Superparticles: A Microsemantic Theory, Typology, and History of Logical Atoms [1st ed.] 9789402420494, 9789402420500

Table of contents :
Front Matter ....Pages i-xix
Introduction (Moreno Mitrović)....Pages 1-13
Construction (Moreno Mitrović)....Pages 15-105
Interpretation (Moreno Mitrović)....Pages 107-188
Grammaticalisation (Moreno Mitrović)....Pages 189-253
Conclusion (Moreno Mitrović)....Pages 255-266
Back Matter ....Pages 267-288

Citation preview

Studies in Natural Language and Linguistic Theory 98

Moreno Mitrović

Superparticles A Microsemantic Theory, Typology, and History of Logical Atoms

Studies in Natural Language and Linguistic Theory VOLUME

98

Managing Editors Marcel den Dikken, Research Institute for Linguistics, Hungarian Academy of Sciences and Department of English Linguistics, Eötvös Loránd University, Budapest, Hungary Liliane Haegeman, University of Gent, Gent, Belgium Maria Polinsky, University of Maryland, College Park, MD, USA Editorial Board Guglielmo Cinque, University of Venice, Venice, Italy Jane Grimshaw, Rutgers University, New Brunswick, USA Michael Kenstowicz, Massachusetts Institute of Technology, Cambridge, MA, USA Hilda Koopman, University of California, Los Angeles, USA Howard Lasnik, University of Maryland, College Park, MD, USA Alec Marantz, New York University, New York, USA John J. McCarthy, University of Massachusetts, Amherst, MA, USA Ian Roberts, University of Cambridge, Cambridge, UK

Studies in Natural Language & Linguistic Theory provides a forum for the discussion of theoretical research that pays close attention to natural language data, offering a channel of communication between researchers of a variety of points of view. Like its associated journal Natural Language & Linguistic Theory, http://www.springer.com/journal/11049, the series actively seeks to bridge the gap between descriptive work and work of a highly theoretical, less empirically oriented nature. The series editors invite proposals for monographs and edited volumes. For more information on how to submit your proposal, please contact the assistant editor, Anita Rachmat, E-mail: [email protected]

More information about this series at http://www.springer.com/series/6559

Moreno Mitrovi´c

Superparticles A Microsemantic Theory, Typology, and History of Logical Atoms

Moreno Mitrovi´c Leibniz Centre for General Linguistics (ZAS), Berlin & Bled Institute Berlin, Germany

ISSN 0924-4670 ISSN 2215-0358 (electronic) Studies in Natural Language and Linguistic Theory ISBN 978-94-024-2049-4 ISBN 978-94-024-2050-0 (eBook) https://doi.org/10.1007/978-94-024-2050-0 © Springer Nature Switzerland AG 2021 This work is subject to copyright. All rights are reserved by the Publisher, whether the whole or part of the material is concerned, specifically the rights of translation, reprinting, reuse of illustrations, recitation, broadcasting, reproduction on microfilms or in any other physical way, and transmission or information storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar methodology now known or hereafter developed. The use of general descriptive names, registered names, trademarks, service marks, etc. in this publication does not imply, even in the absence of a specific statement, that such names are exempt from the relevant protective laws and regulations and therefore free for general use. The publisher, the authors, and the editors are safe to assume that the advice and information in this book are believed to be true and accurate at the date of publication. Neither the publisher nor the authors or the editors give a warranty, expressed or implied, with respect to the material contained herein or for any errors or omissions that may have been made. The publisher remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. This Springer imprint is published by the registered company Springer Nature Switzerland AG The registered company address is: Gewerbestrasse 11, 6330 Cham, Switzerland

To the academic precariat.

Preface

This book is all about the captivating ability that the human language has to express intricately logical (mathematical) meanings using tiny (microsemantic) morphemes as utilities. Languages mark meanings with identical inferences using identical particles and these particles thus creep up in a wide array of expressions. Because of their multi-tasking capacity to express seemingly disparate meanings, I dub them superparticles. These particles are perfect windows into the interlock of several grammatical modules and the nature of the interaction of these modules through time. With a firm footing in the module where grammatical bones are built and assembled (narrow morphosyntax), superparticles acquire varied interpretation (in the conceptual-intentional module—semantics) depending on the structure they feature in. What is more, some of the interpretations these particles trigger are inferential and belong, under the standard account, to the realm of pragmatics. How can such tiny particles, rarely exceeding a syllable of sound, have such powerful and over-arching effects across the inter-modular grammatical space? This is the Platonic background against which this book is set. Why This Book? Well, because there is not one out there in its stead, or even close to dealing with the range, and nature, of things that this book deals with. One thing that is fresh about it is the level of detail for variation in old languages it tries to attain. I cannot wager too much money on the idea that it does succeed, but in case it does not, it sets a trail for future work. The second, and presumably bolder, claim it makes concerns the ways in which logical meanings change, dependent on or independent of the morphosyntactic module. It showcases evidence from a wide array of IndoEuropean languages, as well as the Japonic family. As it turns out, Old Japanese can shed light onto Indo-European. As can Avar, Hungarian, and Tibetan, also. In fact, the facts emerge more fully outside the language-particular box and the problem becomes captivatingly theoretical. Because of this, this book is theoretical, also. Who Is This Book For? Several groups of people. Historical Indo-European linguists will find it fresh for at least two reasons. One, it brings together as wide an empirical range of Indo-European languages as it can. Traditional philological facts and speculations are transplanted, translated, and interpreted using formal vii

viii

Preface

methods from modern generative linguistics. Two, it tries to distil from that data a theory and an analysis of the micro-meanings encoded by some special particles (and particles are parts of the language that tend to go unnoticed). This book is also for modern formal linguists interested, more generally, in how the interaction of syntactic structure and semantics works, in both synchronic and diachronic perspectives (since the formal approach to both of these dimensions is missing in the field, still). Generally, those linguists that are interested in the overall architecture of grammar that the theoreticians build. Lastly, non-linguists might find it insightful for two reasons: the book shows how linguists analyse data and what wild yet evidence-based models we construct. After all, time will tell how our models and ideas about the human language, and the human mind generally, might be verified materialistically using some high-tech machinery that look inside the brain, perhaps the mind itself, in the future (I keep hoping for a lambda-sensitive and tree-detecting EEG-like contraption, accompanied by an app). So we cannot be sure for sure, but we provide some very strong considerations for some intrinsically human mechanics of thought processes that underlie linguistic expressions. This book tries to do that, too. Bled, Slovenia 1 May 2018

Moreno Mitrovi´c

Acknowledgements

This book marks the reincarnation of my doctoral work and, therefore, inherits all the thanks I listed in the dissertation. There are too many thank so let me the following people who have helped through instruction or absorption in significant ways: Edith Aldrige, Theresa Biberauer, James Clackson, Gennaro Chierchia, Noam ˘ Laura Chomsky, Kai von Fintel, Danny Fox, Bjarke Frellesvig, Jovana GajiÄG, Grestenberger, Nina Haslinger, Xuhui Hu, Angelika Kratzer, Manfred Krifka, Adam Ledgeway, Andrew Nevins, Andreea Christina Nicolae, David Pesetsky, Maria Polinsky, Ian Roberts, Kerri Russell, Uli Sauerland, Viola Schmidt, Prods Oktor Skjaervo, Anna Szabolsci, George Tsoulas, Calvert Watkins, Jefferey Watumull, John Whitman, David Willis. My copyeditors Kaylie MacKenzie and Maite Seidel have been instrumental. I am also thankful to the two anonymous reviewers for their critical feedback, as well as the editorial team at Springer for their patient help with production. I would also like to gratefully acknowledge the AHRC, Cambridge University, and the Alexander von Humboldt Foundation for their financial support at different stages of research that led to this book. Finally, my family and friends for their support and love. Berlin, October 2020

ix

Contents

1

Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.1 Microsemantics and Decomposition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.2 Superparticles and Logical Syntax . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

1 2 5 12

2

Construction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.1 The Mystery of the Indo-European Double System of Coordination. 2.1.1 Alternating Positions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.1.2 Counting Morphemes (1M=2P, 2M=1P) . . . . . . . . . . . . . . . . . . . . . . 2.2 The Junction Function of Conjunction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.2.1 Deflattening Coordination . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.2.2 Enter Junction and the Extra Conjunction Position . . . . . . . . . . . 2.3 Back to Indo-European . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.3.1 A 110 Year Old Problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.3.2 JP and Superparticles in Old Indo-European . . . . . . . . . . . . . . . . . . 2.3.3 The Autonomy of the μ-Cycle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.3.4 Deriving the 1P/2P Alternation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.4 Beyond IE . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.4.1 Polyjunctions and the Case of More Than Two (Con)juncts . 2.4.2 CSC Violations Explained. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.4.3 A Parametric Typology of Junction Systems . . . . . . . . . . . . . . . . . . Appendix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

15 16 18 20 23 23 26 35 35 36 38 42 46 46 57 59 61 99

3

Interpretation. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.1 Logic and Grammar: A Starter Kit . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.1.1 Alternatives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.2 J . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.2.1 A Non-Boolean Bullet . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.2.2 Booleanising Junction Syntactically . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.3 μ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.3.1 The Polarity-Sensitive Profile . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

107 109 109 112 112 114 118 127 xi

xii

4

5

Contents

3.3.2 The Free-Choice Implicating Profile. . . . . . . . . . . . . . . . . . . . . . . . . . . 3.3.3 The Universal Quantificational Profile. . . . . . . . . . . . . . . . . . . . . . . . . 3.3.4 The Additive Profile . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.3.5 The Conjunctive Profile . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.4 κ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.4.1 The Existential Quantificational Profile . . . . . . . . . . . . . . . . . . . . . . . 3.4.2 The Interrogative Profile. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.4.3 The Disjunctive Profile . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.5 μ ◦ κ ◦ J and the Composition of Atypically Complex Disjunction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.5.1 Too Many Particles in So Many Languages . . . . . . . . . . . . . . . . . . . 3.5.2 Complex Disjunction as a Superparticle Complex Disjunction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Appendix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

132 136 139 148 153 155 160 162

Grammaticalisation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.1 Change in Construction: The Climb and Decline of μ . . . . . . . . . . . . . . . . 4.1.1 The Decline of 2P. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.1.2 The Derivational Change as Cause for Loss of 2P . . . . . . . . . . . . 4.2 Change in Interpretation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.2.1 Formal Semantics of Grammaticalisation . . . . . . . . . . . . . . . . . . . . . 4.2.2 Developing Superparticles in Japonic . . . . . . . . . . . . . . . . . . . . . . . . . 4.2.3 Losing Superparticles in Indo-European . . . . . . . . . . . . . . . . . . . . . . 4.3 The Evolution of μ: The Rise and Fall of Constraints on Interpretation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.3.1 The Cladistics of the IE μ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

189 189 190 191 197 198 203 230

Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.1 Taking Stock . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.1.1 Construction (Chap. 2). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.1.2 Interpretation (Chap. 3) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.1.3 Grammaticalisation and Diachronic Change (Chap. 4) . . . . . . . 5.1.4 Grammaticised Primitives of Logic . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.1.5 Learnability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.2 The Microsemantics of the Superparticle System . . . . . . . . . . . . . . . . . . . . . 5.2.1 The κ ◦ μ Concert and Expression of Atypical Disjunction . . 5.2.2 The κ and the μ Parts of the System: Commonalities and Divergences . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

255 255 256 257 257 258 259 260 262

164 165 170 178 184

240 245 248

263 265

Historical Texts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 267 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 273

Contents

xiii

General Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 277 Index of Authors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 281 Index of Particles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 283 Index of Languages . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 287

Acronyms, Symbols, and Abbreviations

1P 2P 2FUT A ADN ALL ALTR . ALL

AO ATTR CAUS CL CNCS CON COND CONJ COREL

CI CS CSC CR CSPRT

δ D DE DEB DV EPP EPP EV EVN EXH

First-position Second-position Secondary future Alternatives (alternative set) Adnominal Allative Alternating allative Disjunction to set conversion (following Alonso-Ovalle 2006) Attributive Causative Clitic Concessive Conjunctive gerund Conditional Conjunction marker Correlative marker Conceptual–intentional interface Coordinate structure Coordinate structure constraint Chain reduction algorithm Case particle Sub-domain (alternative) Determiner Downward entailing(ness) Debitative Defective verb Extended projection principle Extra peripheral position (a.k.a. extended projection principle) Evidential Even-based exhaustification operator Exhaustification operator xv

xvi EXT

FA FC FCI FOC HON

IE GC GD IND INDET INT / Q INTRJ

J κ KM KPRT

Λ LF μ MASD MP NML NOM

NPI PERLT

PF PI PLUP POL

PPI PRES 8 PRES 8 PRT

PRS PS PSI PSM

PR PST PV

PQ PWFA QR REL

Acronyms, Symbols, and Abbreviations

External (coordinand, conjunct) Functional application Free-choice (freedom of choice) Free-choice item Focus Honorific particle Indo-European Generalised conjunction Generalised disjunction Indefinite Indeterminate (pronoun) Interrogative Interjectional particle Junction ka-type superparticle Kakari musubi Kakari particle Label Logical form mo-type superparticle Masdar Mediopassive Nominaliser Nominative case Negative polarity item Perlative Phonological form Permutation (automorphism) invariance Pluperfect Polite Positive polarity item Present 8th class present stem Particle Proper strengthening Polarity sensitivity Polarity sensitive item Polarity sensitive morpheme Predicate Past Preverb(al element) Proto-question Point-wise functional application Quantifier raising Relative

References REPR RESTR RETR

σ SG - ACT SI Spec(XP) SUB SUBJC SUPP TENT

TOP VCE

xvii

Representative Restrictive particle Retrospective Scalar (alternative) Singular active Scalar implicature Specifier of some XP category Subordinate Subjunctive Suppletive base Tentative Topic Visibility condition for extension

References Alonso-Ovalle, L. 2006. Disjunction in alternative semantics. PhD thesis, UMass Amherst.

Languages Alb Ant Arm Av BSl CArm Cel Celtib CJ CLuw ClSkt Gaul Eng Gmc Gr Grk Gth Hit Hom IE IIr

Albanian (family and language) Anatolian (family) Armenian (family and language) Avestan Balto-Slavonic (family) Classical Armenian Celtic (family) Celtiberian Classical Japanese (Heian period) Cuneiform Luwian Classical Sanskrit Gaulish English (modern) Germanic (family) Classical Greek Greek (family) Gothic Hittite Homeric Indo-European Indo-Iranian (family)

xviii

Itl Lat MB MdIA MdJ MdW MIr MW Myc OB OE OC OCS OIr OJ ON OW PAlb PCel PGmc PIE PSl PToch Russ SBC Sl Slov TA TA Toch Ven

Acronyms, Symbols, and Abbreviations

Italic (family) Latin Middle Breton Modern Indo-Aryan Modern Japanese Modern Welsh Middle Irish Middle Welsh Mycenaean Old Breton Old English Old Cornish Old Church Slavonic Old Irish Old Japanese (Nara period) Old Norse Old Welsh Proto-Albanian (family and language) Proto-Celtic Proto-Germanic (family) Proto In-do-European Proto-Slavonic Proto-Tocharian Russian Ser-Bo-Croatian Slavonic Slovenian Tocharian A Tocharian B Tocharian (family) Venetic

Symbols: Semantics, Logic, and Other Technicalia 0

[i F] [uF] α [±] α [−] α [+] D ∗ 

Head, minimal category Interpretable feature Uninterpretable feature Optional phonological realisation of α No phonological realisation of α Phonological realisation of α Domain Ungrammatical form Reconstructed form

References

h2 ♥   f o n ⇒ ⇔ ≺ r l #   R ◦ \\ 

xix

The a-colouring laryngeal Innocent exclusion Result of diachronic development/change Source of diachronic development/change Focus semantic (alternative) value Ordinary semantic (alternative) value Example number (in citations, following the page number) Cross-categorical or generalised entailment Biconditional (logically), or vocabulary insertion correspondence (morpho-/post-syntactically), for which also may be used Covering relation Rank function Likelihood function Cardinality (count function) Probability measure ‘Less likely than’, given a  -ordering Recursive function Function composition Line break in inscriptions Missing fragment of the text

Orthographic Notations ABC

ABC

abc abc abc

Transliterated sumerogram in Hittite Transliterated akkadogram in Hittite Supplied grammatical words in Old Japanese Phonographic text in Old Japanese Logographic text in Old Japanese

Chapter 1

Introduction

Abstract The chapter introduces the main goals this book sets out to achieve, the decompositional methodology and conception of microsemantics, recognising that compositional analysis cannot stop at word boundary. Languages mark meanings, which have identical inferences, using identical particles and these particles thus creep up in a wide array of expressions. Due to their multi-tasking capacity to express seemingly disparate meanings, I dub them Superparticles. These particles are perfect windows into the interlock of several grammatical modules. With a firm footing in the module where grammatical bones are built and assembled (narrow morphosyntax), superparticles acquire varied interpretation (in the conceptualintentional module; semantics) depending on the structure they feature in.

This book is all about the captivating ability that human language has to express intricately logical (mathematical) meanings using tiny (microsemantic) morphemes as utilities. Languages mark meanings, which have identical inferences, using identical particles and these particles thus creep up in a wide array of expressions. Due to their multi-tasking capacity to express seemingly disparate meanings, I dub them Superparticles. These particles are perfect windows into the interlock of several grammatical modules. With a firm footing in the module where grammatical bones are built and assembled (narrow morphosyntax), superparticles acquire varied interpretation (in the conceptual-intentional module; semantics) depending on the structure they feature in. What is more, interpretations triggered by these particles are inferential and belong to the realm of pragmatics. How can such tiny particles, rarely exceeding a syllable of sound, have such powerful and over-arching effects across the inter-modular grammatical space? This is the Platonic background against which this book is set (or at least wants to be set).

© Springer Nature Switzerland AG 2021 M. Mitrovi´c, Superparticles, Studies in Natural Language and Linguistic Theory 98, https://doi.org/10.1007/978-94-024-2050-0_1

1

2

1 Introduction

1.1 Microsemantics and Decomposition Conceptually and methodologically, decomposition is a way of attaining as high a level of detail as possible. This level of detail is relevant for our understanding of the basic building blocks that feature in (morphosyntactic) construction and (semantic) interpretation. The morphosyntactic enterprise has moved toward blurring the lines that are traditionally associated with the notion of ‘word’ and ‘word’ boundaries. Three research programmes are relevant in this respect. The first is the enterprise of Distributed Morphology (DM), as galvanised by Halle and Marantz (1993) and culminating in state-of-the-art work such as Arregi and Nevins (2012), who devote a monograph to analysing the internal structure of a couple of auxiliary verbs in Basque and, in doing so, answer some more theoretical questions about the nature of Spellout and the structure of grammar. The second is work by Kayne (2005), who has been independently postulating silent elements in syntax for decades, with great success. The third is the decompositional programme of Nanosyntax (Starke 2009), which treats syntactic terminals as corresponding to submorphemic elements. For our purposes, all latter approaches may be considered methodologically and conceptually on a par and, in tandem, constitute a body of strong motivations, both theoretical and empirical, against the atomicity of ‘wordhood’ in general. In this programmatic respect, this book is aligned with the methodological promise of such decompositional analyses. As an example of this, take Kayne (2005, Chap. 4), who considers the seemingly locative there in English to not in fact be intrinsically locative. Rather, the locative flavour of there is acquired in structural presence of an unpronounced nominal head PLACE, itself the locus of locativity. Such (sub-) morphemic dissection into conspicuous and soundless constituents of a seemingly simplex word is far more than fanciful generative gymnastics. Firstly, this not only assimilates the cases of locative and non-locative incarnations of there, but also naturalises the very diachronic relationship between the contemporarily locative and archaically non-locative semantics of there. Comparatively, this also allows Kayne (2005, Chap. 4) to illuminate and make sense of the morphological connection between the co-occurrence of there with overt prepositions across Germanic. The crucial conclusion here is that synchronic dissections can not only illuminate word-histories and diachronic microsemantics generally, but comparative facts and typological distribution also. By allowing for the preposition to be unpronounced in such cases, we can bring directly into the fold the evidently non-locative uses in French and Italian of the apparently and seemingly locative clitics y and ci. Kayne’s (2005) morphosyntactic exemplar of decomposition achieves two unexpected results. First, the decomposition of a lexical item itself sheds light onto the item’s history. Secondly, it allows one to capture cross-linguistic variation with greater success since the level of structural detail allows for discrete parametric comparison of a new kind. In this book I try to follow these methodological

1.1 Microsemantics and Decomposition

3

and conceptual guidelines by paying special attention to the consequences for interpretation of such micro-structures. So what are the ramifications for semantics? Szabolcsi (2010, 189) states the Microsemantic Principle which describes the microsemantic programme that this book tries to execute. (1)

The Microsemantic Principle Compositional analysis cannot stop at the word level. (Since there is no word-level boundary.)

Everything else being equal, interpretational composition should abide by the morphosyntactic structure that is provided to the conceptual intentional interface. To equip our theory with stipulations that would ‘skip’ nodes or ignore nodes would mean ridding that theory of the principle-based power of explanation it is designed to provide. Naturally, one can pull a cheap trick to salvage this and adopt a view according to which most nodes in a complex structure are vacuous. While I am not bold enough to assume that every morpheme in a structure must contribute to meaning, it seems an undesirable consequence to me to end up with a theory in which most morphemes are void of meaning or some kind of contribution to meaning. Function words sometimes show a fossilised internal structure, which etymologies elucidate and make sense of. Consider the following pieces of evidence, presented using a ‘diamond’ tree, which is doubly rooted. These anti-gravitational trees are simply notational devices. Take the English conjunction and, which, according to McMichael (2006) (but see Lühr 1979), has its semantic roots in a preposition meaning ‘back’ or ‘against’ (cf. Latin ante). As a reviewer kindly notes, this is not entirely undoubtful. The ulterior etymology of ‘and’, as the Oxford English Dictionary notes, as well as its precise semantic development are uncertain. Most scholars assume a development from an adverb or a preposition with originally locative sense to a conjunction, given the comparative evidence from other IndoEuropean languages. I entertain a naïve prepositional etymology of and- in order to sketch the conception and methodology of the approach I develop in this book, not to commit myself to any particular philological school of thought on English ‘and’ (on which I will actually have nothing to say). Consider, then, the word answer, now a noun or a verb, which historically decomposes into and-swere, ‘to swear or declare back’ (McMichael 2006, 48). (2)

V/NP answer and swere P0

N0

NP

4

1 Introduction

While nouns/verbs such as ‘answer’ in (2) have presumably inactive and no longer existing internal morphosyntactic structure it once used to have, there are function words in English with active syntactic and semantic insides. Higginbotham (1991) provided one of the first analyses of either/or, showing that whether is the wh-counterpart of either. What is more, the neither/nor form falls well within this series. (3)

Toy decompositions ‘either/neither/whether’ b. Neg/QP a. Adv/QP

either

Disj0 0/

Disj

DisjP

neither

c.

CP[+Q] whether

n- either

wh- either

Neg0 Disj0

Wh0 Disj0

NegP Disj

WhP Disj

DisjP

DisjP

Historically, all three modern English forms contain whether, which can be traced back to Old English (OE) hwæðer, which can itself (as a reviewer tells me) be further dissected into ‘who’ + the comparative suffix (< PIE  -tero). Hwæðer in OE further built other logical expressions, such as æg(e)hwæðer (or ægðer in contracted form), which is the predecessor of Modern English either. Consider, then, the following data in (4) from Old English (taken from Gast 2013, 87n23) and assume that the emboldened g(e) morphemes are semantically identical and non-homophonic realisations of (what I will call) the ‘μ’ category. (I return to arguments against homophony later.) (4)

was æg[e]ðer, ge heora cyning, ge heora biscop was both=PART.DL-μ-WH μ their king μ their bishop ‘He was both their king and their bishop.’ (Or 238, 14; c893)

se

DET

The OE ge above is, in turn, related to the modern German universal quantifier je, which constitutes an ‘active’ example, that brings us closer to the set of words we will be examining. Rather than functioning as an atomic word, the German quantifier jeder, jede, jedes ‘every.M/F/N’ shows rich internal structure, as Leu (2009) demonstrates. The morphemic ingredients of the quantifier are the independently known distributive particle je- (cf. je-weils ‘each time’), the definite article morpheme d- (featuring in the determiner expressions der, die, das), and the relevant φ-agreement morpheme (-r/-e/-s, ‘M/F/N’). Sketched in (5) is Leu’s (2009) derivation of the quantifier phrase jeder Junge ‘every boy’, which starts out its derivatinal life as a QP (jeP) but moves up the extended adjectival spine (xAP), resembling a small relative clause-like structure.

1.2 Superparticles and Logical Syntax

(5)

5

D2 P D2

xAP D02

jeder

nP NP

n

Junge

n0

xAP jeP

4

3

je

DP1 D01 d-

AgrA P AgrA

NP Junge

Agr0A -er

1

jeP je0

NP

je

Junge

2

The former examples of decompositional approaches served as a model for the kind of microsemantic analyses I pursue in this book. With this conceptual and methodological stance, let us now turn to the data we will be concerned with, namely the superparticles, and the theoretical context (namely, the relation between syntax and logic) in which they are best understood.

1.2 Superparticles and Logical Syntax How logical and meaning-sensititve is syntax? Some expressions, such as (6) are logical in nature and ‘demand’ they be accompanied by sanctioning meanings. In this case, the sanctioning meaning is necessarily a downward-entailing (DE) one, as Ladusaw (1979), building on Fauconnier (1975), discovered. (6)

∗ There are any cookies left.

The morpheme relevant to discussion is the Negative Polarity Item (NPI) any, which cannot yield any grammatical—or for that matter, logical—expression without additional logical markers or structures (such as negation or interrogation, for instance). Any can be said to belong to a special class class of logical words. Logical words can, to the same effect, be defined as “a set of rules or principles that concerns a limited list of special words,” such as any, and, every, etc. (Chierchia 2013, 445) Here is the twist. In many languages, this limited list of special words is a singleton set, containing a single morpheme—a superparticle (I will call this type of

6

1 Introduction

superparticle μ). Consider now the implication of there being a single grammaticallogical formative, μ, that has a special status in grammar. Firstly, it is worth investigating for the latter considerations alone. Secondly, how can clearly distinct meanings of any, every, and and be built from a single μ superparticle? The μ meanings I exhibit in fact contain any, every, and, even, and also. The burden is then to unify not only the morphosyntactic status of the microsemantic superparticle μ but also its multi-tasking semantic nature. To put it simply: How can μ mean so many things? How are also and every connected? Note that I assume (and later buttress the view) that these meanings are ‘simultaneously spontaneous’: that is, there is a single micro-meaning of μ. The various incarnations of meaning are rooted in, and derive from and build on, the common denominational meaning of μ. A reason for my alluding to μ as microsemantic is therefore indicative of the solution to this second problem. I will adopt, and minimally adapt, an interpretational apparatus that can provide us with the desired meaning, in a principled and highly constrained manner. (The apparatus is essentially that of Chierchia 2013.) This apparatus is also the one that bars (6) on both structure-sensitive and inference-verifying grounds. A structural account for the ungrammaticality of (6) can be predicated on the idea that any demands there be a DE operator (a marker of negation, or a question-forming formative) present. Algorithmically, a grammatical principle would search the derivational space for an agreeable operator and return ‘grammatical’ if found, or ‘error’ if not. While this utility is principled and simple (and still a predominant school of thought in the morphosyntactic analyses of NPIs), it does not, and cannot, account for why a logical word like any would require the presence of another word (operator) that ensures DE meaning of the entire expression. The limited list of special words this book investigates are, actually, not words at all, but morphemes that build logical words like any, all, and, demonstrating that there is a common microsemantic denominator to those expressions. Natural language incarnates logical constants such as conjunctive and disjunctive connectives or interrogative, additive and quantificational expressions (i.e., that special logical vocabulary) using a single set of two morphemes—superparticles. Previous research by Szabolcsi (2010, 2015), Kratzer and Shimoyama (2002) and Slade (2011), among many others, has established that languages like Japanese may use only two morphemes, mo and ka, to construct universal/existential as well as conjunctive/disjunctive expressions respectively. Throughout this book, I abbreviate the Japanese mo particle and mo-like particles cross-linguistically as μ and the Japanese ka and ka-like particles cross-linguistically as κ. In Japanese, mo also serves as an additive and ka as an interrogative element. Consider the following three pairs of examples in (7) and (8) featuring the two superparticles. The left column (7) shows the mo-series and the right column (8) shows the ka-series of meanings.

1.2 Superparticles and Logical Syntax

(7)

The μ-system (8) a. Universal quantifier i. Dare-mo hanashi-ta who-μ talk-PST ‘Every-/anyDE -one talked’ ii. Dono gakusei mo which student μ hanashi-ta talk-PST ‘Every-/anyDE -student talked’ b. Additive marker Mary-mo hanashi-ta Mary-μ talk-PST ‘Also Mary talked’ c. Conjunction marker Mary-mo Bill-mo hanashi-ta Mary-μ Bill-μ talk-PST ‘(Both) Mary and Bill talked’

7

The κ-system a. Existential quantifier i. Dare-ka hanashi-ta who-κ talk-PST ‘Some-one talked’ ii. Dono gakusei ka which student κ hanashi-ta talk-PST ‘Some student talked’ b. Interrogative marker Hanashi-ta ka talk-PST κ ‘Did you talk?’ c. Disjunction marker Mary-ka Bill-ka hanashi-ta Mary-κ Bill-κ talk-PST ‘(Either) Mary or Bill talked’

The compositional recipes for the meanings above, using the two superparticle ingredients, seem to be roughly as follows. When a superparticle like mo or ka in Japanese combines with two nominal arguments, like Bill and Mary, coordination obtains (i.e. an expression of conjunction and/or disjunction, respectively). When mo combines with just one determinate, or definite, argument, like ‘Mary’, an additive (antiexhaustive) expression comes about. When a proposition combines with ka, we end up with a polar question (i.e., a set of two propositions). A combination of a superparticle with an indeterminate or indefinite wh-expression, like dare ‘who’ above, delivers a quantificational expression, either with an existential flavour (‘someone’, dare-ka) or a universal flavour (‘everyone’, dare-mo). A combination of a wh-term with μ is, at least prima facie, ambivalent between a universal distributive and a (negative) polar indefinite expression, depending on the polarity of the context in which it features (whether it is DE or not).1I will show in Chap. 4 that the universal distributive meaning of wh-μ is diachronically primary in the history of Japonic. I will accordingly sketch a diachronic analysis of how polarity sensitivity arose from universal meaning.

1 Prosodic

cues to disambiguation have been proposed: see Szabolcsi (2010, 202), Nishigauchi (1990), Yatsushiro (2002), Shimoyama (2006, 2007), among others, for an account of the synchronic distribution of facts.

8

1 Introduction

Similarly, quantificational expressions like ‘some/every student(s)’ in Japanese are constructed by combining an indeterminate wh-phrase, like dono (‘which’), with an NP like ‘student(s)’. I have assumed that the μ and κ superparticles in the three pairs of meaning are, respectively, identical. This is a fundamental assumption and one that is contra many analyses on the topic. Let me discuss this in greater detail in the following paragraph. Against Homophony The pattern underlying the examples in (7) and (8) allows for two contrasting explanations. The first hypothesis is that the semantic contribution of the two kinds of superparticles is uniform in all four of their respective constructions. If this is true, then there is something deeply interesting lying in the morphosyntax and semantics of the two superparticles. On the other hand, the second hypothesis, is that the multifunctional meanings behind the different incarnations of the superparticles in (7) and (8) could simply result from homophony, as Hagstrom (1998) suggests, and Cable (2010) even more explicitly defended on the basis of his analysis of Tlingit. The fact that the Japanese superparticle mo—under this view just a particle—features in conjunction, universal quantification and focal-additives, is a superficial and accidental matter. The different roles that mo performs in (7) stem from the fact that different mo’s are at play. Under such a view, there is nothing ‘super’ or multi-tasking about the μ and κ particles: they are homophonous vanilla particles with different meanings. They just end up sounding the same across many constructions. I oppose this view and defend the first ‘superparticle’ viewpoint. One argument in favour of this view is typological: why would languages consistently manifest homophony of coordinate and quantificational markers? (I will introduce the typological argument below.) Another argument concerns the inconsistency of a pro-homophony analysis in a language like Japanese, as presented in Mitrovi´c and Sauerland (2014). Under a homophony story, there are, at least, two kinds of mo particles: a conjoining one and a quantificational one. In a similar vein, Shimoyama (2006) contends that the universal quantificational and additive mo in Japanese are distinct, given their configurational differences. The same reasoning may extend to ka, which is, under these assumptions, ambiguous between two homophonous particles: a disjunctive one and an existential one. This predicts that mo and ka should not be able to express coordination and quantification simultaneously. As the following two pairs of Japanese examples from Mitrovi´c and Sauerland (2014, 41, ex. 3–4) show, this is not the case.

1.2 Superparticles and Logical Syntax

(9)

a.

b.

(10)

a.

b.

9

dono gakusei mo dono sensei mo hanashita INDET student μ INDET teacher μ talked ‘Every student and every teacher talked.’ ∗ dono gakusei mo mo dono sensei mo mo hanashita INDET student μ μ INDET teacher MO MO talked ‘Every student and every teacher talked.’ dono gakusei ka dono sensei ka ga hanashita INDET student κ INDET teacher κ NOM talked ‘Some student or some teacher talked.’ ∗ dono gakusei ka ka dono sensei ka ka hanashita INDET student κ κ INDET teacher κ κ talked ‘Some student or some teacher talked.’

The data in (9) and (10) is clear evidence homophonous pairs of coordinate and quantificational μ and κ particles do not exist in Japanese. The homophony analysis that Hagstrom (1998) and Cable (2010) most notably defend predicts that coordination of quantificational expressions (9), (10) should, ceteris paribus, yield particle ‘reduplication’: one particle expressing quantification and another expressing coordination. For further arguments against homophony of these two (classes of) particles, see Szabolcsi (2015) and, especially, Slade (2011) for a detailed historical argument based on his diachronic analyses of Japanese, Sinhala and Malayalam particles. In sum, I submit Slade’s (2011) general and methodological argument against homophony: “Entia non sunt multiplicanda praeter necessitatem: let us not suppose the existence of homophonous particles unless we uncover compelling evidence for such multiplicity.” (Slade 2011, 8) Another independent argument which strengthens this view comes from typology. Japanese is by no strech alone in boasting the two superparticles μ and κ. The multi-tasking nature of the two types of superparticles is cross-linguistically well attested. Superparticles are in fact far more common cross-linguistically that it may seem from the European linguistic perspective: Gil (2011) reports that the majority of languages (66% out of the 76 he investigated) that were studied for the World atlas of language structures shows formal similarity between quantificational, focal and coordinate constructions. Why would the grammars of most languages exhibit accidental homophony? If the allegedly homophonous meanings are logically connected, then perhaps it is not homophony but allosemy: a shifting of a common meaning depending on context. (Obviously, understanding the nature of this common meaning is crucial.) Logical Allosemy Superparticles that are cross-linguistically prominent in expressions of various logical meanings and marked with superparticles are not homophonous but rather are best analysed as possessing intrinsic alternative-triggering meaning which may

10

1 Introduction

have differential realisations. Such differential meanings are better understood in the sense of allosemy, being conceptually and theoretically on a par with allophony or allomorphy (following Marantz 2012), depending on the structural context. The various ‘special meanings’ in (7) and (8) can be thought of as being allosemes of the two superparticles. In this regard, consider Marantz’s (2012) ideas on which I will build: (11)

a. It is the structure of the grammar itself that determines the domain of contextual allomorphy: derivation by phase. So the domain of contextual allosemy should also be the phase. b. The additional constraint on contextual allomorphy of phonological adjacency follows if contextual allomorphy is sensitive to a phonological notion of “combines with”—adjacent items combine with each other (directly) phonologically. If we apply this idea to the semantic domain, we predict that contextual allosemy should be restricted to semantic adjacency, i.e., to elements that combine (directly) semantically.

The first desideratum is therefore to unify not only the semantic but also the syntactic distribution of the contextual incarnations of the two kinds of superparticles. From a detailed syntactic structure, the semantic interpretation follows compositionally, in line with Marantz (2011). Coordination will be shown to involve more syntactic material than overt realisations suggest. Under my analysis, the silent structure provides enough room for such non-coordinate meanings as the pairs in (7) and (8) show. Constructions such as quantification, questions and additive focus form substructures of coordination. Evidence will be predominantly drawn from a morphologically rich collection of ancient and modern Indo-European (IE) languages, which—through their morphology—reveal otherwise silent syntactic material that we fail to spot in a language like Japanese. The silent syntax will show itself though a cross-linguistic examination, pivoting on the syntactically and semantically neutral concept of junction. I will take the latter to be structurally and interpretationally the foundational building block of conjunction and disjunction. By breaking down coordination into separate layers, we will capture the syntactic and semantic differences, laying in the amount of layered projections, as well as the core components of the kinds of meanings the two pairs of μ and κ particles dictate. The Historical Dimension All languages change: at the level of sound, word, sentential structure and, presumably necessarily, compositional interpretation. It therefore follows from this that superparticles change, too. The rate, direction, and general nature of change in the superparticle system thus provides excellent grounds for observing, and ‘in fact’ studying, diachronic behaviour of the interfaces between several grammatical modules. Can superparticles die? (Yes.) This book predominantly focuses on Indo-European, which had a superparticle system detectable in the early stages of nearly all Indo-

1.2 Superparticles and Logical Syntax

11

European languages. Unlike modern Japanese, with a fully harmonic superparticle semantics for both μ (7) and κ (8) particles, Indo-European had (and rarely still has) the μ-series only—why κ never fully developed across the branches is unclear (and will remain so in the conclusion).2 In time, the multi-tasking capacity of μ declined in intriguing patterns. Can superparticles be born? (Also yes.) While there is no clear evidence that old Indo-European developed the μ system from an inherited grammar, this is clearly detectable in Japonic. Old Japanese, as the oldest attested form of Japonic, had both μ and κ particles, which were not as ‘super’ as they are in modern Japanese. Instead, the two particles were generally focus markers which, in time, developed the harmonic superparticle system in the Classical period. This leads to another, and more general, question: when and how can superparticles be born? The conclusions I reach are consistent with the claim that the birth of superparticles may result (at least sometimes) from the reanalysis of other elements as a superparticle. Does this indicate that language learners are ‘looking for superparticles’? Furthermore, assuming they are universal micro-building blocks of logical meanings, are superparticles part of our human linguistic endowment? I will return to these questions with some conjectural answers in the conclusion, but many thanks to a reviewer who raises these pertinent points and questions. Insofar as this book is primarily concerned with providing, or describing, an explanatorily adequate theory of superparticles with respect to the wide array of cross-linguistic data, a true answer to the question of the nature of superparticles is meta-theoretical. Answers concerning conditions on genesis of superparticles must, or at least should, after all be tied to and derived from the nature of universality of language (UG) and its division of labour (in the sense of Hauser et al 2002) between what is truly and uniquely linguistic, what is epiphenomenal of the linguistic data, and the general extra-linguistic properties of efficiency (nature). Scope and Limitations of This Book The goal of this book is to arrive at a convincing morphosyntactic, semantic, and pragmatic theory of superparticles in both typological (synchronic) and diachronic perspectives, which is immodest and comes at a cost of depth. In order to provide as complete as possible a microsemantic blueprint of superparticles, I sometimes compromise depth for breadth (I hope this does not frustrate semanticists too much). In future work, the intricacies of how exactly some meanings are calculated and related to other facts and questions, which I revisit in the Conclusion, are left to be worked out in future research.

2 From

Slade (2011) it would seem that at least one IE language (Sinhala) did develop (and retain) κ-particles (more than one realisation of, in fact). In this context, as a reviewer points out, this does raise interesting questions about the role of Dravidian contact, though the pieces in Sinhala κ-marked expressions are all apparently Indo-Aryan. I refer the reader to Slade (2011) for an extensive discussion of the facts.

12

1 Introduction

Roadmap and the Structure of the Book The book comes in three core parts. The first concerns construction (Chap. 2) which examines the morphosyntactic perspective on the superparticle systems specifically and the structure of coordination generally. The second part concerns interpretation (Chap. 3) where the structural morphosyntactic ideas are mapped onto an interpretational and compositional system, deriving the semantics and pragmatics of the superparticle meanings. The last part on grammaticalisation (Chap. 4) takes both morphosyntax and semantics/pragmatics and provides for them a diachronic story which aims to explain the synchronic distribution and historical pathways of variation and change. The reader may want to consult first the conclusion (Chap. 5) and see the major claims of this book, before referring back to the particulars.

References Arregi, K., and A. Nevins. 2012. Morphotactics: Basque Auxiliaries and the Structure of Spellout. Studies in Natural Language and Linguistic Theory. New York: Springer. Cable, S. 2010. The Grammar of Q: Q-Particles, Wh-Movement and Pied-Piping. Oxford: Oxford University Press. Chierchia, G. 2013. Logic in Grammar: Polarity, Free Choice and Intervention. Oxford Studies in Semantics and Pragmatics, vol. 2. Oxford: Oxford University Press. Fauconnier, G. 1975. Polarity and the scale principle. Chicago Linguistic Society 11:188–199. Gast, V. 2013. From æghwæer to either: The distribution of a negative polarity item in historical perspective. In Beyond ‘Any’ and ‘Ever’: New Explorations in Negative Polarity Sensitivity, ed. E. Csipak, R. Eckardt, M. Liu, and M. Sailer, 79–102. Berlin: Mouton de Gruyter. Gil, D. 2011. Conjunctions and universal quantifiers, Chap. 56. In The World Atlas of Language Structures, ed. M.S. Dryer, M. Haspelmath. Munich: Max Planck Digital Library. Hagstrom, P. 1998. Decomposing questions. PhD thesis, MIT. Halle. M., and A. Marantz. 1993. Distributed morphology and the pieces of inflection. In The View from Building 20: Essays in Linguistics in Honor of Sylvain Bromberger, ed. K. Hale, and S.J. Keyser, 111–176. Cambridge, MA: MIT Press. Hauser, M.D., N. Chomsky, and W.T. Fitch. 2002. The faculty of language: What is it, who has it, and how did it evolve? Science 298(5598):1569–1579. Higginbotham, J. 1991. Either/or. In Proceedings of NELS, ed. T. Sherer, vol. 21, 143–157. Amherst: GLSA. Kayne, R. 2005. Movement and Silence. Oxford: Oxford University Press. Kratzer, A., J. Shimoyama. 2002. Indeterminate phrases: The view from Japanese. In The Proceedings of the Third Tokyo Conference on Psycholinguistics, ed. Y. Otsu, 1–25. Tokyo: Hituzi Syobo. Ladusaw, W.A. 1979. Polarity sensitivity as inherent scope relations. PhD thesis, University of Texas, Austin. Leu, T. 2009. The internal syntax of jeder ‘every’. In Linguistic Variation Yearbook, ed. J.V. Craenenbroeck, vol. 9, 153–204. Amsterdam: John Benjamins. Lühr, R. 1979. Das Wort ‘und’ im Westgermanischen. Münchener Studien zur Sprachwissenschaft 38:117–54. Marantz, A. 2011. Locality domains for contextual allosemy, paper presented at the Columbia Lingusitic Society. . 2012. Locality domains for contextual allomorphy across the interfaces. In: Distributed Morphology Today, ed. O. Matushansky, and A. Marantz, 95–115. Cambridge, MA: MIT Press.

References

13

McMichael, A. 2006. The A’s and BE’s of English prepositions. In Syntax and Semantics of Prepositions, ed. P. Saint-Dizier, 42–56. Dordrecht: Springer. Mitrovi´c, M., and U. Sauerland. 2014. Decomposing coordination. In Proceedings of NELS 44, ed. J. Iyer, and L. Kusmer, vol. 2, 39–52. Nishigauchi, T. 1990. Quantification in the Theory of Grammar. Dordrecht: Kluwer. Shimoyama, J. 2006. Indeterminate phrase quantification in Japanese. Natural Language Semantics 14:139–173 Shimoyama, J. 2007. Indeterminate noun phrase quantification in Japanese. Natural Language Semantics 14(2):139–173. Slade, B.M. 2011. Formal and philological inquiries into the nature of interrogatives, indefinites, disjunction, and focus in Sinhala and other languages. PhD thesis, University of Illinois at Urbana-Champaign. Starke, M. 2009. Nanosyntax: A short primer to a new approach to language. Nordlyd 36(1):1–6. Szabolcsi, A. 2010. Quantification. Cambridge: Cambridge University Press. . 2015. What do quantifier particles do? Linguistics and Philosophy 38:159–204. Yatsushiro, K. 2002. The distribution of mo and ka and its implications. In Proceedings of FAJL 3 MIT Working Papers in Linguistics, vol. 41, 181–198.

Chapter 2

Construction

Abstract This chapter investigates the Indo-European (IE) grammar conjunction and an archaeological dig for superparticles in IE. This serves three purposes. Primarily, the conjunction particles will provide a window for a semantic and diachronic analysis in the following chapters. After introducing the mystery, we take a theoretical digression in the second section to motivate some syntactic refinements for the underlying structure of coordination. The third section then returns to IndoEuropean and, against the background of novel motivations, provides an explanation for both the synchronic facts we well as the directions and trends in diachronic development, resting on the notion of loss of the μ-cycle. The final section expands the investigation by looking into the theoretical consequences of the developed analysis for the derivation of polysyndetic coordination: the analysis hinges on leftrecursive structure for junction, which also explains apparent Coordinate Structure Constraint (CSC) violations and provides a window into explaining the typology of junction systems typologically.

This chapter investigates the Indo-European (IE) grammar conjunction and an archaeological dig for superparticles in IE. This serves three purposes. Primarily, the conjunction particles will provide a window for a semantic and diachronic analysis in the following chapters. After introducing the mystery, we take a theoretical digression in the second section to motivate some syntactic refinements for the underlying structure of coordination. The third section then returns to Indo-European and, against the background of novel motivations, provides an explanation for both the synchronic facts we well as the directions and trends in diachronic development. Those readers interested in the theoretical aspects of coordination, and the syntactic structures in which superparticles feature, can skip straight to Sect. 2.2.

© Springer Nature Switzerland AG 2021 M. Mitrovi´c, Superparticles, Studies in Natural Language and Linguistic Theory 98, https://doi.org/10.1007/978-94-024-2050-0_2

15

16

2 Construction

2.1 The Mystery of the Indo-European Double System of Coordination The ‘mystery’ is the following: old Indo-European languages had two configurational means of expressing conjunction. The first one in which the conjunction marker is bimorphemic and placed in the first position (1P) with respect to the second conjunct. The other configuration involved a monomorphemic conjunction particle that was (with respect to the second conjunct) placed in the second position (2P). Why would there exist these two strategies for expressing the same conjunction meaning?1 The mystery of the 1/2P placement of the conjunction marker has important conceptual ramifications for our understanding of the morphosyntactic structures that superparticles feature in. There are two general desiderata for this chapter, namely to show that (P)IE had a rich superparticle system and that the (P)IE grammar of coordination can be characterised with the following two generalisations. (12)

i. There existed two types of conjunction particles and consequently, two types of coordination constructions (1/2P). ii. There existed two types of interpretation for one of the two types of coordinate particles.

The general aim, then, is to tease apart the two taxonomies and have one of the generalisations (the second one in 12ii) automatically fall from the system which I develop (in Sect. 2.2). Consider the minimal pair from Vedic in (13), which shows the two configurational patterns we find across old IE languages. The pivot of variation is the configuration of the conjunction marker and the second conjunct (β in ‘α and β’). The modern and predominant way to express ‘α’ and ‘β’ is to place the conjunction marker between the two conjuncts—and Vedic, too, had a similar configuration, as (13a) shows. In (13b), however, the second conjunct, ‘β’, and the conjunction marker flip. Note, however, that the 2P in (13b) is also the last-position (since, informally, the relevant DP conjunct is a word long). If the second conjunct involved more overt structure, it would be the left-most word within that conjunct that would ‘flip’. (Before I turn to formalising the configurational difference, I will keep using ‘flipping’ as an informal designation of the difference between 1/2P.) (13)

Vedic: a. ma´¯ no maha´¯ ntam utá ma´¯ no arbhakám not us great and not us small ‘[Harm] not either the great and [or] the small of us.’

1 For

(R.V, 6.1.11ab )

details not covered in this chapter, see Mitrovi´c (2018b) and work cited there.

2.1 The Mystery of the Indo-European Double System of Coordination

17

b. va´¯ yav-¯ındra´si -ca ti cetathah. suta¯´ n¯am . v¯ajin¯ıvas¯u Vayu-Indra-and rush.2.DL rich strength-bestowing ‘Vayu and Indra, rich in spoil, rush (hither).’ (R.V 1.002.5a ) The core fact to observe at this point is that the second conjunct and the conjunction morpheme -ca flip: the ‘flipping’ (which I will restate and make more precise later) never occurs between the first conjunct and the conjunction marker. This, itself, is already a window into the underlying structure of coordination.2 The optionality of expressing conjunction using the 1P or 2P configuration, extends well beyond Vedic. We could distinguish the two types of configurations by positing that the 2P conjunction marker triggers some form of movement from right to left, as indicated in (13b). This movement can be assumed to take place either syntactically or postsyntactically. However derived, the fact would remain that the syntactic difference between (13a) and (13b) lies only in the linearisation of the surface placement of the coordinator: the 1P does not, while the 2P marker does trigger flipping, i.e. the change in the linear order. However, at leat two other facts are relevant (otherwise, there would be close to no point to be made in reference to the overall semantic aims of the book). Let me briefly address these two in turn. Aside from the configurational differences between the two ways of expressing conjunction, one of the two configurations uses a conjunction marker which itself contains two morphemes—the 1P. And not just any two morphemes: in (13a), utá comprises the conjunction particle u and another conjunction particle ta. I take this latter particle to be on a par with the palatalised counterpart ca, which features in the 2P configuration (13b). These 2P constructions uniformly use a single (monomorphemic) particle. The question, then, is how and why a particle like ca/ta requires (or does not require) another particle, like u to express conjunction marking? This brings us to the third, and most revealing, difference between 1P and 2P expressions of conjunction. The monomorphemic 2P conjunction markers are μtype superparticles, while the bimorphemic 1P conjunction markers are not. Diachronically, one of the two competing structures (with 2P and 1P placement) dies out across all languages: the 1P type wins. While I address this in detail, and provide a story for how and why this occurs so regularly, consider the intuitive attraction of an economy principle. The 1P conjunction structure requires two conjuncts, α and β, and a conjunction marker placed between them. The 2P configuration requires ‘flipping’ (I will contend this is a reflex of narrow-syntactic movement): the conjunction morpheme needs to see the insides of β and flip with (or move) an element within β. If the interior of β is not visible or unflippable (not extractable), something else needs to rescue the derivation. On the other hand,

2 If

the structure were ‘flat’, that is ternary, we may predict that flipping, and the second position effect that the conjunction markers of this type require, be found in first conjunct configurations, also. This is not what we find and I show how these facts, and a couple of others, are effectively captured.

18

2 Construction

the 1P configuration does not suffer this search-and-rescue predicament. While the metaphor can only go so far, this is the general logic of the economy-based theory for the synchronic account of 1/2P alternation I will pursue in Sect. 2.3. This analysis will be the direct basis for the analysis of historical change and decline of 2P configurations, which I lay out in Chap. 4. In the remainder of this section, I will go through and buttress the first two facts about the contrast between 1P and 2P type of conjunction. After introducing some theory in Sect. 2.2, I will be able to make sense of the third fact.

2.1.1 Alternating Positions The alternation in the placement of the conjunction maker is not entirely random and in free variation. Two additional facts hold. Firstly, 1P markers do not, while the 2P do, tend to repeat for each conjunct. English seems to have a similar feature that distinguishes the bisyndetic (repeating conjunction markers) both/and and the monosyndetic structure with a single ‘and’. Secondly—and this is related to the first fact—the 1P configuration does express conjunctions of clauses (propositions), while the 2P configuration tends not to. We do not have to look beyond English to see that the bisyndetic conjunction requires, or at least strongly prefers, the conjuncts to be category- or type-wise non-clausal or non-propositional.     (14) a. I know both DP,e John and DP,e Mary .     b. I know (∗/?? both) CP,t John knits and CP,t Mary drives fast cars . Klein (1985a,b) has shown for Vedic (cf. the data we started with) that the alternation between 1P and 2P placements of the conjunction correlates with the category of the conjuncts. While the 2P coordinators generally do not coordinate clauses, 1P coordinators can. Given in Table 2.1 is a clausal/non-clausal conjunct distribution with respect to occurrences with 1/2P conjunction markers utá/ca in the R.gveda. The 2P series of conjunction in IE is generally and freely prone to reduplication. As Gonda (1954) and Dunkel (1982) note, a 2P connective like the PIE  k w e is traditionally reconstructed with a twofold syntax: both single (or monosyndetic) (X Y  k w e) and double (or bisyndetic) structures (X  k w e Y  k w e) are confidently reconstructed. The following three pairs from three representative IE languages (Sanskrit, Greek, Latin) show this. Table 2.1 Categorial distribution of non/clausal conjuncts for R.gvedic pen/initial coordinators (numbers from Klein 1985a,b) Coordinator utá (1P) ca (2P)

Distribution 47.64% (N = 705) 52.56% (N = 775)

[+CP] conjuncts 51.66% (N = 364) 7.61% (N = 59)

[−CP] conjuncts 48.34% (N = 341) 92.39% (N = 716)

2.1 The Mystery of the Indo-European Double System of Coordination

19

(15)

Vedic and Classical Sanskrit: a. dharme ca arthe ca k¯ame ca dharma/law.LOC and commerce.LOC and pleasure.LOC and ca bharata r.s.abha yad iha asti tad anyatra moks.e liberation.LOC and Bharata giant which here is.3.SG that elsewhere na tat kvacit yad na iha asti which not here is.3.SG not that anywhere ‘Giant among Bharatas whatever is here on Law, and on commerce, and on pleasure, and on liberation is found elsewhere, but what is not here is nowhere else.’ (Mbh. 1.56.34) b. va´¯ yav ¯ındra´s ca cetathah. suta´¯ na¯Am ˛ . va¯Ajin¯ ˛ ıvas¯u Vayu Indra and rush.2.DL rich strength-bestowing ‘Vayu and Indra, rich in spoil, rush (hither).’ (R.V 1.002.5a )

(16)

Homeric Greek: a. os ede tá te essomena tá te eonta which were (=know.PLUP) the and exist.PART the and exist.FUT pró te eonta before and exist.PART ‘That were, and that were to be, and that had been before.’ (Il. A. 70) b. aspidas eukuklous lais¯aia te pteroenta shields round pelt and feathered ‘The round shields and fluttering targets.’ (Il. M. 426)

(17)

Classical Latin: a. iam tum tendit que fovet que already then pursue and favour and ‘Already then, she both pursued it and (also) favoured it.’ (Aen. 1.18) b. v¯ıam samtem que life safety and ‘the life and safety’ (Or. 1.VI.28-9)

The polysyndetic pattern of enclitic coordinators in (15a), (16a) and (17a) seems to have carried an emphatically distributive component, akin to the modern English bisyndetic distributive conjunction with both. . . and. We find the same reduplicative pattern with emphatic semantics in Old Church Slavonic (OCS), which survives in contemporary Ser-Bo-Croatian, among other contemporary Slavonic languages. It is OCS, and its diachronic descendants, that shows the independence of linear placement and semantic force behind the repetition of the conjunction. Proto-Slavonic has independently syncretised the 1P and 2P coordinators but only lexically (I return to this with evidence). As the following OCS example in (18) shows, conjunction i has both the conjunctive semantics of the 1P coordinators in IE as well as the emphatic/focal semantics of the 2P conjunction. While the

20

2 Construction

dual semantics—to be adequately addressed and derived in the following chapter— is retained in Slavonic, the morpho-lexical difference between the two classes of conjunction has been collapsed. We will return to the syntax of this collapse below. In (18), the first pair (a) shows (reduplicative) polysyndetic coordination with emphatically distributive meaning, while the second pair (b) is an example of a monosyndetic construction. (18)

Old Church Slavonic: a. boite že se˛ paˇce mogo˛ štaago i dšo˛ i tˇelo pogubiti fear but REFL rather be.able and soul and body destroy ‘But rather fear that which is able to destroy both soul and body.’ (CM. Mt. 10:28) b. bo˛ dvete že mo˛dri eˇ ko zmije˛ i cˇeli eˇ ko golo˛ b˘ıe be but wise as serpents and harmless as doves ‘Rather be wise as serpents, and harmless as doves.’ (CM. Mt. 10:16)

Note that the focal additive meaning related to polysyndeticity has been retained in some of the contemporary varieties of Slavonic. The following are parallel examples from the relevant Matthew passage in Ser-Bo-Croatian: (19)

Contemporary Ser-Bo-Croatian: a. Bojte se više onoga koji može i dušu i tijelo pogubiti fear REFL more that which may and soul and body destroy ‘But rather fear that which is able to destroy both soul and body.’ (Mt. 10:28) b. budite dakle mudri kao zmije i bezazleni kao golubovi be therefore wise as serpents and harmless as doves ‘Rather be wise as serpents, and harmless as doves.’ (Mt. 10:16)

So far, I have demonstrated that IE indeed freely allowed reduplication of the conjunction marker. I have also explored the possible semantic side-effect of such reduplication yielding a focus-sensitive and distributive effect identical to the English both/and construction, which is paraphrasable as not only . . . but also. I now turn to another feature of the two types of conjunction systems in IE, namely the morphemic structure of the conjunction markers featuring in the two configurational types.

2.1.2 Counting Morphemes (1M=2P, 2M=1P) There is one additional, and for our purposes crucial, fact distinguishing the 1P and the 2P types of conjunction. The difference also lies in the morphological structure of the two series. Here is the generalisation, in relatively informal terms:

2.1 The Mystery of the Indo-European Double System of Coordination Table 2.2 Dunkel’s (1982) reconstruction of two coordinator series in IE

(20)

Orthotone  kw ó /  kw í  h éw 2  yó  tó

21 Enclitic  − kw e −h u 2  -yo  -te

Markers with one morpheme (1M) are in 2P, while conjunctions with two morphemes (2M) are in 1P.

While 2P coordinators are monomorphemic, the 1P coordinators are not. Initially placed coordinators are bimorphemic and as such are decomposable synchronically (or through reconstruction) into two coordinators, each underlying a morpheme. Greek kai, for instance, derives from  kati, itself being a concatenation of  k w e +  te (Beekes 2010, 614; Boisacq 1916, 390). Conversely, as Dunkel (1982, 2014a,b) etymologises it, Indo-Iranian (IIr.) uta comprises the coordinator u + ta (  h2 (é)u +  te); Gothic coordinators jah and jau result from  yo +  k w e and  yo +  h2 u respectively. Dunkel (1982) reconstructs two enclitic series of four coordinators for PIE. One series is orthotone (i.e, freestanding and 1P) and another enclitic (i.e. 2P) as shown in Table 2.2. The 1P coordinators in IE are generally decomposable into and reconstructable only as a pair of orthotone and enclitic coordinators. Dunkel’s orthotone connectives, however, are not found in independent (not composed) word-level compositions in any of the attested daughter languages, which begs the question of redundancy of the orthotone series (insofar as the conjunction particles go). In its stead I assume a single, inherently enclitic, series, out of which bimorphemic coordinators are derived through concatenation. This reasoning yields the empirical facts in Table 2.3 in a more economical way. Dunkel’s latest and extensive work (Dunkel 2014a,b) contains much more data on composed particles, into which we do not delve further but which are, in fact, in line with the methodological particle decomposition I am attempting here. Therefore, for instance, the OW coordinating particles cen and cet are treated as having a dyadic etymology in line with Dunkel (2014b, 422ff.) who treats the two particles as sharing a common additive particle which can be traced back to PCel.  ko- (Dunkel 2014b, 423; fn. 5 and ref. therein). For details on the etymological decomposition of OW ce-n, see Dunkel (2014b, 410ff.; fn. 55) and for details on the decomposition of OW ce-t, see Dunkel (2014b, 425; fn. 16; ) (cf. also Falileyev 2000, 27 for comparative evidence and further references). The historical decomposition of Slovenian in(u) is supported by Trubaˇcev (1980, 168), Vasmer (1953, 483), Feu (1961), and references therein. There is also Russ. no ‘otherwise’ and Russ. in ‘then’. We will be concerned with the latter form which has the etymology and decomposition given in Table 2.3 (according to Vasmer), but the former is distinct (cognate with the numeral ‘one’ in other languages). Languageparticular particles and their concatenations are explicated in the Appendix.

22

2 Construction

Table 2.3 Clitic combinatorics as strategy for development of freestanding conjunction markers Dependent  kw e

 te

Gr. kai OW cet

 kw e

 te

OIr. to-ch Hit. takku

h u 2

Skt. u ca Lat. atque

h u 2

–

–

IIr. u-ta Gr. aute Lat. au-t

 yo

Gth. ja-h

–

 nu

OIr. nach

OIr. nade

 yo

–

–

–

Gth. j-au

–

–

 nu

Independent [+ε] [−ε]

OW cen

IIr. ca Lat. que Gr. te OIr. ch Gth. uh Gaul. cue Ven. ke Celtib. ku

–

–

Gr. de Alb. dhe Skt. tu

Sl. to

Slov. i-n(u)

IIr. u Gr. au CLuw. a

Sl. i

–

Hit. ya TA. yo Myc. jo

–

–

Hit. nu OIr. no Slav. no

I now take a departure from the IE data in order to discuss the theory behind the coordination structures with an aim to arrive at an analysis which would be

2.2 The Junction Function of Conjunction

23

amenable to particle composites discussed above. I then return to IE and apply the motivated structure in ways which derive the facts.

2.2 The Junction Function of Conjunction The IE data has now been stretched to the point where some theoretical assumptions need to be laid out. These generally concern the structure coordinate markers project. The aim of this subsection is to arrive at a rich enough morphosyntactic structure for coordination generally, and conjunction specifically, which can serve to account for the IE facts laid out in the previous section. I will first start with a brief review of arguments against a flat treatment of coordination in syntax (and I’ve very briefly alluded to such a desideratum already). These arguments motivate a constituency for conjunction phrases that is compliant with the general phrasestructure rules. After this is in place, I ‘upgrade’ the structure without losing any of the power that the previous account provides, yet gaining structural room to capture some facts that stretch beyond coordination.

2.2.1 Deflattening Coordination In perhaps the most recent and simultaneously the most extensive treatment of coordinate syntax, Zhang (2010) has shown that the syntax of coordination involves no special configuration, category, constraint or operation. Based on her work, we will proceed to an assumption that the syntactic structure of coordinate construction is binary, as most notably argued for by Kayne (1994, ch. 6), Zhang (2010) and, with minimal variations, Munn (1993). Earliest arguments for a binary-branching model of coordinate syntax go back to Blümel (1914) with subsequent substantiation from Bloomfield (1933), Bach (1964), Chomsky (1965), Dik (1968), Dougherty (1969), Gazdar et al. (1985), Goodall (1987) and Muadz (1991), and many others in the last two decades. Following Kayne (1994), we take coordinators to be heads, merging an internal argument (coordinand) as its complement, and adjoining an external argument (coordinand) in its specifier, as per (21). &P

(21) XP

coordinand1

&0

YP

coordinator

coordinand2

24

2 Construction

This syntactic model—let me call it the traditional binary model (TBM)— captures, among other things, the (universal) generalisation that there can be no coordinations of heads (22) since the coordinator head &0 requires complements of maximal category (see Kayne 1994, Sect. 6.2 for elaboration, among others). As a brief illustration of this fact, see the Slovenian example in (22), which confirms Kayne’s (1994) ban on clitic coordination (under the assumption that clitics are not maximal categories). 

(22)

Janez me in te gleda J me.ACC.CL and you.ACC.CL watch.3SG.PRES ‘John is looking at me and you.’

Munn (1993) and Zhang (2010) both presented arguments for the TBM at length, which I cannot afford to review in detail. Munn’s analysis has a formal twist that, in regards to the general departure from the TBM, has to do with the derivation of the external coordinand, which is base-generated or raised3 to Spec(&P), as I am assuming here (following Kayne 1994 and Zhang 2010, int. al.). Munn (1993) was among the very first to explicitly treat and discuss the phrase structure of coordination. In his Chapter 2, which we now briefly overview and from which we import the main conclusions, he removes the stipulations about phrase structure of coordinated phrases that refer specifically to coordinate structures. Apart from Gazdar et al. (1985), most studies in syntax of coordination have generally and implicitly assumed a flat syntactic structure, along the lines of (23), taken from Munn (1993, 11n2.1).The coordinator was taken to syncategorematically adjoin to XP(s). (23)

XP

XP

XP

and

XP

The structure in (23) is problematic for several reasons. Firstly, the structure is hardly a phrase since it is not headed and, as such, it violates X-bar theoretic endocentricity. If all lexical items project phrasally, and if a functional item like and is a lexical item, which seems obvious, then not all lexical items project a phrase. This either jeopardises the projection principle or else confines coordinate structures to a position of exceptions and stipulations. Alternatively, if the structure is not flat, 3 Chomsky

(2013), for instance, presents arguments from the view of labeling where a coordinate structure of the type [Z and W] results from the raising of one of the coordinands. Chomsky assumes that the possible base-formed constituency of a coordinate construction, headed by a labellacking terminal (&0 in our discussion), is [α &0 [β Z W ]], capturing the semantic symmetry of coordination. In order to label β (with competing Z and W labels in base-form) and α (unlabellable since &0 has no label, qua inherent categorial feature), one of Z, W must raise, say Z, yielding [γ Z [α &0 [β Z W ]]]. See Chomsky (2013, 46) for details of such a labelling-driven analysis of coordination.

2.2 The Junction Function of Conjunction

25

then the X-bar theoretic template would extend to coordination, which would reestablish theoretical consistency and conceptual advantage of a theory that treats it so. Munn (1993) arguest against ‘flatness’ of coordination, proposing a hierarchical and X-bar conformant structure. Taken from Munn (1993, 13n2.3) and given in (24), are two syntactic structures summarised in Munn (1993): (24) represents a SpecHead structure which Munn first defended, which is in line with the phrase structural assumption we made in (21), based on Kayne’s (1994) and Zhang’s (2010) work. The structure in (24) presents the structure Munn (1993) eventually argues for. (24)

a. Spec-Head BP:

b. Adjoined BP: NP

BP NP B0

BP

NP

B NP

B0

NP

The choice between the two structures is slight if not trivial, from a modern syntactic theoretical perspective in light of the label-projecting theories of Cecchetto and Donati (2010), Chomsky (2013), and Adger (2013), among many others. It is for this reason that we do not revise (21), which will shortly be upgraded, retaining the Spec-Head or Adjunction structure. One of the strongest arguments that Munn (1993, 16–37) puts forth in favour of hierarchical arrangement of coordinate structure, among other pieces of evidence,4 comes from binding asymmetries between the first (external) and the second (internal) coordinand. As shown in (25), the first (external) conjunct can bind into the second (internal) conjunct (25a), but not the other way round (25a). (25)

a. b.

Every mani and hisi dog went to mow a meadow. ∗ Hisi dog and every mani went to mow a meadow.

Under an analysis assuming a flat structure (24), these asymmetries are not predicted, while both the Spec-Head and the Adjunction structures (24) can account for (25) by virtue of the asymmetric c-command relation holding between the two conjuncts. The binary structure with a projecting coordination head accounts for monosyndetic conjunctions, but what about the bisyndetic ones I overviewed in the previous section? Consider a distributive bisyndetic data from Ser-Bo-Croatian: (26)

4 For

i Mujo i Haso and M and H ‘both Mujo and Haso’

further arguments and a general theoretical history of treatment of coordination construction, see Progovac (2014).

26

2 Construction

Either one of the conjunction markers projects, say the second one, or else the other conjunction marker one adjoins to the first conjunct. At this point, the principled generality of structure projection is lost, along with other complications (e.g., under the view that the first i conjunction marker adjoins to, or within, the first conjunct, how come it is a single-argument-taking head, while the second is not?). What is more, consider a minimally modified and fully well-formed version of (26) given in (27). (27)

i Mujo a i Haso μM J=butSBC /J=andCzech μ H ‘both Mujo and Haso (Czech) / Not only Mujo but also Haso (Ser-BoCroatian)’

Where does the adversative conjunction a sit? Perhaps this head is the true conjunction, while the other two are sub-conjunctional heads contained in the conjunct projections. This is the line of thought we build on with an aim to derive some other empirical facts from such a structure. Note also that the same-sounding adversative particle a in Ser-Bo-Croatian (27) is a plain vanilla conjunction in Czech, which also allows the presence of two additive i markers along with the J-level conjunction marker a. The following section is therefore devoted to exploring the idea that a TBM structure in (21) is not quite rich enough, since at least one additional slot for the coordinator is required.

2.2.2 Enter Junction and the Extra Conjunction Position There is a non-negligible amount of data which the TMB structure with a single coordination position cannot account for. Let me lay out the theoretical foundation and then demonstrate how some novel empirical evidence is borne out in light of such assumptions. Assuming a binary branching structure for coordination (21), den Dikken (2006) argues that exponents such as and and or do not in fact occupy the coordinatorhead position as indicated in (21) but are rather phrasal subsets of the coordinator projection, with their origins in the internal coordinand. The actual coordinator head, independent of conjunction and/or disjunction which originate within the internal coordinand, is a junction head, J0 —a common structural denominator for conjunction and disjunction.

2.2 The Junction Function of Conjunction

(28)

27

JP

XP J0

coordinand1

and P

coordinator

and 0

YP

(silent)

and

coordinand2

The core motivation for den Dikken’s postulation of the silent presence of J0 is to capture the distribution of the floating either in English. As Myler (2012) succinctly summarises: (29)

den Dikken’s either is a phrasal category and can be adjoined to any XP as long as: a. XP is on the projection line of the element focused in the first disjunct; and b. XP is not of C category; and c. no CP node intervenes between either and the focused element in the first disjunct; and d. either surfaces to the left of the aforementioned focused element at PF.

This characterisation of either predicts its floatation, i.e. its optional height of adjunction, which is either too high (30) (his ex. 1) or too low (31) (his ex. 2). (30)

a. John ate either rice or beans. b. John either ate rice or beans. c. Either John ate rice or beans.

(31)

a. Either John ate rice or he ate beans. b. John either ate rice or he ate beans.

The floating either may thus move within or, apparently, beyond the first coordinand projection (XP). Note that the subphrasal components to J 0 , namely both-, (wh/n)either-, and- and or-headed phrases that J0 glues together are predicted to be structurally independent by virtue of their phrasal status. This theory covers and explains not only the either. . . or coordinate constructions but also the whether. . . or and both. . . and, which are unified under the structural umbrella of JP structure. Here is where I diverge. Firstly, den Dikken (2006, 58) takes the head introducing the internal (second) coordinand not as the lexicalisation of the J0 but as a phrasal category establishing a feature-checking relationship with the abstract J0 instead. There is no principled reason according to which J0 would resist or be banned from lexicalisation. For den Dikken, J0 is an abstract ‘junction’ category inherently

28

2 Construction

neutral between conjunction and disjunction for which no overt evidence is provided since his account rests on J0 not being lexicalised. I take it as a reasonable hypothesis that there may be languages which overtly realise this junctional component of coordination.5 Secondly, den Dikken’s (2006) subjunctional analysis of English and entails that and-projections are monadic in a sense, insofar as they formally combine with a single argument (the internal coordinand). This itself raises some conceptual issues with regard to the general logical frame of conjunction. The fact that the constituent [andP and XP ] is disallowed in English follows from different lexical specifications on and. (Note that Munn 1993 also suffers the same conceptual drawbacks in regard to his BP analysis). As I develop in the next section, variation in lexical specification need not be invoked, since single-argumental and-XP-like expressions are, in fact, possible. (Such sequences will be very common in the sections to come.) I propose that the true structure for conjunction is the one in (32), which retains all the strengths of den Dikken’s (2006) original proposal while also eliminating the stipulations pertaining to the anti-lexical status of J0 and the monoargumental and-like phrases. These are superparticle-headed projections (μP or κP categories): if independent, or not complementing J0 , μPs (or inversely κPs) express non-conjunctive meanings (additivity, universal quantification, etc., as detailed in Chap. 3). Once embedded within the JP, (at least two) μPs yield conjunction and κPs yield disjunction (how this is achieved is also taken up in Chap. 3). The enriched conjunction structure, featuring the superparticle μ category, is therefore the following. (32)

A structure for rich conjunction: JP

J

μP μ0

XP

J0

μP μ0

YP

In Mitrovi´c (2011a), I proposed a double-headed structure for coordination involving a lexical and a ‘light’ coordinator head, which captures both the polysyndetic data and the mono-argumental μ-expressions. However, the true theoretical antecedent to the fine-grained structure for conjunction, employing den Dikken’s (2006) Junction, can be ultimately traced back to Slade (2011) and was subsequently

5 Proving

this does not, in any way, invalidate den Dikken’s (2006) account. Quite the opposite: it makes a strong case for a JP structure using novel empirical motivation. Where I depart is the conception of necessary abstraction of J, which does not lexicalise in English (while I do not subscribe to this view, I am unable to expand on English conjunction further here).

2.2 The Junction Function of Conjunction

29

followed by Szabolcsi (2013, 2015), Mitrovi´c (2014) and Mitrovi´c and Sauerland (2014, 2016), among others subsequently. Additional empirical motivation for the Junction phrase, along with evidence of the ‘subjunctional’ structure of the conjuncts, comes from several languages. In what follows, I review data from three genetically unrelated language families, showing that conjunction of two arguments may be expressed using three conjunction morphemes (a J and two conjunct-corresponding μ heads)—I may refer to this phenomenon as ‘triadic’ realisation or exponency. This novel data is submitted as critical evidence for the existence of an enriched conjunction structure.

2.2.2.1

Slavonic

Macedonian boasts a rich set of overt coordinate positions. Aside from the standard (English-like) type (33a) and a polysyndetic (both/and-like) type (33b) of conjunctive structure, Macedonian also allows for three conjunction markers to be realised (33d) in the presence of only two conjuncts (the data were provided by Roska Stojmenova and Ivan Stojmenov, pers. comm.). The non-medial i markers are analysed as μ particles. (33)

a. [Roska] i [Ivan] R and I “Roska and Ivan.” b. [i Roska] [i Ivan] and R J and I “both Roska and Ivan.” c. [Roska] i [i Ivan] R and I “Roska and also Ivan.” d. [i Roska] i [i Ivan] and R and I “Not only Roska, but also Ivan.”

(34)

JP J

μP μ0

DP

J0

i

Roska

i

μP μ0

DP

i

Ivan

Snejana Iovtcheva (pers. comm.) also informs me that in Bulgarian, or at least in the Haskovo Region dialect of Bulgarian, three heads of the kind in (33d) may

30

2 Construction

also be realised when conjoining two elements. The question given in (35) can be answered with a rich set of homophonic conjunctive and additive markers as (36) shows (the data was provided by Snejana Iovtcheva, pers. comm.). (35)

ama i Roska li beshe na partito? but also/μ R. Q/κ was at party-the ‘But was R also at the party?’

(36)

i Roska i i Petar i i Dimitar i i also/μ R and/J also/μ P and/J also/μ D and/J also/μ Moreno, vsichki bjaxa na parti-to! M all were at party-the ‘Not only Roska but also Petar and (also) Dmitar and (also) Moreno were all at the party!’

Adversative-flavoured triadic conjunction is also a possibility in Ser-BoCroatian (38), which may optionally co-occur within the repetitive additive construction (37), just like in Hungarian below. Let me repeat the relevant data in order to highlight the cross-linguistic pattern that is emerging. (37)

a. i Mujo i Haso μM (J) μ H ‘both Mujo and Haso’ b. JP μP

(38)

μ0

DP

i

Mujo

J0 [± ]

μP μ0

DP

i

Haso

a. i Mujo a i Haso μM J μH ‘both Mujo and Haso / Not only Mujo but also Haso’ b. JP μP μ0

DP

J0 [± ]

i

Mujo

a

μP μ0

DP

i

Haso

2.2 The Junction Function of Conjunction

31

The Macedonian/Bulgarian facts of triple realisation of i in bisyndetic coordination finds its parallel in Ser-Bo-Croatian also, with the obligatory presence of the cˇ ak, ‘even’—I return to the more precise structure of the exhaustifier underlying the scalar additives in Sect. 3.3.4.2. For now, consider the evidence in (39) with a corresponding analysis (where cˇ ak is taken as an adjunct, preliminarily). (39)

a. Na tulum dodjoš (ˇcak) i Mujo i cˇ ak i Haso. ˇ ˇ To party came CAK μM J CAK μH ‘Even Mujo and even Haso turned up to the party.’ b. JP μP cˇ ak

2.2.2.2

J0 [± ]

μP μ0

DP

i

Mujo

μP cˇ ak

μP μ0

DP

i

Haso

Hungarian

Beyond Slavonic, and in fact beyond Indo-European, we also find triadic realisation of conjunction in Hungarian, which allows the polysyndetic type of conjunction with reduplicative conjunctive markers: (40)

a. Kati is Mari is K μ (J) M μ ‘Both Kate and Mary’ b. JP μP DP

μ0

Kati

is

J0 [− ]

μP DP

μ0

Mari

is

On top of (40), Hungarian allows the optional realisation of the medial connective és co-occurring with polysyndetic additive particles is (Szabolcsi 2014, 17, fn. 21):

32

(41)

2 Construction

a. Kati is és Mari is K μJ M μ ‘Both Kate and Mary’ b. JP μP

2.2.2.3

DP

μ0

J0 [± ]

Kati

is

e´ s

μP DP

μ0

Mari

is

Northeastern Caucasian

Avar, a Northeastern Caucasian language of Dagestan, also provides such evidence for an articulation of a rich conjunction structure. Avar boasts three structural possibilities for conjunction. It first allows coordinate constructions of the polysyndetic (Latin que/que, Japanese mo/mo) type (95), which, according to our JP system, involves two overt μ heads and a silent J 0 [−] . Take the following data reported in a reference grammar (Alekseev and Ataev 2007). (42)

Ravzam gi Umukusum gi R μ (J) U μ ‘Ravzat and Umukusum’

(Alekseev and Ataev 2007, 105)

The following examples, provided by Ramazanov (pers. comm.) and Mukhtarova (pers. comm.), are in line with the enclitic character of the gi marker as described in the reference grammar (Alekseev and Ataev 2007), which we take to be overt instantiations of the μ head. The following example and the accompanying analysis in (95) show this. (43)

a. keto gi ève gi cat μ (J) dog μ ‘cat and dog’ b. JP μP NP

μ0

keto

gi

J0 [− ]

μP NP

μ0

ve

gi

2.2 The Junction Function of Conjunction

33

Taking gi to be of μ category, we predict it to feature independently given the prediction of the subphrasal status of the complement to the J0 . This in fact obtains and the gi-phrase (a μP) independently functions as an additive (the semantics for this is given in Chap. 3). In (44a), the μ marker gi follows the verb and intervenes between the verb (g’yelb, ‘know;) and its object (l’ala, ‘this’). Its presence triggers anti-exhaustive focus on the verb alone. Syntactically, I take the object (assumed to be an NP) to move and left-adjoin to the VP, as indicated by i-indexed movement in (44a). The remnant VP is then taken to left-adjoin to the μP (j -indexed movement), which yields the narrow focus reading of ‘know’. As a focus-sensitive element, the even¯ flavoured μ0 can also be recast as vP-level A-hosting position in the relevant left periphery. In fact, the left-peripheral status of μ heads is a matter I will implicitly return to. (44)

a. Dida [g’yeb gi] l’ala I know μ this ‘I even know this’ (i.e., not just believe) b. TP T

NP Dida

T0

μP μ

VP j g’yelb

μ0 gi

VP VP j

NPi l’ala

V0

NPi

Similarly, the NP/DP-level μ in (45) can be analysed as originating as a Focus head in the nominal left periphery. The nominal category (whether DP or NP) then ¯ A-moves to Spec(μP) which derives the interpretative effect. Alternatively, the μ category can be taken as a separate category, independent of the nominal extended projection, which takes the nominal as a complement. The analysis I am building does not commit to any of the two details since the general properties of the syntaxsemantics, in both synchronic and diachronic terms, have no direct and obvious bearing on this issue. (45)

a. [Dida gi] g’yeb l’ala I μ know this ‘Even I know this’ (as opposed to you knowing this, for instance)

34

2 Construction

b.

TP T

μP NP

μ0

Dida

gi

T0

V V

NP

g’yelb

l’ala

Aside from the polysyndetic type (95), Avar also allows an English-like construction with a conjunction marker placed between the two coordinands, which we take to be a phonological instantiation of J 0 : (46)

a. keto va ève cat (μ) J dog (μ) ‘cat and dog’ b. JP μP J0

μ 0 [− ]

NP

μP μ 0 [− ]

NP

va

keto

ve

The third and last type of construction that Avar allows is typologically rare and, for our purposes, most intriguing as it provides empirical support to the enriched JP analysis I have been motivating. The last type shows that all three of the presumed functional heads may be realised: both μ heads (each heading a conjunct) as well as the J head may be realised simultaneously. (47)

a. keto gi va ève gi cat μ J dog μ ‘cat and dog’ b. JP μP NP

μ0

J0

keto

gi

va

μP NP

μ0

ve

gi

2.3 Back to Indo-European

35

Among the Dagestanian languages, Dargi also boasts bisyndetic conjunction, as van der Berg (2004) reports. Similarly to Avar, Dargi allows realisation of both the J morpheme as well as the μ morpheme in the internal conjunct, as (48) confirms. (48)

k’älijä.li-za-w ca abdal le-w-ni wa eger di-la if [me-GEN castle-ILL-M one stupid present-M-MASD(ABS)] and w-iij-ni nu-ni ka b iz=aq-asli il-ra èu [this-and you(ABS) M-be-MASD(ABS)] me-ERG pose:N=CAUS-COND.1 ‘If I prove there is one fool in my castle and that is you, . . . ’ (van der Berg 2004, 201, ex. 14)

To sum up this section, I take the syntactic refinement of the coordination structure, proposed in (32), to be well motivated from the empirical perspective. Another empirical perspective, with a strong diachronic flavour, is to follow in the next section. I have drawn the triadic conjunction data from North-Eastern Caucasian (Avar and Dargi), Hungarian and Slavonic and thus provided a typologically varied case for establishing a clear morphosyntactic pattern which poses serious challenges for alternative syntactic theories of coordination. The evidence is also strikingly uniform with respect to the predictions that the JP system makes in terms of what heads (morphemes) may be realised. There is currently no alternative syntactic model of coordination which may explain or derive (without further stipulation) the conjunction of two nominals involving three conjunction morphemes. There is simply no syntactic room in the TBM for the second or the third exponent. The finegrained system I put forth here, however, can not only handle such data without any problem, it even predicts their possibility. The system I have set up is now ready to serve as an explicans for the IE mystery, which I revisit in the next section. In Sect. 2.4, I push the JP analysis further as I investigate conjunctions of more than two arguments in a wider typological perspective. Those interested in the theoretical and typological aspects, may move on to Sect. 2.4 from here.

2.3 Back to Indo-European How does the enriched coordination structure, involving a superparticle layer and a juncitonal superstructure, help us understand the problem of the configurational variation in IE conjunction I discussed in Sect. 2.1?

2.3.1 A 110 Year Old Problem Indo-European (IE) scholarship has recognised the semantic and syntactic multifunctionality of certain particles that express meanings which may be more precisely

36

2 Construction

logically characterised as conjunction, quantification, negative polarity, additivity, and free-choice items. The phenomenon in question was informally recognised 110 years ago in the seminal philological work by Meillet (1908), who identified non-conjunctive meanings of prima facie conjunctive particles. Gonda (1954) was among the few philologists who resumed the discussion and formulated the problem more precisely: The question may, to begin with, be posed whether we are right in translating S[anskrit] ca, Gr[eek] te, Lat[in] que, etc., simply by our modern ‘and’ in regarding the prehistoric *kwe as a conjunction in the traditional sense of the term (Gonda 1954, 182).

With this intuition, Gonda (1954, ibid., emphasis mine) observed that “the relation between the copulative [coordinate] te (teA ) and the ‘epic’ [non-coordinate] te (teB ) has never been correctly formulated.” I believe this cannot be properly understood without the correct formulation and theoretical understanding of the syntactic-semantic relation between coordinate and non-coordinate meanings. “It is a matter of general knowledge that many words which at a later period acted as conjunctions originally, or at the same time, had other functions” (Gonda 1954, 182). The methodological nature of traditional IE linguistics has been unable to unify definitively the sense of these logical morphemes. It is therefore understandable that, to date, research on this topic has stood still with respect to solving this problem and formulating a semantic common denominator that would explain the range of meanings these particles express. Empirically, the field has been advanced by Dunkel (2014a,b), who composed a Lexicon containing a rigorous index and etymological commentary of these particles (spanning well over 1000 pages). This allows me to approach the 110-year old issue in a theoretically novel and explanatory manner and to examine the unity and variation of IE particles. This can be achieved if these IE particles are analysed as superparticles, like those in Japanese.

2.3.2 JP and Superparticles in Old Indo-European The mystery we started with can now be understood against the background of the enriched Junction structure, motivated in the previous section. The enriched structure identifies two locally configured conjunction head positions: the μ-marked ‘shell’ of the internal conjunct and the suprastructal Junction projection, headed by a neutral Junction. The old Indo-European conjunction system, with its signature alternation of 1P/2P placement, can be understood in these novel structural terms. Appealing to the two conjunction heads, I develop in this section the analysis which identifies each of the two morphemes, featuring in 1P conjunctions, with the two heads, J0 and μ0 . Such an analysis solves Meillet’s (1908) 110-year old problem since the multifunctional nature of IE conjunction markers is explained elegantly and succinctly by positing that the morphemes occupy, or rather realise, different positions within

2.3 Back to Indo-European

37

the enriched JP, each position carrying a different meaning. Once structurally in concert, the conjunction structure (32) yields conjunction. Additionally, and independently to these semantic effects, the nature of the alternating positions (1P vs. 2P) is solved since the 2P morpheme, corresponding ¯ to μ0 checks its 2P requirement (presumably an A-feature) either by movement (‘move a minimal category’) or selection (‘pronounce J0 ’). We are now in a position to distinguish the three canonical word order types in IE coordination. In monosyndetic coordinations with enclitic particles, the external (first) coordinand (μP) is silent. In coordinations headed by a linearly initial bimorphemic coordinator, the two coordinate morphemes are distributed between J0 and the head of its complement, μ0 , as per Table 2.3. The notation [± ] in Table 2.3 refers to whether a particle is a Wackernagel element, requiring second-position ([+ ]), or not ([− ]). The theory and details behind the notations will be addressed. This idea is summarised in (49) with the three types of coordinate construction; Classical Latin (at)que is taken as an example ([−] is a notation for phonological silence). (49)

a. 1P conjunction constructions i. 2P (15a, 16a, 17a, 18a):  polysyndetic conjunction   [μP μ coord1 ] J0 [μP μ coord2 ] que

que

[−]

ii. 2P monosyndetic conjunction (15b, 16b, 17b, 18b) with phonologically silent μ0EXT :    [μP μ coord1 ] J0 [μP μ coord2 ] [−]

[−]

que

b. 1P (bimorphemic) conjunction constructions with phonologically silent μ0EXT :    [μP μ coord1 ] J0 [μP μ coord2 ] [−]

at

que

The analysis of compound coordinators sketched in (49b), where the morphological components of initial particles like Latin at-que or Sanskrit u-tá are spread between μ0 and J0 , also lends itself to a diachronic analysis of the development of linear placement of coordinators in synchronic IE, which is uniformly head-initial. The analysis put forth here also makes an empirical prediction for IE. By assigning the lower μ-headed coordination structure a category status, aren’t we led to believe that the μP may have a life of its own, without the JP superstructure?

38

2 Construction

2.3.3 The Autonomy of the μ-Cycle According to the model we have been proposing, the syntax of coordination is broken down into categories of two kinds. While the higher J0 is taken to join coordinate arguments, its complement μP is thus, mutatis mutandis, predicted to be an independent phrasal category. By virtue of being junctional, J0 establishes a twoplace relation between coordinands (a formal default of coordination meanings). μP, on the other hand, does not establish a two-place coordinate relation, which leads us to the possibility that there are mono-argumental and morphosyntactically coordination-like constructions headed by μ in IE. Given the generalisation on monomorphemic enclitic coordinators, now treated as μ0 s, we need to find in IE mono-argumental constructions headed by monomorphemic μ particles like Latin que, Sanskrit ca or OCS i. This is in fact what we find in all IE branches. Independent μPs are of four types: universal quantifier constructions (‘Every/each one is happy’), polarity constructions (‘I didn’t see anyone’), freechoice constructions (‘You may have any/whichever one’) and (non/scalar) additive constructions (‘Also/even he is happy’). In the first three, μPs contain a μ0 and a wh-element. The following examples show a consistent spread of μPs, marked with brackets, across the full range of early IE languages. Proceeding from east to west, we start with Indo-Iranian. Both R.gvedic and postVedic Sanskrit show the non-coordinate use of the coordinating particle ca, where it forms a free-choice expression of the wh-ever-type (50a, 50b), or a negative polarity item (50c). When not combined with a wh-host, the particle forms an additive expression with additive semantics, akin to the function of also/even in English, as shown in (50d). (50)

Vedic and Classical Sanskrit: a. prát¯ıdám . vi´svam modate yát [k¯ım . ca] pr.thivya´¯ m ádhi this world exults which [what μ] earth.F.ACC upon ‘This whole world exults whatever is upon the earth.’ (R.V 5.83.9a ) b. yady- abhyupetam [kva ca] s¯adhu as¯adhu v¯aA˛ . if promised to be accepted where μ honest dishonest or kr.tam may¯a . done.PST.PART 1.SG.INSTR ‘If you accept whatever I may do, whether honest or dishonest.’ (BP. 8.9.12) c. na yasya [ka´s ca] tititarti m¯ay¯a? NEG whom. GEN [who. M . SG μ] able to overcome illusions. PL ‘No one [=not anyone] can overcome that (=the Supreme Personality of Godhead’s) illusory energy.’ (BP. 8.5.30) ´ d. [cintayam s ca] na pa´ s y¯ a mi bhavat¯ a m prati vaikr . . . tam thinking.PRES.PART μ NEG see.1.SG you unto offence.ACC

2.3 Back to Indo-European

39

‘Even after much thinking, I fail to see the injury I did unto you.’ (Mbh. 2.20.1) In Latin, too, the combination of a μ particle and a wh-term may yield a freechoice item like ‘whatever’ in (51a). Alternatively, the combination may yield a universal quantificational expression like ‘all’ or ‘each’, as examples which Bortolussi (2013) collected in (51b–51d) show. (51)

Latin (Bortolussi 2013): a. ut, in quo [quis que] artificio excelleret, is in suo genere Roscius that in who [what μ] craft excels, is in his family R diceretur spoken ‘so that he, in whatever craft he excels, is spoken of as a Roscius in his field of endeavour.’ (Or. 1.28.130; Bortolussi 2013 ) b. Sic singillatim nostrum unus quis-que mouetur so individually we one wh-μ moved ‘So each of us is individually moved’ (Lucil. sat. 563; Bortolussi 2013) c. Morbus est habitus cuius-que corporis contra naturam sickness is reside wh-μ body contrary nature ‘The sickness is the situation of any/every/each body contrary to nature’ (Gell. 4.2.3; Bortolussi 2013) d. auent audire quid quis-que senserit want hear what wh-μ think ‘they wish to hear what each man’s (everyone’s) opinion was’ (Cic. Phil. 14,19; Bortolussi 2013)

Note the same free-choice meaning in Gothic, where the combination of a whterm like ‘where’ and a μ particle uh, diachronically deriving from  k w e, yields ‘wherever’ as (52a) suggests. Just as in Latin, and other IE languages, the wh+μ combination may also form a universal quantificational expression as per (52b). (52)

Gothic: a. [þishvad uh] (. . .) gaggis. [where μ] go.2.SG.PRES.ACT.IND ‘wherever you go’ (CA. Mt. 8:19) b. jah [hvazuh] saei hausei waurda meina and who.M.SG and pro.M.SG hear.3.SG.IND words.ACC.PL mine ‘And every one that heareth these sayings of mine’ (CA. Mt. 7:26)

In Old Church Slavonic, we focus on one kind of μ particles: the conjunctive i particle. In non-coordinate uses, i was additive-focal (cf. Sanskrit ex. 50d), and the crucial morphemic ingredient for NPIs. In (53a), we show the additive role of i.

40

(53)

2 Construction

Old Church Slavonic: k nim [i togo] a. posla sent.3.PL.AOR [μ him.M.SG.ACC] to them.PL.DAT ‘He sent also him to them.’ (CM. Mk. 12:6) b. ne mogl bi tvoriti [n-i-¯aeso-e] NEG be-able. PP would.3 SG do [NEG-μ-what-REL] ‘. . . he would not be able to do anything.’ (CM. Jn. 9:33; Willis 2000, 328, ex. 15)

We also find the additive use of the coordinator pe in Tocharian. In the Appendix, I detail the Tocharian superparticle ra, which was not only an additive/conjunction marker, but also formed universal quantifiers. (54)

Tocharian: a. [ñemintuyo ypic olyiyam s¯arth . [jewels.PL.INST full ship.F.SG.LOC caravan.M.SG.OBL Jambudvipac pe] Jambudvipa.M.SG.ALLT and/μ] s.pät kom y¯amuräs., . s¯a having been made.SUPP.ABS.M.SG.ABL seven day.M.PL.PERLT kñukac wram . neck.SG.ALLT water.SG.LOC ‘With a caravan to Jambudvipa also having been made in a ship filled with jewels [. . . ]’ (TA, Pun.yavanta-J¯ataka, 5a )

While Classical Armenian did not possess the enclitic  k w e-type coordinator, we can ascertain its loss in the pre-Classical period, since the remnant  k w e still shows in fossilised non-coordinate form, with the semantics aligned with other IE languages. (55)

Classical Armenian: a. et‘e [o- k‘] . . . if who- μ ‘If anyone [strike (thee) upon thy right cheek . . . ]’ (VT. Mat. 5.39; Klein 1997, 196) b. [erbek‘] . . . [time.LOC μ] ‘At any time/ever.’ (VT. Mt., 5.39; Klein 1997, 191)

Hittite, along with the rest of the Anatolian family, also shows mono-argumental functions of the coordinator, of which there were two kinds: kki/kku and (y)a. In noncoordinate uses in combination with wh-hosts, the former creates negative polarity terms (56a), while the latter creates universal quantificational expressions (151a), a common feature in IE. Note that only the (y)a particle qualifies as μ since it was

2.3 Back to Indo-European

41

able to feature in a range of constructions (particle kki/kku was not). I address the quantificational force behind the ‘original’ IE μ superparticle in Chap. 4. (56)

Hittite: ÚL [kuit ki] sakti a. nu-wa and-QUOT NEG [who PTC] know.2.SG.PRES ‘You know nothing (=not anything)’ (KUB XXIV.8.I.36) b. nu DUMU.MEŠ-ŠU [kuišš-a] kuwatta utn¯a J sons.his who-μ =everything somewhere country.LOC paizzi went ‘Each of his sons went somewhere to a country.’ (KBo. 3.I.1.17–18) c. nu [kuitt-a] arhayan kinaizz[i J what-μ =everything seperately sifts ‘She sifts everything seperately.’ (KUB XXIV.11.III.18)

Old Irish ch, itself a reflex of PIE  k w e, aside from the coordinate function, also creates free-choice (57a) and universal quantificational (57b, 57c) expressions. (57)

Old Irish: a. [ce ch] taibre [what μ] give.2.SUBJ ‘what[so]ever thou may give.’ (Zu ir. Hss. 1.20.15) b. [ce ch] orr [what μ] slay.3.M.SUBJ ‘whichever he may slay.’ (Anecd. ii.63.14.h) c. á

huili duini .i. a [ca-ch] duini VOC all man i.e. VOC wh-μ=every man ‘O, all men i.e. O, every/each man’

(Wb. 10c20)

The morphosyntactic independence of μP across a wide range of IE languages is strong evidence for the J0 -μ0 coordination complex defended here and elsewhere (cf. Slade 2011, Winter 1998, Szabolcsi 2014, inter alia). There is additional semantic evidence for the proposed structure, which gives rise to two different interpretations. In the absence of J0 , μPs are predicted to have independent ‘quantificational’ contribution (additivity, universal quantification, polarity, free choice). By the same token, we predict, for instance, that the Slovenian coordinator in, being derived from a compounding of Proto-Slavonic  i and adverbial-like

42

2 Construction

connective  n˘u,6 is not of μ but of a category which features the concatenation of J and μ, which explains its inability to form a polarity/free-choice item with a whelement (58), unlike Ser-Bo-Croatian (59), which has retained the Proto-Slavonic monomorphemic  i (Derksen 2008, 207), taken here to be of μ category. (58)

(59)

∗ in kdo J who ‘anyone/whoever’ (Slovenian) i (t)ko μ who ‘anyone/whoever’(Ser-Bo-Croatian)

The contrast in the pair above provides a window into how superparticle status may be diagnosed for a given logical marker: if the same marker which is used in expression of conjunction with two arguments may be combined with a single whterm to yield a quantifier, then that marker is μ superparticle. Having identified two syntactic positions in a coordination structure, I will assign them a static meaning via lexical entries and rules for composition in the next chapter.

2.3.4 Deriving the 1P/2P Alternation I have empirically established that there were two canonical constructions available in IE languages: a 1P and a 2P one, the latter with two subtypes (49). Theoretically, given the three properties of the double system (linearisation, additive focus and morphemic structure), all three properties that differentiate the two canonical patterns were derived within the JP system. This section, based on Mitrovi´c (2018b), addresses the syntactic derivation behind the 2P placement of the coordinator. We first investigate the synchronic constructions in IE that feature 2P μ particles, for which a diachronic account is given in Chap. 4, according to which the initial pattern is the surviving one. The second position effect has its traditional aetiology in what is known as Wackernagel’s Law. Wackernagel (1892) is credited with dubbing the one generalisation that applies to the syntax of PIE, namely that some elements consistently occupy the second position in a given string of words, which constitutes solely a descriptive observation pertaining to word count. An explanation is feasible in a theory of syntax which, for instance, attributes all configurational (word order related) differences to differences in movement. There have been two theoretically different approaches to the explanatory account of Wackernagel’s Law. Although both theories identify the source of the second position effect in movement, one 6 For

details on the bimorphemic etymology of Slov. in, see Trubaˇcev (1980, 168), Feu (1961), and references therein.

2.3 Back to Indo-European

43

confines this movement to narrow syntax while another places the movement in the post-syntactic module, relegating the movement triggers to prosody. (See Roberts (2010) and Mitrovi´c and Sideltsev (2017b) for an overview.) The purpose of this section is not to categorically suggest a confinement space wherein the Wackernagel-movement takes place, but rather to entertain the idea of an over-arching factor that accounts for the distribution of the second position effects (at least of the type that the IE coordination data suggests). I will contend that this factor is the phasal architecture, to which not only the syntactic derivation is subject but also the phonological and prosodic processes that follow it. A Wackernagel element like our μ (Lat. -que, Hom. -te, Gth. -uh, Skt. -ca, etc.) has a requirement which demands μ be preceded by a head.7 Let us assume that μ particles come hardwired with an [EPP]-like feature [ε] which, unlike [EPP], attracts and induces movement of the closest and the smallest syntactic object, a terminal/head. Just like [EPP], [ε] must be checked in line with the principle of economy (“as soon as possible”). If there is a syntactically available object satisfying the two ‘movement criteria’—i.e., the syntactic object is (a) the closest (b) X0 —then [ε] is checked syntactically. If there is no eligible local terminal in the syntactic structure, [ε] is checked post-syntactically, as per principles of economy (“better late than never”). The visibility and eligibility of such head targets is determined, as I will suggest, by phasal boundaries. Phases, as domain delimiters for structure building, do not only concern syntactic processes. It is a standard minimalist assumption to view phasal heads as ‘closing off’ a cycle, which is—upon merger of the phasal head, X0π —transferred to the two interfaces for semantic and phonological processing (interpretation and externalisation respectively). A phase, therefore, not only partitions narrow syntactic derivation into logical building blocks but also delimits post-syntactic operations and synchronises them with narrow syntax. In this direction, Samuels (2009, 242) takes as a starting point the conceptual argument laid out in the foundational work by Marvin (2003, 74): “If we think of levels in the lexicon as levels of syntactic attachment of affixes, we can actually say that Lexical Phonology suggests that phonological rules are limited by syntactic domains, possibly phases.” Samuels thus proposes a Phonological Derivation by Phase, which “relies on a cycle that is not proprietary to phonology” (Samuels 2009, 243). Combining Samuels’s theory with the notion of post-syntactic movement, we predict the domain or scope of such operations based on the narrow syntactic derivation. Assume μ in (60) is a Wackernagel-type coordinator specified with [ε], which represents the requirement for 2P placement. Let us further assume that μ takes a phasal complement Xπ P, itself having ZP as its specifier and YP as its complement, as the following configuration suggests.

7 The

clitic host is necessarily (of the size of) a head; we do not come across entire complex categories preceding enclitics.

44

2 Construction

μP

(60)

μ0 [ε ]

Xπ P

ZP

Z0

X

X0π

YP

... a. ε -checkable terminals narrow syntactically: 0/ b. ε -checkable terminals post-syntactically: { Z0 , . . . , X0π } c. closest accessible terminal: Z0 Since the phasal head X0π triggers the transfer of its complement, only the edge of Xπ P is accessible to outside operations. The head of ZP is ineligible for narrow syntactic head movement for reasons to do with anti-locality conditions. Postsyntactically, movement takes place, checking [ε]. Should the ε-accessible domain of heads be non-empty, we predict narrow syntactic incorporation to take place, in line with the aforementioned economy. Nominal coordination of the type in (61) thus gets linearised narrow syntactically since the set of ε-accessible terminals would not be empty, unlike in (60). (61)

ájanayan mánave ks.a¯´ m apái ca ti for.men created.MID.3.SG.M earth (J) water μ ‘For men he created the earth and water.’

(R.V, 2.20.7c )

On the other hand, a structure like the one in (62) could only be an instance of post-syntactic movement. This is so since the target of movement is syntactically inaccessible and head-immovable given that the set of ε-accessible terminals is empty (null C0 ) and does not contain the wh-terminal (itself originating within the specifier of the kártv¯a-headed CP). Assuming “phonology doesn’t have to ‘read’ syntactic boundaries,” since “it just applies to each chunk [terminal] as it is received” (Samuels 2009, 250), the syntactically inaccessible wh-correlative ya´¯ is made available to μ0 post-syntactically, thereby checking via movement the [ε] feature. (62)

ya´¯ i ca ti kártv¯a kr.ta´¯ ni made.PRT. (J) which.REL μ to.be.made.FUT.PART ‘. . . what has been and what will be done.’

(R.V, 1.25.11c )

2.3 Back to Indo-European

45

So far, we have set a system of post-syntactic rescue for ε-checking, appealing to the post-syntactic access of the internal structure of specifiers and availability of post-syntactic incorporation of narrow syntactically frozen specifiers. Now we turn to cases where the edge (understood as containing the specifier and the head) of a phasal category is empty. Take (63): (63)

raks.áso hanti slay.PRES.3.SG demons.ACC.PL ‘He slays the demons.’

(R.V, 5.83.2a )

The present verb hanti seems to sit in T0 with the object, the ‘demons’, lower in the structure, presumably in its V-complementing in situ position. Assuming the category of (63) is that of CP, we see that the CP edge is empty: the indicative C0 is phonologically null and no syntactic material has been extraposed or otherwise moved to any of the left-peripheral CP specifiers, such as a Focus head. Should such a CP undergo coordination, the [ε] feature on μ0 would not be deleted. Given our assumptions, this would cause the derivation to crash.8 The structure in (64) sketches this scenario, where there are no syntactically or post-syntactically accessible terminals within μ0 ’s search domain. The Wackernagel effect is therefore blocked by virtue of there being no suitable post/syntactic material below μ0 . (64)

Derivational accessibility and the impossibility of narrow-syntactic 2P effect: JP ...

a. ε -checkable terminals narrow syntactically: 0/ b. ε -checkable terminals postsyntactically: 0/ c. closest accessible terminal: 0/

J μP

J0

EMP T

Y

μ0 [ε ]

0/

Cπ P

A SS

E MPT Y

IN CC E

0/

C

IB LE

C0π

TP

0/

8 This

is also amenable to the idea that fully spelled out phasal domains are converted into atoms, i.e., they are atomised in the sense of Takita (2010) and Fowlie (2013). While such views are consistent with the analysis I advocate, I leave this aside.

46

2 Construction

The structure in (63) is nonetheless a coordinand: I content that as a last resort operation, the otherwise silent J0 receives phonological realisation for ε-checking reasons. The full internal coordination structure of (63) is given in (65). The last resort mechanism qua phonological realisation of J0 may be analogised to expletive subjects in a language like English. Just as there is no subject (in the vP) eligible to raise to Spec(TP) in sentences like ‘There is no-one here’, an expletive subject is realised as a last resort. Equally, when there are no eligible heads for [ε]-checking, J0 is overt. (65)

u -tá hanti raks.áso J μ slay.PRES.3.SG demons.ACC.PL ‘And he slays the demons.’

(R.V, 5.83.2a )

The proposed analysis is also an explanation of an empirical generalisation that has been extensively shown to hold not only in R.gvedic (Klein 1985a,b; cf. also Klein 1974, 1978) but also in Old Persian (Klein 1988) with validity very likely holding across the vast array of ancient IE languages (Mitrovi´c 2014, Klein 1992, cf. Agbayani and Golston 2010).9 (66)

Categorial generalisation: Peninitial coordinators do not feature in clausal coordinations, where the clausal conjuncts have an empty edge.

This synchronic analysis extends naturally into the diachronic one I spell out in Chap. 4, specifically Sect. 4.1, where the reader may pick up the rest of the story.

2.4 Beyond IE At this point, I put IE on hold and examine how the JP structure works, or would work, when more than two coordinands are coordinated. In this subsections, I explore some comparative and typological considerations.

2.4.1 Polyjunctions and the Case of More Than Two (Con)juncts It seems natural, desirable, and theoretically parsimonious to assume that con/junction of n-many elements, where n > 2, is handled by the same single structural principle as the one legislating con/junctions of (con)juncts (pace Chomsky 2013). 9 Note,

however, that neither Klein (1992) nor Agbayani and Golston (2010) make this claim but provide evidence consistent with the claim, explicated in (66).

2.4 Beyond IE

47

The structure of (67) may correspond to either (67) whether the internal con/junct of J01 is another JP (JP2 ), or (67) where the external con/junct in Spec(JP1 ) is another JP (JP 2 ). (67)

Gandalf, Billy Bob, and Zebedee. a. Rightward recursive J:

b. Leftward recursive J: JP1

JP1

Gandalf J01

J1

JP2

J1

DP

DP

JP2

Gandalf J02

J2

DP Billy Bob J02

J01

J2 DP

DP Zebedee

Billy Bob

DP Zebedee

The nature and conditions of choice between (67) and (67), however, are unmotivated. Hence, let us eliminate the choice and assume, for purposes of consistency of the system developed thus far, that both (67) and (67) are true. Namely, assume that a recursive Junction structure corresponds to (70), where iterativity is permitted to be both leftward and rightward. Call it ambidextrous recursion of J and let JP2 from (67) now correspond to JP3 . (68)

Ambidextrous recursion of J (ver. 1): JP1 J1

JP3 DP Gandalf J03

J01

J3 DP Billy Bob

JP2 DP Billy Bob J02

J2 DP Zebedee

This yields over-generation of argument con/juncts on a descriptive level. For simplicity, let us revert temporarily to con/junction of two arguments. We further assume the principle in (428), for reasons to be expounded upon. (69)

Spec(JPn ) may only be filled by a Junction or sub-junction category.

The principle in (428) reduces all con/juncts to be merged in complement sites, which was already motivated by Kayne (1994) on independent grounds. On a more general level, this principle, as operative in (70), can be seen as a derivational reflex of the Curry-Howard correspondence in mapping a sequence of J-functions to unary

48

2 Construction

functions, with the exception of the matrix J01 . (I will try and eliminate this exception and maintain the principle without loss of generality.) (70)

Ambidextrous recursion of J for n = 2 (ver. 2): JP1

J1

JP3 J03

J01

DP

JP2 J02

Gandalf

DP Billy Bob

Here is a minimalist rationale for this principle as motivated by labelling considerations (Chomsky 2013): [uCAT] on J is categorially valued and labelled as DP (DP being a complement to J0). The label of the JP depends on Λ1 (?, DP ), where the ? is another Junction Phrase, labelled D at J level via complementation and its maximal category is labelled upon merger of the external conjunct, yielding Λ2 . Spec(JP) corresponds to JP until a labellable object obtains. Its label percolates from the most deeply embedded to the matrix JP. For n > 2, the ambidexterity must be broken, since either the right (complement) or the left (specifier) node will recur. That is, in con/junction of three arguments, the structurally highest con/junction will con/join an argument directly and the other two indirectly by virtue of another con/junction of two arguments. Where do we place this latter (embedded) con/junction? I suggest this is the left edge. Consider, then, a revised structure for polysyndetic coordination. (71)

Ambidextrous recursion of J for n = 3 (ver. 2.1): JP1

J1

JP3

J04

J01

J3

JP4 DP Gandalf

J03

DP Billy Bob

JP2 J02

DP Zebedee

Semantically, both right- and left-recursive junctions yield identical meanings, since con/junction is associative (see Mitrovi´c 2014, Sect. 2.4; Appendix D and those he cites). Let’s further distinguish dyadic J categories which take two arguments from monadic sub-Junction categories that take single complements or arguments. The latter are, as independently motivated in previous sections, superparticles of μcategory. Once this is imposed on (71), the resulting phrase structure is the one given in (72).

2.4 Beyond IE

(72)

49

Ambidextrous recursion of J for n = 3 (ver. 2.3): JP1

J1

JP2

μ20

J01

J2

μ2 P J02

DP

μ1 P μ10

DP

Zebedee

Billy Bob

Gandalf

DP

As a final assumption, let us suppose that a two-place function J cannot combine directly with (an extended projection of) a lexical item. Therefore, the complement of J03 is a single-argument-taking μ3 , satisfying (428). (73)

Ambidextrous recursion of J for n = 3 (ver. 3; final): JP1

J1

JP2

μ30

J01

J2

μ3 P DP Gandalf

J02

μ10

μ2 P μ20

μ1 P

DP

DP Zebedee

Billy Bob

Let M be the set of all μ terminals, J the set of all dyadic (two-place taking functor) terminals (J heads), and N the set of all con/juncts (true lexical argument). The following holds for n-many con/juncts (for n > 1, such that n = #(N )). (74)

#(M) = #(N ) = #(J ) − 1

Let me now show that (73), which rests on (74), is an empirically valid generalisation. For conjunctions of n = 2, Hungarian and Avar use three conjunction morphemes. In light of the analysis developed, one of the conjunction markers is a J category and the other two markers superparticles of μ-category. The distinction between the two types is additionally confirmed by the different morphological shape of particles in question: (75)

a. Ser-Bo-Croatian: i Gandalf pa i Bilibob pa i Zebidija vole pasulj μG J μB J μZ love.PL beans b. Hungarian: Gandalf -is és Bilibob -is és Zebedi -is G μ J B μ J Z μ

50

2 Construction

c. Avar: Gandalf -gi va Bilibob -gi va Zebedi -gi G μ J B μ J Z μ

2.4.1.1

Motivating Polysyndetic Left-Recursion

The reader may verify how (74) empirically corresponds to (75). The proposed analysis is in contrast to the traditional derivations in that the basic idea behind the proposal is that recursion is left-ward, which now requires further support. I provide two: one capturing a parameter on single realisations of the conjunction marker in polysyndetic coordinations. The other concerns labelling.

Chain Reduction The first argument pivots on Nunes’s (2004) theory of Chain Reduction (CR), which I generally assume to be operative. More precisely, my question concerns the directionality of reduction with respect to externalisation of the most prominent element of the chain. Let us preliminarily assume that polysyndetic con/junction involves chains of J-heads. We base this on a copy theory of polysyndetic coordination. If we return to the arithmetic relation between the structural amount of J and μ elements (456), this amounts to the following interim generalisation: (76)

a. μ0 -type markers: one marker per con/junct. b. J0 -type markers: one marker per two con/juncts (i.e., per pair).

The formal principle of polysyndetic coordination that Zwart (2005) proposes only states that the number of overt coordinators is equal to the number of coordinands, where the latter count only ranges over J0 heads in our system, ignoring sub-junctional heads like μ0 . In (77), we list some cardinality properties of the proposed system of leftward recursion, taken from Mitrovi´c (2011a, 35n2.48). (77)

Formal correspondence of syntactically and phonologically realised μ0 and J0 coordinators (M) and coordinands (N) in polysyndetic constructions a. the number of phonologically realised μ0 heads: #(μ0 ) = m

b. the number of phonologically realised J0 heads: #(J0 ) = m − 1

2.4 Beyond IE

51

c. the number of all syntactically present con/junction markers, including both covert and overt J and μ elements: #(M) = 2m − 1 Comparing the number of over coordination markers in a language like Ser-BoCroatian with a language like English, we notice that English employs one fewer coordinator for an n-ary sequence of coordinands. We take this observation and the correspondences listed in (77) to be preliminary diagnostics for the coordination structure. Since Ser-Bo-Croatian falls in the (77a) category above, we classify SerBo-Croatian as a μ-marking language; conversely, since English falls in the (77b) category above, we classify English as a J-marking language. Let me now turn to a theoretically implicit generalisation regarding linearisation of coordinate structures. Kayne (1994, 57) states the following pair as asymmetrical and signalling that the coordinator must occur before the last coordinated DP (his 1 and 2) (78)

 I saw John, Bill and Sam

(79)

∗ I saw John and Bill, Sam

While the above symmetry holds for English, there are at least two languages, Tibetan and Amharic, which confirm the non-universality of such an asymmetry. Here, I look at Tibetan only. The following examples are from Classical Tibetan, taken from Beyer (1992, 241). While the first example (80) shows a pattern similar to English, the second one (81) is unlike English in that it is in stark contrast to Kayne’s asymmetry above. While English realises the penultimate conjunction in polysyndetic structures, Tibetan realises the pen-initial (2P) conjunction. (80)

sa-dan¯ thu-dan¯ me-dan¯ rlun earth-and water-and fire-and air ‘earth and water and fire and air’

(81)

sa-dan¯ thu me rlun earth-and water fire air ‘earth, water, fire, and air’

The following two strings, then, are natural linguistic possibilities: (82)

 I saw John, Bill and Sam

[English]

(83)

 I saw John and Bill, Sam

[Tibetan]

The desideratum is to derive both possibilities from a single syntactic structure, that is, assuming on conceptual grounds that all natural language coordination structures are uniform and that the differences in the patterns of realisation may be accounted for by appealing to different mechanisms (e.g., at the level of phonology).

52

2 Construction

The parameter that captures this contrast without reference to different coordination structure motivates my left-recursive theory. The polysyndetic syntax we are assuming is the left-branching one developed in Mitrovi´c (2011a), as shown in (84), where χ 0 is a variable over connectives such as μ0 , κ 0 or den Dikken’s (2006) subjunctional and in English. (84)

Differential monosyndetic realisation of con/junction markers in an n-ary coordination: JP J

JP J

JP J0

J J0

χP χ0

English: Tibetan:

0/ dan¯

χP χ0

χP χ0

χP χ0

XP

earth earth

J0

YP

WP

ZP

wind wind

0/ 0/

f ire f ire

and 0/

water water

We further account for the asymmetry in realisation shown in (84) by appealing to the notion of CR and the distance calculated at two separate modular procedures of the post-syntactic derivation. (We explicate the notion of chain below.) With respect to the structure of the Spell Out, we adopt Arregi and Nevins’s (2012) theory. This theory also allows for parametrisation, which will capture the two exponence types as postsyntactic parametric possibilities, stemming from a single structure underlying the two types. We assume that this ‘parameter’ applies post-syntactically, i.e. at the sensory-motor interface. For independent experimental evidence that a parameter like the one in (85) is likely, see Willer Gold et al. (2017) and those they cite. (85)

Linearity vs. structure sensitivity and the exponence parameters for nary coordination: a. English: Reduce coordinate chain and assign phonological index to the structurally closest terminal (from left to right). b. Tibetan, Amharic: Reduce coordinate chain and assign phonological index to the linearly closest terminal (from left to right).

In contrast, a μ-marking polysyndetic language like Ser-Bo-Croatian, involves an extra conjunction exponent. Compare the number of conjunction markers in (86) with the number of markers in English (87) (86)

Fata poznaje [i Muju i Hasu i Smaju] F knows and M and H and S ‘Fata knows (both) Mujo and Haso and Smajo’

[M = m]

2.4 Beyond IE

(87)

53

John knows [Bill and John and Mary].

[M = m − 1]

The polysyndetic data is problematic for all approaches which assume a single coordination head (such as Chomsky 2013, for instance). For coordination of three coordinands, a μ-marking language like Ser-Bo-Croatian realises three conjunction markers, while a J-marking language like English realises two, in line with (77). A μ-marking language is thus analysed just as Tibetan or English, modulo the shift of phonological realisation from J0 to χ 0 in (84).10 Munn’s (1993) analysis of polysyndeticity, or iterated conjunction as he calls it, assumes a multiple adjunction to his Boolean Phrase (BP). Take a conjunction string like Tom, Dick, Harry and Fred analysed in (88). (88)

Polysyndetic (iterated) conjunction as multiple adjunction to the Munn’s (1993) BP (Munn 1993, 24n 2.18): NP

BP

NP Tom

BP

NP Dick

BP

NP Harry

B0

NP

and

Fred

It is not clear how Munn (1993) would structurally treat polysyndetic conjunction with overt exponents. To maintain endocentricity within a BP analysis, one would have to resort to multiple BPs and not just a single one with multiple adjunction slots. Once this is stipulated, the structure has one other difference, namely right-branching recursion, while I have maintained a left-branching recursion for empirical and theoretical reasons given above. This brings us to another matter implicitly unresolved in (85), where we have introduced a notion of coordinate chain. Conceptually, we would like to capture the core empirical observation that coordination is unbounded and reconcile this universal with a theoretical tool which would allow recursive copying of the coordinate structure. Consider, therefore, a coordination chain generalisation: (89)

JP chain generalisation A JP may freely copy and adjoin a copy of itself to itself and only itself.

The informal stipulation above translates into a generalised structure in (90), which guarantees the left-recursive character of J. 10 I

am not aware of any Tibetan-styled μ-marking language, which would realise the linearly closest μ-marker. This is a possibility our system allows, ceteris paribus.

54

(90)

2 Construction

A copy-theory of polysyndetic derivation: JP1 JP2 J0

JP3 J0 J0 JPn

J0

The triangular structures in (90) are used to specify the coordinand positions. With this structure in place, the linearity vs. structure sensitivity and the exponence parameters for n-ary coordination in (85) can operate differentially in the two language types. In English, the realisation of the penultimate con/junction marker is dictated by structural proximity between the structural closest J-head and an abstract Boolean operator β (which I introduce and explicate in the next chapter). This is shown in Fig. 2.1. Conversely, in a language like Tibetan or Amharic, the realisation of a single con/junction marker in polysyndetic expressions is dictated by linear proximity between the relevant subjunctional head and the abstract β head. This is shown in Fig. 2.2. Let me now turn to the labelling issues, which may be resolved if left-recursive Junction is employed.

Labelling and Extension Recent work has shown that labelling may also trigger narrow syntactic movement (Chomsky 2013, 2015). Movement generally obeys, or rather instantiates, the extension condition, hence instances of root extension are subject to the running of the labelling algorithm. In line with Chomsky (2015) and Rizzi (2015), consider the following visibility condition for extension (VCE): (91)

Visibility condition for extension (VCE): Unlabelled objects do not satisfy the extension condition. (That is, a label

2.4 Beyond IE

55

Fig. 2.1 English-type realisation of con/junction markers (in polysyndetic expressions) dictated by structural proximity

is a precondition for extension; or, an unlabelled syntactic object cannot extend the root.) Con/junctions do not have an intrinsic (categorial) label, nonetheless, conjunctions are categorised syntactic objects (cf. fn. 3). It is standardly assumed that the arguments (con/juncts) provide the search algorithm with categorial candidates for the label of the junction phrase. In rightward-recursive polysyndetic junction, a violation of the VCE principle ensues. Let me exemplify this idea by invoking a simple version of the Labelling Algorithm: consider a syntactic object α with two daughters, each of which has an intrinsic label. Thus {β, γ } is the set of possible labels for α, from which a single label Λ may be derived. The Labelling Algorithm attempts to label α as either β or γ or, should neither be a suitable label, attempts to find a feature x which both daughters have in common. Note that this is a simplification of Chomsky’s (2013) system.

56

2 Construction

Fig. 2.2 Tibetan/ Amharic-type realisation of con/junction markers (in polysyndetic expressions) dictated by linear proximity

(92)

Λ2

β

α γ

Λ1

δ

ε

a. γ needs to be labelled before the merger of β : Λ 1 (γ)({δ ,ε }) = γ,λ | λ

{δ ,ε ,x[δ (x) ε (x)]}

b. α may be labelled only after γ is labelled: Λ 2 (α )({β ,γ }) = γ,λ | λ

{ β ,γ ,y[β (y) γ(x)]}

Therefore, from a labelling point of view, extension of an unlabelled root cannot ensue. How does this translate into the ‘directionality’ choice for polysyndetic con/junction? If labelling of the intermediate Junction projection obtains, then both leftward and rightward recursion of Junction will be equally amenable to explaining the empirical facts.

2.4 Beyond IE

57

One desideratum a theory of coordination should meet is an account of symmetry which is predominantly ensured by grammar.11 Symmetry of con/juncts is guaranteed as long as a special type of attribute for the unvalued categoriality of the Junction head is assumed. In this regard, assume a crash-intolerant categorial lack of valuation [uCAT] on the J head. After percolation of the valued categorial feature to the intermediate projection level, the labelling procedure at the point of merging the external con/junct in the specifier site may output only two possible labelling results: either the label of β or that of α, where α corresponds to the external con/junct.

2.4.2 CSC Violations Explained There is another benefit of the split conjunction system which identifies a lower μcyle, with a derivational and interpretational life of its own if not embedded, and a higher junction superstructure which glues μPs together (for the semantics of J0 , see Sect. 3.2). The split structure resolves apparent violations of the Coordinate Structure Constraint (CSC), as given in its original format in (93), which is generally understood as an inviolable cross-linguistic constraint. (93)

The Coordinate Structure Constraint (CSC): In a coordinate structure, no conjunct may be moved, nor may any element contained in a conjunct be moved out of that conjunct (Ross 1967, 161).

Consider the evidence from Sanskrit in (50) and from Avestan in (95), given as expressions of coordination. (94)

ca ti loka¯´ n ]α upa-hváyate [ tα [ ima´¯ ni these.ACC.SG μ world.ACC.SG summon.2.SG.PRES eta´¯ nij ca tj sa´¯ m¯ani ]CS these.ACC.PL μ chants.ACC.PL ´ 1.8.1.19) ‘He summons these worlds and these chants.’ (SB,

(95)

k¯@ huu¯apå [ raocåsi c¯a ti ]α d¯at [ tα t@måsj c¯a tj ]CS ˜ who artist give.AOR light μ dark μ ‘What artist made light and darkness?’ (YH, 44.5.b)

Both (94) and (95) clearly violate CSC insofar as a conjunct is moved from the ¯ object position to some A-position in the clausal edge. The following analysis is common to both of them.

11 I

ignore here unbalanced and asymmetric coordinations.

58

2 Construction CP

(96)

VP

μPα μ0 D0

V0

DP

JP

μ0

J

μPα μ0 D0

J0

DP

μP μ0

μ0

DP

D0

μ0

CSC v io

la t

io n

If the μ-headed phrases are not analysed as conjuncts but as independent additives, such instances of CSC violations disappear since additive μP are not subject to CSC by definition. Consider the Sanskrit and Avestan data, this time analysed as involving two μ-headed additives. (97)

CP μP

CP

μ0 D0

μ0

VP

μPα DP

V0

D0

μPα μ0

μ0 D0

DP μ0

DP μ0

n o C S C vi ola

ti o

n

Understanding the structurally non-conjunctive μP additives, the reading of the Sanskrit example in (94) would be along the lines of ‘He summons these worlds also, (and) these chants, too’. The Avestan example, similarly, would be, reflecting the CSC-compliant additive structure, ‘What artist also made light? (And) darkness, too?’ Beyond the obscure verses of the Br¯ahman.as (94) and the Yasna, parallel CSC violations have been argued to exist in Japanese and Ser-Bo-Croatian. See Stjepanovi´c (2014); Oda (2017); Boškovi´c (2017) and those they cite to see how the assumption that the presence of coordinators automatically projects a phrase which is subject to the CSC. (98)

Knjigei je Marko [ ti i filmove ] kupio and films bought books AUX M

2.4 Beyond IE

‘Marko bought books and movies.’

59

(Boškovi´c 2017, 2n5b)

While the conjunctive inference exists for sentences like (98), as they do for Sanskrit or Avestan, such sentences are easily analysed as involving an additive adjunct of μ category. I have already demonstrated that the Slavonic i is a μ superparticle (I discuss it further in the chapters to come). My analysis of these alleged CSC violations rests on the enriched JP structure which allows nonjunctional μP additives. Thus, Boškovi´c’s (2017) example in (98) is, I submit, better understood as an expression involving an additive structure as in (99) and in line with the characteristics of the analysis in (97). (99)

Knjige je Marko [μP i filmove ] kupio books AUX M μ films bought ‘It is books, (and) also films, that Marko bought.’

How the semantics of additivity delivers conjunctive inferences is a topic I address in Chap. 3. Before finishing with this chapter, I devote the last portion of this section to tying together the preceding ideas. The main question being, how can the current system predict the typological distribution of facts, i.e. when will a given language express a μP as opposed to a vanilla marker of coordination (What distinguishes Japanese mo or Slavonic i from Modern English and)?

2.4.3 A Parametric Typology of Junction Systems How do we contextualise parametrically the availability and general place of the μ system within a larger system of cross-linguistic expression of conjunction and coordination generally? Since 67% of world languages express μ meanings, i.e. employ conjunction markers which also express non-conjunctive meanings (Gil 2005), I turn to proposing (at least in a programmatic format) a parametric hierarchy (in the spirit of Biberauer and Roberts 2017) for expression of coordination. Organising parameters hierarchically not only models the presumably natural acquisition pathways for learners (who set their grammars deductively, moving down the hierarchy), but provides an implicational system for understanding the typological variation. Thus, if a parameter is set negatively higher up the hierarchy, the grammar cannot posses features arising from an independent parameter lower down in the hierarchy. While most languages of the world instantiate some flavour of μ (Gil 2005), not all do. So what other properties should the grammar of coordination have in order to express μ? I outline the prerequisite ingredients and parametric conditions for this here. Figure 2.3 shows the proposed parametric hierarchy. One of the main questions concerns the semantic switch between conjunction generally and the expressibility of μ meanings (conjunction being one of them). The predictive power of the parametric hierarchy is implicational: a language that

60

2 Construction Is coordination lexicalised?

YES

NO

Middle Egyptian (WALS#=6)

Is the

logical contrast lexicalised?

NO

YES

Is

Warlpiri, ASL

cat.-sensitive?

NO

YES

NO

YES

UNIVERSAL

UNATTESTED

Syncategorematic t-type conj. of all categories.

e-type conj. marker is  superparticle

{most modern IE languages, . . . }

{67% of langs.}

Fig. 2.3 A parameter hierarchy for the junction system

does not lexicalise the semantic contrast between conjunction and disjunction is, logically, predicted to be able to lexicalise μ meanings associated with nominal conjunction and quantificational uses. Another conjectured universal regards the categorial sensitivity of coordination (at least insofar as the morphosyntactic notion of categoriality and semantic notions of type may be parallelised). While conjunction has been shown empirically not to be syncategorematic (Zhang 2010; Mitrovi´c and Sauerland 2016), disjunction appears to be different in this regard. I am not aware of a language which employs different disjunction morphemes for distinct categories. Benjamin Slade (pers.

Appendix

61

comm.) brings the Kannada facts to my attention, as reported in Amritavalli (2003). Kannada uses two types of disjunction markers, -oo and illa: however, the contrast between the two disjunction particles seems to be more complex with regards to the non-disjunctive meanings of illa. The reports of the Kannada facts of disjunction do not suggest that the -oo/illa split is a categorial or a type-sensitive one (as suggested by Mitrovi´c and Sauerland 2014, 2016 for the nominal/propositional distinction in conjunction systems). Even if a language arises, which lexicalises a DP-level and TP/CP-level disjunction, the general typological rarity of such systems should keep future theoretical research going (as well as an important difference with the μ systems, but see my Conclusion on this note).

Appendix The aim of this appendix is to provide some further empirical substantiation to the claims made for IE coordination in the chapter; namely, that it operated a ‘double system’: both the 1P and the 2P configurations were possible expressions of conjunction. I take each of the IE branches in turn and demonstrate this briefly. I will also use a morpheme-breakdown paradigm notation to explicate the general idea put forward, namely the idea of particle compounding, in line with the analysis generally and Table 2.3 specifically. We list in (100) an example of such a morpheme-breakdown for English disjunction (interrogative) markers. (100)

English disjunctive/interrogative morphemes: ∅ either n either wh ether ∅ or n or

In the following subsections, we will aim to provide as concise a discussion as space permits of the connective particle systems for all IE families, focusing on representative languages. As we do so, we also pick up on some synchronic and diachronic topics of interest in light of our analysis.

Greek Throughout the subsections, we will be reviewing the connecting particle sets that each of the language families features. The descriptive aim is to uncover the patterns underlying the particle sets (lexicon), since we are concerned with the morphosyntax of the particles—and in the subsequest chapters, we will add the semantic aspect,

62

2 Construction

which will rest on the morphosyntax, qua compositionality. For a full list of particles attested in Attic in the fourth and fifth centuries AD, see Humbert (1954, 374ff.). A subset of the list is reordered below with respect to the central morphological role of particles like te, toi, etc. It is clear from (101) that a set of six particles can give rise to nine particle forms since the particles combine with others. (101)

Connecting particles in Homeric.: aú aú kaí kaí kaí mén mén é é

te toi te toi te

For an exhaustive descriptive treatment of Greek particles, see Denniston (1950) and, especially, Ruijgh (1971). Homer’s Odyssey and Iliad are dated to 750–650 BCE, following Taplin (1993). Hesiod’s three works (Th, WD, HS) are dated to the same Homeric period (Griffin 1993, inter. al.). Consider the distribution of the two coordinator types in Homer and Hesiod in Table 2.4, where we list the number of occurrences of kaí and te, and the ratio of the occurrences per author, which suggests a rather ‘balanced’ grammar of conjunction where both 1P and 2P markers were used equally frequently. Decker (2013) gives statistical and distributional statistics of the coordinate use in the Greek New Testament, shown below (Tables 2.4 and 2.5). The diachronic facts are taken up in Chap. 4, especially Sect. 4.1.

Epic Meanings What has traditionally been labelled ‘epic te’, is, as Probert (2015, 109), explains an adverb that appears in some relative clauses, normally non-restrictive ones (but see also Goldstein 2014). The observation that non-coordinate te occurs in nonTable 2.4 Grammatical status of 8th c. BCE conjunction system of Greek Homer: Il & Od Hesiod: Th, DW, & SH Σ

kaí 5287(56%) 512(44%) 5799(55%)

te 4091(44%) 554(56%) 4755(45%)

Σ 9378 1176 10, 554

Appendix

63

Table 2.5 Grammatical change in the Greek conjunction system of Greek from 8th c. BCE to 15th c. AD Period 8th c. BCE 5th c. BCE 2nd c. CE 15th c. CE

Text(s) Il, Od, Th, DW, SH Hist GNT Chron

kaí (N ) 5799 3671 7715 1839

te (N ) 4755 1465 200 40

Σ(kaí, te) 10, 554 5136 7915 1879

kaí (%) 54.95% 71.48% 97.47% 97.87%

te (%) 45.05% 28.52% 2.53% 2.13%

restrictive relative clauses (“expressing permanent states of affairs”, Probert 2015, 108) goes back to Ruijgh (1971). Kviˇcala (1864) presents a theory of the ‘generalising te’ according to which the semantics of the indefinite pronominal term coupled with te yields an irgendtype [sic] indefinite semantics. Lammert (1874) later took up Kviˇcala’s (1864) programme and refuted it by demonstrating that the indefinite terms, such as ós te as related to ós tis, are very rare (note that kva ca and kva cit, being a par, are common alternations in Sanskrit and early Indic generally). The term ós te generally functions as a relative-anaphoric term and the indefinite semantics related to the construction is excluded. Termed ‘epic te’, the classical philological scholarship has long recognised the non-coordinate meaning of @mphte (e.g. Hoogeveen 1829), covering a range of constructions found in the Epic period of the language, which we could describe using modern linguistic terms such as additivity, relativisation, and modality.12 In the examples below, we list some non-coordinate uses of te. First, notice the focussensitive role of te in (102). (102)

peletai koros aipsa te phulopidos quick μ battle-cry.FEM.GEN.SG become.PRES.MP.3.SG satiety.M.NOM.SG anthr¯opoisin men.M.DAT.PL ‘ When there is battle men have [suddenly]F their fill of it’ (Il. T. 221)

Consider also the additive function, as shown in (103) and (104), which seems in line with the general emphatic function (103)

ke theois epipeith¯etai mala may.PRT gods.M.DAT.PL be-persuaded.PRES.SUBJ.MP.3.SG very.ADV t(e) ekluov auto¯u μ head.IMPRF.3.PL self.M.3.GEN.SG ‘If any man obeys the gods, they listen to him also.’ (Il. A. 218)

os

REL

12 Philological

scholars use terms like ‘responsive’ (Denniston 1950, 520), or ‘superadditory’ (Hoogeveen 1829, 181).

64

(104)

2 Construction

bothrov oruks ossov te hole.M.ACC.SG dig.AOR.ACT.1.SG REL.N.ACC.SG μ pugousion of-a-length.M.ACC.SG ‘[dug] a pit, and one too of the measure of . . . ’ (Od. L. 25; trans. by Hoogeveen 1829, 182, X)

We also find te as an FCI forming element as shown in (105), which Denniston (1950, 533) calls a ‘general’ function. (105)

shetlie kai men tis te hereioni reckless.M.VOC even indeed.PRT who μ lesser.M.DAT.SG peitheth’ etair¯o trust.IMPERF.MP.3.SG comrade.M.DA.TSG ‘Reckless one, anyone would trust even a lesser comrade’ (Od. Y. 45)

Although we have reviewed the core Greek particles, we cannot, as Morpurgo Davies (1997, 69) warns, assume a priori that all Greek dialects shared the same particles and made the same extensive use of them. To corroborate this claim, we now take into our scope evidence from Arcadian and Mycenaean.

Arcadian As Morpurgo Davies (1997) shows, only two particles occur reasonably frequently: kás/kaí and dé, while others, such as mén, te, atár and allá are rare.13 In what follows, we take each of the representative particles in turn. Arcadian te, although rather rare, shows that it constitutes “a panhellenic phenomenon” since “the history of te everywhere, including Arcadian, cannot be wholly dissociated from that of kás/kaí, largely because of its scarcity of occurrence is determined by the success of kás/kaí” (Morpurgo Davies 1997, 53–54). In Arcadian, te tends to be confined to bimorphemic compound forms, where te is lexically in second position (cf. Lat. at·que), which includes forms such as eí·te, oú·te, and mé·te, where we are employing · to demarcate the relevant morphemes. Outside such compounds, Arcadian te occurs three times in inscriptions and consistently in combination with kaí, forming a (generalised) conjunction structure of the following type:    (106) Z0i te XPj kaí YP , where i ∈ j structurally The following three examples are from those three structures, where te in the first conjunct in (107) takes the form of z’ and ká corresponds to kaí elsewhere. The 13 The supposed atár and the disjunction/alternative marker allá occur only once each in the corpus;

possibly twice for allá—see Morpurgo Davies (1997, fn. 5).

Appendix

65

data, borrowed from Morpurgo Davies (1997, 54, ex. 1–3), is to show that, at least, the ‘simple’ te is no longer part of the language as reflected by Arcadian. This may well explain the fact that te does not reflect the independent semantics of a μP we find across other IE branches. (107)

[kakô]s dz’ heksoloitu ka hodzis tote damioworg¯e [haphae]stai let.perish and horribly and whoever then damioworg¯a let.pay ‘Let him perish horribly and let whoever is then damioworg¯a pay . . . ’ (Dubois 1986: ii, 196; possibly from Pheneos, ca. 500 AD; Morpurgo Davies 1997 op. cit.)

(108)

ha te theos kas hoi dikasstai the and goddess and the judges ‘the goddess and the judges’ (IG-V2 262, 19; Mantinea, 5th c.)

(109)

tas te in E[u]aimoni kai tas i[n Herhomino]i those and in Euaimon and those in Orchomenos ‘those in Euaimon and those in Orchomenos’ Orchomenos, 4th c.)

(IG-V2 343, 49;

Out of the three remaining particles, kás/kaí, dé and mén, the latter is rare, while the other two are not. Morpurgo Davies (1997, 55) establishes the following generalisations, in her own words: i. Arcadian, or at least the Arcadian of the inscriptions, has only a small number of particles, ii. the quasi disappearance of te must be due to a reasonably fast evolution in usage after the Mycenaean period, iii. the form kaí rather than kas of Mantinea and possibly of the rest of Arcadia is due to external influence. For etymological discussion of the kás/kaí alternation, see Lüttel (1981) and Beekes (2010).

Mycenaean Mycenaean operated a system of conjunction using qe as shown in (110), which prima facie resembles the kaí-styled conjunction (in its lexical form) but also the te-type (in its syntactic position) in Greek. Mycenaean data is drawn from Morpurgo Davies (1997, 62–64). (110)

e-ri-ta i-je-reja e-ke e-u-ke-to-qe e-to-ni-jo e-ke-e te-o E priestess has solemnly.affirms-and e-to-ni-jo have.INF god(dess) ‘E the priestess has and solemnly affirms that she has the e-to-ni-jo for the god(dess)’ (PY Ep 704, 5)

66

(111)

2 Construction

[. . . ] do-e-ro pa-te \\ ma-te-de di-wi-ja do-e-ra do-e-ra ma-te pa-te slave father mother-but of-Diwia slave slave mother father ka-ke-u bronzesmith ‘. . . the father (is) a slave, but the mother (is) a slave of Diwia . . . the mother (is) a slave, but the father (is) a bronzesmith’ (PY An 607, 5ff)

It was in Ruijgh (1967) that the Mycenaen de was analogised clearly with Greek dé, as Morpurgo Davies (1997, 63) reports. (112)

ka-pa-ti-ja [. . . ] e-ke-qe ke-ke-me-no ko-to-[no] dwo o-pe-ro-sa-de of-Carpathos has-and divided? ktonai(kek) two having-but wo-zo-e o-wo-ze worzeen NEG-worzei ‘. . . of Carpathos . . . has two ktonai-kek but having to worzeen she does not worzei’ (PY Eb 338, 1–2)14

If the derivation proposed in Morpurgo Davies (1997, fn. 24) to treat dé as deriving from d¯e is correct, being in line with Leumann’s (1949) analysis of alternations in mén/m¯en and dé/d¯e, then the original function of this particle was not adversative. For further discussion, see Ruijgh (1971) whose work remains the most impressive and extant investigation of Greek te.

Germanic The only Germanic language which shows the double system of coordinate construction is Gothic, on which I focus here. Gothic connective particles, as reflected by Jøhndal et al.’s (2014) database, altogether comprise a set of 15 words: (113)

uh, aiþþau, ak, akei, alja, andizuh, aþþan, eiþau, iþ(=uh), jabai, jah, jaþþe, ni, nih, swe, au

Focussing on the second-position -uh, a reflex of PIE  k w e,15 we derive the following paradigm: (114)

Connecting particles in Gothic.:

14 Restoration

as per Morpurgo Davies (1997). Ringe (2006, 117), we assume PIE  k w e  PGmc  hw , cf. Gothic h in postvocalic positions corresponding to postconsonantal uh, also exhibited by (114).

15 Following

Appendix

67

uh andiz uh ja h ni ni h ja þþe þþau ei þþau þþan a þþau a kei a lja a k a As I have already presented in passing, Gothic -uh shows all the traits of the IE Wackernagel coordinator (μ particle) not only in its restriction to second-position but also in its non-coordinate meaning. Repeated below in (115) is uh with the coordinate function. (115)

(galaiþ in praitauria aftra Peilatus jah) came.PRET.3.SG in judgement hall.ACC.SH again P.NOM and Iesu qaþ uh imma wopida called.PRET.3.SG J.ACC said.PRET.3.SG and him.M.DAT.SG ‘(Then) Pilate entered into the judgment hall again, and called Jesus, and said unto him.’ (CA. Jn. 18:33)

Independent in its lower cycle, i.e. not embedded under J0 , and being hosted by a wh-term, uh clearly has a non-coordinate function as indicated in the repeated pair of examples in (116). In combination with a wh-expression, uh delivers a universallike FC expression (116a)16 or a plain vanilla distributive universal (116b). (116)

16 We

GOTHIC: a. [þishvad uh] (. . .) gaggis. [where μ] go.2.SG.PRES.ACT.IND ‘wherever you go’ (CA. Mt. 8:19) b. jah [hvazuh] saei hauseiþ waurda meina and who.M.SG and pro.M.SG hear.3.SG.IND words.ACC.PL mine ‘And every one that heareth these sayings of mine’ (CA. Mt. 7:26)

assume here in line with Chierchia (2013b), and develop further in the next chapter, that FCI are licensed by an interpolating modal. We assume the modal is silent in the case of (116a).

68

2 Construction

Note also that non-coordinate uh is not confined to wh-hosts only, turning them into FCIs or distributes, but rather may combine with other DPs (117) as well as VPs (118). (117)

iþ is ub- uh- wopida but he PRT- μ cried ‘but [he]F cried’

(118)

(CM. Lk. 18:38)

iþ Iesus iddj- uh miþ im but Jesus went- μ with them ‘But Jesus [went]F with them’

(CM. Lk. 7:6)

Eythórsson (1995) assumes that the syntactic position of uh is invariant and heads the CP. The only explicit mention of the non-coordinate form of connective function of Gothic uh is in fact Eythórsson (1995, 81), who assumes that indefinites move to [Spec,CP], where uh is taken to be in C0 . It is unclear how universal quantification over individuals in [Spec,CP] could proceed at all. According to Ferraresi (2005, 150), the clausal role of uh is to introduce a new element into the discourse. Walkden (2012, 123) proposes to treat uh as the lexicalisation of FOC0 so as to allow for focus hosting. The analysis by Walkden (2012, 123) not only makes Eythórsson’s (1995) C-position more precise by virtue of assuming a fine-grained CP structure (Rizzi 1997) but also inadvertently parallels with the intuitions of the present proposals (which will be semantically developed in the next chapter). Our analysis will rest on the exhaustification contribution of μ particles, like the Gothic uh, which will behave like focus (they may as well turn out to be super-/subsets of Focus). For comparative purposes, let us turn to another Germanic language. Other old Germanic languages, such as Old Norse (ON) exemplified in (119) below, have lost the double system of coordinator placement. (119)

Skáli Gunnars var grr af vii einum ok sakir hall Gunnar’s was made with beam one and overlapping-boards tan, ok gluggar hj brnsunum ok snin ar on-outside and windows by ridge-beams.DEF and fastened these fyrir speld. in-front-of shutter ‘Gunnar’s hall was made with one beam and overlapping boards on the outside, and there were windows by the ridge-beams and shutters fastened in front of these.’ (BN, 77)

Note that ON head-initial ok etymologically parallels with Gothic head-peninitial uh. Syntactically, the parametric difference lies in the incorporation-triggering feature (120), which patterns with our prediction of the loss of semantic polyfunctionality of μ superparticles (which ON ok is not). (120)

a. Gth. uh[+ ]

Appendix

69

b. ON ok[− ] (presumably via bimorphemic a-uk) Walkden (2012) is lead to “speculate that the loss of the Gothic (and presumably Proto-Germanic) system of C-domain discourse particles was related to the restricted activation of the expanded left periphery in later Northwest Germanic languages” (p. 125). Aside from Gothic, note that there is evidence that Old English (OE) also possessed a form of μ superparticle, which OE possibly lexicalised as ge. The evidence is distributional: ge forms both universal quantifiers (121) and functions independently as a conjunction marker (122), as the examples show. (121)

Hwa is þætte ariman mæge hwæt þær moncynnes forwearðon on who is that number can what there men.GEN perished on hand? ae-g[e]-ðere PART. DL -μ-which. DAT hand ‘Who (is there that) can number those that fell on each side?’ (Or I, 11; c893)

(122)

se

was æ-g[e]-ðer, ge heora cyning, ge heora biscop was both=PART.DL-μ-WH μ their king μ their bishop ‘He was both their king and their bishop.’ (Or 238, 14; c893) DET

For additional empirical support, see Gast (2013). In the historical Germanic contexts, the superparticle behaviour of the ge morpheme as forming both distributive quantificational as well as correlative conjunctive expressions, is only found in Old Saxon which points to both genetic and historical unity of North Sea Germanic, to the exclusion of West Germanic and Old High German (cf. Walkden 2014, int. al.).

Italic Latin connective particles, as reflected by Jøhndal et al.’s (2014) database, altogether comprise a set of twelve words:17 (123)

ac(=atque), an, atque, aut, et, neu, que, sed, simul, sive, ve, vel

Focusing on the second-position -que, a reflex of PIE  k w e, the following paradigm from a subset of (123) can be derived: (124)

Connecting particles in Latin.:

17 Jøhndal

et al.’s (2014) PROIEL database was used to access the digitalisation of the Latin texts, as well as texts in other languages. Once extracted, the data was statistically analysed.

70

2 Construction

‘AND’, ‘ALL’ que ‘AND’ at que et ‘OR’ aut si ve ve Let us start with coordination patterns. We repeat in (125) the two signature coordination types we are focusing on: one headed by a second-position coordinator que and another in head initial position morphologically containing que. (125)

a. ad summam rem pblicam atque ad omnium nostrum to utmost weal common and to all of us ‘to highest welfare and all our [lives]’ (Or. 1.VI.27-8) b. v¯ıam samtem que life safety and ‘the life and safety’ (Or. 1.VI.28-9)

We list in Table 2.6 the change in the grammar of conjunction from the first century BCE to the fourth century CE with relative (and absolute) values. If we plot this graphically, we may see the decline of one system and the rise of another system of conjunction. We are also averaging over periods—hence, Gal, Att, and Off have been averaged for first century BCE; likewise, Vul and Per. Aeth. have been averaged to give us an idea of fourth century CE conjunction grammar (Tables 2.6 and 2.7). The latest and most extensive description and analysis of the Latin system of coordination is that of Esperanza (2009), who covers a range of coordinate constructions. Since we are focusing on peninitial  k w e and its loss, we report in Table 2.8 Esperanza’s (2009) conclusions, where parentheses mean that the evidence is sporadic while double mark (++) denotes high frequency. Her conclusions are in line with the general trends of loss reported in Table 2.7 specifically and with the trends across IE languages more generally. Let us now consider the non-coordinate forms of Latin que. While wh-que terms are quantificational, forming FCIs or distributive universals, as shown in (126a– 126d), there are also cases where, for instance, quoque performs the additive role (127), which is, as we will defend, the core logical signature and semantic contribution of que. Table 2.6 Grammar of conjunction in Latin: et, que, and atque 1st c. BCE 4th c. CE

Att. & Off. Gal. Vul. Per. Aeth.

et 67.7%(3558) 48.3%(989) 93.2%(5, 662) 89.3%(785)

que 26.8%(1407) 34.3%(702) 6.2%(374) 3.9%(34)

atque 5.6%(294) 17.4%(355) 0.7%(42) 6.8%(60)

Σ 100%(5259) 100%(2046) 100%(6078) 100%(879)

Appendix

71

Table 2.7 Grammatical change from 1st c. BCE to 4th c. CE in the conjunction system of Latin: et, que, and atque

(126)

Romance

Conjunctive coordination et + ++ atque/ac + + -que ++ + Disjunctive coordination ue (+) − aut ++ ++ uel (+) + siue/seu + + Adversative coordination sed + ++ magis − + nisi − −

Late Latin

atque 8.9% 1.5% Postclassical Latin

que 28.9% 5.9%

Classical Latin

Table 2.8 The chronology of Latin coordinators (Esperanza 2009, 482, Tab. 2)

et 62.2% 92.7%

Early Latin

1st c. BCE 4th c. CE

++ (+) −

++ − −

++ − −

− ++ + +

− ++ − +

− ++ − (+)

+ + +

− + +

− ++ −

Latin wh-que as FCIs and distributive universals a. ut, in quo [quis que] artificio excelleret, is in suo genere Roscius that in who [what μ] craft excels, is in his family R diceretur spoken ‘so that he, in whatever craft he excels, is spoken of as a Roscius in his field of endeavor.’ (Cic., de Or. 1.28.130) b. Sic singillatim nostrum unus quis-que mouetur so individually we one wh-μ moved ‘So each of us is individually moved’ (Lucil. sat. 563) c. Morbus est habitus cuius-que corporis contra naturam sickness is reside wh-μ body contrary nature ‘The sickness is the situation of any/every/each body contrary to nature’ (Gell. 4,2,3) d. auent audire quid quis-que senserit want hear what wh-μ think

72

2 Construction

‘they wish to hear what each man’s (everyone’s) opinion was’ Phil. 14,19) (127)

(Cic.

elegia quo-que Graecos prouocamus elegy wh-μ =also Greeks challenge ‘We challenge the Greeks [also in elegy]F .’ (Quint. inst. 10.1.93)

Bortolussi (2013) convincingly argues for a distributive universal semantics of quisque (126a–126d), as I have defended above.

Slavonic The oldest variety of Slavonic and my focus here is Old Church Slavonic (OCS), which comprises the following connective particles, as reported in the Jøhndal et al.’s (2014) database. (128)

a, ali, ašte, da, že, zane, i, ili, li, ljubo, ni, n’, ovo, ta, taže, ti, tože, cˇ ’to, iako

Focusing on three second-position connectives—the disjunctive marker li, and the two conjunctive particles e and i—we derive the following paradigm from a subset of (128): (129)

Connecting particles in OCS.: ‘AND’/‘BUT’

e a ne ta e to e ‘AND’, ‘ALSO’, ‘EVEN’ i ‘BUT’ n’ n i ‘NOT EVEN’ i ako i li ‘OR’ a li ‘BUT’

We will assume that OCS had two markers of μ category: iμ1 and eμ2 .

Negative Polarity Unlike Gothic or Latin, where a combination of μ and a wh-host obtained a distributive universal expression, such morphological constructions in Slavonic are NPIs. I have demonstrated in passing that the OCS i-particle was not only a marker

Appendix

73

of conjunction but also of additivity and polarity. Consider the following two examples below, taken from Willis (2013, 370, ex. 83). (130)

ot˘uvˇeštavaaše n-i-ˇceso-že NEG -μ- WH - REL =nothing answer. IMPF.3. SG ‘He answered nothing.’

(CM Mt. 27:12)

In OCS, the NPI is formed using a wh-stem, like cˇ eso in (130), that hosts the μ particle i, to form i-ˇca¯ eso (μ+what, ‘anything’). Through negative concord of the matrix negative head (following Zeijlstra 2004), the μ is realised with the negative marker as n-i-ˇceso (NEG+μ+what, ‘nothing’, lit. ‘not even one thing’). The following example is cited in Willis (2013, 370, ex. 83) and taken from Veˇcerka (1995, 516). (131)

tvoeje˛ ne zapovˇedi n-i-ko-li-že NEG -μ-who-κ- REL =never command. GEN your NEG prˇesto˛pix˘u. i m˘ınˇe nikoliže ne dal˘u esi transgress.PAST.1.SG and me.DAT never NEG give.PART be.PRES.2.SG koz˘irle˛ te kid.GEN ‘I never broke your command but you have never given me a kid.’ (CM. Lk. 15:29)

The morphological complexity of NPIs can thus be represented in (OCS-NPI), where n-i-koli-že ‘never’, as shown in (132). Following Derksen (2008, 229) and Trubaˇcev (1983, 135–136), I take the etymology of the polarity adverb koli (‘ever’) to consist of the neuter pronoun ko (< PSl.  ko) ‘who’ and the interrogative particle li (< PSl.  li (/ le/ lˇe), Trubaˇcev 1988, 68), which we will independently categorise as a κ particle on the basis of its semantic roles, as discussed in the following chapter.18 We take the n-morpheme to be a concord realisation of the negative operator, which must be locally bound by the negation in the clause (Progovac 1994), following Willis (2013). (132)

Morpho-semantically complex NPIs in OCS and a grammaticalisation pathway for simplification of internal structure:       n i [ [ ko ] li ] že  nikoli(že) : κP DP μP [¬] ADVP REL P

Modern (South) Slavonic NPIs very clearly build out of (at least) a wh-stem and a μ morpheme i (requiring an additional negative concord (NC) element in contexts of clause-mate negation). Old Church Slavonic (OCS) NPIs, as shown, comprise a wh-host and an indefinite NC element ni (which I conjecture contains already at 18 This etymology is contrary to Snoj (1997), who maintains that the ever-particle -koli is originally

a locative of the word for ‘time’, being cognate with Skt. k¯ala (Snoj 1997, 292).

74

2 Construction

that stage of Slavic the μ particle i, see below), as well as (optionally) a marker of contrast and/or relativisation že, which developed into a relativising marker (cf. Mitrovi´c 2016b) across Slavonic. The view that ni in OCS decomposes into a NC morpheme n(e) and a μ superparticle i finds support in Trubaˇcev (1994, 107), who equates OCS ni with i ne (cf. Willis 2013, 393 for a relevant discussion). Additionally, ni is found in OCS as a bisyndetic negative conjunction, analogous to ‘neither . . . nor . . . ’, which parallels the non-negative polysyndetic template i . . . i . . . ‘both . . . and . . . ’, which I take as being the same, module the NC morpheme n-. Assuming, therefore, that ni in NPIs parallels the polysyndetic conjunction with NC, then ni in NPIs is also bimorphemic, by deduction. For further details on the development of polarity in Slavonic, see Willis (2012), Blaszczak (2001, 2002), and Mitrovi´c (2016b) among others, and Haspelmath (1997) for a typological overview.

Indo-Iranian The connecting particles we find in IIr. are listed in (133). (133)

Particles in IIr.: CONJ . ca u ta u ADDIT. ca DISJ . (#1) v a PRT. (NPI) ca PRT. (FCI) ca

Sanskrit Sanskrit, in the narrowest sense, applies to standard classical Sanskrit as regulated by the grammarians but may also be conveniently used more widely as equivalent to Old Indo-Aryan (OIA). In this sense the term traditionally covers both classical Sanskrit and the pre-classical or Vedic language. Middle Indo-Aryan (MIA), that is Prakrit in the widest sense of the term, comprises three successive stages of development: (1) The earliest stage is represented in literature by P¯ali, the language of the canonical writings of the Therav¯ada school of Buddhism. This is a language of the centuries immediately preceding the Christian era. On the same level of development are the various dialects recorded in the inscriptions of A´soka (c. 250 BCE ), and also the language of other early inscriptions. (2) Prakrit in the narrower sense of the word, or Standard Literary Prakrit, represents the stage of development

Appendix

75

reached some centuries after the Christian era. The various literary forms of Prakrit were maintained by grammarians at this period and, as a written language, it has remained essentially unchanged during the following centuries. (3) Apabram . s´a is known from texts of the tenth century AD and was formed as a literary language some centuries prior to that date. It represents the final stage of MIA, the one immediately preceding the emergence of Modern Indo-Aryan languages, which comprise Bengali, Hindi, Gujurati, Marathi, etc. These languages only begin to be recorded from about the end of the first millennium AD and their development can be followed as they gradually acquired their present-day form (Burrow 1955, 2). The diachronic analysis is based on the selected periods ranging from the first attestation of Indo-Iranian (Ir) in the form of Vedic and Avestan to forms of Modern Indo-Aryan (MdIA). As we have seen in Chap. 2, Sanskrit operated a double system of coordination: clausal coordination with medial placement of the coordination, and a sub-clausal system of coordination with non-medial (i.e. second or final) position placement of the coordinator. This double system has been completely lost in MdIA, which operates a single system of coordination with exclusively medial placement of the coordinator. Oberlies (1998, 158) based on ‘cumulative evidence’ dates R.gveda within the range of 1700–1100 BCE. The Veda was originally composed orally and preserved in this fashion for at least 500 years before being written down, which makes dating the R.gveda difficult since it probably emerged as the result of a very long tradition of hymns, some of which go back to before the IndoIranian split, ca. 2000 BCE. It is therefore probable that it had been composed and orally transmitted as early as 1500 and probably no later than 1000 BCE, as Noyer (2011) observes. This section covers the periods listed in Table 2.9. The double system of coordination did indeed undergo loss and replacement by a single system. We will also see that the concept of Old Indo-Aryan as subsuming both Vedic and Epic/Classical Sanskrit cannot possibly be maintained once the syntactic evidence from coordination is considered, since the Vedic and post-Vedic (qua Classical) Sanskrit are grammatically different, at least insofar as the grammar of coordination is concerned. Proto Indo-Iranian (PIIr) or Indo-Iranian (IIr) is the common ancestor of Old Indo-Aryan (OIA) and Avestan. In this section, I attempt to reconstruct the syntax of coordinate complexes in Indo-Iranian. Although this sounds like a problematic task, it may be accomplished if Vedic and Avestan dialects of IIr can be shown to have the same, or at least parametrically similar, syntax of coordination. Table 2.9 Diachronic periods of Indo-Iranian and Indo-Aryan

Time 1700–1100 BCE 1700–1100 BCE 400 BCE–400 AD c. 10th AD Present

Period Vedic Period Avestan Period Classical Period Early modern Period Modern period

Corpus R.gveda Avesta¯a Mah¯abh¯arata Apabram . s´a

76 Table 2.10 Syntactic distribution of Non/Medial coordinate complexes in R.gveda

2 Construction

N R

Medial ca 47.64% 705 1

:

Non-medial utá 52.36% 775 1.10

Vedic and Old Indo-Aryan The earliest document of the linguistic history of Indo-Aryan (IA) is the R.gveda, which is estimated to have been composed between 1700 and 1000 BCE (Oberlies 1998). Vedic is the source language from which all later forms of both Sanskrit and Prakrits developed. The statistical analysis of syntactic patterns in Vedic presented here are based on the entire text of R.gveda. The statistical data on repetition of coordinators and their syntactic positions is adopted from Klein (1985a,b). In Vedic, the distribution of medial and non-medial coordination is equal in that the ratio between initial (utá, ádha, áth¯a) and non-initial (ca) coordinate complexes is generally 1 : 1 as shown in Table 2.10, where N is the sum of all representative tokens and R the ratio between them. Statistically, ca is the most frequently occurring coordinator in R.gveda with 775 occurrences, which Klein (1985a, 46) divides into eight general categories of employment. Let me now turn to the Iranian branch, i.e. to Avestan and Old Persian.

Iranian: Avestan and Old Persian The literature on Avestan and Old Persian, as is often the case with other ancient IndoEuropean languages, is characterised by an absence of in-depth studies dealing with syntax. Both Vedic and Avestan evolved out of an earlier form, which is unrecorded by any direct documentation but it may be reconstructed in considerable detail by means of comparison. By comparing early OIA with the very closely related Iranian, it is possible to form a fairly accurate idea of the original PIIr from which both Iranian and Aryan languages have developed. By comparing IA and Iranian with the other IE languages, it is possible, therefore, to reconstruct in general an outline of the characteristics of the original language from which all IE are derived. In the previous section, I have outlined a statistical sketch of Vedic coordination, which reinforced the theoretical analysis I provided in Chap. 2. In this section, we turn to and compare the established facts about Vedic with the system of coordination in Avestan and Old Persian. Like the R.gveda, the Avesta was composed and preserved orally with rough estimation ranging between 1200 and 600 BCE, and was not put into writing for many centuries (Noyer 2011). The syntax of coordination in the most archaic text of Avesta, that is Yasna, the relation between medial and nonmedial placement of the coordinator is not as balanced as is the case of R.gveda, where the ratio of distribution

Appendix

77

is 1 : 0.91 (Table 2.10). Avestan coordination is overwhelmingly nonmedial: 1530 instances of ca drastically outweigh the mere 9 instances of ut¯a. The ratio is thus: (134)

NON - MEDIAL

: MEDIAL :: 1 : 0.006

The double system of coordination that operated in Vedic is thus replaced by a single system of nonmedial and overwhelmingly polysyndetic coordination. A further diachronic asymmetry comes from the later form of Persian, namely Old Persian texts composed around the 6th c. BCE, which inversely show predominantly medial coordination. Unlike the R.gveda, where utá and ca are competing conjunctions, each characterised with syntactic specialisation with regards to different categories, occurring 705 times and 775 times respectively, the leading exponent of coordination by far in Old Persian is ut¯a, an Iranian cognate counterpart of Vedic ut. Although the coordinators are morphologically and etymologically cognate across Vedic and Avestan (135), their syntax seems very divergent. The etymologies below are based on Misra (1979, 227), where the bimorphemic IE etymology of uta is based on Dunkel (1982, 2014a,b).  (135) a. uta  IIr.  uta  IE  u +  te OP uta, Skt. uta, Gr. ute b. ca  IIr.  ca  IE  k w e Skt. ca, OP c¯a, Lat. que c. v¯a or v  IIr v¯a IE we, cf. Skt. v¯a Gk. e¯ , Lat ve e

The 100 occurrences of ut¯a include 55 instances of sub-clausal coordination, which Klein (1988) subdivides into five subgroups, although only two configurational classes are relevant for our syntactic purposes. Of these, the single most frequent construction is, as expected, X ut¯a Y, as shown in examples (136) through (138). (136)

(137)

P¯arsa ut¯a M¯ada Persian and Median ‘The Persian and Median.’ P¯arsa u[t¯a M]¯ada k¯ara sent.1.SG.PAST Persian and Median ‘(Therefore) I sent off The Persian and Median (army).’

(138)

(DB, 2.18)

Parϑava ut¯a Va.rk¯ana Parthia and Hyrcania ‘Parthia and Hyrcania.’

(DB, 2.81-82)19

(DB, 2.81-82)

The Burgbau inscription (DSf) also involves the same type of head-initial coordination, as shown in the following examples. 19 Cf.

DB 3.29-30.

78

(139)

2 Construction

martiy¯a karnuvak¯a ta[yaiy] aθ an gam akunavan t¯a avaiy Yaun¯a ut¯a men.3.PL stone those made.PAST were those Ionians and [S]pardiy¯a Sardians ‘The stone-cutters who made the stone, those were Ionians and Sardians.’ (DSf, 47-49)

It seems that the overwhelmingly predominant form of simplex nominal coordination is that involving ut¯a, not ca, which is in stark contrast with Avestan and R.gvedic. (140)

(141)

xraθ um ut¯a aruvastam wisdom and activity ‘wisdom and activity’

(DNb, 3-4)

u¯ıy ut¯a fram¯an¯a understanding and command ‘understanding and command’

(DNb, 28)

Statistically, there is no difference between clausal and sub-clausal roles of OP ut¯a, unlike in Vedic. OP ut¯a is also consistently head-initial as there are no occurrences of second or final position placement of ut¯a. Table 2.11 shows the distributional facts about OP ut¯a. There are only 14 instances of c¯a in OP, which amount to merely ten coordinate constructions of which three are uncertain and two are probably wrongly interpreted, that is c¯a in those two cases served a different syntactic role (Klein 1988, 402). We have seen from the Iranian branch of IIr., with Old Avestan as the earliest language, operated a predominantly head-final and 2P system of coordination and that there were only 0.6% instances of medial coordination expressed by ut¯a. In Old Persian, however, the later form of Iranian, the system seems to have undergone a reversal from exclusively non-medial to exclusively medial as less than 10% of instances were headed by head-final c¯a. Now we turn to Classical Sanskrit, where we see that the diachronic asymmetry of coordination that took place in the transition from Avestan to Old Persian is not an anomaly restricted to the Iranian branch of IE but is also clearly indicative of the history of Old IA. Table 2.11 Syntactic distribution of ut¯a in Old Persian

Initial Sub-clausal 63.63% N 35 Clausal 75% N 33

Polysyndetic 36.36% 20 25% 11

Appendix

79

Classical Sanskrit For detailed statistical analyses, I concentrate on the first 151 chapters of the first book of Mah¯abh¯arata. In Vedic, we saw that the earliest form of Sanskrit operated a double system of coordination: the medial placement of the coordinator (such as ut) does not trigger movement, neither of the entire complement, nor an element within it, which is associated with a movement-triggering feature on a coordinating head such as ca or v¯a. In Classical Sanskrit, represented by the extensive texts of Mah¯abh¯arata and R¯am¯ayana, there are altogether 39,369 instances of coordinate structures, almost exclusively involving a nonmedial coordinator. The medial coordination of the uttype is almost entirely non-existent in Classical Sanskrit. For further details, see Mitrovi´c (2011b) and references therein. Classical Sanskrit is, in this respect, a black sheep in the IE family since all other languages that had a Wackernagel-type coordinator lost the enclitic coordinator and replaced it with an orthotonic (freestanding). I conjecturally relegate this exceptional fact to language contact with the head-final Dravidian language family. There are a number of features of Sanskrit which have not been inherited from (P)IE, such as the quotative marker iti (Krishnamurti 2003, 36–37) or the general retroflex phonology.20 Under this conjectured view, the retention of peninitial coordination markers in Sanskrit is contact-fed by Dravidian, which had, and still has, non-initial coordination markers.

The Post-Classical Period It is evident from the earliest MdIA texts, written in early Marathi and/or Hindi, that a single system of medial coordination operated as in the earliest attestations of Marathi (13th c.), coordination was only medial. (Alklujar, p.c.) As Tagare (1948, 334) notes, the tenth century (early MdIA) Prakrit text of Apabram . a possessed the same set of coordinators that were used in earlier Prakrits. We conjecture that the loss of non-medial coordination took place sometime between the post-classical Sanskrit and Prakrit period (4th c.) and thirteenth century AD, leaving the analysis of Prakrit coordination for further research. Modern IA languages all show exclusively medial placement of the coordinator. Hindi coordinate complexes, for instance, are consistently and harmonically headinitial, as shown in examples (142) through (143). (142)

20 For

Manoj lamb¯a aur patl¯a hai. M tall and thin is ‘Manoj is tall and thin.’

further debate and evidence, see Witzel (1999) and Kuiper (1991).

(Snell 2003, 15)

80

(143)

2 Construction

mere pit¯a aur c¯ac¯a donõ net¯a ha˜ı. my father and uncle both politicians are ‘My father and uncle are both politicians.’

(Snell 2003, 19)

Bengali also shows initial headed coordination, whereby ebam . and o are employed for conjunctive coordination and b¯a for disjunctive or contrastive (i.e. adversative) coordination, as indicated in examples (144) through (146). (144)

br.mkara ked.hau k¯alamapke dethila ebam . m¯arila viper hollow tree.LOC saw and killed ‘He saw a viper in the hollow of the tree and killed it.’ (Yates 1849, 118)

(145)

go o mesa o mahisa o ch¯agala carit.eche cow and sheep and buffalo and goat feeding.PL ‘The cow and sheep and buffalo and goat are feeding.’

(Yates 1849, 118)

nauk¯ate b¯a abe sahauba boat or horseback go.1.SG ‘I shall go by boat or on horseback.’

(Yates 1849, 119)

(146)

In Gujarati, there is also no surviving contemporary trace of non-medial coordination, since coordination is consistently medial, that is head-initial.

(Doctor 2004, 68)

ch g n avyo p ï m g n gayo Chagan come.3.SG.PAST but Magan go.3.SG.PAST ‘Chagan came but Magan went.’

(Doctor 2004, 69)

e e

e

e e e e e

(148)

ch g n ï m g n avya Chagan and Magan come.3.PL.PAST ‘Chagan and Magan came.’ e e

(147)

The contemporary Iranian branch of IIr. also shows a single system of coordination, which is harmonically initial, as shown in the example from modern Persian in (149).21 (149)

ya engelisi vä färanse yad begir-id ya almani vä rusi either English and French memory intake.2.SG or German and Russian ‘Either learn English and French or German and Russian.’ (Tabain 1975, 30)

We have seen in this subsection, that Vedic operated a double system of coordination (a sub-clausal postpositive and a sub/clausal prepositive system), which was lost by the time of Classical Sanskrit. Similarly, Avestan also does not 21 Cf.

the disjunction marker ya with the initial interrogative particle a¯ y. See Mauri (2008) and Korn and Öhl (2007) for discussion.

Appendix

81

seem to have operated a double system; it may be conjectured that (a) the double system had been inherited from IE and that Avestan lost it, or (b) that it never developed a double system. In Table 2.12, we list the general statistical details of the double system of coordinate construction in Indic, spanning from Vedic (archaic) to Late Indic.

Anatolian Anatolian is a branch of IE, dated as having split the earliest from the IE core. Wener (1991) presents a branch-internal cladistics for Anatolian, which we show in Fig. 2.4.22 Table 2.12 Development and loss of the double system of coordination in Indic

PERIOD

Archaic Early Epic Classical Medieval Late

uta 45.562% 2.912% 0.838% 2.213% 0.740% 0.699%

ca 54.438% 97.088% 99.162% 97.787% 99.260% 99.301%

PIE

Anatolian

Other IE Languages

South Anatolian

(Proto-Lydian) Hittite Palaic

Cuneiform Hieroglyphic Luwian Luwian

Lydian

Lycian Milycian

Fig. 2.4 A cladistics of the Anatolian branch (Wener 1991)

22 See

Hoffner and Melchert (2008) for further details on history and cladistics of Anatolian.

82

2 Construction

For the Anatolian branch, we will predominantly be focussing on Hittite. There are many connecting particles in Hittite, which we list in (150), following a subword decompositional style we have established by now. (150)

Particles in Hittite: CONJ . (y/m) ADDIT. (y) PRT. (FC, ‘- EVER ’) (y) PRT. (‘∀’) (y) DISJ . (#1) na(u)-m PRT. (NPI) DISJ .

(#2)

a a a a a kki kka (a) ku

We now turn to examine the basics of the syntax and semantics of some of these particles, featuring in wh-expressions as quantificational terms and in non-coordinate structures and contexts, where they perform (non/scalar) additive functions.

Indefinites Indefinite and (negative) polar existentials in Hittite are formed through a morphological combination of a wh-term and a particle. In Hittite, a wh-pronoun kui- ‘who, which’ performs both interrogative and relative functions, and also combines with superparticles to form quantificational expressions. As Hoffner and Melchert (2008, 149) note, the indefinite pronoun ‘some(one), any(one)’ is kuiki, composed of the (inflected) wh-pronoun kui- plus particle -kki or -kka.23

Wh-Quantificational Expressions On the other hand, the universal distributive ‘each(one), every(one)’ quantificational expressions correspond to kuia, comprising of an inflected kui- plus the conjunction -a/-ya ‘also, and’.24 Listed below in (151) are repeated examples of universal quantificational expressions in Hittite: (151)

23 Hoffner

paizzi kuwatta utn¯a a. nu DUMU.MEŠ-ŠU [kuišš-a] who-μ =every somewhere country.LOC went J sons.his

and Melchert (2008, 150) explain the -kki/-kka duality: the particle of kuiki regularly appears as -(k)ki when the vowel in the immediately preceding syllable is i (kuiki, kuinki, kuitki, kuedanikki) and as -(k)ka in other environments. 24 Palaic conjunctions and particles seem to be very similar in this respect—see Carruba (1970, 46–47,60).

Appendix

83

‘Each of his sons went somewhere to a country.’ (KBo. 3.I.1.17–18) b. nu [kuitt-a] arhayan kinaizz[i J what-μ =everything seperately sifts ‘She sifts everything separately.’

(KUB XXIV.11.III.18)

Interestingly, the free-choice ‘whoever’ type expressions are analogous but not identical to universal distributives kuia. As Hoffner and Melchert (2008, 150) show, while universal distributive kuia ( or:  EXH [δA={j,b}] j ∨ b = [j ∨ b] ∧ EXH (j ) ∧ EXH (b)  ¬[j ∧ b] These two exhaustification strategies in (228a-i) and (228b) are equally reasonable means for deriving enriched disjunction in English. The disjunction morpheme or, under very natural assumptions pervading the field, directly maps onto the denotation of ‘∨’, a logical disjunction in itself a scalar-alternative-sensitive operator that is targeted by a silent (probing) EXH. The alternative-targeting domain of EXH may be the σ -dimension, in which EXH denies the conjunctive alternative (228a), or the δ-dimension, in which EXH returns each of the disjuncts are exclusive options (228b). In (228b-ii-a), I show how a local calculation of exclusivity obtains if EXH attaches locally to the disjuncts with wide scope disjunction. The same meaning (with exhaustification applying locally) can occur if the δ-targeting EXH attaches to the root of the disjunction (228b-ii-b). The claim central to this chapter is that the distribution and variety of μmarked expressions are derived from the distribution of the covert exhaustification mechanism. The various applications of EXH over a μ-phrase containing proposition derives those, and only those, meanings we observe typologically (and diachronically which, I address in detail in Chap. 4).

124

3 Interpretation

Before delving into the ways the μ superparticle incarnates its meanings, let me summarise some cornerstone assumptions regarding alternatives. (229)

a. The alternative calculation obtains at a propositional level. b. The alternative set to φ is the extension (set) of φ. i. For a proposition p, the alternatives are, therefore, elements in Dt , i.e. {pw , ¬pw } or {1, 0}. ii. For a wh-term, like who, the extension and the alternative set, subject to contextual restriction, is {x | HUMAN(x)} or all the (relevant) individuals, {a, b, c, · · · } (in the focus dimension, which is the extension of wh-terms).

As above, I use alternative variables a, b, c to stand for propositional alternatives pivoting on the μ-marked element (which is generally but not necessarily an e-type argument) providing the alternative dimension. With an example of how the EXH-based system works in practice, let me now return to μ specifically. The lexical entry for μ in (230) below awkwardly states the aforementioned dual function that μ superparticles have: alternative activation (second line) and exhaustification (third line) against the background of activated alternatives.4 (230)

Lexical entry for or a rule for composition with μ0  (informal; first stab)

While μ-marking is generally a DP-level phenomenon, as μ associates with hosts of type e, the alternatives it obligatorily brings into play must be propositional (of type t), given the technical and conceptual nature of the alternative-based system I am using.5 In order to achieve the seeming type-mismatch between an e-type μhost and its t-type alternatives, I am assuming a Roothian algorithm: “the focus semantic value for a phrase of category S is the set of propositions obtainable from 4 In line with this rationale, note that I am not claiming that μ particles actually lexicalise the otherwise silent EXH operator. I am proposing that the presence of EXH in μ-containing expression is encoded narrow syntactically as an uninterpretable feature on the μ head. 5 One could develop a cross-categorial alternative system for exhaustifying non-propositional alternatives, which (although prima facie feasible and even desirable) is not a standard way of doing alternative semantics. This paper does not try to tweak the alternative-semantic mechanics, it just uses the standard tools that are out there (pivoting on Chierchia 2013b).

3.3 μ

125

the ordinary semantic value by making a substitution in the position corresponding to the focused phrase” (Rooth 1992, 76). Rooth procedure can be formalised in more general terms using Fox and Katzir’s (2011) alternative computation algorithm (which I simplify, following F˘al˘au¸s 2013): (231)

S is an alternative of S if S can be derived from S by successive replacements of sub-constituents of S with elements of the substitution source for S in C. (Fox and Katzir 2011, 97)

In my case, the focused phrase is the μP which also acts as a substitution source. Note that I am not assuming all EXH-based phenomena inherently involve focus or are focus-based. I am, however, assuming that both SIs, NPIs, FCIs, and additivity (i.e., bare μP meanings) are interpreted using the generally identical alternativepruning and -enriching mechanisms relying (inter alia) on the grammatical presence of the EXH-operator. See Fox and Katzir (2011); Xiang (2017), or those they cite, for a discussion and argumentation. The μ is defined in (232) sloppily by ascribing it with an alternative-triggering function with respect to its host (restrictor). Technically, and more precisely, the μ first combines with its host and then the ‘rest’ of the sentence (nuclear scope).6 The third line A(S) should be understood as the set of propositional (sentential) alternatives derived by PFA of the μ-host (restrictor) to the nucleus, resulting in the S set of alternatives. (232)

Lexical entry for or rule for composition with μ0 : [S . . . [μP μ0 XP ]] = XPA   = A XP (by (231) and (222b)) = A(S)



S | S is the result of replacing = μ-hosts in S with its alternatives



(EXH must kick in)  EXH(S )(S)

   = S ∧ ∀S ∈ A S [S  S ] → ¬S

a. Exhaustification operator (EXH), alternate version based on Fox (2007) and Alonso-Ovalle (2006) (where p = S and A(p) = S above) EXH

   A(p) p = p ∧ ∀q ∈ ♥A(p)[¬q]

b. Innocent exclusion (♥), based on Fox (2007) and Alonso-Ovalle (2006) 6 This

is relevant for the discussion of the ‘Gothic paradox’, where I posit a compositional change with respect to the saturation ordering of the two arguments of μ.

126

3 Interpretation

 C ⊆ C : C is a max. consistent subset of C, s.t.  ♥(p, C) = {¬q : q ∈ C} ∪ {p} is consistent

Targeted exhaustification, restricted to δA or σ A, would take the same form as (232a) with A replaced with the relevant alternative restrictor set (same is true for (232b)). The semantic sketch of the meaning of μ in (230) or (232) may be stated more explicitly using the conceptual and notational system in Gajewski (2013). While his system is designed to derive the distribution of exceptives, it provides an amenable route to making (230) more precise and explicit, especially when I turn to the recursive application of EXH. I therefore suppose, following Gajewski (2013), that when a μ particle introduces alternatives, it outputs an ordered pair containing the prejacent and the corresponding set of alternatives. The μ meanings will be derived using a syntactically hardwired obligatory presence of EXH, encoded as [uEXH] on μ. (233)

Lexical entry for μ0  (recycled and formal)

μP = μ μ0

= μ [uEXH]

XP

XP [u EXH]

XP , ( XP ) XP , ( XP )

EXH

EXH EXH

EXH

XP , ( XP ) XP , ( XP

The recycled entry in (233) thus makes more explicit the input meaning-pair to the second instance of EXH, which I will invoke to analyse non-NPI meanings of μ-expressions. Adopting Gajewski’s (2013) format also makes more direct the connection between the alternative-triggering expression—our μ—and the exhaustifier. I return to the technical details and implementation in the subsections to follow. The inferential procedure stated in the last line of (233) can be understood as an economy consideration or a lexical entry. Under an economy view, the second EXH applies if the first-order EXH leads to a contradiction or is, generally, vacuous. If we opt for the definition of EXH which has a hardwired innocentexclusion (♥) proviso, contradiction can generally not be generated after the first layer of exhaustification (but see Gajewski 2009). For an economy principle along these lines to be operative, we need a version of EXH which sometimes generates contradictions.7 The other option, which does not preclude the possible 7 For

a discussion of this, see Gajewski (2013, 203).

3.3 μ

127

first assumption, however, is to hardwire the recursive exhaustification as a separate lexical entry for EXHR , following Gajewski (2013, 205n65). (234)

For a proposition p, set of propositions Π and a set of proposition-set of proposition pairs Σ: EXH

R



 p, Π , Σ = EXHEXH(p, Π ), {EXH(ς ) : ς ∈ Σ}

The μ particles come about in at least four incarnations or profiles: as builders of PSIs/NPIs,8 FCIs, distributive universal quantifiers, and additives. In the following subsections, I address each in succession.

3.3.1 The Polarity-Sensitive Profile In their polarity-sensitive incarnation, μ-markers build Polarity Sensitive Items (PSIs) out of wh-words. PSIs are not only licensed in the scope of negations (NPIs) but are in fact felicitous under any DE operator, as Ladusaw (1979) (building on Fauconnier 1975) discovered. Von Fintel (1999) formalises the DE nature of contexts that license NPIs/PSIs: (235)

(Von Fintel 1999, 100) a. A NPI is only grammatical if it is in the scope of an α such that α is DE. b. A function F of type ς, τ is DE iff for all x, y of type ς such that x ⇒ y : F (x) ⇒ F (y).

The DE pattern stated in (235) arguably does not hold in interrogative contexts which license PSIs but do not appear to be DE. Our analysis of μ-marked PSIs rests on Chierchia’s (2013b) system, which treats NPIs, or PSIs generally,9 as lexical items with obligatorily active alternatives. In 8I

will generally be concerned with polarity-sensitive indefinites in downward-entailing (monotonic) contexts. Since negation is most salient in such downward-monotone context, the term NPI has been the most common theoretical designation for such polar items. Nonetheless, NPIs are also licensed in (presumably) non-negative and non-downward entailing contexts, such as in interrogatives (but see fn. 9), hence I terminologically also follow Chierchia (2013b) in referring to such items more broadly and correctly as Polarity-Sensitive Items (PSIs). Note, however, that PSIs (in my treatment) do not include Positive Polarity Items (PPIs), but see Nicolae (2017a,b) on how PPIs could be integrated. I will generally steer clear from PPIs, but see Szabolcsi (2004), Nicolae (2017a,b), and those they cite for details. 9 NPIs are also licensed in questions, which are not DE contexts. (1)

a. b. c.

Albrecht didn’t see anyone. ∗ Albrecht saw anyone. Did Albrecht see anyone.

128

3 Interpretation

(most of the) old IE languages, NPIs had a transparently bipartite morphology, as we have seen: a wh-term and a μ superparticle. My claim is that it is the μ operator that is responsible for obligatory activation of alternatives of its indeterminate whhost. Note that an opposite view is generally taken in the literature: special types of whterms, termed indeterminates (following Kuroda 1965), are themselves alternativeintroducing elements. This has been claimed by Ramchand (1997) for Bengali and several colleagues for Japanese (e.g., Hagstrom 1998; Shimoyama 2001; Kratzer and Shimoyama 2002; Shimoyama 2006). Consider the two examples in (237) from Shimoyama (2006) (employing our superparticle notations) which find a natural explanation for the (question/quantifier) variability in the idea that indeterminates’ alternatives grow until bound by the suitable quantificational binders, i.e. μ (mo) or κ (ka). (236)

[Dono gakusei mo] odot-ta. which/INDET student μ dance-d ‘Every student danced.’

(237)

[Dare-ga odorimasu ka? κ who-NOM dance ‘Who dances?’

Hoping to build on such views, let me start by departing from some of the assumptions inherent to this system. Firstly, the indeterminate wh-terms10 are assumed to perform the alternative triggering and alternative-growth. The role of the superparticle is solely to act as a suitable binder or ‘quantificational closer’, where κ presumably performs existential and μ universal quantification or closure. My theoretical claim is not as grand with regard to the wh-meanings. If all whterms denote, more or less, the same things cross-linguistically, then indeterminates in languages without superparticle meanings can only be understood as lacking the relevant superparticle as the quantificational binder, while the wh-component with universally grown alternatives is a stable dimension of meaning across languages. Additionally, what about languages where bare wh-terms automatically, and without the need of superparticles or other (overt) quantificational binders, denote existentials? Those languages are generally in the majority typologically. Consider some basic data from Slovenian, with an ambiguous meaning of the indeterminate. (238)

Kaj vzamem what take.IND.1.SG

Questions, however, can be analysed as PSI licensing by virtue of their requiring a strongly exhaustive answer. See Nicolae (2013, 2015) for an EXH-based analysis compatible with views advocated here. For other approaches compatible with the analysis presented here, see also Guerzoni and Sharvit (2014) who contend that interrogatives are DE. 10 I use the terms wh-term/pronoun/phrase and indeterminate indifferently.

3.3 μ

129

a. ‘I (generically) take something.’ b. ‘What do I take?’ There are ways of assuming covert binders and type-shifters which could covertly do the job of converting the wh-term in (238) into an existential quantifier (238a). While I do not oppose such methodological issues in the typological and crosslinguistic treatment of indeterminates and their various incarnations, I pursue a different approach at least as regards the superparticle systems. The distribution of labour in the quantificational expression featuring indeterminates and superparticles is shifted onto the superparticle, which triggers the indeterminates’ alternatives and their growth until they are suitably bound or acted upon. This section applies this view to the μ-marked PSIs. In other sections, I retain these assumptions. In line with Karttunen (1977), and many others, we will treat wh-terms as existentials. The presence of the μ operator activates the alternatives of the existential (wh) host. We take the relevant syntactic feature on—and semantic ‘dimension’ of—the activated alternatives to be [iδ], i.e. the restriction of EXH is that of sub-domain alternatives. Any, in English, looks an existential, having developed from an Old English numeral for one. Chierchia (2013b), for various additional reasons, treats any as an existential quantifier, like some or a(n). What distinguishes any from some, however, is the polarity sensitivity (PS) of the former but not of the latter. In this regard, Krifka (1995) and Chierchia (2004, 2013b) take the PS existentials like any to be characterised by obligatorily active alternatives which need to be cached into the meaning (e.g., by EXH) and cannot be ignored. Alternatives will not be ignored in our calculations; however, this will still not lead to an informationally stronger meaning which is a desirable result insofar as NPIs are taken to be indefinites and the analysis reflects this. Take an example like the one in (239), where in the scope of negation, all of the alternatives to p, ‘John didn’t see anyone’, become entailed by the assertion. The resulting exhaustification therefore returns the original proposition, i.e. there was no-one (at all in the domain and/or its subdomains) that was seen by John.11    (239) EXH[δA] John didn’t see anyone[δ] a. Assertion: (= p)  ¬∃x ∈ D PERSON(x) ∧ SEE(j, x) b. Alternatives:     A(p) = ¬∃x ∈ D PERSON(x) ∧ SEE(j, x) | D ⊂ D

11 In

LF notations, I abbreviate the feature-checked domain restriction of the alternative dimension (σ vs. δ) on EXH as EXH[σ/δA] , which is a shorthand for EXH[uA:σ/δ] . Conversely, the μ-hosts carrying interpretable counterparts of the features on EXH are notated as [σ/δ], standing for [iσ/iδ].

130

3 Interpretation

c. Result: EXH [δA] (p)

=p

(∵ ∀q ∈ A(p)[p  q])

The non-entailment clause on alternative denial is therefore crucial for an EXHbased analysis of NPIs since under negation all alternatives are entailed, as per the definition of EXH in (221). This gives the effect of returning the negative proposition (239c). The alternatives, therefore, play a role in the exhaustification procedure (as they should). Given the nature of subdomain alternatives, such exhaustification leads to no informational enhancement of meaning. Let us see how this applies to our IE data. Given in (240) and (241) are two instances of μ-marked NPIs in Classical Sanskrit and Hittite, respectively. (240)

na yasya [ka´s ca] tititarti m¯ay¯a? NEG whom. REL . GEN [who. M . SG μ] able to overcome illusions. PL ‘No one [=not anyone] can overcome that (=the Supreme Personality of Godhead’s) illusory energy.’ (BP. 8.5.30)

(241)

nu-wa ÚL [kuit ki] sakti and-QUOT NEG [who PSM] know.2.SG.PRES ‘You don’t know anything’

(KUB XXIV.8.I.36)

The NPIs seem to behave in the same way as they do in the English example (239), modulo one difference. I will treat the quantificational contribution of μ-marked NPIs as universal, which return the same result. A Hittite example from (241) is paraphrased in (244). Assume a toy domain (of propositional alternatives) containing three individual-level things one may know: ‘Archimedes’ proof of Pythagoras’ Theorem’ (let knowing this be a), ‘Boole’s expansion theorem’ (knowing this is b), and ‘the name of Chrysippus’ dog’ (knowing this is c). (242)

You (don’t) know something = ∃x ∈ De [(¬)KNOW(x)] = {a, b, c}

Using a PSI instead of a plain indefinite makes the negation somehow stronger: what one doesn’t know is not just something but anything. Chierchia (2013b, 27) informally describes this difference as being one of emphasis or of tolerance of exceptions. An expression with an NPI does not tolerate exception: in case of the ‘not knowing’ example from Hittite, nothing is tolerated as being known without communicating any real truth-conditional difference, as (243b) tries to show in reference to (242). The PS character of such items requires there to be a DE marker like the negation which is not grammatically obligated with plain indefinites (242). (243)

You ∗(don’t) know anything   a. LF: EXH[δA] [ You don’t know anything[iδ] ] b. = ¬∃x ∈ De [KNOW(y)(x)] {¬a, ¬b, ¬c} ¬a ∨ ¬b ∨ ¬c

3.3 μ

131

c. = ¬∃x ∈ D e [KNOW(y)(x)] where D e ⊆ De {¬a, ¬b, ¬c} ¬a ∨ ¬b ∨ ¬c In the μ-marked wh-phrases, I take the relevant feature bundle [δ, σ ] to be present on the μ head. Everything else can be transplanted from Chierchia’s (2013b) system. (244)

You don’t know what-μ =    a. LF: EXH[δA] You don’t know [what-μ[δ] ] b. ASSERTION  : (= p)  ¬∃x ∈ D THING(x) ∧ ¬KNOW(YOU, x)     c. A(p) = ∃x ∈ D THING(x) ∧ ¬KNOW(YOU, x) | D ⊂ D d. EXH[δA] (p) = p

The force of a wh-host of μ remains existential after the exhaustification. In case of recursive application of EXH, as I discuss in the following subsections, universal force arises. We will develop these reasons for treating μ as a universal quantificational marker. The interpretation of PSIs in Chierchia’s (2013b) system fall out naturally and consistently as existentials. Hittite data can be taken to support this view given the evidence from morphology: -ki forms NPIs, not FCIs or universals. This can also be seen in English since any is historically developed from an existential (one). There is overwhelming morphological evidence that other μ-expressions are built on wh-terms using the conjunctive/universal μ and not the existential κ particle. Note that the opposite is predicted on an indefinite analysis of PSIs which takes PSIs to be negated existentials with a ‘widened domain’ in the sense developed in Chierchia (2004, 2006, 2013b). The μ superparticle is, on the other hand, an essentially universal quantificational logical builder. I will develop a diachronic analysis of how universals become existential PSIs in the history of IE. One result of this development is the status of μ markers in the Satemic group of IE languages, such as the Slavonic or the (old) Indo-Iranian branch (but more on this in the next chapter). I do not delve further into the intricacies of how negative words, PSI and μmarking interact here, but see Gaji´c (2015, 2016) whose analysis is generally compatible with the one I developed. I will return to discussing NPIs once more in the next chapter when making a connection between the so-called Strong NPIs and scalar additives. Strong NPIs require, in Chierchia’s (2013b) system, a mode of exhaustification different to EXH, namely EVN, which strictly operates on the scalar dimension (such as likelihood). The requirement, encoded as a presupposition, of anti-exhaustivity on emphatic NPIs, however, suggests that recursive EXH and EVN may be more connected than it is suggested in Chierchia (2013b). The natural switch between EXH- and EVNbased NPIs brings into question the mechanical unity of the theory we are pursuing

132

3 Interpretation

and applying. In Sect. 3.3.4, I will motivate a transition in exhaustification between weak and strong NPIs. Before I discuss how this connection is borne out, both synchronically and diachronically, let me continue overviewing μ-meanings, addressing how μ builds FCIs in the next section. Recursive EXH is, as it turns out, required to derive the FC effect. Recursive EXH can also be shown to build the EVN-operator which derives NPIs, unifying the explananda technically and, I would like to suggest, conceptually, too.

3.3.2 The Free-Choice Implicating Profile The freedom of choice (FC) implicature, communicated using disjunction or Free-Choice Items (FCIs), is a linguistically captivating and cross-linguistically prominent expression which is not validated in classical logic. I presented several μ-marked FCIs in the previous chapter. Let me briefly show how FC disjunction is derived before turning to wh-μ FCIs which inherit the treatment. The implicature of a FC disjunction, for instance, is the following: expressing permission with disjunction implicates the conjunction of the permitted options: (245)

You may have ice cream or cake. (Let p correspond to ‘You may have ice cream’ and q to ‘You may have cake’.

a. Logical meaning: (p ∨ q) b. FC implicature: p ∧ q ‘You may have ice cream and you may have cake (and you may not have both) What the modal element is presumably able to do is relax the sub-domain effect of the denial that the scalar implicature generates. Here is how the system derives FC disjunction. Assume that for (245) two dimensions of alternatives are relevant: the σ -dimension with the scalar alternative to the disjunctive prejacent, namely conjunction (246a). In line with Sauerland (2004), disjunction is assumed to have a defined set of δ-alternatives, constituted by the two disjuncts. (246)

A(245) relevant to EXH:   a. σ A = [p ∧ q]   b. δA = [p], [q]

Exhaustifying the prejacent against the entire alternative set, comprising δ- and σ -sets along the two alternative dimensions, yields a contradiction.

3.3 μ

(247)

133

a. EXH [σ ] EXH [δ ]

p 

q

  b. [EXH[A]  [p ∨[A] q] = EXH[σ A(246a)] EXH[δA(246b)]  [p ∨[σ,δ] q] = [p ∨ q] ∧ ¬  p ∧ ¬  q ∧ ¬  [p ∧ q] =⊥ 

To derive the effect correctly, Chierchia (2013b) and Fox (2007), among others, assume that the alternatives are not quite those as set in (246). Instead, they are exhaustified—exhaustifying (over) such a set yields doubly exhaustified alternatives (via EXHR ). Fox (2007) shows that a double application of EXH ranging over the subdomain δ-alternatives yields an anti-exhaustive effect.12 Chierchia (2013b) uses the exact same mechanism but labels the exhaustified alternatives which undergo another round of exhaustification, instead, pre-exhaustified. Conceptually, the effect is the same. Chierchia (2013b) thus assumes that the alternatives relevant for the calculation of FC disjunction are the following, which revises (246).13 (248)

A(245) relevant to the second instance of EXH:   a. σ A = EXH([p ∧ q])   b. δA = EXH([p]), EXH([q])

The calculation now proceeds to return the desired result with the scalar effect, i.e. the denial of conjunction as the scalar alternative to the disjunctive prejacent. (249)

a. EXH R [σ ] EXH R [δ ]

p q

12 Why

is recursive exhaustification not relevant for the conjunctive σ -alternative for a disjunctive prejacent? Simply because EXH(p ∧ q) is vacuous since there is nothing meaningful to deny (EXH(p ∧ q) = p ∧ q). 13 These can be considered pre-exhaustified or the alternative set as constituted as the result of the first-round exhaustification.

134

3 Interpretation

 b.

EXH R [A]



 [p ∨ q]

   R =EXHR [σ A(248a)] EXH [δA(248b)]  [p ∨[σ,δ] q]

= [p ∨ q] ∧ ¬EXH(p) ∧ ¬EXH(q) ∧ ¬EXH([p ∧ q])          prejacent δ (i) σ (ii) =  [p ∨ q] ∧ p ∧ q ∧ ¬  [p ∧ q] i. ¬EXH(p) ∧ ¬EXH(q) = ¬[p ∧ ¬  q] ∧ ¬[q ∧ ¬  p] = [p → q] ∧ [q → p] = [p ↔ q] = p ∧ q ii. ¬EXH([p ∧ q]) = ¬  [p ∧ q] Therefore, while each is a permissible option, you cannot have both ice cream and cake. If the reading we are after does not exclude the two options, then iterative EXH with a δ-restriction ignores the scalar dimension and yields, along the same lines, the inclusive reading. While the switch between FC disjunction and FCIs seems intuitively substantial, or at least non-trivial, in terms of meanings it is not since disjunctions are on a par with existentials, which is what I take indeterminate wh-words to mean, i.e., they correspond to (discrete) disjunctions. Furthermore, let me show what the compositional role of μ is in FC expressions. In the context of our icecream-cake example, notice how FC disjunction and an FCI like whichever are truth-conditionally identical. (250)

a. You may have ice cream or cake. b. You may have whichever (desert).

Just like FC disjunctions, FCIs carry with them a scalar and a FC implicature— the two contradict each other. Thus, the interpolation of a modal element solves this contradiction by allowing a weakening of the scalar component. Without the modal, or additional subtrigging structure, FCIs are barred. The presence of an existential modal or additional subtrigging sanctions the Fluctuation of the context, which I address below. (251)

You ∗(may) have whichever (desert).

Via interpolation, Chierchia predicts existential and otherwise universal FC readings.14 (252)

14 For

a. EXH . . .  . . . FCI[σ,δ] . . . b. EXH . . . FCI[σ,δ] . . .  . . .

∃-FCI ∀-FCI

an extensive logical discussion of FC and FC-like phenomena, see Humbertstone (2011, 793–808).

3.3 μ

135

The claim I make, specifically for old IE languages, is that μ-FCIs are consistently universal. Syntactic evidence for universal quantification in FC μPs in, say Old Irish, comes from the linear ordering of the FCI and the modal. As shown in (253a) and (253b), μ-marked FCIs are consistently of the universal type. (253)

FCIμ >  a. Old Irish: i. [ce cha] taibre [what μ] give.2.SUBJC ‘what[so]ever thou mayst give.’ ii. [ce cha] orr [what μ] slay.3.M.SUBJC ‘whichever he may slay.’

(Zu ir. Hss. 1.20.15)

(Anecd. II.63.14.H)

b. Classical Sanskrit: yady abhyupetam [kva ca] s¯adhu as¯adhu v¯a . if promised to be accepted what μ honest dishonest or kr.tam may¯a . done.PST.PART 1.SG.INSTR ‘If you accept whatever I may do, whether honest or dishonest.’ (BP. 8.9.12) Universal FCIs require wide-scope (more on that below) and are derived through iterative sub-domain exhaustification (just like distributive universal quantifiers, as discussed in the next subsection). They are additionally constrained by Fluctuation (254): this constraint guarantees that the intersection of the set of individuals in the restriction (P ) and the scope/nucleus (Q) of a proposition is non-identical across the worlds (modal base), i.e. that it fluctuates. (I take Fluctuation to be a privative character of (universal) FCIs.) (254)

The Fluctuation Constraint (Dayal 2009, 241,   5b) ¬∃X∀w [wa ≤ w ]λx P (w )(x) ∧ Q(w )(x) = X

An episodic proposition such as Any student smokes has no worlds relevant for the evaluation other than the actual world and Fluctuation thus explains its infelicitous character. A subtrigged version of the same proposition, such as Any student who’s not a vegan smokes, now contains two distinct worlds which salvages the sentence and the FC inference is sanctioned.15 Another argument for the universal character of FCIs in IE comes from the configurations that guarantee (252b). This is related to the subtrigging contexts 15 This

characterisation of mine leaves out a lot of formal details which make up a stronger argument for Fluctuation or Modal Containment, Chierchia’s (2013b) principles which derives the same effect as Fluctuation. (Dayal 2013 also refers to Fluctuation as Modal Distributivity under the heading Viability Constraint.) See Dayal (2009, 2013) and Chierchia (2013b) for details.

136

3 Interpretation

and the idea that subtrigging involved clausal adjunction with the wh-μP in its edge. Note that the two Old Irish examples in (253a) are subtrigging fragments, i.e. clausal adjuncts to the matrix clause which hosts the relevant EXH. In Sects. 4.1–4.2 of Chap. 4, I posit a diachronic reanalysis of the wh-μP in the clausal edge (with additional considerations from Modern Greek). In a biclausal structure, the placement of an FCI in the edge of the embedded clause containing the relevant modal source (an actual modal, subjunctive, or even covert assertoric modal) ensures universal FC readings. The following example from the R.gveda demonstrates this. (255)

idám a¯ pah. prá [kím ca] duritám máyi vahata | yát this water.NOM PV.away carry COREL what μ trouble me.LOC ‘O Waters, carry this away, whatever trouble is in me—’ (R.V, 1.23.22ab ; trans. Jamison and Brereton 2014, 118)

The proposition in (255), stripped of the relative-correlative structure, is essentially (were it that) ‘water may carry away whatever trouble is in me’. Assuming, again, a troubled domain as a toy of three alternatives for this example, then ‘waters may carry away trouble a that is in me’, ‘waters may carry away trouble b . . . ’, ‘waters may carry away trouble c that is in me’ are all contextually desirable options in this invocation. Recursively exhaustifying the domain {a, b, c} yields these options as the reading suggests. The worlds of evaluation are thus not only the troubles in the actual world, but also the troubles in me, which also satisfies Fluctuation. Note that the denial of the scalar alternative is relaxed in this case, which bring me to the last wh-based meaning that μ may impose: the distributive universal quantificational.

3.3.3 The Universal Quantificational Profile I will contend that universals derive in the same fashion as FCIs do, modulo the Fluctuation constraint which the former do the latter do not have to meet. Universals are built on indeterminates which we take to be existentials. One way, therefore, to look at μ+wh universal quantifiers is to see μ as combining with ‘∃’ and yielding a ‘∀’. (256)

μ

wh

The system, as it stands, can do the above by recursively exhaustifying the δdomain of the wh-host. Let me first give some brief data from Old Irish (257), Gothic (258) and Hittite (259) before demonstrating how our technical apparatus derives the desired meanings. (Although this has partly already been demonstrated for μ-FCIs in the previous subsection.)

3.3 μ

(257)

137

a. na

ei-plet

hua-n

ADV PRT -dies. IMPV.3. PL PREP. DEF - PRO . DAT. SG

bás death.DAT.SG

coitchen hua-n-epil common.DAT.SG PREP.DEF-PRO.DAT.SG-PRT-dies.PRES.IND.3.SG [cá -ch] acht foir-cniter hua-sain [wh μ] but PRT-finish.PASS.IMPV.3.PL PREP.DAT-distinct sech [cá -ch] bás death.DAT.SG PREP [wh μ] ‘Let them not die by the common death whereby everyone dies, but let them be ended by a special death different to all.’ (MCA, 73d.7) b. hi[cá -ch] -du in.DAT [wh μ] place.DAT.SG.F ‘in every place’ (MCA, 24c.9) (258)

uh] saei hausei waurda meina jah [hvaz and who.M.SG μ pro.M.SG hear.3.SG.IND words.ACC.PL mine ‘And every one that heareth these sayings of mine’ (CA., Mat. 7:26)

(259)

nu kuitt-a arhayan kinaizz[i J what-μ separately sifts ‘[She] sifts everything seperately.’

(KUB XXIV.11.III.18)

The morphological formula for universal quantifiers is clear across IE, as is well attested beyond IE. Consider the Hittite example above, the one in (259). Without the μ particle -a, the object would be a wh-DP with indefinite or existential meaning. Since “indefinites of all colouring receive meanings identical to those of or . . . as they are just potentially infinite disjunctions,” (Chierchia 2013b, 357) (259) sans the meaning contributed by μ would be something along the lines of (260), where ‘sand’, ‘flour’, and ‘ash’ in (260iii) are some possible contextual extensions of some things that Hittites may have sifted. We adopt a shorthand for these discrete disjunctions at propositional level in (260iv), where ‘she sifts sand’ is abridged as a, ‘. . . flour’ as b and ‘. . . ash’ as c. The fronted quantificational object μP is indicative of the LF. For ease of exposition, I work with a two-membered domain in (261). (260)

(261)

(259) without the contribution of μ =   i. LF: wh1 . . . (she) sifts t1  = ∃x[SIFTSw (s)(x)] ii. she sifts something iii. she sifts the sand ∨ she sifts the flour ∨ she sifts the ash iv. a ∨ b ∨ c  (259) =  EXHR [iμ,ıδ] ]1 . . . she sifts t1  = [uμ,uA:δ] . . . [what-μ

R EXH [δA] [(259)\μ], {a, b} = EXH R [a ∨ b] = [a ∨ b] ∧ ¬ EXH (a) ∧ ¬ EXH (b) = [a ∨ b] ∧ ¬[a ∧ ¬b] ∧ ¬[b ∧ ¬a]

138

3 Interpretation

= [a ∨ b] ∧ [¬a ∨ b] ∧ [¬b ∨ a] = [a ∨ b] ∧ [a → b] ∧ [b → a] = [a ∨ b] ∧ [a ↔ b] = [a ∨ b] ∧ [a ∧ b] = [a ∧ b] The presence of the μ particle thus entails two procedures: (i) the activation of δ-alternatives and (ii) their recursive exhaustification. Therefore, FCIs share with universal quantifiers the iterative mode of being derived, to the exclusion of NPIs, which also do not informationally strengthen meaning, unlike universal quantifiers or FCIs built on indeterminates. There is further empirical support for the disjunction to conjunction shifts of the kind formally explored above. Bowler (2014) reports evidence of Warlpiri lacking a morphosyntactic distinction of conjunction and disjunction. Rather, Warlpiri employs a single functional word, manu, which shifts its meaning from disjunctive to conjunction depending on some particular (local) contextual constraints. The core proposal of Bowler (2014) is that the Warlpiri coordinator manu has a disjunctive denotation which is pragmatically strengthened to conjunction, where the pragmatic strengthening procedure is carried out by Fox’s (2007) EXH operator. (262a) and (262b) show the main facts. (I return to the details of the mechanics of iterative EXH in the next section.) (262)

Warlpiri (Bowler 2014): a. Cecilia manu Gloria=pala yanu tawunu-kurra. C manu G-3DU.SUBJ go.PST town-to ‘Cecilia and Gloria went to town.’ b. Cecilia manu Gloria kula=pala yanu Lajamanu-kurra. C manu Gloria NEG-3DU.SUBJ go.PST Lajamanu-to ‘Neither Cecelia nor Gloria have been to Lajamanu.’

Bowler (2014) accounts for the above facts by specifying the lexical meaning of manu as disjunctive and thus analyses the conjunctive reading as a derived one via recursive exhaustification. She additionally assumes that the disjunctive alternative set in Warlpiri does not contain the conjunctive scalar alternative, i.e. [p ∧ q] ∈ {p ∨manu q}A , making it therefore different to the alternative set for disjunction English. Since her recursive disjunction relies solely on subdomain alternative exhaustification—subtitling her system with Chierchia’s (2013b)—the absence of the conjunctive scalar alternative to the inherently disjunctive marker manu in Warlpiri makes no computational difference. It is sufficient to note that her computation independently makes use of the recursive exhaustification of the subdomain alternatives to disjunction, yielding the same effect that we derived in (261). While we have done so for discrete generalised conjunction and disjunction, Bowler (2014) derives it for actual and explicit conjunction and disjunction, which gives my analysis additional if not grounding support.

3.3 μ

139

3.3.4 The Additive Profile Additivity derives in the system in the same way FCIs and universal quantifiers do: through recursive exhaustification. Additivity is in stark contrast to the previous profile of μ in one crucial respect which may even be stated without reference to semantics: μ associates with ‘determinates’, not wh-indeterminates. This, already, gives the system an appealing flare: alternative activation and exhaustification are common across the various derivations of meaning, what is parametrically elastic is the nature of μ-association itself. There are two sub-types of μ marking additives which are detailed in this section: the non-scalar additive markers like ‘also’ and the scalar ones like ‘even’. Each of them is analysed in turn with an additional aim: to unify the mechanisms that derive them in a way which allows for the non/scalar elasticity in a principled way. I will contend that this is an improvement of the system I am adapting from Chierchia (2013b) and goes back to the discussion on EXH vs. EVN based exhaustification I started in Sect. 3.3.1. The empirical motivation for this line of pursuit is given in (263). An explicans for (263) constitutes one of the aims of this subsection. (263)

3.3.4.1

The scalarity parameter in pronouncing additives: Languages which mark scalar and non-scalar additivity with a μ morpheme generally employ a greater number of morphemes to express scalar additivity than they do to express non-scalar additivity.

The Non-scalar Profile

Chierchia (2013b, 424) assumes that the exhaustification procedure has nothing to do with deriving the additive meaning and interpretation of ‘too’. The meaning of ‘too’ is, in his words, better captured by treating it as a partial function, that is, a meaning not derived by EXH. Since the μ meanings thus far have been derived using unitary tools, μ-marked additivity seems like it should fall within the semantic domain of those, and hopefully only those, tools. I am not proposing to treat the English additive particle differently, necessarily, but that the additive meanings which are consistently μ-marked across languages should derive their interpretation along the same conceptual lines as previous μmarked meanings. Additives featuring the μ superparticle should, then, have a strong inherent link with exhaustification, indeed. Let me demonstrate how an EXHbased additive meaning of μ obtains. Pre-theoretically, consider the following μ-additive from Japanese, which derives from the negation of the exhaustified reading of the prejacent. (264)

[μP Mary mo ] genki desu. Mary μ well is ‘[Mary, too] is happy.’  ‘Mary is happy and not only Mary is Happy.’

140

3 Interpretation

How are additive meanings technically different from other μ-meanings? I now turn to explicating the commonalities and differences. The main difference can be stated in purely syntactic terms: μ-encoded meanings of universals, FCIs, and NPIs arise when the μ-associate is a wh-phrase. This same distribution can also be captured semantically by caching in on the existential quantifier meaning of wh-phrases, which is what I assume (see Xiang 2016 and those she cites for evidence and further discussion). A proper noun like ‘John’ has no hardwired existential meaning: therefore, when μ associates with ‘John’ it will trigger its alternatives and recursive exhaustification takes place to yield an anti-exhaustive, or additive, meaning. Additive expressions of wh-phrases are therefore predicted to be impossible μ-meanings. (265)

∗ Znam i ko i kako je napravio zada´cu know.1.SG μ who μ how AUX.3.SG made homework ‘I know both who and how one did the homework.’ (Ser-Bo-Croatian)

The differentiating property of additives is therefore the restriction to nonwh-associates, which can be modelled semantically as a possible restriction on type-association (e-type versus non-e-conjoinable/t-conjoinable types), or a restriction on association with existential quantifiers (assuming wh-phrases are, indeed, existential quantifiers). I adopt a lighter take here, assuming that the differentiating property is that of ‘indeterminacy’ (in the sense of Shimoyama 2006): while whphrases are indeterminate, proper DPs, such as ‘John’, are not. How does anti-exhaustivity derive? From the system that is already in place, it follows straightforwardly from the iterative application of EXH. The determinate μ-associate has to be recursively exhaustified, since a single layer of exhaustification yields a contradiction in absence of a negative or a modal operator interpolating within the structure (this follows from the basic set-up we have in place). One way of driving this is to assume that the EXH applies once more after the first cycle, assuming that a μ-associate like ‘John’ has a blocked access to the contextually determined alternatives to ‘John’. (This reasoning requires a stipulation of this type, but let me sketch it nonetheless.) A single level of exhaustification yields a contradiction in absence of alternatives, since the proposition in question, call it φ, is the only available alternative to itself. Since φ entails itself, EXH cannot negate anything and is therefore vacuous. This would be one reason why the speakers are assumed to rerun the Gricean reasoning and add another layer of exhaustification. At the second level the result of the first level of exhaustification now contains the EXH (φ) (the failed attempt) as an alternative.16 Once this alternative is denied, under standard assumptions, anti-exhaustivity obtains. For ample support in favour of a

16 Unlike

Fox (2007), for whom EXH is constant, in this regard, I follow Mitrovi´c and Sauerland (2014, 43), Gajewski (2013), and those they cite, in assuming that the alternative set in the second step of exhaustification is formed by EXH associating with the result of the first step.

3.3 μ

141

recursive technology of exhaustification, see Fox (2007); Katzir (2007); Gajewski (2013); Mitrovi´c and Sauerland (2014); Bowler (2014), among others. Similarly, Xiang (2016) derives the same anti-exhaustivity effects by positing EXH operates on pre-exhaustified alternatives, just as Chierchia (2013b) does. Such approaches are completely on a par with mine and those cited in whose footsteps I follow. Mitrovi´c and Sauerland (2016) derive the anti-exhaustivity effect by following Bade’s (2015) view that sentences are interpreted obligatorily exhaustified. Such a view constitutes EXH as operating, generally, on pre-exhaustified alternatives, in the sense of Chierchia (2013b). Again, this would yield additivity in a similar fashion as the desired inference by avoiding some technical difficulties. Whatever the aetiology for iteration of EXH is, we have strong reason to be able to consider μmarked additivity as stemming from the same microsemantic components inherent to μ superparticles.

3.3.4.2

The Scalar Profile

I have been relying on Chierchia’s (2013b) technical and conceptual apparatus for deriving the semantic/pragmatic inferences (such as SIs and PSIs) narrow syntactically. This is done by supposing there is a silent counterpart to the focussensitive operator ‘only’, i.e. EXH, which negates as many alternatives as it can and, if doubled, can return the assertion conjoined with other alternatives. In this last step, we have extended the range of where and how EXH may apply. In this subsection, I adapt, tweak and push the system further—this time, in order to improve on it conceptually via an optimisation of the inferential procedures assumed to be at play.

One Exhaustifier Is Better Than Two: Optimising Optimal Fit Consider Chierchia’s (2013b) Optimal Fit principle (267), which governs the choice between EXH and EVN following an economical principle, itself presumably a variant of Maximise Presupposition principle. Recall that unlike EXH, EVN is a different exhaustifier that operates on the scalar dimension and, once it combines with a proposition, returns that proposition as (informally) the least likely proposition among the alternatives. (266)

EVN A (p)

= p ∧ ∀q ∈ A[σ ] [p q]

In Chierchia’s (2013b) system, EXH is the default choice. If applying it leads to triviality and there exists a salient probability measure  , then EVN kicks in. Note that if (267ii) holds, then so does, in essence, (267i), since total ordering of a domain constitutes it as a chain as whenever A ⊆ B, then  (A) ≤  (B) is also true for some  and a partial ordering relation ‘≤’. Chierchia (2013b) dubs this ‘structural salience’.

142

(267)

3 Interpretation

Optimal Fit (Chierchia 2013b, 153n20) In exhaustifying φ, use EXH unless EXH(φ) is trivial (i.e., contradictory or vacuous) and there is a salient probability measure  . A probability measure  is salient iff one of the following holds: i.  is salient in the context ii. A is totally ordered by ‘⊆’

Chierchia (2013b) shows how meanings involving some minimisers are derived using the same logic of exhaustification but, crucially, using a different exhaustification operator. While EXH (which he calls O) is the silent counterpart of ‘only’, there is allegedly another exhaustifier EVN (which Chierchia 2013b dubs E), taken to be a silent version of ‘even’. Here, I try to show that one (EVN/E) can be derived from the other (EXH/O). Why? A system with a single (adequate) exhaustifier is simpler and more parsimonious than the system with two (or more), this much is clear. The second argument for preferring a single exhaustifier is empirical. In this connection, I contend it’s EXH, since it is the default, in line with Chierchia’s (2013b) general system and (267) specifically. In very many languages where μ can derive additive meanings, those meanings can be non-scalar (‘also’-type) or scalar (‘even’-type), depending purely on the context. Whether the context-dependent ‘also’/‘even’ switch is taken to derive from the type of exhaustifier silently present or an economy-driven ‘shape’ of a single exhaustifier, reduces to the same parametric mechanism. What is advantageous, and presumably more parsimonious in the latter case, is the size of the inventory of exhaustifiers (bringing us back to the first argument). The third argument is diachronic: there is a demonstrable historical change from an exclusively scalar and ‘even’-type meaning to a non-scalar ‘also’type meaning. For these three reasons, I will keep reducing EVN into EXH. If a meaning can be dissected into compositional components that in concert return the relevant meaning, then structural dissection can be relegated to diachronic morphosyntax. I believe there are compelling synchronic reasons for assuming that strong NPIs are not always EVN-calculated but rather that they can be. My proposal is that EVN is built from EXH. (I will use lattice-theoretic apparatus, e.g, ≺ is a covering relation). Here are the technicalia briefly for how the reduction of EVN into EXH occurs. The prerequisite is the additive meaning which derives via a recursive application of EXH[δ] , or via a single application of EXH[EXHδ] defined to be restricted to preexhaustified alternatives, i.e. for the readers following Chierchia’s (2013b) system. To arrive at the desidered ‘switch’ between a non-scalar additive like ‘also’ in μ-form and its scalar ‘even’ counterpart μ, the partially ordered δ-alternative set (which I assume is an anti-chain, A) is mapped onto a totally ordered set (i.e., a chain, C). I assume that ordering is achieved through ranking, r, to which I return below. See Stanley (1984) or Wild (2005), among others, for details which are not relevant for the basics of the proposed mechanism.

3.3 μ

(268)

143

¬ EXH[u

:δ ] (p)

¬EXH[u

r

:δ ] (p)

EXH R [ u :δ ]

EXH [u :δ ] (p) EXH [δ ]

p

μ [iδ ] a. ¬EXH[uA:δ] (p) =

EXH R [uA:δ] (p)

 A 

= EXH EXHp, A(p) , pμ , A(pμ ) : μ ∈ Dδe(t)

b. r = ∀x, y ∈ A.∃i ∈ C : C = {0, · · · , n}[xi ≺ yi+1 ∨ yi ≺ xi+1 ] 0 for all minimal x, i.e. iff ¬∃y[y ≺ x] i. r(x) = n for all maximal x, i.e. iff ¬∃x[y ≺ x] ii. r(y) + 1 ⇔ ∃y[y ≺ x] c.  : l2 ◦ l1 , s.t.,  : r −1 (0) defined on a two-element chain C2 = {0, 1}, ≤ i. l1 : A !→ C ii. l2 (p) = ∀q ∈ σ A(p)[p  q].p The meaning encoded by this decomposed scalar additive is the one derived by a standard silent even-based EVN exhaustifier. The lower anti-exhaustive structure ensured the additive meaning which is generally accepted but is in standard accounts not encoded in the EVN operator. This we now encode (268a). In line with Gajewski (2013), Mitrovi´c and Sauerland (2014) and Bowler (2014), and in technical contrast to Fox (2007); Chierchia (2013b), I assume that the alternative set is not held constant, nor is the actual position of the second-level EXH. Nothing much hinges on this, hence the first-level EXH (which iterates) is given a dashed ghost-position. The second part of the meaning, coded by the  -operator, is the expression of the least likelihood. This meaning comes in two steps (268c): the antichain of alternatives A is mapped onto a chain C (268c-i) and the prejacent is given the lowest point on a two-element chain C2 (268c-ii). There is additional evidence that some even-meanings in English and elsewhere cannot be accounted for in terms of ‘unlikelihood’ alone, which is what  does in this system. Rather, as Elliott et al. (2015), Mitrovi´c (2018a), and those they cite show, some even-expressions

144

3 Interpretation

Fig. 3.1 Maps and partial homomorphisms from completely lattice-structured alternatives to a ranked chain and two-element chain

(a)

(b)

(c)

2

make reference to low-rankedness. This can be captured in the proposed system by positing a ranking function r (268b). Formally, let r be an order preserving bijection from the lattice A onto a rank-selected (maximal) chain C of A. The unlikelihoodtype chain is then a map from n-ranks onto a two-element chain (i.e.,  : C !→ C2 ). This is sketched in Fig. 3.1. Such a microsemantic dissection of the exhaustifier assists in providing a unitary meaning of μ across languages and across histories of a language, such as the history of the Japanese μ, as I discuss in Chap. 4. Note that the LF dissection in (268) is not a hardwired lexical entry for the scalar μ particles (a case of Old Japanese), but rather the representation of the components that scalar μ brings into play. The evidence for the proposed dissection of even or even-like operators, which relies on its additive component being derived from recursive application of the EXH -operator, is also supported by empirical motivation from English (Francis 2017, and those cited), Hungarian (Szabolcsi 2017b), and Ser-Bo-Croatian (see below). In English, the additivity of even is more or less standardly assumed and taken for granted (cf. Rullmann 1997, int. al. for an opposing discussion). In fact, obligatory additivity of even derives for free by virtue of even’s bringing into play likelihood comparanda. This amounts to adapting Chierchia’s (2013b) Optimal Fit. Nothing really changes apart from our understanding of how EVN and EXH are structurally connected. (269)

Optimal Fit (revised/optimised) In exhaustifying φ: (a) > (b) > (c) a. Use EXH unless EXH(φ) is trivial (i.e., contradictory or vacuous). b. Otherwise, repeat exhaustifying φ. c. As a last resort, resort to a salient probability measure  . A probability measure  is salient iff any of the following holds: i.  is salient in the context (determined pragmatically or syntactically) ii. A is totally ordered (chained) by ‘⊆’ (use r or  to map A onto a chain)

3.3 μ

145

The link between the primary (a) and the secondary (b) option is in fact independently motivated in Chierchia (2013b, 278n57) by the ‘Blocking principle’, which I give in (270) using my notation. (270)

EXH [δA]

> EXHR [δA]

The third (c) Optimal option relates non-scalar additivity arising as a result of iterative exhaustification with the scalar additive like ‘even’. The argument for the intrinsic relation between the two is empirical and as such gives support to why positing two distinct apparatuses for deriving them is undesirable. Szabolcsi (2017b) introduces novel evidence from Hungarian which is on a par with Ser-Bo-Croatian with regards to the fact that two particles must act in tandem to deliver the even-meaning of low-rankedness (or related concessive meaning, CNCS; cf. Crniˇc 2011) or unlikelihood. These facts are not only consistent with but also directly motivate the scalarity parameter stated in (263). (271)

‘Even Mary’ a. Hungarian i. még Mari is EVEN

ADD

ii. még/akár csak Mari is EVEN 1

ONLY

ADD

b. Ser-Bo-Croatian cˇ ak/makar i Marija EVEN / EVEN CNCS ADD In these NP-associating contexts, the least likelihood meaning is expressed with a single even particle in English, two particles in Ser-Bo-Croatian (271b), and up to three particles in Hungarian (271a-ii). The Ser-Bo-Croatian i and Hungarian is are independently additive. The még and akár are, according to Abrusán (2007) ‘even’type particles (which is what Szabolcsi 2017a also assumes), hence labeled EVEN. Note that the Hungarian csak and Ser-Bo-Croatian cˇ ak are cognate borrowings from Turkish (Skok 1971, 289). The original borrowed form, cak, had (and still has in modern Turkish) the low-likelihood meaning even. It is curious that Hungarian borrowed, or developed, the only-type meaning. In this connection, the only-type particle sam- ‘only, alone’ shows an interesting connection in Ser-Bo-Croatian also, bringing the Hungarian and Slavonic meanings of csak/ˇcak (‘only’/‘even’, resp.) synchronically closer in their distribution (272). Also note, however, that the particles may co-occur (273), which is in line with our proposed dissection: (272)

Skok (1971, 289) a. cˇ ak

do zore to dawn.ACC ‘even till the dawn’ EVEN

same b. do zore to dawn.ACC only/alone.F ‘even till the dawn’

146

(273)

3 Interpretation

cˇ ak

do same zore / zore same to only/alone.F dawn.ACC dawn.ACC only/alone.F ‘even till the dawn’ EVEN

Fruthermore, Ser-Bo-Croatian makar is cognate, and (as far as I am able to establish) synchronically equivalent to the Slovenian concessive scalar particle magar(i) (Crniˇc 2011). Skok (1972, 359) notes that makar is “of unknown origin from the domain of syntax,” as it’s found in Romanian m˘acar, and even modern Greek makári. One analysis that Skok entertains is the coupling of the neo-Persian negative morpheme ma and agar (from Old Persian  hakaram, ‘once’; cf. Turkish me˘ger).17 Without reference to the original direction or form of the borrowing, I conjecture that makar/magar(i) are cognate with Hungarian particle sequence még and akár. Historically, the Turkish contact is relevant and likely. This is parallel to cˇ ak/csak and its potential borrowing from Turkish.

The Non/Scalar Additivity and Strong/Weak NPIs One argument for not assuming EVN is an independent formative comes from the natural switch between two μ expressions: the scalar additive (‘even’), presumably derivable by EVN and its non-scalar counterpart (‘also’), derivable through recursive EXH R . Another argument is the presumed ‘strength’ of the so-called strong NPIs such as the Hindi koii bhii, which is the building formative of ‘strong NPIs’ and which is semantically parallel with minimisers in English (e.g., ‘in weeks’ or ‘at all’). Drawing on Lahiri (1998), Chierchia (2013b, 156ff) develops an EVN-based treatment of strong NPIs along the following lines. Note that bhii looks a lot, in its distribution, like the μ I’ve been developing. (274)

17 As

‘some’-series a. Ram bhii aayaa R also/even came ‘Also/even Ram came’ b. ∗ koii bhii aayaa some also/even came ‘anyone came’ c. koii bhii nahiin. aayaa some also/even not came ‘no-one came’

a reviewer tells me, “in Hindi/Urdu agar is ‘if’, presumably a borrowing from Persian (but potentially mediated by some Turkic language). And Hindi/Urdu magar is ‘but’ (which seems to agree with Turkish me˘ger, and may in fact like agar represent a Persian form mediated through some Turkic language)”.

3.3 μ

(275)

147

‘one’-series a. ek bhii aadmii nahiin. aayaa one also/even man not come ‘no man came’ b. ∗ ek bhii aadmii aayaa one also/even man come ‘any man came’

Chierchia (2013b, 156) treats ek (275) as a cardinality predicate, meaning ‘one’, and whose alternative set is purely scalar. Based on this, bhii is akin to ‘even’, underlying the silent meaning of EVN, and its combination with a purely scalar item is predictable and expected since EVN is defined as a purely scalar exhaustifier. (276)

i. ek = λP .λQ.∃x[ONE(x) ∧ P (x) ∧ Q(x)]   ii. ekσ A = λP .λQ.∃x[n(x) ∧ P (x) ∧ Q(x)] : n ≥ 1 b. bhiiA (p) = EVNA (p) = p ∧ ∀q ∈ A[p  q] a.

Taking bhii as a sentential-scope even-type operator, then the strong NPI in (275a) is derived via the negation of the least likely end-point scale member, ek ‘one’. Chierchia (2013b) provides this analysis. What about the non-scalar ‘some’ NPI in (274)? These, too, are treated as strong NPI with the use of some additional stipulations that serve to obtain the NPI meaning “in terms of a hierarchy of quantificational domains” (Chierchia 2013b, 157). The  -functor is, in his terms, a probability measure for how likely something may be found in a domain. Note that Chierchia’s (2013b) aim is to derive Hindi bhii-marked NPIs as inherently strong in nature. However, Hindi also has weak NPIs which are morphologically indifferent from the allegedly strong ones. Consider the data from Kumar (2006, 159ff), showing that bhii-based NPIs are fine in questions (277). (277)

(Kumar 2006, 159n16) a. us kamre men. koi bhii st.ud.en.t. thaa (kyaa) that room in some also/even student was what ‘Was there even one/any student in that room’ b. aap-ne kisii bhii st.ud.en.t. ko dekh-aa (kyaa) you-ERG some also/even student to see-PERF what ‘Did you see any student?’

If bhii-marked NPIs are not inherently strong and EVN-derived per se, then they must derive in the present system via EXH also. This clearly makes the system less tractable and principled with respect to the empirical facts it aims to explain since the choice between EXH and EVN cannot be lexically determined.

148

3 Interpretation

Under my conception of the decomposed exhaustification procedure, where derived from EXH, Optimal fit determines the distribution of non/scalar EXH incrementally. This way, at least, wh-based bhii-NPIs need not be computed via domain measures, as in Chierchia’s (2013b) system, but by a single application of EXH alone. Thus, the core benefits and architectural advantages of Chierchia’s (2013b) system are retained, and salvaged in light of additional data. Let me now turn to the last μ-meaning: conjunction. EVN

3.3.5 The Conjunctive Profile We assume that the distributive bisyndetic conjunction with the doubling of the μ particle obtains from the Junction of two additive μ-associating propositions. Recall that we derived the additivity of a μ-marked expression via iterative exhaustification that delivered the desired meaning.   (278) EXHR δA . . . [ μ John ] . . . → p ∧ ¬ EXH (p) → also John (for some p containing ‘John’) The idea behind deriving the allosemy of additivity and conjunction is to motivate a structural notion of context whereby the additive or anti-exhaustive clause of each μP finds the relevant alternative, by anaphora, in the other μP.18 The alternative resolution within the JP-generated tuple is thus the following where the alternatives are still contextually determined but structurally so (within the JP). (279)

[p ∧ ¬EXH(p)], [q ∧ ¬EXH(q)] → [p ∧ q], [p ∧ q] → · · · → p ∧ q

In what follows, I implement this idea using some Hittite data in which the Hittite particle -(y)a corresponds to the μ superparticle: the distributive bisyndetic conjunction with the doubling of the -(y)a particle obtains from the Junction of two additive μ-associating propositions, where -(y)a is the Hittite lexicalisation of μ. (280)

D ISTAR URU Šamuha=ma=za GASAN=YA apiya=ya par¯a hantatar Istar Samuha=but=REFL lady=my then=μ divine.providence tikkussanu-[t] show-3.SG.PST ‘Then too Istar of Samuha, my lady, showed her divine power.’ (NH/NS (CTH 85.1.A) KBo 6.29+ obv. ii 29–30; Sideltsev 2017a, 184, ex. 18)

18 The idea that two additives make up a conjunction and whereby each conjunct’s ‘presupposition’

is satisfied by the other, goes back to Kobuchi-Philip (2008).

3.3 μ

149

The basic meaning of (280), stripped of μ, as given in that ‘Istar showed (her) power on 1st March’, where I assume the deictic element apiya ‘then’ denotes some definite date, hence only δA defined since no σ -alternatives are lexically specified.19 With the μ meaning plugged in, the resulting meaning is the set of alternatives to p obtained by replacing ‘1st March’ with ‘2nd March’, etc. Note that the connection between terminals does not indicate movement but rather the Agree chains required for feature valuation. The dotted line indicates iteration of the EXH-application. (The world-dependency has also been removed from the computation for simplicity of exposition.) TPIII

(281)

TPII

EXH [u :δ ]

TPI

O

N

EXH [u :δ ]

DP1

T¯ vP

ITERA

TI

Iˇstar T0 [PAST] μP

vP

AdvP1,[δ ] μ 0 t1

v¯

ya apiya

VP

v0

then

DP

V0 tikkussanut

par hantatar

showed

divine power

(282)

19 I

a.

i. TPI  without μ0  = p = SHOW(Ištar)(her-power)(1-March) = Ištar showed her powers then (on 1st March)

assume that μPs with definite DPs, as the Hittite examples here show, involve recursive exhaustification against the δA only. One argument for assuming that σ A are irrelevant (and undefined) comes from the nature of the lexical specification: if a DP like ‘John’ hosts a μP, then δA is the only possible domain restriction for the relevant EXH (or its reapplication), since definite DPs have no lexically specified scalar alternatives. If the host has a specified interpretable σ -feature, as is the case with indefinite wh-existentials (see Mitrovi´c 2018b), only then is exhaustification against a scalar alternative set possible (and defined).

150

3 Interpretation

  ii. A TPI  =

⎫ ⎧ ⎪ p = SHOW(Ištar)(her-power)(1-March) ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎬ ⎨ q = SHOW(Ištar)(her-power)(2-March) ⎪

r = SHOW(Ištar)(her-power)(3-March) ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ . ⎭ ⎩ ..    b. TPII  = EXH δA(p) p    = EXHδ A(p) p = p ∧ ∀q ∈ ♥A(p)[¬q] = p ∧ ¬q ∧ ¬r = SHOW(Ištar)(her-power)(1-March) ∧ ¬SHOW(Ištar)(her-power)(2-March) ∧ ¬SHOW(Ištar)(her-power)(3-March) ∧ · · · # Ištar showed her powers then (on 1st March) and only then    c. TPIII  = EXHδ EXHδ (p)) p = p ∧ ¬EXH(p) # Ištar showed her powers then (on 1st March) and not only then # Ištar showed her powers also then (on 1st March)

To see the post-suppositional composition of two additive μPs at play, consider a sentence in (283). We assume that each of the two μ-associating DPs are independently additive, as per (280) above. (283)

NH/NS 176, (CTH 176) KUB 21.38, letter to Ramses II) DUMU.MUNUS-YA=ya iwaru=ya pehhi

daughter-my=μ

dowry=μ give.1.SG

‘I will give you both my daughter and the dowry.’ The structural presence of the J-projection ‘joins’ the two μ-containing phrases. Note that the μ-conjunction will yield effects analogous to Conjunction Reduction (CR; see Schein 2017 for a recent and very extensive discussion). Given the adopted definitions of the J and μ formatives, CR need not be stipulated since by the pointwise mechanism, alternatives ‘grow’ from within each μ-marked conjunct to yield a set of alternatives with the μ-marked restrictor DP as a substitution source of alternatives. Similarly, the tuple-forming • (J) will ultimately conjoin DP-level μPs as propositional arguments (see Mitrovi´c 2014, Appendix D; for a similar implementation within Hamblin semantics, see also Alonso-Ovalle 2006, Appendix C, int. al, for discussion and details).

3.3 μ

151

(284)

VP JP

V0 pehhi will give

J¯

μP DP

μP

μ 0 J0 ya DP

DUMU.MUNUS-YA my daughter

μ0 ya

iwaru dowry JP J¯

TPIII TPII

EXH [u :δ ] EXH [u :δ ]

TPIII

J0 TPI

EXH [u :δ ]

VP μP

= DP

DUMU.MUNUS-YA my daughter

TPII EXH [u :δ ]

TPI VP

V0 pei μ0 ya

will give

μP DP

V0 pehhi μ0 ya

will give

iwaru dowry

By the theoretical design of our assumptions regarding the structure of alternatives, (283) recreates the effect of conjunction reduction since each (inherently additive or anti-exhaustive) μ-conjunct constitutes a set of propositional alternatives by virtue of PFA (I return to this below). Note that the μ-marker ya does not undergo movement at all—the dashed line signifies the chain of an Agree relation which obtains the suitable alternativerestricting set (via feature-valuation [uA : ]], [uA : δ] ), against which exhaustification takes place. The EXH-operator is, primarily, a TPII -level operator and, secondarily, applying at TPIII -level. For the two instances of EXH, I follow Fox (2007) in assuming that each of the two EXH operators is introduced separately, therefore through External Merge (and not though movement). They do, however, enter into an Agree relation to ensure the featural valuation of the higher EXH.

152

3 Interpretation

As conjuncts, the μPs in (283) are assumed to have their additive meanings as in (280). The existential presupposition that expressions with an additive particle, like μ in (280), express is contextually satisfied generally. In conjunction structures, we programmatically follow Brasoveanu and Szabolcsi (2013), who analyse the existential presupposition of additives dynamically in order to derive the additive requirement as a postsupposition (itself being a delayed update, in a dynamic sense). For our purposes, it suffices to adopt an account which will allow for each inherently additive conjunct to satisfy the other conjunct’s additive/existential presupposition. We assume that under JP the exhaustive (or, alternative negating) conjunct in the meaning of the recursively exhaustified μP is interpreted as a postsupposition (notated with ‘ # ’). Given the additive meaning we derived above and the postulated lexical entry for J0 (212), we derive in (285) the conjunction in (283). Take the first conjunct in (284) to be p, and the second conjunct q.    (285) (284) = J0  μP1  μP2  = μP1  • μP2 

= μP1 , μP2  = [p ∧ ¬EXH(p)], [q ∧ ¬EXH(q)] = [p ∧ ¬EXH(p)] ∧ [q ∧ ¬EXH(q)]   = [p ∧ ¬EXH (p)] ∧ [q ∧ ¬EXH (q)]   = p ∧ q ∧ ¬EXH (p) ∧ ¬EXH (q) =p∧q Ultimately, do we need the post-suppositional mechanism? If it is naturally motivated, on the ontological basis of presuppositions, then their role in the theory should not be problematic, albeit not traditional. These need brief mentioning, at least, since the competing analyses to my own20 assume such mechanism is operative in order to derive the correct distribution of meaning. Postsuppositions are in fact taken as core devices that derive coordinate meanings of the superparticles (both μ and κ). On the other hand, the meaning of a recursively exhaustivised μP with antiexhaustive clause adjoined to the prejacent, technically, has no presuppositional content, which is a technical and conceptual advantage of using a covert ‘only’type exhaustifier EXH which does nothing but assert. (Presupposition projection of an anti-exhaustive proposition is not obligated by any lexical specification of μ.) Diachronically semantically, English too and South Slavonic i or Hittite -(y)a— while all additive markers—thus may be best analysed to give the same (additive) meaning using different entries. In (286), I submit a possible view of developing

20 Essentially

passing.

those of Brasoveanu and Szabolcsi (2013); Szabolcsi (2015, 2018) explicitly or in

3.4 κ 

153

additivity from an assertion-based (and EXH-utilising) additive marker like our μ to a strongly presupposing additive marker, like ‘also’/‘too’ in English. (286)

A developmental view of anti-exhaustive μP: a. μ1  = λp[p ∧ ¬EXH(p)] (no projection, static meaning) b.  μ2  = λp : ¬EXH(p)pstsp [p] (postsupposition projection, dynamic) c.  μ3  = λp : ¬EXH(p)prsp [p] (presupposition projection, static) d.  μ4  = λp : ∃q ∈ A(p | q = p)[p ∧ q] (∃-presup. projection, static) e.  μ5  = λp∃q ∈ A(p | q = p)[p ∧ q] (static ∃-assertion)

With this last conjecture, which has no bearing on the preceding discussion, I conclude the μ section and move on to the obverse of μ. The historical dimension of μ-meanings is undertaken in the next chapter.

3.4 κ  We now turn to the κ-series, which morphosyntactically covers disjunctive, existential and interrogative constructions. The microsemantic core, common to all these instantiations is the ‘foregrounding’ of alternatives. Unlike μ, κ is not an e-level particle and whatever it combines with, κ returns the ordinary semantic value of its alternatives. Additionally, κ ensures that the alternative set is not singleton (presumably via anti-singleton presupposition, in line with Alonso-Ovalle and Menédez-Benito 2010).21 Let me state the general entry I will be defending: (287)

Lexical entry for κ 0 

κP = κ κ0

XP = =

21 Note

XP

f o

XP XP

o

XP s.t. #( ) > 1 o

however, as a reviewer reminds me, that the anti-singleton constraint of Alonso-Ovalle and Menédez-Benito (2010) is used to explain a particular type of ‘epistemic indefinite’ in Spanish and they do not utilise it to make a general analysis for disjunction or superparticles, as I do here. While κ does not appear in the formation of Spanish indefinites, I adopt Alonso-Ovalle and Menédez-Benito’s (2010) theory differently and argue that the anti-singleton meaning is one of the components of an overarching characterisation of κ superparticles.

154

3 Interpretation

The κ does not associate with EXH by virtue of any narrow syntactic specification, unlike μ. Given that κ-based indefinites resist negative, or generally DE, contexts and are characterised as Positive Polarity Items (PPIs) in the wh-domain, the analysis I put forth is compatible with previous approaches to PPIs (Nicolae 2017b, int. al.). Recall the assumptions I laid out in (229), which I repeat below. (288)

a. The alternative calculation obtains at a propositional level. b. The alternative set to φ is the extension (set) of φ. i. For a proposition p, the alternatives are, therefore, elements in Dt , i.e. {pw , ¬pw } or {1, 0}. ii. For a wh-term, like who, the extension and the alternative set, subject to contextual restriction, is {x | HUMAN(x)} or all the (relevant) individuals, {a, b, c, · · · } (in the focus dimension, which is the extension of wh-terms).

The minimal semantic ascription in (287) is sufficient, at least for the present purposes, to derive the three differential meanings of κ-featuring constructions: indefinite existentials, (polar) interrogatives, and disjunctions.22 The exemplar language through which we navigate this section is Japanese and in it I, in the general analysis, follow Uegaki (2018), starting with his descriptive generalisation: (289)

Generalisation: When the ka-phrase is syntactically smaller than a CP, its semantic contribution is an existential quantifier (without the question force); when it syntactically forms a CP, its semantic contribution is to form a question involving alternatives expressed by the wh-item/disjunction. (Uegaki 2018)

Since the proposed microsemantics of the κ returns on the ordinary semantic value of alternatives, I also use the fraction notation, oA fA , representing the ordinary semantic dimension above the focus dimension of meaning.23 Firstly, κ will thus always return an ordinarily value alternative set, even when it associates with extra-ordinary, or focus, values (although κ-hosts tend to be nearly exclusively expressions with an inherently non-void focus alternative value). Using evidence from counterfactuals, disjunctions under modals, and the exclusive disjunctive meanings, Alonso-Ovalle (2006) motivates a Hamblinian treatment of disjunction, according to which they denote sets of ordinary alternatives. In simple terms: (290) 22 The

p ∨ q = {p, q}oA

κ-system also features, at least in some languages, in focus constructions. See Slade (2018) for a historical and typological overview and Mitrovi´c (2018a) for a focus-like treatment of κ in Ser-Bo-Croatian. While focus-sensitivity of κ is not within the scope of this chapter, Chap. 4 investigates focus as a historical origin of the κ system in Japonic. 23 Uegaki (2018) uses tuples but I’ve already made extensive use of tuple notation.

3.4 κ 

155

I rely on his result and view heavily since κ is automatically disjunction-like by virtue of it forming sets with ordinary values, which is exactly how AlonsoOvalle’s (2006) disjunction is designed. (For a review of advantages of a Hamblinian alternative semantics for disjunction, see, for instance Slade 2011, 199, 244 and those he cites.) While ordinary declaratives denote singleton sets whose only member is the relevant TP-denoting proposition, wh-words themselves denote nonsingleton sets, composing with the rest of the sentence, TP, in such a fashion that allows the alternatives ‘to grow’ (such compositionality is discussed below). The result of such a combination is a wh-corresponding set of propositions (290), that is, answers. Secondly, κ meanings cannot be singletons (here my take is inspired by AlonsoOvalle and Menédez-Benito 2010). There are three ways this supposition is structurally checked and enforced in κ-expressions. In association with indefinites and expressions of existentials or wh-interrogatives, the anti-singleton fact comes for free by virtue of the focus-alternative set denoted by the wh-term. In polar interrogatives, the anti-singleton nature of the meaning of questions is structurally provided by a Qt operator (which I borrow from George 2011 and) which maps a proposition to a doubleton set of truth-valued.24 In disjunctions, the anti-singleton requirement that κ imposes is checked at JP level.

3.4.1 The Existential Quantificational Profile The existential constructions with κ seem to be e-level meanings, unlike the other two, which I will claim are inherently propositional in nature, also against the background of other scholars’ work. In the field, semanticists have been converging on a focus-based characterisation of wh-meanings. Crucially, wh-words are said to denote (Hamblin) sets in the focus-dimension, while their denotations are undefined in the ordinary dimension, following Beck (2006); Beck and Kim (2006); Cable (2007, 2010); Kotek (2014) (among many others). (291)

what =

oA:UNDEFINED fA:{x ∈ De | x ∈ NON-HUMAN}

The proposed meaning for κ, which is in line with proposals found in Kratzer and Shimoyama (2002) or Kotek (2014, 66n14), rests on its role to define the otherwise undefined ordinary semantic values for a given term. In case of wh-hosts, κ ‘copies’ its semantic values from f to o, resulting in the existential quantificational expression. Note that on Cable’s (2007, 2010) account, the indefinite and interrogative uses of κ (a Q-particle in his terms) are derived using a different mechanism than the 24 Any

approach to interrogative semantics that derives the Boolean doubleton of a polar question is compatible with my views here.

156

3 Interpretation

one I invoke. The allosemy25 of wh-κ, realising either as a indefinite (dare-ka, ‘someone’) or interrogative (dare . . . ka?, ‘who . . . ?’), is determined externally to the κP. Namely, κPs that raise to Spec(CPQ ) are interpreted as interrogative while those that remain undergo existential closure (kicking in at the TP level) in Cable (2007, 2010). Under my account, being closer to that of Kratzer and Shimoyama (2002), κ ‘closes off’ the focus-dimension of a wh-phrase locally. The idea here being a Hamblinian one, namely that the wh-alternatives, being focal, keep ‘growing’ until a suitable binder is found. For technical soundness, we need Hamblin’s (1973) tool of Point-wise Functional Application which allows for such alternative growth compositionally, the succinct definition of which I borrow from Xiang (2016, 6n1.11). Definition 3.1 Point-wise Functional Application (PWFA) If α : ς, τ and β : ς , then i. αg = Dς,τ ii. βg = Dς

  iii. α(β) : τ and α(β) = f (d) | f ∈ αg , d ∈ βg Given the syntactic assumptions, the κ takes an indeterminate DP as complement and projects a κP, hence the binder to the wh-alternative set is locally present within the κP.26 Technically, and also conceptually, I will assume that there exists a natural principle of ‘translation’ from ordinarily valued Hamblin sets into standard nonalternative semantics. Consider (292) with a toy domain of individuals Abby, Billybob, and Caesar, denoted by the wh-term. For individual-level variables, ae , be , ce stand for the three members in our domain. Corresponding propositional alternatives are a, b, c. (292)

25 In

[Dare-ka] odorimasu who-κ dance ‘Someone is dancing.’

Cable (2007, 2010), the κ (Q) particle building interrogatives and indefinites rests on the assumption that the two types of constructions feature distinct elements (Cable 2010, 215n27). The notion of allosemy is my reinterpretation of how his analysis could be brought closer to the present one. 26 The cross-categorial combination of wh-terms (featuring in DP-level indefinites and CP-level interrogatives) can be achieved using Xiang’s (2016) Hybrid approach according to which whterms are of an ambivalent type τ, τ .

3.4 κ 

157

a.

b. Entries: i. dare = ‘who’ = UNDEFINED {x|x∈De } = UNDEFINED {ae ,be ,ce } ii. ka =

∅ x

!→

x x

iii. odorimasu =

λx[DANCEw (x)] {λx[DANCEw (x)]}

Note that the denotation of κP satisfies the anti-singleton supposition that κ lexically imposes (since the wh-host denotes the three-membered domain). Therefore, no additional closure is needed since the κ alone provides the ordinary values for the host it associates with. At a DP level, combining with a wh-word, κP denotes an ordinarily valued set of alternatives, corresponding to disjunction of individuals (Alonso-Ovalle 2006) which, in turn, corresponds to an existential quantifier. Similarly, a complex wh-κ indefinite expressions, such as the one in (293), would deliver the same result, modulo the following. Following Shimoyama (2006), I assume dono gakusei ‘which student’ is interpreted as a set of students of type et . (293)

[Dono gakusei ka] odorimasu which student κ dance ‘Someone is dancing.’

The κ then copies the focus-valued extension of that set (assume it’s Abby, Billybob, and Caesar again) into ordinary values, at which point we arrive at the denotation similar to the denotation of κP in (292a), where the restrictor is STUDENT .

3.4.1.1

The Semantic Values of wh-Terms

Before moving onto the implementation of how the κ-superparticle forms and derives other meanings, let me briefly discuss the alternative structures that whterms denote. In my assumption that wh-terms have undefined ordinary semantic meaning, I am trying to remain as uncontroversial as possible and maintain this

158

3 Interpretation

view, as advocated in (294), taken (with slight modification) from Beck and Reis (2018) which further assumes a decomposition of wh-terms into their wh proper component and a restrictor component (the sister). (294)

a. who = undefined b. whf = {x | x ∈ Dξ } (where ξ is the type of wh’s sister’s argument)

The view for further decomposition of wh-terms seems to be necessary in order to capture some non-canonical focus constructions featuring wh-terms. There are at least three types of focus constructions which do not derive their meanings from the focus-alternative extensions of the wh-terms. Consider the following three: (295)

Non-canonical wh-focus expressions: a. Echo wh-questions: Boris did WHAT? b. Higher-order alternative wh-expressions: I want to know WHERE the party is (and not WHEN)? (Beck and Reis 2018, 404n83b, Slade 2011) c. Rhetorical wh-questions: Ko li je to? who κ is that ‘Who the fuck/on earth is that?’ (Ser-Bo-Croatian; Mitrovi´c 2018a)

The three types of non-canonical wh-alternative meanings in (295) cannot be accounted for using the standard assumptions concerning the denotation of whterms. For echo wh-questions (295a), Beck and Reis (2018) claim the set of relevant alternative semantic value of who is not a set of things of the type of the ordinary semantics of who, such as {Boris, Donald, . . . }, but rather a unique relevant entity, e.g., the echo-wh-alternative set of what in (295a) is a singleton containing ‘that’, which Beck and Reis (2018) derive by decomposing wh-terms along the lines I mention above. Furthermore, they demonstrate that the actual wh-component of the wh-pronoun is Focus-marked in echo wh-questions. With regard to the ‘higher-order’ wh-expressions, the composition only seems to be derived using the decompositional view. Data of the kind given in (295b) involve focus on the restrictor alone: ‘person vs. thing, place vs. time or perhaps some other participant in the event’ (Beck and Reis 2018, 404l but cf. also Slade 2011 and Eckardt 2007). The negative-biased rhetorical questions which feature both the wh-term and the κ-superparticle in Ser-Bo-Croatian can also be derived using similar assumptions, as sketched in Mitrovi´c (2018a), in concert with the pragmatic apparatus of surprise and likelihood-ordering. I have shown in Mitrovi´c (2018a) that rhetorical questions in Ser-Bo-Croatian (which obligatorily communicate negative bias

3.4 κ 

159

and surprise) involve both question- and focus-marking separately and overtly (assuming C0 [+FORCE] marks interrogativity and C0 [+FOC] focus) as shown the following piece of evidence from a Bosnian dialect: (296)

l(i) je to? Ko li Who C.FORCE C.FOC is that ‘Who the fuck/on earth is that?’

The reading can be derived in a fashion similar to the composition of echo whquestions, by assuming that the wh-component, and not the restrictor NP base, is focus-marked. All three types of non-canonical wh-focus constructions require, at least, a decomposition that a reviewer suggests and Beck and Reis (2018) advocate, whereby allowing for a sub-component of the relevant wh-term to undergo focusmarking and -association. This way, the wh-base NP has the relevant ordinary semantic value which may bring about the relevant readings. Given in (297) is a sketch of this wh-anatomy and a coarse typology of wh-focus meanings that can be derived by assuming that various components of the whP may be focus-marked. (297)

A sketch of the anatomy of wh-terms and the typology of wh-focus: CP FOC

(a) (b)

whP

a. Canonical wh-focus b. Non-canonical wh-focus (295a, 295c) c. Non-canonical wh-focus (295b) NP

wh0

-o

PERSON NON - PERSON

wh-

PLACE

-at

-ere

...

(c)

While I do not pursue a unified semantics of non-canonical wh-focus constructions here, I am nonetheless led to maintain the assumption in (294). The nature and possible universality of the anatomy of wh-terms, possibly along the lines of (297), is something I will leave for future research since the seemingly counter-evidentiary data in (295) do not shake the underlying assumption I am adopting. In any case, see Slade (2011), Eckardt (2007), Beck and Reis (2018), and those they cite for further discussion on this matter.

160

3 Interpretation

3.4.2 The Interrogative Profile The κ-marking of questions is essentially a t-level meaning as it associates with a propositional host, unlike the e-level host type involved in the formation of indefinites, overviewed in the previous subsection. In interrogative contexts, I take κ to sit in the edge of the clause, spelling out the interrogative C head. Let me review two types of questions in which κ features (at least in Japanese; cf. Mitrovi´c 2018a and the discussion on non-canonical wh): wh-questions and polar questions. Wh-questions denote sets of possible wh-corresponding answers. A proposition like ‘Billybob smokes’ can be turned into a wh-question, for instance, by abstracting over the subject. This way, the meaning of the question ‘Who smokes?’ is the set of all possible (full) answers, namely ‘x smokes’ for all individuals in our threemembered domain. This is traditionally achieved by assuming that the relevant interrogative head in the clausal spine, say Force, is interpreted as a Proto-Question (PQ) formation (Karttunen 1977) which takes the declarative proposition and returns a set of true propositions that are identical to the proposition. (298)

λ p. x[P(x) λp

p = SMOKESw (x)] x[P(x)

p = SMOKESw (x)] λ x[p = SMOKESw (x)]

x1 [iWH]

p = SMOKESw (x)

λ1

SMOKES (x)

ForcePQ [uWH] λ p.λ q[p = q]

p

x1 smokes

Thus the answer set for a wh-question, such as the one exemplified in (298), comprises of propositions answering ‘who smokes’. Let’s see an example of a κ-marked wh-question from Japanese and how we derive it in the bidimensional approach (some structure and technicalia is omitted since I’ve explicated it above; see George 2011 or Xiang 2016 for technical details which are not relevant to my goals).   (299) [Dare-ga odorimasu] ka ? who-NOM dance κ ‘Who is dancing?’

3.4 κ 

161 λ p. x { λ p. x

(300)

e [P(x) e [P(x)

p= DANCEw (x)] p= DANCEw (x)]}

λp { λ p}

e [ DANCE w (x)] e [ DANCE w (x)]}

x { x

{ x

UNDEFINED

e [P(x)

p= DANCEw (x)]}

x x

λ x[DANCEw (x)] { λ x[ DANCEw (x)]}

UNDEFINED

{ x

0/ x

e [P(x)]}

Consider now polar questions which also denote a set of two possible answers. (301)

Is Billybob a communist? a. Billybob is a communist. b. Billybob is not a communist.

The polar questions are interpreted along the same lines as wh-questions, modulo the wh-movement. Using the Proto-Question entry and an additional truth-value identifying operator, we derive the desired meaning of polar questions: the set of (the two) possible answers. Let’s see how this works with the κ entry using SerBo-Croatian (which induces 2P movement of the auxiliary in (302), which I ignore, interpreting κ li in its C-initial position). (302)

Je

li Bilibob komunista? κB communist ‘Is Billybob a communist?’ AUX .3. SG

λ p[p= COMMUNISTw (b)] {λ p[p= COMMUNISTw (b)]}

(303)

p= COMMUNISTw (b) {p= COMMUNISTw (b)}

λp {λ p} 0/ x

x x

p= COMMUNISTw (b) {p= COMMUNISTw (b)} b {b}

λ x[COMMUNISTw (x)] {λ x[ COMMUNISTw (x)]}

Recall the meaning I proposed for κ with two characteristics. The first was that κ ensures an ordinarily valued alternative set. In (303), this is delivered and is true throughout the edge of the CP, resulting in the Hamblin form. The second characteristic for κ meanings I submitted was that the alternative set be antisingleton. The resulting question in (303) has no hard-wired alternatives. In order to turn the result into a set of corresponding answers, I follow George (2011, 180) in specifying Qt , an operator that delivers that.

162

(304)

3 Interpretation

Qt : λαstt .λpst .∃vt [p = λws (α(w ))(v)] a.

λ p. vt [p= COMMUNISTw ( b)= v] { λ p. vt [p= COMMUNISTw ( b)= v]}

Qt

λ p[p= COMMUNISTw ( b)] { λ p[p= COMMUNISTw ( b)]}

... b. The polar answers: i. λw [p = 1] ii. λw [p = 0] I assume κ is the lexicalisation of PQ (which is static across interrogative interpretations) with the Qt formative satisfying its anti-singleton presupposition (whose valuation may be structurally delayed, as I explore in the next subsection). For polar questions, therefore, the κ ends up with the meaning given in (305), which derives the standard signature meaning of PQ, namely the disjunction of, or set (of ordinarily valued), the two answers. (305)

κw (p) = pw ∨ ¬pw = pw , ¬pw

In Sect. 3.5, I investigate a curious case of both κ and μ particles taking part in expressions of disjunctions.

3.4.3 The Disjunctive Profile What additives are (structurally) to conjunction, interrogatives are to disjunction, modulo the non-trivial difference in types or categories: additives are nominal, interrogatives are propositional. I contend that disjunctions, too, are strictly propositional. Strong and independent arguments for such a treatment can be found in Alonso-Ovalle (2006), Hirsch (2017), and those they cite. Consider the relevant example from Japanese which is derived and interpreted as in (307) and (308) respectively. (306)

Taro ka Akira ka Tomoaki (ka) ga tabe-ta T κ A κ T κ NOM eat-PST ‘Taro, Akira, or Tomoaki ate.’

3.4 κ 

(307)

163

Reduction analysis: (a) as a reduction of (b)

a. Taro or Akira or Tomoaki ate.

TP T

JP2 J

JP1 J0

J

κP

T0 ga

κ0

ka J0

Taro

tabeta

κP

κ0 DP

vP

κ0 ka

DP

ka

DP

κP

Tomoaki

Akira

b. Taro ate or Akira ate or Tomoaki ate. JP

β [0u BOOL:κ ]

JP2 J

JP1 κP TP Taro tabeta

J0

J κ 0 J0

κP TP

κP TP

κ0

κ0

Tomoaki tabeta

Akira tabeta

Recall that Qt ensured the ordinary alternative set—qua disjunction in AlonsoOvalle’s (2006) sense—of a polar question. This suppositional requirement is somewhat relaxed in disjunction. The anti-singleton imposition of κ is structurally provided by virtue of the JP superstructure. Therefore, the anti-singleton suppositions of each κP in JP1 are satisfied at the tuple level, with c entering into the computation via • at JP2 level. This can be analysed as an instance of allosemy on satisfying the anti-singleton requirement between polar questions and disjunctions. Informally, what Qt does in PQs, J does in disjunctions.

164

(308)

3 Interpretation

(307b) =

JP3 = [a • b] • c β [ GD]

JP2 = [a • b] • c JP1 = a • b ao λ x[x • b]

λ x[x • c] • co

• bo

= β [ao • bo ] • bo = ao • bo • bo o = a bo co

(distributivity of product domains)

3.5 μ ◦ κ  ◦ J and the Composition of Atypically Complex Disjunction This final section essentially serves two purposes. Firstly, it introduces novel data which go against the ‘coin’ view of superparticles: in a superparticle expression, either something is κ-marked or it is μ-marked. This novel data shows simultaneous marking by the κ and μ particles (as well as the J superstructural operator). Secondly, this data provides testing ground for the microsemantics of the two superparticles I have been proposing and illustrating. Can the simultaneously (κ and μ) marked meanings, which are in line with the proposal involve five formative builders, deliver those, and only those, meanings that derive from the compositional analysis of the five heads? I will illustrate an affirmative view. To start, consider the exclusive disjunction data from Ser-Bo-Croatian in (309)

Mujo jede (i -li) jagnjetinu i -li sladoled — znam da (ne) M eats.3.SG μ κ lamb μ κ ice-cream know.1.SG that not oboje. voli like.3.SG both. ‘Mujo is eating (either) lamb or ice-cream — I know he does (not) like both.

We may assume a treatment on a par with other forms of ‘long [or bisyndetic] disjunction’, such as the soit/soit-type in French. The motivation is borne out by virtue of the particle complexes i-li repeating in (309) and we could apply Spector’s (2014) analysis and state that ili/ili triggers obligatory scalar exhaustification (EXHσ ). In fact, in (228) in Sect. 3.3 this was illustrated for exposition. While faithful to such types of analyses as is Spector’s (2014), I contend in this section that each of the repeated disjunction markers signals the presence of a local nonscalar (δ-confined) exhaustification. In (228), it was also illustrated how local

3.5 μ ◦ κ  ◦ J and the Composition of Atypically Complex Disjunction

165

exhaustification may derive the exclusive component in disjunction, which is what I will rely on. However, this is not sufficient in terms of the overall microsemantic ambition of this book, which tries to be as faithful and observant as possible of the tiny details in the makeup of logical operators. Apart from the repetitive syntax (which I tie in with local exhaustivity effects), the very disjoining ‘word’ i-li contains a conjunctive and a disjunctive particle. And not just any conjunctive and disjunctive particles, but μ and κ particles. As demonstrated in this chapter, μ and κ don’t only encode conjunction and disjunction but also encode, and are central building blocks of, related logical meanings, such as additivity, universal quantification, PS (μ) and disjunctions, existential quantification, and questions (κ). Additionally. de/compositional semantic technology ought to be blind to the arbitrary word boundaries (cf. Halle and Marantz 1993, and the vast Distributed Morphology corpus of research), which makes me want to consider microsemantically that i-li is a particle complex. The prima facie problem in a nutshell is the following: the μ superparticle has no business in expressing or helping express disjunction. There are other languages apart from Ser-Bo-Croatian (and presumably Slavonic more generally) that show the patterns in (309). First, I will go through the set of data from these languages in Sect. 3.5.1 Ultimately, as I am unable to transplant any existing approaches to account for all the facts, I propose a compositional account using the two familiar lexical entries in Sect. 3.5.2.

3.5.1 Too Many Particles in So Many Languages The puzzling Slavonic quirk in (309) is not an isolated case. It is not only found still in varieties of Slavonic, but in other (though extinct) IE languages, such as Homeric Greek, Hittite, and Tocharian (both A and B). Beyond, I found such constructions in the languages of the Northeastern Caucuses, with at least three languages showing the same type of expression: Avar, Dargi, Husniq,27 and Lezgian.

3.5.1.1

Indo-European

Homeric Greek We start with Homeric Greek, where one of the disjunction markers, e¯ te, is morphologically complex in the sense that is comprises the disjunctive/interrogative κ-particle e¯ and a conjunction-signalling μ-particle te.

27 Note, however, that the Husniq complex disjunction morphemes are borrowed from Avar (Forker

2013), which is why I am excluding the relevant data here.

166

3 Interpretation

While the e¯ particle is independently interrogative and te additive (Denniston 1950, 282–284), e¯ -te is traditionally taken as encoding exclusive disjunction: ‘(either. . . ) or’, or ‘whether . . . or’ (Autenrieth 1895, 134). ¯ E-t(e) ehremen para soi κ-μ keep with self ‘. . . or to keep with yourself’

(310)

(Il. T. 148)

Another particle complex in Homeric eite, comprising of a conditional ei (‘if’) and the μ-particle te: boulesthe polemein emin ei-te filoi einai wish to be at war for myself COND-μ friend be ‘whether you wish to wage war upon us or [else] to be our friends’ (Cyrop. 3.2.13.)

(311)

ei-te

COND -μ

Old Church (and Modern) Slavonic In Old Church Slavonic (OCS), as well the contemporary descendants of Old Common Slavonic, the disjunction marker ili comprises a μ marker i and an κ marker li. (312)

i d i tlo μ soul (J) μ body ‘both body and soul’

(313)

li bratrijo li sestry κ brothers (J) κ sisters ‘brothers or sisters’

(314)

i -li enj i -li dti μ -κ wife (J) μ -κ children ‘. . . either wife or children’

(CM. Mt. 10:28)

(CM. Mk. 10:29.1)

(CZ. Mt. 19:29)

Hittite In Hittite, too, the disjunction ‘word’ contains a conjunctive μ morpheme.28 As Hoffner and Melchert (2008, 405) note, disjunction is regularly expressed in Hittite by našma ‘or’ or by naššu . . . našma ‘either . . . or’. The ma marker is a conjunction marker and indeed a μ-superparticle, while its morphological presence in the expression of disjunction is quite mysterious. Diachronically, we also know that the marker našma must have developed by syncope from naššu+ma, which definitely contains the conjunction (and universal distributive) marker -(m)a. 28 The

-(m)a morpheme is an adversative conjunction (see Mitrovi´c and Sideltsev 2017a,b, 2018).

3.5 μ ◦ κ  ◦ J and the Composition of Atypically Complex Disjunction

(315)

167

nu-šši naššu adanna peškezzi naš-ma-šši akuwanna peškezzi now-him κ ? -(μ) =either eat give give κ ? -μ-him drink ‘He either gives him to eat or he gives him to drink’ (KUB 13.4 i 24)

But what is the na- morpheme? While we cannot provide any definitive evidence on the meaning of the other particle, naššu, Kloekhorst (2008) takes the namorpheme to instantiate negation.29

Tocharian Tocharian A (TA) also shows the same morphological complexity of its disjunction marking. In TA, the additive marker is pe. The complex disjunction marker clearly featuring pe is e-pe, with an additional e- morpheme. The pair of examples in (316) and (317) show the additive and (exclusive) disjunctive construction, respectively. (316)

(317)

pe klo´säm n¯añi μ ears.DU 1.GEN ‘also my ears’ ck¯acar e-pe s´äm . e-pe sister κ-μ wife κ-μ ‘(either) sister or wife’

(TA 5: 53, b3/A 58b3)

(TA 428: a4, b2)

What does the e- morpheme contribute to the conjunctive meaning and what is its role in complex disjunction marking? Just like Hittite, e- can be understood as a negation, according to Adams (2013).30 See Adams (2013, 89) and Edgerton (1953) for philological evidence. In Tocharian and Hittite, disjunctions contain conjunctive and negative particles. So the disjunction ‘word’ need not even contain a disjunction but rather a μ marker and negation seem to be sufficient. The analysis I give derives the logical presence and role of both the disjunctive marker and the negative marker. Thus, the Hittite and Tocharian data, while

29 Another

pair of enclitic disjunctive markers, attested from the oldest written stage of the language, is -(a)ku . . . -(a)ku translating as ‘whether . . . or’. While Hoffner and Melchert (2008) do not remark on the morphological composition of the expression, the -ku component reflects the PIE conjunctive (super)particle *kw e (Kloekhorst 2008, 483), which is of μ character across IE (cf. Chaps. 2 and 4). (1)

30 My

LU LU =ku GUD =ku [ UD ] U =ku e¯ šzi human being-(κ+)μ ox-(κ)-μ [she]ep-(κ)-μ be ‘. . . whether it be human being, ox or [she]ep.’ (KBo 6.3 iv 53)

thanks to Laura Grestenberger for bringing this to my attention.

168

3 Interpretation

constituting a seemingly separate class, will derive the disjunctive inference in the same fashion.

3.5.1.2

North-Eastern Caucasian

Our last set of decomposition-supporting data comes from a non-IE and non-extinct group of North Eastern Caucasian: Dargi, Avar, and Lezgian. It seems likely that looking beyond these three languages (or four, including Husniq) would result in finding more evidence for the frequency of the pattern where κ-μ complexes feature in marking complex disjunction.

Dargi Take an example featuring negative disjunction of the “neither. . . nor”-type, which shows the disjunctive morpheme ya head-initially and bisyndetically when two DPs (‘pilaf’ and ‘hen’) are coordinated. (318)

amma ya pulaw, nu-ni umˆxu sune-la mer.li-ˇci-b b-arg-i-ra, me-ERG key.ABS self-GEN place-SUP-N N-find-AOR-1 but κ pilaf.ABS ya ‘är‘ä he-d-arg-i-ra κ hen.ABS NEG-PL-find-AOR-1 ‘I found the key at its place, but neither the pilaf nor the chicken was there.’ (van der Berg 2004, 203)

On the other hand, and just like disjunction, conjunction can also be expressed polysyndetically using a μ particle ra: (319)

buruš ra yurˇgan ‘änala ra ra il.a-la take-GER this-GEN mattress.ABS μ blanket.ABS)μ pillow.ABS μ kas-ili sar i be.PL ‘(They) took his mattress, [and] blanket and pillow.’ (van der Berg 2004, 199)

Exclusive disjunction, however, and perhaps by now not as surprisingly, features both ya (κ) and ra (μ) particles, as evidence in (320) shows. (320)

ya ra pilaw b-ir-ehe, ya ra nerˇg b-ir-ehe κ μ pilaf(ABS) N-do-FUT.1 κ μ soup(ABS) N-do-FUT.1 (‘What shall we make for lunch?’) ‘We’ll make (either) pilaf or soup.’ (van der Berg 2004, 204)

3.5 μ ◦ κ  ◦ J and the Composition of Atypically Complex Disjunction

169

Avar The same pattern can be found in Avar: expression of exclusive disjunction is achieved with both a μ and a κ morpheme. Let me repeat some variants on the cat-and-dog example from Chap. 2, coupled with the exclusive disjunction particle complex. (321)

(322)

keto gi va hve gi cat μ J dog μ ‘(both) cat and dog’

(Ramazanov, p.c.)

ya gi Sasha ya gi Vanya κ μS (J) κ μ V ‘either Sasha or Vanya.’

(Mukhtareva, p.c.)

Lezgian The data thus far showed adjacent and potentially continuous particle strings. How do we know there’s no lexical unity of μ and κ? This could invalidate the account I am pursuing since the particle string might constitute a single word and its internal structure might be semantically (compositionally) and pragmatically irrelevant. The following datum from Lezgian demonstrates that these particle strings cannot be treated as singular logical atoms incarnating, say, ‘∨’ since they do not appear in adjacent positions with respect to the con/dis/junct. (323)

Zun ja juxsul-ni tuš, ja kesib-ni. I.ABS κ lean-μ COP.NEG κ poor-μ ‘I am neither lean nor poor.’

(Haspelmath 1993, 334, ex. 917)

Lezgian clearly shows κ and μ are autonomous logical formatives. Furthermore, Dimitry Ganenkov (pers. comm.) informs me that Aghul and Tabasaran behave similarly with respect to μ and κ exponents featuring in expressions of disjunction.

3.5.1.3

Data Summary

What I have shown is that a range of languages, living and dead, do something very unintuitive, and even bizarre: they express disjunction using a conjunction marker. We provide in Table 3.2 a summary of morphosyntactic facts. Recall that the [ ] feature on the Slavonic li particle in Table 3.2 refers to its second-position status which is derived via incorporation of the conjunctive/universal i particle. (Consult Chap. 2 for details.)

170

3 Interpretation

Table 3.2 Complex disjunction markers and their morphosyntax cross-linguistically JP κP κ0

μP

J0

κP κ0

μ0

μP μ0

Homeric

e¯

te

0/

(e¯

te)

OC & Modern Slavonic

li[+ε]

i

0/

li[+ε]

i

Hittite

naˇs

(ma)

0/

naˇs

ma

Tocharian A

e

pe

0/

e

pe

Dargi

ya

ra

0/

ya

ra

Avar

ya

gi

0/

ya

gi

Lezgian

ya

ni

0/

ya

ni

3.5.2 Complex Disjunction as a Superparticle Complex Disjunction What the preceding sections have demonstrated is that coordination expressions (especially polysyndetic) result from a rather rich morphosyntactic structure. This syntactic structure has been plotted using a Junction phrase, which pairs up two superparticle-headed coordinands. In the last section, evidence from five languages has empirically signalled a complexly headed structure for each of the coordinands, where both a μ and κ are present. The microsemantic idea behind this section is that both μ and κ meaningfully contribute to the expression of disjunction, albeit in a complex and logically gymnastic manner. Given in (324) is the compositional transposition of the motivated syntactic structure for polysyndetic (exclusive) disjunction in Table 3.2, sketching the corresponding compositional pathway of structural interpretation. The β-operator turns the JP, a tuple, into a disjunction, via GD determined by the minimality-compliant checking relation between β 0 and κ 0 . A level below, a tuple-forming J0 takes two coordinands as arguments, each of which is a composite function of κ ◦ μ functions applied over respective coordinands (XP, YP).

3.5 μ ◦ κ  ◦ J and the Composition of Atypically Complex Disjunction

(324)

171

Complex disjunction as a superparticle complex The morphosyntactic structure represents the LF of an exclusive-disjunctive expression. $ $ %$ %%        0 0 0 0 0 J  κ1  μ1  XP κ2  μ2  YP    XP ∨ YP ∧ ¬ XP ∧ YP

There is a proof to this theorem at the end of the section. For the reader that decides not to skip, I will go through the calculation incrementally. Our alternative tree involves two alternative-triggering operators, incarnated by the μ and κ superparticles, and one alternative-insensitive Junction head which will pair coordinands and let a c-commanding β operator turn the tuple into a Boolean expression, as per (218) and (219). The no-look-ahead principle will thus allow for ‘embedded’ alternatives, where a κ operator will function over a μ-triggered and exhaustified set of alternatives.31 We will therefore end up computing and composing the meaning of a complexlymarked disjunction in four steps, as the morphosyntactic analysis from the previous section suggested. These compositional steps are shown in (325) and paraphrased in (326). (325)

The compositional steps in interpreting JP+ : 4 β0

3 2 κ0

J0

1 μ0

XP

2 κ0

1 μ0

31 As

YP

a matter of methodological principle, stemming from idealised parsimony, we will also assume that there are no semantically vacuous morphemes: therefore a derivation adds compositional meaning. Alternatively, we assume that the inferential system pays close attention to every step of the interpretational (and derivational) procedure, in the form of triviality checks, in the sense of Romoli (2015), and references therein.

172

(326)

3 Interpretation

Paraphrasing the compositional steps in interpreting JP+ : 1 2 3

4

μP as FA of μ0  and its argument (coordinand) κP as FA of κ 0  and μP JP as tuple-forming FA of J0  and two κPs (structural coordinands) JP+  as FA of β 0  and JP

In the paragraphs that follow, we take each of the compositional steps in turn, starting with the first.

3.5.2.1

Step 1

The first compositional step concerns the μP. Since the μ-associate is not an indeterminate, additivity via iterative exhaustification obtains. (327)

Composing μP (a sketch): JP+ β0

JP κP

J

κ0

J0

μP μ0

XP

κP κ0

μP μ0

YP

Assume a standard additive μ expression, where μ combines with a DP, like John, and which is point-wise ‘lifted’ to propositional level. μP is, or indeed has to be, recursively exhaustified, in line with the rationale from Sect. 3.3.4.

3.5.2.2

Step 2: Interpreting κP

We now take a structural step higher, where the result of step 1, μP, namely (327), is fed into κ, which has the anti-singleton foregrounding meaning, as discussed in Sect. 3.4.

3.5 μ ◦ κ  ◦ J and the Composition of Atypically Complex Disjunction

(328)

173

Composing κP (a sketch): JP+ β0

JP J⬘

κP κ0

J0

μP μ0

XP

κP κ0

μP μ0

YP

κ takes the μP with the denotation [p ∧ ¬EXH(p)] as complement and foregrounds it in an anti-singleton format. The second contribution is the anti-singleton supposition: I assume this is not inherent to the μP, nor is the valuation delayed until the third step involving JP. I therefore assume κ in these constructions functions as it does in interrogatives and that the μP is treated as a proposition. As for polar interrogatives, assume the presence of Qt , which combines with the anti-exhaustive μP and returns a doubleton set, containing μP and its negation. (329)

Composing κP:   κP = κ 0  μP    = λp pw ∨ ¬pw [p ∧ ¬EXH(p)] = [pw ∧ ¬EXH(pw )] ∨ ¬[pw ∧ ¬EXH(pw )] (by DEM)

(by elem. log.)

= [pw ∧ ¬EXH(pw )] ∨ [¬pw ∨ EXH(pw )] = [pw ∧ ¬EXH(pw )] ∨ [pw → EXH(pw )]

The result of (331), even though it constitutes an inconsistent set, is true for both of the disjuncts, hence a pair of such sets is paired up by J0 .

3.5.2.3

Step 3: Interpreting JP

We now pair up the two κ-marked coordinands, with an embedded μP each, via the Junction head.

174

(330)

3 Interpretation

Composing JP (a sketch): JP+ β0

JP κP

J

κ0

J0

μP μ0

XP

κP κ0

μP μ0

(331)

YP

Composing JP:    JP = J 0  κP 1  κP 2      (by Lex. it.) = λyλx x • y κP 1  κP 2  (by FA)

= κP 1  • κP 2 

= κP 1 , κP 2   ' & [pw ∧¬EXH(pw )]∨[pw →EXH(pw )] ,  =  [qw ∧¬EXH(qw )]∨[qw →EXH(qw )]

3.5.2.4

Step 4: GD via β

In the last step, we complete the composition by turning the JP-pair into a Boolean expression by applying GD to the product.

3.5 μ ◦ κ  ◦ J and the Composition of Atypically Complex Disjunction

(332)

175

Booleanising JP and Composing JP+ (a sketch): JP+ β0

JP κP

J

κ0

J0

μP μ0

XP

κP κ0

μP μ0

YP

Minimality ensures that the uninterpretable feature [uF : ] on β 0 is checked by bearing [iκ]. The checked feature [u F : κ] is then interpreted as an instruction to map JP via GD. κ0

(333)

Composing JP+ :   J P +  = β 0  J P   

 (by F − check.) = λP  P κP 1 , κP 2   

& [pw ∧¬EXH(pw )]∨[pw →EXH(pw )] , '   (byF A&GD) =  = (byAO) =

[qw ∧¬EXH(qw )]∨[qw →EXH(qw )]



[pw ∧¬EXH(pw )]∨[pw →EXH(pw )] ∨   [qw ∧¬EXH(qw )]∨[qw →EXH(qw )]

  [pw ∧ ¬EXH(pw )] ∨ [pw → EXH(pw )] ,   [qw ∧ ¬EXH(qw )] ∨ [qw → EXH(qw )]

The generated alternative set is inconsistent and contains several subsets. While we will appeal to the ♥-algorithm to avoid inconsistencies, we need to make sure that the input to ♥ is a flat set. I now take a brief excursus in order to obviate this technical difficulty by positing a set-flattening function.

176

3 Interpretation

We aim to derive a set of propositional alternatives from a set of sets of propositional alternatives.32 To do so, we define a function, BÜ, which will flatten the generated set containing alternative sets (333) and return an elemental set of alternatives to which ♥ may apply. Let BÜ be a recursive injective function which flattens a set A, i.e. creates a list of elements in A. Formally, BÜ is a recursive morphism from ⊆-reflexive objects33 to ⊆-irreflexive objects (i.e., to only those objects which are not subsets of themselves). (334)

BÜ (xx⊆x )

= yy⊆y

¯ in (335), where This derives the flattened alternative set, which we label A, the anaphoric resolution of the anti-exhaustive meaning tied to each of the μP is provided in (335b-ii) (335)

¯ a. BÜ : (333) !→ A b. i. Beforealternative resolution of the anti-exhaustive clause: ¯ = [p ∧ ¬EXH(p)], [q ∧ ¬EXH(q)], [p → EXH(p)], [q → A  EXH (q)] ii. Afteralternative resolution of the anti-exhaustive clause: ¯ = [p ∧ q], [q ∧ p], [p → EXH(p)], [q → EXH(q)] A

The alternative set is still inconsistent but in the form on which we now may impose the ♥-function which will negate an optimal amount of alternative subsets until consistency obtains.  [p ∧ ¬EXH(p)], [¬p ∨ EXH(p)], + . . . . . . . . . . . . . . . .  ¬CONS (336) JP  = [q ∧ ¬EXH(q)], [¬q ∨ EXH(q)]   a. [p ∧ ¬EXH(p)],[q ∧ ¬EXH(q)] = [p ∧ q], [q ∧ p] . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . excludable: HC   b. [¬p ∨ EXH(p)], [¬q ∨ EXH(q)] = [p → EXH(p)], [q → EXH(q)] . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .  We assume that, since the entire set (336) is inconsistent, one of the two maximal consistent subsets is the resulting denotation. The first consistent set in (336a), however, is excludable for two reasons. For one, (336a) violates HC, as briefly sketched in (337). (337) 32 This

Sketch of a proof:

was inspired by a comment by Daniel Büring, to whom many thanks. take a standard assumption that all sets are subsets of themselves (see Halmos 1960, int. al.):   (1) TH. ∀S S ⊆ S   ∵ ∀x x ∈ S → x ∈ S ⇒ S⊆S  

33 We

3.5 μ ◦ κ  ◦ J and the Composition of Atypically Complex Disjunction

177

a. With alternative resolution As per our assumptions, let p, q ∈ C.   The alternative set [p ∧ ¬EXH(p)], [q ∧ ¬EXH(q)] thus comprises of the two disjunct candidates. The first, [p ∧ ¬EXH(p)] entails q since ¬EXH(p)  q, and [q ∧ ¬EXH(q)] entails p since ¬EXH(q)  p. This violates HC.   b. With alternative resolution. Similarly and even more straightforwardly, for p, q ∈ JP. [q ∧ ¬EXH(q)]  [q ∧ p], and [p ∧ ¬EXH(p)]  [p ∧ q], therefore [p ∧ q] ∨ [q ∧ p] does not only violate HC, but is also trivial.   Another possible reason for exclusion of (336a) is, perhaps, the strengthening condition we stipulated in fn. 31, which amounts to stipulating, on the grounds of possibly natural principles, that alternative-sensitive morphemes, like μ and κ, enrich (and not simply maintain) meaning structurally. This view finds support in recent work on the general aspects of and interaction of triviality and grammaticality (Gajewski 2002), or more explicit statements of structural triviality verifications (Romoli 2015).34 A μP with an anti-exhaustive meaning, once fed into κ, should not, then, yield a κP with the meaning identical to that of μP alone. Our resulting denotation, however, contains the μP meanings of each disjuncts and, may, be excluded for reasons of structural enrichment. The other consistent subset in (336b) has a clear meaning of exclusivity: whenever one disjunct is true, only that disjunct is true (EXH(p)). This is the desired result with the exclusive component. Proof (324) The result of each of the two μPs is an anti-exhaustive formula for the two juncts (i) and (ii). i. p ∧ ¬EXH(p) ii. q ∧ ¬EXH(q) The result of each of the two κPs is an ordinarily valued alternative set, containing the propositional meaning of its host and the alternative to it. Since the host of κ 0 is the resulting μP, the composition of each of the two κP juncts results in the two denotations in (i) and (ii). i. [p ∧ ¬EXH(p)] ∨ ¬[p ∧ ¬EXH(p)] ii. [q ∧ ¬EXH(q)] ∨ ¬[q ∧ ¬EXH(q)] By De Morgan Law: i. [p ∧ ¬EXH(p)] ∨ [¬p ∨ EXH(p)] 34 Note,

however, that Romoli (2015) assumes that triviality checks kick in with every (covert) movement (i.e., internal merge) operation. We hypothesise, in a rather similar vein, that triviality scrutiny applies with every external merge, or at least with every externally merged non-terminal node.

178

3 Interpretation

ii. [q ∧ ¬EXH(q)] ∨ [¬q ∨ EXH(q)] By definition of the β-valuation of the J0 , determined by minimality, the two juncts in the JP-denoting tuple are mapped onto GD, resulting in the following disjunction of the two κPs.     [p ∧ ¬EXH(p)] ∨ [¬p ∨ EXH(p)] ∨ [q ∧ ¬EXH(q)] ∨ [¬q ∨ EXH(q)] The anti-exhaustive inference picks up the alternative(s) tuple-internally anaphorically. Assume the anaphora cannot be resolved until the juncts are fully composed.     [p ∧ q] ∨ [¬p ∨ EXH(p)] ∨ [q ∧ p] ∨ [¬q ∨ EXH(q)] Applying the function to flatten the alternative sets yields:  [p ∧ q] ∨ [¬p ∨ EXH(p)] ∨ [q ∧ p] ∨ [¬q ∨ EXH(q)] Due to inconsistency of the resulting set, ♥ applies, returning a consistent set (where ♥ = {p ∧ q}):  [¬p ∨ EXH(p)] ∨ [¬q ∨ EXH(q)] By elementary logic, this amounts to  [p → EXH(p)] ∨ [q → EXH(q)] which has the exclusivity component and which completes the proof.

 

Note that if alternatives were resolved at μP-level, that is at the embedded site, the exhaustivity inference could not percolate upward. Resolution at the root ensures a stronger meaning. I assume the structural presence of κ blocks this resolution before the junct meaning if, fully composed. The result could be obtained if resolution (¬EXH(p) → q) were suppressed and then negated: this is the outlier pattern we find in Hittite and, possibly, Tocharian where the relevant complex disjunction marker comprises of a μ and a negation particle. Therefore this is accounted for, also.

Appendix This appendix outlines the formal foundation of the semantics that is utilised in this thesis. The semantic language I use is adopted in full from Chierchia (2013b, 136–139) where a version of two-sorted type theory known as TY2 is adopted.

Appendix

179

A partial version of TY2 is used along with Kleene’s (1950) strong logic of indeterminacy (K3 ) for connectives and quantifiers, enriched so as to include indexical expressions (like I or here). Definition 3.2 (Types) i. Basic types: e (entities), t (truth values), w (worlds/situations) ii. Functional types: if a, b are types, a, b is a type (of functions of things of types a into things of type b) Definition 3.3 (Syntax) i. Lexicon: for any type a, we have denumerably many variables and constants of that type ii. Functional application: if β is of type a, b and α is of type a, β(α) is of type b iii. λ-abstraction: if α is a variable of type a and β is an expression of type b, λα[β] is of type a, b iv. If φ, ψ are of type t, and α is a variable of any type, then the following are expressions of type t: ¬φ, (φ ∧ ψ), (φ ∨ ψ),(φ → ψ),(φ ↔ ψ),∀α[φ],∃α[φ]. Definition 3.4 (Domains) i. De = U ii. Dt = {0, 1} iii. Dw = W iv. Da,b = [Da ⇒ Db ] Definition 3.5 (Model) A model M is a triplet U, W, F , where W, U are as per definition above and F is a function such that for any w ∈ W and any constant α of type a, F (α)(w) ∈ Da . An assignment g maps each variable of type a into a member of Da . For any well-formed expression β, β is the value of β relative to M, an assignment to the variables g and a world/situation w, if defined. We will generally omit M from the superscript. The world w in the exponent of the interpretation function ·g,w is to be understood as playing the role of the context of evaluation. This means that expressions like man or walk (of type s, e, t ) are going to have a constant value across contexts, while the value of the expressions like I (of type e) are going to vary across contexts. Examples: i. For any w, F (walk)(w) is that member of d of Ds,e,t such that for any w ∈ Dw and any u ∈ De , d(w )(u) = 1 iff u walks in w ii. For any w, F (I)(w) = u, where u is the speaker in w

180

3 Interpretation

Definition 3.6 (Semantics) i. if α is a variable of type a, αg,w = g(α) ii. if α is a constant, αg,w = F (α)(w) iii. β(α)g,w = βg,w (αg,w ), if defined (else undefined) α

iv. for any u, λα[β]g,w (u) = βg[ u ],w , if defined (else undefined) α

v. ∃α[φ]g,w = 1 iff for some u, φg[ u ],w = 1 α

vi. ∃α[φ]g,w = 0 iff for all u in Da (a being the type of α), φg[ u ],w = 0, else undefined vii. ¬φg,w = 1 iff φg,w = 0, else undefined viii. φ ∧ ψg,w = 1 iff φg,w = ψg,w = 1 ix. φ ∧ ψg,w = 0 iff φg,w = 0 or ψg,w = 0, else undefined x. Truth: an expression φ of type t is true relative to w (i.e., φw = q) iff for any appropriate function g to the free variables in φ relative to w, φg,w = 1. An assignment g to the free variables of φ relative to w is appropriate iff for all variables w of type s occurring free in φ, g(w) = w.35 Definition 3.7 (Generalised Entailment)



i. if φ, ψ are of type t, φ ⊆ ψw = 1 iff for any w if φw = 1, then ψw = 1 ii. if β,  γ are of type a,  b , where b is the type that ends in t, then: β ⊆ γ := ∀α [β](α) ⊆ [γ ](α) , where α is a variable of type a Definition 3.8 (Consistency) A set of formulae Φ is consistent (CON(Φ)) iff there is no formula φ such that Φ  φ and Φ  ¬φ. Otherwise, Φ is inconsistent (INC(Φ)) i. Φ is simply consistent iff for no formula φ of Φ, both φ and ¬φ are theorems of Φ. ii. Φ is absolutely consistent iff at least one formula of Φ is not a theorem of Φ.   iii. Φ is maximally consistent iff for every formula φ, if CON Φ ∪ φ , then φ ∈ Φ iv. Φ is said to contain  witnesses iff  for every formula of the form ∃xφ, there exists a term t such that ∃xφ → φ xt Definition 3.9 (Equivalence Classes) i. Commutativity: a. φ ∧ ψ ⇔ ψ ∧ φ b. φ ∨ ψ ⇔ ψ ∨ φ 35 The

effect of this definition of truth is that a formula like RUN(John)(w) is true relative to w iff

RUN(John)(w)g,w = 1, for any g such that g(w) = w.

Appendix

181

ii. Associativity: a. φ ∧ (ψ ∧ χ ) ⇔ (φ ∧ ψ) ∧ χ b. φ ∨ (ψ ∨ χ ) ⇔ (φ ∨ ψ) ∨ χ iii. Distributivity: a. φ ∧ (ψ ∨ χ ) ⇔ (φ ∧ ψ) ∨ (φ ∧ χ ) b. φ ∨ (ψ ∧ χ ) ⇔ (φ ∨ ψ) ∧ (φ ∨ χ ) Definition 3.10 (Quantifier Distribution and Boolean Identities) i. Law of quantifier distribution #2 (Partee et al. 1990, 149, ex. 7-9):   ∀x[φ(x) ∧ ψ(x)] ⇐⇒ ∀x[φ(x)] ∧ ∀x[ψ(x)] ii. Law of quantifier distribution #3 (ibid.):   ∃x[φ(x) ∨ ψ(x)] ⇐⇒ ∃x[φ(x)] ∨ ∃x[ψ(x)] Following from the two laws above are the following homomorphisms (see Hammond 2006, 84): |D|    i. ∀x φ(x) = φ(xi ) i=1 |D|    ii. ∃x φ(x) = φ(xi ) i=1

The truth-functional proof in IL for the homomorphism between quantificational and the Boolean terms is given below and is adapted from Takeuti (1987, 63).   Proof (3.9.i) It holds that ∀d ∈ D ∀x[φ(x)]M,g,w ≤ φ(d)M,g,w since   ∀x φ(x) → φ(d) is provable in IL. Now let CM,g,w ≤ φ(d)M,g,w for every d ∈ D. Then Γ, C → φ(d) is provable for every d ∈ D. Take   d to be a free variable, which does not occur in Γ, C, ∀x φ(x) . Then Γ, C, ∀x φ(x) is provable in IL.   Proof (3.9.ii) Along the same lines. This can also be indirectly proven by appealing to the Generalisation Theorem, following Enderton (2001, 117–118), which is stated below. Theorem 3.1 (Generalisation Theorem) If Γ  φ and x does not occur free in any formula, then Γ  ∀xφ. Proof (Generalisation Theorem) Consider a fixed set Γ and a variable x not free in Γ . We will show by induction that for any theorem φ of Γ , we have Γ ∀xφ. For this it suffices (by the induction principle) to show that the set 

φ | Γ ∀xφ



includes Γ ∩ Λ and is closed under modes ponens. Notice that x can occur free in φ. If φ is a logical axiom, then ∀xφ is also a logical axiom. Then, so is Γ ∀xφ.

182

3 Interpretation

Definition 3.11 (De Morgan Laws) In set-theoretic terms:   i.  A ∪ B ⇔ (A) ∩ (B)   ii.  A ∩ B ⇔ (A) ∪ (B) Proof (3.10.i) The proof is adapted  from Mendelson (1990,  6). Let  A ⊂ S and B ⊂ S. Suppose x ∈  A ∪ B . Then x ∈ S and x ∈ A ∪ B . Thus x ∈ A and  x ∈ B, or x ∈ (A) and x ∈ (B). Therefore x ∈ (A) ∩ (B). Therefore,  A ∪ B  (A) ∩ (B). Conversely, suppose x ∈ (A) ∩ (B) . Then x ∈ S and x ∈ (A) and x ∈ (B). Thus x ∈ A and x ∈ B, and therefore  x ∈ A ∪ B . It follows thatx ∈  A ∪ B and, consequently, (A) ∩ (B)   A ∪ B . It therefore  holds that  A ∪ B ⇔ (A) ∩ (B), which completes the proof.   Proof (3.10.ii) This proof follows along the same lines. A shorter proof is obtained  if weapply (def. 3.10.i)   to the two subsets (A) and (B) of S. Therefore:  (A) ∪ (B) =  (A) ∩  (B) = A ∩ B. Taking complements again, we arrive at       (A) ∪ (B) =   (A) ∪ (B) =  A ∩ B , which entails  A ∩ B ⇔ (A) ∪ (B), thus completing the proof.

 

The set-theoretic expressions trivially translate into propositional-logical terms:   i. ¬ φ ∨ ψ ⇔ ¬φ ∧ ¬ψ   ii. ¬ φ ∧ ψ ⇔ ¬φ ∨ ¬ψ In what follows, we outline how alternatives are recursively defined in Chierchia’s (2013b) system. Definition 3.12 (Alternative Set) αA , for any expression α, where A is a function from expressions to a set of interpretations. i. Base clause. Sample lexical entries: a. or/and • orA = orδA ∪ andδA  λpλq[p ∨ q] • orδA = λpλq[p] λpλq[q]  λpλq[p ∧ q] δA • and = λpλq[p] λpλq[q] b. some/a • someδ A,g = someδA ∪ everyδA • someδ δA,g =  λP λQ∃x ∈ δ[P (x) ∧ Q(x)] : δ ⊂ g(δ) • everyδ δA,g =  λP λQ∀x ∈ δ[P (x) ⇒ Q(x)] : δ ⊂ g(δ)

Appendix

183

Other scalar items are defined along similar lines.36 i. For any lexical entry α, unless otherwise specified: (A)  αA = αM,g,w , i.e. any lexical entry is an alternative to itself. ii. Recursive clause.   (Point-wise functional application/PFA) β(α)(δ)A = b(a) : b ∈ β(δ)A and a ∈ α(δ)A iii. Strict σ -alternatives and contextual pruning:   • ασ A = p ∈ αA : for no q ∈ αδA , q ⊂ p • αC/σ A = C, where C is a subset of ασ A such that for any q, if q is a strongest member of ασ A , then q ∈ C iv. Alternative sensitive operators. For each operator we define (i) its truth conditional import (which is always the same) and (ii) its σ -ALT and (iii) its δA.  

a. • EXHσ A φg,w = φg,w ∧ ∀p ∈ φσ -ALT p ⇒ λw [φg,w ] ⊆ p   • EXHσ A φσ -ALT = φg,w 37   • EXHσ A φδA = EXHσ A(δ) (pδ ) : pδ ∈ φδA , where pδ has domain δ and σ A(δ) are the σ -alternatives to p with domain δ. The (sub)domain alternatives (δ As) of an exhaustification of the form EXHσ A φ are the result of applying EXHσ A φ to the δ-alternatives of φ, point-wise.  

b. • EXHδA φg,w = φg,w ∧ ∀p ∈ φδA p ⇒ λw [φg,w ] ⊆ p   • EXHδA φδA = φM,g,w   • EXHδA φσ A = φσ A 38 

c. EXHExh-δ A φg,w = φg,w ∧ ∀p ∈ φExh-δ A p ⇒ λw [φg,w ] ⊆    p , where φExh-δA = φδA|R = EXH♥-δA (p) : p ∈ φδA . For the definition of EXH♥-A (p), see definition below. In defining preexhaustification, we exhaustify p ∈ φδA relative to the members of p ∈ φδA that are innocently excludable (♥) relative to p. The δ-alternatives and σ -alternatives of recursively exhaustified EXHδA|R φ (EXHExh-δA φ) are defined just like those of EXHδA φ in (b). Definition 3.13 Innocent Exclusion (♥) The set of ♥-A relative to p is defined as follows:     a. ♥-Ap = X⊆ ALT : CONS(p ∧ ¬ X) ∧ ∀q ∈ A CONS(p ∧ ¬ X ∧ ¬q) ⇒ q ∈ X   b. EXH♥-A (φ) = φ ∧ ∀p ∈ A p ∈ ♥-Aφ ∧ ∈ p) ⇒ ¬p

36 In

what follows, we omit the value assignment to variables from the superscript of the recursive calculation of the alternatives. 37 That is, once EXH σ A applies to some expression φ, φ’s σ -alternatives (other than φ itself) are no longer available. 38 The σ -alternatives of EXH φ are just the σ -alternatives of φ. δA

184

3 Interpretation

Theorem 3.2 (Recursive Exhaustification Leads to Anti-Exhaustivity) The following is from Fox (2007, 113). Let the following hold.   C = p|p ∈ C (C is a set of propositions with p ∈ C) I = ♥(p, C) = ∅

I = (C − I − p) = ∅   ANTI -EXH = ¬EXHC (q) : q ∈ I ∩ EXHC (p)   If ANTI-EXH = ∅ (is consistent), then EXHC EXHC (p) = EXH2C (p) = ANTI - EXH (p). Proof (Th. 3.2) By definition of EXH:

EXH C (p)

 ∀q ∈ I [¬q]

¬q  ¬EXHC (q)  ∀q ∈ I [¬EXHC (q)]     ANTI -EXH = ¬EXHC (q) : q ∈ I ∩ ¬EXHC (q) : q ∈ I ∩ EXHC (p)   ¬EXHC (q) : q ∈ C − {p} ∩ EXHC (p) = ANTI -EXH

 ♥(EXHC (p), C ) = C | C := {EXHC (q) : q ∈ C}   2 ¬EXHC (q) : q ∈ C − {EXHC (p)} ∩ ¬EXHC (p) EXH C (p) =

CONS ( ANTI -EXH )

= ANTI-EXH  

References Abrusán, M. 2007. Even and free choice any in hungarian. Proceedings of Sinn und Bedeutung 22:1–15. Adams, D.Q. 2013. A Dictionary of Tocharian B. Leiden Series in Indo-European. Amsterdam: Rodopi. Alonso-Ovalle, L. 2006. Disjunction in alternative semantics. PhD thesis, UMass Amherst. Alonso-Ovalle, L., and P. Menédez-Benito. 2010. Modal indefinites. Natural Language Semantics 18:1–31.

References

185

Autenrieth, G. 1895. A Homeric Dictionary for Schools and Colleges. New York: Harper & Brothers. Bade, N. 2015. Obligatory presupposition triggers in discourse. PhD thesis, Universität Tübingen. Beck, S. 2006. Intervention effects follow from focus interpretation. Natural Language Semantics 14:1–56. Beck, S., and S.S. Kim. 2006. Intervention effects in alternative questions. Journal of Comparative Germanic Linguistics 9:165–208. Beck, S., and M. Reis. 2018. On the form and interpretation of echo Wh-questions. Journal of Semantics 35:369–408. Bowler, M. 2014. Conjunction and disjunction in a language without ‘and’. In Proceedings of SALT 24, ed. T. Snider, S. D’Antonio, and M. Weigand, 137–155. Brasoveanu, A., and A. Szabolcsi. 2013. Presuppositional Too, Postsuppositional Too. In The Dynamic, Inquisitive, and Visionary Life of φ, ?φ, and φ: A Festschrift for Jeroen Groenendijk, Martin Stokhof, and Frank Veltman, ed. M. Aloni, M. Franke, and F. Roelofsen, 55–64. Amsterdam: Institute for Logic, Language and Computation, ms. UC Santa Cruz and NYU. Cable, S. 2007. The grammar of Q. PhD thesis, Massachusetts Institute of Technology. . 2010. The Grammar of Q: Q-Particles, Wh-Movement and Pied-Piping. Oxford: Oxford University Press. Chierchia, G. 2004. Scalar implicatures, polarity phenomena, and the syntax/pragmatics interface. In Structures and Beyond: The Cartography of Syntactic Structures, ed. A. Belletti, vol. 3, 39–103. Oxford: Oxford University Press. . 2006. Broaden your views: Implicatures of domain widening and the “logicality” of language. Linguistic Inquiry 37(4):535–590. . 2013a. Free choice nominals and free choice disjunction: The identity thesis. In Alternatives in Semantics, ed. A. F˘al˘au¸s, 50–87. London: Palgrave Macmilan. . 2013b. Logic in Grammar: Polarity, Free Choice and Intervention. Oxford Studies in Semantics and Pragmatics, vol. 2. Oxford: Oxford University Press. Chierchia, G., D. Fox, and B. Spector. 2012. Scalar implicatures as a grammatical phenomenon. In Handbook of Linguistics and Communication Science, ed. C. Maienborn, K. von Heusinger, and P. Portner, vol. 3, 2297–2332. New York: Mouton de Gruyter. Chomsky, N. 1995. The Minimalist Program. Cambridge, MA: MIT Press. Crniˇc, L. 2011. On the meaning and distribution of concessive scalar particles. In Proceedings of NELS 41, ed. N. LaCara, L. Fainleib, and Y. Park, 1–14. Amherst, MA: GLSA. Dayal, V. 2009. Variation in English free choice items. In Universals and Variation: Proceedings of GLOW in Asia VII, ed. R. Mohanty, and M. Menon, 237–265. Hyderabad: EFL University Press. . 2013. A viability constraint on alternatives for free choice. In Alternatives in Semantics, ed. A. F˘al˘au¸s, 88–122. London: Palgrave Macmilan. Denniston, J.D. 1950. The Greek Particles. Oxford: Oxford University Press. Eckardt, R. 2007. Inherent focus on wh-phrases. In Proceedings of Sinn und Bedeutung 11, ed. E. Puig-Waldmüller, 209–228. Barcelona: Universitat Pompeu Fabra. Edgerton, F. 1953. Buddhist Hybrid Sanskrit: Grammar and Dictionary, vol. 2. New Haven: Yale University Press. Elliott, P.D., E. McCready, and Y. Sudo. 2015. Discourse even vs. attitude even. In SALT 25, Stanford University. Enderton, H. 2001. A Mathematical Introduction to Logic, 2nd ed. London: Academic Press. F˘al˘au¸s, A. 2013. Introduction: Alternatives in semantics and pragmatics. In Alternatives in Semantics, ed. A. F˘al˘au¸s, 1–35. London: Palgrave Macmilan. Fauconnier, G. 1975. Polarity and the scale principle. Chicago Linguistic Society 11:188–199. Forker, D. 2013. A Grammar of Hinuq. Berlin: de Gruyter Mouton. Fox, D. 2007. Free choice and scalar implicatures. In Presupposition and Implicature in Compositional Semantics, ed. U. Sauerland, and P. Stateva, 71–120. London: Palgrave Macmilan. Fox, D., and R. Katzir. 2011. On the characterization of alternatives. Natural Language Semantics 19:87–107.

186

3 Interpretation

Francis, N.C. 2017. There’s something odd about presupposition-denying even. In Ling Lunch (28 September 2017). Cambridge, MA: MIT. Gajewski, J. 2002. On analyticity in natural language, ms. MIT. Accessible at http://gajewski. uconn.edu/papers/analytic.pdf. Gajewski, J. 2009. Innocent exclusion is not contradiction free, unpublished Ms. University of Connecticut. Gajewski, J. 2013. An analogy between a connected exceptive phrase and polarity items. In Beyond ‘Any’ and ‘Ever’: New Explorations in Negative Polarity Sensitivity, ed. E. Csipak, R. Eckardt, M. Liu, and M. Sailer, 183–212. Berlin: Mouton de Gruyter. Gaji´c, J. 2015. N-words and the wide scope illusion. In Proceedings of the 33rd West Coast Conference on Formal Linguistics, ed. K. min Kim, P. Umbal, T. Block, Q. Chan, T. Cheng, K. Finney, M. Katz, S. Nickel-Thompson, and L. Shorten, 128–136. Somerville, MA: Cascadilla Press. Gaji´c, J. 2016. Coordination and focus particles (re?)united. In Proceedings of Sinn und Bedeutung, ed. R. Truswell, C. Cummins, C. Heycock, B. Rabern, and H. Rohde, vol. 21, 427–444. Semantics Archive. George, B.R. 2011. Question embedding and the semantics of answers. PhD thesis, UCLA. Guerzoni, E., and Y. Sharvit. 2014. “Whether or not anything” but not “whether anything or not”. In The Art and Craft of Semantics: A Festschrift for Irene Heim, ed. L. Crniˇc, and U. Sauerland, 199–224. Cambridge, MA: MITWPL. Hagstrom, P. 1998. Decomposing questions. PhD thesis, MIT. Halle, M., and A. Marantz. 1993. Distributed morphology and the pieces of inflection. In The View from Building 20: Essays in Linguistics in Honor of Sylvain Bromberger, ed. K. Hale, and S.J. Keyser, 111–176. Cambridge, MA: MIT Press. Halmos, P.R. 1960. Naive Set Theory. Princeton: Van Nostrand. Hamblin, C.L. 1973. Questions in Montague English. Foundations of Language 10(1):41–53. Hammond, M. 2006. Introduction to the Mathematics of Language, unpublished Ms. University of Arizona. Haspelmath, M. 1993. A grammar of Lezgian. Berlin/New York: Mouton de Gruyter. Hirsch, A. 2017. An inflexible semantics for cross-categorial operators. PhD thesis, Massachusetts Institute of Technology. Hoffner, H.A., and H.C. Melchert. 2008. A Grammar of the Hittite Language. Part 1: Reference Grammar. Languages of the Ancient Near East. Winona Lake: Eisenbrauns. Humbertstone, L. 2011. The Connectives. Cambridge, MA: MIT Press. Jamison, S.W., and J.P. Brereton. 2014. The Rigveda. Oxford: Oxford University Press. Karttunen, L. 1977. Syntax and semantics of questions. Linguistics and Philosophy 1:1–44. Katzir, R. 2007. Structurally-defined alternatives. Linguistics and Philosophy 30:669–690. Keenan, E.L., and L.M. Faltz. 1985. Boolean Semantics for Natural Language. No. 23 in Synthese Language Library. Dordrecht: Reidel. Kleene, S.C. 1950. Introduction to Metamathematics. Princeton: D. Van Nostrand. Kloekhorst, A. 2008. Etymological Dictionary of the Hittite Inherited Lexicon. Leiden IndoEuropean Etymological Dictionary Series, vol. 5. Leiden: E.J. Brill. Kobuchi-Philip, M. 2008. Presuppositional compositionality with Japanese mo. In Proceedings of Semantics and Linguistics Theory, ed. T. Friedman, and S. Ito, vol. 18, 496–509. Kotek, H. 2014. Composing questions. PhD thesis, MIT. Kratzer, A., and J. Shimoyama. 2002. Indeterminate phrases: The view from Japanese. In The Proceedings of the Third Tokyo Conference on Psycholinguistics, ed. Y. Otsu, 1–25. Tokyo: Hituzi Syobo. Krifka, M. 1995. The semantics and pragmatics of polarity items. Linguistic Analysis 25:1–49. Kumar, R. 2006. The Syntax of Negation and the Licensing of Negative Polarity Items in Hindi. Studies in Linguistics. London: Routledge. Kuroda, S.Y. 1965. Generative grammatical studies in the Japanese language. PhD thesis, MIT. Ladusaw, W.A. 1979. Polarity sensitivity as inherent scope relations. PhD thesis, University of Texas, Austin.

References

187

Lahiri, U. 1998. Focus and negative polarity in Hindi. Natural Language Semantics 6:57–125. Mendelson, B. 1990. Introduction to Topology, 3rd ed. New York: Dover. Mitrovi´c, M. 2014. Morphosyntactic atoms of propositional logic: A philo-logical programme. PhD thesis, University of Cambridge. . 2018a. Biased wh-interrogatives, manuscript. Bled Institute. . 2018b. Configurational changes in Indo-European coordinate constructions, Chap. 2. In Word Order Change, ed. A.M. Martins, and A. Cardoso, 19–44. Oxford: Oxford University Press. Mitrovi´c, M., and U. Sauerland. 2014. Decomposing coordination. In Proceedings of NELS 44, ed. J. Iyer, and L. Kusmer, vol. 2, 39–52. . 2016. Two conjunctions are better than one. Acta Linguistica Hungarica 63(4):471–494. Mitrovi´c, M., A. Sideltsev. 2017a. Discourse structure & contrast in Hittite, ms. University of Cyprus/Bled Institute and The Russian Academy. . 2017b. Wackernagel effects on discourse structure & supra-clausal phase boundaries in hittite, ms. University of Cyprus/Bled Institute and The Russian Academy. . 2018. Allosemies of the Anatolian conjunction particle. In Logical Vocabulary & Logical Change, Current Issues in Linguistic Theory, ed. M. Mitrovi´c. Amsterdam: John Benjamins. Nicolae, A. 2013. Any questions? polarity as a window into the structure of questions. PhD thesis, Harvard University. . 2015. Questions with NPIs. Natural Language Semantics 23:21–76. . 2017a. Deriving the positive polarity status of plain disjunction. Semantics & Pragmatics 10(5):1–21. . 2017b. A new perspective on the shielding property of positive polarity. In Proceedings of SALT 27, ed. D. Burgdorf, J. Collard, S. Maspong, and B. Stefánsdóttir, 266–281. Partee, B., and M. Rooth. 1983. Generalized conjunction and type ambiguity. In Meaning, Use, and Interpretation of Language, ed. R. Bäuerle, C. Schwarze, and A. Stechow, 361–383. Berlin: Mouton de Gruyter. Partee, B.H., A. ter Meulen, and R. Wall. 1990. Mathematical Methods in Lingusitics. Dordrecht: Kluwer. Payne, J. 1985. Complex phrases and complex sentences. In Language Typology and Syntactic Description, Vol. 2, ed. T. Shopen, 3–41. Cambridge: Cambridge University Press. Ramchand, G.C. 1997. Aspect and Predication: The Semantics of Argument Structure. Oxford: Oxford University Press. Rizzi, L. 1990. Relativized Minimality. Cambridge, MA: MIT Press. . 2001. Relativized minimality effects. In The Handbook of Contemporary Syntactic Theory, 89–110. Oxford: Blackwell. Romoli, J. 2015. A structural account of conservativity. Sematntics-Syntax Interface 2(1):28–57. Rooth, M. 1992. A theory of focus interpretation. Natural Language Semantics 1:75–116. Rullmann, H. 1997. Even, polarity, and scope. In Experimental and Theoretical Linguistics, ed. M. Gibson, G. Wiebe, and G. Libben, vol. 4, 40–64. Edmonton: Department of Linguistics, University of Alberta. Sauerland, U. 2004. Scalar implicatures in complex sentences. Linguistics and Philosophy 27:367– 391. Schein, B. 2017. ‘And’: Conjunction Reduction Redux. Cambridge, MA: MIT Press. Shimoyama, J. 2001. Wh-constructions in japanese. PhD thesis, University of Massachusetts at Amherst. . 2006. Indeterminate phrase quantification in Japanese. Natural Language Semantics 14:139–173. Sideltsev, A. 2017a. Accented clitics in Hittite? Transactions of the Philological Society 115(2):176–211. Skok, P. 1971. Etimologijski rjeˇcnik hrvatskoga ili srpskoga jezika, vol. 1: A–J. Zagreb: Jugoslovenska akademija znanosti i umjetnosti. . 1972. Etimologijski rjeˇcnik hrvatskoga ili srpskoga jezika, vol. 2: K–poni’. Zagreb: Jugoslovenska akademija znanosti i umjetnosti.

188

3 Interpretation

Slade, B.M. 2011. Formal and philological inquiries into the nature of interrogatives, indefinites, disjunction, and focus in Sinhala and other languages. PhD thesis, University of Illinois at Urbana-Champaign. . 2018. History of focus-concord constructions and focus-associated particles in sinhala, with comparison to dravidian and japanese. Glossa 3(1):2.1–28. Spector, B. 2014. Global positive polarity items and obligatory exhaustivity. Semantics & Pragmatics 7(11):1–61. Stanley, R. 1984. Quotients of peck posets. Order 1(1):29–34. Szabolcsi, A. 2004. Positive polarity–negative polarity. Natural Language & Linguistic Theory 22(2):409–452. . 2015. What do quantifier particles do? Linguistics and Philosophy 38:159–204. . 2017a. Additive presuppositions are derived through activating focus alternatives. In Proceedings of the 21st Amsterdam Colloquium, University of Amsterdam/Semantics Archive, ed. A. Cremers, T. van Gessel, and F. Roelofsen, 455–464. . 2017b. Disjunctive and conjunctive particles meet their negative concord relatives. In Workshop on Logical Vocabulary & Logical Change, ICHL23, San Antonio, TX. Handout available at http://mitrovic.co/logvoc. . 2018. Two types of quantifier particles: Quantifier-phrase internal vs. heads on the clausal spine, to appear in Glossa. Manuscript version as of March. Takeuti, G. 1987. Proof Theory, 2nd ed. Mineola, NY: Dover. Uegaki, W. 2018. A unified semantics for the japanese Q-particle ka in indefinites, questions and disjunctions. Glossa 3(1):14.1–45. van der Berg, H. 2004. Coordinating constructions in Daghestanian languages. In Coordinating Constructions, ed. M. Haspelmath, 197–226. Amsterdam: John Benjamins. Von Fintel, K. 1999. NPI licensing, Strawson-entailment, and context dependency. Journal of Semantics 16:97–148. Wild, M. 2005. On rank functions of lattices. Order 22(4):357–370. Winter, Y. 1995. Syncategorematic conjunction and structured meanings. In Proceedings of SALT 5, ed. M. Simons, and T. Galloway. Ithaca: Cornell University & Linguistic Society of America. . 1996. A unified semantic treatment of singular np coordination. Linguistics and Philosophy 19(4):337–391. . 1998. Flexible boolean semantics: coordination, plurality and scope in natural language. PhD thesis, Utrecht University. Xiang, Y. 2016. The Mandarin particle dou: A pre-exhaustification exhaustifier. In Empirical Issues in Syntax and Semantics, ed. C. Piñón, vol. 11, Paris: Colloque de Syntaxe et Sémantique à Paris (CSSP). . 2017. ONLY: An NPI-licenser and NPI-unlicenser. Journal of Semantics 34:447–481.

Chapter 4

Grammaticalisation

Abstract This chapter provides a grammaticalisation-based analysis of superparticle behaviour in Indo-European and Japonic, aiming to derive some common and presumably universal principles of semantic change. In doing so, this chapter not only shows the synchronically explanatory and predictive power of an exhaustification-based approach to the polarity and scalarity systems, but also the application of this conception in historical linguistics generally and the cross-linguistic diachronic semantic/pragmatics specifically. The chapter sketches a diachronic theory of μ-meanings: how they are born and how they shift. (Two such shifts are demonstrated.) In Japonic, the meaning of μ is that of scalar particle: the best means of capturing this distribution is by positing an uninterpretable scalar feature on the Old Japanese μ head. Once this restriction to scalar complements is lost in Classical Japanese, non-scalar meanings arise in Japanese. One such meaning is that of negative polarity or polarity sensitivity more generally. One of the results of this chapter is a hierarchical organisation of a set of syntactically-informed semantic and pragmatic parameters which are used to model the evolution of meaning of the μ superparticle in Indo-European.

The goal of this chapter is to extract principles that govern the nature and directions of variation and historical change in the superparticle system. It comes in two parts. The first (Sect. 4.1) is a diachronic take on the syntactic analysis from Chap. 2. As a semantic obverse of the diachronic analysis, the remainder of this chapter (Sect. 4.2–4.3) is devoted to questions of semantic change. In Sect. 4.2, I introduce two case-studies: one where superparticles diachronically arose (in Japonic), and another where the superparticle system underwent loss (in IE).

4.1 Change in Construction: The Climb and Decline of μ This subsection is a diachronic resumption of the synchronic analysis of alternation between the bimorphemic 1P and the monomorphemic 2P conjunction marker © Springer Nature Switzerland AG 2021 M. Mitrovi´c, Superparticles, Studies in Natural Language and Linguistic Theory 98, https://doi.org/10.1007/978-94-024-2050-0_4

189

190

4 Grammaticalisation

given in Sect. 2.3.4. I’ll first demonstrate that 2P configuration declined over time, and then address the derivational causes for this change as an instance of grammaticalisation.

4.1.1 The Decline of 2P Let me start by sketching the diachronic decline of the 2P configurations of conjunction, in reference to three classical IE languages (Latin, Greek, and Sanskrit). For the sake of exposition, each of the three languages is taken in turn. (Some of the details may be found in the Appendix to Chap. 2.) In Latin, both the 1P et or atque and the 2P que conjunction markers are observable, as the minimal pair of examples in (338) show. (338)

Classical Latin: a. ad summam rem p¯ublicam atque ad omnium nostrum [. . . ] to utmost weal common and to all of us ‘to highest welfare and all our [lives]’ (Cic., Or., 1.VI.27-8) b. v¯ıam sam¯utem que life safety and ‘the life and safety’ (Cic., Or., 1.VI.28-9)

Diachronically, the 2P marker is lost with the initial et becoming the predominantly single device for expressing conjunction, which is plotted in Fig. 4.1. For a general treatment of the historical change in configurationality in Latin, see Ledgeway (2015). The relevance of the bimorphemic and que-containing 1P marker atque will become relevant in Sect. 2.3 where an explicit analysis is given. An identical synchronic and diachronic pattern is found in the history of Greek where the frequently found 2P configuration in Homeric (339) declines in the postHomeric period, which is plotted in Fig. 4.2. (For more details on the historical trajectory of the loss of the 2P te conjunction in Greek, see Goldstein 2016.) (339)

a. aspidas eukuklous lais¯eia te pteroenta shields round pelt and feathered ‘The round shields and fluttering targets.’ (Il. M. 426) b. ke¯ıs’ e¯ımi kaí anti¯o polemoio there go and meet battle ‘Go thither, and confront the war.’ (Il. M. 368)

4.1 Change in Construction: The Climb and Decline of μ

191

relative occurence (%)

100 80 60 40 20 0

1 BCE

0

1 CE 2 BCE 3 BCE 4 BCE century et (1P) que (2P) atque (1P)

Fig. 4.1 Grammar of conjunction in Latin: et, que, and atque from 1st c. BCE to 4th c. CE

R.gvedic, the earliest attested variety of Indic, also allowed both the 1P and the 2P expressions of conjunction. Sanskrit, however, especially in its post-Vedic periods, is an exception to the IE diachronic typology of the decline of 2P conjunctions. In the post-Vedic period, Sanskrit lost the 1P configuration and kept almost exclusively the 2P system. In Fig. 4.3, the decline of 1P in the history of Sanskrit is plotted. This exceptionality can be explained by understanding (Classical) Sanskrit as a non-vernacular language. In Prakrit vernaculars, which underwent a natural historical change (which Sanskrit did not, at least to a comparable extent), the loss of 2P occurs, in line with the diachronic typology of IE. The Iranian part of Indo-Iranian also shows a natural decline of 2P conjunctions. Consider just the change of the relative frequency of the 1P ut¯a and 2P ca/c¯a in Avestan (cca. 16–12 c. BCE), the earliest attested form of Iranian, compared to the later Old Persian (cca. 6 c. BCE), as Fig. 4.4 demonstrates.1

4.1.2 The Derivational Change as Cause for Loss of 2P Since clauses (CPs) are inherently phasal (Chomsky 2001, et seq.), they provide the selecting head μ with far less search space, or in the case of (65), an empty set of possible incorporees. In non-CP coordinands, [ε] may be checked by virtue of 1 The

Avestan data are my own, while the Old Persian data are based on Klein (1988).

192

4 Grammaticalisation

Fig. 4.2 Grammar of conjunction in Greek: kaí and te from 8th c. BCE to 15th c.

100

relative occurence (%)

CE

80 60 40 20

15 CE

2 CE

5 BCE

8 BCE

0

century kai (1P)

100 80 60 40 20

uta/utá (1P)

ca (2P)

la te

cl as sic al m ed iv al

ep ic

ea rly

0

ar ha ic

relative occurence (%)

Fig. 4.3 Grammar of conjunction in the history of Sanskrit: uta and ca from the Vedic to the late period

te (2P)

4.1 Change in Construction: The Climb and Decline of μ

100

relative occurence (%)

Fig. 4.4 Grammar of conjunction in the history of Iranian: 1P uta/ut¯a and 2P ca/c¯a Avestan and Old Persian

193

80 60 40 20

Pe rs ia n

period

O ld

Av es ta n

0

uta/utā (1P) ca/cā (2P)

access to terminals in μ0 ’s complement’s interior. The derivation of non-clausal coordination is therefore strictly cyclical: once an XP is derived (cycle I), it is selected by μ0 (cycle II) whose [ε] feature is checked Agree-wise. The μ category is in turn incremented by J0 (cycle III), as shown in (340a). The external coordinand2 is merged in Spec(JP) (cycle IV) in line with cycles II. and III. Stopping off the derivation at the point of the second cycle obtains bare μPs. The third J0 -cycle obtains a syntactic structure for coordination. Diachronically, the change occurs in the collapsing of the second and third cycles, whereby μ0 and J0 feature in a single cycle and thereby inherently yield bimorphemic coordinators, morphologically and lexically deleting [ε] on μ0 , which in time gets buried under J0 , as instantiated in (340b). The interdependence of the J-μ complex is empirically and technically analogous to proposals by Chomsky (2008) and Richards (2007), among others, who claim that T0 is lexically defective, bearing no φ-features of its own, and instead inherits its φ-features from the phase head C0 . In light of this, μ0 can be analysed as lexically defective, requiring an overt J0 to delete [ε].

2 The

derivation of the external coordinand is ignored here.

194

(340)

4 Grammaticalisation

Change: (a) !→ (b)

a.

III

II

I

μ PINT

JP

μ0 [ε]

...

XP ZP

J0

X0

YP ...

b.

II

I JP

XP

...

0/ J0 [+ ]

μ PINT μ0 [ε]

X0

YP

0/

...

I propose that the [ε], reflective of the Wackernagel (2) effect, on μ may be checked by two distinct means: either by selection or internal merge. In the latter case, μ is 2P and structurally retains its compositional variety (understood parametrically, as discussed in Sect. 4.3) and projectional independence. The grammaticalisation of the former (change in checking by selection and not internal merge) entails that the conjunction markers are bimorphemic and in 1P. With such ¯ a change, the [A]-related properties are lost along with the gradual disappearance of [ε] and, with it, no discourse-related interpretations (such as those associated with the independent μ-cycle) are possible. This is in fact borne out across IndoEuropean. The only exception is the Slavonic branch where the μ particle i was never a Wackernagel particle, thus such changes never took place in the Slavonic JP-μP conjunction structure. This view predicts that the loss of enclitic monomorphemic coordinators, and the inverse rise of the inherently initial bimorphemic coordinators, entails the loss of independent μP, which features in focal additive, polar and scalar construction, as in (94)–(57). This is in fact confirmed. The only exception to this diachronic interlock between changes in word order and semantics would be a case where μ0 would not carry [ε] and thus would not get buried under J0 in time. The Slavonic branch is such an exception, which has lexically syncretised the entries for J0 and μ0 as i but the semantics of the coordinate/non-coordinate constructions clearly

4.1 Change in Construction: The Climb and Decline of μ

195

shows that two forms of i existed in OCS, which is preserved in most branches of synchronic Slavonic. This is schematised in generalised form in Fig. 4.5, as Δ3 . Diachronically, the last resort option of realising an overt J0 to host the μ-particles (340) becomes the first response. I use the term ‘first response’, again, very pre-theoretically to label any form of movement which is not triggered by last-resort economy. Clausal coordination type generalises to all categories as μ0 comes preinstalled with a hosting morpheme. Historically, this entails the loss of Wackernagel movement and the development of lexicalised J-morpheme. Figure 4.5 represents the culmination of this section, with the following description acting as an explanation, commentary and discussion of the view of grammaticalisation I have defended. Stage I Primarily, all early IE languages, and presumably this was inherited from PIE, had an e-type μ superparticle, which was able to deliver not only elevel conjunction, via the J superstructure, but independently (as μPs) delivered additivity in association with a determinate host (say, DP) and a quantificational expression (universal, FCI, PSI) in association with an indeterminate (wh) host. Which of the three quantificational meanings was primary is discussed in the next chapter, where I defend the idea that it is universal quantification. Stage II This first change, leading to the second stage via Δ1 , is based on evidence we find in Gothic, where the otherwise e-level μ particle is found in the fixed clausal position. The other focus-sensitive nature of μ in the nominal extended projection is best analysed, in order to account for the innovation in Gothic, as having undergone reanalysis: from a nominal to a clausal focus element. Semantically, this gives rise to a problem, which I address below and dub ‘Gothic paradox’. Note that other IE languages did not undergo this change of ‘climb’ (so Stage II is does not apply) but went straight into the ‘decline’ change (from Stage III to IV). Stage III The third stage is far more common in the history of early IE languages (with a single exception, Slavonic). It pertains to the loss of 2P in coordinate structures. In the synchronic analysis, I accounted for the nature of 1P conjunctions by demonstrating how this is derivationally a last-resort option, invoking economy principles in the (synchronic) derivation. Diachronically, I proposed in this section that the Second change (Δ2 ), from Stage III to IV, is best understood as a change in the economy. What was the last-resort option (realising the Jmorpheme in order to check [ε], or the 2P requirement on μ) becomes the first-response principle. Stage IV The economy analysis for the third stage predicts the diachronic takeover of 1P bimorphemic conjunction markers. This is empirically reflected in all branches of IE. Stage V The 1P strategy, operative as a first-response realisation of the Junction head, underwent grammaticalisation by fusing the two heads into one. This may be understood as resulting from the ‘heads-over-phrases preference’, proposed

196

4 Grammaticalisation

ge

VI

sta

YP

&P

0

&

Δ5

XP

ge

V

sta

DP JP 0

CP

μ

0

J

Δ4

ion

fus

ge

IV

sta

CP JP ×

Δ 3 : decline

0

CP

μ

0

J

nse

po

res

ge

III

sta

αP JP

nt me

0

0

0

e ov

-m

nse

po

res

J

ort

Δ2

βP

X

μ es t r

las

ge

DP CP

II

sta

0

μ

JP 0

Δ 1 : climb

DP

ge

sta

0

μ

J

P)

(D

I

DP

JP

nt me

0

DP

0

μ

0

X

e ov

-m

J

Fig. 4.5 A diachronic theory of the 6-stage development of the conjunction structure in IndoEuropean

4.2 Change in Interpretation

197

by van Gelderen (2004, 61) which legislates “Be a Head rather than a Phrase (if possible).” Another explicans could be analogical reanalysis which “eliminates derivational opacity by reconstructing the [structure] involved as derived more transparently by an already existing rule” (Lightfoot 1979, 372). Stage VI Previous changes result in a single coordination formative &0 are reflected in nearly all modern IE languages, including English.

4.2 Change in Interpretation We know far more about syntactic change than we do about, formal and nonlexical(ist), semantic change. The remainder of this chapter outlines a way of doing formal diachronic microsemantics of superparticles by looking into the history of Indo-European and Japonic. There are two empirical cases of diverging diachronic development. In one, the superparticles underwent development (the case of Japonic) and, in the other case, the superparticle system was lost (essentially, Indo-European), which we are familiar with on the basis of synchronic and diachronic analyses from previous chapters. I identified two general historical patterns in the diachrony of superparticles in some of my previous work (Mitrovi´c 2018b, 2019), which I summarise in (341). (341)

Two secondary grammaticalisation patterns of superparticles a. The deflational pattern: the loss of superparticle system in Indo-European b. The inflational pattern: the development of superparticle system in Japonic

While the direction of change is not universal, I explore in this chapter the components of the historical drive which are universal, as I will end up suggesting in (349). There are two theoretical preliminaries which help in the transition between diachronic syntax and compositional change, which I now briefly discuss. The first stems from accepted and well considered assumptions surrounding the modular nature of grammar. Specifically, interpretation is determined by a homomorphism between an algebra of syntactic representations and an algebra of semantic objects. Assume, rather standardly, that those objects form a set (Chomsky 1995, et seq.) or, conversely, a partially ordered set and lattice . A seemingly trivial function φ, then, is an order-preserving map between the structures, or ‘trees’, in syntax ( SYN ) and the trees relevant to semantics ( SEM ). (342)

!  SEM φ : SYN → such that, ∀x, y ∈ [x ≤

SYN

y] ⇒ φ(x) ≤

SEM

φ(y)



198

4 Grammaticalisation

The idea of constituency-preservation then entails that any form of syntactic (at least the forms affecting the structure) are, by default, instances of compositional semantic change. The second is that hierarchical syntactic structure applies “all the way down” (Halle and Marantz 1993, int. al.). ith the advent of decompositional schools of morphology, such as Distributed Morphology (Halle and Marantz 1994; Embick and Noyer 1999, 2001; Embick 2010) (which I adopt), the demarcation of syntax and morphology, and the very notion of word and word boundary got blunted, and nearly eliminated. Formal semantics, however, has lagged behind such advances although it too necessarily relies on precise morphosyntactic structures it takes as its own compositional objects of enquiry. Given the two ideas, microsemantics, as programmatically advocated here, cannot be done independently from formal morphosemantics. While much of pragmatic business has not been backtracked to semantics, recent work (Chierchia 2013) has tied it closely to a syntactic reality, blurring the lines between that which is ungrammatical and that which is illogical. Independently, yet in parallel, work on the relation between morphosyntactic structure and interpretation has resulted in an explicit statement, which we termed the Microsemantic Principle in (1), repeated below in (343). (343)

The Microsemantic Principle Compositional analysis cannot stop at the word level. (Since there is no word-level boundary.)

Recent work on the anatomy of quantifiers (Leu 2009) and quantificational morphemes (Kratzer and Shimoyama 2002; Szabolcsi 2015; Mitrovi´c and Sauerland 2016, int. al.) has implicitly, or explicitly, aligned itself with the programmatic thrust of (343) and this paper aims to contribute in this direction, also.

4.2.1 Formal Semantics of Grammaticalisation Before we turn to the data in the following subsection, I devote the remainder of this section to a preliminary overview of and theorisation on semantic change, aiming to tie it to minimalist models of syntactic change generally and the diachronic syntactic analysis from Sect. 4.1. Research on grammaticalisation has solidified the view that linguistic change, not only phonological, is regular or at least systematic. As Gianollo et al. (2014, 4) note, morphology and the lexicon played a very important role in this enterprise, but mostly in their interface with phonology: meaning was only taken into consideration insofar as it contributed evidence for establishing diachronic and cross-linguistic links between lexical and grammatical items. While there have been subsequent and steady advances made in the area of diachronic generative morphosyntax, especially in the principles and parameters approach, lifted from comparative to diachronic syntax (Roberts 2007), diachronic semantics in the 1990s was still seen as infeasible

4.2 Change in Interpretation

199

in light of the wide-spread impression from lexical semantics that semantic change is fairly unconstrained, as well as the belief that semantics generally is crosslinguistically more invariant than, say, syntax or phonology. Hock (1991, 305) in fact states this quite explicitly: “By and large, semantic change operates in a rather random fashion, affecting one word here (in one way), and another form there (in another way). Given the ‘fuzzy’ nature of meaning, this is of course not surprising. What is surprising is that there should be any instances at all in which semantic change exhibits a certain degree of systematicity. But some such cases can be found.” Grammaticalisation is generally understood as the diachronic process of changing lexical material into functional material. The notion has, wrongly, been associated or indeed equated with the concept of ‘semantic bleaching’, i.e. loss of semantic content.3 The type of grammaticalisation I am concerned with in this book, however, is not of the plain-vanilla variety, under which it is maintained that the semantic change is that from lexical to functional meaning. Rather, we are dealing with secondorder grammaticalisation, since my focus is on inherently functional meanings, i.e., those meanings characterised by high types and permutation invariance (PI) that change into other (and possibly more) functional meanings. In literature, this form of grammaticalisation is also known as secondary grammaticalisation and has been recognised, at least, ever since Kuryłowicz (1965): “Grammaticalization consists in the increase of the range of a morpheme advancing from a lexical to a grammatical or from a less grammatical to a more grammatical status, e.g. from a derivative formant to an inflectional one” (Kuryłowicz 1965, 69). For an overview of theoretical developments in the field of grammaticalisation, see Hopper (1996) and for recent advances in the area of secondary grammaticalisation, see Traugott (2002), Breban (2015) and references therein. I do not ruminate much on this question but simply pick up on, and expand, the idea that functional, as opposed to lexical, meanings have high types, as most notably argued in Chierchia (1984) and Keenan and Stavi (1986), among others (see also Keenan and Faltz 1985 for a boolean perspective or Beghelli 1994 or Keenan and Westerståhl 2011 for a general discussion). From the structure-building perspective, functional and PI items are endocentric—at least those I am concerned with here, namely logical words

3 See,

for instance, Sweetser (1988) and the numerous references therein. I do, however, concede that this notion is also quite an outdated view within mainstream grammaticalisation. For a more recent rectification of grammaticalisation, see Eckardt (2006, 2007, 2011) and references therein. For a case-based account of semantic change, see Eckardt (2006) and references therein; for a historical overview of diachronic semantics, consult Hopper and Traugott. (2003, 19–38), and for a cross-linguistic overview of the grammaticalisation patterns, Heine and Kuteva (2004). Cournane (2014) provides experimental evidence from language acquisition which supports the hypothesis that child learners drive historical change. An overview of various theories of meaning change can be found in Fritz (2012). For a concise overview of diachronic syntax/semantics, I refer the reader to Gianollo et al. (2014).

200

4 Grammaticalisation

like coordinators, quantifiers, and possibly focus, (generalised/Chierchian) exhaustification and question markers. The notion of endocentricity (which I define in 344) entails two beneficial effects, both in syntax as well as semantics. Note that the definition in (344) can also be recast in terms of the labelling principles; see Collins and Stabler (2016, 64, Theorem 12).4 (344)

Endocentricity Each maximal category α max must be a projection of a corresponding minimal category α min , and vice versa, i.e. ∀α[α max ] ⇔ [α min ]

An endocentric functional element does not change the category of its argument (we have also been using the term host(s), which is of equal status in this respect). Among such endocentric elements are sentence modifiers, such as negation markers, which are of type t, t as they take sentences (propositions) and return sentences (propositions), without changing the category or type of their arguments (hosts). We could add to this set of endocentric elements markers of coordination, quantification, and interrogativity, to which I add markers of focus-sensitivity and exhaustification operators—these can uniformly be identified as logical terms.

ψ |ψ

(345)

π , , ¬, Q, FOC, EXH We maintain that syntactically, endocentric elements do not change the categories of their hosts, due to PI. How is this categorial anti-tampering idea implemented in syntax? Everything else being equal, an endocentric element which projects a maximal category (344) simultaneously projects its label upon extension (cf. fn. 4). The projection of the categorial label may be obviated if the relevant endocentric item is syncategorematic (in the syntactic sense of Biberauer et al. 2014 and the semantic sense of Winter 1995). Since, as Biberauer et al. (2014, 202) note, coordination markers in many languages do not appear to c-select specific complements, they therefore cannot be associated with an independent categorial specification (see Cinque (1999) for a syncategorematic treatment of negation). Let me spell out how these two ideas work in concert in formal minimalist terms: logical constants, being PI (see below), are endocentric by virtue of their carrying an intrinsic formal feature (in the sense of Chomsky 1995) specifying their logical constancy as well as syncategorematic since that intrinsic formal feature does not constitute an intrinsic

4 Collins

and Stabler (2016) do not define endocentricity using labelling principles in their published paper, but see the pre-published version for a proof, omitted in publication. Exocentricity, on the hand, is a property of a structure which does not meet or violates (344).

4.2 Change in Interpretation

201

categorial feature. This allows the gross conjunction category to be labelled in line with the coordinands’ categorial feature (with the desirable consequence of yielding, for instance, the category of nominal coordination nominal). This can be implemented using the Labelling Algorithm of Chomsky (2013, 2015) or Cecchetto and Donati (2010); Donati and Cecchetto (2011), which I do not pursue further here. Let me define permutation invariance (PI; also known as automorphism invariance) using two complementary definitions for completeness’ sake. (346)

Permutation (automorphism) invariance a. Given E, F ∈ TYPE 1, 1 is permutation invariant iff for all permutations π of E and all A, B ⊆ E, F (π A)(π B) = F (A)(B). (Keenan and Westerståhl 2011, 869; def. 19.2.3) b. In standard logics, if all models with the same universe interpret an expression d as the same object, then d denotes a permutation invariant element of its denotation set. Thus logical constants always denote permutation invariant elements of their denotation sets. (Keenan and Stavi 1986, 309; prop. 20)

Endocentricity, defined syntactically using the pivotal notion of category, shares logical properties with PI. Additionally, I propose to relate the former two notions with G-triviality, which I define informally in (347a) and formally (originating as Gajewski’s (2008) L(ogical) Analiticity) in (347b). (347)

a. G-triviality [informal] A sentence is G-trivial if its truth value is constant in any situation, no matter how we interpret its lexical material. (Chierchia 2013, 444) b. L-analyticity (Gajewski 2008, 14, Def. 28) [formal] An LF constituent α of type t is L-analytic iff α’s logical skeleton receives the denotation 1 (or 0) under every variable assignment. (Gajewski 2008, 14, Def. 28)

Functional meanings mentioned above and listed in (345)5 therefore have at least the following three properties (von Fintel 1995, 183): (348)

5 Not

a. Functional meaning are permutation-invariant (PI) b. Functional meanings have high types. c. There are universal semantic constrains on possible functional meanings (conservativity).

all functional meanings are the same, so a qualification is in order. For instance, various valency increasing verbal affixes are not PI: e.g. a causative affix (in Turkish, Chichewa, Japanese, etc.) may map intransitive verbs to transitive ones, and transitive ones to ditransitive ones. Equally in Bantu, applicatives would be regarded as functional, qualifying these affixes as substantive, not ‘logical’, in the sense above, and so are not PI.

202

4 Grammaticalisation

I am mostly concerned with (348a–348b). Although von Fintel (1995) does not state it explicity but only states that “an item that becomes a functional morpheme has to assume a higher type”, we are led to conclude that grammaticalisation never results in type-lowering, which is, to the best of my knowledge, empirically borne out. The last property on the list (348c) mentions conservativity: one of the aims of this paper is to consider economy conditions as a possible universal property of semantic variation and change. The hypothesis that serves as conceptual inspiration is the one in (349): (349)

There are universal semantic constrains on possible or probable pathways of change of functional meanings (economy).

These semantic considerations on interpretational change can also be programatically integrated with morphosyntactic aspects of language change, more specifically, with the minimalist diachronic system of Roberts and Roussou (2003). After surveying a rich collection of cross-linguistic constructions, Roberts and Roussou (2003, 212) find a signature common denominator in diachronic syntax in loss of movement and a new exponence of a structurally higher functional head. This head that is formed corresponds to the target slot of movement from a previous stage. In general terms, a formerly lexical head ‘turns into’ () a functional head occupying a structurally superior position, as sketched in (350).     (350) XP Y + X [YP . . . tY . . .]  XP Y = X [YP . . . Y . . .] (Roberts and Roussou 2003, 212n19) Gergel (2009, 261) explicitly proposes that Roberts and Roussou’s (2003) theory of syntactic reanalysis (350) “in fact instantiates a general schema of a semantic counterpart (based on the core of QR) to such syntactic considerations”. This view may be too strong in that it predicts two interlocking facts: high-type elements that covertly QR should end up composing in a high structural position (such as quantifiers), without semantic reconstruction effects, and that grammaticalisation results in higher-types. While the latter seems to be a plausible conjecture (a version of which I submit below), the former may be too strong a claim to maintain (but I leave the discussion on this for future). The structural height, at least within a given categorial spine, can generally be tied to the preceding type-theoretic discussion. I therefore submit the following hypothesis, relating diachronic syntax with type-theoretic compositional semantics. (351)

Diachronic loss of narrow-syntactic movement of an element x at a target site from position y results in x having a higher type since secondary grammaticalisation never results in type-lowering.

To defend (351), I present evidence for a reanalysis of a nominal superparticle μ as a clausal functional head (Focus0 ) in Gothic and Old Irish, along the lines of (350). In the following two subsections, I present two diachronic case studies, buttressing (341), one in which the superparticle system underwent development, and another in which the superparticle system underwent loss.

4.2 Change in Interpretation

203

4.2.2 Developing Superparticles in Japonic Modern Japanese has a very harmonic semantic system of quantificational/logical expression, as we have shown in (7) and (8). We repeat the core patterns in (352) and (353). (352)

The μ-system (353) a. Universal quantifier i. Dare-mo hanashi-ta who-μ talk-PST ‘Every-/anyDE -one talked’ ii. Dono gakusei mo which student μ hanashi-ta talk-PST ‘Every-/anyDE -student talked’ b. Additive marker Mary-mo hanashi-ta Mary-μ talk-PST ‘Also Mary talked’ c. Conjunction marker Mary-mo Bill-mo hanashi-ta Mary-μ Bill-μ talk-PST ‘(Both) Mary and Bill talked’

The κ-system a. Existential quantifier i. Dare-ka hanashi-ta who-κ talk-PST ‘Some-one talked’ ii. Dono gakusei ka which student κ hanashi-ta talk-PST ‘Some student talked’ b. Interrogative marker Hanashi-ta ka talk-PST κ ‘Did you talk?’ c. Disjunction marker Mary-ka Bill-ka hanashi-ta Mary-κ Bill-κ talk-PST ‘(Either) Mary or Bill talked’

The construction of universal, or indeed polar, terms follows the standard pattern of combining a wh-word and the particle mo (μ) and extending to other types of wh-terms. Just as in the previous section, we continue to focus on the μ-series. Compositionally, the semantic role of the μ particle gives rise to a universal reading roughly along the following lines as we have explored in the previous chapter: in the structure [μP μ0 wh ], μ obligatorily activates the alternatives (A) of its complement, i.e., the wh-indeterminates with an existential presupposition, and asserts that all alternatives be true, via recursive exhaustification and, in case of conjunction, a possible silent Junction superstructure which is eventually mapped onto GC. What remains unexplored, however, is the historical dimension of this compositional behaviour in light of the absence of the modern pattern in the earliest stage of the language, which has to the best of knowledge remained unexplored formally. The neat, logically harmonic and linear morphosemantic behaviour of the two series of superparticles μ and κ in (352) and (353) developed over time through change. The original functions of the two particles was not as serial as they are in Modern Japanese.

204

4 Grammaticalisation

This section is devoted to history of Japanese μ (mo) and κ (ka) superparticles. The role of diachronic Japanese in this chapter is to gain comparative-diachronic insight into the nature of change of superparticles. I will propose a view of ‘quantificational split’ for IE, which descriptively explains the distribution of two types of IE languages. One group of IE shows universal quantificational semantics of [wh+μ] expressions, whereas the rest of the IE group shows polarity sensitivity of the same expression. By seeing and understanding what happened in Japonic, we may have independent motivation to contextualise these questions in Sect. 4.2.3. A working assumption, however, is that superparticles undergo common changes, regardless of linguistic genetics, mutatis mutandis.6 A more detailed view of the economy pressures on diachronic semantics is laid out in Sect. 4.3 In this section, we will first review the particle system of OJ and then focus on three construction types: one featuring the ka-particle, another featuring the moparticle, and the third featuring both ka and mo particles, with an aim of accounting for the simultaneous realisation of the two superparticle classes. Note, however, that the ‘super’ attribute was rather absent in OJ since the two kinds of particles, κ and μ, did not feature in cross-categorially and semantically harmonious ways. We will first look into the semantics of the mo (μ) particle, showing that it was an inherently scalar operator. We will then move on to look into the ka (κ) particle, showing that the interrogative/disjunctive signature property, visible in modern Japanese, was absent—instead, ka was a focus marker, which will lend itself to a diachronic syntactic/semantic analysis of how the interrogativity developed in the language. In the last paragraph, we will remark on the syntactic positions of the two particles so as to account for the third type of particle construction, namely the co-occurrence of mo-ka. (A similar counter-intuitive empirical fact was investigated in Sect. 3.5.) Before we engage with any particle in detail, let us make a brief and general remark on the taxonomy of particles in Japanese. Particles are traditionally divided into six types, excluding inflectional endings (Frellesvig 2010, 125): (354)

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

case particles (kaku-joshi) topic and focus particles (kakari-joshi) restrictive particles (fuku-joshi) conjunctional particles (setsuzoku-joshi) final particles (sh¯u-joshi) interjectional particles (kant¯o-joshi)

Our quest for the diachronic origins of the μ (mo) and κ (ka) superparticles will revolve mostly around the ‘topic and focus’ (354c) and final (354e) classes of particles, although we will make reference to conjunctional and restrictive particles as we proceed to the two superparticles. 6 As

a reviewer notes, my case is based on IE and Japonic only and the question, to what extent this view can be maintained, remains to be answered once a wider diachronic and cross-linguistic study is conducted.

4.2 Change in Interpretation

4.2.2.1

205

Developing the μ-System

The oldest text in Japonic dates back to the 8th century CE and allows us to see how the contemporary superparticle μ-system (352) developed. Unlike the contemporary Japanese μ superparticle mo, Old Japanese (OJ) mo did not have the role of performing conjunction, nor encoding negative polarity or freedom of choice. The theory we are relying on in fact predicts such an absence of meanings since the rise of conjunction, additivity, negative polarity, or freedom of choice is precluded in absence of the δ-feature on μ. This is borne out, as this section shows. (Recall that [δ] designates strictly non-scalar and [σ ] strictly scalar alternative sets.) What is more, OJ μ shows that the stipulated [σ ] feature and the general scalar dimension of meaning is empirically motivated. I will first introduce the data in Old Japanese before addressing the changes that occurred in the Classical Japanese period. In the earliest OJ corpus (Man’y¯osh¯u MYS, 8th c.), the [wh+μ] quantificational expressions were confined to inherently scalar (σ ) complements, as first noticed by Whitman (2010). Not only was the polar construction absent from the μ-system in the OJ corpus, but μ0 subcategorised for scalar hosts only. That is, the only eligible hosts of mo were either numeral nominals or inherently scalar wh-terms: how-many and when. The combination of a numeral n and μ, yielded the least likelihood reading along the lines of ‘even n’. In the wh-domain, the μ particle created universal quantificational expressions as shown below. Chierchia’s (2013) system gives us the descriptive power to label this μ as carrying [uσ ] since non-scalar complements were disallowed (recall that the uninterpretable [uσ ] feature captures μ-phrases with only inherently scalar hosts): (355)

itu-mo itu-mo omo-ga kwopi susu when-μ when-μ mother-GEN yearning by ‘I always, always think of my mother [i.e. at all times]’ (MYS, 20.4386; trans. by Vovin 2013, 146)

(356)

sa-ne-si [ywo-no ikuda mo] ara-ne-ba PRE -sleep- PAST [night- SUB how many μ] exist- NEG - COND ‘As there have been few nights in which we slept together . . . ’ (MYS 5.804a, ll. 46–47)

To buttress the fact that only scalar wh-terms were allowed, see Table 4.1 where counts of μ hosts are given. Following Scontras (2013, 548), we take cardinal numerals to be restrictive modifiers in that they compose with predicates and restrict their denotation to those elements with the appropriate cardinality, as indicated in (357a). For further discussion of and arguments for numerals as modifiers (357a), see Scontras (2013), Link (1987), Verkyul (1993), Carpenter (1995), Landman (2003), and those they cite. Conversely, the appropriate wh-term, corresponding to a cardinal numeral, should be a type-equivalent variant with an open cardinality slot (n), as per (357b), where I have generally followed Preuss (2001).

206

4 Grammaticalisation

Table 4.1 Distribution of ±scalar μ-hosts in OJ

(357)

Scalar [wh+μ] itu mo ‘when μ’ iku mo ‘how much/many μ’ Non-scalar [wh+μ] ado/na/nado mo ‘what/why μ’ ika mo ‘how μ’ ta mo ‘who μ’

  a. one = λP λx P (x) ∧ |x| = 1   b. how many = λP λx∃n P (x) ∧ |x| = n

# of attestations Total 23 12 11 Total 0 0 0 0

(Scontras 2013, 548n3a)

The LF of a wh-abstract like how many, then, is inherently scalar insofar as the extension in the answerhood of, say, questions containing such a wh-term (Q: ‘how many x φ?’) is a discretely infinite domain of natural numbers, n ∈ N, from which an answer takes its value (A1 : ‘three’). Hence, the only kind of alternatives such a wh-term has is scalar, which leads us to stipulate that there is a single available possibility for the feature specification of how many, in English, or ikuda, in OJ for that matter, namely [±σ ]. An answer to a how many/much question can also be proportional (A2 : ‘most/many/. . . ’), which we take to be getting its values from a cardinal relationship among subsets of N. Let us then assume two, ontologically interdependent, subkinds of scalar sets, also known as Horn Scales, where the scalar strength (informative prominence) is arranged by height, where ‘higher is stronger’. (358)

Scales for ‘how many’ with (a) discrete and (b) proportional granularity: ⎡

⎤ ∞ ⎢ .. ⎥ ⎢.⎥ ⎢ ⎥ a. ⎢ 3 ⎥ ⎢ ⎥ ⎣2⎦ 1

⎡

⎤ all ⎢ most ⎥ ⎢ ⎥ ⎢ ⎥ b. ⎢much/many⎥ ⎢ ⎥ ⎣ ⎦ few some

In (356), the wh-complex ywono ikuda ‘nights-how-many’ has the denotation paraphrasable as ‘n(-many) nights’, which serves as a host to μ. We maintain an anti-exhaustive semantics for μ, with the scalar EVN-type deriving from EXH, whereby the compositional presence of μ0 activates the alternatives of its (wh-)host. Technically, this is implemented Agree-wise, resulting in positive specification [+σ ] of the scalar feature on the wh-term ikuda. Furthermore, μ asserts that all the alternatives be true. As per the Horn Scale in (358b), recursively exhaustifying over such a scale yields the strongest scalar alternative member all. The strongest scalar alternative thus entails all the weaker members since ‘Mary saw John on all nights’ entails that ‘Mary saw John on some nights’, or using actual numbers, if

4.2 Change in Interpretation

207

‘Mary saw John on three occasions’, then it is logically true that ‘Mary saw John on two occasions’. Therefore, the entire μP ywodo ikuda mo is predicted to mean something along the lines of ‘∀n-many-nights φ’, where φ is a shorthand notation for the rest of the sentence; or in more natural terms, ‘for all, or very many, counts (or amounts) of nights (i.e. all nights), φ’. The presence of negation, seen as the sentence-final existential inflection (ara)-neba7 obtains a SI via negation of the strongest scalar alternative member. This delivers a denotation ‘not for all counts of nights φ’, which infers ‘for some counts of nights φ’, naturally translating into ‘few (counts of) nights’, being on a par with ‘not (very) many nights’. Such SI also obtains in English, where (359b-i) and (359b-ii) go hand in hand. (359)

a. Mary saw John on many occasions. b. i. Mary did not see John on all/many occasions. ii. Mary saw John on some/few occasions.

This brings us to another matter: we have assumed that the role of μ is twofold: to activate the alternatives of its host and, via recursive exhaustification, asserts that they all be true. For (356), I have assumed that in absence of negation, the μP would be universal (‘for all nights’) by virtue of EXHR . If we were on the right track, then the negation of a universal term would yield an existential term as a SI: (360)

¬∀∃

In (356), however, the resulting quantificational interpretation is not ‘some’ but ‘few’, which obtains under negation of ‘many’: (361)

¬ many  few a. Ξ1σ = all, some b. Ξ2σ = many, few

This notion of variability in size (truncatability) of the scale is analogous to a solution of Abrusán (2014, 127), who adopts scalar truncation to derive the correct interpretation for degree interrogatives. In Abrusán (2014), a truncated scale is taken to be a “scale from which an initial segment of the lexically determined scale [. . . ] has been removed.” She follows and minimally revises the proposal by Rett (2007), who proposes an operator EVAL, whose meaning is essentially a function from sets of degrees to sets of degrees. Once applied to a degree d, EVAL returns D , where D ⊂ d and which consists of all and only those members of d, which are higher than a certain contextually determined degree. As Abrusán (2014) notes, the EVAL operator is essentially a truncation operator. Abrusán (2014) upgrades the proposal by assuming that the truncation operator, essentially  Rett’s(2007) EVAL at its core, is presuppositional. This truncation operator, T λd[φ(d)] , which we adopt from 7 We

are ignoring the conditional semantic import of the negative-conditional inflection and limit our analysis to the negative component alone.

208

4 Grammaticalisation

Abrusán (2014), is defined so that the resulting denotation after applying T to a degree, which is below the threshold s, is undefined, making T a partialisation of φ, which is not a propositional formula in Abrusán’s (2014) terms. T therefore turns a function that is defined for a given scale into a function that is defined only for a proper subpart of the same scale (the segment above s), and is otherwise identical (Abrusán 2014, 127). In formal terms, we define T below as a two-place operator that combines with a scale (Ξ σ ) supplied by the proposition (φ) and a scalar threshold (s), which demarcates the truncation. (362)

T(runcation) operator (Abrusán 128, ex.83):  2014,   T = λPd,e,t λxd : x > s λye P (x)(y)

The purpose of having T is to get us from a total Horn Scale to its subsets Ξ1σ (361a) and Ξ2σ (361b). The scalar semantic form of a wh-abstract like how many provides a scale of nmany nights, i.e. scalar variables provided by the denotation of the complement. If we maintain our semantics for μ, we assume that the compositional presence of μ activates and asserts the truth of all scalar alternatives to ‘n-many-nights’. The presence of negation negates the universal-scalar term with the denotation ‘for all n, n-many nights φ’ (where φ denotes the rest of the sentence so as to allow for proposition-level alternatives), yielding a denotation ‘not for all n, n-many nights φ’, which carries a scalar implicature and delivers the correct interpretation ‘few nights’, i.e. ‘not (very) many’. Therefore, OJ quantificational μPs are scalar [wh+μ] constructions. As such, they are analysed as universals since they give rise to a scalar implicature under negation, ‘not all nights’ implying ‘few days’, delivering as the denotation the weakest member of the scale all, many, few . In (364), we give the full composition of (363). We assume that the noun ywono ‘nights’ simply denotes a predicate, which is the argument of the cardinal wh-word ikuda ‘how many’, itself assumed to be a function from (sets of) properties to cardinalities (of those sets). The resulting phrase, containing ywono ikuda ‘how many nights’, is assumed to deliver a property- or set-denoting λ-abstract (Groenendijk and Stokhof 1983) over the cardinality of set, whose extension is a scalar domain of natural numbers or quantity expressions, i.e. ywono ikuda ‘how many nights’σ A = N. The wh-phrase now combines with the μ particle, whose semantic role it is to activate scalar alternatives and induce exhaustification. The resulting set, ywono ikuda mo ‘how many nights μ’ denotes an entire set of scalar alternatives (364a-b). Let me also repeat the relevant datum: (363)

(364)

sa-ne-si [ywo-no ikuda mo] ara-ne-ba PRE -sleep- PAST [night- SUB how many μ] exist- NEG - COND ‘As there have been few nights in which we slept together . . . ’ (MYS 5.804a, ll. 46–47) a. ywono ikuda mo ‘how-manyμ nights’σ A =

4.2 Change in Interpretation

209

  a. LF: EXHR [σ A] . . . [wh-μ[σ ],i . . . [nights ti ]] b. Alternatives:  x ∈ DD | NIGHT(x) ∧ #(NIGHT(x)) = n] : n ∈ N c. EXHR [1 ∨ 2 ∨ . . . n] = [¬EXH(1) ∧ ¬EXH(2) ∧ . . .] = [1 ∧ 2 ∧ . . . ∧ n] = MAX(n)[#(NIGHTS(x)) = n] # ∀n[#(NIGHTS(x)) = n] The remaining part of the sentence in (363) which is relevant for our computation is the negative (verbal) morpheme ne. The contribution of negation is assumed to negate the denotation of the μP, i.e., the universal expression, paraphrasable as ‘for all n, n-many nights’. The resulting denotation carries a scalar implicature, namely ‘not for all n, n-many nights . . . ’, which delivers a SI through negation of the strongest member of the Horn scale. Note the analogy of the OJ type with English SIs as in ‘not all nights’  ‘some nights’ but, crucially, ‘no nights’ does not obtain. The universal quantification is additionally borne out in positive contexts (365) when, in absence of negation, a universal temporal term (‘always’) obtains, i.e. the strongest member of the scale always, sometimes, rarely with an existential scale (cf. English ‘not always’  ‘never’ but  ‘sometimes/rarely’). Note that the polarity system associated with [whσ +μ] expressions is absent in the earliest stage of recorded Japonic, as are other wh-universal terms we find in MdJ. When not embedded under negation, the scalar μ + whσ terms are universal, as shown in (365). (365)

itu-mo itu-mo omo-ga kwopi susu when-μ when-μ mother-GEN yearning by ‘I always, always think of my mother [i.e. at all times]’ (MYS, 20.4386; trans. by Vovin 2013, 146)

As John Whitman (p.c.) informs me, the repetition of phrases may bring about pluralisation or quantification, hence we must investigate the distribution of universal quantificational labour between the μ particle and the repetition/doubling of the μP. Vovin (2005, 107) shows that reduplication is not a productive means of expressing universal quantification since not all nouns can have a reduplicated form. In fact, the list of reduplicative nouns is rather short and includes the following:8 (366)

8I

Reduplicative nouns in OJ (Vovin 2005, 107): a. ka-Nka ‘(all) days’ (