) 602 cognitive functions, higher (horizontal) and lower (vertical), how are they identified? 605 Fodor (1983) Fodor's faculty psychology married with →computation theory (Turing-von Neumann) 603 language is a modular cognitive function (as opposed to a central system) 624 modules are biologically real and innate 627 Descartes, Fodor & Chomsky vs. Darwin, Pinker & Jackendoff two camps: some properties of the mind remain forever unintelligible (Descartes, Fodor, Chomsky) vs. everything is accessible to human reason (Pinker, Plotkin, Sperber, Smith) 608

modularity 586 (continued) Descartes, Fodor & Chomsky vs. Darwin, Pinker & Jackendoff (continued) peripheral (Fodor, Chomsky) vs. massive (Pinker, Plotkin, evolutionary psychology) modularity: is there a non-modular core? 606 central system made of modules? 605 evolutionary psychology 609 is the mind (are modules) the result of Darwinian adaptation? 609 communication among modules 649 translation always practised, even in absence of modular background 693 is the →spell-out mechanism a module? 799 is the interface a(n) (intelligent) module (Jackendoff)? 23 of vocabulary (input) vs. structure (result of modular activity) 652 phonology can see morphosyntactic (tree) structure, but not its labels 398 of and in language 622 grammar-internal modules and their identification 628 →Language Faculty →nested modules identification of language-related modules by double dissociation 642 baseline of the 60s: the inverted T 629 phonology is distinct: domain specificity (Jackendoff, Chomsky) 644 syntax vs. semantics, two distinct modules (Newmeyer 2006, Higginbotham 1987) 642 two macro-modules: phonology vs. morpho-syntax-semantics (Starke) 643

814 Subject index modularity 586 (continued) referee for interface theories 33, 37, 39 refereeing potential of lies waste in the 80s (summary) 682 settled in verb, but not in practice 690 is the generative baseline 725 if wrong, the generative enterprise must be rethought 524 one of the deepest layers of generative thinking 37 surprisingly absent from the (generative) interface literature and theory design 587, 844 in the history of generative grammar 38 in early generative grammar 623 early expression in generative linguistics: the "levels" of LSLT 625 input-output device (Chomsky & Miller 1963) 626 from LSLT (Chomsky 1955-56) to the inverted T and SPE 625f all modules have the concatenative privilege in LSLT (Chomsky 1955-56) 625 in SPE 84 definition of phonology, a forerunner of Fodorian modules 613 in Lexical Phonology violated by the reintroduction of brackets 170 in GB subcomponents (→inverted T) vs. subsystems (GB modules) 631 are GB-subtheories cognitive modules? 632 GB modules introduced without reference to the modularity of the cognitive system 630 GB subtheories and their perception as cognitive modules 628, 635 perception of GB-modules in nonlinguistic quarters: puzzlement 636

modularity 586 (continued) in OT 523 OT is a strong modularity offender 524 connectionist roots of OT: the D of PDP (→Parallel Distributed Processing) induces the violation of modularity 599 radical scrambling in OT-CC: one ranking for phonology and morphology (Wolf 2008) 472 rarely discussed in OT 524 violation: direct syntax 525 violation: mapping done in the phonology 454, 526 violation: morpho-phonological scrambling among constraints and within their body 527 in Distributed Morphology DM is a strong modularity offender: PF movement, phonology interleaved with nonphonological PF operations (→linearisation) 32, 580, 739 in minimalism & biolinguistics minimalism does away with GBmodules 638 against language-internal modularity (Hornstein 2009) 633 biolinguistics: grammar reduces to morpho-syntax, PF and LF are neither language- nor speciesspecific 639 →inverted T untouched by the new biolinguistic delineation of grammar 639 trying to make sense of PF from the modular point of view 736 the minimalism-born PF intermundia violates →domain specificity 738 Idsardi & Raimy (forth) modules after syntax: phonology plus two distinct PF-modules, related by a three-step →linearisation 740

Subject index 815 modularity 586 (continued) generative modularity offenders summary 702 typology of modularity offenders 704 modularity offenders: labelled brackets (SPE) 98f violated by the reintroduction of brackets (Lexical Phonology) 170 OT is a strong modularity offender 524 OT, induced by the D of PDP (Parallel Distributed Processing) 599 violated in OT: →direct syntax 525 violated in OT: mapping done in the phonology 454, 526 violated in OT: morphophonological scrambling among constraints and within their body 527 radical scrambling in OT-CC: one ranking for phonology and morphology (Wolf 2008) 472 DM is a strong modularity offender: phonology interleaved with non-phonological PF operations (→linearisation) 739 PF movement 32, 580 the minimalism-born PF intermundia violates domain specificity 738 violated by Structural Analogy? 705 modularity offenders summary 702 →modularity/ generative modularity offenders monism vs. dualism 591 monostratalism in Sign-Based Morphology 513 Coloured Containment (Oostendorp) 510 in OT: radical scrambling, one ranking for everything (Russell) 528 morpheme recognition →parsing cues (→Kaye 1995) 263

morpheme-specific phonologies →computational systems morphological adjuncts Kaye (1995): independent spell-out of terminals 318 Morphological Merger Marantz (1984), forerunner of →PF Movement 580 morphology vs. syntax morphology ≠ syntax Lexical Phonology: distinct computational systems 155 same computational system: Sproat (1985), Halle & Vergnaud (1987a) 223 distinct engines enforced by Nespor & Vogel's (1986) peaceful coexistence with Lexical Phonology 439 morphology = syntax Distributed Morphology 534f, 537 Lexical Integrity Hypothesis, contra Distributed Morphology 539 ancestor of DM: Selkirk (1984) (contra Lexical Phonology) 432, 540 phonology-free syntax or morphosyntax? 412 just one spell-out mechanism (morphology and syntax are the same computational system), or two (they are not)? 858 one or two engines is decisive for an argument against non-selective spell-out 769 spell-out your sister! (→phase edge) applies to all chunk sizes 766 morpho-phoneme structuralism 62 morpho-phonology SPE: common underlying form for heart - cardiac etc. 126 position of Lexical Phonology 156 Kaye's (1995) take in his proceduralonly system 276 multiple exponence allomorphy (Lowenstamm 2008), note 140 (571)

816 Subject index multiple mini-grammars →computational systems multiple spell-out →derivation by phase

N natural classes of boundaries →mapping puzzle 111 Natural Generative Phonology phonetic boundaries 127 boundaries have a phonetic correlate 669f boundary abuse in Natural Generative Phonology 127 reference to in OT (Green 2007) 527 revival of structuralist →Level Independence 127 NDEB (nonderived environment blocking, Kiparsky 1993) →Strict Cycle Condition, →derived environment effects 196 nested modules violate (informational) encapsulation 632, 738 language-internal modularity →modularity/ grammar-internal modules and their identification →Language Faculty Neumann, John von improved the Turing machine 595f, 603 neural networks (→connectionism) artificial neural networks 591 how they work 595 activation value of a neuron 595 connection weight 595 neuropsychology 611 New Synthesis Psychology Pinker, Plotkin 607 new vs. old (strings, properties) →no look-back devices/ modificationinhibiting no look-back 307

no escape from syntax slogan of Marantz' (Distributed Morphology) 535 no look-back devices →computational systems/ chunkspecific phonologies →Strict Cycle Condition (SCC) →Phase Impenetrability Condition/ PIC à la carte (chunk-specific) →Phase Impenetrability Condition/ PIC à la carte (process-specific) overview 16 summary 783 vs. →morpheme-specific miniphonologies (LP): a major front line 829 →morpheme-specific miniphonologies vs. no look-back (PIC): refereed in favour of the →PIC 830 the old vs. the new overview 307 who identifies old and new portions of the string? 797-99 history since Chomsky (1973) 287 absent in SPE 176 origin in the early 70s: reaction against overgeneration 288 Chomsky's (1973) →Strict Cycle Condition imported into phonology: Kean (1974), Mascaró (1976) 189f Head Constraint (Riemsdijk 1978) 298 in Lexical Phonology: bracket erasure 157, 175 Mohanan's brackets vs. Kiparsky's →SCC-K 203 summary in Lexical Phonology 214 relation of Mohanan's bracket erasure with other no lookback devices 292

Subject index 817 no look-back devices (continued) history since Chomsky (1973) 287 (continued) SCC-K in Halle & Vergnaud (1987a) 237 general presentation of →Phase Impenetrability 304 modification-inhibiting no look-back first timid steps in the 70s and 80s 293 early (70s): Riemsdijks' (1978) Head Constraint 298 early (80s): Free Element Condition (FEC) 295 early (80s): structure preservation 295 →Kaye (1995): "don't undo!" 299 and domain-final empty nuclei 332 motivates the phase edge in syntax and phonology (→Kaye 1995) 766 process-specific no look-back which processes do, which do not violate the SCC and no lookback? 186, 315 typically violated by stress assignment 315 Chomsky's "spell-out and forget" too strong for phonology: two candidates for a weaker version 302 stress: Free Element Condition (FEC) 296 stress: Marvin (2002) in Distributed Morphology 554 English: stress (word-bound) vs. flapping (cross-word) 241 relevant barrier: the word 239, 257 PIC also applies to phonology, but Chomsky senses it is too strong a condition on phonological computation 306, 780, 797, 826

no look-back devices (continued) PIC (continued) Chomsky's "spell-out and forget" too strong for phonology: two candidates for a weaker version, process-specific PIC and Kaye's "don't undo!" 302, 825 no reference made by modern phase theory to previous phonological (and syntactic) implementations 309 no confusion with Kiparsky's SCC 195 PIC à la carte summary 823 process- and chunk-specificity 826 is computational →economy à la carte, not a very minimalistic take 780, 797, 826, 830 non-analytic vs. analytic (domains, morphology, affixes) →Kaye (1995) 272 non-cyclic affixes →Halle & Vergnaud (1987a), affixtriggered interpretation Halle & Vergnaud (1987a), = wordlevel rules in SPE 219, 233 non-isomorphism →Prosodic Phonology 416, 580, 709 the classical example: cat-rat-cheese 418 in the Prosodic Phonology literature, motivation for Indirect Reference 417 is a fact about domains, not about language: it evaporates when (local) boundaries are used 419 and privativity 759 domain abuse: confusing data and domain-based interpretation 420 Nuclear Stress Rule (NSR) early cases of →selective spell-out: Bresnan (1971) 308 eliminated by Cinque (1993): →intonation is a purely syntactic matter 806

818 Subject index number sense →double dissociation 620

O old vs. new (strings, properties) →no look-back devices/ modificationinhibiting no look-back, →Phase Impenetrability 307, 797-99 Coloured Containment (Oostendorp) 505, 507 One-channel translation →Direct Interface 23, 49 opacity predictable vs. unpredictable 561f, 568 result of distinct geometric properties (DM) vs. distinct computation (everywhere else) 582 selective spell-out and the PIC fare better than Marantz' →direct merge 563 in Lexical Phonology 156f in OT opacity killers (and cyclicity) killers 475 in Distributed Morphology 541 result of structure, rather than of computation (Distributed Morphology, Marantz) 542 result of direct merge to a root (merge to an xP produces transparency) 544 direct merge vs. selective spell-out + PIC (summary) 542, 582 concomitance of LF- and PF opacity (summary) 583 PF - LF opacity: all combinations attested 565 combinations of PF - LF opacity: neither the PIC nor direct merge can account for all patterns 566 PF vs. LF opacity: maybe not the same thing 562 LF opacity 542 does not exist in a lexicalist perspective: cómparable - compárable 573

opacity (continued) in Distributed Morphology 541 (continued) predictable = PIC, unpredictable = direct merge? 568 in Coloured Containment (Oostendorp) 508 Opacity Principle Mohanan (1982), note 44 (168) opaque stress placement (English) 311 open juncture →juncture Optimal Interleaving Wolf (2008) (OT-CC): revival of interactionism with one single constraint hierarchy that merges morphology and phonology 472 Optimality Theory connectionist roots 28, 529, 678 like connectionism, has no genuine symbolic vocabulary 28, 452 mapping constraint-based (vs. rule-based) 455, 457 phase-based (Kratzer & Selkirk 2007) 462 status of representations representations marshalled by constraints 453 non-mainstream work where representations (which are not "emergent") are independent of computation 8 computation in OT 269 anti-cyclicity (anti-derivationalism) 28 overview 454 cyclic derivation rejected 466, 468 cyclic derivation (interface) ≠ phonological computation 470 OT less anti-derivational now than it used to be note 127 (454), 686 opacity killers and cyclicity killers 475 cyclicity killers co-phonologies 478 indexed constraints 482

Subject index 819 Optimality Theory (continued) cyclicity killers (continued) Interface Constraints 494 OO-correspondence 497 direct syntax 494, 501 →derived environment effects 516 violates modularity summary 523 misty relationship with modularity 28 systematic violation of →Indirect Reference 28 multiple, unreflected and systematic violation of modularity 471 happy coexistence of prosodic constituency and direct syntax 525 issue rarely discussed 524 Stratal OT favours clear-cut modular contours, but still violates modularity 527 direct syntax 525 scrambling trope: morphophonological scrambling among constraints and within their body 28, 527 scrambling trope in OT-CC: one ranking for phonology and morphology (Wolf 2008) 472 scrambling trope: mapping done IN the phonological computation: ALIGN 28, 457, 526 radical scrambling: one ranking for phonology, morphology and syntax (Russell) 528 morpheme-specific mini-phonologies 473 parallel (co-phonologies, indexed constraints) vs. reranked incarnations (Stratal OT/DOT) 476 reranking (Stratal OT, DOT) 476 serial vs. parallel implementations of morpheme-specific phonologies: arguments are missing 484 co-phonologies 478 indexed constraints 482 reranked (Stratal OT, DOT) 483

OT-CC OT less anti-derivational now than it used to be, note 68 (269), 686 Optimal Interleaving (Wolf 2008): revival of interactionism with one single constraint hierarchy that merges morphology and phonology 472 outer vs. inner morphology Distributed Morphology (Marantz) 546 Output-Output correspondence designed as a cyclicity killer, but cannot do the job alone 497 used for derived environment effects in OT 519 overgeneration →abstractness debate 144 Kiparsky's revisionist position 144 antidote in the early 70s: →no lookback 288 of →PF movement 580

P paradigmatic gaps produced by root-based (as opposed to word-based) morphology 546 Parallel Distributed Processing (PDP) connectionism 597, 599 parallel modules (Jackendoff) contrast with inverted T 29 alternative to the inverted T model 722 implementation by Bendjaballah & Haiden 724 parsing analysis of →affix class-based phenomena by Fabb (1988) and Hay (2002) based on selectional properties of affixes, perception, parsing 247 parsing cues →Grenzsignale (Trubetzkoy) functional and perception-oriented approach to the interface 263f Kaye (1995) 263 theory-independent (clusters) 341 theory-dependent (GP): detection of empty nuclei 342

820 Subject index parsing cues (continued) Kaye (1995) 263 (continued) undetected morpheme breaks 345 morpheme identification by (compute) vs. look-up in the lexicon 353 processes that typically act as boundary detectors 265 parsing device cyclic derivation (inside-out) 675 particle physics 592 perception analysis of affix class-based phenomena by Fabb (1988) and Hay (2002) based on selectional properties of affixes, perception, parsing 247 perception-oriented theories of phonology and the interface →Kaye (1995), →Grenzsignale (Trubetzkoy) 264 percolation of lexical properties of affixes to the morpho-syntactic tree Halle & Vergnaud (1987a) 227 performance →competence Peripherality Condition no extra-Xal elements morphemeinternally, note 106 (400) PET (Positron Emission Tomography) localisation of cognitive functions in the brain 617 PF 726 clean syntax, dirty PF (phonology)? 727 what exactly do we gain? 31, 747 genesis and evolution summary: a minimalism-born monster whose contours and function are murky 31, 747 the minimalism-born intermundia violates →domain specificity 738 evolution of: coextensive with phonology until minimalism, then loaded with a lot of nonphonological things 729 minimalism shrinks (narrow) syntax and pumps up PF 728f

PF 726 (continued) definition PF ≠ phonology 23, 31 (minimalist) PF = phonology + "something", but nobody knows what the "something" really is 729 (minimalist) PF ≠ what phonologists call phonology 729 incommunication and misunderstanding: the syntacticians' phonology has got nothing to do with what phonologists call phonology 730f confusion and mistiness in Chomsky's writings: what does PF mean, what does it contain? 732 Phonetic Form or Phonological Form? a confusion that is not innocent 732 content operations that syntacticians want PF to do, and restrictions on these operations 735 case of PF-outsourcing: clitic placement 730 used as a dustbin by syntacticians 729f properties of PF: what kind of animals live in the intermundia? 733 if spell-out from narrow syntax linearises, in which way can the result be a PF-movement tree? 738 internal structure is a set of serially ordered computational systems 737 the "PF branch" (Distributed Morphology) 734 Idsardi & Raimy (forth): phonology plus two distinct PFmodules, related by a threestep linearisation 740 and modularity trying to make sense of PF from the modular point of view 736

Subject index 821 PF 726 (continued) and modularity (continued) DM is a strong modularity offender: phonology interleaved with non-phonological PF operations 739 PF movement (DM) Distributed Morphology 574 instrument of the →minimalist philosophy: (narrow) syntax shrunk, →PF pumped up 575 syntactic motivation 575 phonological motivation: phonology impacts syntax 577 violates modularity (domain specificity) 580, 738 challenges phonology-free syntax 577 undermines DM's claim that syntax = morphology introduces distinct syntactic and morphological Merge 538, 580 if spell-out from narrow syntax linearises, in which way can the result be a PF-movement tree? 738 morpho-syntactic well-formedness is determined by phonological content (Lowenstamm 2008) 579 syntactic trees with phonological terminals: Lowenstamm (2008) 579 phonology-internal triggers (syntactically undetectable, Piggott & Newell 2006 on Ojibwa) 578 phase boundary →phase theory phase edge early modification-inhibiting no lookback: Head Constraint (Riemsdijk 1978) 298 in syntax and in phonology (Kaye 1995): spell-out your sister! 765 referee for competing phonological theories (intermodular argumentation) 857

Phase Extension (den Dikken 2007) →phase theory 781 phase head →phase theory phasehood →phase theory Phase Impenetrability Condition (PIC) →no look-back devices →computational systems history of: modification-inhibiting no look-back →no look-back devices/ modification-inhibiting no look-back general presentation 304 instrument of active memory economy 306 managed by the spell-out mechanism (not by syntax or phonology) 799 the old vs. the new, memorykeeper: who identifies old and new portions of the string? 307, 797-99 no reference made by modern phase theory to previous phonological (and syntactic) implementations 309 no confusion with Kiparsky's SCC 195 also applies to phonology, but Chomsky senses it is too strong a condition on phonological computation 306, 780, 797, 826 Chomsky's "spell-out and forget" too strong for phonology: two candidates for a weaker version, process-specific PIC and Kaye's "don't undo!" 302, 825 referee for competing phonological theories (→intermodular argumentation) 857 argument in favour of →selective spell-out 770 in Distributed Morphology: spell-out at every xP is much too strong a claim 555

822 Subject index Phase Impenetrability Condition (PIC) (continued) PIC à la carte two types of PIC à la carte: process-specific and chunk-specific 797-99 process- and chunk-specificity: summary 823, 826 is computational →economy à la carte: not a very minimalistic move? 780, 797, 826, 830 PIC à la carte (process-specific) →no look-back devices/ processspecific no look-back summary 780 in syntax: long distance Agree 780 Distributed Morphology (Marvin 2002) 555 PIC à la carte (chunk-specific) summary 809 PIC à la carte vs. →Praguian segregation 818-27 (chunk-specific) PIC à la carte in Samuels (2009a), but hidden by a specific formulation 798 phase theory →no look-back devices →Phase Impenetrability Condition (PIC) →derivation by phase →interactionism →selective spell-out general presentation 304, 771 convergence between phonological and morpho-syntactic evidence for phase boundaries? 19 independent PF and LF phases (asymmetric spell-out)? 567, 779 voices against phase theory 772 history phase in early generative grammar: LSLT (Chomsky 1955-56) 625 delineation of interpretational units: SPE and Lexical Phonology 160

phase theory (continued) history (continued) no reference made by modern phase theory to previous phonological (and syntactic) implementations 309 evolution in syntax 19, 771 phasehood (phase head) trend towards atomisation 775 atomisation marshalled by the antilocality of movement 777 convergence-based phasehood (Svenonius 2001), note 178 (781) extra-syntactic control called into question 774 extra-syntactic criteria for identifying phases ("independence", propositionality) 774 is there a lower limit for atomisation? 776 node-driven phase 774 parameterised (Gallego 2010) 775 Phase Extension (den Dikken 2007) 781 phasehood as a projection of a lexical (phasehood) feature 782 phasehood in Distributed Morphology DM has a peculiar way to do →selective spell-out: functional (yes) vs. non-functional heads (no) 558 all xPs (categorisers) are phase heads 555 functional heads (VoiceP etc.) are not phase heads, hence DM ≠ spell-out-as-you-merge (Epstein) 555 the DM take (every xP is a phase head) does not allow to account for affix class-based phenomena 583 different distribution of phasehood below and above the word level 559

Subject index 823 phase theory (continued) phase boundaries convergence between phonological and morpho-syntactic evidence for phase boundaries? 19 phase-based mapping (Kratzer & Selkirk 2007) 462 phase edge early modification-inhibiting no look-back: Head Constraint (Riemsdijk 1978) 298 in syntax and in phonology (Kaye 1995): spell-out your sister! 765 referee for competing phonological theories (intermodular argumentation) 857 node-driven vs. piece-driven phase 767 Phase Extension (den Dikken 2007) 781 unification by a phasehood feature 782 phi-function →Kaye (1995) 267 phone structuralism 62 phoneme structuralism 62 phonemic clause phonological domain (Chomsky et al. 1956) 75 phonetic boundaries Natural Generative Phonology 127 phonetic correlate of morpho-syntactic information →juncture there is no (summary) 668 phonetic hypothesis rejected by Kaye (1995), note 66 (263) phonetics vs. phonology the same or distinct modules? Domain specificity 647 "phonetic" rules →Alternation condition/ exceptions of →Strict Cycle Condition (SCC) 186, 193

phonological cycle →cyclic derivation term coined by Mascaró (1976), replacing SPE's Transformational Cycle 102, 189, 290, 673 phonology definition: is phonological what is the output of phonological computation 804, 806 phonology-free syntax 412 is in fact melody-free syntax: holds only for phonological properties below the skeleton 412 melody and morpho-syntax are incommunicado in both directions 660 is morphological concatenation sensitive to derived phonological properties? 253 explained if structure, but not vocabulary, is translated 653 and intonation 801 challenged by →PF movement 577 identifies a phonological module 645 phrasal idiom in Distributed Morphology 544 phrase structure rules 625 phrenology →modularity, origins 601 PIC →Phase Impenetrability Condition plasticity of the brain 616 pool of rules Lexical Phonology: one single pool of rules that "intervene" at different strata or postlexically 151 possible word blick vs. lbick (→Kaye 1995) 349 post-cyclic lexical rules (Lexical Phonology) (Rubach & Booij (1984), = SPE's →word level rules 194 postlexical phonology birth of the idea in Lexical Phonology 154f post-SPE 109

824 Subject index pragmatics module interacting with grammar 629 evidence for independent status from →double dissociation 642 Praguian segregation word- vs. sentence phonology, →postlexical phonology 154f summary 816 and the →Strict Cycle Condition (SCC) 192 motivation: process-specific no lookback (→PIC à la carte) 240 dismissed by Halle & Vergnaud (1987a) 219, 248 dismissed by Kaye (1995) 283 vs. PIC à la carte 818-27 assessed by →double dissociation 818 →word-spell-out-mystery supported by the word-spell-outmystery 817 not good enough to explain what is going on with the word-spellout-mystery 808 precedence relationships →directed graphs (Raimy) Precedence Resolution Principle →linearisation: Epstein et al. (1998), Richards (2004, 2007) 745 precedence rules →linearisation: Bobaljik (2002) 745 precompilation theory Hayes (1990) 408 privativity five arguments in favour of privativity (representational and procedural) 761 on the procedural side: selective vs. non-selective spell-out 760 on the representational side: underfeeding of phonology is an undisputed fact, but all theories since SPE are non-privative 757 not every morpho-syntactic division is projected onto phonology (Chomsky et al. 1956) 79

privativity (continued) SPE is non-privative: →boundaries present without phonological effect 90 the non-privativity of SPE led to confusion 132 Lexical Phonology is non-privative 173 Prosodic Phonology is non-privative: all layers of the Prosodic Hierarchy are always constructed 383, 400 procedural vs. representational analyses →balance of procedural and representational analyses, →Interface Dualism 320 Procedural First if both representational and procedural analyses are available, choose the latter 6, 18, 316, 320, 850 procedural management of interface phenomena →Interface Dualism proceduralisation of the interface Lexical Phonology 213 Halle & Vergnaud (1987a) 255 processors Jackendoff's name for modules 644 process-specific external sandhi →sandhi (external) 239 process-specific no look-back →no look-back devices/ PIC à la carte, →Phase Impenetrability/ PIC à la carte projection principle →Government Phonology 705 prosodic government →Government Phonology 705 Prosodic Hierarchy →Prosodic Phonology properties and function 378f , 399 resounding success of 361, 364 transition boundaries > prosodic constituency (summary, →Prosodic Phonology) 711

Subject index 825 Prosodic Hierarchy (continued) superfluous? Issue with →direct syntax (→Prosodic Phonology) 409 internal structure its six layers 382 Strict Layer Hypothesis: classical (monolithic) and OTed (made of 4 independent constraints) 383 the Prosodic Hierarchy is a heterogeneous blend of bottom-up (moras, syllables, feet) and top-down constructions (the rest) 374 Clitic Group, abandoned, note 96 (382) is non-privative: all layers of the Prosodic Hierarchy are always constructed 383, 400 the Prosodic Hierarchy is a heterogeneous blend of bottom-up (moras, syllables, feet) and top-down constructions (the rest) 374, 401 domains are necessarily diacritic: →Prosodic Phonology is a diacritic (from the PrWd upwards) definition of diacritics 405 →diacritics do not qualify 449 diacritic →translation (summary) 697 illusion of a diacritic-free phonology since the 80s 698 diacritic issue settled in verb, but not in practice: No Diacritics! 690 prosodic constituents are diacritics, if autosegmental ones 361, 402 prosodic constituents are the autosegmental version of boundaries 403 prosodic constituents share all formal properties with boundaries 404 only purpose: storage of morphosyntactic information (the buffer) 400

Prosodic Hierarchy (continued) is a diacritic (from the PrWd upwards) (continued) prosodic constituents are wrongly sold as "truly phonological objects" 405 non-diacritic carriers of morphosyntactic information exist goal for further study: to make carriers of morpho-syntactic information local AND nondiacritic (→Indirect Interface) 366 →initial CV 44, 49, 713 Prosodic Lexical Phonology Inkelas' (1990) compromise 433 Prosodic Morphology 442 forerunner of OT 361 transposed into OT: Generalized Template Theory 448 two periods: representational and antirepresentational 443 typical phenomena covered: reduplication, templatic morphology 444 Prosodic Phonology 360 →non-isomorphism →Prosodic Hierarchy diacritic character of: →Prosodic Hierarchy overview/summary 361, 449, 697 general architecture 379 has made translation systematic and consensual 361 black box (non-isomorphism) 378, 380 buffer (=Prosodic Hierarchy) 378 three ways to make reference to the PH: domain span, domain juncture and domain limit rules 384, 709 history, genesis overview of its roots in the 70s 110 all architectural properties come from SPE 385 Indirect Reference 385

826 Subject index Prosodic Phonology 360 (continued) history, genesis (continued) continuity with SPE, but a tabula rasa self-understanding 361 ancestors: Selkirk adapts Liberman & Prince (1977) to the interface 363 is a child of autosegmentalism 368 major historical break: boundaries, not domains, have always been the carriers of morphosyntactic information since the 19th century 361, 366 early period (up to 1986) 362 summary of Selkirk's early work 363 summary of the second strand: Nespor & Vogel 364 vs. direct syntax: Indirect Reference 378, 385 why Indirect Reference? 407 phonology can see morphosyntactic (tree) structure, but not its labels 398 alliance of Lexical Phonology and Direct Syntax against Prosodic Phonology, note 121 (432) mapping how it works 387 mapping rules 378, 380 is done in modular no man's land 381 for rhetorical purposes, mushy and uncontrolled mapping is hidden by a clean and universal Prosodic Hierarchy 393 is prosodic phrasing sensitive to the length of the string? 421, note 142 (577) Nespor & Vogel (1986): two sets of mapping rules 439 Selkirk's (1986) end-based mapping 394 Selkirk's (1986) end-based mapping: origin of alignment in OT 395

Prosodic Phonology 360 (continued) mapping (continued) Selkirk's (1986) end-based mapping puts to use the technology of the 80s: X-bar-based mapping 395 mapping puzzle poorly understood 387 mapping: mechanics, atomising parametric evolution, absence of cross-linguistic generalisations 386 and cross-linguistic variation: progressive parameterisation of mapping rules 388 variation regarding ω and φ 390f Selkirk's (1986) end-based mapping: no advance 392, 396 (local) boundaries vs. (non-local) domains (→translation) local boundaries vs. non-local domains (overview) 361 boundaries are the primitive currency for mapping: prosodic domains are built on the basis of boundary grouping 397 boundaries replaced by autosegmental domains 361 boundaries not an issue anymore for Nespor & Vogel (1986) 370 domains have an independent motivation: stress, rhythm, musical properties 374f (non-)discussion of boundaries 369 major historical break (that went almost unnoticed): boundaries, not domains, have always been the carriers of morphosyntactic information since the 19th century 361, 366f non-arguments against boundaries 371f nothing is gained when boundaries are replaced by prosodic constituency since both are diacritics 373

Subject index 827 Prosodic Phonology 360 (continued) (local) boundaries vs. (non-local) domains (→translation) (continued) the diacritic argument against boundaries 373 domains are necessarily diacritic →Prosodic Hierarchy definition of diacritics 405 diacritic translation (summary) 697 illusion of a diacritic-free phonology since the 80s 698 settled in verb, but not in practice: No Diacritics! 690 prosodic word and higher prosodic constituents are the projection of nothing 715 top-down constructions are diacritic by definition 714 syllable and feet are entirely determined by their terminals 716 only local boundaries can be nondiacritic: syllabic space is the solution 717 non-diacritic carriers of morphosyntactic information exist goal for further study: to make carriers of morpho-syntactic information local AND nondiacritic (→Indirect Interface) 366 →initial CV 44, 49, 713 relations with Lexical Phonology and the metrical grid: overview 424 relations with Lexical Phonology summary competition and outlook (Stratal OT) 441 officially peaceful cohabitation, but a conflict in actual fact 361 above vs. below the word level: no concurrence above 401, 430

Prosodic Phonology 360 (continued) relations with Lexical Phonology (continued) Nespor & Vogel (1986): peaceful coexistence, but no rationale for dividing labour 435 Rubach & Booij: peaceful scrambling of everything, happy violation of Indirect Reference 440 Hayes (1989 [1984]): no Prosodic Phonology below the word level 431 Selkirk (1984): Lexical Phonology is redundant and has to go 432 Inkelas (1990): reducing Lexical Phonology to an empty shell 433 alliance of Lexical Phonology and Direct Syntax against Prosodic Phonology, note 121 (432) elimination of boundaries in Lexical Phonology remains unreflected 367 Nespor & Vogel (1986): two sets of mapping rules 439 relations with the metrical grid 425 Nespor & Vogel (1986): peaceful cohabitation with the grid 428 Selkirk (1984): the grid is superior 426 Selkirk (1986): the grid is derived from prosodic constituency 427 prosodic phrasing sensitive to the length of the string? 421, note 142 (577) prosopagnosia →double dissociation 620 psychology study of language is a sub-discipline of psychology (Chomsky) 627

Q quarks (physics) 592

828 Subject index R rationalism vs. empiricism 591 readjustment SPE: ancestor of →non-isomorphism and →Indirect Reference (→Prosodic Phonology) 91 SPE, replicated in Prosodic Phonology as the Black Box 380, 385 in Distributed Morphology: allomorphy/suppletion 571 readjustment rules →boundary mutation rules, →readjustment recursion also →intonation privilege of morpho-syntax, absent in phonology: no phonological phenomenon qualifies 45f, 801, 803 absence in phonology predicted by flat CVCV (Vol.1) 805 definition: repetition of the same item, limited only by performance restrictions 803 confusion: recursion is a phenomenon whose existence is established by pre-theoretical properties, not by analyses that happen to use recursive constructions, note 5 (46), 803 recursive prosodic structure is no evidence for recursion in phonology because it is not created by phonological computation 804 is the result of concatenation: since phonology does not concatenate anything, there can be no recursion in phonology 803, 805 Merge must be absent from phonology: Neeleman & van de Koot (2006) 805 Strict Layer Hypothesis allows for recursive prosodic structure since Selkirk (1996) 802 reductionism reduces the mind to the brain 591

relation-based mapping vs. end-based mapping (Selkirk 1986) 396 Relative Access Memory (RAM) →Turing (von Neumann) machine Relativized Minimality atomisation of phasehood marshalled by the anti-locality of movement 777 in phonology 705 Representation Theory Williams (2003) 632 representational vs. procedural analyses: →balance of representational intervention →local vs. domain-based intervention representational management of interface phenomena →Interface Dualism Representational Modularity (Jackendoff) alternative to the inverted T model 644, 723 representations independent of constraint action 507 reranking Stratal OT, DOT 476 Revised Alternation Condition (RAC) also →Alternation Condition 129, 187 against overgeneration 144 back to, after the bankruptcy of the SCC 197 Rhythm Rule →stress clash rhythmic properties of the linear string ancestors of Prosodic Phonology: Liberman & Prince (1977) 363 robustness →Kaye (1995) 279 Root Faithfulness used for derived environment effects in OT (Anttila) 520 roots do not possess category (Distributed Morphology) 544

Subject index 829 rule types also →computational systems summary 234-36 equivalences between SPE and Lexical Phonology: →cyclic, →word-level, →level 1, →level 2, →postcyclic lexical, →non-cyclic 106 →word-level rules (SPE) = →noncyclic rules (Halle & Vergnaud 1987a) = →post-cyclic lexical rules (Lexical Phonology) 219, 233 →post-cyclic lexical rules (Rubach & Booij 1984) = →non-cyclic rules (Halle & Vergnaud 1987a) = SPE's →word-level rules 154 cyclic rules summary 236 SPE 105 Lexical Phonology 158, 193 Halle & Vergnaud (1987a) 219, 233 lexical rules 153 level 1 rules (→rule-blocking pattern) 149-51 level 2 rules (→rule-triggering pattern) 149-51 rule-blocking boundaries SPE 93 rule-blocking pattern definition: Kenstowicz & Kisseberth (1977, 1979) 50 summary English processes 311 analysis in SPE 93 in Lexical Phonology 163 level 1 rules 164 (unintended) elimination by procedural management 165 analysis by Halle & Vergnaud (1987a) 227-29, 249 analysis by Kaye (1995): underapplication achieved by no look-back 312f

rule-list fallacy →"Cognitive" Linguistics (Langacker 1987), usage-based (Bybee 2001) 596 rules ranked by →boundary strength (McCawley 1968) 114 rule-triggering boundaries SPE 94 rule-triggering pattern definition: Kenstowicz & Kisseberth (1977, 1979) 50 summary English processes 311 analysis in SPE 94 in Lexical Phonology 166 level 2 rules 168 brackets and bracket erasure needed 168 can be done by the SCC (but this is ignored in the literature) 203 cannot be done by Halle & Vergnaud (1987a) 250 cannot be done by Kaye (1995) in absence of additional conditions 325

S s+C clusters lexical access (→Kaye 1995) 347 sandhi expression of →local (rather than domain-based) intervention 707 sandhi (external) process-specific, →no look-back devices, PIC à la carte, →Phase Impenetrability, PIC à la carte 239 and modification-inhibiting no lookback 302 English: flapping 241 function in perception: I know what ewe I want (→Kaye 1995) 263 savant syndrome double dissociation 621 SCC-K (Kiparsky's / Halle's version) →Strict Cycle Condition, →derived environment effects 190

830 Subject index SCC-M (Mascaró's version) →Strict Cycle Condition, →derived environment effects 189 scrambling trope →Optimality Theory, violates modularity sealed suitcases vs. late insertion 536, 646 selectional properties of affixes analysis of affix class-based phenomena by Fabb (1988) and Hay (2002) based on selectional properties of affixes, perception, parsing 247 selective rule application Lexical Phonology: a given rule applies only at a given stratum →domain assignment, →computational systems 149, 151 selective spell-out →computational systems overview 15 summary, two options: spell-out of the sister or of the mother 764 and privativity 760 unifying perspective with syntax 770 referee for competing phonological theories (intermodular argumentation) 857 argument against non-selective spellout: spell-out is selective in syntax 769 Halle & Vergnaud (1987a) birth of a groundbreaking idea 220, 225 = derivation by phase 220 what exactly is spelled out 226 early cases of: Bresnan (1971) 308 in Kaye (1995) 273 comparison Kaye's vs. Halle & Vergnaud's implementation 277 Distributed Morphology spell-out at every xP 559 peculiar: categorisers (nP, vP, aP) are phase heads, but functional heads (VoiceP etc.) are not 555

selective spell-out (continued) Distributed Morphology (continued) distinct computation (selective spell-out) vs. distinct representations (DM, →direct vs. indirect Merge)) 560 serial computation vs. →Parallel Distributed Processing (PDP, →connectionism) 597 short term (working) memory 596 Sign-Based Morphology (Orgun) 512 monostratalism 513 everything is melted into one, no interface left 515 cyclic effects 514 instantiation of →direct syntax 502 Silent Demibeat Addition SDA Selkirk (1984) 426 simplicity derived environment effects 198 single engine hypothesis →Distributed Morphology, →morphology vs. syntax 534f sister spell-out your sister! (→Kaye 1995, →phase edge) 282 SLI (Specific Language Impairment) →double dissociation of language 621 sonority sequencing of word-final clusters 331 spatial reorientation →double dissociation 620 SPE 81 →boundaries →computational systems general architecture 86 general architecture is an implementation of →modularity 84, 99 practises (intermodular) →translation 84 inverted T model and its plug-in "all concatenation before all interpretation" 13, 86 readjustment (component) 91 absence of →no look-back device (PIC) 83 labelled brackets 95

Subject index 831 SPE 81 (continued) general architecture (continued) →cyclic derivation 100 definition of the cycle by major categories (N,V,A) 103 one single →computational system (due to the representational definition of words and affix classes) 107 →rule types cyclic vs. word level rules 105 word-level rules ≈ lexical postcyclic rules (Rubach & Booij 1984) 194 word-level rules ≠ postlexical rules in Lexical Phonology 106 →boundaries are [-segment] segments without →phonetic correlate 88 feature [±segmental] heralded in Chomsky et al. (1956) and Halle (1959) 78 different types: #, +, = 89 boundary erasure (at the end of the derivation) 90 boundary mutation rules (readjustment rules) 112 non-privative boundary management (boundaries present without phonological effect) 90 = boundary without posterity 89 reduction of boundary clusters to two 89f reduction of the number of boundaries in post-SPE (boundary economy / abuse) 129 mapping mapping algorithm (insertion of #) 90 mapping: # restores full morphosyntactic information except for consecutive affixes of the same major category 90

SPE 81 (continued) affix classes representational management 92 rule-blocking boundaries 93 rule-triggering boundaries 94 representational vs. Lexical Phonology's procedural management 106 post-SPE frustration with boundaries 110, 138 overview main strands until the 80s 110 socialist revision (Kiparsky) vs. communist revolution (Stampe) 125f restoration by Halle & Vergnaud (1987a) 217, 219 dismissal of Praguian segregation 238 interpretational units 242 spell-out →selective spell-out →affix-triggered interpretation →phase theory/ phasehood Kaye (1995): domain structure is richer than what spell-out can produce 274 of terminals: procedural first 316 asymmetric spell-out (independent LFPF), summary 779 spell-out mechanism →word-spell-out mystery →Phase Impenetrability/ PIC à la carte asymmetric spell-out (independent LFPF), summary 779 common properties of morpho-syntax and phonology due to spell-out? 21 just one (morphology and syntax are the same computational system), or two (they are not)? 858 →Phase Impenetrability, a property of the spell-out mechanism, rather than of syntax or phonology 799

832 Subject index spell-out mechanism (continued) who decides to ignore "old" strings (memory keeper)? 21, 307, 799 spell-out-as-you-merge Epstein et al. (1998), extreme version of →derivation by phase 305, 775-77 spurious cluster Kaye (1995) 343 storage vs. computation no distinction in →connectionism 596 strata (lexical) in →Lexical Phonology: general introduction 147 definition 147-52 procedurally ordered (level ordering) 147 stratal (Lexical Phonology) vs. noninteractionist (Halle & Vergnaud 1987a) architecture 218, 235 stratal (Lexical Phonology) vs. noninteractionist (Halle & Vergnaud 1987a) architecture 251 testing ground 1: lexical phonology sensitive to syntactic information? 252 testing ground 2: phonology-free syntax 253 Stratal OT general presentation 483 morpheme-specific mini-phonologies in OT: parallel vs. reranked incarnations 476 does away with the →SPE-legacy of →Lexical Phonology 489 concern for unrestricted access to morpho-syntactic information 490 and modularity Indirect Reference yes, but ALIGN is allowed to violate this principle 491 favours clear-cut modular contours, but still violates modularity 527 how different can mini-grammars be? 492

Stratum Domain Hypothesis →domain assignment (→Lexical Phonology) stress a strange guy: strategies devised to cope with it 556 and structure preservation: the Free Element Condition 295 stress clash 233, note 64 (240), 302 evidence for process-specific PIC (→phase theory, PIC à la carte) 556 independent of word stress assignment: two distinct processes, note 64 (240), 820 not necessarily a modification of word stress 820 Stress Copy Halle & Vergnaud (1987a) 231 Halle & Kenstowicz (1991), compensation vs. condensation 549 Stress Erasure Convention Halle & Vergnaud (1987a) 230 Halle & Kenstowicz (1991), compensation vs. condensation 549 stress placement problem child: particularly apt to violate →no look-back and the →Strict Cycle Condition 315, 556, 814 transparent vs. opaque 311 stress-shifting vs. stress-neutral affixes →affix classes, note 32 (142) Strict Cycle Condition (SCC) →derived environment effects →no look-back devices →Lexical Phonology properties Chomsky's (1973) original version: "use new material!" 289 fusion of Chomsky's SCC and derived environments 178, 188 genesis and birth Chomsky's (1973) original version: "use new material!" 289

Subject index 833 Strict Cycle Condition (SCC) (continued) genesis and birth (continued) overview and properties 188 import of Chomsky's (1973) device into phonology: Kean (1974), Mascaró (1976) 189, 290 Kiparsky sells Mascaró's SCC for what it isn't (Mascaró's SCC has got nothing to do with derived environments) 190 Halle (1978) combines Chomsky's SCC and derived environments, but his contribution went unnoticed 190 birth: fusion of Chomsky's SCC (→no look-back) and derived environments 190, 291 confusion disentangled: SCC-K(iparsky) vs. SCCM(ascaró) 195 often confused: Chomsky's (1973) original and Halle/Kiparsky's version 291 instrument against overgeneration 144 further career posterity of 200 adopted by Halle & Vergnaud (1987a) 237 Kiparsky's SCC is void of empirical content 195 bankruptcy of, declared by Kiparsky (1993) 197 eliminated in Stratal OT 489 in the architecture of Lexical Phonology and lexicalism 192 and Praguian segregation 192 lexical post-cyclic rules (Rubach & Booij 1984) 194 can do the rule-triggering pattern (but this is ignored in the literature) 203 stratum-specific cyclicity / respect of the SCC 194, 233 violation of and derived environments correlation between derived environments and lexical vs. postlexical rules 192

Strict Cycle Condition (SCC) (continued) violation of and derived environments (continued) counterevidence for "lexical = applies only to derived environments" 192 no correlation between the lexical character of rules and their application to underived items 194 violated by stress assignment (English) 556 violation of: structure-building vs. structure-changing rules SCC-K weakened 193, note 53 (199) derived environment created by structure-building rules? 193 structure-changing rules that apply in the Lexicon and to underived items 194 Halle & Vergnaud (1987a) 237 English →stress placement is a problem child 556, 814 Strict Cyclicity Chomsky (1973), →Strict Cycle Condition, genesis and birth Strict Layer Hypothesis classical (monolithic) and OTed (made of 4 independent constraints) 383 and privativity 759 in OT: decomposed into four constraints (Selkirk 1996) 461 allows for recursive prosodic structure since Selkirk (1996) 802 argument by Hayes (1989 [1984]) against boundaries based on the SLH 373 string theory (physics) 592 Structural Analogy Dependency Phonology, Government Phonology 705 morpheme-specific mini-phonologies vs. no look-back (PIC): refereed in favour of the PIC 830 violation of modularity? 705 structuralism (American) 59

834 Subject index structure preservation →Lexical Phonology (eliminated in →Stratal OT) 489 early modification-inhibiting no lookback (stress, syllabification) 193, 295-97 Chomsky's "spell-out and forget" too strong for phonology: two candidates for a weaker version 302 structure vs. labels morpho-syntactic structure may, but content (labels) may not bear on phonology (?) 752 structure-building vs. structure-changing rules →Strict Cycle Condition/ violation of Structure-Constrained Modularity (Jackendoff) alternative to the inverted T model 644, 723 structured modular programming modern computers based on the standard model of Cognitive Science 603 subcomponents (→inverted T) vs. subsystems (GB modules) →modularity/ in GB 631 successive cyclic movement linearity-driven (Fox & Pesetsky 2004) 744 superheavy rhymes banned in Government Phonology, parsing cue for empty nuclei 343 suppletion lexical structure of go - went: look-up (→Kaye 1995) 354 suppletion, allomorphy in →Distributed Morphology 571 syllabification algorithm in →Government Phonology, note 165 (714) syllable structure arboreal vs. lateral account (→deforestation) 43 early modification-inhibiting →no look-back 297 symbolic representations in OT (connectionist roots of) 529

symbolic vs. anti-symbolic views of the cognitive system introduction 591 adult science without symbols? 592, 594 syntactic hierarchical structure all the way down a slogan of Marantz' (DM) 535 syntactico-centrism Jackendoff argues against 723 syntax vs. morphology →morphology vs. syntax (one or two computational systems?) syntax-first model Selkirk (1984), syntax = morphology, contra Lexical Phonology 540 Synthesis Psychology New Synthesis Psychology: Pinker, Plotkin 607

T templates work by Bendjaballah & Haiden 724 terminal nodes spell-out in isolation of 274 spell-out in isolation of: procedural first 316 theory-external ways to get a handle on interface theory history, modularity, intermodular argumentation 845 third factor explanations →biolinguistics empties UG 633 eliminate GB-modules (→modularity/ in GB) 638f minimalism: extra-linguistic explanations 298 applied to phonology (Samuels 2009a,b) 639 tone is not melodic: it can be the output of translation (Manfredi 2010) 662 traces (syntactic) are able to influence phonology? 372 transformational cycle →cyclic derivation (inside-out) 673 transistor (computer) 592

Subject index 835 translation (insertion of a carrier of morpho-syntactic information) summary 692 doing away with: →direct syntax 134 phonological theories refereed by their (absurd) behaviour at the interface 701 among cognitive modules by computation (Translator's Office) vs. by lexical access (dictionary) 652, 655 is selective, the choice of translated pieces is arbitrary 654 vocabulary (as opposed to structure) excluded from translation? 661 translation of vocabulary (input) vs. structure (result of modular activity) 652 structure, but not vocabulary, is translated 653 enforced by domain specificity 413 of morpho-syntax into phonology local vs. non-local encoding of morpho-syntactic information 405 Indirect Reference is a consequence of domain specificity 650 outlook Vol.2: One Channel Translation 655 history and evolution invented by structuralism (→Level Independence) 699 always practised, even in absence of modular background 693 output: always the representational vocabulary of the current theory 5, 72 in SPE 84f post-SPE, output: non-diacritics, i.e. truly phonological objects 134 summary post-SPE 138 output has always been a →diacritic overview: juncture phonemes, boundaries 695

translation (insertion of a carrier of morpho-syntactic information) (continued) output has always been a →diacritic (continued) a non-diacritic output enforces privative translation 761 privativity: underfeeding of phonology is an undisputed fact, but all theories since SPE are non-privative 757 generative modularity offenders: reference to untranslated information (summary) 702 into non-diacritic phonological objects nature of the output: diacritics do not qualify 655 a structuralist invention 72 →Direct Interface: output of translation must be vocabulary items that are used in phonology in absence of extraphonological factors 655 possible outputs reduce to syllabic space 717 morpho-syntax and melody are incommunicado 660 morpho-syntax cannot read or parse melody 662 carriers of morpho-syntactic information do not include melody 663 (phonological) melody is never the output of 653 tone is not melodic: it can be the output of translation (Manfredi 2010) 662 trees →deforestation two different tree-building devices (morpho-syntax and phonology)? (→morphology vs. syntax) 45 Trubetzkoy's Grenzsignale 55 Turing (von Neumann) machine (universal) 595f, 603 two-place approach of Lexical Phonology (Marantz 2000) (→Distributed Morphology, →morphology vs. syntax) 534f

836 Subject index U underapplication morpheme-specific phonologies and selective spell-out are incompatible 303 in →Lexical Phonology 156 →affix class-based phenomena derived by the underapplication to →morpheme-specific strings comparison analyses in →Lexical Phonology, →Halle & Vergnaud (1987a), →Kaye (1995) 303 impossible to achieve in Distributed Morphology 559 →rule-blocking pattern (→level 1 strings) Lexical Phonology 168 Halle & Vergnaud (1987a): achieved by →selective spellout 227f, 249 Kaye (1995): achieved by no lookback, comparison with other models 279, 312 Lexical Phonology 164 →rule-triggering pattern (→level 2 strings) Kaye (1995): impossible without additional condition (stringfinality in English) 322 Halle & Vergnaud (1987a): cannot be done 250 in Lexical Phonology: eliminated in Stratal OT 489 →derived environments created by structure-building rules? 193 Kiparsky's (1993) solution for →derived environment effects 198 Universal Turing machine 595f, 603 usage-based theories 596

V via-rules Natural Generative Phonology and Kaye (1995): structure of lexical entries keep - kept 353 vocabulary (items) in →Distributed Morphology 536

vocabulary insertion in →Distributed Morphology 536 von Neumann-Turing model computation 603

W Williams Syndrome double dissociation of language 621 word possible word: blick vs. lbick (Kaye 1995) 349 SPE: representational definition (##...##) 105 autonomous unit the word a natural barrier? 238f, 257, 302 absence of a sealing mechanism for words in theories where word phonology also applies to word sequences 240 and modification-inhibiting →no look-back 302 SPE 105 Lexical Phonology 153 Halle & Vergnaud (1987a) 239 process-specific treatment by Praguian segregation 240 word formation phonology sensitive to syntactic information? 252 word phonology →computational systems, →word/ autonomous unit word-final clusters heteromorphemic: analysis with →domain-final →empty nuclei 331 unlimited by the phonology: only morphology imposes restrictions (→Kaye 1995) 332 word-formation based on roots vs. on existing words 546 word-level phonology →computational systems, →no lookback devices, PIC à la carte (chunk-specific), →Phase Impenetrability

Subject index 837 word-level rules →SPE 105 SPE, (absence of) in Lexical Phonology 154 SPE: representational definition of the word (##...##) 105 = non-cyclic rules in Halle & Vergnaud (1987a) 219 Lexical Phonology: ≠ postlexical rules 106 Kaye (1995) 283, 338 word-spell-out mystery 786, 794 overview 22 stands in the way of →intermodular argumentation 861 facts: what the mystery is absence of cyclicity-induced external sandhi: a fact that nobody talks about 787 empirical generalisation wrong? 796 external →sandhi has only representational treatments (Prosodic Phonology) 787 the paradox: why does phonology refuse to react on piecemeal fire? 795 cyclic interpretation of word sequences? baseline position of SPE: both spell-out and interpretation are cyclic 789 Lexical Phonology: non-cyclic interpretation of word sequences imposed without argument or discussion 158, 790f

word-spell-out mystery 786, 794 (continued) cyclic interpretation of word sequences? (continued) Halle & Vergnaud (1987a) and Kaye (1995) restore SPE: both spell-out and interpretation are cyclic 792 DM is agnostic 793 how theories react in Lexical Phonology: why is postlexical phonology noncyclic? 156, 158, 790f Strict Cycle Condition (SCC) 192 an argument in favour of Praguian segregation 817 word-spell-out mystery 786, 794 (continued) how theories react (continued) Praguian segregation is not good enough to explain what is going on 808 analysis: PIC à la carte chunk-specific PIC (summary) 796f, 809 (chunk-specific) PIC à la carte in Samuels (2009a), but hidden by a specific formulation 798 the PIC is managed by the spellout mechanism 799 working (workbench, active) memory 596 Wrap (OT) →alignment (OT)

X, Y, Z Y model →inverted T

865

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. – cross-reference to other entries of the language index are introduced by "→": look up here. – reference to footnotes is made by indicating their number, but also by mentioning the number of the § in which they occur: "note 258 (415)" refers to footnote 258 which occurs in §415. A, B Berber (Kabyle) negative marker, analysis by Bendjaballah (2001) 713 Bulgarian clitic placement, (PF-based) analysis by Franks & Bošković (2001) 730

C, D Catalan voicing (like Cracow/Poznań voicing → Polish): anti-cyclic (outside-in) processes 822 Chi Mwi:ni (Bantu) prosodic phrasing, analyses by Kisseberth & Abasheikh (1974), Hayes (1989 [1984]), Selkirk (1986) 396

E English affix class-based phenomena →process-blocking pattern (level 1 rules)

English (continued) affix class-based phenomena (continued) process-triggering pattern (level 2 rules): →nasal cluster simplification - four varieties: 1) →n-deletion (damn, damning2 vs. damn-ation1), 2) →g-deletion (sign, sign-ing2 vs. sign-ature1), 3) →post-nasal b-deletion (bomb, bomb-ing2 vs. bombard1), 4) →post-nasal g-deletion (sing, sing-ing2 vs. long-er1) aspiration example of an "automatic process" 186 Kiparsky (1993) 199 bracketing paradoxes, allomorph selection: un-happier vs. *beautifuller anti-interactionist ammunition 243 and independent spell-out of terminals 319

Language index 839 English (continued) bracketing paradoxes, allomorph selection: un-happier vs. *beautifuller (continued) analysis by Rubach & Booij (1984) 440 category-sensitive stress: récord (noun) vs. recórd (verb) analysis by Nespor & Vogel (1986) 438 melody-free syntax 752 complex word-final clusters (sixths, röntgst) analysis in Government Phonology with empty nuclei 331 compounds progressive erosion of internal structure 339 stress and vowel reduction: dialectal and idiolectal variation due to variable domain structure (Kaye 1995) 335-37 black bóard vs. bláckboard, analysis in SPE 96 epenthesis of schwa: wickəd, nakəd analysis by Kaye (1995) 344 flapping (cross-word phonology) 241 g-deletion gn-n: sign, sign-ing2 vs. sign-ature1 SPE: boundary contrast longer vs. longer "person who is longing" 118 analysis in SPE 94 facts and analysis in Lexical Phonology 167 great vowel shift Kiparsky: example of an "automatic process" 186 intonation (sentence stress) Nuclear Stress Rule (NSR), analysis by Bresnan (1971) 308 cat-rat-cheese: readjustment in SPE 91 cat-rat-cheese, analysis by SPE and Nespor & Vogel (1986) 418

English (continued) linking r across sentence boundaries, analysis by Nespor & Vogel (1986) 418 nasal assimilation in- vs. unfacts and analysis in SPE and Lexical Phonology 164 sensitive to affix classes, but not to derived environments 182 phonological and morpho-syntactic effects of the asymmetry 319 analysis by Nespor & Vogel (1986) 438 analysis by Halle & Vergnaud (1987a) 249 analysis in Kaye's system 314 nasal cluster simplification four varieties: 1) →n-deletion (damn, damning2 vs. damn-ation1), 2) →g-deletion (sign, sign-ing2 vs. sign-ature1), 3) →post-nasal b-deletion (bomb, bomb-ing2 vs. bombard1), 4) →post-nasal g-deletion (sing, sing-ing2 vs. long-er1) post-nasal plosive deletion (postnasal b- and g-deletion): pattern suffers exceptions, note 43 (167) analysis by Halle & Mohanan (1985): stratum-specific cyclicity/SCC 194 analysis by Kaye (1995): based on the domain-final condition 322 analysis by Kaye (1995): why do the clusters implode? 324 n-deletion mn-m: damn, damn-ing2 vs. damn-ation1 SPE 94 facts and analysis in Lexical Phonology 167f night rate vs. nitrate 63

840

Language index

English (continued) opacity, simultaneous PF and LF (cómparable vs. compárable) analysis in Distributed Morphology 545 no opacity in a lexicalist perspective 573 post-nasal b-deletion mb-m: bomb, bomb-ing2 vs. bomb-ard1 analysis in SPE 94 facts and analysis in Lexical Phonology 167f pattern suffers exceptions, note 43 (167) post-nasal g-deletion ŋg-ŋ: sing, sing-ing2 vs. long-er1 facts and analysis in Lexical Phonology 167f Kiparsky: example of an "automatic process" 186 cannot be done by Halle & Vergnaud (1987a) 250 pattern suffers exceptions, note 43 (167) process-blocking pattern (párent parént-al - párent-hood) summary triggering conditions 311 boundary invisible 51 analysis in SPE 93 strata-based analysis in Lexical Phonology 147 analysis in Halle & Vergnaud (1987a) 227-29 analysis by Kaye (1995): no lookback 279 Rhythmic Rule →stress clash sandhi (external) t → tÉʃ / __ j: I know what you want vs. I know what ewe I want, Kaye (1995) 263 flapping 241 linking r, analysis by Nespor & Vogel (1986) 418 stress (sentence) →intonation

English (continued) stress (word) 814 4 stress phonemes (Chomsky et al. 1956) 74 black bóard vs. bláckboard, analysis in SPE (→compounds) 96 and →vowel reduction: dialectal and idiolectal variation due to variable domain structure (Kaye 1995) 335-37 English stress (predictable) vs. Vedic stress (lexical), analysis by Halle & Vergnaud (1987a) 237 not frozen when class 1 affixes are added (órigin - oríginal), analysis in Distributed Morphology (Marvin 2002) 555 secondary stress (compensation vs. condensation, →vowel reduction), analysis in SPE 97 →process-blocking pattern sealed at the word level 240 example of an "automatic process" 186 →stress clash stress clash not the same process as stress placement, note 64 (240) evidence for process-specific PIC 556 trouble for "don't undo" (processspecific no look-back) 302 domains vs. boundaries 374 incidence on Marvin's (2002) analysis of word stress 556 redistribution of prominence among primary, secondary etc. word stresses, rather than change of primary stress 820 theatricality derivation in SPE (3 cycles) 103 trisyllabic shortening literature, note 28 (127) sensitive to both derived environments and affix classes 182 SPE, reanalysis by Kiparsky (1982) 129

Language index 841 English (continued) trisyllabic shortening (continued) facts and analysis in SPE and Lexical Phonology 164 derived by the Elsewhere Condition 191 analysis by Halle & Vergnaud (1987a) 237, 249 analysis by Kaye (1995): lies outside of phonology 276 analysis in Kaye's system 313 velar softening (electri[k] vs. electri[s]ity) literature, note 28 (127) abstractness debate, (anti-)lexicalism 570 Kiparsky: example of an "automatic process" 186 analysis by Kaye (1995): not phonology 276 vowel reduction analysis in SPE (word-level rule) 105 and stress: dialectal and idiolectal variation due to variable domain structure (Kaye 1995) 335-37 vowel reduction, stress-conditioned (condensation vs. compensation) analysis in SPE 548 analysis by Halle & Kenstowicz (1991) 549 analysis in Distributed Morphology (Marvin 2002) 550f vowel shortening mean - meant vs. paint, pint (Kiparsky 1985a) 129 word stress →stress (word)

F Finnish t → s / __i derived environment effect, analysis by Kiparsky (1973a,b) 180 analysis by Kiparsky (1993) 198

Flemish (West) voicing (like Cracow/Poznań voicing → Polish): anti-cyclic (outside-in) processes 822 French gender markers, possessive pronouns (mon armoire, *m'armoire) analysis by Lowenstamm (2008) 579 gliding: prefix-suffix asymmetry li-j-a vs. *anti-j-alcoolique 53 intrusive t in French (tennis-t-ique) analysis by Pagliano (2003) 713 liaison Selkirk (1972), boundary mutation rules 112 nasal deletion in in-: /in-légal/ → illégal SPE, Selkirk (1972) 122 vowel nasalisation: mon ami vs. bon ami analysis by Prunet (1986) / Kaye (1995) 295

G, H, I, J German aspiration analysis by Moulton (1947) 64 complex word-final clusters (sixths, röntgst) analysis in Government Phonology with empty nuclei 331 distribution of X-ç analysis by Moulton (1947) 64 glottal stop analysis by Moulton (1947) 64 glottal stop insertion at prefix boundary (auf-ʔessen) analysis by Kleinhenz (1998) 448, 459 tense marker in German strong verbs analysis by Bendjaballah & Haiden (2003a,b) 713 voicing eliminated: Teil = /d#eil/ (Harris 1951) 67

842

Language index

Greek agreement PIC-violating long-distance agreement 780

K Korean (word-initial) stop voicing in external sandhi analysis by Cho (1990) 391 affrication before i analysis by Oostendorp (2006) 508

L Latin gemination due to morpheme-internal juncture (Hill 1954, juncture abuse) 69 stress placement with enclitic elements: liminá-que 295

M, N Macedonian clitic placement, (PF-based) analysis by Franks & Bošković (2001) 730

O Ojibwa (Eastern Algonquian) hiatus avoidance analysis by Piggott & Newell (2006) 578 Old English fricative voicing Lass (1971): # = [-voice] 135

P, Q Polish Cracow (Poznań) voicing anti-cyclic (outside-in) processes 822 palatalisation derived environment effect, analysis by Kiparsky (1973a,b) 181 analysis with bracket-sensitive rules (Mohanan) 202

Polish (continued) velar palatalisation in (phonologically) derived environments: analysis by RubachŁubowicz (brydż-ek) 518 Puyo Pongo (Quicha, Eastern Ecuador) obstruent voicing after heteromorphemic nasals 54

R Romanian agreement PIC-violating long-distance agreement 780

S Spanish nasal assimilation in external sandhi in Spanish analysis by Nespor & Vogel (1986) 420 s → h in codas analysis by Harris (1993): structure preservation 297

T, U Tagalog (Austronesian, Philippines) infixation in CC-initial loans analysis by Zuraw (2007) 412 Turkish disharmonic roots analysis by Oostendorp (2006) 508 minimal word constraint analysis by Orgun 514 Tzeltal (Maya) infixation analysis by Samuels (2009) 746

V, W, X, Y, Z Vedic stress (word) English stress (predictable) vs. Vedic stress (lexical), analysis by Halle & Vergnaud (1987a) 237

866

Index of phenomena

– 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. – cross-reference to other entries of the index of phenomena are introduced by "→": look up here. – reference to footnotes is made by indicating their number, but also by mentioning the number of the § in which they occur: "note 258 (415)" refers to footnote 258 which occurs in §415. A affix class-based phenomena (in English) →process-blocking pattern (level 1 rules) process-triggering pattern (level 2 rules, →nasal cluster simplification): four varieties 1) →deletion of n (damn, damning2 vs. damn-ation1) 2) →deletion of g (sign, sign-ing2 vs. sign-ature1) 3) →deletion of b in post-nasal position (bomb, bomb-ing2 vs. bomb-ard1) 4) →deletion of g in post-nasal position (sing, sing-ing2 vs. long-er1) affrication before i in Korean analysis by Oostendorp (2006) 508 agreement in (Mod.) Greek and Romanian PIC-violating long-distance agreement 780 anti-cyclic (outside-in) processes Cracow/Poznań voicing, also in Catalan and West Flemish 822 aspiration (English) example of an "automatic process" 186 Kiparsky (1993) 199

aspiration (German) analysis by Moulton (1947) 64 aspiration (s → h) in codas (Spanish) analysis by Harris (1993): structure preservation 297 assibilation in Finnish: t → s / __i analysis by Kiparsky (1993) 198 derived environment effect, analysis by Kiparsky (1973a,b) 180

B bracketing paradoxes, allomorph selection in English: un-happier vs. *beautifuller anti-interactionist ammunition 243 and independent spell-out of terminals 319 analysis by Rubach & Booij (1984) 440

C category-sensitive stress in English: récord (noun) vs. recórd (verb) analysis by Nespor & Vogel (1986) 438 melody-free syntax 752 clitic placement in Bulgarian and Macedonian (PF-based) analysis by Franks & Bošković (2001) 730

844 Index of phenomena compounds (English) black bóard vs. bláckboard, analysis in SPE 96 progressive erosion of internal structure 339 stress and vowel reduction: dialectal and idiolectal variation due to variable domain structure (Kaye 1995) 335-37

D deletion of b in post-nasal position (English): mb-m bomb, bomb-ing2 vs. bombard1 analysis in SPE 94 facts and analysis in Lexical Phonology 167f pattern suffers exceptions, note 43 (167) of g in post-nasal position (English): ŋg-ŋ sing, sing-ing2 vs. long-er1 facts and analysis in Lexical Phonology 167f Kiparsky: example of an "automatic process" 186 cannot be done by Halle & Vergnaud (1987a) 250 pattern suffers exceptions, note 43 (167) of g: gn-n sign, sign-ing2 vs. signature1 (English) SPE: boundary contrast longer vs. longer "person who is longing" 118 analysis in SPE 94 facts and analysis in Lexical Phonology 167 of n: mn-m damn, damn-ing2 vs. damn-ation1 (English) SPE 94 facts and analysis in Lexical Phonology 167f disharmonic roots in Turkish analysis by Oostendorp (2006) 508

E ellipsis, sluicing (PF-based) analyses 731 epenthesis of schwa in English: wickəd, nakəd analysis by Kaye (1995) 344 extraposition motivation for PF movement 575, 577

F first vowel of words alternation with zero of, analysis with an empty CV unit 685 flapping in English (cross-word phonology) 241 fricative voicing in Old English Lass (1971): # = [-voice] 135

G gemination (Latin > Italian) due to morpheme-internal juncture (Hill 1954, juncture abuse) 69 gender markers, possessive pronouns in French (mon armoire, *m'armoire) analysis by Lowenstamm (2008) 579 gliding in French: prefix-suffix asymmetry li-j-a vs. *anti-j-alcoolique 53 glottal stop in German analysis by Moulton (1947) 64 glottal stop insertion at prefix boundary in German(auf-ʔessen) analysis by Kleinhenz (1998) 448, 459 great vowel shift (English) Kiparsky: example of an "automatic process" 186

H heavy NP shift motivation for PF movement, note 142 (577) hiatus avoidance in Ojibwa (Eastern Algonquian) analysis by Piggott & Newell (2006) 578

Index of phenomena 845

I, J, K

N

ich- [ç] and ach [X] Laut in German: distribution analysis by Moulton (1947) 64 infixation cross-linguistic conditioned by phonology (but not by melody) 412 in CC-initial loans in Tagalog (Austronesian, Philippines) analysis by Zuraw (2007) 412 in Tzeltal (Maya) analysis by Samuels (2009) 746 intervocalic voicing boundary-defined domains of rule application (McCawley 1968) 114 intonation (sentence stress) Nuclear Stress Rule (NSR) in English, analysis by Bresnan (1971) 308 cat-rat-cheese (English): readjustment in SPE 91 cat-rat-cheese (English), analysis by SPE and Nespor & Vogel (1986) 418 intrusive t in French (tennis-t-ique) analysis by Pagliano (2003) 713

nasal assimilation in English: in- vs. unfacts and analysis in SPE and Lexical Phonology 164 sensitive to affix classes, but not to derived environments 182 phonological and morpho-syntactic effects of the asymmetry 319 analysis by Nespor & Vogel (1986) 438 analysis by Halle & Vergnaud (1987a) 249 analysis in Kaye's system 314 in external sandhi in Spanish analysis by Nespor & Vogel (1986) 420 nasal cluster simplification (English) four varieties 1) →deletion of n (damn, damning2 vs. damn-ation1) 2) →deletion of g (sign, sign-ing2 vs. sign-ature1) 3) →deletion of b in post-nasal position (bomb, bomb-ing2 vs. bomb-ard1) 4) →deletion of g in post-nasal position (sing, sing-ing2 vs. long-er1) post-nasal plosive deletion (post-nasal b- and g-deletion): pattern suffers exceptions, note 43 (167) analysis by Halle & Mohanan (1985): stratum-specific cyclicity/SCC 194 analysis by Kaye (1995): based on the domain-final condition 322 analysis by Kaye (1995): why do the clusters implode? 324 nasal deletion in French in-: /in-légal/ → illégal SPE, Selkirk (1972) 122 nasalisation of vowels in French: mon ami vs. bon ami analysis by Kaye (1995) 295

L liaison (in French) Selkirk (1972), boundary mutation rules 112 linking r in English across sentence boundaries, analysis by Nespor & Vogel (1986) 418 locality conditions in phonology, similar to syntax (Truckenbrodt 1995) 577

M minimal pairs (in structuralism): night rate vs. nitrate (English) 63 minimal word constraint in Turkish analysis by Orgun 514

846 Index of phenomena negative marker in Kabyle Berber analysis by Bendjaballah (2001) 713

O opacity, simultaneous PF and LF (English: cómparable vs. compárable) analysis in Distributed Morphology 545 no opacity in a lexicalist perspective 573

P, Q palatalisation in Polish analysis with bracket-sensitive rules (Mohanan) 202 derived environment effect, analysis by Kiparsky (1973a,b) 181 possessive pronouns, gender markers in French (mon armoire, *m'armoire) analysis by Lowenstamm (2008) 579 process-blocking pattern (English párent parént-al - párent-hood) summary triggering conditions 311 boundary invisible 51 analysis in SPE 93 strata-based analysis in Lexical Phonology 147 analysis in Halle & Vergnaud (1987a) 227-29 analysis by Kaye (1995): no look-back 279 prosodic phrasing in Chi Mwi:ni (Bantu) analyses by Kisseberth & Abasheikh (1974), Hayes (1989 [1984]), Selkirk (1986) 396

R Rhythmic Rule (in English) →stress clash

S sandhi (external) English t → tÉʃ / __ j: I know what you want vs. I know what ewe I want, Kaye (1995) 263 flapping in English 241 linking r in English, analysis by Nespor & Vogel (1986) 418

sluicing, ellipsis (PF-based) analyses 731 stress (sentence) →intonation stress (word) in English 814 →process-blocking pattern →stress clash 4 stress phonemes (Chomsky et al. 1956) 74 black bóard vs. bláckboard, analysis in SPE (→compounds) 96 and →vowel reduction: dialectal and idiolectal variation due to variable domain structure (Kaye 1995) 335-37 English stress (predictable) vs. Vedic stress (lexical), analysis by Halle & Vergnaud (1987a) 237 example of an "automatic process" 186 not frozen when class 1 affixes are added (órigin - oríginal), analysis in Distributed Morphology (Marvin 2002) 555 sealed at the word level 240 secondary stress (compensation vs. condensation, →vowel reduction), analysis in SPE 97 stress clash (in English) domains vs. boundaries 374 evidence for process-specific PIC 556 incidence on Marvin's (2002) analysis of word stress 556 not the same process as stress placement, note 64 (240) redistribution of prominence among primary, secondary etc. word stresses, rather than movement of primary stress 820 trouble for "don't undo" (processspecific no look-back) 302 stress placement in Latin with enclitic elements: liminá-que 295

T, U tense marker in German strong verbs analysis by Bendjaballah & Haiden (2003a,b) 713

Index of phenomena 847 trisyllabic shortening (English) literature, note 28 (127) sensitive to both derived environments and affix classes 182 SPE, reanalysis by Kiparsky (1982) 129 facts and analysis in SPE and Lexical Phonology 164 derived by the Elsewhere Condition 191 analysis by Halle & Vergnaud (1987a) 237, 249 analysis by Kaye (1995): lies outside of phonology 276 analysis in Kaye's system 313

V velar palatalisation in Polish in (phonologically) derived environments: analysis by RubachŁubowicz (brydż-ek) 518 velar softening (English: electri[k] vs. electri[s]ity) literature, note 28 (127) abstractness debate, (anti-)lexicalism 570 Kiparsky: example of an "automatic process" 186 analysis by Kaye (1995): not phonology 276 voicing in German eliminated: Teil = /d#eil/ (Harris 1951) 67 in Korean: word-initial stops in external sandhi analysis by Cho (1990) 391 in Polish (Cracow / Poznań voicing), Catalan and West Flemish anti-cyclic (outside-in) processes 822 of obstruents after heteromorphemic nasals in Puyo Pongo (Quicha, Eastern Ecuador) 54

vowel nasalisation in French: mon ami vs. bon ami analysis by Prunet (1986) / Kaye (1995) 295 vowel reduction in English analysis in SPE (word-level rule) 105 and stress: dialectal and idiolectal variation due to variable domain structure (Kaye 1995) 335-37 in English, stress-conditioned (condensation vs. compensation) analysis in SPE 548 analysis by Halle & Kenstowicz (1991) 549 analysis in Distributed Morphology (Marvin 2002) 550f vowel shortening (English) mean - meant vs. paint, pint (Kiparsky 1985a) 129

W, X, Y, Z Weight Increase analysis of heavy NP movement (Selkirk 2001), note 142 (577) word stress (in English) →stress (word) word-final clusters, complex (e.g. English, German: sixths, röntgst) analysis in Government Phonology with empty nuclei 331 word-initial consonant clusters restrictions of, analysis with an empty CV unit 685 word-initial consonants, strength of analysis with an empty CV unit 685