Micro-change and Macro-change in Diachronic Syntax (Oxford Studies in Diachronic and Historical Linguistics) [1 ed.] 9780198747840, 0198747845

Citation preview

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

Micro-change and Macro-change in Diachronic Syntax

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

OX F O R D S T U D I E S I N D IAC H R O N IC A N D H I S T O R I C A L L I N G U I S T I C S general editors: Adam Ledgeway and Ian Roberts, University of Cambridge advisory editors: Cynthia Allen, Australian National University; Ricardo Bermúdez-Otero, University of Manchester; Theresa Biberauer, University of Cambridge; Charlotte Galves, University of Campinas; Geoff Horrocks, University of Cambridge; Paul Kiparsky, Stanford University; Anthony Kroch, University of Pennsylvania; David Lightfoot, Georgetown University; Giuseppe Longobardi, University of York; George Walkden, University of Konstanz; David Willis, University of Cambridge recently published in the series 16 Syllable and Segment in Latin Ranjan Sen 17 Participles in Rigvedic Sanskrit The Syntax and Semantics of Adjectival Verb Forms John J. Lowe 18 Verb Movement and Clause Structure in Old Romanian Virginia Hill and Gabriela Alboiu 19 The Syntax of Old Romanian Edited by Gabriela Panˇa Dindelegan 20 Grammaticalization and the Rise of Configurationality in Indo-Aryan Uta Reinöhl 21 The Rise and Fall of Ergativity in Aramaic Cycles of Alignment Change Eleanor Coghill 22 Portuguese Relative Clauses in Synchrony and Diachrony Adriana Cardoso 23 Micro-change and Macro-change in Diachronic Syntax Edited by Éric Mathieu and Robert Truswell For a complete list of titles published and in preparation for the series, see pp. 320–2.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

Micro-change and Macro-change in Diachronic Syntax Edited by É R IC M AT H I E U A N D R O B E RT T RU S W E L L

1 i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

3

Great Clarendon Street, Oxford, ox2 6dp, United Kingdom Oxford University Press is a department of the University of Oxford. It furthers the University’s objective of excellence in research, scholarship, and education by publishing worldwide. Oxford is a registered trade mark of Oxford University Press in the UK and in certain other countries © editorial matter and organization Éric Mathieu and Robert Truswell 2017 © the chapters their several contributors 2017 The moral rights of the authors have been asserted First Edition published in 2017 Impression: 1 All rights reserved. No part of this publication may be reproduced, stored in a retrieval system, or transmitted, in any form or by any means, without the prior permission in writing of Oxford University Press, or as expressly permitted by law, by licence or under terms agreed with the appropriate reprographics rights organization. Enquiries concerning reproduction outside the scope of the above should be sent to the Rights Department, Oxford University Press, at the address above You must not circulate this work in any other form and you must impose this same condition on any acquirer Published in the United States of America by Oxford University Press 198 Madison Avenue, New York, NY 10016, United States of America British Library Cataloguing in Publication Data Data available Library of Congress Control Number: 2016961203 ISBN 978–0–19–874784–0 Printed in Great Britain by CPI Group (UK) Ltd, Croydon, CR0 4YY Links to third party websites are provided by Oxford in good faith and for information only. Oxford disclaims any responsibility for the materials contained in any third party website referenced in this work.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

Contents Series preface Acknowledgements List of figures and tables List of abbreviations List of contributors . Micro-change and macro-change in diachronic syntax Éric Mathieu and Robert Truswell . Syntactic theory and syntactic change . Overview of the volume . In defence of the child innovator Ailís Cournane . . . .

Introduction The survival and diffusion of L input-divergent properties Unidirectionality of diachronic pathways and the mapping problem Conclusion

. Where do relative specifiers come from? Nikolas Gisborne and Robert Truswell . . . . .

Introduction Relative specifiers in English and elsewhere Relative specifiers and noun phrase accessibility Unsystemic change Conclusion

. Diachronic interpretations of word order parameter cohesion John Whitman and Yohei Ono . Introduction . Crosscategorial word order generalizations . Reanalyses that propagate crosscategorial word order regularities . Statistical analysis . Discussion and diachronic interpretation . The rise and fall of Hungarian complex tenses Katalin É. Kiss . Goals . Complex tenses in Old Hungarian . The traditional view of the origin of complex tenses: Latin influence . An alternative explanation: Old Turkic contact effect . The evolution of Hungarian complex tenses

ix x xi xiv xx                          

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

vi

Contents . The emergence of situation aspect marking . The fall of complex tenses . Theoretical implications

. Modelling transient states in language change Gertjan Postma . . . . . .

Introduction Three-state models The logistic model The three-state model Competing grammars approach Conclusions

. Modelling interactions between morphosyntactic changes Hezekiah Akiva Bacovcin . Introduction . Failed changes and the Constant Rate Hypothesis . English dative to . A blocking model of failed change and probability multiplication . Conclusions . From Latin to Modern French: A punctuated shift Michelle Troberg and Heather Burnett . Introduction . Early Latin as a satellite-framed language . Modern French as a verb-framed language . From satellite-framed to verb-framed: The case of Medieval French (– ce) . Two reanalyses involving Path . Theoretical implications . Case in diachrony: Or, why Greek is not English Nikolaos Lavidas . Introduction . Case, aspect, and definiteness, and their interrelation . Case, aspect, and definiteness: A comparison of the diachrony of Greek and English . Conclusions . Leftward Stylistic Displacement (LSD) in Medieval French Marie Labelle and Paul Hirschbühler . Leftward Stylistic Displacement (LSD) . Mathieu’s analysis . Informational role of the fronted constituent . Intervention effects . Subject condition . VP and remnant VP movement

                                  

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

Contents . Positions of the LSD constituent . Summary and Conclusion

vii  

. Diagnosing embedded V in Old English and Old French Christine Meklenborg Salvesen and George Walkden



. Introduction . Types and analyses of V in modern and historically attested languages . Investigating complement clauses . Results and discussion . Conclusion



. The pragmatics of demonstratives in Germanic Caitlin Light . Introduction . A puzzle in historical English . Corpus data on syntax and information structure . Types of topicalization . The special status of demonstratives . Analysing demonstratives . Conclusion . Persistence as a diagnostic of grammatical status: The case of Middle English negation Aaron Ecay and Meredith Tamminga . Introduction . Background: Persistence . Background: Negation . Results . Conclusion

                 

. The origins of the Romance analytic passive: Evidence from word order Lieven Danckaert . Introduction: Passives from Latin to Romance . The development of Latin BE-periphrases: The state of the art . Some overlooked evidence: Word order . Towards an analysis of the word order facts . An interesting parallel: The genesis of the Romance synthetic future . Conclusion



. Reconciling syntactic and post-syntactic complementizer agreement Sarah G. Courtney . Introduction . Complementizer agreement as a PF interface phenomenon . Complementizer agreement in the narrow syntax . Complementizer agreement and microvariation . Complementizer agreement and grammaticalization . Conclusion



     

     

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

viii

Contents

. On the grammaticalization of temporal–aspectual heads: The case of German versprechen ‘promise’ Łukasz Jędrzejowski . . . . .

Introduction: Versprechen in present-day German Selected differences between v and v Versprechen as a functional verbal head Reanalysis Conclusion

References Index

       

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

Series preface Modern diachronic linguistics has important contacts with other subdisciplines, notably first-language acquisition, learnability theory, computational linguistics, sociolinguistics, and the traditional philological study of texts. It is now recognized in the wider field that diachronic linguistics can make a novel contribution to linguistic theory, to historical linguistics, and arguably to cognitive science more widely. This series provides a forum for work in both diachronic and historical linguistics, including work on change in grammar, sound, and meaning within and across languages; synchronic studies of languages in the past; and descriptive histories of one or more languages. It is intended to reflect and encourage the links between these subjects and fields such as those mentioned above. The goal of the series is to publish high-quality monographs and collections of papers in diachronic linguistics generally, i.e. studies focussing on change in linguistic structure, and/or change in grammars, which are also intended to make a contribution to linguistic theory, by developing and adopting a current theoretical model, by raising wider questions concerning the nature of language change or by developing theoretical connections with other areas of linguistics and cognitive science as listed above. There is no bias towards a particular language or language family, or towards a particular theoretical framework; work in all theoretical frameworks, and work based on the descriptive tradition of language typology, as well as quantitatively based work using theoretical ideas, also feature in the series. Adam Ledgeway and Ian Roberts University of Cambridge

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

Acknowledgements The papers in this volume were presented at the 15th Diachronic Generative Syntax conference, at the University of Ottawa, August 2013. Our thanks to the conference participants and graduate volunteers (particularly Nova Starr), as well as to Ottawa’s Faculty of Arts for financial support. We gratefully acknowledge the input of a large team of anonymous reviewers, including several dozen reviewers of the original conference submissions and approximately twenty reviewers of the chapters in this volume. Finally, thanks to Vicki Sunter and Julia Steer for their patience and support as this volume lurched towards completion.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

List of figures and tables Figures 3.1. Demonstrative relatives and wh-relatives with locative gaps, as a proportion of all relatives with locative gaps, over time

34

3.2. Demonstrative relatives and wh-relatives with argument gaps, as a proportion of all relatives with argument gaps, over time

35

4.1. Scree plot of eigenvalues from MCA of the WALS dataset

51

4.2. First and second dimension coordinates of 450 feature value categories generated by MCA

52

4.3. First and third dimension coordinates of 450 feature value categories generated by MCA

53

4.4. First and third dimension coordinates of 450 feature value categories generated by MCA (features 81A, 83A, 85A, 86A, 87A, 88A, and 89A only)

53

4.5. First and second dimension coordinates of 201 languages generated by MCA

55

7.1. A logistic curve and its first derivative

96

7.2. Graph of English to data with the fits from the probability multiplication model

98

7.3. Best fits from the first derivative and probability multiplication models for the recipient–theme data

100

7.4. Simulation results showing the statistical power in the English ditransitive data

102

8.1. Replacement of avant/arrière by en avant/arrière

118

10.1. Percentage of null subjects, and of LSD of nonfinite V and adjectives/predicates/quantifiers in XV embedded clauses, prose texts only

154

10.2. Percentage LSD in Subject Relatives

154

10.3. Percentage LSD in thematic null subject embedded clauses

155

12.1. Structure of the discourse tree (Büring 2003)

196

12.2. Implicit moves in the d-tree (Büring 2003)

196

13.1. Results from Estival (1985), showing that English lexical and transformational passives prime themselves, but not each other, in conversational speech

205

13.2. The change from ne to not in Middle English, as measured in negative declarative sentences in the PPCME2

207

13.3. The analysis of three stages in the evolution of English negation, from Frisch (1997)

208

13.4. The analysis of three stages in the evolution of English negation, from Wallage (2008)

209

13.5. The predictions of the two-atom model measured in the corpus data

212

13.6. The predictions of the three-atom model measured in the corpus data

212

13.7. Priming predictions and actual results of the two models

213

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

xii

List of figures and tables

13.8. Structure with ne as sentence negation in combination with not or another reinforcing adverb 13.9. The three-way priming results, after applying Frisch’s divide-by-0.16 adjustment for adverbial not 13.10. Priming effects on ME negation in the period 1350–1400

213 214 214

14.1. Relative frequency (in percentages) of the order ‘possum–infinitive’ over time

223

14.2. Relative frequency (in percentages) of the order ‘debeo–infinitive’ over time

224

14.3. Relative frequency (in percentages) of the order ‘BE–past participle’ in passive and deponent E-periphrases over time

225

14.4. Relative frequency (in percentages) of the order ‘BE–past participle’ in passive and deponent F-periphrases over time

226

14.5. Relative frequency (in percentages) of deponent and passive F-periphrases (as compared to E-periphrases) over time

227

14.6. Relative frequency (in percentages) of the order ‘Aux–PaPa’ in E-periphrases over time, passives and deponents compared

229

15.1. Feature-sharing between T and C via head movement

242

15.2. A possible syntactic structure for the sentence in (22)

250

15.3. An alternative syntactic structure for the sentence in (22)

250

15.4. Movement of a pronoun to Spec,TP

251

15.5. Cliticization of a pronoun to C

251

15.6. Reanalysis of a pronominal clitic as an output of agreement

251

16.1. Base positions of v1 and v2 in present-day German

262

16.2. v2 as Aux head and its merge position according to Wurmbrand (2001)

264

16.3. The development of versprechen as a subject raising verb

276

16.4. The grammaticalization of versprechen

277

Tables 2.1. Summary of proposed alignments between L1A and change

23

3.1. Dependent relative specifiers in 172 languages (based on De Vries 2002)

29

4.1. Sources for Chinese adpositions

47

4.2. Classification accuracies for Cluster A

56

4.3. Classification accuracies for Cluster B

57

4.4. Classification accuracies for Cluster C

57

4.5. Classification accuracies for Cluster D

58

7.1. Comparison of Heavy NP Shift models using Likelihood Ratio, AIC, and BIC

99

7.2. AIC values for the different models of the recipient–theme failed change

100

7.3. Example outputs of the three grammars in competition in the Middle and Modern English period

101

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

List of figures and tables

xiii

7.4. Likelihood Ratio and AIC comparisons between different parameterizations of the probability multiplication model

102

8.1. Locative prepositions and directional/aspectual prefixes in Latin

107

8.2. Medieval French aspectual prefixes and their prepositional counterparts

110

8.3. Place–Path meaning alternation in prepositions in Medieval French

113

8.4. Evolution of resultative secondary predication from Latin to Modern French

116

8.5. Replacement of directional avant by directional uses of en avant (by date of author’s birth)

117

8.6. Replacement of directional arrière by directional uses of en arrière (by date of author’s birth)

117

9.1. Archaic vs. Pre-Koine Greek vs. Post-Koine Greek system: the aspectual features

135

9.2. English vs. Greek: Case, definiteness, and aspect in diachrony

136

9.3. Archaic vs. Pre-Koine Greek vs. Post-Koine Greek system: case and aspect

141

9.4. Overview of the factors affecting the accusative subject examples

144

10.1. Fronted infinitives, participles, and small adverbs

155

10.2. SXV vs. XSV in the parsed corpus

156

10.3. Percentage of VO word order w.r.t. OV order in LSD with a participle, in cases where the fronted (remnant) VP contains more than a head

159

11.1.

175

Embedded V2 in OE and OF by class of embedding predicate: overall results

11.2. Embedded V2 in OE and OF by class of embedding predicate: only non-accidental V2

176

12.1. Topicalization of different DP types in the IcePaHC corpus

192

12.2. Topicalization of different DP types in the PPCEME corpus

192

13.1.

Counts in each prime/target position of our primary negation dataset

14.1. Passives in Latin and Romance (indicatives)

211 217

14.2. E- and F-periphrases in passives: the perfectum

218

14.3. Passives in the infectum

219

14.4. Frequency of the orders ‘PaPa–Neg–Aux’ and ‘Neg–PaPa–Aux’ in BE-periphrases of the E-type (only authors/texts with at least four negated E-periphrases included)

232

16.1. Selected differences between the lexical and functional uses of versprechen

262

16.2. Subject-to-subject raising verbs in present-day German according to Wurmbrand (2001)

265

16.3. Grammaticalization of v2 according to Heine and Miyashita (2004)

271

16.4. The use of versprechen in the GerManC corpus

274

16.5. The use of versprechen in the BFK corpus

274

16.6. The use of versprechen in the GOE corpus

275

16.7. The use of versprechen in the KHZ corpus

276

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

List of abbreviations 1

1st person

2

2nd person

2L1

two first languages

3

3rd person

AAVE

African American Vernacular English

ABL

ablative

ACC

accusative

ACT

active voice

Act. apost.

Actus apostolorum

Adj

adjective

ADV

adverbial

Æhom

Ælfric, Homilies

Agr

agreement

AgrO

object agreement

AgrOP

object agreement phrase

AgrS

subject agreement

AgrSP

subject agreement phrase

AH

accessibility hierarchy

AIC

Akaike information criterion

Ap.

Apology

APPL

applicative

ART

article

ARTFL

American and French research on the Treasury of the French Language

Asp

aspect

AspP

aspectual phrase

ATILF

analyse et traitement informatique de la langue française

Aug.

Augustinus

AUX

auxiliary

BFK

Bonner Frühneuhochdeutschkorpus

BFM

base du français médiéval

BIC

Bayesian information criterion

C

complementizer

CA

complementizer agreement

Cas.

Casina

CD

complementizer doubling

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

List of abbreviations Chev. papegau

Le Chevalier du papegau

Chron.

Chronique

Chron. Scand.

La Chronique scandaleuse

CIA

child innovator approach

Cic.

Cicero

CL

clitic

COND

conditional

Contr

contrast

CP

complementizer phrase

CPR

Corpus Papyrorum Raineri (Wessely 1895)

CRH

Constant Rate Hypothesis

CT

contrastive topic

D.

Demosthenes

D

determiner

d-tree

discourse tree

DA

double agreement

DAT

dative

DE

differential equation

De vera relig.

de Vera Religione

Deg

degree

DEM

demonstrative

DeReKo

Deutsches Referenzkorpus

Dev

deviance

Df

degrees of freedom

DiGS

Diachronic Generative Syntax

Dir

directional

DirP

directional phrase

DMF

dictionnaire du moyen français

DO

direct object

DP

determiner phrase

DUR

durative

EME

Early Modern English

Emph

emphasis

EP

European Portuguese

EPP

extended projection principle

F

focus

F

foot (ch. 14)

F

feminine

Foc

focus

xv

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

xvi

List of abbreviations

FocP

focus phrase

FM

formal movement

FOFC

final-over-final constraint

FP

functional projection

FUT

future tense

FUTPRF

future perfect tense

GDV

gerundive

GEN

genitive

GER

gerund

GOE

Goethe-Korpus

Grg.

Gorgias

Hom.

Homer

HT

hanging topic

i-

interpretable

I

inflection

IcePaHC

Icelandic parsed historical corpus

IE

Indo-European

Il.

Iliad

IMP

imperative

INF

infinitive

infl

inflection

INS

instrumental

Int

interrogative

INTR

intransitive

IO

indirect object

IP

inflectional phrase

IPFV

imperfective

JC

Jespersen’s cycle

KHZ

Mannheimer Korpus Historischer Zeitungen

L1

first language

L1A

first language acquisition

L2

second language

Lay.

Layamon

LD

left-dislocation

LF

Logical Form

Liv.

Titus Livius

LOC

locative

LSD

Leftward Stylistic Displacement

Lucr.

Lucretius

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

List of abbreviations M

masculine

MCA

multiple correspondence analysis

MCVF

modéliser le changement: les voies du français

ME

Middle English

MHG

Middle High German

MOD

modal

ModG

Modern German

ModP

modality phrase

MP

mediopassive

N

neuter

N

noun

Nat.

Historia Naturalis

NEG

negation

NegP

negation phrase

NOM

nominative

NP

noun phrase

NPI

negative polarity item

O

object

Obj

object

OBL

oblique

OC

Old Chinese

OCOMP

object of comparison

Od.

Odyssey

OE

Old English

OF

Old French

OHG

Old High German

Op

operator

OSp

Old Spanish

P

preposition

P&P

Principles and Parameters

P. Flor.

Papiri Fiorentini (Comparetti and Vitelli 1906–15)

P. Par.

Les papyrus grecs du musée du Louvre et de la Bibliothèque Impériale (Jean et al. 1866)

p. ptc.

past participle

PaPa

past participle

PART

partitive

PASS

passive

PathP

path phrase

PCA

principal component analysis

xvii

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

xviii

List of abbreviations

PDE

present-day English

Percef.

Perceforest

PF

Phonological Form

PFV

perfective

ϕ

phonological phrase

Phorm.

Phormio

PIE

Proto-Indo-European

PL

plural

Pl.

Plato

PlaceP

place phrase

Plaut.

Plautus

PLC

Penn Linguistics Colloquium

PLD

primary linguistic data

Plin.

C. Plinius Secundus

POSS

possessor

PP

prepositional phrase

PPCEME

Penn–Helsinki parsed corpus of Early Modern English

PPCME2

Penn–Helsinki parsed corpus of Middle English, 2nd edition

Pr.

probability

PredP

predication phrase

PRET

preterite

PRO

personal pronoun

PRS

present

PRT

particle

PST

past tense

PTCP

participle

QMIII

quantification method III

QUD

question under discussion

Quinct.

pro Quinctio

REFL

reflexive

REL

relative

Resid.

residual

S

subject

SBJV

subjunctive

SF

stylistic fronting

SFX

suffix

SG

singular

SIR

susceptible–infected–removed

Spec

specifier

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

List of abbreviations ST

scenic topic

SU

subject

Subj

subject

SubjP

subject phrase

t

trace

T

tense

TAB

true A-bar movement

Ter.

Terence

TFA

textes du français ancien

Th.

Thucydides

Top

topic

TopP

topic phrase

TP

tense phrase

TR

transitive

Trans

transitivity

TransP

transitivity phrase

u-

uninterpretable

V

verb

V1

verb-first

V2

verb-second

V3

verb-third

VP

verb phrase

ω

prosodic word

WALS

World Atlas of Language Structures

YCOE

York–Toronto–Helsinki parsed corpus of Old English prose

xix

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

List of contributors Hezekiah Akiva Bacovcin is a Ph.D student at the University of Pennsylvania. He primarily works on the interaction between statistical models of language variation and theories of language cognition. His dissertation, focusing on modelling variation in the syntax of ditransitive in Germanic languages, is titled Parameterising Germanic ditransitive variation: A historical– comparative study. Heather Burnett is a CNRS researcher in the Laboratoire de Linguistique Formelle at l’Université Paris 7-Denis Diderot. In 2012, she completed her Ph.D thesis in the Department of Linguistics at the University of California, Los Angeles under the supervision of Edward Keenan and Dominique Sportiche. From 2012 to 2014, she was a SSHRC postdoctoral fellow at l’Université de Montréal; from 2014 to 2015, she was a CNRS postdoctoral researcher at l’Université de Toulouse 2-Jean Jaurès; and in 2015, she was a Banting postdoctoral fellow at the University of Toronto. Her work on formal semantics, formal syntax, and language variation and change has appeared in Linguistic Variation, Linguistics and Philosophy, and the Journal of Semantics. Ailís Cournane is an Assistant Professor at New York University. She holds a Ph.D in Linguistics from the University of Toronto. Her research is about how children learn what functional words mean, and how this learning process may relate to the way languages change over time. She is particularly interested in variation and change of the modal system within and across languages. At the core of her research is an interest in uncovering what patterns of development reveal about how meaning is mapped onto grammatical forms. Sarah G. Courtney is a graduate student in linguistics at Cornell University. She is writing a dissertation on the cross-linguistic syntax of complementizer agreement, focusing on its variation and diachronic development. Lieven Danckaert graduated in Classical Philology at Ghent University in 2006, and went on to obtain a Ph.D degree at the same institute, with a thesis on the syntax of the left periphery in Latin. He is currently working as a researcher (chargé de recherche) at the CNRS/University of Lille, and he recently finished a book on the syntax and diachronic development of the Latin verb phrase (The Development of Latin Clause Structure: A Study of the Extended Verb Phrase, OUP, 2017). Aaron Ecay is a postdoctoral research associate at the Centre for Language History and Diversity in the Department of Language and Linguistic Science at the University of York. He received his Ph.D in 2015 from the University of Pennsylvania. His primary research interest is in the field of historical syntax. Specifically, he is interested in investigating how diachronic patterns of language use can answer questions about the architecture of synchronic grammar. He is also interested in the refinement of statistical methods as applied to linguistic data and syntactic questions more generally.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

List of contributors

xxi

Nikolas Gisborne teaches in the department of Linguistics and English Language at the University of Edinburgh. He received his Ph.D in Linguistics from University College London in 1996 and is the author of The Event Structure of Perception Verbs (Oxford University Press, 2010). Paul Hirschbühler is a retired professor from the Université d’Ottawa. After graduating with a Licence en Philosophie et Lettres (section Philologie Romane) from the Université Libre de Bruxelles (1970), he earned his Ph.D at the University of Massachusetts in Amherst (1978). His research interests include French synchronic and diachronic syntax, comparative syntax, and cross-linguistic differences in the expression of arguments, especially in the locative alternation. Łukasz Jdrzejowski is a Research Associate at the University of Potsdam, where his research focuses on theoretical diachronic syntax and the historical linguistics of Germanic and Slavic languages. Between 2010 and 2013 he was at the Zentrum für Allgemeine Sprachwissenschaft in Berlin, working in the project Lexical conditioning of syntactic structures: Clause-embedding predicates. In 2015, he defended his Ph.D thesis on subject-to-subject raising verbs in the history of German. He is currently working on adverbial clauses in diachronic West Germanic. Katalin É. Kiss is Professor at the Research Institute for Linguistics of the Hungarian Academy of Sciences and at Pázmány Péter Catholic University, where she is also head of the Doctoral School in Linguistics. Her publications include The Syntax of Hungarian (Cambridge University Press, 2002), Discourse-Configurational Languages (Oxford University Press, 1995), Event Structure and the Left Periphery (Springer, 2006), Adverbs and Adverbial Adjuncts at the Interfaces (de Gruyter, 2009), and The Evoution of Functional Left Peripheries in Hungarian Syntax (Oxford University Press, 2014). Marie Labelle is professor of linguistics at the Université du Québec à Montréal. She received her Ph.D at the Université d’Ottawa in 1989. Her research interests include French syntax (synchronic and diachronic), psycholinguistics (particularly first language acquisition and child second language acquisition), the semantics of tense and aspect, the mapping of semantic arguments to syntax, and negation. Nikolaos Lavidas is Assistant Professor of Historical Linguistics at the Department of Theoretical and Applied Linguistics, School of English, Aristotle University of Thessaloniki. His research interests include (Proto-)Indo-European syntax and syntactic change, in particular, the development of transitivity and voice in Indo-European languages as well as the role of language contact. He is co-editor of Proto-Indo-European Syntax and its Development (2015: Benjamins Current Topics 75, 2013: as a special issue in the Journal of Historical Linguistics) and Typology of Labile Verbs: Focus on Diachrony (2014: as a special issue in Linguistics). Caitlin Light is a Lecturer in English Linguistics at the University of York. Her research interests include variation and change, syntax, and information structure, with a particular focus on the Germanic language family and the history of English. She received her Ph.D from the University of Pennsylvania in 2012. Éric Mathieu is Associate Professor at the University of Ottawa. He completed his Ph.D in 2002 at University College London. His research focuses on French (Modern and Old) and Ojibwe (an Algonquian language). He has published in Linguistic Inquiry, Natural Language &

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

xxii

List of contributors

Linguistic Theory, Syntax, International Journal of American Linguistics, Journal of Linguistics, Lingua, Probus, Studia Linguistica, and Linguistic Variation. He is also the author of numerous chapters in books (with Oxford University Press, Cambridge University Press, John Benjamins, Routledge, and Springer, among others) and co-author of several books; a monograph on island effects entitled The Syntax and Semantics of Split Constructions, a special edition of Lingua on noun incorporation, and an edited volume on Romance languages Variation Across and Within Languages. Yohei Ono is visiting research fellow of the Institute of Statistical Mathematics in Tokyo, Japan. His main interest is the application of statistical methods to linguistic data, especially in stylometry, dialectology, and typology. Gertjan Postma is senior researcher in variational linguistics at the Meertens Institute in Amsterdam, a research institute of the Netherlands’ Academy of Sciences. He specializes in quantitative and qualitative models in diachronic syntax, especially formal properties, such as negation, pronominalization, and inalienable possession. He built an automatic parser of Middle Dutch, and made an annotated historical corpus (Low Saxon dialect, 1400–1500). In recent years, he has been investigating language change in Dutch and Low Saxon language islands in Brazil, and is currently writing a contrastive grammar of Brazilian Pomeranian, spoken in Espirito Santo. Christine Meklenborg Salvesen is Associate Professor at the Department of Literature, Area Studies, and European Languages at the University of Oslo. She is currently directing the Traces of History project, which compares verb-second word order in Germanic and Romance languages in a diachronic perspective. Her research pivots around diachronic French syntax. Salvesen has edited several books and is a member of the editorial board of Oslo Studies in Language. Meredith Tamminga is an Assistant Professor in the Department of Linguistics at the University of Pennsylvania, where she is also director of the Language Variation and Cognition Lab. She received her BA from McGill University and her Ph.D from the University of Pennsylvania. Her research integrates experimental psycholinguistic methods, computational modelling, and the quantitative analysis of natural speech data to learn how speakers process, store, and produce linguistic variation. Michelle Troberg is an Assistant Professor, Teaching, at the University of Toronto Mississauga. She received her M.A. and Ph.D in French Linguistics from the University of Toronto and is also an ancienne pensionnaire of the École Normale Supérieure (Paris). Her research and teaching interests involve comparative Romance syntax, particularly from a diachronic perspective, treating such topics as clitic doubling, resultative secondary predicate constructions, and various other types of argument realization involving the verbal complement. Robert Truswell is a Chancellor’s Fellow in the school of Philosophy, Psychology, and Language Science at the University of Edinburgh and Adjunct Professor in the department of Linguistics at the University of Ottawa. His research covers a range of topics associated with syntax-external influences on syntactic phenomena, and his previous Oxford University Press publications are Events, Phrases, and Questions (2011) and Syntax and its Limits (2013, with Raffaella Folli and Christina Sevdali).

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

List of contributors

xxiii

George Walkden is a Lecturer in the Department of Linguistics and English Language at the University of Manchester. His research is in historical and comparative Germanic syntax. George’s monograph Syntactic Reconstruction and Proto-Germanic, a revised version of his doctoral dissertation, was recently published with Oxford University Press. He is also Associate Editor of Language, with responsibility for its Historical Syntax section. John Whitman is professor and chair of the Department of Linguistics, Cornell University, USA, and was previously director of the Department of Crosslinguistic Studies in the National Institute for Japanese Language and Linguistics in Tokyo, Japan. He works on syntactic variation and change, with a specialization on East Asian languages.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

 Micro-change and macro-change in diachronic syntax É R I C M AT H I E U A N D R O B E RT T RU S W E L L

. Syntactic theory and syntactic change Syntactic theory has come a long way in the past couple of decades. Despite superficial disagreements, almost every researcher in every major framework is now committed to a conception of a grammar as a large, language-particular, and open-ended population of lexical items (richly specified correspondences between sets of phonological, syntactic, and semantic properties), related by a small, universal set of syntactic operations (see, e.g., Chomsky 1995; Steedman 2000; Bresnan 2001; Sag 2012). Gone are the phrase-structural or transformational rules of Chomsky (1957) or Gazdar et al. (1985); the parameters of Chomsky (1981) are on their way out, too, except insofar as they can be reduced to properties of individual lexical items (e.g. Borer 1984) or perhaps to properties of interfaces between syntax and neighbouring domains (Berwick and Chomsky 2011). By and large, though, this pared-down syntactic theory has to hold itself accountable to the same range of data as ever. In particular, the chapters in this volume are concerned with a series of logical and empirical problems relating to syntactic change. Studying syntactic change can help us test and refine our synchronic syntactic theories; it can also remove some of the empirical burden from synchronic theories, by providing a diachronic basis for attested facts, whether at the level of individual languages or typological generalizations. Finally, it can provide a new source of insight into related matters such as language variation or acquisition. All of these issues are daunting, because modern syntactic theory gives us so little to work with. If a grammar is just a handful of universal operations or relations, plus a lexicon, then all syntactic change must ultimately reduce to lexical change (see also Walkden 2012). More specifically, most of the researchers in this volume adopt the feature-based approach to syntactic relations from Chomsky (1995, 2000, 2008). For those researchers, in most cases, syntactic change must reduce to a change in the featural specification of one or more lexical items. That really is all there is: other, apparently different modes of explanation (for example, claims that a certain constituent used to Micro-change and Macro-change in Diachronic Syntax. First edition. Éric Mathieu and Robert Truswell (eds). This chapter © Éric Mathieu and Robert Truswell 2017. First published 2017 by Oxford University Press.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i



Éric Mathieu and Robert Truswell

move to the left periphery, but now merges there directly, e.g. Roberts and Roussou 2003; or that a certain constituent now moves to a new landing site) are just abridged descriptions of changes in particular configurations of lexical items with certain, interacting featural specifications. This is the most basic type of diachronic syntactic research: reduce changes in surface syntactic patterns to a series (minimally two) of synchronic ‘snapshots’ of the underlying grammars, conceived of as sets of lexical items which interact to yield the surface patterns. Many chapters in this volume engage in this kind of detailed analysis. As well as its own intrinsic interest, this endeavour feeds into synchronic syntactic theorizing: minimally, every postulated synchronic ‘snapshot’ of a grammar G carries with it the implicit claim that G is a possible human grammar: we cannot postulate illegitimate synchronic states, just to get ourselves more smoothly from A to B. Although this sounds trivial, it need not be: to take two familiar examples, van Gelderen’s (2004) analyses of certain cyclical changes in terms of tendencies for specifiers to become heads, and for interpretable features to become uninterpretable, if accurate, commit us to a syntactic architecture in which specifiers can be distinguished from heads, and uninterpretable features from interpretable. Many varieties of Minimalism make the distinctions that van Gelderen needs; other theories may not. In this way, a theory of syntactic change implies a theory of syntax. At one step further removed from synchronic syntactic theory, snapshot-based analyses of particular syntactic changes, when aggregated, imply notions of ‘possible (and probable) syntactic change’. Given a series of snapshots G1 , . . . , Gn , we can infer a crude analogue (to be refined immediately below) of a series of grammatical changes, as sets of paired grammars {G1 , G2 , G2 , G3 , . . . , Gn−1 , Gn }. It is possible, in principle, to aggregate such sets of changes, and take them to imply a roadmap of sorts through the space of possible grammars: grammars like Gi tend to develop into grammars like Gj or Gk , but not grammars like Gg or Gh . Once we have enough pairs of snapshots, we may expect to feel confident making such statements. Something like the model described in the previous paragraph is visible in several of the most lively current approaches to grammatical change, including grammaticalization theory (Heine and Kuteva 2002) and cyclical change (van Gelderen 2011a). In fact, though, no one currently works in exactly the way sketched in the preceding paragraph. Once we understand why, a range of further diachronic questions open up which point to the real challenge, and real interest, posed by trying to reconstruct a theory of syntactic change within such a minimal architecture. First, for all the architectural minimalism, grammars remain extremely complex objects. We are all told in LING101 that the average college student knows 30,000 words, or maybe 60,000, or 100,000. On the lexicalist approach to syntax espoused above, that means that a grammar consists of tens of thousands of statements about correspondences between bits of phonology and semantics with bundles of syntactic features. Any change from G1 to G2 is doomed to be a hapax legomenon when G1 and G2 are objects of such complexity. We can reduce the problem somewhat, by making a few further assumptions about syntactic theory. For instance, the exoskeletal approach (Marantz 1997; Borer 2005a,b), which pairs category-neutral roots with c-commanding categorizing heads, means

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

Micro-change and macro-change in diachronic syntax



that the majority of those tens of thousands of lexical items have no syntactic features, and, a fortiori, can only participate in syntactic change by gaining syntactic features. That makes the syntactic hapax legomenon problem an order of magnitude smaller, but still quite real. We could take a step further, if we remember that a grammar is just a collection of lexical items. Rather than comparing entire grammars, we could extract pairs of ‘corresponding’ lexical items, and note any changes. Rather than seeing syntactic change as a pair G1 , G2 , we would thereby construct a more fine-grained object, a set of pairs of sets of syntactic features associated with corresponding lexical items {LI11 , LI12 , LI21 , LI22 , . . . , LIn1 , LIn2 }. Regularities in syntactic change can be established at this finer grain: maybe LI11 can correspond to LI12 in an immediately subsequent snapshot, but never LI13 ; or perhaps if LI11 corresponds to LI12 in a subsequent snapshot, then LI21 must correspond to LI22 . Recurring pathways of change boil down to statements of the form: if grammar G1 contains LI11 , then with greater than chance frequency, grammar G2 will contain either LI11 or LI12 . And in the ideal case, possible and likely long-term syntactic changes would be nothing more than the transitive closure of these ‘instantaneous’ paired snapshots. As Lightfoot (2002) has vigorously argued, the notion of ‘correspondence’ is far from unproblematic. However, Walkden (2012) shows that phonological and semantic similarities, as well as independently established properties of probable syntactic changes, have significant heuristic value. All of this can be recast as a theory of reanalysis in acquisition: statements about possible and probable changes, about contingently related sets of changes, and so on, can be seen as statements of the form ‘If speaker S1 has a grammar containing LI11 , and if S2 learns from S1 , then S2 may/will/will never induce a grammar containing LI12 .’ At this point, we have something approaching a formally coherent theory of syntactic change, with at least the hope of avoiding the hapax legomenon problem and having some predictive power. (Of course, the more restrictive our theory of syntactic features, the more restrictive our theory of change in the featural composition of lexical items will independently be, but this is one of the more contentious areas of current synchronic syntactic theory, and we prefer to avoid wading in). However, such a theory would be far from satisfactory to any specialist in syntactic change. Here are just a few examples of problems that cannot yet be analysed in the above, snapshot-based approach. • It is not straightforward to state when a language has undergone a syntactic change, because of synchronic variation. • Many syntactic changes play out over centuries, or even millennia. • Complex, long-term changes form recurrent ‘pathways’, often composed of series of subsidiary changes. Many of these facts cannot even be stated in the vocabulary developed above, without doing violence to the presumed cognitive basis of these groups of corresponding lexical items, rooted in the acquisition process. To give an example, imagine an exceptionally well-behaved cycle, in which Stage 1 always progressed to Stage 2, and Stage 2 to Stage 3. We can describe this using the ideas we have developed, and have done so above. However, such a cycle would put language learners in an awkward position:

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i



Éric Mathieu and Robert Truswell

they can innovate and create a Stage 1 or Stage 2 grammar, but only if they promise that some subsequent generation will go further and create a Stage 3 grammar. Of course, we must not say things like this, but the empirical phenomena will not go away. And perhaps perfectly well-behaved cycles do not exist, with some languages getting stuck at Stage 1 or Stage 2, and others going somewhere else altogether, but the phenomenon of long-term change is real enough: it is not an accident that researchers see Jespersen’s Cycle, for example, as a set of changes with an internal cohesion over and above the usual, despite the fact that we can take dozens of generations to get from the start to the end. It is now common to take the first two of the problems enumerated above as indicative of interactions between multiple grammars. This has its origin in the seminal results of Kroch (1989b), which introduced the Constant Rate Hypothesis, and derived the CRH from a model of grammar competition. Again, given the limited resources at our disposal, it makes sense to make grammar competition do as much for us as possible. As for the third problem, that of recurring pathways of change, one possibility is to see this as evidence for biases in syntactic acquisition. Here, too, there is a link to multiple grammars. As described above, a complex, cyclical change can be decomposed into a series of ‘instantaneous’, simple changes. If transitions like LI11 , LI12 , LI12 , LI13 , and LI13 , LI14  are all very likely (either likely to happen in the first place or likely to spread through a population), then the result will be a long-term pathway of change which is still likely to be followed. The goal at this point is to explain why these changes are very likely to occur, rather than others. One very common answer is that certain biases in acquisition entail that syntactic change is not a free-for-all. This is our area, then. We start small, obsessing about the internal structure of lexical items, the set of features which can be lexically associated with a head, and so on; but basic considerations of empirical adequacy force us to consider population dynamics, interactions between multiple grammars, and acquisition biases. Neither half is complete or satisfying without the other: we have seen that individual lexicalist grammars alone lead to incomplete theories of change, but the larger scale considerations only have any predictive teeth if they are related to well-defined basic objects. The micro and the macro: the two sides of a modern syntactic theory of change.

. Overview of the volume We have not attempted to group these chapters into parts: the major themes are too intertwined, and any attempt to draw boundaries just leads to artificiality. However, the chapters at the beginning of the volume are ‘more macro’: they concern gradual change, among individuals and among larger populations, often over long periods of time. As the volume progresses, we focus increasingly on the micro-changes: particular proposals about particular groups of features or lexical items, in a couple of languages, over shorter time periods. The two styles of analysis cannot easily be separated: macro-level work builds on concrete diachronic grammatical analysis; concrete fine-grained diachronic work gains an extra dimension of interest from being considered in the big picture.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

Micro-change and macro-change in diachronic syntax



Although diachronic syntax is far from a predictive science, we are not scared as a discipline of asking why. In other words, although we cannot predict what will happen next in any individual case, we can, and should, attempt to describe what will tend to happen. And given the uniformitarian perspective of most diachronic syntactic research, every micro-level piece of research here is an implicit example of the kind of thing that tends to happen, grist to the mill of macro-level theorizing. We begin with a series of chapters that explore the relationship between acquisition and pathways of gradual change in nonuniform populations of linguistic agents. Cournane focuses on the relationship between acquisition biases and the typology of change. She distinguishes two ways in which acquisition pathways can relate to diachronic pathways: they can align (root modals develop before epistemic modals in acquisition; epistemic modals often arise from root modals by a process of grammaticalization), or they can oppose (monoclausal structures with parentheses are acquired before subordinating multiclausal structures, but multiclausal structures can grammaticalize into monoclausal structures with auxiliary and main verb). Cournane proposes that the two patterns relate to two different areas of grammar, and so to two different acquisition tendencies: alignment results from semantic overextension during acquisition, while opposition between acquisition pathways and diachronic pathways results from conservativity biases during morphosyntactic acquisition. Gisborne and Truswell share a focus on acquisition biases, but with a view to explaining typological particularities, rather than universals. Their focus is on parallel syntactic evolution, the establishment of a recurring pathway within a language family which is only rarely attested outside that family. For example, interrogative forms and demonstrative nominals repeatedly develop into phrasal headed relative markers in Indo-European languages, but not in other languages. Gisborne and Truswell claim that this can serve as a new source of evidence about acquisition biases: parallel changes in the syntax of cognate lexical items suggest that syntactic change, and so the acquisition of syntax, is biased. This typological perspective is shared by Whitman and Ono, who take up the longstanding idea (Givón 1975; Aristar 1991) that crosscategorial word order generalizations, like many of Greenberg’s (1963) implicational universals, can be attributed to recurring patterns of change. Whitman and Ono perform a clustering analysis on the properties described in the World Atlas of Language Structures Online (Dryer and Haspelmath 2013), in order to reveal robust patterns of co-occurrence between word order parameters. Whitman and Ono show that the only robust crosscategorial generalizations to emerge (other than those restricted to unmarked values) concern the order of a head and its arguments. These can be reduced to ‘relabelling’ instances of grammaticalization (for instance, prepositions developing from verbs and maintaining the relative position of the verb’s complement). A relatively restricted and well-attested set of recurring reanalyses can therefore be taken as the motor driving crosscategorial word order generalizations. The next two papers address the range of effects that can arise from language contact, maintaining the above focus on increases in complexity arising from interactions among multiple grammars, but in a somewhat different form. É. Kiss shows through an analysis of changes in the Hungarian verbal complex that contact can lead

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i



Éric Mathieu and Robert Truswell

to pervasive changes in syntax. She demonstrates specifically that the Hungarian tense system became more complex under Turkic influence, and then simplified under subsequent Slavic influence, raising questions about structural conditions for borrowing of such systems. É. Kiss argues that the evolution of complex tenses started with a micro-change: the reanalysis of the feature content of a verbal suffix. This step initiated further processes of reanalysis, analogical extension, and abstraction, as a consequence of which the tense system inherited from Uralic, distinguishing only past and nonpast, developed into a complex system marking both tense and viewpoint aspect. A longstanding obsession in diachronic syntax has been the S-curve as a model of the time course of linguistic change (Osgood and Sebeok 1954; Weinreich et al. 1968; Bailey 1973; Altmann et al. 1983; Kroch 1989b; Blythe and Croft 2012). Three chapters here investigate an alternative pattern, in which a change initially appears to move away from its eventual end-state. Postma and Bacovcin both address the dynamics of failed changes, where a change gets underway but is never fully actualized, from apparently incompatible perspectives. Postma develops his earlier work (Postma 2010) tying failed changes to successful changes, presenting a reconceptualization of Kroch’s Constant Rate Hypothesis, expanded to include a transient state (reflecting the failed change) between the initial and final grammars in a competition process. Bacovcin argues, based on the diachrony of a short-lived Middle English pattern in which a ditransitive verb phrase is realized as V to-NP NP, that at least some failed changes are an emergent result of interactions between two changes which go to completion. On Bacovcin’s analysis, the V to-NP NP pattern emerged as a result of a reanalysis of to as a dative marker, and disappeared as a consequence of a second reanalysis, a couple of centuries later, of the recipient in recipient–theme orders as accusativemarked. The V to-NP NP order spread with the first reanalysis, and disappeared as the second reanalysis spread. Meanwhile, Troberg and Burnett show that the development of verb-framed modern Romance languages from satellite-framed Latin does not reduce to an incremental accumulation of verb-framed constructions. Rather, Medieval French, en route to the verb-framed behaviour of modern French, passed through a stage in which it exhibited even more satellite-framed behaviour than Latin. This is not a U-shaped curve in the same sense as Bacovcin, then: Troberg and Burnett are not concerned with fluctuations in frequency over time. Rather, they are concerned with trajectories between two idealized grammar ‘types’. They analyse the unusual progression from Latin to Medieval French to Modern French as a series of two reanalyses: first, place prefixes (which can be supplemented with a null path morpheme) are reanalysed as Path heads, leading to the innovation of resultative and verb–particle constructions. Secondly, the Path head is reanalysed as an inseparable part of the verb, leading to the loss of the whole range of satellite-framed constructions. Lavidas aims to explain the appearance of an unusual pattern of accusative-marked subjects of finite predicates in Early Byzantine Greek. The predicates in question are usually mediopassive verbs; Lavidas claims that the unusual, and quite short-lived, option of an accusative subject of such verbs arose as a result of developments in the case system: accusative case was initially interpretable (in opposition to partitive) and would subsequently become reanalysed as a structural case marking objects;

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

Micro-change and macro-change in diachronic syntax



meanwhile, the voice morphology is argued to show a complementary change: initially arising as a result of agreement with the accusative subject, once accusative case became structural, voice morphology became inherently determined. The unusual intermediate system arose during the transition between the two more standard systems. By now, we are approaching the ‘micro’ level of diachronic syntax. The meat of Bacovcin’s analysis (if not the implications for the modelling of change) consists of small changes in the English case system; and the analyses of Troberg and Burnett, and Lavidas, are based on equally small changes, in the French system of path/place prefixes and prepositions and the Greek aspectual system respectively. Much of the current fascination of diachronic syntax comes from the large-scale effects arising from such small lexical changes: the relationship between surface phenomena and underlying analyses is far from straightforward, as has been emphasized repeatedly since Andersen (1973) and Lightfoot (1979). One example of the complexity of this relationship, currently the subject of intense analysis, concerns the range of fronting operations in Germanic and Romance. It has become increasingly clear that the bynow well-established descriptive label ‘verb-second’ covers multiple possible featural specifications for T and C heads, and beyond. At this stage, even the synchronic typology is not definitively established, and it so happens that many ‘old’ languages (including Old French, Old English, and Old High German) are of particular interest in refining that typology. Labelle and Hirschbühler concentrate on the related phenomenon of Stylistic Fronting, attested in modern insular North Germanic varieties and described in Old French by Mathieu (2006b). Labelle and Hirschbühler dispute this analysis, claiming that the construction described by Mathieu does not share certain properties with North Germanic Stylistic Fronting, and actually does not reflect a single syntactic structure, but rather a range of structures differentiated by patterns of word order, derived by phrasal movement of VP, possibly in combination with short scrambling of the object. There is thus not one Leftward Stylistic Displacement construction but different constructions allowed by the grammar. Salvesen and Walkden focus on the distinction between symmetric and asymmetric V2 languages. In contrast to modern German, the V2 language par excellence, Old French and Old English show a number of V3 and other orders. This has led to different analyses being proposed for Old French and Old English: some researchers aim to maximize the similarity to German, by postulating V-to-C movement in all cases, together with housekeeping measures to account for V3 orders; other analyses place the verb in these languages lower, for example in T0 , leaving more scope for multiple constituents to appear on its left. Moreover, the range of V3 constructions in Old French and Old English is not identical: Old English is largely as predicted by this description, while Old French V3 orders are more limited in scope, and largely reduce to the constructions which Labelle and Hirschbühler address in their chapter. Because of this, it is unclear whether a unified analysis of Old French and Old English is possible, or even desirable. Salvesen and Walkden’s contribution is to examine the distribution of embedded V2. If V in these languages does not raise past T, then embedded V2 should be quite unremarkable. If, on the other hand, V has raised to C, then we are inclined to treat embedded V2 as an embedded root phenomenon,

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i



Éric Mathieu and Robert Truswell

of the sort first described by Hooper and Thompson (1973): only compatible with certain classes of matrix predicate. Salvesen and Walkden show that the latter prediction better fits the corpus data, generating an expectation that V3 orders can be explained elsewhere, for instance within a more articulated syntax of the left periphery. Finally, Light uses current ideas about the syntax of V2 in Germanic to argue that demonstrative pronouns across historical Germanic varieties are interpreted as contrastive. The argument runs as follows: many Germanic varieties allow fronting of contrastive elements, but only German allows fronting to preverbal position of noncontrastive elements other than the subject (a fact which Light links to the wider range of possibilities for scrambling to the Mittelfeld in German). However, sixteenthcentury Bible translations in Early New High German, Early Modern English, and Icelandic all contain instances of fronting of object demonstrative pronouns to preverbal position. Light takes this as evidence that object demonstratives are interpreted as contrastive in these languages, as the only preverbal noncontrastive elements are subjects. She links this to a technical characterization of contrast, building on the pragmatic notion of subinformativity (Gast 2010). Ecay and Tamminga develop a new method for relating surface phenomena to underlying grammatical patterns, similar in its scope and outlook to the Constant Rate Effect of Kroch (1989b). The logic of the Constant Rate Effect is that, if two surface phenomena are linked to a common underlying mechanism, the rates of change of the two phenomena across time should be similarly linked. Ecay and Tamminga’s new diagnostic relies instead on persistence, or non-independence of sequential utterances as a consequence of priming-like effects. If persistence can be demonstrated between two surface phenomena, that could indicate that they have underlying elements in common. Ecay and Tamminga apply this logic to the analysis of Middle English negation, which surfaced in three variants: ne, ne . . . not, and not. Competing accounts have analysed this surface alternation either in terms of two possibly co-occurring underlying atoms, or three independent forms. The evidence from patterns of persistence supports the three-atom model. At the most micro level, diachronic syntax involves tracking developments in the syntactic composition of a single form and its descendants. For example, Danckaert deals with a typical problem in diachronic syntax: there are two candidate sources in an ancestral language for a later construction, and the task is to distinguish which is the actual source. His study involves the Latin sources of Romance analytic passives. Late Latin had two analytic passive forms: the infectum (e.g. amatus sum), and the perfectum (e.g. amatus fui). It has typically been assumed that the infectum is the source of the modern Romance analytic passive (e.g. French Je suis aimé ‘I am loved’), because of the formal continuity between sum and suis. However, Danckaert shows that word order facts support the opposite conclusion: forms like fui, along with most auxiliaries, tend towards preverbal position in Late Latin, while sum remains largely postverbal. This suggests that largely preverbal fui is the source of the modern Romance construction, while the sum form came to behave syntactically more like the ancestors of the future marker. This in turn raises a number of questions about how the modern paradigms emerged from these formally distinct ancestors.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

Micro-change and macro-change in diachronic syntax



Courtney discusses the Dutch inflectional paradigm, with a focus on complementizer agreement. Competing syntactic and post-syntactic analyses of complementizer agreement exist in the literature; Courtney’s contribution is to argue that the two types of analysis are each appropriate to different dialects, and that it is plausible to construe the synchronic microvariation between syntactic and post-syntactic agreement patterns as a reflex of gradual grammaticalization of the agreement relation, in the sense of recent Minimalist approaches to grammaticalization such as that of Roberts and Roussou (2003). Finally, Jędrzejowski tracks the directional change of a single lexical item, German versprechen ‘promise’. As in many languages, versprechen is polysemous: its bestknown interpretation is as a control verb, with the speaker committed to a state of affairs described in a complement clause. However, it can also have a raising interpretation, typically with inanimate subjects like das Wetter ‘the weather’, an interpretation without speaker commitment, and other changes in argument structure. Jędrzejowski uses well-established grammaticalization mechanisms to establish a pathway in which the raising interpretation of versprechen emerges first in constructions embedding a nominal, rather than clausal complement. By now, we have come a long way from the very general patterns of change in the earlier chapters. But the point of modern diachronic syntax is that you can’t do one without the other. Specific long-term time courses of change must emerge from discrete, local, intergenerational changes in the lexical specification of grammatical features. No paper in this collection is far from a reduction of syntactic change to ‘imperfect’ lexical acquisition; no paper concentrates on a phenomenon so small as to have no broader architectural implications. The counterpoint between small, discrete changes and large-scale, emergent, multigenerational diachronic phenomena is the core of this new discipline.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

 In defence of the child innovator A I L Í S C O U R NA N E

. Introduction For over a hundred years researchers have appealed to the child innovator to explain language change (Meillet 1912; H. Paul 1920; Halle 1964; Andersen 1973; Kiparsky 1974, i.a.). Today, first language acquisition (L1A) is central to generative theories of change (e.g. Lightfoot 1979; Roberts and Roussou 2003; van Gelderen 2004, 2011a). I will call this approach the Child Innovator Approach (CIA), in contrast to approaches that place explanatory weight on adult speakers (Bybee and Slobin 1982a; Heine et al. 1991; Ziegeler 1997; Givón 2009; Diessel 2011, 2012, i.a.). Proponents of the CIA argue that innovative child analyses of the input language, if and when they survive and spread, become the changes reflected in the historical record. The CIA developed from empirical observations of historical data, but places the burden of explanation for diachronic patterns on the L1A process. Despite this, the consequences for language acquisition have rarely been considered beyond the anecdotal (though see Clark and Roberts 1993; Weerman 1993, 2011; van Gelderen 2011a: 21–6; Meisel et al. 2013), and even more rarely assessed using targeted L1A studies and child data (Baron 1977; Cournane 2014, 2015a). In this chapter, I address two common criticisms of the Child Innovator Approach that persist in the (near-)vacuum of communication between researchers working on L1A and historical linguistics. The first is that the CIA is untenable since inputdivergent1 analyses that occur during L1A must survive into adulthood. While everyone agrees children make input-divergent analyses in the process of learning, critics of the CIA argue these resolve prior to a time when diffusion is possible (adulthood; 1 I use the term input-divergent to refer to any analysis by the child that does not conform to the grammars of the speakers who comprise the child’s input. Input-divergent analyses are normally referred to as child ‘errors’. The term ‘error’, though convenient, is problematic for many reasons, perhaps the most egregious of which is that these apparent errors only become erroneous when compared to the ‘correct’ analyses of other people’s grammars; when analysed within the grammar which produces them (i.e. the child grammar at its relevant state) then they are fully consistent. The term ‘error’ also carries a judgement of inaccuracy relative to some goal that is not consistent with a descriptive and non-teleological view of language development.

Micro-change and Macro-change in Diachronic Syntax. First edition. Éric Mathieu and Robert Truswell (eds). This chapter © Ailís Cournane 2017. First published 2017 by Oxford University Press.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

In defence of the child innovator



e.g. Traugott and Dasher 2005; Diessel 2012: 1600). In Section 2.2, I respond to this criticism by drawing attention to bodies of evidence for (a) peer-to-peer acquisition throughout childhood, (b) bilingual and heritage language effects on L1A that prolong input-divergent analyses, and (c) longstanding sociolinguistic evidence that the teenage years are most relevant for the spread of innovations (as well as evidence that even pre-adolescent children likely participate in advancing sociolinguistic changes). The second common criticism comes from the assumption that parallels must hold between child innovations and diachronic innovations (‘ontogeny recapitulates phylogeny’) in order for the CIA to hold. We see parallel alignments in some domains (e.g. morphological regularization, semantic extensions), but not others (constructional reanalysis, as from biclausal structures to monoclausal, the development of agreement markers from pronouns). The heterogeneity of alignments between L1A pathways and diachronic pathways across linguistic domains is taken to be evidence against the CIA (e.g. Diessel 2012). In Section 2.3, I argue that the CIA does not predict that the L1A process should mimic completed diachronic changes. Rather, the CIA predicts that contemporary children make input-divergent analyses that advance diachronic cycles (which we can look back on in the future).2 In an effort to capture why semantic changes show parallels but morphosyntactic ones do not, I formulate the child language predictions of the CIA in terms of the mapping problem (see also Cournane 2014). The mapping problem is a primary problem in L1A, referring to how children learn which linguistic forms to associate with which meanings (entities, events, properties, etc.; see E. Clark 1977, 1993; Gleitman 1990; Bloom 2000). Broadly, parallel alignments occur when the child posits a new meaning for an extant form (e.g. overextension), while oppositional alignments occur when the child posits a new form for an extant meaning (parsimonious or conservative structure postulation). This approach stresses that different necessary processes in typical L1A give rise to different types of innovations.

. The survival and diffusion of L input-divergent properties For the CIA to be supported, input-divergent analyses during L1 acquisition must diffuse to other speakers and become the changes we see in the historical record. For this to happen the child must either succeed in spreading her input-divergent analyses to mature speakers during childhood or fail to correct input-divergent properties and spread innovations as an adult. The first possibility is seen as problematic because children are not influential members of the speech community (older speakers will not model their speech after children) while the latter is problematic because children are excellent language acquirers who make few possible errors (Snyder 2007, i.a.) and resolve those errors that they do make prior to adulthood (see Kerswill 1996). Thus, the CIA is untenable because children cannot spread errors as children and errors resolve by adulthood when they theoretically could spread (Diessel 2012). I argue that this 2 In other words, we expect children to make new changes, not repeat old ones. For example, overextend current spatial terms like beside to temporal meanings, rather than use, e.g., after first with spatial meanings then later extend it to temporal meanings.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i



Ailís Cournane

assessment misconstrues the sociolinguistic evidence and overestimates the swiftness and accuracy of child development, especially in multilingual or contact situations. I appeal to three lines of evidence: (a) children learn from other children via peer-topeer acquisition, (b) adolescents diffuse innovations, not adults, and (c) multilingual learning conditions may prolong divergent features of first language acquisition. First, the notion of social prestige is important to clarify. The idea that children lack the social prestige necessary to be linguistically influential is sometimes taken as an argument against the CIA (e.g. Diessel 2012: 1600, who cites Labov 2001). However, prestige applies within all groups, not only in adulthood or with those that hold conventional power (see Labov 2001, 2012 for discussion). Children, pre-adolescents, and adolescents may all hold power roles and differing influence within their groups. Children begin talking like their peers (instead of like their parents) from earliest socialization, approximately kindergarten age in Western societies (e.g. Labov 2012; Cekaite et al. 2014). Taking peer-to-peer acquisition seriously, it is possible that young siblings or school-age children diffuse and reinforce input-divergent analyses in their grammars. Children with similar input show similar acquisition patterns so reinforcement may be particularly plausible. Even if we assume a critical period for syntactic or semantic acquisition (perhaps ending at puberty), input-divergence may spread to peers, or overlapping input-divergent analyses may be reinforced among peers, within the childhood years. Sociolinguistic studies in both real and apparent time repeatedly show that it is primarily adolescents, not adults, who embrace and spread innovations (‘lead sociolinguistic changes’; e.g. Labov 2001, 2007). For example, quotative be like (He was like ‘X’) was embraced first by teenage girls (e.g. Tagliamonte and D’Arcy 2004), as with most changes. If we assume innovative variants originate in L1A input-divergence, then the sociolinguistic evidence significantly decreases the timespan within which input-divergent properties of the I-language need to survive before they spread to the speech community. Further, as input-divergent properties are internally conditioned by the input and learning mechanisms, children and teens could have the same inputdivergent properties and reinforce each other, as with peer-to-peer acquisition. Thus, for the CIA to be supported, input-divergence need only survive to adolescence, possibly only to pre-adolescence (see e.g. Levey 2006 for evidence that pre-adolescents may lead sociolinguistic change). L1A is often remarkably accurate when measured against adult speakers of the input, but it is also persistently and systematically inaccurate in some areas despite the input (e.g. derivational morphology usage and comprehension continues to develop through the early teen years, Derwing and Baker 1986). Even in monolingual learners with relatively uniform linguistic input, child input-divergence has been demonstrated to persist in areas like semantic development of modals until at least 12 years old (see Papafragou 1998). More commonly, divergent patterns have been shown to persist into at least late childhood (age 7 or 8). For example, comprehension studies with 6–9-year-old children show persistent non-adult scope biases (van Koert et al. 2015), and overextensions of direct evidential marking (de Villiers et al. 2009), to give just two semantic examples. These properties of first language acquisition are a plausible source of innovations even in monolingual populations (which cannot be taken

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

In defence of the child innovator



to be without variation), but are likely exacerbated in bilingualism contexts. These should still be taken to be instances of first language acquisition (in the case of heritage languages, or 2L1). In bilingual children evidence shows that certain areas of the grammar may be susceptible to prolongation (relative to monolinguals) of divergence periods (see Gathercole and Thomas 2009). Critically, these are cases of first language acquisition that show delayed (but not deviant) development relative to monolingual L1 development. This prolonged divergence may be another means through which divergent analyses persist long enough to be reinforced and/or diffuse.3 In sum, child input-divergent analyses are plausibly mutually reinforced during childhood through peer-to-peer acquisition. Furthermore, input-divergent analyses need only persist into the early teenage years when they can be picked up by the sociolinguistic change powerhouse of teenage peer groups. The variables that diffuse in the teenage years still need to originate from somewhere (the actuation problem, Weinreich et al. 1968). Even in monolingual populations, input-divergence sometimes persists into late childhood or pre-adolescence, a property of L1 acquisition that is likely exaggerated in bilingualism contexts,4 lending further plausibility to this avenue of research. The debate over whether input-divergent child analyses may diffuse into the speech community is far from settled.

. Unidirectionality of diachronic pathways and the mapping problem Proponents and critics alike regularly see parallel development between ontogeny and phylogeny as a necessary condition for the CIA, for example: ‘if language acquisition is the source of diachronic change, child language should include the same types of changes and developmental patterns as the diachronic evolution of language’ (Diessel 2012: 1600, emphasis mine). However, for the CIA to be supported by child patterns, child input-divergence needs to advance new diachronic pathways, not mimic (or parallel) completed ones. While the links between L1A and change are perhaps more obvious in parallel alignments (discussed in Section 2.3.1), I show that both parallel alignments and oppositional (= nonparallel) alignments between L1 development and diachronic development can be reduced to two sides of the form ↔ meaning mapping problem. On one side, children attribute new meanings to existing forms as part of the normal process of meaning extension (and parallel pathways occur), while on the other, children attribute new forms to meanings as part of the normal process of developing grammatical complexity (and oppositional pathways occur). The diachronic domain of inquiry that forms the empirical basis for my arguments is the development of modal expressions in child language (Kuczaj 1977; Stephany 3 Cf. Weerman (, ); Meisel et al. (), who argue that sequential bilinguals (age of onset at age  or older) as well as L speaker input in the primary linguistic data are the primary triggers for morphosyntactic changes. If L speakers make up much of the input then the child’s grammar will correct the inconsistencies (compare with creole formation or language creation, as with Nicaraguan Sign Language, e.g., Senghas and Coppola ). 4 Another population of interest as potential innovators is the Specific Language Impairment population (Petra Schulz, p.c.), which makes up approximately  of the population and exhibits prolonged and divergent LA in several areas (Tomblin et al. ).

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i



Ailís Cournane

1979; Shepherd 1982; Wells 1979; Bassano 1996; Papafragou 1998; Heizmann 2006; Cournane 2014, 2015a,b), and along the diachronic Modal Cycle (Jespersen 1924; Lightfoot 1979; Traugott 1989; Bybee et al. 1994; Gergel 2009, i.a.). The modal cycle has both a semantic component (lexical > functional (root > epistemic)) and a syntactic one (V > v > AUX/INFL).5 The semantic pathway has parallels in child language, while the syntactic pathway in children is in opposition to diachrony; the Modal Cycle thus serves as a useful illustration for a discussion on diachronic directionality and the CIA. .. Parallel alignments: Mapping existing forms to new meanings Parallel alignments between acquisition and change have generated many discussions since the early days of historical linguistics (see Baron 1977: ch.1, for an overview). In parallel alignments the same descriptive pathway is charted in child development as in historical development. For example, in both domains, spatial meanings of adpositions arise before temporal (afterspatial > aftertemporal ; Ziegeler 1997; Clark and Carpenter 1989), motion paths arise before temporal paths (be-going-tomotion > begoing-tofuture ; Fleischmann 1989; Schmidtke-Bode 2009), and root modal meanings precede epistemic meanings (Traugott 1989; Papafragou 1998). In this section we focus on modal semantic development. Modals (e.g. must, can, have to, gonna) are quantifiers (i.e. must is universal with a necessity meaning, while might is existential with a possibility meaning) that express two broad types of meanings (Kratzer 1977, 1981, 2012, i.a.): root meanings like abilities (1a), deontic obligations (1b), or future intentions ((1c); Copley 2002), and epistemic meanings (evidence-based inferences, like (1d)). Polysemous6 (or functional) modals are those modal expressions that express both root and epistemic meanings, depending on context and/or contributing grammatical factors (Kratzer 1981; Perkins 1983; Sweetser 1990; Hacquard 2006, i.a.). For example, English must can express deontic (2a) or epistemic (2b) meanings. Root meanings arise when a polysemous modal scopes below Tense (over the verbal event), while epistemic meanings arise when a polysemous modal scopes above Tense (over a propositional event) (see Hacquard 2010 for analysis). Polysemous modals are more grammatically complex than monosemous modals that consistently express only one modal meaning ((3); see Hacquard 2013). (1)

a. b. c. d.

Olivia can speak Spanish. Erin has to be home before 11pm. I’m going to go to the shops before working. You must be going skating, since you’re holding your hockey stick.

5

I do not address any claims about sound change. This is shorthand; the same modal word (e.g. must) expresses both root and epistemic meanings in combination with different modal bases and ordering sources but the modal itself is not polysemous in the Kratzerian tradition, the lexeme is specified only for quantificational force (e.g. strong necessity or weak possibility meaning; Kratzer , ). 6

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

In defence of the child innovator (2) The baby must be fed. . . a. (i) . . . because he’s hungry. (ii) The baby mustRoot [be fed]VP b. (i) . . . because he’s not hungry. (ii) mustEpistemic [the baby be fed]TP



Root Deontic Epistemic

(3) Monosemous modal expressions a. I know Darwin is hungry. b. Darwin is probably hungry. c. It’s likely (that) Darwin is hungry. Diachronically, lexical modal verbs with root meanings like want or know how (‘premodals’, see Lightfoot 1979) grammaticalize into functional modals. Within functional modals, root meanings dominate early on and epistemic meanings increase over time (see Traugott 1989; Tagliamonte and D’Arcy 2007). For example Old High German musan (> müssen) (4a) was a main verb with the root meaning [[to be able]]. In (4), we see that müssen has changed from ability meaning in OHG, to deontic meaning in MHG, and finally occurs with epistemic meanings in Modern German. This pattern of semantic developments for lexical items is robustly attested (Bybee et al. 1994). Currently, the verb müssen is used to express both deontic and epistemic meanings, like English cognate must. The meanings expressed by a lexical item go from lexical > functional (root > epistemic). Focusing here on the development from root > epistemic, müssen first appears with root meanings and later with epistemic. Root > epistemic reanalysis reduces to the following: the modal first scopes over the verbal event at LF (5a), expressing root meanings, and later extends to the higher scope position above the proposition (5b), expressing epistemic meanings (see Hacquard 2013; Hacquard and Cournane 2015 for details). (4) The Semantic Modal Cycle (German; Fritz 1997, cited in Traugott 2006: 111): OHG müssenability → MHG müssendeontic → ModG müssenepistemic a. Sie ni musan gan so fram zi themo heidinen man. they not were-able go so far to the heathen man ‘[For religious reasons] They were not able to proceed further to [the palace of] the heathen man (Pilate).’ (9th c., Otfrid IV.20.4) b. Tie minnera habeton die muosan gan. those no.money had they had.to walk ‘Those who had no money were obliged to walk.’ c. Du must wohl müde sein. you must mod.prt tired be ‘You must be tired.’ (5) LF scope interpretations The baby must be fed. a. The baby must [be fed]VP b. must [the baby be fed]TP

(c.1000 Notker I.152.1)

(S. Gargova, p.c.)

Root Deontic Epistemic

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i



Ailís Cournane

In the individual L1 learner, polysemous modals like English must or French devoir (Bassano 1996) first appear in naturalistic child speech with root meanings (6a), then later with epistemic meanings ((6b); after the third birthday). This developmental pathway, like the diachronic one in (4), is robustly attested (for Turkish, Finnish, and Greek, Stephany 1979; for French, Bassano 1996; for English, Wells 1979; Papafragou 1998; for Mandarin, Guo 1994, i.a.). (6) First root must and first epistemic must used by Sarah (Brown 1973) a. he mus(t) talk [Deontic; Sarah 2;9, referring to her broken talking-doll] b. must be gone [Epistemic; Sarah 3;0, referring to missing toy plates] If the CIA is correct and children are responsible for innovations like root > epistemic, then children are expected to overextend the meaning coverage of root lexical items. In other words, meaning extensions must create the new mapping relationships for form ↔ meaning that arise in the historical record. To develop semantic productivity, we know that the child must extend meanings learned within constrained contexts to novel contexts. For example, when learning the word ‘kitty’ with one referent (the family cat Darwin) the child is tasked with extending that meaning (to all cats, including Darwin). How far does the meaning extend, or, which set of referents is correct? Extension is necessary for productivity, however, if the child also calls dogs kitty, she has overextended the meaning coverage of kitty to include too many possible referents. Overextension happens in normal acquisition of both lexical and functional items (e.g. Bowerman 1985; E. Clark 1993). For example, even if must only has root meanings in the input (as was once the case in English and for West Germanic cognates), the semantic grammar still allows functional modals to merge in two positions (see Hacquard 2006, 2010; Hacquard and Cournane 2015 for details). The child’s grammar can productively extend the meaning of must to cover epistemic usage.7 The compositional semantics predicts the grammatical availability of root > epistemic extension (nothing formally restricts a functional modal from scoping over either a VP or a TP; see Hacquard 2006 for details).8 As a normal part of development, children must learn and constrain the extent of meaning coverage of lexemes as part of solving the form ↔ meaning mapping problem; sometimes the grammar will allow productive patterns that are input-divergent but internally consistent. If and when these diffuse, they are expected to give rise to noticeable parallels between L1A patterns and change patterns (the former begets the latter).

7 With modals, as with be-going-to or adpositions, the earlier meanings typically remain when the newer meaning arises, giving rise to one-to-many form ↔ meaning mappings (Hopper and Traugott ), as is relatively common in the modal domain (van der Auwera and Ammann ). 8 Why might the child do this? The child is under pressure from expressivity forces: she needs to express herself with a limited lexicon and grammar and thus may be more likely to co-opt morphemes with meanings close to the desired meaning in her thoughts (and the grammar allows it).

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

In defence of the child innovator



.. The CIA and parallel alignments: Response to critics The child meaning mapping extensions from root > epistemic, space > time, or motion > future are parallel to diachronic extensions for lexemes. In both cases, more concrete meanings extend to more abstract meanings in predictable, metaphorical, ways. For Diessel (2011, 2012), parallel alignments are parallel only because in both diachrony and acquisition the same psychological mechanisms (analogy, entrenchment, and categorization) are involved, not because the child extensions are the diachronic ones (as the CIA argues). For adult language users to be responsible for parallel pathways, adults must be shown to overextend meanings in the direction predicted by the diachronic trajectories. While adult innovation is plausible (also for many generative researchers, adults may be responsible for at least some types of change), there are three hurdles the adult approach faces that the child approach does not. First, adults have more experience with language and with existing speaker variation within and across dialects. When adult speakers produce what appear as innovations, as in over-regularization of morphology (went > goed; see Bybee and Slobin 1982a), it is always possible that the speaker acquired that variant in development or has been exposed to other speakers who have those variants natively. Adult usage of variants is more likely participation in sociolinguistic borrowing, incrementation, or diffusion of extant variants. Further, once the existent variants from the speech community are eliminated as candidates for adult innovation, the remaining candidates for adult innovations may be localized events that stand out as processing errors or poetic/playful acts; this is different from the child who systematically produces internally consistent, productive, input-divergent forms. While the child approach must demonstrate that child innovations can become fixed in their grammar and/or spread to other speakers (see Section 2.2 above), the adult approach faces this problem in addition to the burden of showing that adults are true innovators of new variants. Second, the adult innovator approach potentially violates Ockham’s razor, as it needs to demonstrate that adults also make systematic meaning extensions that align with diachronic innovations. There is agreement that children make creative extensions as in (7), in the direction compatible with diachronic patterns; but only the CIA view assumes that these are ultimately the same extensions that become diachronic innovations. In other words, the CIA hypothesizes that parallels suggest not only the same kinds of extensions in L1A and change, but also maintains that the simplest solution for the source of semantic innovations is the input-divergent meaning extensions of the developing learner. There is only one mechanism underlying overextension; we do not need to argue that the child makes psychological extensions (which, as discussed below, are suspect when not treated with enough grammatical detail) and then later adults make similar extensions. It is less clear what would motivate adults to overextend meanings along conceptual routes, while this is obvious in child development (meaning extension to the set of possible referents in the world underlies the construction of productivity in the grammar). (7) Can I have any reading behind the dinner? (= after)

(Bowerman 1985: 1292)

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i



Ailís Cournane

It is commonly accepted that the L1A root > epistemic pathway reflects conceptual development in Theory of Mind (Moore et al. 1990; Papafragou 1998; cf. de Villiers 2007; Cournane 2015b); children do not use epistemic modals until they can entertain epistemic thoughts. However, previous literature, on which the Theory of Mind approach is based, focused almost exclusively on canonical English-type polysemous modals. Recall that polysemous modals enter into more complex grammatical (compositional semantic) relationships than monosemous modals. Modal meanings (root vs. epistemic) are linked to LF scope for polysemous modals (e.g. must, can, have to, gotta), but not for monosemous modals (maybe, probably). Crucially, epistemic interpretations for polysemous modal verbs depend on the ability to embed a proposition (= sentential complement) under a modal (5b). Cournane (2015a,b) shows that a child’s first epistemic use of a polysemous modal (e.g. must) correlates strongly with first markers of sentential embedding (first uses of embedding verbs like see or think, productive infinitival marking on the second verb, disjoint subjects in matrix and embedded clauses). These grammatical complexity-based reconsiderations of the root > epistemic dogma call into question the practice of considering uses of forms like must as directly signifying conceptual states, without considering the mediating or confounding effects of grammatical complexity development (see also de Villiers 2007). Even the 2-year-old child has other means of expressing reasoned guesses (grammatically simpler monosemous modals maybe, peut-être, think; Bassano 1996; O’Neill and Atance 2000; Cournane 2015a,b; see de Villiers 2007 for discussion on the lack of straightforward alignments between modal developments and Theory of Mind milestones). Thus, it is questionable to argue that conceptual extensions underlie child development for root > epistemic, and is likewise suspect for diachronic extensions. In both the child and diachrony, extensions are linked to local components of the grammar. The child extends the meaning coverage of polysemous modals to higher scope positions (above propositional content, minimally TP). This is a grammatical development where a form goes from a one-to-one meaning relation to a one-to-many meaning relation. As with the child, ‘conceptual routes’ in diachrony are localized to lexemes entering new grammatical relationships, they are not conceptual changes. In both the child and diachrony, the trajectory from root > epistemic describes grammatical development, not conceptual development; it captures the extension of the meaning coverage of polysemous forms from low scope to high scope interpretations. I conclude that for semantic diachronic pathways where we see parallel alignment, child evidence is compatible with the CIA if we find that the child learner extends the meaning mappings of a form in order to advance the cycle. Children must necessarily extend meanings; they are ‘extenders’ and we know they sometimes overextend in ways consistent with their grammar. Further research is needed to explore whether functional overextensions in more domains of L1A provide the diachronically expected innovations. The assessment metric for whether children make the right kinds of overextensions is previous language changes, but we must never fall for the trap of expecting recapitulations with the same lexical items. Only renewing items (see van Gelderen 2009a) that will forge truly new changes are under consideration when we assess the CIA in child data.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

In defence of the child innovator



.. Oppositional developmental alignments: New forms for existing meanings Many historical changes also occur in a unidirectional manner, but fail to show a parallel in L1 acquisition. Diessel (2011) has argued that such lack of parallels in child language constitutes evidence against the CIA. Two common such changes from the historical record are (1) reanalysis of biclausal structures into monoclausal structures (e.g, IP[modal verb] + IP[main verb] > IP[modal aux + main verb] , see Roberts 1985; or CP[+Q,cleft] + CP[+Q] > complex CP[+Q] , see Tailleur 2012); and (2) reanalysis of independent morphemes into bound morphemes (e.g. pronoun > clitic > agreement, see Fuß 2005). These morphosyntactic changes appear to lack L1 parallels, unlike the semantic extension changes discussed above. Generalizing, we see structural simplification processes in diachrony (see Longobardi 2001; Roberts and Roussou 2003), but the child develops syntactic complexity ((8b); e.g. Brown 1973; Diessel 2004; PérezLeroux et al. 2012). In these domains we see oppositional, not parallel, developmental pathways (8). I will focus on the development from biclausal structures to monoclausal structures, as attested from the historical record for modal verbs in the history of English (Lightfoot 1979; Roberts 1985; Roberts and Roussou 2003, i.a.). (8) Oppositional developmental pathways: a. Diachrony: Complex Structures > Simpler Structures b. L1A: Simpler Structures > Complex Structures The canonical set of English auxiliary modals (e.g. must, may, should, can, will, might) is currently realized as inflectional markers (in INFL or T; Pollock 1989, i.a.). They underwent wholesale reanalysis from premodal verbs to inflection markers during the late OE and ME periods (see Lightfoot 1979; Roberts 1985; Warner 1993; Denison 1993, i.a). For example, in the historical record we see that in OE and into ME these modals appeared with subject agreement and infinitival complements (9a). Currently, these modals are invariant in form9 and take bare verbal complements (9b). In (10a) we see an example from OE with the premodal sceolde (currently shall) taking a direct object (hundteontig mittan hwætes). Later in the historical record, from ME, these modals no longer appear with direct objects, as today (10b). (9) Loss of inflection a. sone hit mæi illimp-en soon it may.3sg happen.inf ME a1225 (?a1200) Lay. Brut 2250 (Denison 1993: 299) b. It may happen soon. (10) Loss of object argument a. he cwæð þæt he sceolde him hundteontig mittan hwætes he said that he shall him hundred bushels of-wheat ‘He said that he owed him a hundred bushels of wheat.’ OE Æhom 17:26 (Fischer and van der Wurff 2006: 147) b. *I shall [DP] 9 There is arguably still past marking on some of these modals, for example could, would, and should may retain at least some reflexes of previously productive tense marking (see Cowper ).

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i



Ailís Cournane

These changes (among others, see Lightfoot 1979; Roberts 1985) reveal a syntactic reanalysis for the canonical modals from verbs that inflect for tense and phi-features and take arguments, to exponents of INFL, which are themselves an expression of tense features and have lost their argument structure (see Pollock 1989). The reanalysis from V > INFL10 was necessarily concomitant with a reanalysis of the entire clausal structure. Prior to reanalysis, constructions with modal verbs were biclausal, with the modal verb embedding an infinitival verb. After reanalysis, the structure is monoclausal because the (newly) INFL modal becomes the inflection marking for the main verb (11). (11)

Structural reanalysis of two clauses into one (higher V > INFL).

IP… Vmod

IP…

IP… Vmain

Imod XP

Vmain

XP

L1 development, on the other hand, progresses from simple clauses > complex clauses (e.g., Brown 1973; Bloom 1991). The development process is highly contingent: with each new development, new expansions that were previously closed become possible (e.g. Gleitman et al. 2005). For example, Sarah (Brown 1973), like other children, begins using the premodal verb want in simple constructions followed by a bare noun (12a) or bare verb (12b,c), later we see want with multi-verb complements (12d), and later still with a possible subject for the second verbs (12e), and finally with a second verb with its own subject and inflectional marking (12f). This is the opposite trajectory from what we saw above for reanalysis from biclausal to monoclausal structures: structural reanalysis in diachrony appears as simplification but the child gradually develops structural complexity. (12) a. b. c. d. e. f.

want Bobo [2;3,07] I wanna11 ride my horse [2;3,07] I wan ride a horsie! [2;4,12] wan go read it [2;7,28] I want that write on [2;10,05] I want Daddy to help me [3;3,13]

What does the CIA predict for child data, if the child is responsible for innovations from complex biclausal structures to simpler monoclausal clauses? In oppositional pathways, the child needs to stay at a structurally simpler stage of analysis and never 10 I am simplifying, it is likely that these modals actually followed the more stepwise trajectory: V > v > INFL (see Tollan  for discussion). 11 In this usage the transcriber wrote wanna but when assessed against all other uses, and considering Sarah’s low MLU of . (relative to other typically developing English-learning children) it is likely that wanna at this point is an unanalysed whole. This debate over the correct analysis of early forms may equate to input opacity for the child too, not just for the linguist (thanks to an anonymous reviewer for pointing out this possibility).

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

In defence of the child innovator



posit a more complex structural analysis for the string in question (see Roberts and Roussou 2003 and van Gelderen 2004, 2011a for differing economy-based approaches). For example, children can use INFL (e.g. modal auxiliaries), as in (13a) earlier than they can embed sentences, as in (13b) (around age 2 vs. around age 3). Now, imagine that the 2-year-old child, unable to subordinate, analyses input strings like (13b) as monoclausal (want dance; cf. (12c) from Sarah). If the input is opaque with regards to key pieces of evidence that want is a subordinating verb, like the infinitival marker, then the child’s monoclausal analysis may persist as her final analysis. Note this is precisely what happened in the history of English (oversimplifying for expository purposes): the infinitive marker on the second verb in biclausal modal constructions was phonologically eroded, leading to reanalysis of biclausal modal structures into monoclausal structures with INFL modals (14). (emerges at approximately age 2; e.g. Wells 1979) (13) a. [(I) canINFL danceV ] b. [(I) wantV [to danceV ]] (emerges at approximately age 3; e.g. Diessel 2004) (14) [I wantV [to danceV ]] > [I wantINFL danceV ] This type of reanalysis is fed by phonological changes that level inflectional marking paradigms, as in the history of English (see Lightfoot 1979; I. Roberts 1985 for effects of loss of infinitival and subjunctive marking on verbal reanalysis). Critically, the loss of infinitival marking makes the simplex analysis sufficient and the conservative learner will not posit more structural features than necessary (see van Gelderen 2011a). The meaning of the string remains consistent, but the syntactic composition is reduced (fewer nodes) (see von Fintel 1995 for a similar approach using semantic types; see Fuß 2005 for comparable meaning-preserving string reanalyses for the pronoun > agreement cline). In sum, throughout development the child has interim analyses (forms) for syntactic strings consistent with the stage of her grammar; an interim analysis becomes her ultimate analysis if it: (a) captures the compositional meaning of the string, (b) is consistent with the rest of the grammar, and (c) is not sufficiently cued by the input to be complexified. In morphosyntactic reanalyses the child posits a new form (= structural analysis) for an existing meaning (the compositional meaning of a modalized verb). The new form in diachrony aligns with a simpler, earlier, stage in child development. .. The CIA and oppositional alignments: Response to critics Critics of the CIA argue that morphosyntactic oppositional pathways provide evidence against the CIA because these changes do not show child language parallels. For example, Diessel (2012: 1609) argues: While the semantic developments of grammatical markers are often parallel in language acquisition and diachronic change, the morphosyntactic developments are different. There is no evidence that grammatical markers originate from lexical expressions in language acquisition as they do in diachronic change. . . . The acquisition of the English present perfect, for instance, does not originate from an attributive construction as in diachronic change; and the be-goingto future is acquired in the context of a simple clause, whereas the historical development

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i



Ailís Cournane

originates from a bi-clausal purposive construction. . . . If children were the main instigators of language change, one would expect a closer match between the two developments. In fact, since the categorical and constructional changes of grammaticalization have no parallels in early child language, we can be fairly certain that grammaticalization processes do not originate from changes in language acquisition. (Diessel 2012: 1609)

This is an argument against a straw man; under the CIA, the child is not expected to recapitulate completed changes (be-going-to), neither for parallel alignments nor for oppositional ones. The child is expected to forge new innovations (see van Gelderen 2004, 2009a, 2011a), a point which is even more critical for oppositional pathways. When a child learner progresses from one stage of development to another, her only accountability is to the present system of her grammar, to the input (remembering she has access to different information from the input as she develops, see Gleitman et al. 2005), and to her language-learning capacity. She is tasked with creating productivity (e.g. categorizing objects, events, and propositions in the world around her and mapping meanings to forms). The child inevitably makes analyses that are input-divergent throughout this process (the example in (7) constitutes just such a not-yet-realized potential innovation for behind).12 In order to confidently claim that child input-divergence can or cannot be responsible for changes we need to look at systematic biases in, for example, the entire modal system of learners, and look to renewing items and how they are treated by learners. For modals, renewing items are lexical verbs that denote states of obligation, potential, desire, etc., such as to owe, to know how, or to want (see Lightfoot 1979; Tollan 2013). If we are to study child development for innovations, we need to look at current renewing items, like to know how, or to want. These verbs currently express the meanings that cunnan, willan exhibited in OE. As these items provide renewals to the linguistic cycle, the CIA predicts that children should show directional biases with these items in development (and not with be-going-to). For example, the CIA predicts that the verb want may become a future marker, so we should look at how children treat want. In sum, the existence of oppositional pathways does not refute the CIA because the CIA does not predict uniform relationships between all child phenomena and all change phenomena. The child must necessarily determine the compositional relationships between elements in a sentence; if an early (= less complex) analysis is successful at capturing the semantic composition of a sentence and is string-compatible (see von Fintel 1995; Fuß 2005), there is no internal motivation to posit a more complex analysis. Rather, different aspects of child development are predicted to underlie semantic overextensions and morphosyntactic reanalyses. This serves to illustrate that we must be domain-specific about our predictions, maintaining sensitivity to the expected renewing innovations and the actual child patterns observed in that domain of language development. As many have cautioned before me, overadherence to biology principles such as ‘ontogeny recapitulates phylogeny’ obscures the more 12 Anecdotally, use of behind for after appears regularly in the Baltimore, Maryland-based TV show The Wire, e.g. They lost their daddy behind what happened (SE). I have not yet pursued whether this is a feature of native speakers from the region or AAVE more generally.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

In defence of the child innovator



Table .. Summary of proposed alignments between L1A and change Type of input divergence Type of diachronic innovation Parallel alignment

Means of diffusion

Overextension Semantic change Old form > New meaning

Oppositional Structural reduction alignment Old meaning > New form

Peer-to-peer reinforcement Persistent input-divergence Morphosyntactic change >>sociolinguistic diffusion

fine-grained relationships between individual development and language changes. Table 2.1 summarizes my arguments.

. Conclusion Child input-divergent analyses remain a promising source of both directional semantic changes and directional morphosyntactic changes. The data shows compatibility once we allow that different developmental mechanisms for productivity and analysis underlie different types of innovations. For parallel alignments, the child posits new meanings for existing forms, extending the meaning coverage of a lexical item. For oppositional alignments, the child posits a new form for an existing meaning, thus changing the structural relationship between compositional elements (see Hale 1998 for why reanalysis of functional elements is particularly robust). While this approach is still relatively broad, it moves in the direction necessary for testing the validity of the CIA; generative diachronic theory can be brought to the level of domain-specific, fine-grained predictions for child language if we consider what child mechanisms align with advancements (not recapitulations) along a grammaticalization cycle. This broad approach provides opportunities for meaningful progress to generative change theory. By taking complexity in first language acquisition seriously, we can test specific hypotheses for generative change theory with data from contemporary child corpora and experiments. Furthermore, several avenues of research provide possible avenues for the reinforcement or diffusion of inputdivergent analyses (peer-to-peer first language acquisition, bilingual language acquisition, and established sociolinguistic diffusion mechanisms). More targeted research is needed to assess the empirical basis of diffusion of child input-divergent analyses for these populations, but the mechanisms and extant language contact situations are present. This chapter has responded to criticism of the generative approach to language change by recasting the CIA as part and parcel with the fundamental mapping problem in language acquisition, stressing that both child developments and diachronic

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i



Ailís Cournane

innovations reduce to redistributions of form ↔ meaning grammatical relationships. I suggest that forces that create productivity in the child’s developing grammar lead to input-divergent properties which later diffuse through normal peer-to-peer acquisition or sociolinguistic mechanisms of change (diffusion). After 100 years, the child innovator is still the most compelling source for innovative grammatical analyses, though we must assess more areas of child development at the right level of targeted formal analysis.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

 Where do relative specifiers come from? N I KO L A S G I S B O R N E A N D R O B E RT T RU S W E L L

. Introduction This chapter argues that the emergence of dependent wh-relatives in Middle English (ME) was an instance of diffusion of a novel syntactic property through the set of wh-forms, construed as a series of reanalyses of individual lexical items. This seemingly parochial claim takes on a broader significance in the context of two wellestablished findings about the typology of relativization. First, similar constructions (dependent relatives containing a filled Spec,CP, which we will call a relative specifier) have an unusual typological distribution: they are largely clustered in Indo-European languages (IE), but probably not present in Proto-IndoEuropean (PIE). This leads us to ask where relative specifiers come from, or more broadly, how daughter languages can share an otherwise rare property which they did not inherit from a common ancestor. Second, the research tradition stemming from Keenan and Comrie (1977) associates relative specifiers with positions low on the Accessibility Hierarchy (AH). Romaine (1980, 1982, 1984) describes the genesis of English dependent wh-relatives in just those terms: they were initially confined to low-accessibility functions, and subsequently spread up the hierarchy. This pattern is very common among languages with relative specifiers (Hendery 2012), and may be a diachronic universal. These findings invite analyses based on two tenets: (1) a near-complementarity between the initial distribution of dependent wh-relatives and that-relatives, the primary relativization strategy in Keenan and Comrie’s terms, which suggests that wh-relatives emerged to replace the earlier demonstrative series of relativizers in functions which could not be relativized by that (Romaine 1982: 450); (2) work by Keenan and Hawkins (1987); Hawkins (1995); and Kirby (1996) grounding the Accessibility Hierarchy in processing factors, which suggests that patterns described in the terms of the Accessibility Hierarchy should have explanations in those terms. These AH-based analyses ask ‘in language L at time t, which forms were available for which types of relativizaton?’. This approach leads to a surprising position, where formally Micro-change and Macro-change in Diachronic Syntax. First edition. Éric Mathieu and Robert Truswell (eds). This chapter © Nikolas Gisborne and Robert Truswell 2017. First published 2017 by Oxford University Press.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i



Nikolas Gisborne and Robert Truswell

distinct elements (such as demonstratives and interrogatives) are considered part of the same system, provided that they do the same job. Accordingly, we find quotes like this: distinct [relativization] strategies in [a complementary] relationship are no more different than complementarily distributed allophones. And just as different allophones of a particular phoneme are phonetically similar to each other, different [relativization] strategies in a given language must be syntactically similar. (Maxwell 1982: 142–3)

In contrast, we pursue a lexicalist approach (Kroch 1994): we consider a dynamically evolving population of forms and ask how they were used at a given time. By giving primacy to individual forms and the syntagmatic relations that they enter into, we walk away from the prospect of construing this aspect of language as ‘un ensemble où tout se tient’ (Meillet 1908). Likewise, we cannot construe this change as an instance of Meillet’s ‘renouvellement formel’. We believe this is for the best, for several reasons. First, modern generative syntax is lexicalist, almost without exception. In the past twenty-five years, theorists have converged on a model of syntax with two components: (1) a large amount of lexical items, each consisting of information stored in a particular format, and (2) a limited and probably invariant set of schemata for recursively combining lexical items. Such a description encompasses all flavours of Construction Grammar (Goldberg 1995; Culicover and Jackendoff 2005; Boas and Sag 2012), Categorial Grammar (Steedman 2000), and Dependency Grammar (Tesnière 1959; Hudson 2007), as well as Minimalism (Chomsky 1995). However, it excludes approaches such as the Standard Theory (Chomsky 1965) or the framework of Huddleston and Pullum (2002), in which a larger and in principle open-ended set of bespoke rules for combining syntactic units is specified. Second, the panchronic typological distribution of dependent relative specifiers can be explained in terms of two components: a general theory of directional change, and a distinctively Indo-European initial state. We claim that this initial state concerns the PIE *kw i-/kw o-forms, used in relative correlative constructions and elsewhere, from which the English wh-series, along with cognates in other IE languages, are descended. If this is accurate, a history of this family of lexical items can form part of a theory of the typological distribution of dependent relative specifiers. Finally, we believe that the facts (at least, the fine-grained details about the development of dependent wh-relatives in English) support this approach. In many respects, the development of ME dependent wh-relatives out of Old English free hw-relatives is an unremarkable, gradual syntactic and semantic change, largely explicable in language-internal terms. In contrast, certain details of the genesis of dependent wh-relatives do not lend themselves to a purely AH-based analysis. In this chapter, we focus on the diffusion of this change across lexical items, rather than across grammatical functions. Section 3.2 summarizes the diachronic typology of relative specifiers. Section 3.3 discusses work relating this typology to the Accessibility Hierarchy. Section 3.4 presents new data that suggest that a lexicalist alternative should complement the accessibility-based approach. Finally, Section 3.5 concludes.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

Where do relative specifiers come from?



. Relative specifiers in English and elsewhere .. Definitions We are concerned with three binary distinctions among relative clauses. First, dependent relatives (1a) are clauses adjoined within noun phrases, whilst free relatives (1b) are clauses with the external distribution of noun phrases (or of certain other categories— see Bresnan and Grimshaw 1978). (1)

a. I’ll have [NP the food [CP that she’s having b. I’ll have [NP [CP what she’s having ]].

]].

We assume the structure in (2) for either variety of relative clause.

CP

(2)

XPi C

IP ... ti ...

Two distinguished positions in (2) are the complementizer position, filled by that in (1a), and the Spec, CP position, filled by what in (1b). Either position can be filled or empty independently of the other, which gives us a 2×2 subclassification of dependent relatives.1 (3) the food

which that which she’s having that ∅

Although the food which that she’s having, with a doubly-filled comp, is ungrammatical in present-day English, structurally similar examples like (4) are attested in the fourteenth to fifteenth centuries,2 as described by Keyser (1975). (4) blisfulnesse is [NP that thing [CP [for whiche] that [IP alle thise othere thinges ben desired]]] (cmboeth, 434.C1.226) We call elements like for whiche ‘relative specifiers’, and elements like that ‘relative complementizers’. We also talk about dependent relative specifiers (relative specifiers in dependent relatives), and so on. The analysis of wh-phrases as specifiers and that as a complementizer correctly predicts that wh-phrases always precede that when they co-occur. That is, examples like (5) are never attested. 1 For reasons that we only partly understand, free relatives typically have a filled specifier and empty complementizer. 2 All examples are from the York–Toronto–Helsinki Corpus of Old English Prose (YCOE, Taylor et al. ) and the Penn–Helsinki Parsed Corpus of Middle English (PPCME, Kroch and Taylor ), unless otherwise stated. Identifiers such as cmboeth in () refer to individual files; following material identifies particular sentences within a file.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i



Nikolas Gisborne and Robert Truswell

(5) *the food [that which she’s having] Wh-specifiers are also phrasal. (4) contains the wh-PP for whiche, and examples as elaborate as (6) can be found in Middle English. (6) þe þrid degre and maner of leuing . . . [CP [PP In [NP þe whiche solitari forme and (cmcloud, 14.18) maner of leuyng]] C0 [IP þou maist . . . ]] Such complexity is no longer possible in present-day English relatives. However, we should be reluctant to analyse PDE wh-phrases as relative specifiers if none of them were phrasal. In contrast, complementizers, being heads, are syntactically simple. This lies behind Allen’s (1980) observation that that (and its Old English predecessor þe) never piedpipe prepositions. (7) a. the person [that we spoke [to b. *the person [[to that] we spoke

]] ]

Furthermore, relative specifiers, being the heads of filler–gap relations, may display connectivity effects, whereby their form reflects properties of the gap site. Accordingly, der in (8) shows nominative case, as determined by the gap, rather than the accusative case of the containing NP. (8) Ich fürchte den Herrn [der I fear the.acc man.acc who.nom ‘I fear the man who carries a gun.’

eine Pistole trägt]. a gun carries (De Vries 2002: 118)

As heads do not undergo phrasal movement, there should be no such connectivity effects with relative complementizers. However, complementizer alternations can still be conditioned by the gap site. For example, French qui occurs with subject gaps and que with object gaps. (9) a. l’ homme [qui est venu] the man that is come ‘the man that arrived’ b. l’ homme [que t’ as vu the man that you have seen ‘the man that you saw’

]

Following Rizzi (1990), we assume that qui and que are nevertheless relative complementizers, and the alternation results from local agreement rather than nonlocal connectivity. This distinguishes our relative complementizers from the traditional class of invariant relative particles (as opposed to inflecting relative pronouns). The structure we adopted in (2) makes the specifier–complementizer distinction a more natural one, despite the complications in interpreting alternations like that in (9). .. Typology Dependent relative specifiers fall into two main classes: demonstrative phrases and phrases built around interrogative pronouns like the wh-series. Both are largely

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

Where do relative specifiers come from?



Table .. Dependent relative specifiers in 172 languages IE Spec Wh Dem No Spec

27 19 8 13

(67.5) (47.5) (20) (32.5)

Other 7 3 4 125

(5.3) (2.3) (3) (94.7)

Source: based on De Vries (2002).

Indo-European phenomena.3 A 172-language survey (De Vries 2002) reveals that each subtype of dependent relative specifier is very significantly concentrated in IE languages (Table 3.1).4 However, this does not reflect inheritance from a common ancestor, as it is unlikely that Proto-Indo-European had dependent relatives at all. Based on comparison of attested early IE languages, Clackson (2007) argues that PIE probably had only adjoined relatives, for example in correlative constructions. Adjoined relatives are functionally similar to dependent relatives. However, their internal structure is closer to that of free relatives (Srivastav 1991). The synchronic typological skew of dependent relative specifiers therefore most likely reflects a recurring process whereby certain types of adjoined relatives are reanalysed as dependent relatives. However, the demonstrative and interrogative forms found in dependent relatives are directly descended from PIE ancestor forms. .. English Throughout its written history, English has had an invariant dependent relative complementizer, þe in Old English (OE), subsequently that. OE þe always strands prepositions (Allen 1980): there are 826 examples like (10a) in YCOE, but none like (10b). Partly because of this, the relative complementizer typically co-occurs with an NP gap. of ] geworhte] (10) a. þæt ribb [þe he þæt wif [ of made that rib that he that woman ‘the rib that he made that woman from’ (coadrian, Ad: 3.1.9) b. *þæt ribb [[of þe] he þæt wif geworhte] Two series of dependent relative specifiers have complemented the relative complementizer over the history of English. In OE, a full system of demonstrative relatives is 3 Comrie () claims that they are European, rather than Indo-European phenomena, or that they cluster areally rather than genetically. This is partly true, in that some Finno-Ugric languages also have dependent relative specifiers. However, they are also found in Indo-Aryan languages, so there is clearly a genetic component to their distribution too. 4 Indo-European languages are overrepresented in De Vries’ sample. However, this does not affect our conclusions, as we are only interested in a distinctive property of IE in comparison to other families.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i



Nikolas Gisborne and Robert Truswell

attested. Allen (1980) shows that these are relative specifiers, in our terms. First, they are phrasal, with obligatory piedpiping of prepositions (11). Second, they show connectivity effects, with the case marking typically determined by the gap site. For instance, seo in (12) is nominative, as mandated by the subject gap in the relative, rather than the accusative case associated with the object position of arærde. Third, the specifier precedes the complementizer ðe. (11)

his huse, [[of þam] ðe he his house of that.dat that he ‘his house, out of which he went’

ut ferde] out went (coaelhom, +AHom_4: 44.542)

(12) Ure Drihten arærde anes ealdormannes dohtor, [seo ðe our lord raised up a.gen alderman.gen daughter that.nom.f that læg dead digellice on his huse] lay dead secretly in his house ‘Our lord raised up an alderman’s daughter, who lay dead in the alderman’s house in secret.’ (coaelhom, +AHom_6: 176.960) Demonstrative relatives mainly disappeared with the collapse of case inflections in Early Middle English. Around the same time, dependent wh-relatives started to appear. OE wh-phrases had three functions which they shared with their cognates in many early IE languages: they occurred as NPI-like indefinites (13), as interrogative forms (14), and in free relatives (15). (13) and gif hwa hyt bletsað, þonne ablinð seo dydrung. and if who it blesses then ceases dem illusion ‘And if anyone blesses it, then the illusion is dispelled.’ (coaelhom, +AHom_30: 4.4082) (14) Saga me on hwilcne dæig he gesingode say me on which day he sang ‘Tell me which day he sang on.’

(coadrian, Ad: 2.1.4)

(15) eal swa hwæt swa ic þe gehet eal ic hit gesette all so what so I thee promised all I it appoint ‘Whatever I promised you, I will do it all.’ (coblick, LS_20_[AssumptMor[BlHom_13]]:147.155.1807) Although it is notoriously difficult to pinpoint the emergence of dependent whrelatives, (16) is a convincing early example, from the early twelfth century. (16) ungewædera [[for hwan] eorðwestmas wurdon swiðe amyrde] bad.weather for which earth.fruits were very damaged ‘bad weather, which seriously damaged the crops’ (cochronE, ChronE_[Plummer]: 1110.25.3499) Because dependent wh-relatives emerged as demonstrative relatives disappeared, it is tempting to see the two changes as two halves of a larger, systemic change, as English replaced one set of forms with another. Indeed, Romaine (1982), probably the bestknown work on the diachrony of wh-relatives, suggests as much. We summarize this

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

Where do relative specifiers come from?



work in Section 3.3. However, we think that such an account is incomplete, and give our reasons for moving beyond it in Section 3.4.

. Relative specifiers and noun phrase accessibility Dependent wh-relatives emerged gradually, over several centuries. The first examples have adverbial or oblique gaps, and examples with argument gaps only occur c.200 years later. This is surprising because dependent relatives with argument gaps, once they emerge, are the more common variety. A cluster of publications relate this to Keenan and Comrie’s (1977) Accessibility Hierarchy, given in (17). (17) SU > DO > IO > OBL > GEN > OCOMP (Keenan and Comrie 1977: 66) (SU: subject; DO: direct object; IO: indirect object; GEN: genitive; OCOMP: object of comparison) Among the findings related to this hierarchy are the following: • Relativization on positions higher in the hierarchy is more common than on lower positions. • If a language can relativize on a given position, it can relativize on any higher position.5 • Every language can relativize on subjects (the highest position). • For any position on the hierarchy, there are languages which relativize on that position, but no lower. • A language may have multiple relativization strategies, but every strategy covers a convex subset of the hierarchy. Keenan and Hawkins (1987) claimed that the Accessibility Hierarchy reflects a hierarchy of processing ease, with subject-gap relatives the easiest to process. This means that the typology implied by the Accessibility Hierarchy is an example of Performance– Grammar Correspondence (Hawkins 1995): parsing preferences in one language are reified as grammatical constraints in another. Kirby (1996) presents a series of simulations demonstrating that, given a parsing-based hierarchy like the Accessibility Hierarchy, Performance–Grammar Correspondence is predicted to emerge diachronically, as a product of a conflict between tendencies to minimize parsing complexity and morphological complexity in the ‘trigger experience’ on the basis of which a child induces a grammar. OE þe-relatives contained a gap in a high-accessibility position: subject, direct object, or complement of P. Demonstrative relatives were available across the whole hierarchy. When dependent wh-relatives first emerge, they always contain a gap in a low-accessibility position (genitive, adverbial, or oblique). That is, early dependent wh-relatives constitute a secondary relativization strategy, confined to the low end

5

This was subsequently slightly amended, but not in a way which concerns us.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i



Nikolas Gisborne and Robert Truswell

of the accessibility hierarchy, and in complementary distribution with the relative complementizer.6 Romaine (1980, 1982, 1984) presents the most fully developed account along these lines, based on a variationist analysis of relatives in Middle Scots texts from c.1530–50. She finds (1982: 151) that wh-relatives in her corpus are heavily concentrated in low-accessibility functions: 75 per cent of restrictive relatives with genitive gaps are wh-relatives, compared to only 14 per cent with subject gaps. The figures for nonrestrictive relatives are similar, but with a much higher incidence of wh-relatives across the board. Romaine also demonstrates that wh-relatives can be used as an index of the syntactic complexity of a text. She therefore comments that the wh-strategy enters ‘into the system “by the back door”, since it enters the most complex and least frequently occurring positions in the case hierarchy’ (1982: 152).7 Hawkins (1995: 448, fn.8) relates this to the following possible universal (see also Hendery 2012: 50): ‘if a language permits relativization using a [+case] strategy on a high position on AH, it permits it on all lower positions that the language permits relativization on.’ A [+case] strategy is one where the filler inflects to reflect information about the gap site; this includes relative specifiers. Hawkins notes that this prediction is borne out by all the languages in Keenan and Comrie’s 49-language sample, except for Tongan. Kirby (1996) explains this pattern in terms of complexity: a [+case] relativizer is more morphologically complex than a [−case] relativizer, but that added complexity can contain information which helps to identify the gap. Such information reduces parsing complexity, and is therefore most useful in low-accessibility positions where parsing complexity is greatest. A parallel analysis was developed by Maxwell (1979) for the fact that resumptive pronouns occur in low-accessibility positions, and Maxwell (1982) demonstrated that this implied that resumptive pronouns could only advance up the hierarchy or retreat down it: no other change is compatible with a processing-based universal that confines them to the bottom of the hierarchy. Kirby showed that the same logic can be applied to any [+case] strategy. We find all of this compelling. However, as an explanation of the genesis of dependent wh-relatives, it is still incomplete. The main reason is that Kirby’s simulations, which ground the above explanations of typological reflexes of the accessibility hierarchy, are models of selection among pre-existing variants. They do not contain an equivalent of mutation, or actuation. From a functionalist perspective, dependent whrelatives did not need to emerge: they may have expanded the structural repertoire of English by adding a strategy for relativization on low-accessibility positions, but

6 Like much research in this tradition, we do not stick to the letter of Keenan and Comrie’s original hierarchy: we ignore the OCOMP role; we collapse the IO and OBL functions; and we include gaps of categories other than NP, in particular PP and adverbial gaps, as this is essential for any complete account of early dependent wh-relatives. 7 Romaine describes this as syntactic diffusion in the sense of Naro and Lemle (), where endogenous syntactic changes begin in low-saliency environments, while borrowings occur first in highly salient environments. This appears to constitute indirect evidence that dependent wh-relatives are an internal development, rather than a borrowing from Latin or French, as assumed by Mustanoja () and Romaine herself. See also Truswell and Gisborne () for a description of aspects of this change in endogenous terms. However, a full discussion of borrowing is beyond the scope of this chapter.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

Where do relative specifiers come from?



languages do not need to relativize on low-accessibility positions (see Keenan and Comrie’s claim that only subject-gap relatives are universal). This problem is also reflected in the typological distribution of relative specifiers, as reported in Table 3.1. Relative complementizers tend to be examples of Keenan and Comrie’s [−case] relativization strategies, in the same way that relative specifiers tend to be [+case] strategies. [−case] strategies are universally associated with highaccessibility positions, just as [+case] strategies occur with low-accessibility positions. We might therefore ask why this low-accessibility strategy is so heavily concentrated in IE languages. Other languages have other strategies for low-accessibility relativization (for example, resumption, Maxwell 1979), so why do they tend to shun this one? Matters become worse if we consider the death of demonstrative relatives and the emergence of dependent wh-relatives jointly (a step presumably not taken by Romaine because demonstrative relatives were long gone by the sixteenth century). Both demonstrative relatives and dependent wh-relatives are [+case], and so predicted to be confined to the low end of the Accessibility Hierarchy. Processing-based ætiologies of the Accessibility Hierarchy therefore concur that multiple [+case] relativization strategies will compete with each other at the low end of the hierarchy. Diachronically, demonstrative relatives receded down the hierarchy at the same time as dependent whrelatives were climbing up it. In other words, dependent wh-relatives first emerged in those environments in which demonstrative relatives persisted longest. It is hard to imagine what functional pressure could motivate the innovation of a new [+case] strategy in such circumstances. The pattern is messy in places, because of orthogonal facts about the diachrony of inflected demonstratives, but we illustrate it with two clear examples. First, whererelatives like (18) are attested from the fourteenth century. (18) þe þyestre stedes [huer hi zelleþ hare cloþ the dark places where they sell their cloth ‘the dark places where they sell their cloth’

] (cmayenbi, 45.751)

However, relatives with there survive for at least a century beyond this: (19) is from the mid-fifteenth century. (19) every place [there as inquesyscyon (= inquisition) was made] (cmgregor, 201.1651) The emergence of dependent where-relatives may well have been delayed by competition between where and there. However, the fact remains that there was competition: the two forms coexisted as dependent relative specifiers, even in individual grammars (there are seventeen texts in PPCME2 containing both there-relatives and whererelatives). In contrast, demonstrative relatives with high-accessibility gaps died out before their wh-counterparts emerged. Late examples of demonstrative relatives with argument gaps can be found in conservative texts dating from around 1200 ad, such as (20). However, even the status of these examples is open to question, as they come from copies or translations of earlier texts. Demonstrative relatives with argument gaps are extremely rare in original texts from this period.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i



Nikolas Gisborne and Robert Truswell

her cumet on drihtenes nome]. (20) he is iblesced [þe þe here comes in lord’s name he is blessed that that ‘He that comes here in the lord’s name is blessed.’ (cmlambx1, 5.30) Dependent wh-relatives with argument gaps are, if anything, even rarer in Early Middle English. Isolated examples occur throughout ME, but they only occur with any regularity from the mid-fourteenth century onwards, for example in Wycliffe’s New Testament. (21) a very liÇt, [which

liÇtneth ech man that cometh in to this world] (cmntest, I, 1.19)

A graphical illustration may make this clearer. Figures 3.1 and 3.2 plot the rate of occurrence of demonstrative and wh-relatives over time, as a proportion of all relatives with locative gaps (Figure 3.1) and argument gaps (Figure 3.2). In Figure 3.1, dependent whrelatives with locative gaps are all but nonexistent until 1250–1500, when they abruptly become the majority variant. However, demonstrative relatives with there do not disappear at that time. Rather they continue to account for around 20 per cent of the dependent relatives with locative gaps throughout that period, before disappearing. Figure 3.2 shows that argument wh-relatives are all but unattested until 1250, when they start to increase in frequency, roughly following an S-curve. However, demonstrative relatives with argument gaps are long gone by then, disappearing abruptly around 1150. So while Figure 3.1 shows a 250-year period (1250–1500) where both types of relative specifier co-occur, Figure 3.2 shows a 100-year period (1150–1250) where neither type of relative specifier occurs.8 100

Frequency (%)

80 60 40 20 0 950–1000

1050–1100 1150–1250 1350–1420 1500–1570 1640–1710 Period Dem

Wh

Figure .. Demonstrative relatives and wh-relatives with locative gaps, as a proportion of all relatives with locative gaps, over time 8 The rates of occurrence for the two types of dependent relative at a given time period rarely sum to . This is mainly because of dependent relatives with empty specifiers, such as that- or ∅-relatives.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

Where do relative specifiers come from?



100

Frequency (%)

80 60 40 20 0 950–1000

1050–1100 1150–1250 1350–1420 1500–1570 1640–1710 Period Dem

Wh

Figure .. Demonstrative relatives and wh-relatives with argument gaps, as a proportion of all relatives with argument gaps, over time

This poses a real challenge to an account of the genesis of wh-relatives in terms of their relation to other relativization strategies. In the thirteenth century, English had a demonstrative strategy for forming relatives with locative adverbial gaps, but not with argument gaps. It developed a second strategy for locative gaps, which also initially did not extend to argument gaps. From this perspective, early dependent wh-relatives look quite useless. Moreover, the diachronic treatments of the Accessibility Hierarchy discussed above actually predict this situation: the two [+case] strategies are expected to cover the same portion of the hierarchy. The same would be true for two [−case] strategies, which would be expected to converge on the top of the hierarchy.9 This indicates that, although processing-based work on the Accessibility Hierarchy does an admirable job of predicting aspects of the diachrony of English relatives, it cannot account on its own for the actuation of those changes. We therefore explore an alternative, based on diachronic changes in the functions associated with a given form, rather than the forms that realize a given function.

. Unsystemic change Truswell and Gisborne (2015) argue that dependent wh-relatives emerge in English through reanalysis of clause-final free wh-relatives. Free wh-relatives in Old English can be clause-initial or clause-final, but rarely if ever clause-medial. Only clause-final free relatives are ever interpreted anaphorically. Example (22) shows the functional overlap between free and dependent relatives. The example contains two sentences, 9 This latter prediction appears slightly inaccurate, given the restricted distribution of [−case] zerorelatives with subject gaps in standard PDE.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i



Nikolas Gisborne and Robert Truswell

the first apparently with a dependent relative þa byrigeles hwar ic þe leigde, and the second with a free relative hwar ic hine byrede. There is only a small semantic difference between the two: the free relative denotes a goal, while the dependent relative denotes a location. (22) Þa cwæð ic to him, æteowe me þa byrigeles [hwar ic þe leigde]. then said I to him show me the tomb where I you laid Se Hælend me þa beo þære rihthand genam and me ut lædde The Saviour me then by the right hand took and me out led [hwar ic hine byrede] where I him buried ‘Then I said to him, “Show me the tomb where I laid you”. The Saviour then took me by the right hand and led me out to where I buried him.’ (conicodC, Nic_[C]: 149.161–2) Moreover, free relatives can be used in apposition to a preceding NP.10 In (23), for instance, it is unclear whether the hwanon-clause is an extraposed dependent relative or an appositive free relative. In fact, the same applies to the first relative in (22): although we stated above that it appears to be a dependent relative, it could equally well be a free relative in apposition to þa byrigeles. (23) þæt se ungesewena wulf infær ne gemete, [hwanon he in to Godes that the unseen wolf entrance ne find whence he in to God’s eowde cume & þær ænig scep of abrede] herd come.sbjv and there any sheep off snatch ‘that the unseen wolf may not find an entrance from where he might come into God’s herd and snatch any sheep.’ (cochdrul, ChrodR_1: 11.1.232) We claim that this is the ambiguous context which permits the emergence of dependent wh-relatives through reanalysis. Two further points strengthen this claim. First, free wh-relatives in OE and early ME are not universally interpreted as generalizing, as has often been claimed, e.g. Mitchell (1985: 65–6).11 They admit definite readings, as in (22), and are particularly likely to be interpreted as definite in clause-final position. Although definite free relatives do not have identical denotations to those standardly assumed for dependent relatives, the difference is quite small, and easily bridged. Again, we refer the reader to Truswell and Gisborne (2015) for details.

10 De Vries (, , ) has argued that appositive relatives should universally be analysed as false free relatives modifying a null determiner. Although this analysis clearly has many features in common with our own claim that appositive wh-relatives emerged through reanalysis of free relatives, the fine details of the diachrony are actually incompatible with De Vries’ synchronic analysis. False free relatives are essentially restrictive relatives with a syntactically light antecedent, whereas the history of English includes a stage in which wh-relatives were used appositively, but not yet restrictively. 11 We assume that a tendency to treat OE free wh-relatives as indefinite stems from the fact that OE whforms are also used as indefinite pronouns.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

Where do relative specifiers come from?



Second, adverbial wh-phrases are unusually common among clause-final definite free wh-relatives. These were the first dependent wh-relatives to emerge, as discussed in Section 3.3. This suggests that adverbial wh-relatives have distinctive distributional properties which may favour their reanalysis as dependent relatives, without direct reference to the Accessibility Hierarchy. The ambiguous context which we have just identified allows us to account for two facts that are puzzling on a purely AH-based account. First, the spread of dependent wh-relatives does not progress smoothly up the Accessibility Hierarchy, but rather from lexeme to lexeme, in a manner consistent with the hierarchy but not determined by it. Second, we will suggest (although scarcity of data will hinder us to an extent) that the later spread of dependent wh-relatives mimicked their initial emergence in ways which can only be explained by reference to the ambiguous context described above. .. Spread from form to form The emergence of dependent wh-relatives shows many hallmarks of gradual diffusion. If accessibility were the sole factor conditioning the spread of dependent whrelatives, we may expect the change to spread gradually up the accessibility hierarchy. That, however, is not what we find. In PPCME2, argument-gap wh-relatives are first attested in the period 1250–1420. There are seventeen texts in that period that contain argument-gap wh-relatives. All seventeen texts include both object-gap wh-relatives and subject-gap wh-relatives. The available data therefore suggest that the two constructions emerged simultaneously.12 The gradual spread of dependent wh-relatives proceeds instead from wh-form to wh-form. The few very early possible tokens, scattered through OE, are headed by when or where, like (24). (24) & bit [ðære tide, [hwonne he ðæs wierðe sie ðæt he and asked the time when he that.gen worthy be.sbjv that he hine besuican mote]]. him deceive may.sbjv ‘and asked when he would be fit to deceive him’ (cocura, CP: 33.227.10.1487) In Early ME, dependent wh-relatives encompass a range of new prepositional and oblique forms, with a few possessive forms also in evidence. Most texts, for example the Katherine group, use forms such as þurh hwam ‘through which’ with a range of prepositions (25), although some texts use forms such as hwer þurch ‘wherethrough’. 12 As with much research into Early Middle English, the force of these conclusions is weakened by the lack of corpus data for the period –. PPCME contains only three texts from this period: the late thirteenth-century Kentish Sermons, and the early–mid fourteenth-century Ayenbite of Inwyt and Earliest English Prose Psalter. The former contains no wh-relatives of either variety; both constructions are found in the latter, and in almost all later texts. Strictly speaking, then, the corpus data only show that any delay in the introduction of subject-gap dependent wh-relatives in comparison to object-gap dependent wh-relatives is very short, in the order of a few decades.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i



Nikolas Gisborne and Robert Truswell

(25) þer nis buten [an godd [þur hwam witerliche ha alle weren there neg.is but one god through which truly they all were iwrahte]] created ‘There is only one god, through whom, truly, they were all created.’ (cmkathe, 22.56) Around the middle of the fourteenth century, relatives headed by (the) which emerge. This looks like a gradual case of lexical diffusion. From an accessibility perspective, though, it is quite a dramatic leap, as dependent wh-relatives, previously confined to low-accessibility functions, abruptly become available throughout the hierarchy. (26) a. our Lord, [þe which makeþ sauf þe ryÇtful of heret] our lord, the which makes saved the pious of heart ‘our lord, who saves the pious of heart’ (cmearlps, 7.241) b. þe folk þat shal ben borne, [which our Lord made] the people that shall be born which our lord made ‘the people that shall be born, which our Lord made’ (cmearlps, 25.1031) However, as has frequently been observed, not all argument relativizers emerged at the same time. Whom-relatives with object gaps (27) are first attested around 50 years after which-relatives, particularly in the Wycliffite Bible. Last to emerge were subjectgap relatives with who, first attested in the late fifteenth century (28). (27) [he [whom God hath sent]], spekith the wordis of God

(cmntest, III, 20.234)

(28) seke euery man vpon his feblest and wekest / [who otherwyse wylle now haunte and vse the world] (cmreynar, 61.687) The spread from whom to who follows the Accessibility Hierarchy, of course, but only at a time when dependent wh-relatives with other relativizers had already generalized across the hierarchy. In other words, although every development we observe is compatible with established accessibility-based generalizations, those generalizations do not yield a complete account of the spread. The diffusion of dependent wh-relatives from lexeme to lexeme is gradual, while the spread up the hierarchy is quite abrupt. Moreover, the one clear example of a gradual spread up the hierarchy (the fact that object-gap whom precedes subject-gap who by over a century) occurs at a time when wh-relatives are already available throughout the hierarchy. To account for it in Maxwell’s (1982) terms, we would need to define who(m) as a distinct relativization strategy from (the) which. This would approximate a lexeme-by-lexeme account of the spread. .. Transmission and the reanalysis context Further evidence for the lexeme-by-lexeme nature of the spread comes from the position of the relative in the host clause. We discuss this below, but with some reservations due to the scarcity of corpus data from the crucial time periods.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

Where do relative specifiers come from?



We begin by giving a working definition of final. Many early English sentences come in two loosely related parts, with left-peripheral material such as clauses, NPs, or PPs prosodically isolated from the following host clause, as in (29). (29) [Þe stille nicht [hwenne me ne seið nawicht þt lette þe the silent night when man neg sees nothing that hinders the bone]]. þe heorte is ofte se schir prayer the heart is often so bright ‘On a silent night when one sees nothing that hinders prayer, the heart often shines so brightly.’ (cmancriw-1, II.115.1446) We consider a constituent to be final if it is string-final within either the left-peripheral material or the host, or if it is followed within either constituent by any combination of adverbials, parentheticals, conjuncts, and extraposed constituents. According to this characterization, the relative in (29) is final, because it is final within the left-peripheral material. The most obvious, but not the only, place where nonfinal material can be found is in any preverbal argument, or between such an argument and the verb. The relative in (30) is nonfinal, for example. (30) [þe eareste Pilunge [hwer of al þis uuel is]] nis buten of prude. the first stripping where of all this evil is neg.is but of pride ‘The first stripping, from where all this evil comes, is nothing but pride.’ (cmancriw-1, II.119.1506) The few dependent wh-relatives in English texts up to the Peterborough Chronicle are final, in this sense. This is expected, as the context of reanalysis identified at the start of Section 3.4 gives final position a special status. The four earliest nonfinal examples in the corpus, including (30), come from the early thirteenth century, roughly a century later. This pattern repeats itself with argument-gap wh-relatives. The earliest texts with any number of argument-gap wh-relatives are the mid-fourteenth-century Earliest English Prose Psalter and Ayenbite of Inwyt, which have over fifty tokens between them. All but one are clause-final, and nonfinal argument-gap relatives only occur with any regularity in the Wycliffite Bible, a generation later. If this is accurate, it suggests that argument-gap wh-relatives enter English through a reanalysis parallel to that which introduced adverb-gap dependent wh-relatives. Again, this would suggest that a description of the change in terms of spread up the Accessibility Hierarchy would be incomplete—after all, why would argument-gap wh-relatives be confined to final positions if they are just an extension of adverb-gap dependent wh-relatives, which at that time were no longer confined in that way? However, we stress that this conclusion remains fragile until confronted with further data. PPCME2 contains little material from the twelfth and thirteenth centuries, and often there are substantial temporal and dialectal discontinuities between texts. More data is required to strengthen this argument.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i



Nikolas Gisborne and Robert Truswell

. Conclusion We have argued for quite a subtle take on the Accessibility Hierarchy. Every change discussed here is compatible with the diachronic work on the hierarchy from Maxwell (1982) to Kirby (1996). Romaine’s (1982) initial glimpse of an accessibility-based account of the development of dependent wh-relatives is confirmed when confronted with an expanded dataset and a broader empirical context. This is particularly striking because the changes are counterfunctional in places: there is a global noncomplementarity in the distribution of the two series of relativizers, in that demonstrative relatives disappear last where wh-relatives appear first, and vice versa. This noncomplementarity arises from local functional considerations: demonstrative and interrogative forms are both [+case], and [+case] strategies are confined to the bottom of the hierarchy for functional reasons. To our knowledge, competition between two [+case] strategies has not previously been examined in detail. The fact that independently established generalizations predict such a counterintuitive pattern is a triumph for this work. At the same time, the explanations grounded in the Accessibility Hierarchy are incomplete. In particular, they have nothing to say about actuation, and they do not capture the fact that the spread of dependent wh-relatives proceeds from lexeme to lexeme. These are not accidental gaps: previous diachronic work on the Accessibility Hierarchy has focused to a very large extent on the dynamic interactions of multiple strategies, and has blurred or obliterated internal differences among exemplars of those strategies, except insofar as those differences reflect different positions on the hierarchy. However, we believe that we can discern the outlines of an analysis that preserves the best of the accessibility-based research, within an integrative lexicalist approach. The key lies in a very simple answer to the titular question of this chapter, ‘Where do relative specifiers come from?’. The answer is that the forms were there all along, being used in other ways. The genesis of dependent wh-relatives is not just the association of an established function (relativization) with a new set of forms; it is also the association of an established set of forms with a new function. Much the same is true from a more fine-grained perspective: the genesis of dependent which-relatives involves associating an established form with a function which it did not previously have.13 A learner who has figured out that his target language contains a form such as where has arguably done the easy bit. Next, he must associate the form with some specification of how it figures compositionally in syntactic and semantic structures. Clearly, a form may be specified in such a way that it can appear in interrogatives, restricted indefinites, and free relatives, as with most OE wh-forms. Equally, a form may be specified in such a way that it can appear in all of the above, and also in dependent relatives. These two potential specifications of the syntactic and semantic behaviour of a form are in competition as the learner induces his grammar. The competition is orthogonal to the competition generally taken to explain gradual grammatical change (Kroch 1989b, 1994), in that it concerns competing specifications of a given form, rather than 13 Sometimes, the form, the function, or both can be absolutely new to the language. When hwer þurch was first used as a relativizer in Early Middle English, the form was previously unattested and there was no established way of constructing such relatives. We hope to return to this in future research.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

Where do relative specifiers come from?



competing realizations of a given function. As emphasized in Section 3.4, though, these are two complementary ways of construing the same opposition. Any single instance of spread of dependent wh-relatives can be conceptualized in these terms: the learner chooses a less restrictive specification of the environments in which the form in question can occur. However, this leaves a further question: why do these individual changes occur in an order specified by the Accessibility Hierarchy? Our proposal relies again on competition, and on certain biases affecting the outcome of competition. Keenan and Comrie (1977) state that every language has a primary relativization strategy that relativizes functions high on the accessibility hierarchy. In Early Middle English, this strategy used the complementizer þe, with an NP gap. This primary strategy then acts as a brake on the emergence of other strategies for relativizing high-accessibility positions, because of the preference for a one-to-one association between forms and functions (Slobin 1985; E. Clark 1988; Markman and Wachtel 1988). In Early Middle English, after demonstrative relatives have mainly disappeared, there is no other widely attested strategy, and so no comparable brake on strategies for relativizing low-accessibility positions. This means that there is no hindrance to reanalysis of clause-final free relatives as dependent relatives in these cases. Once dependent wh-relatives with nonargument gaps are established, they may spread with the help of whichever biases underpin analogical extension, a topic which has been investigated in some depth with respect to phonological and morphological acquisition (Hare and Elman 1995; Gildea and Jurafsky 1996; Albright and Hayes 2003; Albright 2009), but should be equally relevant to syntax as conceived of here. The intuition is as follows: initially, certain wh-forms can appear in dependent relatives, but wh-forms are conceived of as a natural class because they tend to behave in similar ways. An expectation that wh-forms behave in similar ways could facilitate the spread of dependent relatives to new wh-forms. An account along these lines may have a hope of explaining the typological distribution of dependent relative specifiers. As noted in Section 3.2.2, the use of interrogative forms in dependent relatives is a largely Indo-European phenomenon. This makes little sense in purely accessibility-based terms: all languages have interrogative forms, so why do only IE languages use them in this way? One possibility is as follows: the early English use of wh-forms in clause-final free relatives creates an ambiguous context which allows dependent wh-relatives to emerge through reanalysis. Cognates of the wh-forms (descended from PIE *kw i-/kw o-) are widespread among IE languages, with similar distributions, so it would be unsurprising if such an ambiguous context were to recur across IE. More broadly, if there is something distinctive about *kw i-/kw o- which tends to be conserved in daughter languages, we can hope to build an account of the distribution of dependent relatives with interrogative specifiers on that distinctive property. Such an account would surely once again be phrased in terms of competition among possible analyses of the forms in question, as it is the forms, not the functions, that are distinctively Indo-European. There is a lot to do, then: diachronic typological work to establish how well the story sketched for English generalizes (both across languages and across series of relative specifiers), modelling work to determine whether the biases identified in this chapter

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i



Nikolas Gisborne and Robert Truswell

really do interact as described here, and regular diachronic linguistic work to fill in the gaps in the chronology and in the formal syntactic account thereof. We hope to address at least some of this in future research. This chapter has offered a starting point: it has emphasized the need for a lexicalist theory of syntactic change to complement the concerns underpinning most work on the Accessibility Hierarchy, and sketched a way in which patterns compatible with work on the hierarchy can emerge from a lexicalist approach to syntactic change.

Acknowledgements Thanks to the participants at DiGS 15 and ICHL 21, and to an anonymous reviewer. The title of this chapter is a slight modification of a section header in Romaine (1984): ‘Where do relative markers/clauses come from and where do they go?’.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

 Diachronic interpretations of word order parameter cohesion J O H N W H I T M A N A N D YO H E I O N O

. Introduction Since the 1970s, a body of research in the functional/typological tradition has claimed that crosscategorial consituent order regularities result from patterns of syntactic change (Givón 1975; Aristar 1991). The idea that Universal Grammar might not include crosscategorial parameters referring solely to word order has gained adherents in generative work over the past decade as well (cf. Newmeyer 2005; Whitman 2008). The objective of this chapter is to apply statistical tools to examine which word order features truly do ‘hang together’ from a cross-linguistic standpoint, and to explore how such correlations might be explained from the standpoint of generative accounts of acquisition and change. The main results we report in this chapter are the following: (i) Word order features show the clearest groupings in a multiple correspondence analysis (MCA) of the dataset in WALS Online (Dryer and Haspelmath 2013). (ii) MCA shows a clear grouping of head-initial versus head-final feature values and languages. (iii) Further analysis shows that NP-internal word order features cluster separately from word order features at the clausal level. However the distinction between NP-internal word order features and clause-level word order features is not simple. Order of head noun and genitive consistently clusters with clause-level word order features. We provide a diachronic explanation for this fact. The chapter is structured as follows: Section 4.2 presents our view of the status of crosscategorial word order regularities. Section 4.3 gives a brief outline of the kinds of syntactic change that result in such regularities, focusing on a language with relatively well-recorded historical material and time depth, Chinese. Section 4.4 presents our statistical study of the typological features in WALS Online (Dryer and Haspelmath 2013). Section 4.5 gives our diachronic interpretation of the results.

Micro-change and Macro-change in Diachronic Syntax. First edition. Éric Mathieu and Robert Truswell (eds). This chapter © John Whitman and Yohei Ono 2017. First published 2017 by Oxford University Press.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i



John Whitman and Yohei Ono

. Crosscategorial word order generalizations Crosscategorial generalizations (Dryer’s 1992 ‘Greenbergian correlations’) relate the internal properties of two or more categories irrespective of their relationship in a particular structure (Whitman 2008). Greenberg’s (1963) Universal 3 is an example of a crosscategorial generalization: Universal 3. Languages with dominant VSO order are always prepositional. (Greenberg 1963: 78) Universal 3 claims that in languages where the main clause is verb-initial, PPs are head-initial. This relationship is claimed to hold regardless of any relationship between clauses and PPs. If, for example, PPs in some language are head-initial when they are generated internal to the clause, but head-final when they are extracted from the clause, it might be possible to appeal to some aspect of the syntactic derivation to account for the constituent order. This is the strategy adopted by Biberauer et al. (2014) in the statement of their Final-Over-Final Constraint (FOFC): although the FOFC is a constraint on constituent order, it applies to the word order properties of one constituent embedded within another; Biberauer et al. are thus able to appeal to the derivational relationship between the two constituents to explain the word order facts. In general, however, this kind of strategy is not available to explain the kinds of crosscategorial generalizations stated by Greenberg. To take an additional example, Greenberg’s Universal 2 refers to the internal constituent order of PPs and NPs: Universal 2. In languages with prepositions, the genitive almost always follows the governing noun, while in languages with postpositions it almost always precedes. (Greenberg 1963: 78) Universal 2 is obviously designed to state a generalization about NPs and PPs as independent categories, not about, for example, PPs inside NP. Within generative grammar, the earliest attempt to capture crosscategorial generalizations was Chomsky’s (1981) Head Parameter, which was stated as a kind of metaconstraint on head/nonhead order in underlying structure. The Head Parameter was called into question by Kayne (1994), whose approach posits that head-final orders are derived, not underlying. Perhaps the most fundamental theoretical problem for a metaconstraint such as the Head Parameter is posed by a Minimalist approach to syntax, where cross-linguistic variation is localized in the features of functional heads. On this approach, whether the complement of a verb appears to its left or to its right is determined by the presence or absence of a feature on the functional head dominating VP that determines placement of the complement. These features are independent across lexical categories: there is no reason to expect that a feature in the functional head (say, v) determining the placement of complements of V is also found in a functional category heading adpositional phrases. From an empirical standpoint, the plausibility of crosscategorial generalizations or principles attempting to capture them such as the Head Parameter as components of Universal Grammar has been undermined by the discovery that all purely crosscategorial (that is, nonderivational) generalizations of this type appear to be statistical

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

Diachronic interpretations of word order parameter cohesion



(Dryer 1998; Whitman 2008). They all have exceptions. Our vastly improved access to typological information, due in no small part to the work of Dryer and Haspelmath (2013) and colleagues, makes it possible to confirm this. For example, a check of the strength of Universal 3 using the data in Dryer and Haspelmath (2013) finds six examples of VSO/postpositional languages distributed across five language families, as opposed to seventy-six VSO/prepositional languages. VSO/postpositional languages are relatively rare, but are not a geographically or genetically localized phenomenon. Greenberg himself states Universal 2 as a statistical generalization and a similar check confirms this: Dryer and Haspelmath (2013) list fifty-four prepositional languages with genitive–noun order and thirteen postpositional languages with noun–genitive order. We might amplify the point by considering a third oft-cited crosscategorial word order generalization, the one correlating order of head noun and relative clause with main clause order verb–object (Dryer 1992). Verb–object order correlates with noun–relative order: Dryer (2011) finds 415 VO languages with noun–relative order as opposed to only five VO languages with relative–noun order.1 Of these five, three are Sinitic; the other two, Bai (Tibeto-Burman) and Amis (Austronesian), have had intensive contact with Chinese. Bai has switched from OV to VO order under Chinese influence (Li and Yang 1999: 46), while Amis (VSO) is under intensive pressure from Chinese bilingualism. It seems likely that outside of contact-induced change, the combined word order property VO/relative–noun has arisen only within Sinitic.2 Yet aside from word order, there is no evidence that Chinese-type prenominal relatives are otherwise aberrant, for example that they are different from prenominal relatives in OV languages. Nor is there evidence that the exceptionality of the pattern makes acquisition of relative clauses by children learning Sinitic languages more difficult. Chen and Shirai (2015: 420) report production of relative clauses by three of the five Mandarin-acquiring children they studied at 1;4–2;0 (MLU 2.10), comparable to results for children acquiring English.3

. Reanalyses that propagate crosscategorial word order regularities The idea that sharing of word order properties between, for example, VPs and PPs might be the result of diachronic reanalysis originates in work such as Lord (1973) and was first articulated by Givón (1975). Givón and Lord discuss cases where a verb in a head-initial serial construction is reanalysed as a preposition. In such cases, reanalysis of original [VP V [VP V NP]] order results in [VP V [PP P NP]], that is, prepositional order. Whitman (2001) pointed out that under an approach that rejects radical reanalyses, the higher verb in a serial construction cannot be reanalysed as a P: assuming 1 Michael Cysouw points out to us in a very helpful review that from a statistical standpoint, Greenberg’s Universal  (P–NP → N–Gen) is actually the strongest of the correlations discussed here, followed by Universal  (VSO → P–NP) and VO → N–Rel. See the discussion in Cysouw (). 2 Or possibly Sino-Tibetan. Wright () identifies Hkongso as SVO, Rel–Noun. 3 Chen and Shirai do report a difference between the early Chinese and English data: early relative clauses in English tend to be predicate nominals (‘Here’s the book Daddy bought’), while Mandarin early relative clauses are isolated NPs (‘book Daddy bought’). But this difference may simply be due to the different status of pronominal subject and copula in the two languages. The copula is a clitic element in English, while Chinese has subject pro-drop.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i



John Whitman and Yohei Ono

successful acquisition of basic structural relations, the higher verb must continue to c-command the projection of the second verb. This prediction seems to be borne out by verbs with original lexical meanings such as ‘take, hold’, in the pattern [vP take DP [VP V eDP ]] such as Akan de (Campbell 1989), Chinese bˇa (Whitman 2001; Whitman and Paul 2005), and Late Old Chinese yˇı (Aldridge 2012). Instead of being reanalysed as a preposition, such verbs are reanalysed as a higher functional projection dominating the second VP. For example, Aldridge (2012) analyses yˇı as a high applicative. Functional/typological work on Chinese has shown that adjunct clauses in this language provide a rich source for V > P reanalysis (Sun 1996). Thus Mandarin cóng ‘from’ originates as a verb meaning ‘follow’, ‘accompany’. In Late Old Chinese both the verbal pattern (1a) and the PP (1b) pattern occur in preverbal adjunct clauses (Whitman and Paul 2005). (1)

zh¯ı dàif¯u [vP [TP PRO [vP cóng Jìn a. Xià, [TP zh¯uhóu feudal.lord gen high.offical follow Jin summer Qín]]] hóu]] [vP fá attack Qin duke ‘In summer, high officials of the feudal lords followed the duke of Jin and attacked Qin.’ (Zuozhuan: Xianggong 14; 5th–3rd c. bce) b. [VP [PP Cóng tái shàng] [VP tán rén]] from platform top shoot people ‘pro shot people from on top of the platform.’ (Zuozhuan: Xuangong 2; 5th–3rd c. bce)

However in the earlier Shang bone inscription texts (14th–11th c. bce) only the verbal ‘follow’ function is found. In contemporary Mandarin only the prepositional ‘from’ function remains. Chinese is a good example of how the word order properties of the source constructions in a change like cóng ‘follow’ > ‘from’ in (1) are inherited in the output constructions. It is a language with both prepositions, which can be shown to come from head-initial constructions, and postpositions, which can be shown to come from head-final constructions. Table 4.1 is a list of the prepositions and postpositions in contemporary Mandarin from Djamouri et al. (2013).4 As we see in Table 4.1, all of the monosyllabic prepositions have transitive verbal counterparts in contemporary or earlier stages of Chinese. (By transitive here we mean predicates taking two arguments; many, such as zài ‘be at’, have theme–location argument structure). Some of the verbs are still productive in the modern language, but others, like cóng ‘from’, have disappeared as verbs.5 The disyllabic prepositions can either be derived from the combination of a verb and preposition, such as duìyú ‘with 4

See W. Paul () for more extensive lists. In addition to cóng and wàng W. Paul (: ) lists two monosyllabic prepositions that lack verbal counterparts in the modern language: comitative hé ‘(together) with’ and benefactive wèi ‘for (the sake of)’. Comitative hé first appears in the Tang period (Ota : ). The source is a verb ‘harmonize with’. The preposition wèi (OC *Gw (r)aj-s; Baxter and Sagart : ) is morphologically derived from the verb wéi ‘make, do, be’ (OC *Gw (r)aj; Baxter and Sagart : ). Although there are various theories about the semantics of the derivation, the direction of derivation through the OC suffix *-s is clear. 5

i

i i

i

i

i

i

i Table .. Sources for Chinese adpositions b. Postpostion < Relational noun, Vintr

cháo cóng d¯ang(zhe)

‘facing’ ‘from’ ‘at’

cháo cóng d¯ang

‘face’ ‘follow’ ‘correspond’

hòu lái lˇı

‘behind; after’ ‘for, during’ ‘in’

hòu lái lˇı

dào duì duìyú gěi g¯en g¯enjù gu¯anyú lí tì wˇang wèi(le) xiàng yán(zhe) zài zhìyú zìcóng

‘to’ ‘toward’ ‘with respect to’ ‘to; for’ ‘with’ ‘on the basis of ’ ‘concerning’ ‘from, away’ ‘instead of, for’ ‘toward’ ‘for the sake of ’ ‘to, toward’ ‘along’ ‘in, at’ ‘concerning’ ‘since, from’

dào duì duì yú gěi g¯en g¯enjù guan yú lí tì wˇang *Gw (r)aj-s xiàng yán(zhe) zài zhì yú zì-cóng

‘arrive’ ‘be opposite, face’ ‘face to’ ‘give’ ‘follow’ ‘follow+’ ‘relate to’ ‘depart’ ‘replace’ ‘go toward’ ‘do-sfx’ ‘face’ ‘go along’ ‘be at’ ‘extend to’ ‘from-follow’

nèi páng qián qiánhòu shàng shàngxià wài xià zh¯ong yˇılái yˇınèi yˇıqián yˇıshàng yˇıwài yˇıxià yˇıhòu zh¯ıji¯an zuˇoyòu

‘inside’ ‘next to’ ‘in front of ’ ‘around’ ‘on’ ‘around’ ‘outside’ ‘under’ ‘amidst, in’ ‘since, during’ ‘inside, within’ ‘before, ago’ ‘above, over’ ‘outside, beyond’ ‘under, below’ ‘after’ (time) ‘between’ ‘around’

nèi pang qián qiánhòu shàng shàngxià wài xià zh¯ong yˇı-lái yˇı-nèi yˇı-qián yˇı-shàng yˇı-wài yˇı-xià yˇı-hòu zh¯ıji¯an zuˇo+yòu

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi

a. Preposition < Vtr

‘rear’ ‘come’ ‘village, measure of distance’ ‘interior’ ‘side’ ‘front’ ‘front–back’ ‘rise, above’ ‘above–below’ ‘exterior’ ‘below’ ‘middle’ ‘appl-come’ ‘appl-go.in’ ‘appl-go.before’ ‘appl-go.up’ ‘appl-go.out’ ‘appl-go.down’ ‘appl-go.after’ ‘gen-interval’ ‘left + right’

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i



John Whitman and Yohei Ono

respect to’ < duì ‘face’ + yú ‘to’; or from disyllabic verbs, such as g¯enjù ‘on the basis of ’ < ‘based on’. In the case of more recent reanalyses, we can find evidence for the internal structure of the source adjunct clause. For example, d¯ang(zhe) ‘at’ < ‘correspond’ and yán(zhe) ‘along’ < ‘go along’ optionally incorporate the durative particle -zhe, while wèi(le) ‘for the sake of ’ can include the perfective particle -le. Assuming that such aspectual morphology was external to VP in the source construction, this indicates that the reanalysed adjunct projection was higher than VP: it contained at least aspect, and a position for the external argument, which as shown in (1a) was controlled by the matrix subject. Reanalysis to P involves not just relabelling, but ‘pruning’ of the functional structure above VP (Whitman 2001). The sources for postpositions are more complex. Examples like pang ‘next to’ < pang ‘side’ are derived from relational nouns. Disyllabic zh¯ıji¯an ‘between’ < gen-‘interval’ involves the same derivation except that the genitive particle zh¯ı is retained between the relational noun and its complement. Disyllabic zuˇoyòu ‘around’ involves a reanalysis of the compound relational noun zuˇo + yòu ‘left + right’. In all these cases, the fact that NPs in Chinese at all stages of its history are head-final explains why N > P reanalysis results in postpositional phrases. The existence of N > P reanalyses in particular is important for the discussion in this chapter, because it is a diachronic point of contact between NP-internal word order and word order elsewhere in the clause. Djamouri and Paul (2012) point out that postpositions like lái ‘for, during’ < ‘come’, also qˇı ‘starting from’ < ‘arise’, result from intransitive verbs. Again, the reanalysis preserves the source word order: NP–V > NP–P. Disyllabic postpositional phrases in yˇı derive from the applicative head yˇı, the fronted applicative object, and verbs such as lái ‘come’ or shàng ‘rise, go up’. The variation in headedness in Chinese PPs, then, is a direct diachronic result of the difference between head–complement order in VP and NP in Chinese, and the involvement of both transitive and intransitive verbs as sources for Ps. It appears to be cross-linguistically common for both verbs and nouns to be sources for P.6 A count of the grammaticalization patterns in Heine and Kuteva (2002) involving VP, NP, and PP gives forty-eight patterns involving N > P reanalyses and twenty involving V > P. Returning to the general issue of what kinds of reanalyses propagate word order regularities across categories, at first glance what appears to be missing is N > V and V > N reanalyses. Aristar (1991) points out that grammaticalization in the conventional sense of reanalysis from lexical to functional category cannot account for constituent order symmetries involving N and V, since both are lexical categories. However, there is one attested mechanism for NP-to-VP (or clause) reanalysis: reanalysis of a nominalized clause as a main clause. Harris and Campbell (1995: 246–7) cite cases including Gildea’s work on Cariban (cf. Gildea 1998); other examples include Starosta et al. (1982) and Kaufman (2009) on Austronesian, and Claudi (1993) for Niger–Congo. The prevalence of reanalysis of nominalizations as main clauses in East and Southeast Asia was first pointed out by Matisoff (1972). Individual studies of nominalization > main clause reanalyses in these areas are collected in Yap et al. (2011). 6 See, for a combination of serial verb and relational noun sources particularly similar to Chinese, DeLancey () on Burmese and Tibetan.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

Diachronic interpretations of word order parameter cohesion



Taking into account the reanalysis of nominalizations as main clauses, and our discussion of Chinese PPs, we see that there are two routes for NP-internal word order to provide a source for word order internal to other categories. NPs headed by relational nouns can be reanalysed as PPs, or nominalizations can be reanalysed as clauses. In both cases the reanalysis involves the head noun, reanalysed as a P or V, and its arguments, typically realized as a genitive. A transparent example is the case of Chinese zh¯ıji¯an ‘between’ in Table 4.1, where the earlier Chinese genitive marker zh¯ı is exceptionally retained as part of the reanalysed P. This fact might lead us to predict that noun–genitive order could play a special role in any diachronic link between NP-internal word order and word order in other categories. The statistical analyses we present in the following section bear this prediction out. Our next objective in this chapter is to present a statistical study of how word order feature values in WALS pattern together, using the basic methodology pioneered by Tsunoda et al. (1995). We return to the issue of how the synchronic patterns we find relate to patterns of change in Section 4.5.

. Statistical analysis .. Overview Much of the focus in statistical typology (see the review in Cysouw 2005) has been on quantitative methods to reassess issues debated in earlier, mainly qualitative work, such as the veridicality of implicational universals. Tsunoda et al. (1995) was a pioneering attempt to identify the main component in a set of typological parameters. Rather than attempting to group languages by typological features, Tsunoda et al. investigate the statistical relations between parameters across a sample of languages.7 We use a similar approach to investigate interrelationships between the values of 450 features for a sample set of 201 languages in The World Atlas of Language Structures Online (Dryer and Haspelmath 2013; below WALS). We apply a form of multiple correspondence analysis (MCA) to the feature values in WALS, in addition to several types of cluster analysis. These analyses shows three notable results. First, even given a set of features that include not only word order but the entire set of syntactic, morphological, lexical, and phonological features collected in WALS, word order feature values show the clearest groupings. Second, MCA groups together head-initial and head-final feature values, along the lines of the Head Parameter. Finally, MCA shows four distinct groupings of word order parameters. Features involving word order at the clausal level (order of subject, object, and verb; object and verb, adposition, and NP complement), as well as order of noun and genitive, make up one cluster each for head-initial and head-final values. Features referencing a subset of word order features internal to NP (order of demonstrative and noun, adjective and noun, and numeral and noun) form two additional, distinct clusters, one each for head-initial and head-final orders.

7 Cysouw (: ) calls this a ‘reversed’ approach, grouping parameters and then languages, as opposed to the standard approach in statistical typology of first grouping languages and then parameters. Cysouw himself provides an example of this approach using a set of morphological parameters.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i



John Whitman and Yohei Ono

.. WALS (Dryer and Haspelmath ) The complete list of features in WALS is accessible at http://wals.info/feature. Features are coded so that 1A, for example, designates the feature [consonant inventories]. Feature 1A has five values: small, moderately small, average, moderately large, and large. We represent the values as 1A_1, 1A_2, 1A_3, 1A_4, and 1A_5. (For the sake of processing convenience using the R software environment used for this research, we denote, for example, 12A_1 as X12A_1).8 .. Multiple Correspondence Analysis We applied a form of multiple correspondence analysis (MCA; Benzécri 1973) to the WALS data. MCA is a non-metric principal component analysis for categorical data. Principal component analysis (PCA) yields a multidimensional figure showing the interrelationships among components—in the case of WALS, typological feature values and languages. Each dimension in the figure is associated with a quantity called an eigenvalue which indicates how much of the variance in the total dataset that dimension accounts for. Usually only a few dimensions with larger eigenvalues are taken into account. Since the WALS dataset is categorical, we used MCA, specifically, a technique known as Hayashi’s Quantification Method III (QMIII below; Hayashi 1952). For further details, see the Online Appendix at http://www.oup.co.uk/companion/DiGS15. Applying MCA required transforming the original categorical data into binary data. The total number of values associated with the 192 features gives some 1,200 categories. We reduced the number of these categories to 489 to eliminate those features for which WALS gives values for fewer than ten languages. We also removed ‘complex’ word order features which correlate more than two parameters, since these duplicate information contained in other features. For example, information contained in features 83A (order of O and V) and 85A (order of adposition and noun) is duplicated in feature 95A (the relationship between order of O and V and the order of adposition and noun).9 The resulting total number of feature values was 450. Our sample of 201 languages was chosen based on the recommended WALS 200-language sample (http://wals.info/languoid/samples/200), which is balanced by genetic affiliation and region. MCA yields a set of multidimensional coordinates, each of which shows the location of a category in a multidimensional space. In this case, the categories consist of feature values and languages in WALS. The coordinates show the interrelationships between categories: the shorter the distance between a pair of categories, the closer their interrelationship. In practice, it turned out that the number of feature values was too large to generate an overall figure clearly representing the interrelationships between all the values under consideration. 8 The R software environment is an open source language and environment widely used in statistical computing. For more information see http://www.r-project.org/about.html. For an introduction to the use of R to analyse linguistic data, see Baayen (). 9 The complex word order features eliminated were, B–G, A–A, and A–Y, (except A, E, F, G, B, V and X). We thank an anonymous reviewer for pointing out the desirability of excluding complex features.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

Diachronic interpretations of word order parameter cohesion



eigenvalues

0.12 0.08 0.04 0.00 0

50

100 Index

150

200

Figure .. Scree plot of eigenvalues from MCA of the WALS dataset

Figure 4.1 shows the scree plot of eigenvalues in descending order. Eigenvalues and contribution ratios for MCA up to the eighth dimension in ascending order are: (1) 0.1429, (2) 0.1200, (3) 0.1081, (4) 0.0831, (5) 0.0659, (6) 0.0635, (7) 0.0575, and (8) 0.0552. The cumulative contribution ratios from the first through eighth dimensions are: (1) 4.35, (1–2) 8.01, (1–3) 11.31, (1–4) 13.84, (1–5) 15.85, (1–6) 17.84, (1–7) 19.54, and (1–8) 21.22. There is a gap in the relative difference of eigenvalues between the third and fourth dimension, so we chose the third dimension as a cutoff point for calculating the set of distances between pairs of feature values on MCA coordinates.10 In comparison to PCA, the cumulative contribution ratio usually results in a lower score in MCA.11 Generally, the output of the MCA coordinate system is normalized with mean = 0.0 and variance = 1.0 along each dimension so as to represent the figure in a reasonable space.12 We thus obtain a set of distances {dij } (distance here is what is known as Manhattan distance). (2) shows the three-dimensional coordinates for the a sample set of word order values:13

10 We carried out several statistical trials on the first four, five, six, seven, and eight dimensions. We then observed the stability of the results of clustering with respect to dimensionality, as well as with respect to the type of clustering technique used. We finally decided to consider the dimensions up to the third dimension. We cannot exclude the possibility that some other choice of dimension cutoff might lead to more illuminating results. 11 We explain this problem in detail in the Online Appendix. 12 In mathematical terms, this means that the output consisting of the coordinates in each dimension of MCA is multiplied by the inverse of the corresponding eigenvalue. Thus, we had to weight back the eigenvalue for each coordinate dimension when we calculated the ‘real distance’ dij between pairs of categories of features, i and j, as below:  200 3    √  dij = ωk Xik − Xjk , where ωk = λk λt

k=1

t=1

λk is the eigenvalue of the kth dimension, and Xik is the value obtained for feature value i in the kth dimension. (The maximum rank of the data matrix is ). 13 We actually performed MCA up to the th dimension. The coordinates listed here correspond to the first three dimensions of the -dimensional space.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i



John Whitman and Yohei Ono X130A_1

X7A_6 X103A_5

X100A_4 X18A_3

0.5

X83A_3 X49A_9

X4A_1

X24A_1 X1A_1

X48A_4

X84A_6

X124A_2

X125A_1

X33A_1 X69A_5

X50A_1

X31A_3

X49A_1

X51A_9 X88A_2

X26A_1

X91A_3 X83A_2 X105A_4

X111A_1

X110A_1

X112A_4 X84A_1X85A_2 X13A_3

X86A_2 X81A_2

X78A_4

X104A_5

X58A_1 X8A_4 X2A_1 X108B_1

X5A_1

X38A_4

X143A_2 X33A_9

X101A_4

X98A_1 X117A_3

X21A_5

X66A_4

X91A_2 X14A_7

X105A_2 X28A_1

X110A_3

X123A_3

X81A_3

X16A_1

X116A_6

X99A_1

0.0

X102A_5

X108A_1

X109A_1 X131A_6 X87A_3 X75A_2 X105A_3 X109B_1 X108A_2 X82A_3 X33A_8 X104A_3 X40A_2 X92A_3 X39A_5 X122A_2 X59A_2 X22A_5 X109B_4 X74A_1 X1A_2 X74A_3X107A_2 X29A_3 X101A_2 X109A_3 X106A_1 X98A_4 X76A_3 X80A_3 X136B_1 X118A_1 X117A_4 X90A_7 X15A_7 X51A_6 X32A_2 X50A_3 X77A_2 X49A_7 X109B_2 X112A_2 X71A_3 X66A_2 X131A_3 X136A_1 X3A_4 X45A_1 X124A_4 X25A_3 X3A_2 X123A_2 X144B_4 X15A_8 X99A_4 X78A_2 X34A_4 X6A_2X19A_1 X44A_6 X28A_2 X10A_1 X137B_2 X17A_1 X129A_2 X94A_2 X79A_4 X68A_4 X36A_2 X11A_1 X143E_1 X64A_2 X5A_4 X47A_2 X4A_2 X111A_4 X20A_7 X12A_2 X94A_5 X75A_3 X70A_1 X46A_1 X6A_4 X27A_1 X13A_1 X6A_1 X24A_4 X43A_2 X143F_4 X33A_7 X1A_5 X13A_2 X86A_1 X16A_2 X9A_3 X3A_5 X143A_14 X103A_2 X87A_2 X43A_1 X41A_3 X119A_1 X83A_1 X12A_1 X127A_3 X88A_1 X60A_4 X113A_1 X114A_7 X103A_4 X143F_1 X30A_1 X31A_1 X32A_1 X112A_6 X35A_6 X50A_2 X9A_1 X42A_1 X143G_4 X89A_2 X2A_2 X85A_1 X22A_4 X50A_4 X71A_4 X35A_7 X126A_1 X62A_7 X55A_3 X79B_5 X70A_5 X37A_2 X7A_1 X33A_6 X73A_2 X25B_2 X35A_5 X37A_5 X38A_5 X126A_3 X144B_3 X3A_1 X113A_3 X16A_6 X81A_1 X21B_1 X17A_5 X137B_1 X7A_2 X137A_1 X52A_2 X23A_3 X20A_1 V72A_4 X10A_2 X120A_2 X113A_2 X143E_4 X93A_1 X15A_3 X69A_2 X77A_1 X78A_1 X65A_2 V72A_1 X48A_3 X82A_1 X111A_2 X58B_1 X18A_1 X127A_1 X67A_1 X92A_6 X16A_7 X37A_1 X14A_1 X118A_3 X14A_2 X116A_1 X100A_2 X143A_15 X77A_3 X28A_4 X25A_5 X12A_3 X44A_2 X51A_1 X70A_4 X4A_3 X93A_2 X128A_1 X3A_3 X143F_2 X71A_2 X40A_3 X92A_2 X30A_5 X22A_3 X120A_1 X55A_1 X114A_3 X41A_2 X90A_1 X57A_2 X27A_2 X29A_2 X125A_3 X69A_4 X89A_1 X35A_4 X71A_1 X42A_2 X143A_3 X114A_1 X94A_1 X76A_2 X58A_2 X63A_2 X62A_1 X8A_2 X108A_3 X108B_4 X80A_1 X53A_5 X39A_3 X49A_8 X33A_2 X29A_1 X36A_3 X36A_1 X27A_3X143E_2 X26A_3 X130A_2 X1A_3 X122A_4 X91A_1 X16A_4 X19A_5 X23A_4 X84A_3X112A_1 X48A_2 X44A_3 X40A_1 X56A_1 X106A_2 X55A_2 X67A_2 X87A_1 X129A_1 X43A_5 X21A_1 X63A_1 X2A_3 X115A_1 X21B_2 X107A_1 X131A_1 X73A_1 X123A_4 X49A_2 X31A_2 X9A_2 X65A_1 X30A_2 X59A_1 X110A_2 X74A_2 X56A_2 X53A_6 X26A_2 X54A_1 X109B_5X30A_3 X109A_8 X119A_2 X64A_1 X1A_4 X17A_4 X92A_1 X114A_5 X5A_2 X100A_1 X103A_1 X102A_1 X15A_5 X54A_4 X44A_1 X35A_8 X62A_2 X38A_1 X101A_5 X46A_2 X47A_1 X124A_1 X53A_4 X118A_2 X57A_4 X117A_5 X68A_3 X104A_1 X36A_4 X5A_3 X131A_2 X24A_2 X121A_1 X61A_2 X105A_1 X79A_1 X52A_3 X14A_6

X104A_2

X82A_2

X69A_1 X144X_4 X144V_1 X26A_6

X34A_3 X106A_4 X62A_8

X57A_1

X104A_4

X100A_6 X26A_4 X81A_7

X53A_1 X138A_3 X8A_1

X59A_3

2dim

X85A_5

X48A_1

X25A_1

X35A_3 X23A_1

X40A_5 X114A_2 X121A_3 X86A_3

X26A_5

X143A_1

X121A_2

X125A_2

X70A_2

–0.5

X32A_3

X142A_2

X126A_2

X43A_3

X136B_2

X101A_1

X4A_4

X79B_2

X52A_1

X98A_2 X66A_1

X34A_6 X56A_3

X54A_2 X23A_2

X102A_2 X117A_1

X138A_1

X143A_4

X94A_4 X49A_6

X50A_6 X25A_2

X42A_3

X22A_2

X135A_1 X138A_2

X116A_2

X90A_2

X99A_2

X60A_6

X37A_4

X127A_2

X24A_3

X38A_2

X128A_2

X45A_2

X136A_2 X62A_3

X75A_1

X28A_3

X53A_7 X142A_1

X76A_1 X79A_3

X121A_4 X123A_1

–0.5

0.0 1dim

0.5

Figure .. First and second dimension coordinates of 450 feature value categories generated by MCA

(2) Output of MCA showing sample word order feature values X81A_1 [SOV] = (0.3687 0.0205 0.2327)14 X81A_2 [SVO] = (−0.5496 −0.2091 −0.0019) X83A_1 [OV] = (0.3635 0.0629 0.2270) X83A_2 [VO] = (−0.5139 −0.0486 −0.1205) X85A_1 [Postpositions] = (0.3177 0.0436 −0.1803) X85A_2 [Prepositions] = (−0.5676 −0.1265 −0.1418) As shown in (2), the first dimension coordinates for word order feature values all group around 0.3 or −0.6. [SOV], [OV], and [Postpositions]—all head-final feature values— group on the positive side of the first and second dimensions, while head-initial values [SVO], [VO], and [Prepositions] group on the negative side. Figure 4.2 shows the first and the second dimensions of MCA. The circle on the positive side of the first dimension encloses head-final feature values, while the circle on the negative side of the first dimension encloses head-initial values. Figure 4.3 shows that head-initial and head-final feature values group on opposite sides of the first dimension. It also shows that two distinct subgroups can be distinguished in each. The leftmost and rightmost circles in Figure 4.3 group clause-level head-initial (SVO, VO, P–NP) and head-final (SOV, OV, NP–P) values respectively. The circle second from the left groups head-initial values (N–Adj, N–Demonstrative, N–Numeral) while the circle second from the right groups head-final values (Adj–N, Demonstrative–N, Numeral–N), internal to NP. Only one NP-internal feature, 86A 14 Strictly speaking, the feature ‘order of subject, object, and verb’ is a complex feature, as it includes information also contained in the features ‘order of subject and verb’ and ‘order of object and verb’. However ‘order of subject, object, and verb’ is the only feature in the WALS database that directly encodes information about verb-initial (VSO, VOS) and verb-final (SOV, OSV) order. We decided therefore to retain it in our analysis.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

Diachronic interpretations of word order parameter cohesion



X101A_5

X100A_1 X102A_1

0.5

X103A_1

X29A_1

X143A_2 X112A_4

X40A_1 X5A_4 X26A_1 X55A_3

X144B_4

X69A_5

X121A_2

X53A_4 X13A_3 X143F_1

X143E_4

X89A_2

X91A_2

X33A_7 X70A_5 X23A_4

X33A_9

X70A_4

X34A_4

X104A_1

X125A_1 X105A_4

X110A_3 X51A_9

3dim

X110A_1 X50A_1

0.0

X49A_1

X81A_2

X142A_2

X101A_4

X111A_1 X84A_1

X33A_1

X123A_3 X86A_2 X26A_6 X85A_2

X30A_5

X83A_2

X69A_1

X31A_3 X94A_1

X70A_2 X82A_2

X81A_3 X144X_4 X144V_1

–0.5

X124A_2

X116A_2 X143A_4

X51A_6 X90A_2 X28A_4 X124A_4 X81A_1 X143F_2 X98A_4 X105A_2 X83A_1 X25A_2 X9A_1 X94A_5 X123A_2 X64A_2 X143A_14 X106A_1 X48A_2 X60A_4 X20A_7 X77A_3 X3A_1 X27A_2 X114A_1 X37A_4 X57A_4 X112A_6 X85A_1 X66A_2 X107A_2 X114A_5 X55A_2 X8A_1 X123A_4 X63A_2 X112A_1 X124A_1 X93A_2 X1A_1 X87A_2 X101A_1 X71A_2 X131A_6 X86A_1 X4A_2 X126A_3 X50A_2 X84A_3 X109B_4 X23A_2 X13A_2 X14A_7 X3A_2 X82A_1 X5A_3 X111A_4 X92A_6 X12A_2 X30A_1 X31A_1 X32A_1 X54A_4 X74A_3 X120A_2 X33A_8 X26A_2 X43A_1 X79A_4 X38A_1 X44A_6 X76A_3 X118A_3 X114A_3 X119A_1 X113A_2 X53A_5 X6A_1 X127A_3 X47A_1 X67A_2 X1A_3 X36A_3 X118A_1 X131A_3 X49A_7 X73A_1 X105A_1 X52A_2 X66A_4 X21A_1 X37A_5 X42A_3 X35A_5 X90A_7 X143E_2X38A_5 X78A_2 X129A_1 X24A_2 X28A_1 X19A_1 X65A_2 X75A_3 X9A_2 X126A_1 X127A_1 X71A_4 X69A_2 X56A_2 X54A_2 X113A_3 X7A_1 X62A_1 X15A_5 X3A_3 X79B_5 X42A_2 X76A_2 X125A_3 X59A_2 X18A_1 X8A_2 X75A_2 X59A_3 X117A_4 X1A_2 X11A_1 X62A_8 X122A_4X39A_5 X78A_1 X77A_1 X50A_3 X106A_2 X74A_1 X99A_1 X7A_2 X116A_6 X34A_3 X115A_1 X5A_2 X143G_4 V72A_4X138A_3 X4A_4 X22A_2 X35A_4 X73A_2X121A_3 X21B_1 X45A_1 X128A_1 X98A_2 X62A_7 X60A_6 X68A_3 X46A_2 X10A_1 X131A_1 X98A_1 X74A_2 X80A_1 X4A_1 X39A_3X41A_2 X16A_2 X27A_1 X49A_9 X136A_1 X5A_1 X10A_2 X2A_2 X116A_1 X111A_2 X99A_2 X25A_5 X44A_2 X108B_4 X108A_3 X136B_1 X22A_4 X143A_15 X80A_3 X121A_1 X43A_3 X138A_1 V72A_1 X58A_1 X18A_3 X1A_4 X58B_1 X117A_1 X62A_2 X109A_3 X58A_2 X106A_4 X54A_1 X43A_2 X2A_1 X109B_5 X53A_1 X109A_8 X3A_5 X16A_1 X53A_6 X49A_8 X71A_1 X37A_1 X137B_2 X13A_1 X113A_1 X114A_7 X25B_2 X4A_3 X68A_4 X78A_4X49A_2 X77A_2 X90A_1 X17A_5 X20A_1 X56A_1 X55A_1 X137A_1 X108A_2 X28A_2 X15A_8 X137B_1 X3A_4 X21A_5 X6A_2 X46A_1 X41A_3 X48A_3 X14A_1 X57A_1 X130A_2 X17A_4 X52A_3 X120A_1 X22A_3 X16A_4 X88A_1 X33A_2 X51A_1 X59A_1 X143E_1 X109B_2 X14A_2X117A_5 X9A_3 X57A_2 X105A_3 X37A_2 X109A_1 X16A_7 X22A_5 X38A_4 X107A_1 X70A_1 X79A_1 X15A_3 X87A_3 X143A_1 X63A_1 X29A_2 X15A_7 X26A_3 X36A_2 X1A_5 X61A_2 X143F_4 X108A_1 X100A_2 X42A_1 X103A_2 X14A_6 X109B_1 X67A_1 X104A_2 X40A_5X23A_1 X64A_1X110A_2 X102A_2 X143A_3 X65A_1 X12A_3 X100A_4 X35A_6 X108B_1X87A_1 X47A_2 X32A_2 X144B_3 X38A_2 X112A_2 X8A_4 X101A_2 X89A_1 X118A_2 X44A_1 X6A_4 X75A_1 X48A_4 X129A_2 X44A_3 X29A_3 X102A_5 X135A_1 X50A_4 X91A_1 X119A_2 X85A_5 X40A_3 X136B_2 X114A_2 X104A_4 X16A_6 X33A_6 X24A_1 X100A_6 X103A_4 X17A_1 X92A_3 X45A_2 X23A_3 X66A_1 X35A_3 X131A_2 X24A_3 X128A_2 X26A_5 X104A_3 X93A_1 X104A_5 X34A_6 X142A_1 X86A_3 X26A_4 X138A_2 X43A_5 X125A_2 X30A_2 X25A_1 X25A_3 X31A_2 X52A_1 X48A_1 X19A_5 X130A_1 X84A_6 X50A_6 X126A_2 X79B_2 X32A_3 X27A_3 X103A_5 X21B_2 X127A_2 X69A_4 X71A_3 X81A_7 X30A_3 X28A_3 X92A_1 X36A_4 X7A_6 X40A_2 X12A_1

X88A_2

X99A_4

X35A_8

X36A_1 X122A_2

X35A_7

X91A_3

X94A_2

X56A_3

X92A_2

X2A_3

X117A_3

X94A_4

X24A_4

X49A_6

X136A_2

X76A_1

X83A_3 X82A_3

X79A_3 X62A_3

X53A_7 X121A_4

X123A_1

–0.5

0.0 1dim

0.5

Figure .. First and third dimension coordinates of 450 feature value categories generated by MCA. Circled left to right: clause-level head-initial feature values; NP-internal head-initial feature values; NP-internal head-final feature values; clause-level head-final feature values

X89A_2

0.3

Head-initial(NP-internal) except X86A_2

3dim

0.2

X88A_2

0.1 0.0 –0.1

X81A_1 X83A_1 X85A_1 X86A_1

X87A_2 Head-final(others)

X81A_2 Head-initial(others)

Head-final(NP-internal) except X86A_1

X86A_2X83A_2 X85A_2

–0.2

X88A_1 X87A_1

X89A_1

–0.6

–0.4

–0.2

0.0

0.2

0.4

1dim

Figure .. First and third dimension coordinates of 450 feature value categories generated by MCA (features 81A, 83A, 85A, 86A, 87A, 88A, and 89A only). Circled left to right: clauselevel head-initial feature values; NP-internal head-initial feature values; NP-internal head-final feature values; clause-level head-final feature values

(order of genitive and noun), groups with the clause-level features, N–Genitive with clause-level head-initial feature values and Genitive–N with clause-level headfinal values. This result is further confirmed by the cluster analysis presented in Section 4.4.4. We return to the diachronic significance of the grouping of order of genitive and noun in Section 4.5. The full set of feature values circled in Figure 4.3 is listed in (3).

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i



John Whitman and Yohei Ono

(3) Feature values on the positive and negative side of the first dimension Clause-level features plus 86A (order of Genitive and N) Positive Negative 81A_1 83A_1 85A_1 86A_1

[SOV] [OV] [NP–P] [Genitive–N]

81A_2 83A_2 85A_2 86A_2

[SVO] [VO] [P–NP] [N–Genitive]

NP-level features except 86A (order of Genitive and N) Positive Negative 87A_1 88A_1 89A_1

[Adj N] [Dem N] [Num N]

87A_2 88A_2 89A_2

[N Adj] [N Dem] [N Num]

By way of contrast, non-word order feature values do not show similar groupings around the first three dimensions. A sample set is given in (4).15 (4) Sample of non-word order feature values close to the first dimension X2A_1 [Vowel Quality Inventories: Small] = (0.2468 0.3784 0.0629) X2A_3 [Vowel Quality Inventories: Large] = (−0.1857 −0.1493 0.2575) X28A_1 [Case Syncretism: No case marking] = (−0.2057 0.1544 0.0783) X28A_2 [Case Syncretism: Core cases only] = (0.2815 0.1113 −0.0918) X112A_1 [Negative Morphemes: Neg affix] = (0.4186 −0.1224 0.1743) X112A_2 [Negative Morphemes: Neg particle] = (−0.0931 0.1506 −0.1985) Of the features in (4), only X112A_1 [Negative affix] shows a first dimension coordinate close to the (head-final) word order features in (3). Other features are broadly distributed. As noted above, the values of the output of MCA up to the third dimension in descending order are 0.1429, 0.1200, and 0.1081. The cumulative contribution ratio is 11.31 per cent up to the third dimension. The first dimension generated by MCA corresponds to the largest factor in the data. Feature values located in this region form what may be interpreted as a one-dimensional component, and as we have seen, they primarily involve word order. Since the data in WALS also includes information about languages, we can also inspect the outcome of MCA to see how languages are distributed. The first dimension also separates head-final and head-initial languages, as shown in Figure 4.5. Figure 4.5 shows the first and second dimensions for the languages in the sample. The circle on the positive side in the first dimension groups head-final languages like 15 That is not to say that a focused statistical study of non-word order feature values would not produce linguistically interesting results. In a preliminary investigation of the WALS data subtracting word order features, we found a tendency toward genetic and areal grouping, rather than the typological grouping found with word order features.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

Diachronic interpretations of word order parameter cohesion 0.8



Bunuba

0.6 0.4

Cree(Plains)

Yuchi Wardaman

Wambaya Nunggubuyu

Oneida

Wichita Yurok

2dim

Ekari

0.2

Ungarinjin Cayuvava

Kiribati

Arapesh(Mountain)

Paamese

0.0

Batak(Karo)

Semelai

Khmu'

Luvale Rapanui

Diola-Fogny Mixtec(Chalcatongo) Bagirmi Chinantec(Lealao) Kayah_Li_(Eastern) Nkore-Kiga

Drehu

Indonesian Ju|'hoan Hmong_Njua

Khmer Vietnamese

Ndyuka

Sango Yoruba

Tagalog Chamorro

Chukchi Mundari

Martuthunira Suena

Zulu Krongo

Ewe

Malagasy

Kera Koromfe

Arabic(Egyptian)

Hausa

Yidiny Ngiyambaa Nivkh Kayardild Carib

Rama Warao

Khoekhoe

Grebo

Iraqw

Bambara

Araona

Barasano Abkhaz

Meithei Beja Basque

Burmese

Burushaski

Hunzib Nubian(Dongolese)

Nenets Brahui Hungarian

Armenian(Eastern) Georgian

Japanese Kannada

Ingush

Khalkha Greek(Modern)

Hindi

Spanish English

Pomo(Southeastern)

Evenki

Korean

Oromo(Harar)

Persian

Irish

Huitoto_(Minica)

Garo Ladakhi

Lak Quechua(Imbabura)

Hebrew(Modern)

–0.4

Aymara Kewa

Yukaghir(Kolyma)

Fur Kanuri

Lepcha

Supyire Mandarin

Thai

Yup'ik(Central)

Miwok(Southern Sierra) Ika

Awa Pit Shipibo-Konibo Epena_Pedee Greenlandic(West)

Yaqui Kunama

Berber(Middle Atlas) Swahili Maori

Khasi Igbo

Haida Comanche Trumai

Qawasqar

Imonda Ket Maba

Koasati

Cahuilla Una

Sanuma Yimas Mapudungun Kobon

Amele

Murle Koyraboro_Senni

Paumari

Pirahi Alamblak Lavukaleve Bribri

Tunica Zoque(Copainale)

Maricopa Kiowa Dani(Lower_Grand_Valley)

Selknam Hixkaryana

Sentani

Canela-Krah

Bawm

Otomi(Mezquital) Tukang Besi

Kongo

–0.2

Fijian

Tlingit

Pitjantjatjara

Hamtai

Apurine Navajo Gooniyandi Passamaquoddy-Maliseet Mangarrayi Asmat Usan Urubu-Kaapor Squamish Daga Ainu Makah Slave Marind

Paiwan Guarani

Lango

Maybrat Taba

Ngiti

Jakaltek Tiwi Maung Nahuatl(Tetelcingo)

Abipon

Kilivila

Lakhota Wichi

Nez_Perce

Yagua

Kutenai Tsimshian(Coast) Wari

Karok Coos(Hanis)

Acoma

Maranungku

Lezgian Turkish

Finnish Latvian German

French

–0.4

–0.2

Russian

0.0 1dim

0.2

0.4

0.6

Figure .. First and second dimension coordinates of 201 languages generated by MCA

Japanese or Khalkha, while the circle on the negative side groups head-initial languages like Thai or Vietnamese.16 (5) Feature values on the positive and negative side of the first dimension Japanese = (0.2183 −0.2447 0.2711) Khalkha = (0.2951 −0.3086 0.2200) Thai = (−0.3427 −0.2085 0.2625) Vietnamese = (−0.3779 −0.1365 0.2668) It is not our purpose in this chapter to investigate language groupings derived from the WALS features; we merely note here the fact that this distribution confirms the status of the first dimension in the MCA analysis as based primarily on word order. .. Cluster analyses Cluster analysis of the WALS dataset allows us to further investigate the interrelationships between features. We explored two approaches for the WALS data. The first is direct cluster analysis, using techniques known as Ward’s method (Ward 1963) and the complete distance method (Sørensen 1948). We also experimented with applying the same techniques to the output of the MCA described in the previous section (see the Online Appendix for figures).17 16 The coordinates for  languages for the first and third dimensions show a less clear grouping of head-initial and head-final languages (see the Online Appendix). Following the recommendation of a reviewer, we produced a heat map (also in the Online Appendix) based on a cluster analysis of this data. This revealed an intriguingly compact grouping in the second and third dimensions made up of all the Indo-European languages in the sample (Spanish, Greek, Irish, English, Latvian, German, Russian, French, Armenian, Persian, Hindi) as well as all of the non-IE European languages (Hebrew, Georgian, Hungarian, Finnish). We hope to investigate the significance of the latter grouping in future research. 17 We applied cluster analyses directly to the WALS dataset on the recommendation of a reviewer, in addition applying them to the output of MCA. We also experimented with different clustering methods: Euclidean distance using Ward’s method, Squared Euclidean distance with Ward, and Euclidean distance

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i



John Whitman and Yohei Ono

Table .. Classification accuracies for Cluster A Parameter

Values

Classification accuracy ()

Area

X81A_2

2: SVO

81.09

Word order

(Complex feature)

X83A_2 X84A_1

2: VO 1: VOX

X85A_2 X86A_2 X1A_3 X28A_1 X105A_3

2: Preposition 2: Noun–Genitive 3: Average 1: No case marking 3: Secondary-object construction

89.55 80.60

Word order Word order

(Complex feature)

88.06 84.08 61.69 58.21 65.17

Word order Word order Phonology Morphology Syntax

As noted in the text, Cluster A lacks simplex phonological, morphological, or syntactic features to meaningfully calculate classification accuracy. We use three parameters from other clusters to provide a tentative standard of comparison.

Through these analyses, we identified four main clusters, which we have labelled A, B, C, and D. Tables 4.2, 4.3, 4.4, and 4.5 show the feature values in each of the clusters, providing classification accuracies for only the highest-ranked non-word order features. Clusters A and B are grouped together in a higher order cluster containing all head-initial values for word order features, and clusters C and D are likewise grouped together in a higher order cluster containing all head-final values. In both cases the clause-level word order feature values form a compact cluster. Cluster A contains five word order features with head-initial values: 81A_2 (SVO), 83A_2 (VO), 84A_1 (VO–Oblique), 85A_2 (P–NP), and 86A_2 (N–Genitive). It contains only two non-word order feature values: 123A_3 (pronoun retention strategy for relativization on obliques), and 111A_1 (no morphological or compound causatives). The correlations between these two feature values and head-initial, SVO word order values are worthy of future investigation. As shown in Table 4.2, all five word order features have classification accuracies above 80 per cent. This compares to the much lower classification accuracies for the non-word order features found in other clusters. Cluster C contains four of the word order features we saw grouped in the MCA (Figure 4.3) with head-final values: 81A_1 (SOV), 83A_1 (OV), 85A_1 (NP–P), and 86A_1 (N–Genitive). Each has a classification accuracy above 70 per cent. All but three of the remaining values in Cluster C pertain to word order features as well, but all have classification accuracies under 50 per cent. Six have typical head-final values, while using the complete distance method. The results are given in the Online Appendix at http://www.oup.co.uk/ companion/DiGS15. Although results varied, all of these analyses consistently group the word order features A (order of subject, object, and verb), A (order of object and verb), A (order of adposition and NP), and A (order of genitive and noun).

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

Diachronic interpretations of word order parameter cohesion



Table .. Classification accuracies for Cluster B Parameter

Values

Classification accuracy ()

Area

X87A_2 X88A_2 X89A_2 X90A_1 X91A_2 X91A_3 X143A_2 X144B_4 X144F_1 X3A_1 X28A_1 X105A_2

2: Noun–Adjective 2: Noun–Demonstrative 2: Noun–Numeral 1: Noun–Relative Clause 2: Adj–Degree word 3: Adj–Deg: No dominant order 2: V–Nega 4: Immediate postverbal 1: V–Neg 1: Low (Consonant–Vowel Ratio) 1: No case marking 2: Double-object construction

57.71 72.64 55.22 78.61 58.71 64.68

Word order Word order Word order Word order Word order Word order Word order Word order Word order Phonology Morphology Simple clause

62.69 58.21 60.20

a We omitted classification accuracies for the negation-related word order features, because some of these features are complex (specifically WALS features 144D–S, e.g. 144D ‘the position of negative morphemes in SVO languages’. Rather than present a possibly misleading subset of negation-related word order features, we omitted classification accuracies for them all. They are in most cases low. Note: There is no syntactic feature in cluster B, so we use 100A_1 to provide a standard of comparison.

Table .. Classification accuracies for Cluster C Parameter

Values

Classification accuracy ()

Area

X81A_1

1: SOV

73.63

Word Order

(Complex feature)

X83A_1 X84A_3

1: OV 3: XOV

70.15 46.77

X85A_1 X86A_1 X90A_2 X94A_2 X94A_4 X94A_5 X143A_4 X143A_14 X143E_4 X143F_2 X7A_2 X26A_2 X126A_3

1: Postposition 1: Genitive–Noun 2: Relative Clause–Noun 2: Final Subordinator word 4: Subordinating suffix 5: Mixed 4: V–Neg 14: ObligDoubleNeg 4: None (Complex feature) 2: V–Neg 2: Ejective only 2: Strongly suffixing 3: Deranked

80.60 75.12 49.75 49.25 43.78 46.77

Word Order Word Order

(Complex feature)

44.78 60.20 56.22

Word Order Word Order Word Order Word Order Word Order Word Order Word Order Word Order Word Order Word Order Phonology Morphology Syntax

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i



John Whitman and Yohei Ono

Table .. Classification accuracies for Cluster D Parameter

Values

Classification accuracy ()

Area

X87A_1 X88A_1 X143A_14 X143G_4 X3A_4 X23A_3 X100A_6

1: Adjective–Noun 1: Demonstrative–Noun 14: ObligDoubleNeg 4: None 4: Moderately High 3: Double Marking 6: Split

48.26 68.66

Word order Word order Word order Word order Phonology Morphology Syntax

48.76 45.77 42.79

the third, 94A_5 (a mixed pattern of adverbial subordinators with none dominant), is moderately unmarked in the WALS sample. The three non-word order features in Cluster C all have lower classification accuracies. The highest of these, 26A_2, has the value ‘strongly suffixing’, again a not unexpected result for a cluster characterized by head-final values. Both of the clause-level word order clusters A and C are thus quite compact. Both group N–Genitive with clause-level features. In Cluster A, 86A_1 (N–Genitive) is sister to 85A_2 (P–NP). In contrast, Clusters B and D, containing NP-internal feature values, are more diffuse. Cluster B, with head-initial features internal to NP, contains values for phonological, morphological, and syntactic features including word order. N–Demonstrative and N–Relative Clause contribute the highest classification accuracies. Cluster D, containing 87A_1 (Adj–N) and 88A_1 (Demonstrative–N), also contains features across phonological, morphological, and syntactic domains. The third NPinternal word order feature value, 89A_1 (Numeral–N), is contained in a completely different subcluster within the larger head-final cluster. We draw two main conclusions from the cluster analyses discussed in this section. All of the cluster analyses produced groupings of word order features based on headinitial and head-final values, although not all of them reproduced the four distinct groupings we found along the first dimension of MCA. We found no comparable clusters based on non-word order features, and calculations of classification accuracies in the clusters we investigated in detail always showed word order features with the highest classification accuracies.18 18 A reviewer points out to us that the prominence of word order features in the MCA and cluster analyses of the WALS data might be attributable to their relatively large number in the WALS dataset,  out of the total of  features. To check this possibility, we calculated classification accuracies for the clusters generated by Ward’s method and Manhattan distance applied to the output of MCA, but reducing the number of word order features to just three. The results of this calculation are shown in the Online Appendix, ‘Classification Accuracy Comparison Table’. Summarizing these results, whichever three of the word order features A (order of subject, object, and verb), A (order of subject and verb), A (order of object and verb), A (order of adposition and noun phrase), and A (order of genitive and noun) were retained,

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

Diachronic interpretations of word order parameter cohesion



. Discussion and diachronic interpretation Below we summarize the results of the MCA and cluster analyses in Section 4.4. (i) Word order features form the main component in a multiple correspondence analysis of the WALS dataset. (ii) MCA shows a clear grouping of head-initial versus head-final feature values and languages. (iii) A cluster analysis on the output of MCA shows the same dominant effect of word order features, and headedness; however: (iv) In both the MCA and clearest cluster analysis, NP-internal word order features cluster separately from other constituent order features, with the exception of the feature ‘order of genitive and noun’, which consistently groups with clauselevel word order features. The result most relevant for our discussion in Sections 4.2 and 4.3 is (iv), but before looking at this result in more detail, it is worthwhile emphasizing once again the takeaway point from (i)–(iii): at least in an analysis of the features in the WALS dataset, word order features show clearer correlations with one another than other typological features. This is not to say that a more fine-grained analysis might not reveal similar correlations between other sets of features (see footnote 15). But at least at the level of analysis reported here, word order features show overwhelmingly more salient correlations with one another than other features. In a theory of Universal Grammar that included a Head Parameter, the results in (i)–(iii) might be taken to reinforce the centrality of such a parameter. But as we explained in Section 4.2, previous research shows that a parameter governing crosscategorial constituent order patterns constrains neither the typological distribution nor the acquisition facts. This points to the conclusion that ‘relabelling’-type changes that propagate word order patterns across categories have been very common throughout human linguistic history. Result (iv) suggests a more nuanced view of the role of nominal and verbal projections in changes of this type. NP-internal word order features group separately from clause-level word order features, with the exception of genitive–noun order. A similar result was obtained by Albu (2006: 40–2), who found that WALS features 83A (order of object and verb), 85A (order of adposition and NP), and 86A (order of genitive and noun) cluster together, while the NP-internal word order features 87A (order of adjective and noun) and 88A (order of demonstrative and noun) cluster separately (we are grateful to Michael Cysouw for bringing Albu’s research to our attention). Our interpretation of this fact is that constituent order patterns in the nominal domain cluster separately from the clausal, verbal, and adpositional domain when the former involve nonargument constituents such as demonstratives, attributive adjectives, and numerals. Arguments in the nominal domain, on the other hand, pattern values for those features displayed the highest classification accuracy (ranging from . to .), with the exception of XA_ (little affixation), which showed classification accuracies in these conditions ranging from . to .. In contrast, the highest classification accuracy for phonological feature values under the same conditions was . for XA_ (average consonant inventories).

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i



John Whitman and Yohei Ono

with arguments in the projections of other categories. The diachronic explanation we have suggested is that nouns and their arguments, realized as genitives, may undergo categorical reanalysis, either to PPs, or in the form of nominalization > main clause reanalyses. We saw in Section 4.3 how [NP N] patterns provide a source for postpositional phrases in Chinese, and we noted that in the cross-linguistic examples collected by Heine and Kuteva [NP N] / [N NP] patterns appear to provide a wide range of sources for reanalysis as PP. We also commented that nominalization-to-main clause reanalyses provide a route for [NP Argument–N] patterns to be reanalysed as [VP Argument–V]. The results of our cluster analysis suggest that the patterning of arguments (typically realized as genitives) in the nominal projection is the route for crosscategorial symmetries linking NP and other categories.

Acknowledgements Ono’s work on this paper was supported by the Japan Society for the Promotion of Science (JSPS): Grant-in-aid for Science Research (S) No. 22223006. Whitman’s research was supported by the Academy of Korean Studies Grant funded by the Korean Government (MEST) (AKS-2011-AAA-2103). We are grateful to the editors and two highly conscientious reviewers; all errors are of course our own.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

 The rise and fall of Hungarian complex tenses KATALIN É. KISS

. Goals This chapter aims to reconstruct how and why complex tenses appeared in Hungarian grammar in the late Proto-Hungarian period, and how and why they disappeared a thousand years later. It will show that the evolution of complex tenses started with a micro-change: the reanalysis of the feature content of a verbal suffix. This step initiated further processes of reanalysis, analogical extension, and abstraction, as a consequence of which the tense system inherited from Uralic, distinguishing only past and nonpast, developed into a complex system marking both tense and viewpoint aspect. The chapter will argue that both the appearance of complex tenses, and their disappearance, that is, the replacement of morphological viewpoint aspect marking by situation aspect marking via verbal particles, were triggered by language contact, and will speculate about the conditions that make a grammatical construction susceptible to foreign influence. The chapter is structured as follows: Section 5.2 introduces the complex tenses of Old Hungarian, marking both tense and viewpoint aspect. Section 5.3 argues against the traditional view that they are the artificial creations of scribes translating from Latin. Section 5.4 shows that they must have appeared in Proto-Hungarian under the influence of Old Turkic. Section 5.5 attempts to reconstruct which Old Turkic construction served as a model for Proto-Hungarian, what changes it triggered in ProtoHungarian, and how the changes unfolded. Section 5.6 describes a change that took place in Hungarian a thousand years later, in the course of which the extension of telicity marking by resultative verbal particles, presumably induced by Slavic influence, developed into an alternative aspectual system, marking situation aspect. Section 5.7 shows how situation aspect marking made viewpoint aspect marking redundant, and led to the disappearance of complex tenses. Section 5.8 summarizes the theoretical implications of the changes observed.

Micro-change and Macro-change in Diachronic Syntax. First edition. Éric Mathieu and Robert Truswell (eds). This chapter © Katalin É. Kiss 2017. First published 2017 by Oxford University Press.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i



Katalin É. Kiss

. Complex tenses in Old Hungarian Whereas Modern Hungarian only has two tenses: past and nonpast, Old Hungarian had a complex tense system marking both tense and viewpoint aspect. In Old Hungarian texts, we attest the following tenses: (1)

i.

Simple Present, e.g.:

ii. Simple Past: iii. Present Perfect: iv. Past Imperfective/Continuous: v. Past Perfect:

mond-∅-om say-prs-1sg mond-á-m say-pst-1sg mond-t-am say-pfv-1sg mond-∅-om val-a say-ipfv-1sg be-pst mond-t-am val-a say-pfv-1sg be-pst

In the simple tenses, the lexical verb bears tense and agreement morphemes. In the complex tenses, the lexical verb is marked for aspect and agreement, and the tense morpheme is borne by the copula serving as a temporal auxiliary. Old Hungarian also had simple conditional and conditional perfect verb forms, in which the tense morpheme is replaced by a mood suffix. Thus in the simple conditional, the verb bears mood and agreement morphemes. In the conditional perfect, the verb is marked for aspect and agreement, and the auxiliary is marked for mood: vi. Simple Conditional: vii. Conditional Perfect:

monda-ná-m say-cond-1sg mond-t-am vol-na say-pfv-1sg be-cond

In Old Hungarian, the different tense–aspect combinations had similar values as they have for example in present-day English. Thus the simple present was used to describe events that include or follow the speech time. The simple past, also called narrative past, was the default past tense used for example in storytelling. The present perfect marked past events with a result still in effect at the speech time. The difference between the functions of the simple past and the present perfect is illustrated by a Biblical citation from the Book of Ruth, where the storyteller relates the event of Orpah turning back by using the simple past, whereas Naomi, witnessing Orpah walking away, refers to the event with the present perfect. (2) Orpha megapol-a o napat & m˙eg´fordol-a. Rvt Orpah kiss-pst.3sg ´her mother.in.law.acc and turn-pst.3sg Ruth Kinc. m¯od-a Noemi Ime te ˙eggèsolè o napaual ´ ´her mother.in.law.with whom tell-pst.3sg Noemi behold your united rokonod m˙eg´fordol-t relative turn-pfv.3sg ‘Orpah kissed her mother in law and turned back. Ruth clave onto her mother in law. Naomi told her: Behold, your relative has turned back.’ (Vienna Codex 1416, p.2)

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

The rise and fall of Hungarian complex tenses



The past continuous verb forms marked past events in progress—see (3a), habitual past events—see (3b), and incomplete past achievements ((3c), a sentence from the story of Adam and Eve). (3) a. mend czudal-yak ual-a Es halgat-yak ual-a ewtett all admire-3pl be-pst and listen-3pl be-pst him ‘They were all admiring him and listening to him’ (Jókai Codex 1370, p.37) b. Ez nemesseges zent zvz . . . hetet tart-∅ val-a az this noble saint maid week.acc keep-ipfv.3sg be-pst the cohnyan fevz-∅ val-a az sororoknak kitchen.at cook-ipfv.3sg be-pst the sisters.dat ‘This noble saint maid would have turns on duty at the kitchen, she would cook for the sisters’ c. Es oz gimilsnec wl keseruv uola vize. hug turchucat mige and that fruit.dat so bitter was water.3sg that throat.3pl.acc prt zocozt-ia vol-a. burst-ipfv.3sg be-pst ‘And that fruit had such a bitter juice that it was bursting their throats.’ (Funeral Speech and Prayer 1195) The past perfect marked past events preceding a past reference point, for example: (4) Es megemlékez-é-c Peter az igéro¸l, kit mondo-t val-a and commemorate-pst-3sg Peter the word.about that say-pfv.3sg be-pst ‘And Peter commemorated the word that he had told them.’ (München Codex 1416, p.103) As illustrated by these data, the Old Hungarian tense system encoded not only the external tense of events, but also the viewpoint of the speaker. It marked whether the speaker’s perspective included the whole event, or only an internal section of it, that is, it marked, in addition to tense, also viewpoint aspect.

. The traditional view of the origin of complex tenses: Latin influence Since the majority of the Uralic languages, including Mansi and Khanty, the closest relatives of Hungarian, only have two tenses: past and nonpast, traditional historical linguists assume that the complex tenses of Old Hungarian were artificially created by scribes translating from Latin in order to distinguish the different Latin past tenses (E. Abaffy 1991: 109–10; Sárosi 2003: 367). This view, however, is untenable primarily for theoretical reasons. If grammar changes when children acquiring a construction analyse it differently from the way their parents analyse it, then a scholarly second language learned at an older age is unlikely to lead to major changes in the native grammar. Empirical considerations also argue against deriving the complex tenses from Latin influence. First of all, Latin has no complex tenses in the active voice. The Neo-Latin languages have complex tenses, but they are structured differently from those of Old Hungarian (the agreement

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i



Katalin É. Kiss

suffix is on the auxiliary, instead of the lexical verb). Secondly, the complex tenses were present in Old Hungarian before large scale translation from Latin (e.g. the translation of the Bible) began. The very first surviving Hungarian text, the Funeral Sermon and Prayer from 1195, which is believed not to be a translation, contains both past perfect and past continuous verb forms: (5) a. es odut-t-a vol-a neki paradisumut hazóá and give-pfv-3sg be-pst him Paradise.acc house.for ‘and had given him Paradise for his house’ b. turchucat mige zocozt-ia vol-a throat.3pl.acc up split-ipfv.3sg be-pst ‘it was splitting up their throat’. Thirdly, the complex tenses are also found in Old Hungarian private letters, as testified, for example, by the letters edited by Hegedűs and Papp (1991). Fourthly, although the complex tenses disappeared from most varieties of Hungarian in the sixteenth–seventeenth centuries, they are still attested in the most archaic East Hungarian dialects, including the Csángó dialect spoken in Moldavia, which has never had any written form or any literate speakers—see for example the Csángó corpus collected by Szegő (1988). Standard Hungarian has also preserved a complex verb form, the conditional perfect, which has been reinterpreted as a past conditional: (6) mond-tá-tok vol-na. say-pfv-2pl be-cond ‘you would have said’ In view of these facts, the possibility that the Old Hungarian complex tenses evolved as a consequence of Latin influence, can be excluded.

. An alternative explanation: Old Turkic contact effect Bereczki (1983, 1993) proposed an alternative explanation for the emergence of complex tenses in Hungarian. He observed that there are five Uralic languages that have developed complex tense–aspect systems: those that have had close contact with Turkic languages sometime in the course of their histories. Hungarians were in close contact with West Turkic tribes in the seventh and eighth centuries, when they belonged to various Turkic tribal alliances in the area between the Dneper and the Dnester. The Hungarian tribal alliance settling in the Carpathian Basin in the ninth century incorporated a Kabar tribe and other Turkic fragments, and—as is reported in De administrando imperio by Constantine Porphyrogennetos—Hungarians and Turks spoke each other’s languages (cf. Sándor 2011). The other four Uralic peoples with Turkic contacts, the Udmurts, the Komi, the Mari, and the Mordvins, have shared their habitat with Turkic peoples along the Volga and Kama rivers for the past thousand years. The Old West Turkic language that could influence Hungarian is known from inscriptions and other documents, and a present-day descendant, the Chuvash language. According to Erdal’s Old Turkic Grammar (2004), Old Turkic had complex tense forms structured the same way as the complex verbs of Old Hungarian,

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

The rise and fall of Hungarian complex tenses



consisting of a lexical verb bearing an aspect marker and an agreement marker, and a copula bearing a tense or mood suffix. In Old Turkic complex verb forms, the aspect marker on the lexical verb and the tense marker on the copula appear to be identical, and they are both glossed as tense suffixes, but the one on the lexical verb is said by Erdal to express taxis, that is, relative tense. Example (7) contains an Old Turkic past perfect, and (8) is an example of Old Turkic conditional perfect. (7) öŋdün sözlä-di-∅ är-di earlier say-pst-3sg be-pst ‘he had said (it) earlier’ (8) te-di-miz är-sär say-pst-1pl be-cond ‘we would have said (it)’

(Erdal 2004: 245)

(LeCoq 2011)

In Chuvash, the only present-day descendant of West Old Turkic, the be+past complex has cliticized to the lexical verb marked for taxis and agreement, as a result of which the agreement suffix appears in the middle of the inflected verb: (9) şyra-tt-ăm-čč˘e1 work-dur.pst-1sg-be.pst ‘I was working’ şyr-satt-ăm-čč˘e work-pret-1sg-be.pst ‘I had worked’ The complex verb forms of the Uralic languages along the Volga–Kama rivers are structured similarly. As is illustrated by the Udmurt and Mari examples in (10) and (11) cited from Bereczki (1983), the agreement marker is on the lexical verb; the copula is only marked for tense. (10) Udmurt: a. mi.niśk-em go-pfv.1sg ‘I went’ (11)

Mari: a. tol¨әn-am come-pfv.1sg ‘I came’

b. mi.niśk-em val go-pfv.1sg be.pst ‘I had gone’ b. tol¨әn-am ¨әl’e come-pfv.1sg be.pst ‘I had come’

Aspect is marked on the lexical verb, preceding agreement, as shown by the following Komi minimal pair of Bereczki (1983). (12) Komi: a. mun-a ve¸li go-1sg be.pst ‘I was going’

1

b. mun-e¸m-a ve¸li go-pfv-1sg be.pst ‘I had gone’

The Chuvash data have been provided by Klára Agyagási, Debrecen University.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i



Katalin É. Kiss

The fact that the Old Turkic and the Old Hungarian verbal complexes are structured similarly could also be a coincidence, although the close contact of Turks and Hungarians in the Proto-Hungarian period is documented in historical sources. However, if the complex tenses of Old Hungarian were the result of internal development, we would not expect to also find similar complex tenses in the four Uralic languages that happen to be spoken in an area cohabited by Turkic peoples. The fact that verbal complexes of the Turkic type are attested in all and only the languages of the Uralic family that have come into close contact with Turkic peoples suggests that the complex tenses of these languages have evolved owing to contact with Turkic.

. The evolution of Hungarian complex tenses Grammatical borrowing presupposes a bilingual situation (cf. Bowern 2005), which can be plausibly assumed for large sections of the seventh–ninth-century Hungarian population. (The administrative and military governability of the Turkic–Hungarian tribal alliances required a common language shared at least by the elites of the allied tribes.) In a bilingual situation, a scenario of the grammar of L2 influencing the grammar of L1 involves the reanalysis of a construction of L1 according to the rules of L2. This is how the borrowing of complex verb forms from Old Turkic might also have happened. The Uralic languages abound in participles and gerunds, which can have overt subjects, and can agree with them. Hungarian, for example, has a type of gerund derived by the suffix -t, which has a subject of its own, eliciting possessive agreement on the gerund. This type of gerund must be of Ugric, or perhaps Uralic, heritage, as Khanty, a Ugric sister language of Hungarian, also has it (see Nikolaeva 1999: 47). For example: (13) Hazafelé men-t-em-kor eleredt az eső. homewards go-ger-1sg-at started the rain ‘During my going home, it started to rain.’ This type of gerund could—and still can—easily occur as the subject in a copular clause, for example: (14) Nyug-t-om val-a-∅2 rest-ger-1sg be-pst-3sg ‘I had some resting’ [Lit.: My resting was] Nyug-t-od val-a-∅ rest-ger-2sg be-pst-3sg ‘You had some resting’

Nyug-t-unk val-a-∅ rest-ger-1pl be-pst-3sg ‘We had some resting’ Nyug-to-tok val-a-∅ rest-ger-2pl be-pst-3sg ‘You had some resting’

2 This example is a reconstructed Old Hungarian structure. In the Modern Hungarian version, the past tense marker of the copula is -t, and the gerund has a somewhat idiomatic meaning, i.e.: (i) Nyug-t-om vol-t-∅. rest-ger-1sg be-pst-3sg ‘I had some resting [I wasn’t disturbed]’

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

The rise and fall of Hungarian complex tenses



Nyug-t-a val-a-∅ Nyug-t-uk val-a-∅ rest-ger-3sg be-pst-3sg rest-ger-3pl be-pst-3sg ‘He had some resting’ ‘They had some resting’ The construction in (14) is structured similarly to the past perfect verb form of Old Turkic. The -t- gerundive suffix appears in the same position as the Old Turkic -diaspectual suffix, and is also formally similar to it. Consequently, children acquiring Proto-Hungarian in a biligual situation could easily identify the feature content of the Hungarian -t- with that of the Turkic -di-, that is, they could interpret the -tas a perfective morpheme. The possessive agreement on the gerund is nondistinct from verbal agreement, hence it could easily be reanalysed as such.3 The possessor, represented by a silent pro in (14), was caseless, hence it could also be interpreted as a nominative subject. The 3rd person singular agreement morpheme on the copula is phonetically null, which facilitated the reanalysis of the copula as the carrier of a mere tense morpheme. That is: (15) Reanalysis [proi V + gerund.suffix + poss.agri ] [copula + tense + subj.agr] → proi [V + aspect + subj.agri ] [auxiliary + tense] For example: (16) pro men-t-em val-a-∅ → 1sg go-ger-poss.1sg be-pst-3sg ‘My going was’

pro men-t-em val-a 1sg go-pfv-1sg be-pst ‘I had gone’

The change in the feature content of -t-, and the reanalysis of the gerund + copula string as a past perfect verb form must have triggered further changes. The -t- perfect morpheme could be removed from the past perfect paradigm, and in the resulting verb form, the lack of an overt element in the position of the aspectual morpheme was interpreted as the marker of imperfective/continuous aspect. Thus the analogical extension of the past perfect paradigm yielded an imperfective/continous paradigm, as well: (17) Analogical extension men-t-em val-a → go-pfv-1sg be-pst ‘I had gone’

men-∅-ek val-a4 go-ipfv-1sg be-pst ‘I was going’

The establishment of the past perfect paradigm opened up a further possibility: omitting the past tense morpheme and its copula carrier resulted in a present perfect verb form:

3 More precisely, Hungarian uses both definite and indefinite verbal paradigms, depending on whether or not the verb has a definite object. The singular possessive agreement suffixes are identical with the singular agreement suffixes of the definite verbal paradigm, and the plural possessive agreement suffixes are identical with the plural agreement suffixes of the indefinite verbal paradigm. 4 The stem of men-ek also has a megy allomorph.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i



Katalin É. Kiss

(18) Abstraction: men-t-em val-a go-pfv-1sg be-pst ‘I had gone’

→

men-t-em go-pfv-1sg ‘I have gone’

In sum: in a bilingual situation, the structural and morphological similarity of the Proto-Hungarian -t- gerund + copula string and the Old Turkic past perfect verbal complex made it possible for children acquiring Proto-Hungarian to assign the structure of the Old Turkic construction to the Hungarian expression. The emergence of the past perfect paradigm in Proto-Hungarian initiated further changes: the removal of the perfect morpheme yielded a past imperfective paradigm, and the removal of the past tense auxiliary yielded a present perfect paradigm.

. The emergence of situation aspect marking The tense–aspect system that evolved in Proto-Hungarian combines tense marking with viewpoint aspect marking, expressing morphologically whether the speaker’s perspective includes a whole, completed event, or only the internal section of an event. At the same time, the very first Old Hungarian documents already carry the germs of an alternative aspectual system, as well, which distinguishes telic events, having an inherent endpoint, from atelic events with no inherent endpoint, by means of a resultative or terminative verbal particle. Verbal particles are claimed to have already existed in the Ugric proto-language (Honti 2001). They are attested in the Ugric sister languages of Hungarian, too. Their occurrence in the early Old Hungarian documents is still sporadic. In the Funeral Sermon and Prayer (1195), we find a single resultative verbal particle: (19) turchucat mige zocozt-ia vol-a throat.3pl.acc prt rive-ipfv.3sg be-pst ‘It was splitting up their throat.’ In the rest of the clauses of this text, telicity is marked by perfect verbal morphology (20a), by the context (20b), or by the lexical meaning of the verb (20c). (20) a. es odut-t-a vol-a neki paradisumut hazóá and give-pfv-3sg be-pst him Paradise.acc house.for ‘and had given him Paradise for a house’ b. es vetev-e wt ez muncas vilagbele and throw-pst.3sg him this laborious world.into ‘and threw him into this laborious world’ c. hug isten ív uimadsaguc-mia bulsas-s-a w bunet that God they prayer.3pl-because.of forgive-sbjv-3sg he sin.3sg.acc ‘so that God forgive his sin because of their prayer’ In the fourteenth–fifteenth centuries, we attest the fast spreading of verbal particles. Jókai Codex, containing a translation of the Legend of Saint Francis from around 1370, already abounds in particle verbs. We have an interesting ‘snapshot’ from the end of

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

The rise and fall of Hungarian complex tenses



the fifteenth century of their gaining ground, the so-called München Relic (Haader 2004). A German monk interested in the Hungarian language described the Lord’s Prayer in Hungarian in two versions. One version seems to have been copied in Hungarian spelling from a written source. The other version is the transcription in German spelling of how the scriptor heard the prayer from a Hungarian native speaker. The version copied from a written source, representing somewhat earlier usage, still has no verbal particles. The version recording the oral prayer, on the other hand, already has three of them: (21) ës meg-bozässät mi vëtkenkët. mikëpen ës mi mag-boczätunk and prt-forgive.imp.2sg we sin.1pl.acc as also we prt-forgive.1pl vëtëtëknek . . . de säbädicz-mk mikët a gonostwl sinners.dat but free.imp.2sg-prt us the evil.from ‘and forgive us our debts, as we also forgive our debtors . . . but deliver us from evil.’ The Lord’s Prayer in the Bible translation of Gáspár Károli from 1590 contains four more verbal particles: (22) szenteltesséc meg à te neued. Iöijön el az te hallow.pass.imp.3sg prt the you name.2sg come.imp.3sg prt the you országod: Légyen meg à te akaratod . . . Az mi minden napi country.2sg be.imp.3sg prt the you will.2sg the our every day kenyerünket add meg nékünc ma. bread.acc give.imp.2sg prt us today ‘Hallowed be your name. Your kingdom come, your will be done . . . Give us this day our daily bread.’ By the Middle Hungarian period (1526–1772), most process verbs had developed particle versions.5 The particle verbs assumed telic, accomplishment meanings, and the bare verbs came to be confined to atelic contexts, as shown by the following minimal pairs from the Legend of Saint Margaret (1510). (23) a. ker vala hust . . . es meg fevzÿ ask.ipfv.3sg be.pst meat.acc and prt cook.ipfv.3sg pro6 it.acc ‘She would ask for meat . . . and would cook it.’ b. mert akoron az kÿs setet kohnyaban fevznek because then that small dark kitchen.in cook.ipfv.3pl ‘because then they used to cook in that small dark kitchen’

vala be.pst

(p.26 verso) vala be.pst (p.66 verso)

5 Unergative verbs are exceptions. The particle predicates the resulting state of the theme argument, hence unergative verbs can only be bounded by a verbal particle if they take a non-thematic object—see É. Kiss (). 6 The presence of this object pro can be reconstructed from the definite conjugation of the verb.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i



Katalin É. Kiss

(24) a. es meg allanak ez beteg soror agÿanal az zentek and prt stand this sick soror’s bed.at the saints.nom ‘and the saints would stop at the bed of this sick soror’ (p.63 recto) b. Mykoron ez soror egÿ napon az karban allana when this soror one day the chorus.in stand.cond.3sg az ablacnal the window.at ‘When this soror was standing in the chorus at the window one day’ (p.73 verso) The development of atelic–telic verb pairs must have been inspired by Slavic. Hungarians settling in the Carpathian basin in the ninth century found a Slavic population there, which they absorbed, and this process must have involved the bilingualism of large sections of the population for more than one generation. Hungarian borrowed hundreds of words from Slavic, and the grammar of the Slavic substratum is also likely to have influenced Hungarian grammar. In the Slavic language, telicity is systematically marked by a verbal prefix, for example, čitat’ means ‘to read’, pro-čitat’ means ‘to read from beginning to end’; delat’ means ‘to do, to make’, sdelat’ means ‘to do, to make completely’ (Borik 2002). Old Hungarian verbs supplied with a prefix-like preverbal verbal particle were also all telic (see (19)), but bare verbs could be either atelic, or telic (as shown by examples (20a)–(20c)). That is, whereas the aspectual properties of prefixed/particle verbs were similar in Slavic and in Hungarian, the aspectual roles of bare verbs were different: Slavic verb types: Old Hungarian verb types:

prefixed verbs ↔ bare verbs [+telic] [−telic] particle verbs — bare verbs [+telic] [±telic]

Slavic contact must have resulted in the reanalysis of the aspectual feature of Hungarian bare verbs from [±telic] to [−telic]: (25) Reanalyis: particle verbs — bare verbs [+telic] [±telic]

→

particle verbs ↔ bare verbs [+telic] [−telic]

The resulting aspectual contrast was analogically extended to all transitive and unaccusative verbs (unergatives being unable to take particles—see footnote 4); practically all bare verbs developed versions with a resultative or terminative particle.7 The reanalysis of verbal particles as the carriers of telicity did not affect their syntax though. Whereas Slavic telicizing prefixes are bound morphemes, the Hungarian verbal particle has preserved the syntactic independence that it had in early Old Hungarian. In neutral sentences, it immediately precedes the verb, but in negative sentences and in focus constructions, the verb is raised across the particle.

7

For evidence, see É. Kiss ().

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

The rise and fall of Hungarian complex tenses



. The fall of complex tenses The micro-change in the feature content of bare verbs, resulting in the evolution of aspectual verb pairs and the systematic distinction of [±telic], has eventually led to a macro-change: the emergence of situation aspect marking made viewpoint aspect marking redundant, and led to the disappearance of complex tenses. In most cases, atelic events are imperfective, and telic events are perfective. However, fifteenth century Hungarian was also capable of systematically encoding mismatches between the two types of aspect, describing, for example, incomplete telic events (26) or incomplete series of telic events (27), by combining a telicizing verbal particle with an imperfective verb form (Gerőcs 2011). (26) Ki meg-foguan mg ¯ foit’-a uala otet monduan who prt-grabbing prt throttle-ipfv.3sg be.pst ´him saying Ad meg miuèl tartoz-ol give.imp.2sg back what.ins owe-2sg ‘Who, having grabbed, was throttling him, saying, Pay me that thou owest.’ (München Codex 1416, p.24 verso) (27) kikèt akar uala meg-ol uala kikèt ´ who.pl.acc want.ipfv.3sg be.pst prt-kill.ipfv.3sg be.pst who.pl.acc akar uala meg-uèr uala want.ipfv.3sg be.pst prt-beat.ipfv.3sg be.pst ‘whom he would he slew; whom he would he put down.’ (Vienna Codex 1416, p.143) Such mismatches between situation aspect and viewpoint aspect, however, were not common enough to support the coexistence of two aspectual systems. It was viewpoint aspect marking that started fading. As a first step, the present perfect blended functionally with the simple past, eventually supplanting it completely. Val-a ‘be-pst’, the temporal auxiliary of the past perfect and past continuous paradigms, was also more and more often replaced by vol-t ‘be-pfv’, as a consequence of which the same suffix appeared in the positions of both the tense marker and the aspect marker in complex verb forms, increasing the confusion. (28) Ki hallot-t-a vol-t valamikoron ezt hog az zentoknec ´ who hear-pfv-3sg be-pfv ever this.acc that the saints.dat coronaztassec coronaia toviskel ´ crown.pass.sbjv.3sg crown.poss thorn.with ‘Who had heard before that the saints’ crown be crowned with thorns.’ (Döbrentei Codex 1508, p.5 verso) The use of volt in the past continuous (29) indicates that by beginning of the sixteenth century -t had lost its perfectivity feature; it simply marked past events, rather than past events with a present result. (29) Wgh mond zenth agoston ky thaneyt-ya vol-t oteth so says Saint Augustin who teach-ipfv.3sg be-pfv ´him ‘Saint Augustin, who was teaching him, says so.’ (Winkler Codex 1506, p.107 recto)

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i



Katalin É. Kiss

By the end of the Middle Hungarian period (1526–1772), the -t-marked past tense had become practically the only productively used nonpresent tense. The -a/e marked simple past and the complex tenses occurred less and less frequently; they were mostly used as stylistic variants without any clear aspectual value in elevated, literary texts. By now, Hungarian has returned to the dual Uralic tense system consisting of a past and a nonpast, except that the Uralic past tense morpheme has been replaced by -t-, the Old Hungarian perfectivity marker. Viewpoint aspect morphology has been lost; its function has been taken over by situation aspect marking. A similar process, that is, the replacement of morphological viewpoint aspect marking by situation aspect marking via verbal prefixes, has been attested in several Slavic languages, as well. Kiefer (2010) regards the aspectual function of preverbs as a Sprachbund phenomenon whose central area is Slavic, and which also comprises Hungarian, Lithuanian, Yiddish, German, and Romani. The changes that have taken place in the morphosyntax of Hungarian verbal inflection can be interpreted structurally as follows: the projection hosting the verbal particle, labelled as PredP (cf. Koster 1994), has been reanalysed as AspP, whereas the projection hosting the -t- suffix, labelled as AspP, has been reanalysed as TP. The original TP projection, harbouring the auxiliary vala, disappeared. That is: (30) a.

TP

T vala

→ AgrSP

AgrS -tok

AgrSP

b. AgrS -tok

AgrOP AgrO -á

AgrOP AgrO -á

AspP Asp -t

TP T -t

PredP

Spec meg Pred

Spec meg

AspP

Pred mond

Asp Asp mond

VP

VP V

V meg-mond-tá-tok val-a prt-say-pfv-2pl be-pst ‘you had said’

→

meg-mond-tá-tok prt-say-pst-2pl ‘you said’

. Theoretical implications Claims of contact-induced syntactic changes appear to be rarer in historical linguistics than claims of lexical and morphophonological contact effects. This chapter has demonstrated through the case of the Hungarian verbal complex that contact effects

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

The rise and fall of Hungarian complex tenses



can lead to pervasive changes in syntax, too. The syntactic interference of a second language presupposes the acquisition of the mother tongue in a bilingual environment; but bilingualism is believed to have been very common throughout the history of mankind. Perhaps the most basic question of contact linguistics is what can, and what cannot change in a contact situation; what triggers and what restricts contact-induced changes. The licensing conditions of the two contact-induced syntactic changes of Hungarian discussed in this chapters, the evolution of verbal complexes marking both tense and viewpoint aspect, and a thousand years later, the replacement of viewpoint aspect marking by situation aspect marking via verbal particles, suggest a possible answer. In both cases, Hungarian had a syntactic construction that resembled a construction of the lender language formally and functionally. In both cases, a translinguistic reanalysis took place: Hungarian speakers assigned to the Hungarian construction the structural–functional properties of the construction of the contact language. That is, these contact-induced syntactic changes involved the same mechanism that is attested in internal changes—apart from the fact that the trigger came from a second language. Originally, the change may have been a micro-change altering the feature content of a single element: the reanalysis of a nonfinite suffix as a perfectivity marker, or, the reinterpretation of the [±telic] feature of verbs with no resultative particle as [−telic]. These micro-changes, however, had major consequences. The featural change altered the category of the given morpheme, which necessitated the reanalysis of the phrasal constituents subsuming it. Thus the reanalysis of the -t- gerundive suffix as a perfectivity marker led to the reanalysis of the gerund phrase as a finite verbal complex, and this also involved the reanalysis of the possessive agreement on the gerund as verbal agreement, the reanalysis of its genitive subject as a nominative subject, and the reanalysis of the copula as a temporal auxiliary. In the course of the emergence of situation aspect marking, the assignment of a [−telic] feature to verbs with no verbal particles led to the reinterpretation of the verbal particle as the canonical marker of telicity, and the reanalysis of the projection harbouring it as an aspectual phrase. The reanalysed constructions could serve as input to further changes, for example, analogical extension and abstraction. Thus the removal of the perfectivity marker from the ‘perfect lexical verb + past auxiliary’ complex expressing past perfect yielded a past continuous paradigm, whereas the removal of the past tense auxiliary yielded a present perfect paradigm. A thousand years later, the development of viewpoint aspect marking by means of verbal particles led to the obsolescence of the complex tense system marking both tense and viewpoint aspect, leaving in place only two simple tenses: past and nonpast. Sources of examples DÖBRENTEI CODEX (1508). Abaffy, Csilla, Szabó, Csilla, and Madas, Edit (eds., 1995). Régi magyar kódexek 19. Budapest: Argumentum. FUNERAL SPEECH AND PRAYER [Halotti beszéd és könyörgés] (1192–5). In Magyar nyelvemlékek, (ed. József Molnár and Györgyi Simon, 1977), p. 27. Budapest: Tankönyvkiadó. JÓKAI CODEX (1370/1448). Balázs, János (ed., 1981). Budapest: Akadémiai Kiadó.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i



Katalin É. Kiss

LEGEND OF SAINT MARGARET [Margit-legenda] (1510). Balázs, János, Dömötör, Adrienne, and Pólya, Katalin (eds, 1990). Régi magyar kódexek 10. Budapest: Magyar Nyelvtudományi Társaság. MÜNCHEN CODEX (1416–35/1466). Nyíri, Antal (ed., 1971). Budapest: Akadémiai Kiadó. VIENNA CODEX [Bécsi kódex] (1416–35/1450). Mészöly, Gedeon (ed., 1916). Budapest: Magyar Tudományos Akadémia. WINKLER CODEX (1506). Pusztai, István (ed., 1988). Codices Hungarici 9. Budapest: Akadémiai Kiadó.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

 Modelling transient states in language change G E RT JA N P O S T M A

. Introduction Failed changes are often left out of studies of language change. This is understandable as studies of change are inherently post hoc: we want to know why and how a certain change has come about. Changes that did not occur do not have such a post hoc legitimation, and failed changes, that come and fade away, similarly lack such a post hoc legitimation and interest, and are equally ignored in the research. With no good reason: what doesn’t succeed and why is very relevant to any comprehensive theory of change. For instance, a failed change 0 → T → 0 can tell us something about the stability of the grammar with respect to the feature [±τ ] that is connected to T. The relevance of failed changes, however, exceeds this stability issue. Failed changes are important to understand successful changes. This can be seen as follows. Successful changes A → B, going from a stable state A to a stable state B, must have a trigger. Call this trigger T. This trigger T is necessarily temporary: it only exists as long as the transition A → B is underway. To the extent that T is linguistic—and our theory models that case—T is a failed change, and the change 0 → T → 0 must have some causal connection to the overall process A → B. In order to understand the linguistic nature of 0 → T → 0 and its interaction, with A → B, our challenge is to model this causal interaction, that is, one must build a model that includes A, T, and B: we have to design a three-state model of linguistic change. Apart from the standard two-state model of linguistic change, A → B, as discussed in Weinreich et al. (1968); Kroch (1989b); and others, three-state models have been proposed with an intermediate transient state T: A → T → B (Andersen 1973; van der Wurff 1990; Weerman 1993), where the transient state causes and fuels the transition from one stable state A into a new stable state B. The latter causal models include a linguistic trigger for the change, and these models are theoretically superior to the more descriptive two-state models. However, while the two-state model has received algebraic implementation, for example the logistic model (Bailey 1973; Altmann et al. Micro-change and Macro-change in Diachronic Syntax. First edition. Éric Mathieu and Robert Truswell (eds). This chapter © Gertjan Postma 2017. First published 2017 by Oxford University Press.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i



Gertjan Postma

1983), no algebraic implementation of the three-state model has been developed. This is what we are going to do in this chapter. The roadmap is as follows. Building further on Postma (2010), who observed that the dynamics of the transient state T (‘failed change’) has an algebraic linking to the dynamics of the overall change A → B, we present a generalized algebraic model that includes both the failed change and the successful change A → B. To extend the logistic model, we must take a step back. Instead of taking the parameterized logistic function model as basic, we propose that linguistic change A → B is ruled by a differential equation (DE) that can be interpreted as the law that governs the change. This DE has a bundle of time-shifted logistic curves as its solution. This will be identified as Kroch’s Constant Rate Hypothesis. By modifying this DE, it is possible to describe the dynamics of the entire A → T → B process, that is to say, we have a model that includes the successful and the failed change. The algebraic link found in Postma (2010) between failed change and successful change (failed change is first derivative of the successful change) turns out to hold by approximation.

. Three-state models The existing three-state models all implement the idea that natural language should be split into an I-language or Grammar of L, and an E-language, or set of utterances of L, which are generated by L’s Grammar (Chomsky 1986). In stable situations without change, the process of language transmission proceeds along the arrows drawn in the figure in (1). (1)

(2)

Grammar 1

Grammar 1

Grammar 1

Grammar 2

Output 1

Output 1

Output 1

Output 2

generations

generations

The recognition that the grammars themselves are not transmitted to the next generation is probably Andersen’s. A grammar is acquired via its output. The acquisition process of the second generation is based on the set of utterances that the following generation is exposed to. It is in this step where language acquisition can lead to change, if something distorts the perfect learning. This is represented in (2), taken from Andersen (1973: 767). The imperfection is represented by the dotted arrow, which causes the new generation to arrive at Grammar 2 instead of the target Grammar 1.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

Modelling transient states in language change



Various theories have been proposed about the nature of this imperfect learning, that is, about the nature of the dotted line. We discuss two types here. The first type concerns those theories that situate the imperfection in the learner, the person who acquires Grammar 2 instead of the target Grammar 1. This type of theory is the oldest and is well expounded in Andersen (1973). The second type of theory situates the change in the adult speakers of the previous generation, who distort their language under sociolinguistic pressure. This is well expounded in Weerman (1993), who is certainly not the first, but who has described it most explicitly. We discuss these below. .. Andersen () Andersen’s conception of change situates the change in the new generation that acquires the target grammar by trial and error. Those errors are permanently corrected. Andersen draws his examples from phonology and notes that it is difficult to infer from the perceived acoustic features the underlying articulatory features in the production. The child of the new generation hypothesizes an (articulatory) grammar and corrects it until it has reached the target Grammar 1. Or not. Failure to duplicate Grammar 1 happens if the child assumes a grammar that approximates Grammar 1, and makes a tentative correction by an Adaptive Rule (or A-Rule as Andersen coins it), probably interpreted as an Adaptive Rule on the output. This is represented in the diagram in (3).

Grammar 1

A Grammar 2 R u l e

Output 1

Output 2

(3)

generations The quality of Andersen’s proposal comes from the insight that Grammars are not transmitted directly but are acquired through their output. Secondly, and this feature is not represented in the diagram, the fact that Universal Grammar is active in the acquisition process. Thirdly, there is the attractive feature that the change in grammar is situated in the child, as only children before maturation age can set formal parameters. This makes the model in (3) an L1 model of language change, further elaborated in Lightfoot (1999) and Anderson and Lightfoot (2002). Despite these qualities, at least two objections have been raised to this model. The first objection is formulated by Bybee and Slobin (1982b), who noticed that children before maturation age do not form communities and are, therefore, unable to transmit their language. The presupposed accidental ‘imperfect learning’ therefore cannot

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i



Gertjan Postma

spread. Only adolescent and adult peer groups can.1 The second objection emerges from the fact that L1 learners are usually seen as perfect learners. Imperfect learning exists but is situated in L2 learning and other adaptive strategies not executed by the Acquisition Device of language but by some other cognitive device (say, the ‘problemsolving device’ of Chomsky 1965). This view complies with the metaphor of children as ‘little inflection machines’ (Wexler 1998: 44). This criticism is forwarded most consistently by Weerman’s observation that ‘in “normal” transmission from generation to generation children are simply too good to be responsible for transmission errors’ (Weerman 2011: 149). This leads us to the second theory of the source of language change. .. Weerman () The other conception of transmission errors situates the source of language change in adult language use (van der Wurff 1990; Sankoff and Blondeau 2008), but the most clear advocate is Weerman (1993), who designs a model that we will formalize. Weerman tries to reconcile two or three seemingly incompatible ingredients, mentioned in the previous section. These ingredients are given under (4). (4) • Only children can set linguistic parameters. • Children are perfect learners and cannot be responsible for transmission errors. • Children do not have the social infrastructure to spread a change. So, only children can change the language, but they are too good learners to do so. We will call this Weerman’s Oxymoron. To resolve this oxymoron of language change, Weerman assumes that adult speakers who have acquired a perfect Grammar 1, embellish it during the lifespan with peripheral rules, which are not part of Grammar 1, and presumably not compatible with any parametrization of UG, for instance, because it is an Output Rule that operates in the post-syntax. They do so under peer group pressure, or accommodation in language contact, or by any fashionable linguistic innovation. When exposed to this embellished grammar, Grammar 1 + Peripheral Rules, the new generation, being perfect learners of UG, (re)set the parameters in order to comply with (1) their parents’ output and (2) the requirements of UG. As Weerman writes in discussing changes in OV and VO order in the history of English: (T)hese (VO) leakages were so to speak exaggerated via L2-M acquisition, both quantitatively and qualitatively. The relevant speakers do not change their internalized setting of the head parameter. From the perspective of their L1 grammar these overgeneralizations are ungrammatical. What they do is add a peripheral rule. A next generation, however, could set the head parameter differently. (Weerman 1993: 923)

1 This claim has been challenged, e.g. in Johnson (), who argues that children whose parents speak a dialect different from the surrounding community, initially pattern with their parents, but when they start forming a peer group (i.e. once they start schooling) pattern with their peers. This might suggest that children do form communities.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

Modelling transient states in language change



This model with peripheral rules is compatible with suggestions as early as Halle (1962), when he writes: The language of the adult—and hence also the grammar that he has internalized—need not, however, remain static: it can and does, in fact, change. I conjecture that changes in later life are restricted to the addition of a few rules in the grammar and that elimination of rules and hence a wholesale restructuring of his grammar is beyond the capabilities of the average adult. (Halle 1962: 64)

The idea is also present in Lightfoot (1999: 80): Although adult innovations may not affect grammars, they . . . reflect changes to the primary linguistic data, the input experience for the next generation of language learners. Adult innovations, then, constitute one reason why an individual might be exposed to PLD which differ from what his mother was exposed to. (Lightfoot 1999: 80)

This model shares the idea of Adaptive Rule and the idea that the grammatical system is active in language change, that we encountered in Andersen (1973). The two theories, however, differ on the locus of these Peripheral (P-Rule) or Adaptive rules, as temporary deviations from true grammaticality. It is given in (5a). (5) a.

Grammar 1

Grammar 2

Output 1

Output 2

P-Rule generations Without any lack of generality, we can describe this model as a three-state model: a system A, consisting of a pure parameter setting of UG, is temporarily embellished with Peripheral Rules and gets into the excited state T. This situation, which is marked from the perspective of UG, relaxes to a new, pure grammatical state B, produced by a parameter setting of UG.

accommodation

(5) b.

G{a}

*[G{a} + PR]

G{b}

(initial state) A

(transient state) T

(final state) B

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i



Gertjan Postma

The idea is now that every change passes through a transient state. A transient state is a state that does not fully comply with the principles of UG. This transient state must be seen as the initiator of the change. Moreover, this initiator is also fully part of the change: it is a subpart of the successful change. To see this, it is important to notice that T and B together materialize the change away from A. As to the PLD, T and B are not necessarily different and are, in fact, often indistinguishable. T and B are only essentially different in the way these PLD are generated: by G{a} +PR in the case of T; by G{b} in the case of B. But as to their outer form, T and B may be equal or similar: T + B together is the change away from A. Initially, T comprises the majority of the new T + B (= non-A), while in the aftermath of the change, T comprises only a very small part of T + B (= non-A). (6) The transient state is a subpart of the successful change. In two-state models, T and B are simply collapsed into one. For the purpose of the study of successful changes, T’s splitting off from T + B (= non-A) is ignored. For the study of the causes of successful changes, however, it is essential to identify T. The fact that T and B are different in the way they are generated, but not esssentially different in the way they manifest, makes it notoriously difficult for the researcher to study transient states. In many cases, only highly marked grammatical contexts will tease apart T and B. Only in rare cases are T and B simply distinguishable at surface level. In the next section, we discuss an instructive case from the literature. .. Illustration An illustrative instance of this three-state change is found in the emergence of the reflexive in Middle Dutch. Middle Dutch had no reflexive pronoun to express contexts with argumental reflexivity, as in ‘John saw himself in the mirror’, and simply used the pronoun hem/om ‘him’ (as Old English did). In the fifteenth century, the North Eastern dialects (e.g. in the Drenthe area) began the change. The pronoun hem/om ‘him’ could not be used anymore as a reflexive, and the language borrowed a reflexive pronoun sick from a neighbouring (Low German) dialect. However, this pronoun was not well-formed because of phonological reasons, and was gradually replaced by sich, borrowed from yet another (Central German) dialect, which completed the change. The curves of the various occurrences are displayed in the figure under (7), taken from Postma (2011). During this century, when hem/om/. . . was gradually replaced by an s-reflexive, we observe a transient period with sick that emerges and fades away. These two curves maintain a formal relation to each other, which is interpreted as the sick-line to be the cause of the rise of the s-reflexive. To make the various discussions in this chapter more concrete, we will regularly refer to this simple case.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

Modelling transient states in language change



Rise of s-reflexive and sick in fifteenth century Drenthe

(7)

1.0 0.8

rate

0.6 0.4 0.2 0.0 1400

1440

1480

1520

period % s-reflexive

logistic fit to s-reflexive % sick logistic peak fit to sick

In what follows, we propose an algebraic model that captures the basic ingredients of the three-state model. Before we can do so, we need to briefly recapitulate the two-state model: the logistic model with its advantages and disadvantages. This will be done in the next section.

. The logistic model The logistic model, proposed in Altmann et al. (1983) and Kroch (1989b), is a threeparameter algebraic model that implements the insights in Weinreich et al. (1968) quantitatively. While the latter authors take any S-shape curve as the drawn line in (7) as part of their modelling, the former authors specifically opt for the logistic model. The logistic function is given in (8) for further reference. Smax (8) S(t) =   (t−t 0 ) 1 + e− a Kroch’s choice of the logistic function is motivated by pragmatic reasons. It is a widely used model in the life sciences, and it is easy to manipulate graphically: logistic functions turn into straight lines when plotted on bi-logarithmic scales. The logistic function has three parameters: the saturation level Smax , the parameter that situates the change in time t0 , and the parameter that defines the speed of the change a. It also represents the broadness of the bell curve in (7).

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i



Gertjan Postma

The saturation level Smax is usually scalable to 1.2 We then have two parameters. The parameter that situates the process in time, t0 , or actuation time, is the inflection point in (7). On a logarithmic plot, where the S-curve is a straight line, it is the intercept. Variation in actuation time causes a horizontal shift, illustrated with a linear plot in (9).3

1.0

(9)

rate

0.8 0.6 0.4 0.2 0.0

1400 actuation 1

1440 Year actuation 2

1480 actuation 3

The parameter that defines the speed of the change, or rate in Kroch (1989b), is the steepness a of the S-curve. On a logarithmic plot it is the slope of the straight line. In (10) we illustrate variation in a in a linear plot.

1.0

(10)

rate

0.8 0.6 0.4 0.2 0.0

1400 slope 1

1440 Year slope 2

1480 slope 3

2 By dividing both members by S max . But this is not always possible, for instance if the change from A to B involves an increase in the use of the construction. In that case, one does not know when the change completes. 3 Our parameters are natural, i.e. interpretable, a and t , which are both measured in years/time units. 0 Typical values in () are a ≈ 25 years for (half of) the broadness of the bell curve, while t0 ≈ 1460. There is a straighforward conversion to Kroch’s (b) regression parameters s (slope), which has ‘per time unit’ as its dimension, and k (intercept) which is dimensionless. The conversion is: t0 = −k/s, a = −1/s or s = −1/a; k = t0 /a.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

Modelling transient states in language change



In the next section we dicuss these parameters in relation to the Constant Rate Hypothesis. .. Kroch’s Constant Rate Hypothesis A change in the grammar does not always show up in the E-language in all contexts at the same time. Usually a change starts in a certain context, and extends its scope to other contexts step by step, until it comes to completion. A famous example is do-support in the history of English, which starts in negative wh-questions and then proceeds to other contexts. Negative declaratives are the last context affected. Kroch (1989b) suggests that changes in the E-language that are manifestations of one and the same change in the I-language share a property: the speed with which they proceed (their rates) is equal although their actuation times may be distinct. In terms of the logistic function in (8), it turns out that the various curves share the slope parameter a, while their actuation times t0 are different. In other words, the various curves in (9) may be manifestations of the same change in parameter, while the various curves in (10) cannot. This is the Constant Rate Hypothesis (CRH).4 The CRH has turned out to be a major tool in the analysis of linguistic change. However, it has not been derived from deeper principles. Let us finish this section with a summary of the advantages and disadvantages of the logistic model. (11)

Advantages of Kroch’s logistic model • simple logistic two- or three-parameter model; • well-known from life sciences; • can be easily displayed and graphically manipulated; • Constant Rate Hypothesis ties E-language process to the I-language (P&P model).

There are also disadvantages of this model. The first disadvantage is the subject of this chapter: the relation with transient states under S-curves cannot be made, as was first noticed by Anthony Warner: ‘. . . the “Constant Rate Effect” . . . [has] the difficulty that the relationship between affirmative declarative do and the other contexts is not clear’ (Warner 2006: 49). This is related to another property of the logistic model: it is a rigid three- or two-parameter model, that cannot be modified, for example when we deal with datasets that deviate from this ideal case, nor is it clear under what circumstance it is expected to hold. The disadvantages are listed under (12). (12) Disadvantages of Kroch’s logistic model • relation with the transient state is not expressed; • Constant Rate Hypothesis remains a stipulation; • the model cannot be modified; • no instructions are provided under what circumstances it holds. In the next section, we sketch how we can transcend these limitations without losing the advantageous features.

4 The CRH could more transparently be called the Equal Rate Hypothesis. For convenience, we stick to the traditional terminology.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i



Gertjan Postma

.. The underlying differential equation As said, Kroch’s CRH is an empirical generalization. It has not been derived from underlying principles. At this point, it must be noted that logistic curves describe the kinematics of a change in the time. They do not describe the actual process behind the change: how do the various players interact dynamically? It may be clear that a description of the outer manifestation cannot give insight into the connection between these kinematics. In this section we show how a shift from functions (describing the kinematics of a change) to differential equations, describing the fundamental interactions, sheds light on the CRH. To illustrate the step, let us first take a simple case of Malthusian growth, as described kinematically by an exponential function under (13). (13)

100

Growth according to Malthus

population

80 60 40 20 0 1300

1400 Year Malthusian growth

1500

As Malthus showed in 1826, the exponential growth is a consequence of a fundamental property of life: the increase over time of a population S is proportional to the size of the population S. The bigger the population, the quicker it grows. This can be expressed mathematically by a differential equation: the incremental change in S, dS, over an incremental lapse of time dt, that is, dS/dt, is proportional to S, as given in (14a). This differential equation has a solution, or rather, a bundle of solutions, as given in (14b). (14) a.

dS dt

= aS

(Malthus 1826)

b. S(t) = ea.(t−t0 ) This is the exponential function. For every parameter a in (14a), there is a set of timeshifted solutions, defined by the various t0 . This model of exponential growth was modified by Verhulst in 1845. Verhulst noticed that there is a counterforce that limits the growth. The more a population grows, the fewer commodities there are to serve it. Verhulst, therefore, added a factor (1−S) to the equation in (14a), which reduces the growth when S becomes bigger. The new differential equation is given in (15a). Its solution is an S-curve, the logistic function. It has, once again, not one solution, but a bundle of time-shifted solutions, defined by the various t0 . It is given in (15b).

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

Modelling transient states in language change dS = aS(1 − S) dt 1 b. S(t) =   (t−t 0 ) 1 + e− a

(15) a.



(Verhulst 1845)

Verhulst’s strategy may be clear: DEs have interpretations, which are usually rather easy to inspect from their mathematical shape. In our case: the change in S is proportional to two factors: the available procreators S and the available commodities. This makes DEs modifiable in a nonarbitrary way. One can argue against or in favour of each factor or term in the DE. Subsequently one finds an analytic solution or a solution by computer simulation, instead of modifying the functions directly. The interpretability of the differential equation is essential. Let us now return to the language change of A → B, which is described by an Scurve (15b). The related DE in (15a) has at least two interpretations that are relevant for the present discussion. There is the interpretation by Verhulst: apart from the factor that fuels a change, say the wish for offspring/innovation, there is a counterforce, say, food/social acceptability, or whatever. These two factors interact in a certain way. Adding (1 − S) to (14a) leads to (15a) but (14a) can in principle be modified by another type of factor. There is a second interpretation, which is statistical in nature. If S represents the number of people that have shifted from A to the innovative B, 1 − S is the number of people that has not undergone the change. We therefore can give the following interpretation to the underlying DE in (15a). Let us assume that speakers influence each other when they meet. If a B-speaker meets an A-speaker, A may change to B. The increase in the number of speakers of B is proportional to the probability that a speaker of B meets a person that is still an A-speaker. If we assume the extremely simple model where all speakers meet each other with equal probability, the DE in (15a) holds. Notice that the DE in (15a) is the underlying equation that has the logistic function as its solution. Notice further that this solution is not unique but has a bundle of time-shifted solutions. The constant a in (15b) is fixed by a in (15a), but t0 in (15b) is completely free. So, if the equation in (15a) describes a language change, we predict many solutions that instantiate this change. These may have different actuation times, t0 , but must have equal slopes or rates, a. This derives Kroch’s Constant Rate Hypothesis. We also now know when we may expect the CRH to hold: only in the case that all (relevant) speakers meet each other by equal chance, that is, without geographical effects and without diffusion, is (15a) the applicable DE and does the CRH hold. In the next section we apply this promising method to the transient state. We may summarize the conditions of the logistic model and the CRH as in (16). (16) Kroch’s logistic model The DE underlying the logistic model (Verhulst’s equation) only holds iff: • there are just two variants in competition: A and not-A; • there is only a first-order interaction, i.e. no inherent decline or increase (e.g. increase of literacy might favour a certain construction);

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i



Gertjan Postma • everyone meets with everyone with equal chance (no geographical effects, no diffusion, etc.); • all logistic solutions of the DE have the same slope a, but may differ in actuation time t0 .

Before we can modify the underlying DE of the logistic model with the presence of a transient state in Section 6.4, we reflect for a moment on the interpretability of the models that use differential equations. .. On the interpretation of differential equations The history of differential equations is intimately linked to the idea of conservation laws. In fact, Isaac Newton developed both. Differential equations are supposed to describe the deeper reality modulo the actual realization in space and time. Let me give an example. If we throw a ball, we know that it follows a curved trajectory, in fact, we know it follows a parabola. The actual trajectory of the ball is a complex function of the initial conditions of the throwing event and properties of the gravitational field, properties of the air, etc. We observe a complex world where everything changes: the position as well as the speed of the ball is different at every moment. There is, however, a deeper reality, which transcends the reality of space and time. In this reality—the world of energy (E) and momentum (p)—things do not change. The description of the throwing process is much simpler in this energy–momentum space. For instance: Newton taught us that the total energy does not change in time, it is a constant in time. This conservation law is in fact a differential equation. The Newtonian E = p2 /2m (or the Einsteinian E = mc2 ), together with the conservation laws, and the energy stored in the ball in the gravitational field (V = gx) provide us with a differential equation, (dp/dx)2 = 2mgx, of which the actual complex trajectory, for example the parabola of the Newtonian space, is a solution. There are, in fact, infinitely many solutions, an infinite number of trajectories. The boundary conditions of the throwing event select one of these. While the trajectories are the visible space, the Newtonian laws represent the interpretable space. We can add terms to the differential equation with specific interpretations: terms that represent the gravitational field, terms that represent frictional terms, etc. In the actual trajectory these terms add up in a complex way. In the trajectories, these contributing factors are not easy to separate. Although the parabola can be considered ‘a correct model of trajectories of ball throwing’, it does not represent a full understanding of what is going on. The differential equations, on the other hand, do offer such an interpretive framework. This step from trajectories to differential equations which represent abstract conservation laws is a step from kinematics to dynamics. The latter represents the forces that are active in the system at hand. An anonymous reviewer raises the interesting question how differential equations as mathematical models differ from other statistical models in their interpretability. This is an interesting question to which only a provisional answer can be given. All models must of course have an interpretation, or at least should have, but not all models are equally interpretable, in the sense that one can (easily) assign an

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

Modelling transient states in language change



interpretation to every subpart. To express this issue in linguistic terms: a kind of composition principle seems to hold in the case of differential equations. A logistic function is more a wholesale parametrized model. Call it ‘construction grammar’. If we apply this way of reasoning to linguistics, we do recognize Kroch’s logistic function of linguistic change as a trajectory, but the model does not provide us with an interpretation. As we have seen, it is much easier to give an interpretation to the differential equation of which the logistic is a solution. The DE can be modified with other, equally interpretable terms.

. The three-state model In the previous section we reported a well-established method from physics and the life sciences to modify processes in a nonarbitrary, interpretable way. One adds a well-understood term or factor to the ruling differential equation and solves the new equation analytically or by computer simulation. In this section we apply it to failed changes. .. The failed change model (Postma ) A failed change is defined as a combination of a monotone increasing logistic function and a subsequent monotone decreasing logistic function with the same slope. A change is ‘inherently failing’ if the actuation time of the increasing and decreasing logistic function coincide. If we denote the increasing logistic function with A, then the decreasing logistic function with equal actuation time and opposite slope is (1−A). The failed change is therefore A(1 − A), and by using the DE in (15a), repeated here under (17), we conclude that the failed change T = A(1 − A) is proportional to the first derivative of the successful change. (17)

dA = aA(1 − A) dt

So, according to the failed change model, (17) has two interpretations. (1) It can be interpreted as a differential equation that describes the successful change A. (2) It can be interpreted as a relation between the successful change A and the failed change defined as T = A(1 − A). The failed change is the first derivative of the successful change. dA = a.T dt It has turned out that this model gives quite precise predictions about the transient state that occurs during a successful change. Not only is the example given above of the rise of reflexive in Middle Dutch nicely described (cf. (7)), the failed change of do-support in positive affirmative contexts also comes out rather precisely. The defect of this model is that the failed and successful change are not described by one set of (coupled) DEs, but by one DE (17) which is interpreted in two different ways. In the next section we will correct that by developing the transient state model. (18)

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i



Gertjan Postma

.. Preparatory step: Transition diagrams Verhulst’s equation describes the transition of A → B by describing the rise of B or the decline of A together with a conservation law of speakers: A + B = constant. In the model no persons die, dissolve in the air, or become non-speakers. This can also be expressed by two coupled differential equations as in (19), which is fully equivalent to (17). (19)

dA = −a.A.B dt dB = a.A.B dt

The equivalence can be inspected by taking the sum of both members in (19), resulting in d(A+B)/dt = 0, which states that the change in A+B is 0, that is, A+B is constant.5 By substituting B = 1 − A in the first equation of (19) we derive (17). Similarly we can go from (19) to (17) by substitution of 1 − A = B in the first equation in (19). A two-state change A → B can, hence, be expressed by two coupled DEs. The equations in (19) have an interesting interpretation. The transition away from A (e.g. the hem reflexive) and towards B (s-reflexive) comes about by the interaction of A and B, that is, the chance that A meets B (i.e. a hem speaker meets a sick/sich speaker), multiplied by the chance a that this encounter leads to a change from A (hem) to B (sick/sich), that is, the transition coefficient a. In this case the relation between the two equations in (19) is trivial: the amount by which A (hem) diminishes is equal to the amount by which B (sick/sich) increases. We can visualize it in the transition diagram in (20), as a kind of visual shorthand for (19). (20)

Transition diagram A → B aA.B

or even simpler

Transition diagram A → B a

In this model, we have glossed over the distinction of the two s-reflexives. We have collapsed them as representatives of the new s-reflexive but glossed over their different behaviours: sick comes up and disappears, while sich increases. We equally gloss over the fact that sick is initially the dominant form of the s-reflexive, while sich is the dominant form of the s-reflexive at the end. In sum, we ignore that sick acts as the (ephemeral) trigger of the change. In the next section, we apply this transition diagram visualization to the three-state model (i.e. we make a model that takes track of the function of sick). .. Modelling transient states Let us assume a simple stable model with no overall increase or decrease in speakers and where children simply replace their parents in society numerically and linguistically. So, if the child enters society, we consider the parent to be removed. This is 5

This constant is . We always normalize to  in this chapter.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

Modelling transient states in language change



a rough approximation which excludes interactions between generations in society. By using the techniques developed in the previous section, we can model a three-state change A → T → B by assuming the transition diagram in (21). (21)

Transition diagram A → T → βA.T αT.B

B

This leads to the coupled differential equations in (22). (22) a.

dA = −βA.T dt

b.

dT = +βA.T − αT.T − αT.B dt

c.

dB = dt

+ αT.T + αT.B

Let us interpret the terms. The term A.T is the chance of an encounter of an A-speaker and an innovator T (say the sick speaker in our example), who has changed the language with a peripheral rule. If these two speakers meet, speaker A turns into an innovator with chance β. We may call this an L2-interaction, and β the L2 coupling constant. The term T.B turns an innovator into a B-speaker. This involves a parameter reset, which is only possible upon L1 acquisition, that is when the child replaces the adult. One should keep in mind that the innovation T is not really grammatical, as was assumed in Weerman’s model, and upon this L1interaction, the innovator cannot resist turning into the more grammatical B (sich in the case of the example). It comes about with factor α. Finally, there is the term T.T which represents the encounter of an innovator with another innovator. To understand what this means, it is important to notice that we assume that nothing happens whenever A meets A, or whenever B meets B: no terms A.A or B.B are included. These are grammatical utterances in a homogeneous speech interaction, which do not cause a change. These are encounters of two hem-sepakers or two sich-speakers: no change happens. This is different, however, in the case of a T–T encounter. It is generally assumed in innovation/rumour theory that innovators stop using the innovation as soon they meet other innovators (Piqueira 2010)—as if a woman in a Viktor & Rolf dress meets another woman in a similar dress. She will never wear it again. In the case of language change we must split this term into an adult–adult interaction and an adult–child interaction (or rather child–adult replacement in our model). The adult–adult interaction does not contribute a change as adults cannot reset parameters, which is needed to switch to B. The adult–child interaction, or rather, when the child replaces the parent in society, results in B. A similar reasoning holds for the interaction between T and B. Switching to B needs a child. So, while the transition coefficient β involves L2-interaction, the transition coefficient α involves L1. So we fundamentally have built in the transient intermediate state with its properties into the model. This model has been described and computer-simulated in Piqueira (2010). It

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i



Gertjan Postma

is an implementation of the famous SIR-model6 in biology modelling, the spreading of diseases (Kermack and McKendrick 1927).7 The coupled differential equations in (22) are not solvable by analytic methods. We have solved them using numerical methods.8 The solution is displayed in the figure in (23).9

Piqueira Model (σ=0.25)

(23)

1.0 0.8

rate

0.6 0.4 0.2 0.0 0

50

100 time T

150

200

T+B

We see a rather similar picture to the figure in (7), which displays the failed change of sick in relation to the successful change of the s-reflexive in the failed change model, where the failed change is the first derivative (or slope function) of the successful change. In the new model, however, this relation is an approximation. Notice that (23) is not a fit to the historical data, which we have not performed yet. It would mean an in-tandem solving of differential equations and a fitting process. We hope to carry that out in further research. What the present discussion tells us, is that we succeeded in making a consistent model of a realistic linguistic process that includes the successful and the failed change (transient state), where the transient state is the initiator and driving force of the successful change. A final comment on (23) is in order. It must be noticed that the T + B line does not reach the full saturation level of 100 per cent. This means that a small number of A-speakers remain after the T-innovation has swept through the community. The reason is that we did not include a term of direct interaction between A and B, say 6

Susceptible–Infected–Removed in the epidemiological model. Piqueira’s model is an application of this biological model to the spreading of information: the Daley– Kendall model. The three groups are called Stiflers, Spreaders, and Ignorants. The differential equations in Piqueira’s . and ours (cf. ()) are identical, modulo the error in Piqueira’s .. 8 The Runge–Kutta method. We used the program SolveDiffEq . for MacOS by Bob Delaney’s Science Software . 9 SolveDiffEq does not have a graphical component. We used ProFit .. for MacOS. 7

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

Modelling transient states in language change



γ A.B. If we add such a term with negative sign to (22a) and with positive sign to (22c), full completion of the change is guaranteed. However, this simply superposes our new model with the terms of the old model (19). These terms are inherently asymmetric. Alternatively, one might add a symmetric change like A.B.B → B and A.A.B → A to the system. This three-person interaction would add an effect of accommodation to the majority. In terms of the equations in (22) it would add a term AB(A − B) to (22a) and AB(B − A) to (22c). However, as these changes can only happen within the family (from A to B is a parameter switch), we must assume core families with more than two adults.

. Competing grammars approach The competing grammars interpretation of the logistic function (Kroch 1989b; Yang 2002; Fruehwald and Wallenberg 2013) does not situate the variation found in the E-language as a distribution of A speakers and B speakers (as discrete properties), but as a relative activation of two grammars A and B present in each individual. This interpretation can be characterized as an internalizaton or psychologization of the variational space, that is, as a property of I-language. In the extreme case, it is possible that a completely homogeneous community with an equal A–B grammar contribution in each individual, shifts from the initial value 1 to the final value 0. If we assume that the A/A + B and B/A + B ratios represent the chance that, given a particular individual x, the A grammar or the B grammar is activated in x, and if we assume that these ratios are equal in all speakers,10 we end up with a chance that the A grammar in individual 1 ‘meets’ the B grammar in individual 2. We can now simply leave out the level of individuals from the theory and model the encounter of grammar A and grammar B, as if they were meeting directly. Mathematically we end up with the same differential equations and (logistic) solutions. Put differently, this is another interpretation of the same mathematical modelling. However, in principle, the competing grammars interpretation allows for modelling both the individual-internal variational space and the individual-external variational space, and probably some interaction between these. This model is therefore potentially a richer model of which the mathematical complexities exceeds the scope of this chapter. We therefore limit our discussion to the external varational space. This raises the question of whether the failed change model has a similar internalized interpretation. This is problematic. While it is perfectly reasonable to assume two grammars in one individual, or that one construction may receive a double analysis in one individual, it is difficult to understand how a specific surface construction might have both a marked and an unmarked analysis: a marked one with grammar A + peripheral rules, which is transient, and a neutral unmarked analysis that is produced by grammar B. If so, a construction might have a marked, say peer group, reading in grammar A and a completely neutral reading in grammar B. It would be reasonable of optimizing parses that only the most neutral, most optimal 10 We ignore that fact that it is not a logical necessity that the activation of grammars, A or B, be equal in speakers and hearers.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i



Gertjan Postma

parse surfaces. Moreover, we assume that reanalysing the construction with the Bgrammar was the prerogative of a young pre-maturity individual, while analysing it with peripheral rules is possible for a post-maturity individual with the A grammar. We, therefore, must use the external variational space between the individuals if we want to include failed changes in the competing grammars interpretation. This is not necessarily a disadvantage, as the internal and the external degrees of freedom do not necessarily add independent complexity to the model, and we cannot exclude the external variational space in the competing grammars approach without ad hoc stipulations anyway. This brings us to the interpretation of the Constant Rate Effect in the competing grammars approach. Notice that the Constant Rate Effect makes crucial use of variation in the intercept parameter of the logistic, which we interpreted as the actuation time according to standard assumptions. Interestingly, Kroch (1989b) proposes a different interpretation of this intercept parameter, which is crucial in his explanation of the Constant Rate Effect. For Kroch, the intercept does not reflect the different actuation points for different contexts, but instead the differential advantage of the two forms in various contexts. Thus, some contexts will favour the incoming variant, which will be used more frequently in that context at all time periods, while others will favour the outgoing variant. This is a crucial part of Kroch’s theory that only one change is actually occurring with one actuation, but with distinct advantages. The change actuates only once, and then progresses at the same rate through all the contexts. Different contexts, however, can have preferences for one form which impacts the E-language distribution. Thus, Kroch argues that since there is only one change, it can have only one slope. This intercept implementation of the ‘advantages’ is not compulsory, however. It is possible to implement the ‘differential advantages and context favourings’ by varying the intercept, which Kroch opts for. It is equally possible to implement the ‘differential advantage’ by varying the rate. In fact, the first option is problematic. In the first place, varying the intercept varies the actuation time, and for dimensional reasons alone, one cannot hold the intercept directly responsible for a favouring upstep. The relation is indirect due to the monotone relation between actuation time and occurrence threshold. One can only calculate a corresponding threshold by using the monotoneincreasing relation between both given by the logistic function. Moreover, in Kroch’s interpretation the ‘favouring’ by certain contexts only happens once at the actuation time, where some contexts get a favouring threshold, an upstep at the very beginning, from which point on, all contexts are treated equally. It is not clear, however, why this ‘favouring’, ‘advantage’, or ‘upstep’ would not be present at intermediate times, or infinitesimally at all times. This would in fact change the rate. But the major argument against the threshold interpretation is empirical. There is no indication that some contexts start out with a threshold in the E-language. All observed curves are simply complete logistic curves that are slightly shifted with respect to each other. An obvious objection against this counter-argument is that the threshold occurs within the individual speaker, in the internal dynamics of the change, not in the data of the E-language. This objection is legitimate but shows that the competing grammars interpretation crucially uses both the internal and external

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

Modelling transient states in language change



variational space. If this model is mathematically expounded in full detail, it will have a complexity of which our modelling is only a simple subcase.

. Conclusions It is possible to generalize the logistic model of language change from a kinematic modelling of the change to a dynamic differential equation that rules the change: the differential equation captures the respective interactions as a process law. In the simplest case, this law takes the shape of Verhulst’s equation. The solution of this equation is a logistic function, or rather, a bundle of the logistic functions, which are timeshifted with respect to each other. This derives the Constant Rate Hypothesis. By modifying Verhulst’s equation with an intermediate state, we made a unified model (coupled differential equations) that models both the successful change and the failed change within one model. The failed change is the first derivative of the successful change, by approximation only.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

 Modelling interactions between morphosyntactic changes H E Z E K IA H A K I VA BAC OVC I N

. Introduction This chapter revisits the modelling of ‘failed changes’ (Postma 2010). Instead of showing the characteristic S-shaped curve of successful changes, failed changes show a rise–fall pattern. Postma (2010) proposes a model for failed changes as the first derivative of the logistic model of its successful counterpart. I start by discussing the statistical underpinnings of the logistic model, its relationship to the Constant Rate Hypothesis for successful changes, and how Postma uses these components to develop his model. Ultimately, however, the first derivative model proves inadequate for describing the data from the rise of English dative to (e.g. ‘I gave the book to John’). Instead, I implement one of Postma’s other suggested models, the probability multiplication model. The failed change in English dative to is modelled as the interaction between two countervailing changes, that is, one change with a positive slope and another change with a negative slope. These two combine via multiplication of their probabilities to generate the failed change’s characteristic rise–fall pattern. Since the two changes can be disentangled, it is possible to test the Constant Rate Hypothesis on the slopes of the component changes. This probability multiplication model provides an example of how micro-changes can interact to create a larger, more complex change.

. Failed changes and the Constant Rate Hypothesis Postma (2010) advances his first derivative model as a way of connecting a failed change and a successful counterpart, using a modified version of the Constant Rate Hypothesis. The standard version of the Constant Rate Hypothesis from Kroch (1989a) is a restriction on the possible values for the slope parameter using a logistic model of syntactic change. The logistic has the statistically useful property of being reducible to a linear model of the log-odds of a particular change. The log-odds are simply the natural log of the Micro-change and Macro-change in Diachronic Syntax. First edition. Éric Mathieu and Robert Truswell (eds). This chapter © Hezekiah Akiva Bacovcin 2017. First published 2017 by Oxford University Press.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

Modelling interactions between morphosyntactic changes



odds (i.e. ln(odds)), which for any given year is equal to the probability of the new p variant divided by the probability of the old variant (i.e. 1−p since the probability of the new variant and old variant must sum to 1). As a ratio, an odds of 1 means that the two variants are equally likely (i.e. a probability of .5). The odds range down from 1 to asymptotically close to 0 indicating that the new variant is less likely than the old variant, or from 1 to asymptotically close to positive infinity indicating that the new variant is more likely than the old variant. The logistic provides a transform from the log-odds (defined as Intercept + Slope * Year) to a probability (i.e. a number between 0 and 1). The Intercept reflects a timeinvariant tendency towards or against the new variant. The Slope reflects how much more likely the new variant will be for every increment of the year variable. The Constant Rate Effect is the fact that across a number of different studies of syntactic change in a variety of different languages the Slope parameters in different grammatical contexts were found to be equal (e.g. for English do-support: negative declaratives, questions, imperatives, etc., see Kroch 1989a for an early discussion of these studies). The Constant Rate Hypothesis says that the equal rates found in a number of different studies reflects the adoption of a single grammatical change over time. The differing Intercepts found in different grammatical contexts reflect time-invariant contextual effects on the application of the incoming grammar. The theoretical interpretation of Intercepts given above contrasts with the equally statistically valid interpretation given by Postma (2010) of differing Intercepts reflecting different actuation points for distinct changes. For Kroch, since the Constant Rate Effect is derived via the spread of a single change through the speech community, the change only has one actuation point. The Intercept parameter for each context reflects time-invariant contextual factors favouring or disfavouring the use of the new grammar. Postma in this volume instead derives the Constant Rate Effect via a single change being related to a particular differential equation, which describes the progression of that change. Each context is thus capable of having a different actuation point, but is constrained to behave as previous iterations of the change. Either interpretation of the Constant Rate Hypothesis provides a way of testing whether or not two possibly related surface changes reflect a change in a single underlying process. If two changes are demonstrated to have different Slope parameters, then, by hypothesis, either: (a) they must reflect two distinct changes or (b) some additional factor must be interfering with the simple relationship between the grammar and the surface forms. Occam’s Razor would suggest that without external verification of the interfering factor, the interpretation of different Slopes as differing changes is preferred. It is also possible to use the Constant Rate Hypothesis to argue that two surface changes reflect the same underlying process. The more confident that we are that two surface changes have the same Slope, the more confident we should be that they are the reflection of a single grammatical change. Demonstrating that two slopes are equal requires a test of statistical power in order to demonstrate that there exist enough data to be able to detect a reasonably large difference if it should exist (Streiner 2012). In order to be able to apply this reasoning to failed changes, an appropriate statistical model of failed changes needs to be developed, as well as a way of

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i



Hezekiah Akiva Bacovcin 1.00

Probability

0.75

0.50

0.25

0.00 –10

–5 First derivative

0 X

5

10

Logistic function

Figure .. A logistic curve and its first derivative

relating them to successful counterparts, such that the Constant Rate Effect could be found. Postma (2010) points out that the linear logistic model provides a poor fit for failed changes. When transformed from the linear log-odds to probability space, the logistic model produces the characteristic S-shaped curve found in successful syntactic changes. However, the S-shaped curve is a poor fit for the U-shaped curve found in failed changes. Postma shows that the first derivative of the logistic function produces a U-shaped curve (Figure 7.1). However, requiring a failed change to be related to a successful change by being the successful change’s first derivative essentially reduces to a strong version of the Constant Rate Hypothesis, namely that both the Slope and Intercept parameters be the same between the successful (logistic) change and the failed (derivative) change. The identical intercept parameters forces the peak of the U-shaped curve to be at the same time as the middle of the S-shaped curve. The first derivative ends up being mathematically equivalent to p * (1 − p), where p is the probability given by the S-shaped curve of the successful change. There exists a class of probability multiplication models of failed changes of the form p1 * p2. The first derivative model is the special case of this class in which the two probabilities are drawn from a logistic curve and its negative sloped complement (i.e. a curve that is identical to the positive one except for the sign of the slope parameter). Thus, the first derivative model predicts that a failed change should (a) always be symmetrical about its peak (since the value of the slope parameter is identical in the two changes) and (b) have its peak under the midpoint of the successful change’s S-shaped curve (since the Intercept parameter is the same between the successful change and the two components of the failed change). In the following section, I show that neither of these properties hold in a failed change in the development of English dative to.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

Modelling interactions between morphosyntactic changes



. English dative to The phenomenon under discussion in this section is the rise in the use of to in place of dative case in recipient constructions. Rappaport Hovav and Levin (2008) call the relevant verb classes transfer of possession verbs (e.g. give) and future transfer of possession verbs (e.g. promise).1 The new possessor is the recipient and the newly possessed item is the theme. (1)

a. Recipient–Theme: I gave (to) John the books. b. Theme–Recipient: I gave the books (to) John.

Rappaport Hovav and Levin (2008) show that verbs of (future) transfer of possession only license a recipient argument (new possessor) and not a goal (endpoint of movement). However, a related class of verbs, verbs of sending, license both recipient and goal arguments. The goal arguments reflect the endpoint of a path of motion. In Old English, the two thematic roles were morphosyntactically distinguished. Recipients were introduced with synthetic dative case, while goals were introduced prepositionally with to. This pattern of morphosyntactic marking meant that Old English verbs of transfer (e.g. sellan ‘give’) never occurred with to. McFadden (2002) and Polo (2002) show that the use of to with recipients in verbs of transfer enters the grammar once the morphological marking of the dative/accusative distinction had been completely lost. This process of case erosion went to completion during the twelfth and thirteenth centuries, which are among the worst-attested periods for the English language. During this early period of the change, the texts that still maintain a dative/accusative distinction (usually only on pronouns) resist using to to mark recipients. By the end of the thirteenth century, almost all dialects of English had completely lost the dative/accusative distinction (it survived in some Kentish documents into the fourteenth century). I propose that dative to, used to mark recipients, entered the language via reanalysis. An adult language user produced to-marked goals with verbs of sending, which were erroneously interpreted by the child as a to-marked recipient. Since animate goals are always ambiguous with a recipient interpretation semantically (assuming that the theme actually reaches the goal), the children’s error is entirely expected. Before the loss of synthetic case marking, these errors would be corrected by the existence of another way of realizing dative case (namely the synthetic dative case morpheme). After the loss of the synthetic dative marker, there was no longer any overt form for dative case to contradict the new to-as-dative grammar. (2) a. Adult Intention: I sent the package [goal to John]. b. Child Interpretation: I sent the package [recipient to John]. Goal to introduces a prepositional argument phrase. Throughout the history of English, the default position for prepositional argument phrases follows that of 1 In particular, the data given here are derived from the Middle and Modern English verbs: allot, appoint, assign, ayeven, behieght, bequeath, betake, give, grant, lend, offer, owe, pay, proffer, promise, restore, sell, serve, show, vouchsafe, and yield. For Old English, the verbs agiefan, behatan, giefan, dælan, offrian, and sellan were used.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i



Hezekiah Akiva Bacovcin 100%

T

T

T

TT TTT T T TT

T

Context

Percent to use

80% T

R

60%

R

50%

R

40% T

20%

R R

R

800

R T

R

1000

I gave theme (to) recipient

Less than 50 50 or more

R

R

Context I gave theme (to) recipient

R

T

0%

I gave (to) recipient theme

T

Token number

R

T

R

R

R T R R

1200 1400 1600 Year of composition

R R

R

R

I gave (to) recipient theme

R

1800

2000

Figure .. Graph of English to data with the fits from the probability multiplication model

nominal arguments. While different scrambling operations can alter that order (e.g. Heavy NP Shift), the theme–prepositional object order is dominant. (3) a. I put the book on the table. b. ??I put on the table the book. Since the dative to has its origins in a prepositional object construction, it is expected that the theme–recipient order would lead the rise of dative to (i.e. that it should have a higher Intercept than the recipient–theme order). The expected relationship between the theme–recipient and recipient–theme orders can be seen in Figure 7.2, where to-marking is always more likely in the theme–recipient order than in the recipient–theme order. The historical data discussed here is drawn from Kroch and Taylor (2000); Taylor et al. (2003); Kroch et al. (2004); Taylor et al. (2006); and Kroch et al. (2010).2 This data only includes cases where both the theme and recipient are full noun phrases and follow the verb phrase. Pronominal objects have their own movement and marking patterns, which introduce complications that are irrelevant to the point of this chapter. There is not enough data for noncanonical word orders (wh-questions, topicalization, IP scrambling) to do statistical inference. Throughout the thirteenth and fourteenth centuries both the theme–recipient and recipient–theme order show an increase in to use. In the early fifteenth century, use of to in the recipient–theme order begins to decline, showing the characteristic rise–fall pattern of a failed change. McFadden (2002) proposed that to in the theme– recipient context is the reflex of a prepositional object construction, with to in the 2 The data used for this paper came from the current working files version of these corpora, so ID numbers may not match with older versions. The online supplement at http://www.oup.co.uk/companion/ DiGS contains a copy of all the data used in this paper, as well as the R code used to fit the models, run the simulations, and generate all of the graphs and tables.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

Modelling interactions between morphosyntactic changes



Table .. Comparison of Heavy NP Shift models using Likelihood Ratio, AIC, and BIC Resid. Df Resid. Dev Df Deviance Pr(> Chisq) Null model Only heaviness of theme Year and heaviness of theme

AIC

BIC

967 965

1052.09 1015.79

2

36.30

0.00

1054.09 1058.97 1021 1036

964

948.81

1

66.98

0.00

956

976

recipient–theme context exclusively being derived by Heavy NP Shifting the theme over the recipient. Heavy NP Shift is the main movement operation in Middle and Modern English that changes postverbal word order. McFadden explained the rise in to-marking in the recipient–theme order by arguing that there were more heavy themes in the later Middle English period, and thus more Heavy NP Shift. Before the change has gone to completion in the theme–recipient context, not all cases of Heavy NP Shift would have to, since it could be derived from a to-less theme–recipient construction. After about 1400, however, when to is categorical in the theme–recipient context, all cases of Heavy NP Shift should have to. Assuming that Heavy NP Shift is creating all the to-marking in the recipient–theme context, after 1400 the only statistically significant predictor should be the weight of the theme (measured by number of words or by whether or not the DP contains a CP complement or relative clause). There should be no significant effect of Year, since the variation between different years should reflect (accidental) differences in the weight of the theme over time. Table 7.1 explicitly compares a model which includes both the weight of the theme and year to a model with only the weight of the theme. According to the Log-likelihood Ratio Test (Pr(> Chisq)), the AIC, and the BIC,3 the model including year is the more reasonable fit, that is adding a slope for year significantly improves the model’s ability to predict to use in the recipient–theme context in the post-1400 period. This significant slope parameter indicates that the fall seen in Figure 7.2 is significant independent from any changes in the prevalence of heavy themes and suggests that the rise found in the pre-1400 period also reflected a real change distinct from the amount of Heavy NP Shift. Since the rise–fall pattern in the recipient–theme context is not epiphenomenal of variation in theme weight, it plausibly represents an example of a failed change. All recipients marked with dative case in Old English began to be marked with to, that is, to became the new morphological reflex of dative case. In the theme–recipient context, this change was successful and to became obligatory in the same way that synthetic case was obligatory in Old English. In the recipient–theme context, the change failed. To-marking initially rose, just as in the theme–recipient context, but it subsequently fell again, returning to the original state of null marking. 3 A Pr(> Chisq) less than . indicates that the more complex model is significantly better than the less complex model, while for AIC and BIC, the model with the lowest value is to be preferred.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i



Hezekiah Akiva Bacovcin

100%

Percent to use

80% 60% 50% 40% 20% 0%

800

1000

1200

1400 1600 Year of composition

Model type First derivative Probability multiplication

1800

2000

Token number Less than 50 50 or more

Figure .. Best fits from the first derivative and probability multiplication models for the recipient–theme data

Table .. AIC values for the different models of the recipient–theme failed change

First derivative model Probability multiplication model

AIC

df

1182.63 1165.94

3 4

This rise–fall pattern, however, shows neither of the characteristics predicted by Postma’s first derivative model. The peak of the change is around 1350, which is well after the middle of the theme–recipient change (c.1200). Also, the change in the recipient–theme context lacks symmetry around its peak, that is, the change rises more quickly than it falls. This can be observed in the probability multiplication model line in Figure 7.3. AIC values reflect how well the model fits after taking into account model complexity with smaller numbers indicating better fit to the data. Table 7.2 shows that when comparing the first derivative model and probability multiplication model for just the recipient–theme data, the probability multiplication model is superior. In the next section, I discuss how the probability multiplication model can be interpreted as a blocking model of a failed change, while providing the ability to test the Constant Rate Hypothesis between the theme–recipient and recipient–theme context.

. A blocking model of failed change and probability multiplication As discussed in Section 7.2, the logistic model relates a linear model in log-odds space to the S-shaped curve in probability space (i.e. the range between 0 and 1). Thus, for

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

Modelling interactions between morphosyntactic changes



Table .. Example outputs of the three grammars in competition in the Middle and Modern English period

Null dative grammar All to grammar Split to grammar

Recipient–theme

Theme–recipient

I gave John the books I gave to John the books I gave John the books

I gave the books John I gave the books to John I gave the books to John

any year, the logistic model gives the probability of a particular grammar being used. In addition to the first derivative model, Postma (2010) suggests that the multiplication of two simple changes could produce a failed change. This multiplicative model has one change that reflects the counterpart of the successful change, which under the Constant Rate Hypothesis must have the same Slope as the successful change, but can have a different Intercept. The other change has a negative slope, and bleeds the application of the first change. In the case of English to, this is the probability of a recipient phrase being marked with to. The three relevant grammars can be seen in Table 7.3. The grammar at the beginning of the Middle English period had only a null realization of dative case. The incoming grammar was one in which dative case was always realized as to. This new grammar was the same as the one found with full noun phrases in the western Romance languages (e.g. French, Spanish), where all recipient noun phrases are marked with à ‘to’. Around 1400, a third grammar entered the language community. This grammar was characterized by having a split in the realization of dative case between the theme– recipient and recipient–theme contexts. In this grammar, the theme–recipient context has dative case realized as to, while the recipient–theme context has dative case realized as ∅. This split grammar bled the application of the all-to grammar in the recipient–theme context. In the theme–recipient context, the development of the all-to grammar followed an S-shaped curve moving from 0 around 1000 to 1 around 1400. Around the same time that the change goes to completion in the theme–recipient context, the recipient– theme context reaches its peak and begins to decrease. I propose that when the theme–recipient change went to completion, native language learners reinterpreted the remaining variability in the recipient–theme context as part of a new grammar, instead of the product of grammar competition intended by the older speakers. The dative to analysis for English recipients combined with the Constant Rate Hypothesis predicts that there should be a constant slope parameter between the theme–recipient context and the rising component of the recipient–theme context, since they both reflect a single change in the realization of dative case (from nullmarking to to-marking). Table 7.4 shows that the best model is the one consistent with the Constant Rate Hypothesis. The data is, however, not compatible with any further simplifications of the model towards the first derivative model (i.e. forcing the

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i



Hezekiah Akiva Bacovcin

Table .. Likelihood Ratio and AIC comparisons between different parameterizations of the probability multiplication model

Everything constant Failed Intercepts constant All slopes constant Constant rate effect Everything differs

Tot Df

Deviance

Chisq

Df

Pr(> Chisq)

AIC

2 3 4 5 6

2183.75 1424.85 1358.84 1291.94 1291.74

758.89 66.01 66.90 0.21

1 1 1 1

0.00 0.00 0.00 0.65

2187 1430 1366 1301 1303

% simulation p-values higher than empirical p-value

100%

75%

50%

25%

0%

0

1 2 3 4 Absolute value difference between theme–recipient and recipient–theme slopes

Figure .. Simulation results showing the statistical power in the English ditransitive data

decreasing component of the recipient–theme change to be the negative counterpart of the rising component and forcing the Intercepts of both parts of the recipient–theme change to be the same as the theme–recipient Intercept). In order to be confident that there was enough statistical power in the data to reliably test the Constant Rate Hypothesis, I ran a set of simulations. Each simulation generated artificial data with the same distribution across time as the actual data (i.e. in each simulation every real token has a simulation counterpart). The data was generated using the probability multiplication model with the fitted parameters derived

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

Modelling interactions between morphosyntactic changes



from the actual data. The estimated slope parameter for the theme–recipient change and the rising component of the recipient–theme change was about 5.6. I ran 100 simulations with the slope for the rising component of the recipient–theme change ranging from 1.6 to 9.6 (in steps of .1), while keeping all the other parameters constant (including the slope of the theme–recipient context). This means that the slopes of the theme–recipient and recipient–theme context could differ by as much as 4. I then fit both the full model and the Constant Rate Effect model to each set of simulated data and extracted the p-values. Figure 7.4 shows for each possible absolute value difference in slope values between the theme–recipient and rising recipient–theme context what percentage of the p-values were at least as high as the empirical p-value. As the graph shows, the data only provides enough power to distinguish between slopes whose difference is greater than 1.2.

. Conclusions The probability multiplication logistic model for failed changes builds a complex failed change out of the interaction of two micro-changes. The compositional nature of the probability multiplication model for failed changes allows their components to be directly compared to successful changes. By using simulations to calculate statistical power, it is possible to determine that the slopes of two changes are equivalent within a certain precision. Assuming the Constant Rate Hypothesis, this equivalence can be used as evidence that two surface changes arise from one underlying change. An example of this type of inference can be seen in the shared rise in the use of to in both theme–recipient and recipient–theme recipient ditransitives in the history of English. English before 1400 was developing a Western Romance-style grammar, where all recipients are marked with a prepositional case marker. After 1400, a grammar with dative shift entered the language, which bled the use of to in the recipient–theme context (I gave (to) John the books).

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

 From Latin to Modern French: A punctuated shift M I C H E L L E T R O B E R G A N D H E AT H E R BU R N E T T

. Introduction The goal of this chapter is to illustrate a surprising intermediate stage during the course of the development from the satellite-framed behaviour of Latin to the verb-framed behaviour of Modern French, and to suggest why such a development occurred. Contrary to prevailing assumptions, we show that this typological shift does not simply involve a gradual loss of satellite-framed (henceforth s-framed) properties accompanied by a gradual increase of verb-framed (v-framed) alternatives. Rather, Medieval French presents a remarkable intermediate stage, permitting a cluster of s-framed constructions found neither in Early Latin nor in Modern French and thus requiring a re-examination of the shift. We consider this change from the point of view of micro-changes occurring at the level of lexical items, and the principal claim in our account involves a progression of reanalyses within the prepositional system. The first change involves the Path prefix system, where telicizing preverbs in Latin, syntactically derived from locative prepositions, had been reanalysed as Path heads by Medieval French. These elements were then further reanalysed as an inseparable part of the verb to produce the Modern French system. The second and more surprising development, however, involves a cluster of non-Latin s-framed constructions involving resultative secondary predication: bare goal-of-motion, verb–particle, and complex adjectival resultative constructions. We suggest that the rise and fall of these innovations is linked to the loss of the Latin telicizing preverbs. The chapter is structured as follows: Section 8.2 outlines the s-framed properties of Early and Classical Latin, and Section 8.3 follows by showing how its daughter language, Modern French, is an archetypal v-framed language. Section 8.4 then describes aspects of the grammar of Medieval French, arising midway between these two typological poles and presenting strikingly new s-framed behaviour. Section 8.5 outlines our account of the reanalysis, and Section 8.6 finishes with some remarks on the theoretical implications of the study.

Micro-change and Macro-change in Diachronic Syntax. First edition. Éric Mathieu and Robert Truswell (eds). This chapter © Michelle Troberg and Heather Burnett 2017. First published 2017 by Oxford University Press.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

From Latin to Modern French: A punctuated shift



. Early Latin as a satellite-framed language Following Talmy’s (1975, 1985, 2000) influential typology regarding motion events, languages are characterized in terms of how the core schema, or Path, is lexicalized. In s-framed languages, transition is lexicalized in a so-called satellite to the verb (such as particles, verbal affixes, and prepositional phrases), while in verb-framed languages Path is not so lexicalized, but is rather encoded in the meaning of the main verb itself. Typologically, s-framed languages tend to present a number of syntactic possibilities involving main verbs that express manner combined with a resultative secondary predicate. In Early and Classical Latin (200 bc to 200 ce), prepositional elements could appear as prefixes on manner-of-motion verbs to give a goal-of-motion reading; see Haverling (2003) and Acedo Matellán (2010). For example, in (1) the verbal prefixes encode the result state (the goal) of the theme: the flies come to be at the goat’s udders through flying in (1a), the riders come to be down through jumping in (1b), and the theme comes to be at the gates by riding in (1c). The verbal prefixes in (1) appear to be prepositions that have incorporated into the verb.1 Similarly, the prefixes in (2) encode the result state of various DPs while the verb encodes manner: the shells come to be out through coughing in (2a), and the blood comes to be away through wiping in (2b). The prefixes in (2) could be considered an incorporated particle (or intransitive preposition).2 (1)

a. Caprarum-que uberibus ad-volant goat.gen.pl-and udders-dat.pl at-fly ‘And they fly onto the udders of the goats.’ (Plin. Nat. 10, 115; in Acedo Matellán 2010: 100) b. Repente ex equis de-siliunt. suddenly out horses.abl down-leap.3pl ‘Suddenly they leapt down from their horses.’ (Liv. 22, 48, 2; in Acedo Matellán 2010: 189) c. Qui ubi ad-equitavit portis. who.nom.sg as.soon.as at-ride.pfv.3sg doors.dat ‘This one, as soon as he had ridden to the gates’ (Liv. 22, 42, 5; in Acedo Matellán 2010: 189)

(2) a. Serpentes ova solida hauriunt, [. . . ] atque snake.nom.pl egg.acc.pl whole.acc.pl swallow.3pl, and putamina ex-tussiunt. shell.acc.pl out-cough.3pl ‘Snakes swallow the eggs whole and expel the shells through coughing.’ (Plin. Nat. 10, 197; in Acedo Matellán 2010: 179) 1 A derivational relationship between the prefix and the preposition would assume that the prefix assigns case to the Ground argument. Dative DPs in contexts such as (a) and (c) therefore pose a problem for such an analysis; see Acedo Matellán (), who presents arguments favouring an analysis whereby the prefix governs in some way the dative DP. 2 How the Figure argument receives case is not always clear. In (a) it appears to be the verb that gives accusative case to putamina while in (b) it looks to be the prefix abs- that gives ablative case to cruore. In this article, we leave aside the specific questions of Latin case assignment.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i



Michelle Troberg and Heather Burnett b. Inspectum vulnus abs-terso cruore. examine.ptcp wound.nom.sg away-wipe.ptcp blood.abl.sg ‘That the wound had been examined after wiping the blood off ’ (Liv. 1, 41, 5; in Acedo Matellán 2010: 97)

Such prefixes could not only have directional meaning, as shown above, but also a related aspectual meaning. In Early Latin, they were also used to create hundreds of verb pairs that correspond to atelic/telic pairs. A few examples of this alternation are given in (3), taken from Haverling (2003: 114). (3) a. b. c.

edo ‘eat/eat of ’ bibo ‘drink/drink of ’ tacui ‘have been silent’

vs. vs. vs.

comedo ‘eat up’ ebibo ‘drink up’ conticui ‘have fallen silent’

Interestingly, the complex of prefixes introduced in (1) to (3) are the unique source of resultative secondary predication with manner verbs in Early Latin. Unlike English, for example, Early Latin does not present other common s-framed constructions (Acedo Matellán 2010); it has no morphologically free directional/aspectual verb particles (to eat something up; to walk away) and no adjectival resultative constructions (to hammer something flat).3 In his dissertation, Acedo Matellán (2010) argues that the prefixes are not in fact inherently directional or aspectual, but that instead they have fundamentally locative semantics, noting that when they are used with stative verbs, they have locative interpretations, shown in in (4). Further, these forms also occur independently as locative prepositions, shown in (5). (4) a. Argentum de-erat. silver.nom away-was.ipfv ‘Money was lacking.’ b. Senex ab-est. old.man.nom away-is ‘The old man is missing.’

(Ter. Phorm. 298; in Acedo Matellán 2010: 98)

(Plaut. Cas. 882; in Acedo Matellán 2010: 98)

(5) a. quorum saepe et diu ad pedes iacuit stratus whose often and long.time at foot.acc.pl lay spread.out ‘at whose feet he often lay at that for a long time’ (Cic. Quinct. 96; in Luraghi 2010a: 25) b. quia ab tergo errant clivi, because from back.abl were hills ‘because behind them were hills’ (Liv. 2,65,2; in Luraghi 2010a: 26) In sum, earlier Latin possessed a single kind of resultative secondary predication construction: that involving verbal prefixes. One way to describe the system of prefixes 3 Acedo Matellán and Mateu () present some evidence of resultative secondary predication formed from non-prefixed verbs in (later) Classical Latin. Such cases reveal that the intermediate grammar we describe for Medieval French had developed before the birth of Romance, further supported by evidence that similar kinds of resultative secondary predication found in Medieval French are also attested in Old Catalan and Old Italian; see Acedo Matellán and Mateu (); Iacobini (); and Iacobini and Masini () and references therein.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

From Latin to Modern French: A punctuated shift



Table .. Locative prepositions and directional/ aspectual prefixes in Latin Locative preposition

Path prefix

de ex cum in ad ab ...

deexco(n)inadab...

‘from’, ‘of ’ ‘out’ ‘with’ ‘in’, ‘on’ ‘at’ ‘at’

‘down’ ‘out from’ ‘become’ ‘into’ ‘to’ ‘from’

would be to say that their directional/aspectual meaning arises through a derivational relationship with their locative prepositional counterparts. This relationship is summarized in Table 8.1; in Section 8.5, we describe this relationship in terms of head movement.4 , 5

. Modern French as a verb-framed language While in s-framed languages Path is lexicalized in satellites associated with the verb, Romance languages like Modern French are considered to be verb-framed because transition or result is lexicalized in inherently directional verbs. In some cases, telic Path readings can arise through pragmatic inference as well, but what is clear is that satellites do not productively contribute to such interpretations. For example, goal-ofmotion readings are not possible with manner-of-motion verbs in Modern French.6 Recall the examples in (1) for Latin where goal-of-motion readings arise from the 4 Early and Classical Latin present case variation according to the locative or directional sense of some prepositions (namely in, sub, super). Iacobini () suggests that the loss of such case distinctions may have contributed to the loss of satellite-framed behaviour in Latin. 5 An anonymous reviewer points out that the preposition cum and the aspectual prefix co(n) are likely not etymologically related, as discussed in Rosen (), but that they may later have been reanalysed as derivationally related. 6 In some cases, a bounded Path interpretation can indeed arise with a restricted set of manner verbs denoting movement including courir ‘to run’ and sauter ‘to jump’. Beavers et al. () suggest that possible directional interpretations such as those in (i) and (ii) are not grammatically constructed but are rather products of pragmatic reasoning associated with the inherent aspectual properties of the verb and with the properties of the Ground.

(i) Jean a couru au magasin. Jean has run to/at.the store ‘John ran to the store/John ran while at the store.’ (ii) Jean a sauté dans la flaque. John has jumped in/into the puddle ‘John jumped into the puddle/John jumped while in the puddle.’ Comparable verbs (without prefixes) in Latin can also be used in goal-of-motion constructions. Acedo Matellán () accounts for them in a similar way.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i



Michelle Troberg and Heather Burnett

presence of bounded goal Path prefixes on manner verbs denoting movement such as vol¯are ‘to fly’, sal¯ıre ‘to jump’, and equit¯are ‘to ride’. No such productive system of prefixes is available in Modern French, and prepositional phrases cannot bring about the necessary telic readings either, as shown in the translations below.7 (6) a. *Les mouches survolent aux pis des chèvres. the flies on.fly to.the udders of.the goats *Les mouches volent sur les pis des chèvres. the flies fly on the udders of.the goats For: ‘The flies fly onto the udders of the goats.’ b. *Soudain ils ont desauté de leurs chevaux. suddenly they have down.leapt from their horses For: ‘Suddenly, they leaped down from their horses.’ c. *Il avait achevauché aux portes. he had to.ridden to.the gates *Il avait chevauché aux portes. he had ridden to.the gates For: ‘He had ridden to the gates.’ As is typical of v-framed languages, the Path component of these events would be expressed via verbs whose meaning is inherently directional or telic, while the manner component is expressed either implicitly or explicitly. In (7a), the reflexive use of the verb poser ‘to land’ encodes the result state of the subject, while the manner is implied through the verb and the subject. In (7b), the manner is encoded in the verb sauter ‘to jump’ and the direction and result are implied. The telic directional verb arriver ‘to arrive’ encodes the result state of the subject in (7c), while manner is expressed either implicitly or by mention of the horse. (7) a. Les mouches se posent sur les pis des chèvres. the flies refl land on the udders of.the goats ‘The flies fly onto the udders of the goats.’ b. Soudain, ils ont sauté de leur chevaux. suddenly they aux jumped from their horses ‘They suddenly jumped down from their horses.’ c. Il était arrivé aux portes (à cheval). he aux arrived at.the gates (on horseback) ‘He rode to the gates.’ Latin-style resultative prefixation is thus not available in Modern Standard French, nor are any other Path satellites that would allow the language to express result in a goal-of-motion construction or other typical s-framed constructions.

7 Only two prefixes (iterative re- and change of state de-) are productive in Modern French; see Sablayrolles () and Kopecka ().

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

From Latin to Modern French: A punctuated shift



. From satellite-framed to verb-framed: The case of Medieval French (– ce) It has been widely accepted that as Latin evolved into French, it passed gradually from a satellite-framed language to a verb-framed language (Acedo Matellán 2010; Acedo Matellán and Mateu 2013; Talmy 2000) and involved various stages of hybridization; see Hickmann and Robert (2006); Iacobini and Fagard (2011: 158); Iacobini (2012); Kopecka (in press: 4); Slobin (2004). We demonstrate below that the typological shift from Latin to Modern French is anything but a gradual decline in the degree of s-framed behaviour.8 On the contrary, this shift is punctuated by the birth of a new s-framed system, significantly different from what we understand Early Latin to be, presenting robust occurrences of bare goal-of-motion, verb–particle, and adjectival resultative constructions. Moreover, the loss of this new s-framed system in French appears to be rather abrupt, not gradual as previous accounts claim. In the next sections, we detail the relevant facts from Medieval French. .. Changes in the Latin prefixes By Late Latin (200–600 ce), the robustness of the directional/aspectual prefix system was beginning to erode; prefixes were coming to be reanalysed as part of verb roots. Haverling (2003) demonstrates this process by showing that the telicity alternation in Classical Latin between edo ‘eat’ and comedo ‘eat up’ is lost by Late Latin, where comedo is used in atelic contexts. Likewise, she shows that the Classical Latin alternation tacui ‘have been silent’ vs. conticui ‘have fallen silent’ is less clear in Late Latin; authors confuse the forms, so that conticui is used in atelic contexts and tacui can mean ‘stopped talking’. Path prefixes are still widely used in Medieval French, but their spatial meaning has faded dramatically, leaving them with the unique function of modifying the aspectual value of the event. In (8b), for example, the prefix a- shifts the activity verb penser ‘to think’ toward an accomplishment reading, ‘to realize’. This prefix can also create the activity/accomplishment alternation with manner-of-motion verbs producing goalof-motion constructions similar to those we find in Latin (see examples (1a)–(1c)), illustrated in (9), where voler has an activity reading and advoler has the accomplishment goal-of-motion reading. Martin (2001) discusses hundreds of verbs that occur with the prefix a(d)- in Medieval French, focusing on the distribution, interpretation, and productivity of the prefix.9 , 10 (8) a. car il pensoit bien que aucuns de l’ostel le roi le for he thought well that someone from the-house the king him sivroit. followed ‘For he suspected that someone from the king’s residence was following him.’ (Artu, 13th c., 11; Dufresne et al. 2001) 8

See also Burnett and Troberg (). We draw on two electronic corpora: the TFA (La Base Textes du Français Ancien, ARTFL) and the DMF (Dictionnaire du Moyen Français, ATILF). These facts are also discussed in Troberg (). 10 The prefix a- can also contribute an inchoative flavour to the event (reminiscent of ad- in Latin). 9

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i



Michelle Troberg and Heather Burnett b. Mais quant il ouy la freinte, il appensa tantost que but when he heard the noise, he a.think immediately that Glaudes retournoit, Glaudes return.ipfv ‘But when he heard the noise, he immediately realized that Glaudes was coming back.’ (Arras, 1392–1393, 20; DMF)

(9) a. Et il vola si hautement and he flew so high.adv ‘and it flew so high’

(Mach, D. Aler, a.1349, 356; DMF)

b. l’innumerable nombre de langoustes qui ad-volerent en France the-innumerable number of locusts who to-flew in France ‘the uncountable number of locusts that flew into France’ (Simon de Phares, Astrol, c.1494–1498, fo 107 vo ; DMF) Other productive directional/aspectual prefixes in Medieval French are listed in (10); see Tremblay et al. (2003: 555) and Kopecka (2009) for further discussion.11 (10) parler ‘to speak’ tendre ‘to stretch’ amer ‘to love’ querre ‘to ask, seek’ aler ‘to go’ geter ‘to throw’ nagier ‘to swim’

de-parler ‘to speak ill about something’ e-tendre ‘to stretch out’ en-amer ‘to fall in love’ sor-querre ‘to insist’ por-aler ‘to go all around’ par-geter ‘to throw all the way’ tres-nagier ‘to swim/row across’

The crucial difference between the Latin Path prefixes and the Medieval French reflexes is that the clear derivational relationship with a series of locative prepositional counterparts no longer exists. Latin Path prefixes were essentially identical in form and in meaning to the prepositions save that they carried an additional layer of meaning denoting the attainment of a goal. Table 8.2 summarizes what remains of Table .. Medieval French aspectual prefixes and their prepositional counterparts Locative preposition

Aspectual prefix: generalized meaning of transition/result

a ‘at’ en ‘in’ par ‘through’ — — —

aenpartrespore(s)...

11 Of the list in (), deparler, enamer, sorquerre, poraler, and tresnagier are new formations, not lexical items inherited from Latin.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

From Latin to Modern French: A punctuated shift



the prefixes in Medieval French. Note that only three have a prepositional counterpart and that none of the prefixes have a clear spatial meaning corresponding to an analogous preposition. The development of the preposition–prefix relationship is not surprising; it heralds the well-known reanalysis of the prefixes as an indivisible part of the verbal root. What is rather surprising, however, are the syntactic innovations that accompany it, outlined in the next section. .. Satellite-framed innovations ... Bare goal-of-motion (without prefixes) Similar to other s-framed languages (like Dutch, for example), Medieval French can express both activities and goal-ofmotion using the same verb, presenting unergative syntax for the former and unaccusative syntax for the latter, shown in (11) and (12) for voler ‘to fly’ and courir ‘to run’. Note the auxiliary alternation, where the auxiliary être ((11a), (12a)) is associated with unaccusative syntax and avoir with unergative ((11b), (12b)). (11)

a. Les aeles de vertus avoit [. . . ]. Donc Marie est volee en haut, En the wings of virtue had so Marie aux flew in high in la region ou est chaut the region where is hot ‘She had wings of virtue [. . . ]. So Mary flew up into the region where it is hot.’ (Bestiaire marial, c.1333, 181; TFA) b. Et quant il avoit tant volé que toz li monz le tenoit a and when he aux much flown that all the world him held at merveille wonder ‘And once he had flown around enough so that everyone marvelled at him.’ (Queste del Saint Graal, 1225, 131; TFA)

sont coru por lui rescorre. (12) a. Mais tot li chevalier ensamble i but all the knights together there aux run for him rescue ‘But together the knights quickly ran there in order to rescue him.’ (Vengeance Raguidel, 1200, 33; TFA) b. Tant a coru et porchacié, so.much aux run and pursued ‘So much did he run and chase’ (Saint-Cloud, Roman de Renart Branche 7, c.1175, 5835; TFA) Other manner-of-motion verbs that can occur with a goal-of-motion interpretation in our corpus include marcher ‘to walk/march’, cheminer ‘to make one’s way’, chevaucher ‘to ride on horseback’, ramper ‘to climb’, flouer ‘to flow’, and trotter ‘to trot’.12

12 A subset of such verbs in later Classical Latin can also be used in goal-of-motion constructions without a Path-denoting prefix (Acedo Matellán  cites curro ‘run’ and navigo ‘sail’), but this set appears to have broadened considerably by Medieval French.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i



Michelle Troberg and Heather Burnett

The innovation in the expression of goal-of-motion is that Medieval French does not require the presence of a Path prefix, which was generally the case in Latin, exemplified in (1a)–(1c). Where is Path therefore expressed? We contend that in constructions such as those in (11a) and (12a), Path is expressed in the preposition. Indeed many prepositions alternate between a locative and a directional interpretation in Medieval French; compare the directional interpretations of sur ‘onto’, en ‘into’, and aux ‘to the’ given in (13) with their locative counterparts in (14): sur ‘on/over’, en ‘in’, and au ‘at/in the’. (13) a. il vole sur les rainceaulx ou sur les branches. he flies onto the branches or onto the branches ‘He flies onto small tree limbs or branches.’ (Le Menagier de Paris, c.1392–1394, 163; DMF) b. Et puys après nous troterons en guerre. and then after we will.trot in war ‘And then after we will trot into war.’ (de La Vigne, La Ressource de la Chrestienté, 1494, 133; DMF) c. en passant par la chambre et cheminant aux nopces in passing by the room and making.his.way to.the wedding ‘while passing by the bedroom and making his way to the wedding’ (Cent Nouvelles Nouvelles, c.1456–1467, 122; DMF) (14) a. si aucun oysel vole sur icelui endroit, incontinent chet mort à if certain bird flies on this place immediately falls dead to terre. earth ‘If any bird flies over this place, it immediately falls dead to the ground.’ (Simon de Phares, Astrologues, c.1494–1498, 87; DMF) b. et cevauçans en France nuit et jour and riding in France night and day ‘and riding in France day and night’ (Froissart, Chroniques, c.1400, 569; DMF) c. Ne qu’ on puet au firmament Sans eles voler nor that one can at.the firmament without wings fly ‘Nor can one fly in the heavens without wings.’ (Machaut, Les Lays, 1377, 388; DMF) Table 8.3 contains some examples of prepositions that illustrate the new Place–Path meaning alternation attested in Medieval French. In sum, while telic Path with manner verbs could only be expressed in verbal prefixes in Latin, the data suggest that it can be expressed both in aspectual prefixes (see Table 8.2) and in directional prepositions in Medieval French. ... Verb–particle constructions Along with goal-of-motion constructions such as those discussed in Section 8.4.2.1, the presence of verb particles that express or modify a Path expression are also typical of s-framed languages. Particle-like elements

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

From Latin to Modern French: A punctuated shift



Table .. Place–Path meaning alternation in prepositions in Medieval French

a en sur hors par dedans

Place interpretation

Path interpretation

‘at’ ‘in’ ‘on’ ‘outside’ ‘at’ ‘inside’

‘to’ ‘into’ ‘onto’ ‘to outside’ ‘through’ ‘to inside’

already appear in Classical Latin, but Iacobini (2009) claims that they are marginal and stylistically marked. The following examples are from Iacobini (2009: 37). (15) a. retro regredi ‘to march back; to retreat/return back’

(Bellum Africum 50,2)

b. retro . . . reverti ‘to return/revert back’

(Lucr. 1,785)

c. foras . . . exire ‘to go out’

(Lucr. 3,772)

Early attestations show these elements occurring mainly with prefixed verbs, likely arising to reinforce the directional Path meaning that is decreasingly salient in certain high-frequency verbs. Iacobini (2009) provides some further examples from the same period where these elements are used to express the Path of a non-prefixed generic motion verb, and gives examples from Late Latin, in (16), where their use is more common. It is unclear from the examples, however, that these elements—perhaps still adverbs—create a telic event. What is clear, however, is that they are the precursors of the telicizing Path particles attested in Medieval French. (16) a. Cecidit de tertio cenaculo deorsum ‘and fell down from the third loft’ b. Noli foras ire, in te ipsum redi ‘Do not wish to go out, return into yourself.’

(Act. apost. 20, 9) (Aug., De vera relig. 39)

By Medieval French, a robust system of verb particles exists. These particles are discussed extensively in Burnett and Tremblay (2009); Marchello-Nizia (2002); and Buridant (2000). These elements function similarly to the Late Latin examples above, elaborating Path. The path can be implicit, as in (17a), where sus specifies its direction, and the path can be encoded in the verb (or prefix), as in (17b,c), where hors and jus reinforce the direction of movement. (17) a. il sailly sus le plus tost qu’il peut he leaps up the more fast that-he can ‘He gets up as fast as he can.’ (Percef. III, R., t.2, c.1450 [c.1340], 233; DMF)

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i



Michelle Troberg and Heather Burnett b. A ces criz et a ces noises issi hors Messires At these cries and at these noises exited out Messire-nom Gauvains de son ostel Gauvain-nom from his lodging ‘When he heard the cries and the noise, Sir Gawain went out of his house.’ (Artu, 13th c., p.130; Burnett and Tremblay 2009) c. et le reversa jus a terre. And him re.spill down to ground ‘And he dumped him down on the ground.’ (Froissart, Chron. D., c.1400, 387; DMF)

While the verb particles shown above express spatial notions, they could also have strictly aspectual interpretations, shown in (18). (18) a. arriere les voit consillier again them see converse ‘He sees them conversing again.’ (Bel inconnu, l.2770; Burnett and Tremblay 2009) b. il s’ entrecommencent a regarder et semont li uns they refl between.begin to look and ask the one l’autre de parler avant the-other to talk forward ‘They begin to look at each other and ask each other to start talking.’ (Artu, 13th c., p.13; Burnett and Tremblay 2009) c. boire fors; manger fors; paiier fors drink out eat out pay out ‘to drink up; to eat up; to pay up’ (Buridant 2000: 544) ... Adjectival resultative secondary predication Finally, Medieval French presents complex adjectival resultative secondary predicate constructions, which are often found in s-framed languages. The Medieval constructions involve transitive VPs combined with an adjective that describes the state of the direct object at the end of the event, as shown in the examples in (19). Interestingly, this kind of secondary predication does not occur in Latin, as demonstrated in Acedo Matellán (2010). The following examples are taken from Troberg and Burnett (2014), where this construction is discussed in detail. (19) a. que mort ne l’acraventet. that dead not him-crush ‘so that it didn’t crush him dead’ (Roland, c.1100, 285.3930; MCVF) b. Que tricherie abat jus plate that deception beats down flat ‘that deception beats down flat’ (Pizan, Fortune, t.2, 1400–1403, 29; DMF) c. Une colder trencha par mi, tute quarree la fendi. a hazel.tree sliced through middle all square it cut. ‘He sliced a hazel tree through the middle, he cut it right square.’ (Marie de France, Lais, 1154–89, p.183; TFA)

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

From Latin to Modern French: A punctuated shift



It is less obvious how these adjectival constructions are formally related to the resultative secondary predicates built from prepositional elements, discussed above. We presume that there is in fact a formal connection based on what appears to be a simultaneous loss of both adjectival resultatives and verb particles in the sixteenth century and supported by the typological generalization. However, in this chapter, we focus on the relationships involving prepositional elements. A unified account has been left for future work. .. Discussion Early Latin allowed only one type of construction that is associated with s-framed languages: a goal-of-motion construction where the Path satellite appears as a telicizing verbal prefix. In Latin, these prefixes can be considered to be derivationally related to locative prepositions. As Latin evolves into Medieval French the form–meaning relationship between the prefixes and the prepositions diminishes. At the same time, however, there is robust evidence that prepositions themselves can encode Path meanings, giving rise to bare goal-of-motion interpretations (without directional preverbs) and correlating with (at least) two new syntactic possibilities associated with s-framed languages: verb–particle constructions and adjectival resultative constructions. This burst of innovation, however, does not appear to extend any later than the sixteenth century; Modern French prepositions are predominantly locative, and speakers do not productively admit any of the s-framed resultative secondary predication constructions that we find in Medieval French.13 (20) a. Path prefixes; compare with (9b) *Ils a-volèrent en France. they to-flew in France b. Verb particles; compare with (17b) *Seigneur Gauvain est allé hors de son logement. Sir Gawain aux gone out of his lodging c. Goal-of-motion; compare with (13a) *L’oiseau vole sur les branches. the-bird flies on the branches d. Adjectival resultative secondary predication; compare with (19b) *battre le metal plat to.beat the metal flat The Medieval French facts illustrate that the shift from the Latin s-framed system to the Modern French v-framed system is not at all gradual. The development of the Modern French system involved three distinct systems, summarized in Table 8.4. Each successive system is not simply a lesser degree of s-framedness; rather, the restricted s-framed Latin develops into the less restricted s-framed grammar of Medieval French, which then becomes the v-framed grammar of Modern French. Certainly the tendency to express Path in the verbal root increased gradually over many centuries and involved low-level lexical changes in which Path prefixes came

13 Save vers ‘toward’ and par ‘through’, which have an unbounded Path denotation. Since these prepositions cannot be used to build resultative secondary predicates, we do not discuss them further.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i



Michelle Troberg and Heather Burnett

Table .. Evolution of resultative secondary predication from Latin to Modern French

Directional/aspectual prefixes Goal-of-motion construction Directional/aspectual verb particles Complex adjectival resultatives

(Early) Latin s-framed

Medieval French Modern French v-framed s-framed

  (prefix required) × ×

   

× × × ×

to be reanalysed as part of the root itself.14 In contrast, the disappearance of the new s-framed grammar attested in Medieval French appears to be strikingly abrupt. For example, in order to measure the rate at which verb particles disappeared, we tracked the loss of two directional particles, arrière ‘back’ and avant ‘forward’, as they were replaced by PPs introduced by en. Examples (21) and (22) show arrière and avant as particles in Medieval French, reinforcing the path denotation already present in the verb. In Modern French, these elements are nominal and must occur within the PP en arrière and en avant in order to perform the same function ((21b), (22b)).15 (21) arrière vs. en arrière a. le mers reportoit le nef ariere the sea re.take the ship back ‘The sea pushed the ship back.’ (Clari, early 13th c., p.74; in Dufresne et al. 2003) b. La mer a rapporté le navire en arrière. (Modern French) (22) avant vs. en avant a. Lors saut avant Girflez et dist a la reïne so jump forward Girflet and says to the queen ‘So Girflet jumps forward and says to the queen’ (Artu, 13th c., p. 319; Burnett and Tremblay 2009) b. Alors Girflet saute en avant et il dit à la reine (Modern French) We tracked the frequency of avant, en avant, arrière, and en arrière in directional contexts from the beginning of the fourteenth century to the end of the seventeenth century. Although the particles are productive and used robustly through the fourteenth century, both arrière and avant begin to decline in frequency in the fifteenth century, 14 Dufresne et al. () systematically study the evolution of the preverb a-, by far the most productive and the longest lasting, showing a steady decline from the th c. to the th c. in new verbs formed by the addition of a-. 15 Data from the Old French period (–th centuries), Middle French period (–th centuries), and th and th centuries come from the Textes de français ancien (TFA) database, the corpus associated with the Dictionnaire du moyen français (DMF), and the Frantext database respectively.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

From Latin to Modern French: A punctuated shift



Table .. Replacement of directional avant by directional uses of en avant (by date of author’s birth) 1300–49 1350–99 1400–49 1450–99 1500–49 1550–99 1600–49 1650–99 avant en avant  of particles

165 9 94

85 1 98

98 21 82

25 12 67

6 32 15

6 72 8

0 25 0

0 37 0

Table .. Replacement of directional arrière by directional uses of en arrière (by date of author’s birth) 1300–49 1350–99 1400–49 1450–99 1500–49 1550–99 1600–49 1650–99 arrière en arrière  of particles

126 2 98

63 2 96

43 3 93

39 14 73

21 22 49

13 61 17

7 73 9

0 0 —

and within 200 years, the particles are obsolete.16 Our results are shown in Tables 8.5 and 8.6 and summarized in Figure 8.1. The time-course of the loss of verb–particle combinations is thus not a slow and gradual decline. Rather, it bears similarities to other well-known syntactic changes. A quantitative study of the loss of bare goal-of-motion constructions and adjectival resultatives would allow us to assess a possible correlation between their disappearance and that of verb particles, a correlation we would expect to find if the three types of resultative secondary predicates are indeed structurally related. Here, however, we have hit an impasse; measuring the loss of these constructions has proved challenging due in part to the low number of tokens. Nevertheless, it is worthwhile noting that the last adjectival resultative construction identified in our corpus was written by François Villon, born in the first half of the fifteenth century. This may be evidence that this construction disappeared at approximately the same time as the particles. Finally, our results strongly suggest that neither normative pressures, implied in Iacobini (2009), nor functional reasons (Foulet 1958) lead to the disappearance of Path particles in Medieval French. The uniformity and abruptness of the loss of arrière and avant are also incompatible with proposals that s-framed syntax gradually falls out of use due to semantic and pragmatic aspects of the language favouring v-framed strategies in the expression of motion events (Pourcel and Kopecka 2005; Slobin 2004). In line with the idea that the typological shift has been gradual, it has also been 16 Our results for the loss of the particles avant and arriere confer with Marchello-Nizia’s () claim that the particle aval falls out of use in the th century.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i



Michelle Troberg and Heather Burnett 100

% particles

80 60 40 20 0 1300

1350

1400

1450 1500 Author date of birth avant

1550

1600

arrière

Figure .. Replacement of avant/arrière by en avant/arrière Note: While the loss of the two verb particles follows an S-shaped curve, more data is required to confirm the Constant Rate Hypothesis.

suggested that during the evolution of Latin to Modern French, so-called hybrid systems have arisen, where both s-framed and v-framed behaviour is present. Our position is that hybrid systems do not exist in the sense claimed within the functionalist tradition.17 If we understand v-framed grammars like Modern French to be those which do not have a nonverbal morpheme expressing transition or attainment of a goal and which instantiate the functional head Path, then v-framed and s-framed grammars are in a superset–subset relationship; mixed strategies for the expression of manner and path are thus expected in an s-framed grammar, which will give varying effects of a so-called ‘hybrid’ system. On the other hand, in a v-framed grammar, we predict no productive evidence of s-framed behaviour. The difference in the two grammars is thus categorical from this perspective, and this difference is born out in the history of French. French was likely a satellite-framed grammar until the fifteenth century, and by the seventeenth century, most speakers were only acquiring a verbframed system.

. Two reanalyses involving Path We claim that the typological difference between Early Latin and Modern French involves at least one abstract category belonging to the extended functional projection of the prepositional system, often called Path. It encodes transition/result and can add a layer of meaning to lexical items that otherwise encode location. Lexical items that instantiate this function give information about a trajectory and, in the case of those with a bounded goal denotation, create a resultative reading (see H. Koopman 17

Our position is, however, entirely compatible with work in the ‘competing grammars’ tradition.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

From Latin to Modern French: A punctuated shift



2000; den Dikken 2010; Folli and Ramchand 2005; Son 2007; Son and Svenonius 2008; among others). We propose that Path in Early Latin is lexicalized by a null prefix with a bounded goal denotation, which permits the rich stock of Latin locative prepositions to also function as Path-denoting prefixes. We assume that the null Path head triggers incorporation of the lower Place head, and then this complex, with the morphological specification of a prefix, incorporates into the verb.18 A structure such as that in (23) would derive the Path prefix from a locative preposition, accounting for the pairs in Table 8.1. To illustrate, we use the Latin sentence in (2b) as an example. The Ground, here realized as pro, is coindexed with vulnus ‘wound’. (23)

V Path

PathP V

Place

Path

abs

0/ -

ters-

DP cruore

PlaceP

Path

Place

Path

abs

0/ -

DP cruore

Place

DP

abs

proi

As the sound–meaning relationship is lost between the locative prepositions and their directional prefix counterparts, so is the derivational relationship. This loss produces a new system in Late Latin and Medieval French where the prefixes are reanalysed as dedicated Path morphemes, a classic case of grammaticalization, where internal merge is reanalysed as external merge, as described in Roberts and Roussou (2003). This process is shown in (24).

18 Although our assumptions about head movement differ, the present account is in essence very similar to that proposed in Acedo Matellán (). We adopt a theory of head movement along the lines of Baker’s () P-incorporation, which frames Latin verbal prefixation as the result of a property of a lexical item. Acedo Matellán, on the other hand, treats the prefixation as the presence of a broader PF operation involving lowering of v to Path.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i



Michelle Troberg and Heather Burnett

(24) a. Early Latin derived Path prefixes

Path Place ad de ex ...

→

Path adee(s)...

PlaceP

Path 0/-

b. Medieval French Path prefixes

Place adde ex...

PlaceP

In addition to the Path preverbs, the data suggest the presence of a null non-prefixal Path morpheme into which a locative preposition would incorporate, deriving the directional meaning of many simple prepositions in Medieval French (25a) and licensing verb particles (25b). (25) a. Medieval French prepositions with Path interpretation19

Path Place a sur en hors ...

PlaceP

Path 0/ Place a sur en hors ...

(DP)

19 This system should allow robust occurrences of goal-of-motion constructions with strictly mannerof-motion verbs as we see in English (i.e. wiggle, spin, etc.); however, we have only been able to identify goal-of-motion with manner verbs that themselves imply translative movement. A potential problem with this proposal, therefore, is that it is not constrained enough.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

From Latin to Modern French: A punctuated shift



b. Medieval French Path particle20 See (17c): et le reversa jus a terre ‘and knocked him down to the ground’

DirP PP jus

Dir

PathP DP Path

PlaceP

Place

Path

à

0/

DP Place

DP

à

terre

We tentatively suggest that the erosion of the Latin prefix system (concretely, the loss of incorporation) naturally leads to the development of a non-prefixal null Path (shown above). By Late Latin, the semantic bleaching of some prefixes produced variable input so that prefixed verbs occur in both telic and atelic contexts; recall the brief discussion in Section 8.4.1 of tacui/conticui and edo/comedo. Applied to manner-of-motion verbs, we contend that such variable input would lead the child learner to analyse the source of the telic interpretation as the preposition, not the preverb.21 For example, accourir ‘to run quickly to a destination’ occurs in telic goal-of-motion constructions as expected; two examples of this are given in (26). (26) a. et y fu tantos acourus, car il n’i a que and there aux immediately to.run for it neg-there has that quatre liewes. four leagues ‘And he immediately ran there, for it was only four leagues away.’ (Froissart, Chron. R., VIII, c.1375–1400, 108; DMF)

Following Svenonius (); although not shown here, à would conceivably raise to lexicalize Dir0 . Although our discussion and examples come from Medieval French, the reanalysis we suggest likely occurred much earlier. Already in Late Latin there is confusion between prefixed and non-prefixed forms as described in Haverling () and more extensively in Haverling (). Moreover, Iacobini () offers examples of prefixed manner-of-motion verbs from Late Latin that occur with particles that reinforce the meaning of the prefix; see (). 20 21

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i



Michelle Troberg and Heather Burnett b. environ l’eure de sept heures au matin, [. . . ] vindrent et around the-hour of seven hours at.the morning came and acoururent dedens les trenches de ladicte ville grant quantité desdiz to.ran inside the ditches of the.said city great quantity of.said Bourguignons Burgundians ‘At around the time of seven o’clock in the morning, [. . . ] a great number of Burgundians came and ran quickly into the trenches of the said city.’ (Roye, Chron. Scand., I, 1460–1483, 275; DMF)

In addition, however, accourir comes to be also used simply to express manner: ‘to run quickly, to hasten’. The Godefroy, for example, points out that the present participle, acorrant, used as a noun modifier simply means ‘fast’. (27) Et beles armes et acorranz destriers and beautiful weaponry and fast horses (Garin, ms. Dijon, fo 3d ; Dictionnaire de l’ancienne langue française) Similarly, the prefix a- does not contribute a telic meaning in example (28), in which the hero, Huon, looks back to see the Saracens pursuing him. The verb accourir here can only refer to an atelic event. (28) Quant Hulin voit payens aprés luy acourir when Huon sees pagans after him run.inf ‘When Huon sees the pagans quickly pursuing him’ (Huon Bordeaux B., c.1400–1450, 211; DMF) The uses of the verb accourir to simply mean ‘rush’ or ‘run quickly’ are numerous, particularly in the present participle form when used to modify a directional verb, as shown in (29). (29) et vint acourant par grant ire vers le cerf, and came running by great anger toward the deer ‘And in great anger it came rushing toward the deer.’ (Chev. papegau H., c.1400–1500, 84; DMF) To reconcile this variable use, we suggest that the child learner would have analysed a prefixed verb like accourir as an atelic manner verb based on uses like those in (28) and (29). This would drive the parse for a goal-of-motion construction, such as those in (26), to involve Path located in the prepositional phrase—not the prefix—and would require a null Path morpheme, as shown in (25a). Of course, not all prefixed verbs were reanalysed as manner verbs, but we propose that enough relatively high-frequency manner-of-motion verbs like accourir, avoler, attroter, achevaucher, etc. were used in enough atelic contexts to trigger a Path-inpreposition parse. The appearance of the non-prefixal null Path—and the resultative secondary predication constructions that it entails—would therefore be the result of the variable use of possibly only a handful of high-frequency prefixed manner-ofmotion verbs.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

From Latin to Modern French: A punctuated shift



The Modern French system has neither productive Path preverbs from Latin nor the null Path that permitted the satellite-framed innovations in Medieval French. This verb-framed system is the culmination of the centuries-long reanalysis of the preverbs’ contribution to the telic interpretation of the verb phrase as a contribution derived from the idiosyncratic meaning of the verb root itself. But if it is relatively clear how the telicizing preverb system gradually disappears, it is not so for the system illustrated in (25), which rapidly becomes obsolete by the sixteenth century. We think that the loss of null Path, which underlies the bare goal-of-motion and verb–particle constructions, is linked to the definitive loss of the Medieval French preverbs. If the system in (25) arises out of the variable use of prefixed manner-of-motion verbs, as we have proposed, then the loss of this variation may entail the end of the grammar that produces the new constructions. For example, a great many prefixed verbs disappear as each prefix loses productivity. This holds for manner-of-motion verbs with variable telic and atelic uses such as attroter and avoler. Meanwhile, the prefixed manner-of-motion verbs that remain are reanalysed as inherently telic (accourir ‘to run up’ and affluer ‘to flow in’, for example), resulting in an unambiguous Path-in-verb analysis.

. Theoretical implications The present study examines changes in the syntactic expression of events involving features related to event composition within the vP. From a diachronic point of view, this kind of change has been little explored; studies in generative diachronic syntax have tended to focus on word order changes where the analysis concerns categories and features found within the extended IP and CP domains. As we have emphasized, it is widely accepted that changes involving event structure, such as shifts from s-framed to v-framed or vice versa, occur gradually over very long periods of time. Our study seeks to challenge and qualify this description, demonstrating that, in some cases, gradual drift is but a mere impression. We have shown that there is no gradual unfolding of a v-framed language from a s-framed one, but rather an evolution punctuated by a new system, which then disappears rather abruptly during the fifteenth century. Only a theory that appeals to abstract grammatical formatives can really explain why this occurs. Within such a theoretical framework, we have attributed the loss of the intermediate s-framed system to the reanalysis of telicizing verbal prefixes and the properties of a null element in Path, which encodes transition/result. It is clear that the reanalysis of the Latin directional/aspectual prefixes is gradual and unrelenting; the slow and steady loss of the Path preverbs drives the shift toward a v-framed grammar, but as this happens, a dramatic development occurs, which, we claim, is contingent on the semantic bleaching of the preverbs. The variable use of prefixed manner-of-motion verbs causes a new grammar to emerge, one that includes a null Path morpheme, which itself underlies a cluster of syntactic innovations associated with resultative secondary predication. This new grammar, however, is shortlived. Once the Medieval French preverb system is lost and the prefixes are reanalysed as an inseparable part of the verbal root, the variable use of prefixed manner-of-motion verbs is lost too, the cue, we claim, for null Path. The s-framed grammar disappears

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i



Michelle Troberg and Heather Burnett

as a consequence. The kind of change we observe here is thus in line with Lightfoot’s (1999, 2006) well-known arguments that a shift in the distribution of cues has significant consequences for the abstract system, and further, that certain abstract changes in the grammar spread rapidly through a population of speakers. Our claim is therefore that changes at the syntax–semantics interface involving event structure proceed in much the same way as they do in the IP or CP domains; the reanalysis of functional morphemes or the non-instantiation of a functional head may entail the same kind of dramatic reflexes regardless of its type. We also attempt to show that a fine-grained analysis of the grammatical formatives involved in event composition can account for the development of intermediate microsystems that fall outside broad typological descriptions.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

 Case in diachrony: Or, why Greek is not English N I KO L AO S L AV I DA S

. Introduction The starting point of this study is the common historical development with regard to aspect and definiteness in Greek and English: Greek attests both a change from inner (Aktionsart) into outer (grammatical) aspect as well as a change from demonstratives into definite articles, as English does. Even though aspect and definiteness are connected with Case, we will show that the development of the Case system in Greek differs in several aspects from the development of the Case system in English. Moreover, we will demonstrate the status of Ancient Greek1 as a transitional stage where aspectual features that are related to Case can appear on two different heads ([Inner]Asp[ect] and Trans[itivity]). In this respect, we will discuss the issue of the nature of transitional stages for the development of a language. We follow van Gelderen’s analysis (2000; 2011a,b), according to which structural Case is connected with a probe with an uninterpretable feature, whereas inherent Case is connected with an interpretable feature. We will present evidence that Greek attested similar changes in aspect and definiteness as English did. The initial hypothesis is that the Case system of Greek should also have changed in a similar way as the Case system of English. On the contrary, the Case system of Greek followed a different historical path. This issue has theoretical implications for the mechanisms of language change, as well. We will, therefore, discuss the possibility of a change from structures with aspectual features on Asp into structures with aspectual features on Asp and Trans (Ancient Greek) and, then, structures with aspectual features only on Trans (PostKoine Greek). Section 9.2 presents the main reasons that support the view of an interrelation between Case, aspect, and definiteness. The main argumentation in favour of this interrelation derives from the diachrony of English. We present the development of 1 Ancient Greek: th century bc–rd century bc (Classical Greek th century bc–rd century bc), Koine Greek: rd century bc–th century ad (New Testament Greek st century ad), Early Byzantine Greek: th century ad–th century ad.

Micro-change and Macro-change in Diachronic Syntax. First edition. Éric Mathieu and Robert Truswell (eds). This chapter © Nikolaos Lavidas 2017. First published 2017 by Oxford University Press.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i



Nikolaos Lavidas

the three interrelated categories in English, following the analysis by van Gelderen (2000, 2011a,b). Section 9.3 contrasts the English facts with the development of Greek. In Section 9.3.1, we show that, even though the changes in aspect and definiteness are similar, the picture of the tendencies with regard to the Case system is quite different. Section 9.3.2 supports our analysis with two relevant phenomena that mark the change from the Pre-Koine Greek system of Case to the Post-Koine system: in PreKoine Greek but not in Post-Koine Greek, there are examples of violation of number agreement between the subject and the verb; Pre-Koine but not Post-Koine Greek allows the presence of accusative objects with mediopassive (non-deponent) verbs. Section 9.4 summarizes the main conclusions of the study.

. Case, aspect, and definiteness, and their interrelation The interrelation between Case, aspect, and definiteness is a crucial parameter for the present study. This interrelation has been shown in several analyses; see, for instance, Abraham (1997); van Gelderen (1997, 2011a,b); Kiparsky (1998); among others.2 In the syntactic perspective, the relationship between aspect and argument structure is manifested through the representation of aspect as a functional feature or as a category within the verbal domain (among others, Ritter and Rosen 1998; Sanz 2000). Aspectual features determine the presence and the case of the arguments (Tenny 1987; Arad 1998; Borer 2004), whereas Case can mark definiteness/specificity (van Driem 1986).3 For instance, the contrast between accusative and partitive case in Finnish has been argued in many studies to be very similar to the aspect contrast in Russian (and other Slavic languages): see, among many others, Dahl and Karlsson (1976) and Dahl (1985). According to Kiparsky (1998), Finnish uses the partitive case whereas Russian uses the imperfective aspect if the object is an indefinite bare plural or the verb is atelic (it does not express a completed event). Both the accusative case in Finnish and the perfective aspect in Russian demand that the object is definite plural and that the verb is telic (1a–b). On the other hand, the partitive case in Finnish and the imperfective aspect in Russian result in ambiguous structures in a similar manner (2a–b). (1)

a. Russian On na-pisa-l pis’m-a. he pfv-write-pst.3sg letter-pl.acc b. Finnish Hän kirjoitt-i kirjee-t. s/he write-pst.3sg letter-pl.acc ‘He wrote the letters [. . . and left].’

(Kiparsky 1998: 272)

2 Smith (), for instance, has also argued that the meaning of a predicate is related to the verbal inner aspect and the specificity of the object. Due to space restrictions, our discussion of the theoretical background of the interrelation between Case, aspect, and definiteness has to be brief. See also Lavidas (). 3 See van Gelderen (a: ):

Definiteness and specificity can be marked through Case in, for example, Finnish, Turkish, Persian, Japanese, and Limbu (van Driem 1986: 34), through aspect as in Russian (Leiss 1994, 2000; Abraham 1997; Philippi 1997), through position as in Chinese, through a determiner, and through a combination of position and articles in Arabic, Dutch, and German, for example (Diesing 1992).

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

Case in diachrony: Or, why Greek is not English (2) a. Russian On pisa-l pis’m-a. he ipfv.write-pst.3sg letter-pl.acc b. Finnish Hän kirjoitt-i kirje-i-tä. s/he write-pst.3sg letter-pl-part ‘He wrote [some] letters [. . . and left].’ ‘He was writing letters [. . . when I came].’ ‘He was writing the letters [. . . when I came].’



(Kiparsky 1998: 272)

Similarly, Leiss (2000) has shown that verbal aspect and nominal definiteness are ‘two sides of the same coin’. For this reason, Finnish and Russian do not require articles for definiteness: they both have, instead, an analogous system of Case and aspect. Accordingly, Leiss has argued in favour of a close relationship between Case, aspect, and definiteness. With regard to Russian, she has demonstrated (similarly to Kiparsky 1998) that the definiteness of the object as presented in the contrasted examples in (3a–b), is the result of the perfective aspect (as expressed by the verbal prefix), which demands a definite object. The common characteristics of aspect and definiteness concern the (in)completeness of the event and how the nominal and the event are measured. (3) a. On kolo-l drova. he split-pst wood ‘He was splitting wood.’ b. On ras-kolo-l drova. he pfv-split-pst wood ‘He split the wood.’

(Leiss 2000: 12)

Moreover, with regard to Old Norse, Leiss has shown that the aspectual prefixes are lost in Old Norse and, in parallel to this development, definite articles arise. In this respect, she has claimed that, despite the loss of the aspectual prefixes, Old Norse shows perfective aspect with the historical present tense. According to her analysis, definite articles in Old Norse (as well as verb-first word orders and adverbs) contribute to the perfective aspect. Leiss has also revealed several significant aspects of the role of Case in this type of development in Old Norse. She has argued that the articles emerge with the genitive case because the genitive is (the first case that is) no longer aspectually marked (see, for instance, the use of articles with genitives but not with nominatives in Old High German; Glaser 2000). We turn now to the starting point of our study: van Gelderen has shown that Case, aspect, and definiteness are connected with regard to their development in English. Our discussion of the English facts is based on van Gelderen (1997, 2000: ch.5, 2011a,b). Van Gelderen has argued that the change in the Case system of English, by late Old English, is a change of interpretable features of Case, aspect, and definiteness—marked by an Asp category—into uninterpretable features. Van Gelderen’s analysis is based on the fact that, in Old and Early Middle English, the alternation between the genitive and accusative expresses partial affectedness and definiteness, respectively: the genitive marks partial affectedness of the object (mainly with verbs of deprivation

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i



Nikolaos Lavidas

or mental action, (4a)); on the other hand, the accusative marks total affectedness and definiteness (4b). The genitive–accusative contrast (see Kiparsky 1998 for the partitive–accusative contrast) expresses a link between affectedness and measure, that is, between aspect and definiteness. (4) a. Ðar com eft ongean Swegen eorl to Eadwerde cinge and gyrnde there came back again Swegen earl to Edward king and craved to him landes þæt he mihte hine on afedan. of him land.gen that he might it on sustain ‘Then Earl Swegen came back again to King Edward and wanted land from him so that he would be able to sustain himself.’ (Chronicle D, anno 1049.9) b. se helend þa witende þohtas heora cweþ to heom . . . the healer then knowing thoughts.acc their.acc said to them ‘The Saviour, knowing their thoughts, said to them . . .’ (Rushworth Glosses, Matthew 12.25, van Gelderen 2011b: 129) Moreover, van Gelderen has argued that, in the twelfth century, the genitive case for objects and the pronominal forms that distinguish the dative and accusative are lost. For instance, based on the lists of verbs that have genitive objects in Old English (Bungenstab 1933; Mitchell 1985), she has shown that only very few verbs can take a genitive object in Middle English (if we compare them with the list of over 200 verbs that take genitive objects in Old English). With regard to the parallel development of definiteness and aspect, according to van Gelderen’s analysis, during the same stage of the history of English, definiteness marking on nouns is present in more instances whereas aspectual prefixes disappear. In the same stage (twelfth century), the demonstrative pronoun se is also reanalysed as the definite article, as shown in (5a,b) from van Gelderen (2011b). Evidence for this reanalysis derives from the fact that the earlier demonstratives/articles must not be followed by a noun; the definite articles, in the period after late Old English, appear with nouns, not on their own. For van Gelderen, the definite articles change from having interpretable deictic features into having uninterpretable Case features. Following Leiss (2000) for Old High German, van Gelderen has claimed that, in late Old English, the availability or presence of the definite article in all positions can be analysed as a compensation for the loss of the genitive (inherent) Case. (5) a. Se heora cyning ongan ða singan that their king began then sing ‘That king of theirs began to sing.’ (Orosius 35.14–15) þone abbode Saxulf ⁊ b. Ic Wulfere gife to dæi Sancte Petre ⁊ I Wulfhere give today St. Peter and the abbot Saxulf and þa munecas of þe mynstre þas landes ⁊ þas wateres the monks of the abbey these lands and these waters ‘I, Wulfhere, am giving today to St. Peter and Abbot Saxulf and the monks of the abbey these lands and waters.’ (Peterborough Chronicle anno 656: 40, van Gelderen 2011b: 130)

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

Case in diachrony: Or, why Greek is not English



Turning to the other part of the interrelated parameters, aspect, we should notice that evidence for the change in aspect comes from the perfectivizing/transitivizing prefixes of Old English. Van Gelderen has shown that the perfectivizing/transitivizing prefixes (for instance, ge-) can have the role of adding a Theme argument in Old English. These prefixes are lost in late Old English/early Middle English. Many of these prefixed verbs are replaced by French loanwords in this stage: for instance, bedrincan → absorb / forðbringan → produce (Brinton 1988: 215–31). Moreover, as van Gelderen has mentioned, the prefix ge- changes before disappearing: prefixes in Old English denote telicity, but ge- first becomes a perfect marker for present relevance and then is lost (van Gelderen 2011b: 131–2). Van Gelderen has tied all the above facts (on the development of Case, aspect, and definiteness in English) together to conclude that Old English interpretable measure features in Asp are responsible for all relevant characteristics: (a) the involvement of a Theme in the event, (b) the genitive or accusative case that corresponds to affectedness or definiteness of the object, (c) the aspectual characteristics of the event described in the clause. The diagram in (6) is a representation of the foregoing remarks in the syntactic structure, according to van Gelderen: the interpretable [measure] features in Asp result in the (non)affectedness of the Theme (and the genitive/accusative case). Following this perspective, inherent Case (in contrast to structural Case) is related to one feature that licenses a thematic role: the [measure] feature is able to link the affectedness and definiteness of the nominal to the perfectivity (and transitivity) of the event (van Gelderen 2011b: 132).

AspP

(6)

Asp

DP D þis [+affected]

N land

ge[i-measure]4

VP V fer-

√ fer(van Gelderen 2011b: 132, (45))

On the other hand, van Gelderen has argued that Middle and Post-Middle English have a different Case system that does not include an inherent Case. Following Pesetsky and Torrego’s (2004) approach, she has analysed Middle and Post-Middle English structural Case as a Case that involves uninterpretable and unvalued T/Asp features on the DP (valued by T[ense] for subjects and by v for objects). T and v have uninterpretable phi-features (person and number features) and probe for a feature on the DP.5

4

See also Borer’s () analysis of theta-roles as aspectual features. Notice that agreement is significant for Pesetsky and Torrego’s perspective because an uninterpretable phi-feature (of person and number) in v probes in the structure for a feature on the DP (see Section .. for the relevant Greek data). 5

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i



Nikolaos Lavidas

The tree in (7) represents some parts of van Gelderen’s analysis (VP and AspP are left out in this schema to focus on Case assignment in Post-Middle English). According to this analysis, in the new system of Middle and Post-Middle English, the interpretable aspect features are in v, not in Asp, and the DPs have uninterpretable Case features ([u-Asp], in the case of objects). (7)

TP T T [i-T] [u-Phi]

vP DP [u-T] [i-Phi]

v v [i-Asp] [u-Phi]

DP [u-Asp] [i-Phi] (van Gelderen 2011b: 137, (53))

To summarize, according to van Gelderen’s analysis, Case in English is related to an interpretable Asp feature on a DP object valued by Asp in Pre-Middle English, and to an uninterpretable Asp on a DP object valued by v in Middle and Post-Middle English. Accordingly, in Old English, the alternating genitive and accusative cases mark partial affectedness and definiteness, respectively. In Early Middle English, not only is the case alternation no longer productive, but also a parallel change in definiteness (rise of definite articles from demonstratives) and in aspect (loss of perfectivizing/transitivizing prefixes) is attested. Section 9.3 will compare the diachrony of these three interrelated phenomena (Case, aspect, and definiteness) in Greek with the above remarks on the diachrony of English. We will show that, even though similar tendencies can be observed for aspect and definiteness, Case in Greek followed a different path.

. Case, aspect, and definiteness: A comparison of the diachrony of Greek and English .. The Greek tendencies ... A similar picture for the diachrony of definiteness and aspect Greek shows similar tendencies as English with regard to definiteness and aspect—but not with regard to the Case system. For example, in Standard Greek, the genitive case is retained (and expanded to cover more functions).6 We will first present the similar tendencies that can be observed with regard to definiteness and aspect in the Greek diachrony, if it is compared to the diachrony of English. Then, we will focus on an analysis of the differences between the diachrony of Case in Greek and English. Our proposal will be that the different path in the Greek diachrony can be accounted for with regard to 6

On transitivity and object pronouns and their diachrony in English and Greek, see also Lavidas ().

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

Case in diachrony: Or, why Greek is not English



changes in the Case system of Greek in relation to syntactic properties of the clause, and in particular the realization of ±interpretable features in inner (Aktionsart) and outer (grammatical) aspect. According to our proposal, the change that takes place in the history of Greek involves a transitional stage in Ancient Greek where the relevant features can appear on the Trans7 head (outer aspect) or the Asp head (inner aspect). The completion of the change in Post-Koine Greek involves the loss of the interpretable feature on Asp; thus Case remains (uninterpretable) on the higher Trans head. The diachrony of definiteness in Greek does not differ from the diachrony of other Indo-European languages in that the definite article is derived from demonstratives (Manolessou 2000, 2001; Bakker 2009). Guardiano (2011, 2013), following Bakker (2009) and Napoli (2009), has clearly shown that the definite article (ho, h¯e, tó) in Ancient Greek is not compulsory but connected to several functions. According to Guardiano’s analysis, multiple functions—anaphoric, pragmatic (with retopicalization of an element), informational—can be attributed to the definite article (ho, h¯e, tó) of this stage. It is a common consensus that definiteness is not grammaticalized in Homeric Greek—on the contrary, it is the accusative case and its alternates that contribute to definiteness/specificity. Evidence is provided, for instance, from the fact that ho, h¯e, tó in Homeric Greek do not appear together with a demonstrative—presenting properties of demonstrative pronouns themselves; see (8a,b) from Guardiano (2013), for an example of absence of a definite article in a definite context in Ancient Greek—the bare nouns equal a definite DP in these instances. Ho, h¯e, tó also function as pronouns in Homer, used anaphorically or cataphorically ((9a–c), from Guardiano 2013). As Guardiano has observed, this does not mean that anaphoric definiteness is grammaticalized, because the repetition of a nominal can happen without the ho, h¯e, tó. Moreover, the Homeric Greek ho, h¯e, tó can occur when the nominal phrase includes a relative clause even though it cannot occur with a demonstrative.8 Guardiano’s conclusion is that, in definite contexts, ho, h¯e, tó are present in Homer but not in a systematic manner: they can occur where the demonstrative occurs (anaphoric or topical contexts, for instance), but they are not compulsory (again, similar to demonstratives). Therefore, they cannot be a D-element. (8) a. éntha kaì ¯ematí¯e mèn huphaínesken mégan histón. then and day-by-day prt weave.3sg great.acc web.acc ‘Then day by day she would weave at the great web.’ (Hom. Od. 2, 104) b. karpalím¯os árnas te phérein Príamón te kaléssai. quickly lambs.acc and fetch.inf Priam.acc and summon.acc ‘quickly to fetch the lambs and to summon Priam’ (Hom. Il. 3, 117, Guardiano 2013: 82)

7 We will follow Tsimpli and Papadopoulou () and claim that a Trans head (and not the v head) may contain outer-aspectual features and be responsible for structural Case. See Section .... 8 The Ancient Greek ho, h¯ e, tó have several non-compulsory functions (see also Bakker ; Napoli ). Guardiano (: ) uses the term ‘transition stage’ for Ancient Greek with regard to definiteness: ‘Such an alleged freedom is often justified as the product of transition stages.’

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i



Nikolaos Lavidas

(9) a. L¯etoûs kaì Diòs huiós. ho gàr basilêi khol¯otheís. Leto.gen and Zeus.gen son.nom 3sg.nom prt king.dat anger.pret ‘son of Leto and Zeus. For this reason, he angered at the king.’ (Hom. Il. 1, 9) b. ápoina . . . tà d’ ápoina dékhesthai ransom.acc art.acc prt ransom.acc accept.inf ‘ransom [. . . ] and accept the ransom’ (Hom. Il. 1, 13–20) c. aoidòn . . . dè tóte tòn mèn aoidòn ág¯on es minstrel.acc prt then art.acc prt minstrel.acc take.prt to nêson er¯´em¯en isle desert ‘minstrel [. . . ] then indeed (Aegisthus) took the minstrel to a desert isle.’ (Hom. Od. 3, 267–270, Guardiano 2013: 83–4) On the contrary, in Classical and New Testament Greek, definiteness is grammaticalized with the definite articles: ho, h¯e, tó are systematically present with definite nominals. Accordingly, bare nominals cannot be definite (in the case where they do not have a unique referent or if they are not abstract nouns) (Guardiano 2013: 86). In the first century ad, the process of change in definiteness is completed with the grammaticalization of the indefinite article: the definite and the indefinite articles with singular count nouns are then distinguished for the first time. To summarize, Homeric Greek shows no grammaticalization of definiteness and ho, h¯e, tó are demonstrative pronouns. Definiteness is expressed through the presence of the accusative case and the accusative/genitive alternation in this period (see also below with regard to cases). On the other hand, Classical and New Testament Greek demonstrate grammaticalization of definiteness. In Post-Koine Greek, the change in this system of definiteness is completed and an indefinite article is also present with singular count nouns. The grammaticalization of definiteness in Greek is attested with the emergence of (a) the definite article in Classical Greek, (b) the indefinite article in the first century ad. Demonstratives are reanalysed into definite articles in Classical Greek, whereas indefinite articles derive from numerals in Koine Greek. Turning to the diachrony of aspect, we should notice that a significant difference between Modern Greek and English is the grammaticalization of the perfective–imperfective contrast in Greek. English grammaticalizes the progressive– nonprogressive opposition (See (10) on subdivisions of aspect, from Comrie 1976).9 This difference, for instance, has been shown to affect the preference for null (nonreferential) or overt objects; see Tsimpli and Papadopoulou (2006) and (11a–b).10

9 10

On aspect in Modern Greek, see also Xydopoulos and Tsangalidis (). According to Tsimpli and Papadopoulou (: ):

The perfective/imperfective distinction has been shown to affect object drop in other languages too. Babko-Malaya (1999) shows that, in Russian, null objects are allowed in sentences with imperfective unprefixed verbs but not with perfective ones.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

Case in diachrony: Or, why Greek is not English

Aspect

(10)

Perfective

(Comrie 1976)

Imperfective

Habitual

Continuous

Progressive (11)



Nonprogressive

a. Mary drew *(the painting). (Modern English) b. (i) ?I Maria zoghrafise sto spiti tis. art.nom Maria.nom drew.pfv.3sg in-the home her ??‘Maria drew at her home.’ (ii)

I Maria zoghrafize sto spiti tis. art.nom Maria.nom drew.ipfv.3sg in-the home her ??‘Maria was drawing at her home.’ (Modern Greek)

Moser (2008, 2009, 2014) has discussed in detail a change from a tripartite morphological opposition (aorist–present–perfect stems) that was connected with inner aspect (telicity–atelicity–state) into a morphological opposition (aorist–present stems) that is linked to outer aspect (perfective–imperfective opposition). Her proposal accounts for the distribution of aorist and present in Pre-Koine Greek and for the loss of the perfect in Koine Greek. According to this analysis, in Pre-Koine Greek, the verbs that are attested exclusively in the perfect are states; the movement from an inner aspectbased verbal system to an outer aspect-based verbal system resulted in the loss of the perfect. Hence, the the loss of the perfect stem signals a change in the aspect. The category expressed by this stem is also lost, resulting in a change of a three-part distinction to a two-part distinction (see (12)). Moser’s conclusion is similar to Sihler’s conclusion (1995), for Greek and other Indo-European languages: Homeric Greek shows that the present–aorist–perfect distinction is connected with inner aspect in earlier stages and denotes duration/nonterminativity, instantaneity/terminativity, and stativity, respectively; see (13) on the archaic verbal system, from Moser (2014). (12) gráph-¯o (present stem) / gráps-¯o (future stem) / gégraph-a (perfect stem) → graf-o (imperfective stem) / (tha / eho) graps-i (perfective stem)

Situations

(13)

Stative (Perfect)

(Moser 2014: 76)

Dynamic Dynamic/atelic (Present)

Instantaneous/telic (Aorist)

We should notice that this does not mean that Moser has supported a ‘homogeneous’ inner aspect-based picture for Homeric Greek. Counterexamples can be found in

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i



Nikolaos Lavidas

Homer and these are the result of the fact that Homer’s texts contain elements from different dialects and even different stages of the language. According to Moser, Classical Greek can show the first evidence of the change: verbs can appear in full paradigms in Classical Greek and can be used to express outer aspect as well. It is important, though, that Moser has observed for Classical Greek that there is still dependence from inner aspect. This is obvious if one compares Post-Koine (for instance, Modern) Greek with Classical Greek with respect to the use of past tenses and infinitival (for Classical Greek) complements. Moser has concluded that the differences that can be observed in such comparisons mainly concern the use of imperfective forms in Classical Greek: these forms are attested in many instances and, specifically, mainly with durative verbs in Classical Greek (in contrast to Post-Koine Greek). With regard to example (14), Moser has shown that the Classical Greek imperfective form of epiplé¯o (durative, atelic) cannot appear in Post-Koine Greek (in contrast to what holds for the other verb, peripiplé¯o—durative, telic): this imperfective form can only be accounted for in terms of durativity and the role of durativity in Classical Greek. On the other hand, Post-Koine Greek imperfective forms are not inner aspect-based (but outer aspect-based) and, therefore, are not accepted in relevant contexts. (14) hai dè tri¯e´reis tôn Surakosí¯on háma art.nom.pl prt triremes.nom art.gen Syracusians.gen simultaneously kaì apò ksunth¯e´matos pénte mèn kaì triákonta ek toû and from agreement five prt and thirty from art.gen megálou liménos epépleon . . . kaì periépleon big.gen port.gen sail.ipfv.pst.3pl and sail.around.ipfv.pst.3pl boulómenoi pròs tàs entòs prosmíksai wanting to art.acc inside join.inf ‘Thirty-five triremes of the Syracusians sailed according to the agreement from the big port [. . . ] and sailed around wanting to join the ones inside.’ (Th. 7.22.1, Moser 2014: 77) A clear contrast between Post-Koine and Pre-Koine Greek verbs can also be described with the following remarks (besides the loss of the perfect stem) (Moser 2014: 80): in Post-Koine Greek, (a) all verbs can appear with the perfective–imperfective opposition; (b) a morphological levelling has been applied—mainly with the expansion of the perfective -s-; (c) this aspectual opposition is expanded to the future tense—which also demonstrates this dual aspectual opposition.11 Moser has argued that the link of the outer aspect to inner aspect (as observed, for instance, in the constraints posited by the inner aspect on the verbal forms) is absent from Post-Koine Greek. In Post-Koine Greek, the outer aspect is independent from inner aspect. The connection between inner aspect and outer aspect does not further affect the verbal system in Koine Greek. This can be translated for our analysis, in that there are no interpretable inneraspectual features on Asp in Post-Koine Greek, whereas Ancient Greek has aspectual features on Trans as well as on Asp (Table 9.1). Accordingly, Ancient Greek can be defined as a transitional stage, the crucial characteristic being that aspectual features 11

See also Napoli (), who examines the interaction of inner and outer aspect in Ancient Greek.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

Case in diachrony: Or, why Greek is not English



Table .. Archaic vs. Pre-Koine Greek vs. Post-Koine Greek system: the aspectual features Archaic system (traces in the archaic system of Homer)

Pre-Koine Greek system

→ inner-aspectual features on Asp

→ inner-aspectual features on Asp → outer-aspectual features on Trans

Post-Koine Greek system

→ outer-aspectual features on Trans

are to be found on two heads in this stage. Post-Koine Greek is a system with aspectual features only on Trans—not on Asp. This difference between Pre- and Post-Koine Greek can also explain the fact that passivization was not always available in Pre-Koine Greek (Luraghi 2010b).12 This can be easily explained if we agree that passivization is only linked to the (outer) aspectual features on Trans—and not to the (inner) aspectual features on Asp. According to this assumption, in the case that the object is licensed by Asp (and not by Trans), passivization is not possible.13 Post-Koine Greek shows aspectual features only on Trans. In this respect, the same structures are attested in all periods but aspectual features appear both on Trans and Asp in Classical Greek. This needs to be the definition of a transitional stage: similar features can appear on two heads; in some cases, the object is related to Asp—in other cases, to Trans.14 ... Case in Greek The grammaticalization of definiteness and outer aspect followed similar paths in Greek as in English. We argue, however, that the development of the Case system in Greek is different than in English. This remark can raise questions on the historical correlation of Case, aspect, and definiteness. Table 9.2 summarizes our hypothesis with regard to this diachronic comparison (and change in these particular languages). We claim that Case in Ancient Greek can also be interpretable and reflect the aspectual features on Asp (in the transitional stage of Ancient Greek, the relevant features can appear on the Trans head [outer aspect] or the Asp head [inner aspect]). This is evident in the distribution and interpretation of cases in Ancient Greek (Luraghi 2003, 12 We claim that the development of aspect, from being related to an aspectual feature on Asp and Trans into being related to an aspectual feature on Trans, is reflected in the expansion of passivization. Luraghi (b) states:

. . . the extent to which such [verbs] may passivize was limited in Homer, but increased in later prose, showing that they were being reanalysed as transitive. . . . The extent to which verbs with non-accusative objects may passivize increases after Herodotus. (Luraghi 2010b: 60, 69) 13 We agree with a reviewer that this probably has consequences for the interpretations of the verbs that can(not) passivize in the different stages. We leave the significant issue of the verbal semantic classes and their role for the particular development open for future research. 14 In other words, there can be an interpretable feature on Asp (and inherent Case) or an uninterpretable one on Trans (and structural Case) in Ancient Greek.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i



Nikolaos Lavidas

Table .. English vs. Greek: Case, definiteness, and aspect in diachrony

Definiteness [as seen in Sections 9.2 and 9.3.1.1] Aspect [as seen in Sections 9.2 and 9.3.1.1] Case [our hypothesis]

English

Greek

Demonstratives: loss of the interpretable deictic feature → definite articles inner → outer aspect (for instance, loss of aspectual perfectivizing prefixes)

Demonstratives: loss of the interpretable deictic feature → definite articles inner → outer aspect (and lability with change in the voice morphology)15

[i-measure] → [u-asp]

[i-measure] and [u-outer-asp] → [u-outer-asp]; through Trans (that includes an [i-outer-asp] feature)

2009, 2010b): the accusative marks wholly affected patients; it appears with verbs of killing, destruction, and creation, for instance. The genitive marks partially affected patients. In this respect, the genitive is used for arguments without agentive properties, whereas the accusative is used for arguments with agentive properties. The dative marks objects that are not patients—there is no expression of change of state—but of human participants: recipients, possessors, experiencers, or benefactives.16 The alternation between the cases is a typical characteristic of Ancient Greek: for instance, the genitive (vs. accusative) can be used for the expression of low affectedness, whereas the dative (vs. accusative) can be used for the expression of absence of change of state.

15 A comparison of the directionality of valency in Greek with the directionality of valency in English can provide further evidence on possible similarities or differences in valency and Case systems. Van Gelderen (b) has shown that English changed towards lability (same morphology for the transitive and the intransitive [anticausative] type) with loss of prefixes and suffixes of transitives/causatives. Greek also changed towards lability but with loss of the unmarked status of the mediopassive/nonactive voice morphology for anticausatives. This means that these two Indo-European languages followed the same direction with a significant difference: English mainly changed the form of its transitives whereas Greek changed the voice morphology of its intransitives (Lavidas ). 16 Studies that have examined the historical development of the Greek cases have focused on the dative case. See, for instance, Cooper and Georgala (); Michelioudakis and Sitaridou (); Alexiadou et al. (, ). Due to space restrictions, we cannot present details of these studies here. We should note that, based on passivization properties, Alexiadou et al. (, ) analyse the dative of Ancient Greek as a mixed (structural and inherent) Case. They do not discuss, however, the relationship between Case and aspect or definiteness. Our analysis is in accordance with these studies because they recognize characteristics of an inherent Case in the dative of Ancient Greek, some of them starting with the remarks in Lavidas () on the changes in the passivization of the dative in the diachrony of Greek. Note that, with regard to the dative and passivization in English, passivization of the dative is not possible in Old English, when (as we assume, following van Gelderen , a,b) Case was inherent–interpretable on Asp. It becomes available in Middle English when Case becomes structural–uninterpretable on v (or Trans, in our terms).

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

Case in diachrony: Or, why Greek is not English



Accordingly, in Ancient Greek, the cases affect the interpretation of the constructions and the nominals can have a restricted relation to the verb. But the relation between the verb and the nominals strengthens in Koine and PostKoine Greek. Besides the accusative, which is expanded in Greek,17 the genitive is retained and also expanded. Notice that the clitic pronoun in the genitive appears to have been reanalysed as part of the verbal domain too—and not only of the nominal domain; see (15), from Horrocks (2010), where the clitic in the genitive case can be reanalysed from belonging to the nominal domain to belonging to the verbal domain. This fact demonstrates an overlap between the dative and genitive and a replacement of the dative through genitive, first in the category of pronouns and then of nouns. Horrocks (2010: 184) further provides examples from the fourth century ad where the genitive and accusative pronouns replace the dative as verbal complements (16a–b). (15) tò proskúns¯emá sou poiô art.acc supplication.acc 2sg.gen make.1sg ‘I also make supplication for you.’ (2nd century ad, Horrocks 2010: 180) (16) a. ¯epánt¯eká sou met.1sg 2sg.gen ‘I met you.’ b. eír¯eká sou said.1sg 2sg.gen ‘I said to you’

(Horrocks 2010: 184)

The overlap between the dative and the genitive is also seen in (17). The genitive as a clitic appears in a high position in the structure (second position). It can replace the dative that also appears in a high position and can have a possession meaning. (17) lambáneis mou tà grámmata. receive.2sg 1sg.gen art.acc letters.acc ‘You receive my letters.’

(P. Flor. 127, Horrocks 2010: 116)

To summarize, in the later papyri (fourth century ad), many examples of the genitive and accusative for indirect objects—instead of the Classical Greek dative—are attested. The use of the genitive expanded from the pronouns to the nouns. The functions of the dative started to be expressed with the accusative (indirect object, temporal expressions), or the genitive or PPs (Humbert 1930; Trapp 1965). This period, characterized as a ‘long transitional period’ by Horrocks (2010: 182) with regard to cases, too, includes examples of violations of concord with a preference for the accusative (see also Section 9.3.2). The accusative case (instead of the nominative) appears with singular and plural participles, for instance. Horrocks (2010: 182) describes the relevant change as ‘a growing preference for the accusative as an allpurpose oblique case’.

17 We interpret Luraghi’s ‘extension of transitive features’ as the process of loss of aspectual features on Asp.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i



Nikolaos Lavidas

In our syntactic analysis of the Greek clause, we follow the perspective of Tsimpli and Papadopoulou (2006) and we assume that Spec,TransP is the location of merge for the individuated (definite and/or animate)18 objects; the lower VP is where nonindividuated objects are merged. Trans includes outer-aspectual features: for Tsimpli and Papadopoulou, the outer-aspectual/perfective–nonperfective features are grammaticalized as formal features on Trans in Modern Greek. Accordingly, in Modern Greek, in the case of a Trans [+pfv] verb, the DP object is merged in Spec,TransP; in the case of a Trans [−pfv] verb, the DP object is merged in the lower VP. Case assignment demands movement of the object from VP to TransP with Trans [−pfv] verbs. See (18a,b) from Tsimpli and Papadopoulou (2006) (who analyse the Modern Greek data). (18) a.

vP v

Subj v

TransP Trans

Obj Trans [+pfv]

VP V V

b.

vP v

Subj v

TransP Obji Trans [–pfv]

Trans VP V

ti V

(Tsimpli and Papadopoulou 2006: 1604) 18 On individuated objects, see, among others, Hopper and Thompson (); Comrie (); Aissen (); Næss ().

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

Case in diachrony: Or, why Greek is not English



Following the above analysis, we propose that, in Pre-Koine Greek, similar to Old English, Case, aspect, and definiteness are interpretable through an Asp head: the Pre-Koine Greek accusative—similar to other Indo-European languages—marks Themes whereas the dative marks Goals, and the genitive expresses partial affectedness (in contrast to the accusative, which denotes total affectedness).19 The object has the option to be either in Spec,AspP—for instance, with a partial affectedness interpretation—and/or in Spec,TransP (19a). We will further comment (in Section 9.3.2) on the fact that the Pre-Koine Greek accusative can appear with mediopassive (voice alternating verbs—i.e., not only with deponents) transitive verbs. (19) a. Pre-Koine transitives vP

v

Subj v

TransP

Obj [u-OuterAsp]

Trans Trans

AspP Asp

Gen/Dat/Acc Obj [i-InnerAsp] Asp

VP V V

In this manner, the loss of the dative case is correlated with a reorganization of the Case system (and not, for instance, only with a phonological change/a phonological merger of the dative with the accusative; see also Horrocks 2010: 144ff., 180). Definiteness and aspect are also linked to the development of Case in a clear way. As discussed in Section 9.3.1.1, aspect in Greek changed from expression of inner aspect to expression of outer aspect (±perfective) oppositions. With regard to the chronology of the changes, the new (aspectual) system appears in the first centuries ad to be completed in Early Byzantine Greek (see also Blass and Debrunner 1961). Furthermore, the definite article is assumed (in the relevant bibliography) to have developed in Classical Greek, but the indefinite article appears in the first century ad to take its ‘modern’ status in Early Byzantine Greek. Post-Koine Greek demonstrates a different system of Case, as well: the emergence of definite articles has been completed in Early Byzantine Greek and the distribution of genitive and accusative objects is different than in Pre-Koine Greek. In other words, Case changes from being assigned via inner- and outer-aspectual 19

Cf. the discussion in Section ..

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i



Nikolaos Lavidas

features on Asp and Trans (respectively) to being assigned via an outer-aspectual feature on Trans (19b). (19) b. Post-Koine transitives (on the Trans head in Modern Greek, see Tsimpli and Papadopoulou 2006) vP

v

Subj

TransP

v

Obj [u-pfv/ipfv] [i-phi]

Trans Trans [i-pfv/ipfv] [u-phi]

VP V V

Comparing (19a) for Pre-Koine Greek, and (19b) for Post-Koine Greek, we observe that aspectual features are present only on the Trans head in Post-Koine Greek, and that the objects contain uninterpretable Case features. The new (Post-Koine) system emerges with the change of Case assignment, from being related to Asp and/or Trans into being linked to Trans.20 Table 9.3 summarizes some of the main remarks and assumptions with regard to the development of Greek cases and aspect. .. Further related Greek issues In this section, we present further evidence in favour of the described change (from the Case system of Pre-Koine Greek to the Case system of Post-Koine Greek) and our relevant hypothesis. Our arguments here concern issues of violations of Agree. Pre-Koine cases, which can be interpretable/inherent and related to Asp, allow violations of Agree (on a regular basis), and objects can appear with verbs having mediopassive (nonactive) as well as with active voice morphology without restrictions. Post-Koine cases, which are only related to Trans, disallow any kind of violation of Agree and demand the presence of active voice morphology for verbs that take an object (except for the case of the voice non-alternating deponents). In this respect, we will also explain some examples of noncanonical accusative subjects in Early Byzantine Greek, when the new system of Case emerges.

20 It is obvious that Case is associated with the higher head (Trans) in the new (Post-Koine) stage. This is in accordance with the syntactic changes as described by Roberts and Roussou (): in many cases, Roberts and Roussou have observed a reanalysis of an element as an exponent of a higher head.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

Case in diachrony: Or, why Greek is not English



Table .. Archaic vs. Pre-Koine Greek vs. Post-Koine Greek system: case and aspect Archaic system (traces of the archaic system in Homer)

Pre-Koine Greek system (transitional stage)

Post-Koine Greek system

→ Case alternations between accusative and genitive, accusative and dative

→ (a) Case alternation between accusative and genitive (with interpretable features) as well as (b) structural Case with uninterpretable features

→ Structural Case with uninterpretable features

→ Inner-aspectual features on Asp

→ Inner-aspectual (interpretable) features on Asp → Outer-aspectual (uninterpretable) features on Trans

→ Outer-aspectual features on Trans

Our first remark concerns a type of semantic agreement—but morphosyntactic nonagreement—that is attested in Pre-Koine Greek. Its origin is probably linked to the Proto-Indo-European neuter plural, which was a collective form with a singular meaning. This phenomenon is called skhêma attikón (‘attic schema’; e.g. (20)). Postclassical writers, except Atticists, do not follow this rule. We will claim that the subject (anti)agreement phenomenon is relevant for our analysis, because its development— even though it concerns the subject and not the object—can reveal aspects of the change in the system of Case. (20) tá ge díkaia pánta kalá estin. art.nom.pl prt just.nom.pl all.nom.pl fair.nom.pl be.3sg ‘All just things are fair.’ (Pl. Grg. 476b) Examples with this type of violation of agreement appear already in Homer. For instance, Scott (1929) shows that Iliad (in the Ludwich text) has a neuter plural subject with a singular verb 190 times (75.40), and with a plural verb 62 times (24.60). In Odyssey, a neuter plural subject is attested with a singular verb 185 times (76.76) and with a plural verb 56 times (23.24). Note that, according to Scott’s data, Homer prefers number nonagreement with nouns than with pronouns. Scott also observes a tendency for number nonagreement with verbs with active morphology but agreement with verbs with mediopassive morphology. Moreover, a single sentence in Homer can contain a neuter plural subject with a singular verb and a plural verb (due to metrical reasons, too; e.g. (21)). Animacy has also been considered a crucial factor for this (non)agreement preference: agreement is preferred with animate subjects (Kühner and Gerth 1898: 64ff.; Cooper 1998: 1015ff.; Widmer 2013).

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i



Nikolaos Lavidas

(21) álla mèn en khroì p¯e´gnut’ [. . . ] pollà dè other.nom.pl prt in body stick.ipfv.mp.3sg many.nom.pl prt [. . . ] hístanto stand.ipfv.mp.3pl ‘Some were driven deep in the bodies [of quick-stirring young men] while many [in the space between before they had got to the white skin] stood fast in the ground.’ (Hom. Il. 15, 315–17) Our second remark concerns two seemingly different phenomena that apparently demonstrate the opposition of the old (Pre-Koine) and new (Post-Koine) system, as described in our proposal. The presence of noncanonical accusative subjects underlines the end of the old system: examples from papyri and chronicles from Early Byzantine Greek include subjects of finite verbs bearing the accusative case (22a,b).21 These subjects in the accusative case are subjects of mediopassive finite verbs, whereas, in some cases, there is violation of number agreement between the verb and its subject. (22) Early Byzantine Greek a. taftin tin epistolin eghrafi en Thmui this.acc.sg art.acc.sg letter.acc.sg write.mp.pst.pfv.3sg in Thmui ‘This letter was written in Thmui.’ (P. Par. 18) b. on ta metra ke tas rel.gen.pl art.acc.pl metre.acc.pl and art.acc.pl jitnias dhia ton proktitikon adjoining-area.acc.pl by art.gen.pl previous-owner.gen.pl tetakte define.mp.pfv.3sg ‘the metres and the adjoining areas which have been defined by the previous owners’ (CPR 1 4) Furthermore, in accordance with our proposal, in Pre-Koine Greek, mediopassive verbs can take a direct object in the accusative, and mediopassive voice morphology can appear in a structure with features on Asp and interpretable aspectual features on the object. In Post-Koine Greek, only verbs with active voice morphology are allowed to have a direct object in the accusative, and the mediopassive voice morphology cannot appear in a structure with aspectual features on Trans and on the DP object ((23a) vs. (23b)):22 , 23 the active voice morphology of Post-Koine Greek demonstrates an Agree relation between the verb (Trans head) and the accusative object (the accusative case is a structural Case).

21 The noncanonical accusative subjects are not very productive or frequent; they appear mainly with mediopassive verbs but there exist examples with verbs with active voice morphology, as well. See Lavidas (to appear). On the other hand, their presence only in this stage is indicative for the significant change in the Case system of Greek. 22 We exclude deponents from the analysis; deponents are voice non-alternating types (they do not alternate between mediopassive and active voice). For details on the diachrony of deponents in Greek, see Lavidas and Papangeli (). 23 For a detailed discussion of the particular change, see also Lavidas ().

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

Case in diachrony: Or, why Greek is not English



(23) a. Pre-Koine Greek (example from Classical Greek) Mediopassive verb + DP-Acc D¯emokhár¯es [. . . ] ouk apokékruptai t¯e`n Demochares.nom.sg neg conceal.mp.pfv.3sg art.acc.sg ousían property.acc.sg ‘Demochares . . . has not concealed his property.’ (D., Against Aphobus, 2, 28, 3)24 b. Post-Koine Greek (example from Modern Greek) *Mediopassive Verb + DP-Acc *krivete tin periusia tu conceal.mp.prs.3sg art.acc.sg property.acc.sg 3sg.gen ‘He conceals his property.’ Our proposed analysis attributes the availability of noncanonical accusative subjects in the Early Byzantine examples to two significant changes in Greek that co-emerge: a change in voice morphology and a change in the Case system. The accusative case, not the nominative, is assigned in these examples, and, in parallel, there can be only partial agreement, with respect to the person but not to the number. Baker and Vinokurova (2010) and (Baker 2012, 2014, and subsequent work) have shown that there is no overt object agreement to mark the relationship between a DP in the accusative and any particular functional head in Sakha, which is the language they examine, or in any language in which the accusative is not assigned by a functional category.25 The obligatory active voice morphology of transitive verbs in Koine and Post-Koine Greek functions as a marker of an Agree relation between the DP object and Trans—the object case is assigned by the Trans (active) head; see Table 9.4.

. Conclusions We have shown that, even though the changes in aspect and definiteness are similar in Greek and English, the picture of the tendencies in Greek with regard to the Case system is quite different. We have put forward the hypothesis that in Pre-Koine Greek, Case can also be inherent (with inner-aspectual features), and there is no obligatory Agree between the object and the verb (a transitive verb can be active or mediopassive). In Post-Koine Greek, the accusative is only assigned through Agree with Trans, and Agree is manifest on that category through obligatory active suffixes at 24

This verb is a non-deponent one: apokrúpt¯o (active voice morphology ‘hide’). (i) [. . . ] ekeín¯en t¯e`n sophían apokrúptein that.acc art.acc wisdom.acc conceal.act.prs.inf ‘[. . . ] to obscure that wisdom’

(Pl. Ap. d–e)

25

Moreover, note that in Pre-Koine Greek the accusative can appear in complements of adjectives, in complements of nominalized verbs, or as the morphological case of subjects of infinitives. In this respect, see also Baker (): But accusative case, in addition to not being paired with visible agreement, appears in some surprising contexts: in passive clauses with a covert agent, in agentive nominalizations, and in subject raising constructions even when the matrix verb is not transitive.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i



Nikolaos Lavidas

Table .. Overview of the factors affecting the accusative subject examples → Pre-Koine Greek The accusative can also be related to inner-aspectual features/be an inherent Case (and depend on inner-aspectual features). → Examples with an accusative subject (evidence for the transition from the Pre- to the Post-Koine Greek Case system) The accusative can be reanalysed as a subject marker in the particular constructions (to ‘avoid’ structures with mediopassive verbs + accusative objects and violation of Agree). → Post-Koine Greek The accusative is only assigned by the (active) Trans head; active voice suffixes mark an Agree relation between Trans and object.

PF (mediopassive suffixes are ungrammatical with verbs that take accusative objects). In this respect, we have argued that the change that takes place in the history of Greek also involves a transitional stage in Ancient Greek where the relevant features can appear on Trans (outer aspect) or Asp (inner aspect). The completion of the change in Post-Koine Greek involves the loss of the interpretable feature on Asp; thus, Case remains (uninterpretable) on the higher Trans head. Our proposal is also supported by two relevant phenomena that mark the change from the Pre-Koine to the PostKoine Case system: in Pre-Koine but not in Post-Koine Greek, there are examples of violations of number agreement between the subject and the verb; and Pre-Koine but not Post-Koine Greek allows the presence of objects with mediopassive verbs.

Acknowledgements We would like to thank Éric Mathieu, Rob Truswell, the audience of the 15th Diachronic Generative Syntax conference (DiGS15), and two anonymous reviewers for very useful suggestions and comments. As usual, all errors remain ours.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

 Leftward Stylistic Displacement (LSD) in Medieval French M A R I E L A B E L L E A N D PAU L H I R S C H B Ü H L E R

Un coup de dés jamais n’abolira le hasard ‘A throw of the dice never will abolish chance’ Mallarmé, 1897, transl. Basil Cleveland

. Leftward Stylistic Displacement (LSD) Medieval French is generally considered an asymmetric V2 language, with V2 in main clauses and SVO in embedded clauses, except for a small list of complements which may be V2 (e.g. complements of bridge verbs). But in embedded clauses, some element (nonfinite verb, adverb, adjective, XP) may appear to the left of the fronted verb, as in (1)–(2). A similar word order pattern has been identified in other Medieval and Modern Romance languages (Benincà 2006; Cardinaletti 2003; Egerland 2011; S. Fischer 2010, 2014; Fischer and Alexiadou 2001; Fontana 1993; Franco 2009, 2012; Martins 2005, 2011). (1)

Qant levé furent del mangier when risen were from-the eat ‘When they had finished eating’ [Le Chevalier de la Charrette 1043, in Mathieu 2006b: (7a)]

(2) come cil font qui en queste doivent entrer like those do who in quest must enter ‘Like those who must start the quest do’ [La Queste del Saint Graal, in Mathieu 2006b: (1)] We call this leftward displacement Leftward Stylistic Displacement (LSD), an atheoretical term covering at least three configurations, as we will progressively see. In the present study, we examine LSD in Old French (9th c.–1350) and in Middle French (1350–1600), developing previous work by Labelle (2007, 2013). We mainly Micro-change and Macro-change in Diachronic Syntax. First edition. Éric Mathieu and Robert Truswell (eds). This chapter © Marie Labelle and Paul Hirschbühler 2017. First published 2017 by Oxford University Press.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i



Marie Labelle and Paul Hirschbühler

focus on nonfinite verb fronting, but will provide examples of other types of categories, to illustrate the fact that they behave similarly.1 It has been claimed that LSD is a case of Stylistic Fronting (henceforth SF) (Cardinaletti and Roberts 2002; Dupuis 1989; Mathieu 2006a,b, 2009, 2013; Molnár 2010; I. Roberts 1993; Salvesen 2011, 2013), a construction described initially for Icelandic (Maling 1980), and illustrated in (3). (3) Hver heldur þú að stolið hafi hjólinu? who think you that stolen has the-bike ‘Who do you think has stolen the bike?’

(Holmberg 2006: (10b))

Mathieu (2006a,b, 2013) provides a detailed analysis of the French construction. He argues, following Hrafnbjargarson (2004) on Icelandic and Old and Middle Danish, that, in Old French, two elements may be Stylistically Fronted (4), and he extends the domain of SF to main clauses, particularly when a verbal head is fronted (5) or when a V3 word order is observed (6). (4) Se lieve sus, et cil le voient / Qui [avoec lui] [aler] devoient refl get up and those him see who with him go must ‘He gets up and they, who should have gone with him, see him.’ [Le Chevalier de la Charrette, in Mathieu 2006b: (24b)] (5) Parlé as a ton amant spoken have to your lover ‘You have spoken to your lover.’

[Aucassin et Nicolete, in Mathieu 2006b: (32)]

(6) [Aprés la biere] [venir] voient une rote behind the coffin come see an escort ‘Behind the coffin they see an escort coming.’ [Le Chevalier de la Charrette, in Mathieu 2006b: (51)] We argue that Medieval French LSD has properties distinct from Icelandic SF, and that it covers not one but at least three distinct constructions: (1) a V2 construction, (2) an LSDLeft construction, with LSD to the immediate left of the subject, and (3) an LSDRight construction, with LSD to its immediate right. For embedded clauses in which the subject position is not filled, we propose that the LSD element occupies the unmarked LSDRight position. In Section 10.2, we summarize the most relevant aspects of Mathieu’s analysis, then we turn to a description of our data.

1 Unless otherwise indicated, the French examples are taken from the MCVF corpus (Martineau et al. ) and the Penn Supplement to the MCVF corpus (Kroch and Santorini ) ( verse and prose texts from the th to the th century, over ,, words). Using CorpusSearch (http://corpussearch. sourceforge.net), we extracted the sentences containing in preverbal position the typical elements characteristic of Stylistic Fronting, namely, nonfinite verbs (infinitives, past participles) and small adverbs (pas, point, molt, mie, mult, mut, bien, ben, tant, trop). See Table . in Section ...

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

Leftward Stylistic Displacement (LSD) in Medieval French



. Mathieu’s analysis In early work on SF, it was considered that the fronted constituent occupied the canonical subject position (Maling 1980), and Martins (2005, 2011) defends for Portuguese an analysis in terms of movement to (multiple) Spec,TP. But in recent years, several authors have argued that SF-like elements move to the left periphery: FocP (Benincà 2006; Hrafnbjargarson 2004) or a projection holding focus-like elements (S. Fischer 2014), Laka’s P projection expressing ‘emphatic affirmation’ (Fischer and Alexiadou 2001), FinP (Franco 2009, 2012; Salvesen 2011). Mathieu (2006a,b, 2009, 2013) follows this trend, but places the fronted element in a new specialized Top+P projection. Mathieu’s analysis of (6), shown in (7), exemplifies for him double SF of an XP in Spec,Top+P and an X0 in Top+0 . (7) [Aprés la biere] [venir] voient une rote Top+P Top+

Aprés la biere [−Foc]

Top+ [P][−Foc][uV] venir [V]

FinP Fin TP

Fin [uV] voient +[φ][V][D]

aprés la biere

T T [uφ][uV][uD][p] voient

VP voient venir une rote aprés la biere

For the left periphery, the structure is a variant of Benincà and Poletto (2004), adding a Top+ projection between TopP and FocP (Mathieu 2013: 17), but, for him, TopP hosts the initial constituent of a V2 clause, rather than left-dislocated elements. In V2 clauses the finite V is assumed to occupy Fin (cf. also Labelle and Hirschbühler 2005; Holmberg 2015), while in embedded clauses, it remains under T (except for V2 embedded clauses). The Top+P projection is said to host defocalized elements ([−Foc]). Mathieu (2006b: 246) argues that the [D] and [P] features of the traditional EPP feature of T, where [D] is responsible for subject agreement and [P] corresponds to the requirement of having a filled Spec, may be split between two heads. For SF of an XP, the [P] feature is on Top+. TP is considered a ‘strong phase’. In consequence, an XP must move through an empty Spec,TP to reach Spec,Top+P and check [P]. For SF of a head, Top+ must have some other relevant feature (for example, a [V] feature attracting a nonfinite verb, Mathieu 2006b: 246–7). A head moves directly to Top+0 over an eventual subject in Spec,TP (8).

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i



Marie Labelle and Paul Hirschbühler

peüst, / n’ issir]]] (8) [Pour ce qu’ [Top+P entrer [TP on n’ y enter one neg loc could, nor come-out for it that ‘so that one could not go in nor come out’ [1370, PRISE, .2738]2 The analysis offers a unified account for main and embedded clauses, and it makes clear predictions, namely, the presence of a subject should block the fronting of an XP, but not of a head, and at most two elements can be SF-fronted, in the order XP– X0 . It is claimed that SF disappeared when null subjects were no longer possible: the loss of rich verbal agreement led to the requirement to fill Spec,TP with a subject. In consequence, the [P] feature of T remained on T, and the escape hatch for SF of XPs is lost. This analysis faces both theoretical and empirical difficulties. On the theoretical side, splitting the EPP of T and migrating its [P] feature up to Top+ is an unexpected operation. One wonders why the [P] feature associated with Top+ must come from T0 , since [P] appears to correspond to the generalized EPP feature that every functional head may have. In addition, the [−Foc] feature of Top+ does not seem to play a role in triggering the fronting operation. Moreover, if TP is a strong phase and the finite V occupies T0 in embedded clauses, there is no escape hatch for heads. In embedded clauses, no head should be able to move directly to Top+ without passing through T. Thus, (1) for example (Qant levé furent del mangier) is predicted not to be possible since furent occupies T. The analysis of TP as a strong phase raises other questions. If Spec,TP is filled by a subject as in (9), wh-movement to the left periphery should be impossible, since the escape hatch Spec,TP is occupied, as shown in (9 ). (9) La [u cist furent], des altres i out bien there where those were, of-the others loc had many ‘Where those were, there were many others.’ [c.1100, ROLAND, 8.89] (9)

[ForceP u [tp cist furent u] ×

Some minor revisions of the analysis might solve the above problems, but we think that there are empirical problems with an SF analysis of LSD. Our argumentation will focus on four different points: (a) the informational role of the fronted constituent; (b) intervention effects; (c) the subject condition; and (d) the question of VP-fronting.

. Informational role of the fronted constituent For Icelandic, apart from Hrafnbjargarson (2004), who claims that the fronted expression is focused, most linguists consider that it has no specific informational role (e.g. Maling 1980; Platzack 1987; Holmberg 2006; Ott 2009; Sigurðsson 2010), but is ‘compatible with contrastive focus given the right context’ (Sigurðsson 2010: 180, fn.32). For Old French, S. Fischer (2014) and Fischer and Alexiadou (2001) argue that the fronted constituent is emphasized. By contrast, Mathieu (2006b: 247 and 2

Indications like ‘.’ correspond to the ID number of the example in the MCVF corpus.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

Leftward Stylistic Displacement (LSD) in Medieval French



fn.23) claims that Top+ hosts not foregrounded elements, but rather asserted background topics. This proposal is elaborated in two different ways. On the one hand, SF elements are said to provide non-presupposed asserted information, with the same semantics as appositives (Mathieu 2006b: 247, fn.23; Mathieu 2006a: 303, 2013: 342). Background here is opposed to foreground. On the other hand, they are said to be a means to defocalize a constituent (Mathieu 2006b: 221, 230, 248), to ‘make sure that the focused element in the clause is in final position’ (Mathieu 2013: 341). In this case, background contrasts with information focus. Fronting is also said to facilitate rhymes (Mathieu 2006b: 230, 248). As for the comparison with appositives, a comparison of (10) and (1) (Qant levé furent del mangier) shows that a fronted participle does not have the same semantics as appositive (relative) clauses. The participle phrase in (10) is truly an appositive, a parenthetical constituent contributing an assertion at a background level of predication; this role is clearly different from that of the LSD participle in (1). Consider also (11), where the LSD participle is inserted within an appositive. The appositive provides a background assertion with respect to the main clause; within the appositive, the LSD participle has a completely different role: it is not a parenthesis adding a second level of background information; rather, it provides the only information of the clause. (10) Brutus, [lyez pour la devant ditte victoire], enrichist ses compaignons Brutus, glad for the before said victory, enriched his followers de la despouille des mors of the belongings of-the dead ‘Brutus, glad for the above victory, enriched his followers with the belongings of the dead.’ [c.1330, PERCEFOREST, 81.335] (11)

David cunfortad sa muiller Bethsabéé, [ki deshaitéé fud] David comforted his wife Bethsabée, who suffering was ‘David comforted his wife Bethsabée, who was suffering.’ [c.1170, QUATRELIVRE, 80.3028]

As for the claim that fronting is a means to ‘make sure that the focused element in the clause is in final position’ (Mathieu 2013: 341), consider the case where a participle is fronted. If the participle is the only informative element in the clause, the only element carrying new information focus, as in (11), it would be informationally inappropriate to defocalize it in order to place the auxiliary in focus position. If, on the other hand, there is material after the participle, it would remain in focus after fronting the participle, as would be the case in (1) Qant levé furent del mangier. Moreover, in (1), del mangier is background information, and Qant levé furent is the information focus, the new information.3 Therefore, fronting the participle neither focalizes del mangier, nor does it ensure that levé is not part of the information focus (as suggested by Elsig, p.c.). 3

(i)

Et mangierent ansanble et burent / Tant que del mangier lever durent. / and ate together and drank until that of the eating get up had to Qant levé furent del mangier, . . . when got up were of the eating ‘They ate and drank together until they had to get up from the table. When they had left the table, . . . ’ (l.–)

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i



Marie Labelle and Paul Hirschbühler

In (12), a first conjunct is fronted while the rest of the constituent remains in place; both parts form a discontinuous informational focus, and the first one is not defocalized to allow the second one to be focused. (12) Dreit les meinet a un castel / Qui riches ert e grant e direct them leads to a castle which rich is and large and bel beautiful ‘He leads them directly to a castle which is rich and large and beautiful.’ [c.1120, BRENDAN, 37.159] About half of the collected examples of fronted participles and a third of fronted infinitives are from prose texts, therefore LSD was productive in Medieval French. While there is no doubt that fronting an element may often be especially useful in verse texts (just as preposing in V2), in many cases, the fronted element does not come from a line-final position and its displacement has no effect on rhyme, as in (13), where plus cannot possibly be defocalized. (13) Quan Rollant veit la contredite gent, / Ki plus sunt neirs when Roland sees the accursed people, who more are black que nen est arrement . . . than not is ink ‘When Roland sees the accursed people, who are more black than ink . . . ’ [c.1100, ROLAND, 144.1952; v. 1933] Versification exploits the possibilities of the language and we thus might expect a higher percentage of LSD in verse than in prose texts for metrical reasons. The present corpus does not allow us to estimate this effect, since this factor is confounded with time: all but one of the verse texts are anterior to 1220 in the corpus, while all the prose texts but one are posterior to 1200. A valid comparison between the two genres would require a corpus with more verse texts for the post-1200 period (no other sizeable prose documents exist for the pre-1200 period). In the collected data, the LSD expression has no consistent informational role. It may correspond to the information focus of the clause (or part of it), in main clauses (14) and in embedded clauses (e.g. (11)), or to already known background information (15)–(16). (14) Quant sunt refait, levét s’ en sunt when are refreshed, got-up refl gen are ‘When they are refreshed, they got up.’ [c.1120, BRENDAN, 49.489] (15) é distrent entre sei: ‘Mened únt l’ arche jesque á nus and said among themselves ‘Carried have the ark until to us pur nus ocire . . . ’ to us kill’ ‘and they said to each other: “They have carried the ark up to our place to kill us . . . ” ’ [c.1170, QUATRELIVRE, 13.372]

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

Leftward Stylistic Displacement (LSD) in Medieval French



(16) Si cururent par quinze jurs / Desque li venz tuz lur fud gurz prt sailed for fifteen days until the wind all them was weak Dunc s’ esmaient tuit li frere / Pur le vent qui falit ere. then refl frighten all the brothers for the wind that stopped was ‘They sailed on during fifteen days until the wind calmed on them. Then, all the brothers got frightened because of the wind that had stopped blowing.’ [c.1120, BRENDAN, 36.125, v. 219–222] In addition, in main clauses, the LSD element may be strongly (contrastively or emphatically) focalized, as in (17)–(19). (17) Se fuït s’ en est Marsilies, Remés i est sis uncles if fled refl gen is Marsilie, remained loc is his uncle Marganices, Marganice ‘If Marsilius has fled, his uncle Marganice stayed there.’ [c.1100, ROLAND, 143.1937] (18) Vous m’ avez mort sans recouvrier, mais vaincu ne m’ avez you me have dead without recourse, but defeated neg me have vous mie. you a.bit ‘You killed me without recourse, but you have not defeated me.’ [Merlin en prose, ii. p.52, in Rickard 1962: 19] (19) car par le conseil Joseph avoit il recovree sa terre que since through the advice Joseph had he recovered his land that Tholomers li toloit, et tolue li eust il se . . . Tholomers him was.taking, and taken him would-have he if ‘since through Joseph’s advice, he had recovered his land that Tholomers was taking from him, and he would have taken [it] from him if . . . ’ [c .1225, QUESTE, 113.2974] We conclude that LSD is not associated with a specific informational role, and it is not a means to defocalize an element.

. Intervention effects In Icelandic, if many elements are possible targets of SF, only the highest may be fronted. In particular, SF of X0 or XP is blocked by the negative adverb or by a sentence-medial adverb ((20)–(23); Holmberg 2006: 546, (39a–d)). (20) þeir sem hafa ekki búið í Ósló those that have not lived in Oslo

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i



Marie Labelle and Paul Hirschbühler

(21)

þeir sem ekki hafa búið í Ósló

(22) *þeir sem búið hafa ekki í Ósló (23) *þeir sem í Ósló hafa ekki búið In Medieval French, while a negative adverb like pas/point/mie may front (e.g. Mathieu 2006b: 232), neither heads nor XPs are blocked by its presence (Labelle 2007).4 Notice that in (24), pas, syntactically and semantically associated with the higher verb, should be more accessible than the fronted element, which is part of the lower VP.5 The relevant part of the structure of (24a) is (24b). (24) a. que desreer ne se vost pas as premiers cos, ne since act-wildly neg refl wants not at-the first blows, nor angoissier. trouble ‘since he did not want to lose control at the first blows, nor to get troubled’ [c.1170, YVAIN, 136.4744] b. [desreer] [T0 ne-se-vost] pas [vP1 vost [vP2 se desreer as premiers cos ne angoissier]] (25) Huelin dist une noevele / qui a Gorm[un]d ne fut pas Huelin said a piece-of-news which to Gormund neg was not bele good ‘Huelin brought news which didn’t please Gormund.’ [c.1130, GORMONT, 239–240] Mathieu (2013: 338) argues that (25) is not a problem for his analysis, because pas can front as a head. However, since ekki blocks SF of both X0 s and XPs, if (24) and (25) are cases of SF, another explanation must be given for why negative adverbs like pas, point, and mie do not block LSD of a lower expression. In addition, LSD of a participle is not blocked by a high adverb: (26) car recuvré sunt veirement . since found are certainly ‘since they have certainly been found’ [c.1170, QUATRELIVRE, 18.590] (27) car demuret i unt trop since stayed loc have too-much ‘since they stayed there too long’

. [c.1100, ROLAND, 136.1832]

We conclude that the accessibility constraints typical of Icelandic SF are not operative in Medieval French.6 4

See ()–() for examples from other Old Romance languages. The embedded VP may be a subpart of a full embedded CP. 6 Old French also contrasts with Icelandic in the possibility of fronting secondary predicative adjectives: Blanche avait la barbe ‘white had the beard’ vs. *maðurinn sem rikur er talinn ‘the man that rich is considered’ (Holmberg : (c)). 5

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

Leftward Stylistic Displacement (LSD) in Medieval French



. Subject condition In Icelandic, SF is subject to an empty subject condition (Holmberg 2006: (10b), (11b–c)): (28)

Hver heldur þú að stolið hafi hjólinu? who think you that stolen has the-bike ‘Who do you think has stolen the bike?’

(29) *Hvaða hjóli heldur þú að stolið hann hafi? which bike think you that stolen he has ‘Which bike do you think that he has stolen?’ (30) *. . . að hann stolið hafi? that he stolen has This constraint is strict when the subject is an XP. However, Hrafnbjargarson (2004: 117, (79b)), contrary to other Icelandic linguists, marginally allows SF of a head to the right of the subject in Icelandic, provided it is a weak subject, argued to be cliticized on the complementizer (31).7 (31) ?Allt sem ’ann lesið hafði i bókinni war satt all that he.weak read had in book-the was true ‘Everything that he had read in the book was true.’ Mathieu (2006a,b, 2013) stresses the importance of an empty subject in Old French as a condition for fronting an XP, and he argues that the construction disappeared when the verb’s agreement no longer carried pronominal properties, a condition for null subjects (Mathieu 2006b: 258). .. Link with the disappearance of null subjects S. Fischer (2014) casts doubts on the hypothesis that there is a link between the disappearance of null subjects and that of Stylistic Fronting (LSD for us) by pointing out that both Old Spanish and Old Catalan lost nonfinite verb fronting, while continuing to be null subject languages. In Mathieu’s analysis, only SF of XPs should be affected by the disappearance of null subjects; SF of a head should continue to be possible, since heads do not pass through the subject position and are not dependent on a [P] feature on Top+0 . But Figure 10.1 (with percentages on the vertical axis and text date on the horizontal axis) shows that, if we consider embedded clauses with the finite verb in second position in prose texts: 1) the rate of null subjects remains constant, while 2) the rate of fronted nonfinite verbs and of adjectives (X0 s as well as XPs), predicates, and quantifiers decreases at the same rate. This suggests that there is no relation between the disappearance of embedded XV (at least for X0 ) and the disappearance of null subjects.

7 Sigurðsson (: , fn.) does not accept this construction and he points out that neither do other Icelandic linguists, and that he has never come across an authentic example.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i



Marie Labelle and Paul Hirschbühler

10 9 8 7 6 5 4 3 2 1 0 1150

1200

1250

Pct NullS nf V

1300

adj,prd,q

1350

1400

Linéaire (Pct NullS)

1450

1500

Linéaire (nf V)

1550

1600

Linéaire (adj,prd,q)

Figure .. Percentage of null subjects, and of LSD of nonfinite V and adjectives/predicates/ quantifiers in XV embedded clauses, prose texts only 100 90 80 70 60 50 40 30 20 10 0 950

1050

1150 Pct LSD/Total

1250

1350

1450

Linéaire (Pct LSD/Total)

1550 R2 = 0,47866

Figure .. Percentage LSD in Subject Relatives (N = 3690)

In addition, we might predict a similar rate of LSD in subject relatives and in null subject thematic embedded clauses, since in both cases the subject position is empty. But LSD in subject relative clauses (Figure 10.2) decreases at a regular rate during the period, while in null subject complement clauses (Figure 10.3, p. 155), the rate of LSD remains constant during the period. We conclude that there does not seem to be a link between the disappearance of LSD and that of null subjects. We will argue that there is in fact no subject condition in Medieval French. .. Absence of a subject condition in Medieval French For Medieval French, the distribution of fronted infinitives, participles, and small adverbs in XV sentences is given in Table 10.1. Fronting of these elements is found mostly with a null subject (XV∅) or a postverbal subject (XVS), but fronting to the left of a preverbal subject (LSDLeft ) or to its right (LSDRight ) is attested in over 10 per cent of the cases, both in matrix and in embedded clauses. Here again, French LSD differs from Icelandic SF. As we saw in Section 10.2, Mathieu’s analysis allows SF of a head to the left of the subject in Old French. This word order is the opposite of the one in (31), marginally

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

Leftward Stylistic Displacement (LSD) in Medieval French 100 90 80 70 60 50 40 30 20 10 0 950

1050

1150

1250 Pct LSD

1350

1450

Linéaire (Pct LSD)



1550 R2 = 0,00288

Figure .. Percentage LSD in thematic null subject embedded clauses (N = 991)

Table .. Fronted infinitives, participles, and small adverbs

Infinitives Participles Sm.adverbs

Infinitives Participles Sm.adverbs

XV∅

XVS

274 340 881 73

23a 117b 204e 17

XV∅

XVS

437c 537d 624 83

6a 42b 30 4

Matrix clauses LSDLeft LSDRight 8 1 24 2

43 37 96 9

Embedded clauses LSDLeft LSDRight 10 18 42 4

72 71 35 9

LSDL + LSDR 15 8 10

LSDL + LSDR 16 13 11

a

Inf V S: Matrix: 21 before 1225 (6 weak pron S); Embedded: all before 1200 (2 weak pron S) b Prt V S: Throughout the period c Inf V ∅: Embedded: 144 trace, 293 null d Prt V ∅: Embedded: 225 trace, 312 null e Germanic inversion (Adv Aux S V): 26; Romance inversion (Adv Aux V S): 9. The rest are ambiguous.

accepted by Hrafnbjargarson (2004). As shown in Table 10.2, in Medieval French, LSDLeft is much rarer than LSDRight , and it is almost nonexistent in Old French; it is essentially a Middle French phenomenon. We found the LSDLeft order practically exclusively with pronominal subjects, as in (32) and (8), but there are some exceptions, for example (33), with a nonfinite V, and (34) with an adverb. Both are Middle French examples.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i



Marie Labelle and Paul Hirschbühler

Table .. SXV vs. XSV in the parsed corpus (where X = nonfinite verb (phrase) or small adverb) Main clauses

Old French

Middle French

Embedded clauses

LSDLeft

LSDRight

LSDLeft

LSDRight

< 1150 1150–1199 1200–1249 1250–1299 1300–1349 1350–1399 1400–1449 1450–1499

3 5 1 1 0 13 1 8 1

32 64 22 9 0 43 0 6 0

0 1 2 1 1 50 0 12 3

10 98 15 22 0 20 0 13 0

Totals

33

176

70

178

16th century

(32) Et trop grant temps avoit que point il ne s’ estoient veu and too-much long time had that not he neg refl were seen en parti de bataille avoir in part of battle have ‘And there had been too much time since they had seen themselves to the point of having a battle.’ [1373, FROISSART, 724.15707] (33) Il feïrent une ordenance / Qu’ esleüs xij. homes seroient / Qui they made a decree that elected twelve men would-be who le païs gouverneroient, the country would-govern ‘They made a decree that twelve men would be elected to govern the country.’ [1370, PRISE, .4632] (34) Et se bien li nostre assailloient, / Li autre mieus se and if well the ours attacked the others better refl deffendoient. defended ‘And if our men attacked well, the others defended themselves better.’ [1370, PRISE, .1453] By contrast, the LSDRight order is unmarked and is regularly found with nominal and pronominal subjects: (35) Eüstaces mot n’ en savoit / De ço que Dex sauvé avoit / ses Eustace word neg gen knew of this that God saved had his effanz ... children ‘Eustace did not know a word of it that God had saved his children.’ [c.1212, EUSTACE-FISHER, 31.370]

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

Leftward Stylistic Displacement (LSD) in Medieval French



(36) por savoir s’ il verroit riens de la chose qu’ il plus to know if he would-see anything of the thing that he most desirroit. desired ‘to know if he would see anything of the thing that he most desired’ [c.1225, QUESTE, 82.2217] This LSDRight word order is not only more frequent in Medieval French; it is also attested cross-linguistically. For medieval Romance, examples exist in Old Spanish (37), in Old Catalan (38), and in Old Portuguese ((39), from Martins (2011), who notes that the OS (LSDLeft ) and SO (LSDRight ) orders are possible for what she calls ‘IP scrambling’, the SO order being more common). (37) E pues yo gradescer non vos lo puedo and since I thank not you it can ‘And since I cannot thank you for it’ [Libro del caballero Zifar, p.362, in Rivero 1992: 265] (38) E que aviem guarit que nos dar non.o deviem and that had.1pl considered that we give not.it must.1pl ‘And that we had considered that we did not need to give it’ [XIII, Carta del Bisbe d’Urgell, 1257–9, in Russell-Gebbett 1965: 109, quoted from Batllori et al. 2005: (70a)] (39) de quem.quer que uos alg˜uu enbargo sobrel quiser põer from whoever that you some obstruction over-it wants put ‘[protecting you] from whoever tries to block it [the land] from you’ [Legal document, 1381, in Martins 2011: 146 (22a)] The LSDRight word order is also attested in Old Icelandic (S. Fischer 2014), Old English (S. Fischer 2010), and modern Italian (Cardinaletti 2003). To summarize, in Medieval French, while LSD is favoured by an empty subject position, it is attested with a filled subject over 10 per cent of the time with nonfinite Vs or VPs and small adverbs. Among LSDLeft and LSDRight , the unmarked order is LSDRight . We conclude that the subject condition does not exist in Medieval French LSD (a point also made by S. Fischer 2010).

. VP and remnant VP movement Mathieu assumes that there is head movement of the participle in examples like (40) where a verbal head is fronted and a complement is left behind (see (6) for a main clause example). However, a Remnant VP Movement analysis along the lines of den Besten and Webelhuth (1987, 1990) is also possible, as shown in (41). The object is extracted out of the VP and the remnant VP is fronted (we keep the traditional VP label, though the movement might be more precisely that of vP, as in Salvesen 2011, 2013). A Remnant Movement approach is defended by Salvesen (2011, 2013) for Old French and by a number of other authors for other languages (Holmberg 2006;

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i



Marie Labelle and Paul Hirschbühler

Ott 2009 for Icelandic; Franco 2009, 2012 for Old Florentine). To make the structures more readable, we will not represent the traces within the fronted VPs. (40) sil fiert en l’ elme, ki gemmet fut a or; prt-him struck on the helm, that jewelled was at gold ‘He struck him on the helm, that was ornate with gold.’ [c.1100, ROLAND, 170.2306] (41) . . . [VP1 gemmet]k fut [VP2 [a or]i [VP1 gemmet a or]] Mathieu (2006b) argues against Remnant VP Movement in Old French on the grounds that Old French did not allow V2 topicalization of verbs or VPs (p.226) and VP is not a category that is among those that can be SFronted (p.241). As for V2 ‘topicalization’ of verbs, we will argue in Section 10.7.1 that it is possible. In the next section we show that VP-fronting is attested. .. VP-fronting in OF VP-fronting is attested in various forms in Medieval French, including LSD of a VP with a VO word order (an X0 –XP word order not predicted by double-SF).8 The word order in the fronted VP in (42) shows that we do not have an OV base here. (42) . . . Einsi-comme apres le sarez, Quant [bien leü ce livre] as later it will-know, when well read this book arez. will-have ‘. . . you will know later, when you will have read this book well’ [1370, PRISE, .135] (43) Vos savez bien que [desguarni / De touz avoirs] sonmes ichi you know well that stripped of all belongings are here ‘You know well that we are here stripped of all assets.’ [13th c., Eustace-Petersen, 27.366] Other cases of fronted vPs are illustrated in (44)–(45). In (44), an infinitival clause introduced by de occupies the first position of a V2 clause; in (45), it appears to the left of the subject. (44) Ou que il seit, [de Deu servir] ne cesset. where that he be, to God serve, neg ceases ‘Wherever he is, he does not cease to serve God.’

[c.1090, ALEXIS, 17.174]

(45) Seignors, qui Dex aïdier weult, / [De riens grever] nus ne Lords, who God help wants, of anything harm nobody neg , le puet him can ‘Lords, who God wants to help, nobody can harm him in any way.’ [13th c., EUSTACE-PETERSEN, 56.762] 8

A similar word order is attested in Old Spanish (Rivero : ).

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

Leftward Stylistic Displacement (LSD) in Medieval French



Table .. Percentage of VO word order w.r.t. OV order in LSD with a participle, in cases where the fronted (remnant) VP contains more than a head Date

Number VO

12th c. 13th c. 14th c. 15th c. a b

8 4 19 1

Tot VO+OV (not ambiguous) 81 14 32 4

 VO

Ambiguous casesa

9.9 28.6 59.4 25.0

46 9 31 3

Grand totalb

127 23 63 7 220

Ambiguous cases: LSD constituents larger than a head not containing an object (e.g. constituents of type [Adv V]). Total: Total number of LSD of a VP containing more than a verbal head.

We conclude that there is vP movement to the left in Medieval French. Next, we show that assuming vP fronting and short scrambling we can account for word orders that are different from the XP–X0 order predicted by Mathieu. ... Short scrambling In Medieval French an object may appear to the left of a nonfinite verb. This movement, called ‘Short Scrambling’ by Martins (2011), is analysed by her as adjunction to VP (see also Salvesen 2011, 2013 and Zaring 2011, for Old French; Poletto 2006 for Old Italian). In Medieval French, short scrambling is found with selected and nonselected complements, definite and indefinite, including nonspecific ones. Examples (44) and (45) above illustrate the OV order in nonfinite clauses.9 The AuxXV order is shown in (46)–(47). At this point we remain agnostic as to what triggers this movement. (46) et dist a son seignor qu’ il en avoit .ii. autres gaangniez, and told to his lord that he gen had two more won ‘And (he) told his lord that he had won two more of them.’ [1225, QUESTE, 84.2307] (47) E al ceval a l’ eschine trenchee: and to-the horse has the spine cut ‘And he cut the spine of the horse.’

[c.1100, ROLAND, 107.1385]

This AuxXV order does not correspond to a VP with an underlying OV order. An object may remain after the V (48), and (49) shows that the object of the infinitive may appear to the left of the participle that governs the infinitive. (48) . . . irez . . . de çó qu’ il ourent David eslit á rei é á seignur. angry of it that they had David elected at king and at lord ‘. . . angry . . . for the fact that they had elected David as king and as lord’ [c.1150, QUATRELIVRE, 63.2384] 9

A similar word order is given in Batllori et al. (: (d)) for Old Catalan.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i



Marie Labelle and Paul Hirschbühler

, ... (49) Et puis que Diex vos a ceste honor apareilliee a avoir and since God you has this honour prepared to have ‘and since God prepared you to have this honour . . . ’ = Diex vos a [ceste honor [apareilliee vos [a avoir ceste honor]]] [c.1225, QUESTE, 95.2602] These examples show that there is short scrambling in Medieval French. .. Combining VP-fronting with short scrambling We have seen at the beginning of this section that examples with a fronted participle may be analysed as resulting from Remnant VP Movement. A VP movement analysis can be extended to cases which Mathieu analyses as Double SF. Thus, (50), analysed by Mathieu as in (51), can instead result from fronting a VP containing a fronted object, as in (52). (50) Cele dame une fee estoit / Qui l’ anel doné li avoit that lady a fairy was who the ring given to-him had ‘That woman was a fairy who had given him the ring.’ [Le Chevalier de la Charrette, in (Mathieu 2006b: (24c))] (51) [qui [Top+P l’anel [Top+ doné [TP li avoit doné l’anel]]]] (52) qui li avoit [VP2 l’anel [VP1 doné l’anel]] → qui [VP2 l’anel doné] li avoit In Remnant Movement, the inner VP is fronted, that is, the VP from which the nonV material has been extracted (here, it would be VP1 ). But in (52), the outer VP is fronted, that is, VP2 .10 In some cases, as in (53), after the inner VP1 is fronted, an object is extracted out of it. (53) Apres tout ce, congié pris a / Dou duc & de la after all that, leave taken has of-the duke and of the compaingnie. company ‘After this, he took leave of the duke and of the company.’ [1370, PRISE, .758] = a pris congié dou duc . . . → a [VP2 [dou duc . . . ] [VP1 pris congié dou duc . . . ]] → [VP1 pris congié] a [VP2 [dou duc . . . ] [VP1 pris congié]] → [VP3 congié [VP1 pris congié]] a [VP2 [dou duc . . . ] [VP1 pris congié]] A combination of VP-fronting with short scrambling accounts, in addition, for the XP–XP word order in (54), structurally similar to (52). (54) Se [trestout ce] [bien gardé] unt if all this well kept have ‘if they have well kept all of this’

[1190, BORON, 88.1396]

10 While Hrafnbjargarson () accepts double SF for two adverbs, an adverb and a participle, or a weak subject and an adverb, Holmberg (: ) states that ‘Only one category per clause can be moved by SF’, and Wood (: fn.) observes that a verb and its complement cannot both be fronted.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

Leftward Stylistic Displacement (LSD) in Medieval French



= se unt [VP bien gardé trestout ce] → se unt [VP2 trestout ce [VP1 bien gardé trestout ce]] → se [VP2 trestout ce bien gardé] unt [VP2 trestout ce bien gardé] The V1 V2 word order provides evidence for Remnant VP movement, particularly when a complement is left behind, as in (55). A reader asked why devant sen veissel must be scrambled in this example, as it is in final position in both the initial and in the final representations. This scrambling allows the displacement of a VP constituent containing only aler ourer. (55) Dusqu’a un jour [qu’ alez ourer / Fu Joseph devant sen veissel]; until a day that gone pray was Joseph in-front-of his vase ‘until a day when Joseph went to pray in front of his vase’ [c.1190, BORON, 100.1595] = que fu Joseph [alez [ourer devant sen veissel]] → que fu Joseph [[devant sen veissel] [alez [ourer devant sen veissel]]] → que [alez [ourer]] [fu Joseph [[devant sen veissel] [alez [ourer]]]] The V3 V2 word order in (56) requires movement of V3 over V2 , and the XP–V2 –V3 word order in (57) can only result from a combination of short scrambling and VPfronting. Given the order of the fronted verbs, it cannot be a verb-final construction. (56) Si com dire oy l’ ai celi / Qui y estoit; as like say heard it have the-one who loc was ‘As I heard it said by someone who was there’ [1370, PRISE, .1680] = Si com l’aiV1 [VP [oyV2 direV3 ] celi qui y estoit]. → Si com [oyV2 direV3 ] l’aiV1 [VP [oy dire]] [celi qui y estoit]] → Si com [direV3 oyV2 direV2 ] l’aiV1 [VP [oy dire] [celi qui y estoit]] (57) Un sydoine feit feire avoie a shroud made make had ‘He had a shroud made.’ [c.1190, BORON, 55.850] = avoieV1 [feit V2 feireV3 un sydoine] → avoieV1 [un sydoine [feit V2 feireV3 un sydoine] → [un sydoine feitV2 feireV3 ] avoieV1 [un sydoine feit feire] To summarize, we have shown not only that (remnant) verb phrase movement is possible in Medieval French, but also that it is necessary to assume both short scrambling and (remnant) movement to account for a number of Medieval French examples.

. Positions of the LSD constituent In this section, we argue that three constructions must be distinguished: V2, LSDRight , and LSDLeft . We assume the structure in (58). (58) [ForceP que [FrameP (HT/ST) [TopP LD [FocP Emph/Contr/wh-Op [FinP V [SubjP [TP . . . ]]]]]]] (HT = Hanging Topic; ST = Scenic Topic; LD = Left-Dislocated)

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i



Marie Labelle and Paul Hirschbühler

This combines aspects of Benincà and Poletto’s (2004) structure for the left periphery, Holmberg’s (2015) analysis of V2, with the finite verb under Fin in main clauses, and Rizzi and Shlonsky’s (2007) structure, with the subject in SubjP followed by TP. TP dominates the finite verb in embedded clauses. .. V When a fronted element is found together with a postverbal subject, there is no reason to think that it is not a V2 construction. This is so even when the fronted element is a participial head. Indeed, Remnant VP Movement was developed by den Besten and Webelhuth (1987, 1990) to account for the existence of participles in the first position of a V2 clause in German. (59) Beneit seiez vus de nostre Seignur blessed be you from our Lord ‘May you be blessed by Our Lord.’ (60)

[c.1170, QUATRELIVRE, 63.2370]

Certes, fet il, ma dolce amie, / [morir n’ i voldroie indeed, says he, my sweet friend, die neg loc would-want je mie] I not ‘Indeed, says he, my dear friend, I would not want to die there.’ [c.1170, YVAIN, 48.1634]

Assuming remnant VP movement to Spec,FinP, the derivation of (59) is as in (61). The PP object of beneit is first adjoined to VP, and the remnant VP is fronted to Spec,FinP. (61) [FinP [VP1 Beneit] [Fin seiez [SubjP vus [TP seiez [vP vus seiez [VP2 [de nostre Seignur] [VP1 beneit]]]]]]] When the participle is contrastively focused, as in (17)–(19), it could be in FocP (cf. Frey 2004, 2006a for ‘True A-bar movement’ in German): (62) mais vaincu ne m’ avez vous mie but defeated neg me have you not ‘But you have not defeated me.’

(see (18) above)

Notice that, in (62), the postverbal subject precedes the negative adverb mie, which indicates that it occupies Spec,SubjP. The Participle–Finite V–Pronominal Subject word order is not observed in embedded clauses. Since pronominal subjects are never lower than the highest subject position below the left periphery (Vance 1997), that is Spec,SubjP in the present analysis, it can only result from movement of the V and the participle to the left periphery. This word order could exist only in V2 embedded clauses. .. LSDRight Examples of LSDRight were given in (35)–(36), where it was clear that the LSD element (noun, participle, adverb) follows the subject. We argue here that this LSD position

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

Leftward Stylistic Displacement (LSD) in Medieval French



is internal to TP, rather than being within the left periphery, and that this is the unmarked position even when no subject is present. In embedded questions, the LSD expression follows the wh-operator. Assuming that the wh-Op occupies Spec,FocP, the structure must be [TopP [FocP wh-Op [. . . LSD . . . ]]], with LSD to the right of FocP (and not to its left as in Mathieu 2013: 345). (63) . . . Conter . . . / [En queu terre] [aler] le couvint tell in which land go him was-needed ‘. . . tell . . . in which land he had to go’ [c.1190, BORON, 121.1891] (64) . . . Se c’ estoit voirs [que] [dist] avoit / Dou prophete if it was true what said had of-the prophet ‘. . . if what he had said of the prophet was true’ [c.1190, BORON, 59.912] Since the wh-expression is clearly within the left periphery, we do not expect another A-bar-moved constituent to intervene between it and its trace, which would be the case if the LSD was also in the left periphery (Poletto 2005). There is in fact no need to assume that the LSD is in the left periphery since, in embedded questions, the finite V is under T0 and not under Fin0 . On the contrary, we will see that there is evidence that the LSD occupies a functional position between Subj0 and T0 . In (65) En queu terre could move independently of aller or, as suggested by Salvesen (p.c.), as part of aller en queu terre in the FP position before being extracted to the left periphery. (65) . . . Conter . . . [FocP En queu terre [SubjP [FP aler [TP le couvint]]]] First, the LSDRight construction is the unmarked one, and we take the existence of examples such as (35), repeated here as (66), to indicate that the subject occupies its canonical position, followed by the LSD element. (66) que [SubjP Dex [FP [sauvé] [TP avoit [ses effanz [sauvé ses effanz]]]]] (from (35)) Secondly, in embedded questions, the subject intervenes between the wh in FocP and the LSD: [SubjP vous [FP cest suaire (67) pour Dieu nous dites [FocP Ou where you this shroud for God us tell [TP preïstes]]]]. took ‘For God, tell us from where you took this shroud.’ [c.1190, BORON, 55.847] Thirdly, the LSD element does not interfere with either short-distance or long-distance wh-movement, as expected if it is not within the left periphery. We have just discussed examples with short distance wh-movement (e.g. (67)). Long-distance movement of en sarraguce to a focus position, in the spirit of Chomsky (1977), is illustrated in (68). (68) en sarraguce sai ben [qu’ aler m’ estoet ] in Sarraguce, know well that go me is-necessary ‘I know well that I must go to Sarraguce.’ [c.1100, ROLAND, 23.283]

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i



Marie Labelle and Paul Hirschbühler

Fourthly, if qui in subject relatives occupies Fin0 (Rizzi 1997) or if the -i of qui occupies the canonical subject position (Taraldsen 2001) or Fin0 (Rizzi and Shlonsky 2007), the LSD which follows qui must occupy a position within SubjP rather than within the left periphery: (69) come cil font [FinP qui [SubjP [[en queste] [TP doivent like those do who in quest must entrer en queste]]]] enter ‘like those who must start the quest do’ [La Queste del Saint Graal, in Mathieu 2006b: (1)] Finally, in the following example, there are two displaced elements: Aprés Rollant to the left of the complementizer que, and whose position is unclear, and vive between the pronominal subject and the finite verb: (70) Ne place Deu ne ses seinz ne ses angles / [Aprés Rollant que jo neg please God nor his saints nor his angels after Rolland that I vive remaigne]! alive remain ‘God forbid, and his saints and his angels, that I remain alive after Roland!’ [c.1100, ROLAND, 268.3709] Therefore, in the examples discussed, the LSD element occupies a position within SubjP to the right of Subj0 . The wh–LSD order is also attested in a handful of main exclamatives and questions (two examples anterior to the thirteenth century): (71) A quel dolur deduit as ta juventa! to which suffering live-out have your youth ‘To which suffering have you lived out your youth!’

[c.1090, ALEXIS, 91.821]

(72) Tu, pourquoi feit l’ as? you, why done it has ‘You, why did you do it?’

[c.1190, BORON, 12.178]

Since these are main wh-clauses, the finite V is under Fin0 . If movement to Spec,FinP and movement to Spec,FocP are both A-bar movements, we might expect the two operations to be excluded by minimality, as mentioned above. This consequence is avoided if the VP is raised to Spec,FinP, and if the wh-expression is further extracted from it to Spec,FocP. If this is correct, (71)–(72) are not examples of the LSDRight construction: (73) [HT/TopP Tu [FocP pourquoi [FinP [VP fait pourquoi] [Fin l’as] [SubjP l’as fait pourquoi]]]] Another possibility suggested by Salvesen (p.c.) is that pourquoi is base-generated in an IntP position to the left of FocP, as proposed by Rizzi (2001), in which case this could be LSDRight , with the finite verb under T and the participle in a functional LSD position to its left.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

Leftward Stylistic Displacement (LSD) in Medieval French



LSDRight is also found in main clauses, with a subject neither left-dislocated nor topicalized: (74) . . . car il dist ha / Qu’ au tierz jour resuscitera; since he said has that at-the third day will-resurrect ‘since he said that he will resurrect on the third day’ [c.1190, BORON, 17.269] (75) Car le pueples [assez a faire] ha since the people enough to do has ‘since the people have enough to do’

[1370, PRISE, .2197]

If, following many authors (e.g. Travis 1991; Zwart 1993; Vance 1997; Hirschbühler 1995), the subject in SVO main clauses is not raised to the left periphery, as expected under Rizzi’s view that the subject position is criterial (Rizzi 2006; Rizzi and Shlonsky 2007), this construction is the LSDRight construction. .. LSDLeft Examples with a nonverbal LSD in pre-subject position in embedded clauses were given in (32)–(34). In some cases, for example (34) repeated below as (76), the LSD element is contrastively focalized. (76) Et se bien li nostre assailloient, / Li autre mieus se and if well the ours attacked, the others better refl deffendoient. defended ‘And if our men attacked well, the others defended themselves better.’ [1370, PRISE, .1453] With respect to LSD of verb phrases, LSDLeft in main clauses seems often to correspond to movement to a foregrounding position (Foc or Top): (77) Prendre si tost jo vus defent. take so early, I you forbid ‘I forbid you to take [this water] as for now.’ = [FocP Prendre si tost [SubjP jo vus defent]]

[c.1120, BRENDAN, 647]

Movement of a VP to a Top position is illustrated in (78), where the VP precedes an interrogative in FocP. (78) Se mes sires riens demander / M’ en vouloit ne achoisonner, / if my lord anything ask me gen wanted nor accuse [Respondre de ce] que pourroie? answer of this what could ‘If my Lord wanted to ask me anything about it or accuse me, what could I answer regarding this?’ [c.1190, BORON, 46.700] A topicalized or focalized reading of the fronted infinitive is doubtful in (79). Here, the necessities of the rhyme might explain the displacement (peot rhymes with esteot in the next line).

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i



Marie Labelle and Paul Hirschbühler

(79) . . . Un de nus deus estuet murir, / U ambedeus, estre ceo peot. one of us two have-to die or both be it can ‘. . . one of us will die, or both, possibly.’ [c.1160, MARIE-DE-FRANCE, 173.3521, v.591] In the fifteenth century, when both V2 and LSD are disappearing, we find more complex examples of displacement to the left periphery: (80) mais nexun d’ eulx point veu je n’ ay but none of them not seen I neg have ‘But I have not seen any of them.’ [c.1454, Roi René, Regnault & Jeanneton, in Godefroy (1880: Tome V: 490, nesun)] = mais je n’ay point [VP1 ay [VP2 veu nexun d’eulx]] In embedded clauses, the LSDLeft element is generally highlighted though not necessarily focalized (see also (32)–(34)), and it plausibly occupies a position within the left periphery: (81) Et vendoient les villes . . . a ceuls meismes lesquels boutés hors and sold the cities to those precisely whom thrown out il en avoient, they gen had ‘And they sold the cities . . . even to those whom they had thrown out of them.’ [1373, FROISSART, 858.18847] (82) et furent les nefs recargiés de tout ce que il veoient que and were the boats re-loaded of all that which they saw that point mener il n’ en pooient not bring they neg gen could ‘And the boats were reloaded with everything they saw that they could not take with them.’ [1373, FROISSART, 75.791]

. Summary and Conclusion It was argued that Leftward Stylistic Displacement in Medieval French has properties distinct from those of Icelandic SF, although these phenomena share some superficial features. Three distinct constructions were observed: a V2 construction in main clauses, an LSDRight construction, which is the unmarked one, and an LSDLeft construction. LSD with an unfilled canonical subject position is a subcase of LSDRight . The LSDLeft construction is mainly a Middle French phenomenon. In main clauses it appears to involve focalization. As for the structures involved, it was argued that the displacement may target: 1. Spec,FinP for V2 clauses, particularly those with a postverbal subject; 2. The Topic or Focus projection for foregrounded elements (in V2 or LSDLeft constructions);

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

Leftward Stylistic Displacement (LSD) in Medieval French



3. A position between SubjP and TP in the case of LSDRight . For embedded clauses, this is the unmarked position. There is thus not one LSD construction but different constructions allowed by the grammar. We also argued that a combination of (Remnant) VP Movement and Short Scrambling can account for the complex set of facts involving nonfinite verb fronting. At this point, intriguing questions remain, in particular what are the triggers of the various LSDs, and why did they progressively disappear from the language, in particular as regards the LSDRight construction (which includes the case of a missing subject in its canonical position), which represents the core case? As shown by the sentence in epigraph to the present paper, the LSDRight construction survives marginally in an elevated style.

Editions Alexis (c.1090). Storey, Ch. (ed.) (1968). La vie de Saint Alexis. Droz, Genève. Boron (c.1190). Nitze, W. A. (ed.) (1927). Le roman de l’estoire dou Graal. Champion, Paris. Brendan (c.1120). Short, I. and B. Merrilees (eds) (1979). The Anglo-Norman Voyage of St Brendan. Manchester University Press, Manchester. Eustace-Fisher (c.1212). Fisher, J. R. (ed.) (1917). La vie de Saint Eustache par Pierre Beauvais. The Romanic Review, 8, 1–67. Eustace-Petersen (XIIIe s.). Petersen, H. (ed.) (1928). La vie de Saint Eustache. Poème français du XIIIe siècle. Champion, Paris. Froissart (1373). Diller, G. T. (ed.) (1972). Chroniques. Dernière rédaction du premier livre. Droz, Genève. Gormont (c.1130). Bayot, Alphonse (ed.) (1931). Gormont et Ysenbar. Champion, Paris. Marie-De-France (c.1160). Rychner, J. (ed.) (1973). Les lais de Marie de France. Champion, Paris. Perceforest (c.1330–1350). Taylor J. H. M. (ed.) (1979). Le roman de Perceforest. Première partie. Droz, Genève. Prise (c.1370). Mas Latrie, M. L. de (ed.) (1877). La prise d’Alexandrie, ou Chronique du roi Pierre Ier de Lusignan. Fick, Genève. Quatrelivre (c.1170). Curtius, E. R. (ed.) (1911). Li quatre livre des reis. Gesellschaft für romanische Literatur, no. 9. Max Niemeyer, Dresden, Halle. Queste (c.1225). Marchello-Nizia, Ch. (ed.) Édition électronique basée sur Pauphilet, A., éd. (1923). La Queste del Saint Graal. Champion, Paris. Roland (c.1100). Moignet, G. (ed.) (1972). La chanson de Roland. Bordas, Paris. Yvain (c.1170). Roques, M. (ed.) (1960). Les romans de Chrétien de Troyes, édités d’après la copie de Guiot (Bibl. nat. fr. 794). Champion, Paris.

Acknowledgements We would like to thank Martin Elsig, Christine Meklenborg Salvesen, and Beatrice Santorini for helpful comments and suggestions.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

 Diagnosing embedded V2 in Old English and Old French C H R I S T I N E M E K L E N B O R G S A LV E S E N AND GEORGE WALKDEN

. Introduction The goal of this chapter is to discover the conditions under which embedded verbsecond (V2) was permitted in Old English (OE) and Old French (OF), and to explore the theoretical implications of these conditions for the analysis of the landing site of the finite verb. Specifically, we seek to establish (i) whether embedded V0 -to-C0 movement is attested in OE and OF, and, if so, (ii) whether its occurrence in verbal complement clauses is constrained according to the class of the embedding verb in Hooper and Thompson’s (1973) five-way typology. Both OE and OF exhibit at least a strong tendency towards surface V2 in main clauses.1 For both OE and OF, influential early generative treatments (e.g. van Kemenade 1987 for OE; Vanelli et al. 1986 and M. Adams 1987, 1989 for OF) proposed that the finite verb in such clauses was to be found in C0 —but for both languages this analysis has also been challenged (e.g. Pintzuk 1993, 1999 for OE; Dupuis 1989; Rinke and Meisel 2009 for OF), in favour of an analysis where the verb is in some lower position such as I0 . The two families of analysis, which we will henceforth refer to as CP and IP analyses respectively, differ in terms of a crucial prediction: while CP analyses predict the existence of an asymmetry between main and subordinate clauses, IP analyses predict that there should be no such asymmetry. It is this prediction that the present chapter sets out to test. We limit our study to embedded that-clauses, thus leaving other types of subordinate clauses for future research. Our point of departure is Hooper and Thompson’s (1973) typology of verbs that take clausal complements, and we apply this to historical corpus data. To our knowledge, this has not been done before. The chapter is structured as follows. Section 11.2 lays out the two families of analysis and their motivations in more detail, and explains how these predictions are derived. 1 This is one of the many apparently Germanic properties exhibited by OF; see Mathieu () for an overview.

Micro-change and Macro-change in Diachronic Syntax. First edition. Éric Mathieu and Robert Truswell (eds). This chapter © Christine Meklenborg Salvesen and George Walkden 2017. First published 2017 by Oxford University Press.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

Diagnosing embedded V2 in Old English and Old French



Section 11.3 describes our methodology, presenting Hooper and Thompson’s (1973) typology of verb classes. Section 11.4 presents our results, and explores their implications. We conclude that, for both OE and OF, CP analyses are better able to account for our data than IP analyses, but that there is a crucial difference between the two languages: OE does not permit embedded V2 at all, whereas OF permits it in some contexts. By relating this to the Merge site of finite complementizers, we link a microlevel parametric difference to the macro-level property of V2.

. Types and analyses of V in modern and historically attested languages .. Definitions Before we proceed, a word is in order on what we mean by V2. For us, the term is used in its ‘surface’ sense, to describe a structural configuration in which one and only one constituent precedes the finite verb, as in (1), and in which that constituent need not be the subject.2 A V2 language is then one in which this configuration predominates, and a strict V2 language is one in which it is exceptionless. (1)

Den Mann hat der Hund gebissen. the.acc man.acc has the.nom dog.nom bitten ‘The dog has bitten the man.’

(German)

.. V in modern Germanic German, Dutch, and the Mainland Scandinavian languages are all asymmetric V2 languages: as a general rule, V2 is restricted to main clauses, and is not possible in subordinate clauses (see Holmberg 2015: Section 2.4). Another group of Germanic languages, including Icelandic and Yiddish, are considered to be symmetric V2 languages in the sense that the verb is in second position in both main and subordinate clauses (see Santorini 1994).3 See Holmberg (2015) for detailed discussion of the V2 phenomenon in Germanic and beyond. (2) Asymmetric V2—Norwegian a. I går kom jeg for sent på jobb. in yesterday came I too late on work ‘Yesterday, I was too late for work.’ b. Jeg beklager at jeg kom for sent på jobb i går. I regret that I came too late on work in yesterday ‘I regret that I was too late for work yesterday.’ c. *Jeg beklager at i går kom jeg for sent på jobb. 2 A more theory-loaded definition is used by Poletto (), who sees V as a cover term to refer to any structural configuration in which the finite verb occupies a position in the C domain. 3 It is doubtful whether Icelandic, at least, is truly symmetric; see Maling (). It should also be noted that at least Icelandic seems to be restricted as to embedded topicalization in much the same way as the asymmetric V languages even though Icelandic permits more instances of embedded V than e.g. Mainland Scandinavian (see Hrafnbjargarson and Wiklund  and references therein).

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i



Christine Meklenborg Salvesen and George Walkden

(3) Symmetric V2—Icelandic Jón harmar að þessa bók skuli ég hafa lesið. John regrets that this book shall I have read ‘John is sorry that I’ll read this book.’ (Hrafnbjargarson and Wiklund 2009) The symmetric V2 languages have for the most part been analysed as involving the finite verb in I0 in both main and subordinate clauses: see Diesing (1990) on Yiddish and Rögnvaldsson and Thráinsson (1990) on Icelandic. When it comes to the asymmetric languages, however, the main question has been how to derive the contrast between clause types without stipulation. The key insight came from den Besten (1977) and Evers (1981), who proposed that, in these languages, the finite verb was in complementary distribution with subordinating complementizers, with both in C0 . This type of analysis, in which the finite verb occupies C0 in all main clauses, has been challenged by Travis (1984, 1991) and Zwart (1991, 1993), who argue that the verb only raises to C0 when a nonsubject constituent is in initial position; when the subject is initial, it remains in Spec,IP, and the verb remains in I0 . The idea that the position of the finite verb depends on the nature of its preceding XP (subject vs. nonsubject) is what we will refer to as the split hypothesis. Schwartz and Vikner (1989, 1996) demonstrate that the split hypothesis faces a number of problems not shared by the consistent V0 -to-C0 analysis, and as a result the majority of syntacticians today adopt a version of the latter, though Sells (2001) and Mikkelsen (2010) are exceptions.4 See Poletto (2002); I. Roberts (2004); Frascarelli and Hinterhölzl (2007); and Mohr (2009) for variants of the V0 -to-C0 approach, which differ in implementation but share the basic insight that the position of the finite verb in main clauses is uniform. To sum up: there are three competing analyses that can in principle account for V2 word order in main clauses: (i) the IP analysis, claiming that the finite verb does not move higher than I0 and that the preceding element sits under Spec,IP; (ii) the CP analysis, that claims that the finite verb sits under a head in the C domain; and (iii) the split hypothesis, that suggests that the finite verb is in the C domain when an XP other than the subject precedes it, and under I0 when it is preceded by the subject. The analyses make differing predictions for subordinate clauses, which we will return to in Section 11.2.5. .. V in Old English That OE has a high proportion of V2 main clauses has long been recognized (see e.g. Jespersen 1949). The analogy to the modern Germanic languages was first drawn in generative theory by Canale (1978) and van Kemenade (1987). However, OE differs from the modern languages in a number of ways: in particular, it exhibits a considerable number of V3 constructions. Most notably, with personal pronoun subjects, as

4 More exotic analyses of V exist: see Fanselow () and Müller (). We will not consider these here.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

Diagnosing embedded V2 in Old English and Old French



in (4), main clauses are almost invariably V3; pronominal objects, some adverbs, and full nominal subjects as in (5) may also occupy this position.5 (4) æfter his gebede he ahof þæt cild up after his prayer he lifted the child up ‘After his prayer he lifted the child up.’ (cocathom2, +ACHom_II, _2: 14.70.320) (5) Nu se rica mann ne mæg her habban . . . now the rich man neg can here have ‘Now the rich man cannot here have . . . ’ (coaelive, +ALS[Ash_Wed]: 110.2758) These data have been taken to motivate an analysis in which the verb is typically in the I domain, for instance in Pintzuk (1993, 1999); Eyþórsson (1995); Haeberli (1999, 2002, 2005); Fuß (2003); and Speyer (2008, 2010). If the verb does not evacuate the I domain, the prediction is that there will be no main–embedded asymmetries. Other authors, however, have pointed to the possibility of an expanded CP layer following Rizzi (1997) as an alternative way of accounting for some of these patterns; these include I. Roberts (1996); Westergaard (2005); Hinterhölzl and Petrova (2009); and Walkden (2009, 2014).6 We refer the reader to Walkden (2014: ch.3) for more discussion of the range of constituent orders to be found in OE main clauses. .. V in Old French As for OE, the idea that OF is fundamentally a V2 language has a long pedigree, going back at least to Thurneysen (1892). M. Adams (1987, 1989) and I. Roberts (1993) are two influential early generative analyses exploiting the idea that the verb is in the C domain. When an XP other than the subject precedes the finite verb, the subject is postverbal (6). (6) Lors oste mes sires Gauvains son hiaume de sa teste then removes my sire Gauwain his helmet of his head ‘Then Sir Gauwain removes the helmet from his head.’ (Graal, col. 165d, l. 27) There is no consensus as to the position of the finite verb in main clauses: Vance (1997); Labelle (2007); and Labelle and Hirschbühler (chapter 10, this volume) are proponents of the split hypothesis, whereas in the analyses of Vanelli et al. (1986); M. Adams (1987, 1989); I. Roberts (1993); Benincà (2006); Vance et al. (2010); Salvesen (2013); and Steiner (2014) there is always V0 -to-C0 movement. OF does, however, display numerous exceptions to the V2 rule, in general by allowing V3 structures. These are quite homogeneous in the sense that the initial XP is either a subordinate adverbial clause (Vance et al. 2010; Donaldson 2012a,b; Radwan 2011; Salvesen 2013; Bech and Salvesen 2014) or belongs to a small set of adverbials 5 There are also examples in which the verb occurs in a later position: see W. Koopman (); Pintzuk and Haeberli (); and Walkden (: –). 6 Recent work by van Kemenade and co-authors discussing information-structural variation in subject positions, e.g. van Kemenade and Milićev () and van Kemenade and Westergaard (), proposes that the verb moves to the head of a functional projection FP, but is neutral as to whether this projection is in the C domain or below it. Our data support the view that FP is in the C domain.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i



Christine Meklenborg Salvesen and George Walkden

(Vance 1997; Bech and Salvesen 2014). Examples that are not captured by this generalization are not widespread in the thirteenth century. Example (7) shows a fronted subordinate adverbial clause followed by S–V word order.7 (7) Et quant il fu auques anuitié et il fu hore de dormir li and when it was some begun-night and it was hour of sleep the rois prist Galaad et l’enmena en sa chambre king took Galahad and him-brought in his chamber ‘And when the night had started to fall and it was time to sleep, the king took Galahad and brought him into his own chamber.’ (Graal, col. 164d, l.7) As a fronted subordinate adverbial clause may be followed either by an S–V structure or (rarely) a V2 structure, these clauses have been analysed as occupying a specifier position to the left of the V2 scheme, probably the specifier of the Scene Setting position in the sense of Benincà and Poletto (2004) (see Donaldson 2012b; Salvesen 2013; and Bech and Salvesen 2014). In the cases where the fronted subordinate clause is followed by the finite verb and (optionally) the subject, it must be analysed as the first element in the V2 structure. This implies that it occupies a different position than in the cases where it does trigger a V3 word order. The standard description of OF says that it is an asymmetric V2 language, and that subordinate clauses have S–V word order (8). (8) Et il dient que ce est voirs and they say that this is true ‘and they say that this is true.’

(Graal, col. 214b, l. 19)

In this chapter we will show that embedded V2 in OF is fairly common, and we will examine these structures more thoroughly. .. Embedded clauses: a neglected domain If we assume, standardly, that complementizers originate in the C domain, the predictions made by the two families of analysis are clear (Haeberli 2005: 273–4). All IP analyses predict that XP–V–(S) word order should be as prevalent in subordinate clauses as in main clauses, whereas CP analyses predict a clear asymmetry. IP analyses also predict that V1 structures should be possible in subordinate clauses.8 In addition, Pintzuk’s (1993, 1999) IP analysis of OE and Dupuis’ (1989) and Rinke and Meisel’s (2009) IP analyses of OF, since they treat Spec,IP as an A -position that can host any constituent, predict that V2 should be possible in all subordinate clause types, as should embedded topicalization. By contrast, the ‘split’ IP analyses (for OE see Eyþórsson 1995; Haeberli 1999, 2002, 2005; Fuß 2003; Speyer 2008, 2010; for OF see Vance 1997; Labelle 2007; Labelle and Hirschbühler Chapter 10, this volume), in which Spec,IP is not an A -position but where the finite verb moves to a head in the I domain,

7 The main clause following a fronted subordinate adverbial clause may also be introduced by the particle si, especially if the two clauses contain the same subject (Schøsler and van Reenen ). 8 Pintzuk (: ) acknowledges this problem. Haeberli () is able to evade it by positing, like CP analyses, that the verb moves higher in main clauses (to AgrS0 ) than in subordinate clauses (T0 ).

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

Diagnosing embedded V2 in Old English and Old French



predict that embedded topicalization and non-subject-initial V2 should be ruled out under certain predicates, as do the CP analyses. Following Vikner (1995: 65), asymmetric V2 languages can be divided into two subclasses: languages such as standard German, which prohibit embedded V2 whenever the complementizer is present, and languages such as Mainland Scandinavian, which permit embedded V2 with an overt complementizer only in specific contexts.9 In particular, embedded V2 is ruled out in these languages in complement clauses of verbs that are not ‘bridge’ verbs (Vikner 1995; see also Andersson 1975; Green 1976; Julien 2009). In this chapter we aim to investigate the possibilities of embedded V2 in OE and OF. If embedded V2 exists, is the occurrence of such constructions limited to certain contexts? Instead of evoking the rather blurry term of ‘bridge verb’ (Vikner 1995; see also Andersson 1975; Green 1976),10 we will use Hooper and Thompson’s (1973) classification of different types of verbs that take finite complement clauses. We take all instances of XP–V–S as instances of V2 and do not consider Leftward Stylistic Displacement in the sense that has been proposed by Labelle and Hirschbühler (Chapter 10, this volume).

. Investigating complement clauses .. Classes of complement-taking verb ‘Main clause phenomena’ (Emonds 1970; Green 1976) are found in the complements of some verbs, but not others. These main clause phenomena include Negative Constituent Preposing, as in (9) and (10), as well as VP preposing, topicalization, and more. Example (10) shows that main clause phenomena are ruled out in at least some subordinate clauses. (9)

He’s been out of work before, but never has he had to borrow money.

(10) *The fact that never has he had to borrow money makes him very proud. To account for the empirically observed distribution of main clause phenomena, Hooper and Thompson (1973) divide complement-taking verbs into five classes, labelled A–E, according to the discourse status of their complement clauses: strong assertive verbs (Class A), weak assertive verbs (Class B), verbs that are neither assertive nor factive (Class C), factive verbs (Class D), and semifactive verbs (Class E). Class A verbs in Hooper and Thompson’s schema introduce strong assertions, and are typically verbs of saying, e.g. say, report, exclaim, assert, claim, vow, be true, be certain, be sure, be obvious. Class B verbs introduce weak assertions, and are typically verbs denoting mental processes: examples are suppose, believe, think, expect, guess, imagine, it seems, it happens, and it appears. Hooper and Thompson (1973) find that members of both these classes allow main clause phenomena in their complement clauses. The clauses in (11) both contain a topicalized element. 9

In German, embedded V is the only possibility when the complementizer is not expressed. Vikner himself notes (: , fn.) that the notion of a bridge verb is problematic: in particular, the term was introduced to denote verbs that did not permit extraction from their complement, and this does not always correlate with permitting embedded V. See Biberauer () and Julien () for two critical views. 10

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

 (11)

Christine Meklenborg Salvesen and George Walkden Root: Most embarrassing of all was falling off the stage. a. Class A: Carol said that most embarrassing of all was falling off the stage. b. Class B: I suppose that most embarrassing of all was falling off the stage.

In contrast to Classes A and B, verbs of Class C select complements that are not asserted. In English these are verbs such as be (un)likely, be (im)possible, be (im)probable, doubt, and deny. Hooper and Thompson show that main clause phenomena never occur in the complements of Class C verbs (12). (12) Class C: *It’s unlikely that most embarrassing of all was falling off the stage. The Class D factive verbs resent, regret, be sorry, be surprised, bother, be odd, be strange, be interesting select complements that are presupposed, and typically express an emotion or subjective attitude. Class E verbs also select presupposed complements, and are referred to as semifactive (Karttunen 1971); they include realize, learn, find out, discover, know, see, and recognize, and typically ‘assert the manner in which the subject came to know that the complement proposition is true’ (Hooper and Thompson 1973: 480). Hooper and Thompson find that main clause phenomena are not permitted in the complements of Class D verbs. As for complements of Class E verbs, however, they allow root phenomena, but only when the complement is asserted. The generalization that emerges is that main clause phenomena are possible in asserted complement clauses only. (13) Root: She opened the window and in flew Peter Pan. a. Class D: *Wendy was sorry that she opened the window and in flew Peter Pan. b. Class E: Tinker Bell saw that Wendy opened the window and in flew Peter Pan. Julien (2009) applies their diagnostics to the issue of embedded V2 using corpora of Norwegian, Danish, and Swedish. In all cases, she finds that the contexts in which embedded V2 is permitted are exactly those that count as asserted using Hooper and Thompson’s criteria, that is complements of verbs of Classes A, B, and E. Our prediction is therefore that, insofar as we find embedded V2 in OE and OF, it should be restricted to these contexts. .. Our methods The corpus used for OE was the York–Toronto–Helsinki Parsed Corpus of Old English Prose (YCOE; Taylor et al. 2003). From this corpus we took an exhaustive sample of complement clauses with surface order (non-nominative) XP–V– . . . . In addition, a sample of 881 other complement clauses was taken.11 For OF, the Base du Français

11 Given that the YCOE is not lemmatized and a search for these other complement clauses turned up , examples, an exhaustive manual analysis of non-V clauses was not undertaken for reasons of time. Instead, a sample of non-V clauses was randomly selected for further analysis.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

Diagnosing embedded V2 in Old English and Old French



Médiéval (BFM; Heiden et al. 2010) was used, and all verbal complement clauses from the text La queste del saint graal were analysed (879 in total). We then classified all matrix predicates as falling under one of Hooper and Thompson’s (1973) five categories. Applying this typology to OE and OF was not entirely straightforward, since more verbs in OE and OF take finite clausal complements than in their modern counterparts. In particular, verbs of ‘volition’, as in (14) and (15), are difficult to fit into the system. In Modern English, these verbs take nonfinite complements, as shown by the translations. (14) gyf we willað þæt us gehyre se heofonlica God if we want that us hear the heavenly God ‘if we want heavenly God to hear us’ (OE; coaelhom, +AHom_5: 178.795) (15) Dame, fait il, que volez vos que je vos face lady made he what want you that I you do ‘Milady’, he said, ‘what do you want me to do?’ (OF; Graal, col. 132, l. 36) The complements of these verbs seem to be neither asserted nor presupposed, but express the protagonist’s attitude towards a hypothetical state of affairs. In both OE and OF these clauses typically take the subjunctive mood. For the purposes of our study, we have chosen to treat these ‘volitives’ as a separate group, Class V.

. Results and discussion .. Breakdown of results by verb class The overall distribution of V2 clauses by embedding predicate type is given in Table 11.1.12

Table .. Embedded V2 in OE and OF by class of embedding predicate: overall results Old English

Old French

Total

V2



Total

V2



A B C D E V

476 230 14 96 280 240

172 93 5 13 123 49

3.6 4.3 3.5 1.0 4.9 1.7

286 196 0 35 235 127

78 18 0 0 24 1

28.0 9.2 0.0 0.0 10.2 0.8

Total

1336

455

3.3

879

121

13.8

12 For OE, the italicized percentages in Tables . and . are estimates reached by assuming that the proportions of the different classes represented in the non-V sample can be extrapolated to all , non-V clauses.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i



Christine Meklenborg Salvesen and George Walkden

Table .. Embedded V2 in OE and OF by class of embedding predicate: only non-accidental V2 Old English

Old French

Total V2



Total

V2

476 230 14 96 280 240

11 6 0 2 8 2

0.2 0.3 0.0 0.2 0.3 0.1

286 196 0 35 235 127

60 21.0 6 3.1 0 0.0 0 0.0 12 5.1 0 0.0

Total 1336

29

0.2

879

78

A B C D E V



8.9

It is, however, uninstructive to compare the different classes of embedding verbs without taking into account the type of embedded verb, since we know that several types of constructions will merge the DP subject in a low position that may cause it to surface postverbally. This is the case for unaccusative and impersonal verbs, passive, presentative, and modal constructions, and clauses with copulas (see in particular van Kemenade 1997 for OE, and Legendre and Sorace 2003 for Modern Romance). If we assume that there is no requirement for the DP subject to evacuate the vP domain, these embedded verbs may therefore have a postverbal DP subject without verb movement to the C domain (Salvesen and Bech 2014), yielding a surface V2 word order which we label accidental V2. Therefore, these kinds of clause have been left out. We have also excluded embedded clauses with a pro subject in OF, even though pro is often seen as an indication of V0 -to-C0 movement in main clauses (M. Adams 1987; Vance 1989, 1997; I. Roberts 1993). The reason for this is that there are known cases where pro may occur without V0 -to-C0 movement (Hirschbühler and Junker 1988; Ingham 2012; Salvesen 2014b).13 When we leave out these groups, we get a set of clauses that are clearly instances of embedded V2. We see that in OF there are no instances of embedded V2 after Class C or D predicates. Even though we do not have any occurrences of Class C verbs in the OF data, the differences observed in Table 11.2 are statistically significant: a Fisher’s exact test (Fisher 1922) for Classes A, B, E, and V vs. C and D gives us p = 0.03615. In OE, the difference between Classes A, B, E, and V and Classes C and D is not significant (p = 1), apparently because embedded V0 -to-C0 as a whole is incredibly infrequent with all classes of verbs: in no class does it occur more than 0.3 per cent of the time.

13 OE examples where it is clear that no overt subject is present (instances of subject extraction from the subordinate clause, as well as eleven instances of referential null subjects) were also excluded from the V figures in Table ., as it is not clear that these do not also involve accidental V. Italicized figures for OE are extrapolated estimates.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

Diagnosing embedded V2 in Old English and Old French



... Class A predicates In OE most tokens of Class A verbs involve secgan and cweðan ‘to say’, including nine of the eleven cases of embedded V2. (16) ac hit wæs openlice gecyþed, þæt his forðfore begeat seo but it was openly said that his death obtained the þingung þæs arwyrðan Anastasies intervention the.gen venerable A. ‘But it was openly said that the intervention of the venerable Anastasius caused his death.’ (OE; cogregdC, GD_1_[C]: 8.53.29.608) In OF, 21 per cent of the clauses attached to a Class A predicate display V2. The vast majority of these (51 of 60) are attached to the verb dire ‘say’. (17) Or dit li contes que .iii. jorz fist li preudons Lancelot now says the story that three days made the prud’homme Lancelot demorer o lui stay with him ‘Now the story tells you that the gentleman made Lancelot stay with him for three days.’ (OF; Graal, col. 187s, l. 3) ... Class B predicates In OE, Class B verbs include geliefan ‘to believe’, as in (18), and limpan ‘to happen’, and in OF cuidier ‘believe’, avenir ‘happen’, and sembler ‘seem’. (18) þæt þu gelyfe, þæt þa wiþercorenan bærneþ þæt ece that you believe that the.acc against-chosen.acc burns the eternal fyr of þam dæge heora ændes and forðfore fire from the day their.gen end.gen and death ‘that you believe that the wicked are burned by eternal fire from the day of their death’ (OE; cogregdC, GDPref_and_4_[C]: 29.303.14.4509) (19) Si m’ en est si bien puis avenu que mout i ai si me of-it is so well then happened that much there have.1sg gaaingnié won ‘And it all turned out so well for my part that I won a lot there.’ (OF; Graal, col. 173a, l. 36) (20) et bien i paroit a ce qu’ il avoit le jor fet que and well there seemed of that what he had that day done that d’ iluec en avant porroit il legierement sormonter de proece toz of here in forward could he easily surmount of prowess all les autres chevaliers the other knights ‘And based on what he had done that day, it seemed likely that he would easily surpass the boldness of all the other knights in the future.’ (OF; Graal, col. 163a, l. 32) In (20), the matrix verb is paroir ‘seem’, and the subject of the embedded clause is the personal pronoun il ‘he’. There is a long tradition among OF linguists to consider the

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i



Christine Meklenborg Salvesen and George Walkden

postverbal personal pronoun as enclitic to C0 (I. Roberts 1993; Vance 1997). Thus, the construction in (20) strongly indicates that there is V0 -to-C0 movement. ... Class C predicates Hooper and Thompson’s Class C is incredibly rare in our corpora: there are no examples for OF, and only fourteen in total for OE. In neither language is there a single example of unambiguous V2: (21) below, from OE, appears to be one, but this is an instance of a passive with a low nominative, and hence accidental V2 (see van Kemenade 1997: 335). (21) Forþon se ylca wer wiðsoc, þæt him moste beon ænigu therefore the same man denied that him must be any nydþearfnes geþegnod need served ‘Therefore that man denied that there was any need for him to be served.’ (OE; cogregdC, GDPref_and_4_[C]: 12.276.9.4027) ... Class D predicates In OE, of the ninety-six examples of clauses embedded under Class D predicates, only two appear to contain embedded V2, and both are embedded under the same predicate in the same sentence. The offending example is given in (22) below. (22) Gif hire ðonne se wiðsace, ðonne is cynn ðæt him spiwe if her then dem.nom deny then is proper that him spit.sbjv ðæt wif on ðæt nebb, ðæt is ðæt hine tæle ðæs the woman in the nose that is that him blame.sbjv the.gen folces gesomnung people.gen assembly ‘If he then denies her, then it is proper that the woman should spit in his face, that is, that the people’s assembly should blame him.’ (OE; cocura, CP: 5.45.2.249) These clauses are the only two counterexamples in our corpora to the claim that nonaccidental embedded V2 is ruled out with verbs of classes C and D. Moreover, the predicate here—cynn wesan ‘to be suitable/proper’—is classified as D because it expresses an emotion/attitude, like Hooper and Thompson’s (1973) examples be interesting and be strange; it could also be classified as A, parallel to Hooper and Thompson’s (1973) classification of be obvious. In any case, this sentence alone cannot be taken to indicate that embedded V2 was a productive possibility in OE. In OF, there are no instances of embedded root phenomena after a Class D predicate. Class D verbs are, however, attested. This group typically include douter ‘fear’, plaisir ‘please’, and soffrir ‘suffer’. ... Class E predicates In OE, seon ‘to see’, witan ‘to know’, and ongietan ‘to perceive/understand’ account for most of the Class E predicates. In OF, we find verbs such as savoir ‘know’, conoistre ‘know’, veoir ‘see’, and oïr ‘hear’. The verbs savoir and veoir are the most common verbs to embed a V2 subordinate clause, with ten and nine clauses respectively.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

Diagnosing embedded V2 in Old English and Old French



(23) In þære wisan mæg beon ongyten be þæs ylcan Stephanes In this way may be perceived by the.gen same.gen S.gen life, þæt in him wunnon and campedon þa yfel his lichaman life that in him fought and struggled the evil his.gen body.gen wið þam weorce his ælmesdæda against the work his.gen alms-deeds ‘In this way one can see from Stephen’s life that in him the evil of his body fought against the works of his charitable deeds.’ (OE; cogregdC, GDPref_and_4_[C]: 37.320.16.4809) (24) car tu sez bien que la fille le roi Pellés coneus tu for you know well that the daughter the king Pelleas knew you charnelment carnally ‘For you know well that you were intimate with the daughter of the king Pelleas.’ (OF; Graal, col. 192d, l. 30) ... Class V predicates In the YCOE there are only two unambiguous examples of embedded V2 under Class V predicates; (25) is one of them. (25) gyf we willað þæt us gehyre se heofonlica God if we want that us hear.sbjv the heavenly God ‘if we want to be heard by God’ (OE; coaelhom, +AHom_5: 178.795) In OF, there are no unambiguous examples of embedded V2 after Class V predicates. .. Analysis: Embedded V and the complementizer Our results show a difference in behaviour between OE and OF: while in OF, embedded V2 appears to be possible under certain predicates but not others, much as in Mainland Scandinavian (Julien 2009), in OE embedded V2 is completely ruled out, much as in modern German, with only a handful of non-accidental counterexamples (0.2). For both OE and OF, we would expect embedded V2 to be generally productive with all types of embedding predicates under a pure IP analysis, since Spec,IP is an A -position. Our data falsify this prediction, indicating that only a CP analysis, or a ‘split’ IP analysis, is realistic. Why do certain types of verbs permit V2 word order in the subordinate clause in OF whereas others seem to prohibit this? Using Rizzi’s (1997) model of an exploded CP, we will argue that the difference is caused by the verb’s selectional properties: verbs permitting embedded root phenomena select a larger phrasal unit headed by a higher complementizer—presumably under Force0 . Verbs that do not permit embedded root phenomena select a smaller phrasal unit headed by a lower complementizer presumably merged under Fin0 . This is in the same vein as analyses proposed by I. Roberts (2004); Benincà and Poletto (2004); Rouveret (2004); Paoli (2007); VillaGarcía (2012); Ledgeway (2012); and Dagnac (2012), just to mention a few. Under the CP analysis we would thus expect to find an asymmetry between different types of verbs with respect to embedded root phenomena. This asymmetry would be

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i



Christine Meklenborg Salvesen and George Walkden

unexpected under the IP analysis. We thus believe that the distribution we find, where embedded V2 does not occur under predicates of types C and D, indicates that main clause constituent order involves verb movement to the CP layer. .. Embedded V in OF We argue that in OF there are two options for the CP where ForceP is projected: either the complementizer is merged under Fin0 and moved by standard head movement to Force0 , or it is merged directly under Force0 (Rouveret 2004; Salvesen 2014a). In the former case the subordinate clause crucially has an S–V structure with the subject under Spec,TP (or enclitic to Fin0 , see Vance 1997). In certain cases there may be material in the CP field which is preceded and followed by a complementizer (in boldface)—known as complementizer doubling (CD) (26).14 , 15 In cases where no slots in the left periphery are filled, the complementizer is merged under Force0 , leaving Fin0 open for V0 -to-C0 movement (27). This gives us the structure in (28) for the cases under discussion in this chapter (excessive structure omitted). (26) ( . . . ) car vos savez bien que quant vos alastes a cort que mes for you know well that when you went to court that my sires li rois avoit guerre contre le roi Libran sir the king had war against the king Libran ‘For when you went to court, you knew well that the king was at war with king Libran.’ (OF; Graal, col. 178d, l. 26) (27) Et il respont que ce ne feroit il pas. and he responded that that neg does he neg ‘And he replied that he would not do that.’ (OF; Graal, col. 183a, l. 40) (28) [ForceP [Force0 que] ( . . . ) [FinP [Spec,FinP ce] Fin0 [ ne feroit ( . . . )]]] If we consider the different subtypes discussed here, we see that they all represent different kinds of root phenomena and that these may be linked to the presence of a high complementizer. In structures that express two complementizers, the word order is always S–V. When only the higher complementizer is expressed, it is possible to get V2 structures following the initial element. Both kinds of root phenomena show the

14 As complementizer doubling implies the selection of a higher complementizer (under Force0 ), we could have chosen to include these in our tables as they too provide information about the selectional properties of the matrix verb. For reasons of clarity we have chosen to limit this chapter to the discussion of unambiguous V cases. 15 See also Ribeiro and Torres Morais () for Old Portuguese. These examples are occasionally found for OE too, though they are rare:

(i) Wel mæg gehwa witan þæt gif ahwær is myrcð and wuldor þæt þær is unasecgendlic well may each know that if anywhere is mirth and glory that there is unspeakable wuldor þær se wunað þe ealle ðincg gesceop glory where he lives who all things shaped ‘All may well know that if there is mirth and glory anywhere, there is incredible glory where he who created all things lives.’ (OE; coaelive, +ALS[Ash_Wed]: .)

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

Diagnosing embedded V2 in Old English and Old French



same pattern as we saw when we only considered V2: they exclusively appear after verbs of classes A, B, E, and V. .. Embedded V in OE Finally, OE, as we have seen, basically disallows embedded V2 entirely. The microparametric difference between OE and OF—the same that distinguishes German and Mainland Scandinavian—can be captured lexically if we assume that there is no high complementizer in OE: all complementizers are first merged in Fin0 . Whether or not the complementizer remains there, its lower copy will then block verb movement to the left periphery.16 While there are traces of the structure found in OF—a handful of examples of apparently non-accidental embedded V2, and double-complementizer structures—it does not seem to be a productive possibility in OE.

. Conclusion In this chapter we have tested the predictions of two families of analyses of OE and OF with respect to embedded V2. IP analyses predict that embedded V2 should occur freely, as has been argued for Icelandic and Yiddish—a prediction that is false for both OE and OF. By contrast, CP analyses (and ‘split’ analyses) predict that embedded V2 should either (a) be ruled out entirely except where accidental, or (b) occur only under specific types of predicate. Using Hooper and Thompson’s (1973) typology of predicate classes, we found that OE, like modern German, instantiates option (a): non-accidental embedded V2 is vanishingly rare. OF, like the modern Mainland Scandinavian languages, instantiates option (b), allowing embedded V2 only under certain predicates. We link this difference to the site of first Merge of complementizers in the two languages.

Acknowledgements We would like to express our gratitude to the audiences of DiGS 15 and ICHL 21 as well as to two anonymous reviewers and the volume’s editors for their helpful comments.

16 A reviewer questions what happens in cases of OE relative clauses in which the relative pronoun is in Spec,FinP and there is no overt complementizer. In these cases, the finite verb might be predicted to move to Fin0 ; however, embedded V is vanishingly rare in such contexts as it is in other relative and complement clauses with an overt complementizer. This problem in fact exists for all accounts that seek to translate the den Besten () intuition into X -theoretic terms (cf. Richter and Sailer ). We do not have a solution at present.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

 The pragmatics of demonstratives in Germanic CAITLIN LIGHT

. Introduction This chapter will seek to demonstrate that demonstrative pronouns in Germanic are inherently pragmatically contrastive, in that they conventionally signal a marked and unexpected referent given the existing discourse structure. In this way they can be analysed as subinformative in the sense of Gast (2010), with an information-structural function not unlike contrastive topics. To explore the properties of demonstrative pronouns, I will centre my investigation on object topicalization. Crucially, the phenomena labelled ‘object topicalization’ in Germanic prove to be of two distinct but overlapping motivations. One type of topicalization is information-structurally driven, while the other is purely formal, to satisfy a requirement in the syntax, the so-called verb-second (V2) constraint (cf. Frey 2004, 2006a,b; Light 2012). The perceived unity of these constructions is a consequence of the fact that information-structurally driven topicalization may also serve to satisfy V2. Ultimately, I will show that in information-structurally driven operations, demonstrative pronouns pattern more like contrastive elements than they do like noncontrastive elements, even when we cannot identify a traditionally recognized contrastive interpretation. By revising our understanding of the pragmatic function of demonstrative pronouns, we are able to account for their synchronic and diachronic behaviour with greater success. This investigation will begin and end with a puzzle in the history of English topicalization. While Speyer (2010) offers a thorough and satisfying account of change in the behaviour of topicalization from Old to present-day English, further work has shown that demonstrative pronouns specifically do not behave according to Speyer’s expectations. It is in pursuing this problem, and repairing an apparent inconsistency in this account, that we are led to a new analysis of the pragmatics of demonstratives. The chapter will be organized as follows: in the next section, I will outline the details of the puzzle in the history of English topicalization. Following this I will briefly outline Micro-change and Macro-change in Diachronic Syntax. First edition. Éric Mathieu and Robert Truswell (eds). This chapter © Caitlin Light 2017. First published 2017 by Oxford University Press.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

The pragmatics of demonstratives in Germanic



the methodology of the current study, and the data upon which my conclusions will be based. Then, in Section 12.4, I will take a step back to consider the general properties of topicalization in Germanic. Section 12.5 will present evidence that demonstrative pronouns are consistently behaving against the expectations set out by the existing analysis. I will consider the general properties of demonstratives and present a complete analysis of their behaviour in Section 12.6, and finally, Section 12.7 will conclude, and a solution to our puzzle will be proposed.

. A puzzle in historical English Across the Germanic language family, we find a type of movement traditionally termed topicalization. This phenomenon may be realized in Germanic languages which adhere to the V2 constraint, as well as those without it. (1)

a. Das weiss that.acc know b. það veit that.acc know c. That, I know.

ich. I.nom ég. I.nom

Direct object topicalization in V2 languages is exemplified by (1a), for German, and (1b), for Icelandic. Here, topicalization of the object leads to inversion of the subject and the finite verb, according to the V2 constraint. Modern English, in contrast, is not V2, and thus topicalizing the direct object does not lead to subject–verb inversion (1c). Instead, the result may be superficially described as a sentence with verb-third word order. A number of recent studies have begun to clarify the synchronic and diachronic properties of object topicalization in English. Such work has also revealed some puzzling questions; it is here that the current study begins. Speyer (2010) makes significant progress in picking apart the syntactic and prosodic properties of English topicalization. Superficially, it has been observed that the overall rate of object topicalization declines over time, so that object topicalization in Modern English occurs much less frequently than it did in Old English. An immediate question one must ask is: what leads to this change? Is object topicalization in the process of disappearing from the English language? Speyer argues that this is not the case. According to his proposal, the overall decline in the rate of object topicalization is not an independent syntactic change in the grammar of English: in other words, the syntax of topicalization does not, in itself, change. Rather, the contexts in which object topicalization is prosodically well-formed decline, and the frequency of topicalization declines as a consequence. Old English, as is well-known, had a grammar that generated verb-second word orders not unlike those found in German (although the subject type played a particular role in OE-style V2; cf. Pintzuk 1991; Fischer et al. 2000). Over time, V2like word orders in main clauses disappeared as the grammar of English changed. Speyer claims that this is the key to understanding object fronting in English: the loss of V2 word orders limited the environments in which fronting is prosodically

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i



Caitlin Light

well-formed, leading to a decline in the frequency of use of topicalization across the board. Topicalized DPs bear a prosodic accent in English. When the subject is also accented, object topicalization will result in accent clash, a prosodically dispreferred result. To demonstrate: (2a) is a perfectly acceptable example of object topicalization in present-day English, with the accented object and finite verb ‘saved’ from accent clash by the intervening, unaccented subject pronoun. However, (2b) is dramatically less acceptable, because the adjacency of the accented subject and object leads to accent clash. (2) a. Bèans I líke. b. ??Bèans Í like. The latter type of example is made acceptable in Old English by the availability of V2 word orders, which would allow the unaccented finite verb to intervene between two accented elements. Consequently, topicalization was prosodically acceptable in more environments while V2 word orders were possible in English. Speyer’s theory allows us to link several apparent changes in the English language to a single syntactic development. It also offers a clearer picture of topicalization across the history of English. Because pronoun subjects are resistant to accent, in the absence of the V2 word order option, it is predicted that object topicalization is more likely to be well-formed in clauses with a pronoun subject. And in fact, when only clauses with a pronoun subject are considered, Speyer finds that the rate of object topicalization is nearly stable through the history of English.1 .. The exception: unaccented topicalization Note that Speyer’s theory relies on the assumption that topicalized objects are necessarily accented in English. This assumption is required to motivate the claim that accent clash can have the observed effect on the overall rate of topicalization. In fact, Speyer claims that unaccented objects were capable of topicalizing in Old English. Frequent topicalization of object pronouns is treated as a key demonstration of this, because, as before, Speyer assumes that pronouns resist accent. Examples of pronoun object topicalization in Old English are demonstrated in (3). (3) a. Þone asende se Sunu this sent the son ‘The son sent this one.’

(coaelhom,+AHom_9:113.1350)

b. & hit Englisce men swy3e amyrdon and it English men fiercely prevented ‘and the Englishmen prevented it fiercely.’ (cochronE,ChronE_[Plummer]:1073.2.2681)

1 The relatively stable rate of topicalization reported by Speyer is with the exception of pronominal object topicalization in Old English: the one type of topicalization he argues has been lost.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

The pragmatics of demonstratives in Germanic



Pronoun topicalization of this type occurred at a reasonable rate in Old English, but disappeared rapidly in the Middle English period. Speyer relates this to the proposal of dual topicalization mechanisms presented for German in Frey (2004, 2006a,b). Frey suggests that topicalization is associated with two separate types of movement: (1) True A-Bar Movement (TAB), which results in a contrastive interpretation on the topicalized XP, and (2) Formal Movement (FM), which has no interpretive effect. While TAB may target any constituent in the clause, FM is restricted to targeting only the highest available element below C. Contrastive constituents are frequently associated with a specific accent pattern (cf. Jackendoff 1972; Büring 1997). Speyer therefore associates the availability of a contrastive interpretation with the availability of a contrastive accent, which means that the patterns of English topicalization within his general account may be reduced to TAB. More importantly, however, Speyer proposes that FM was available in Old English, and subsequently lost. This proposal accounts for the availability of unaccented topicalization in previous stages of the language, as well as the abrupt drop in pronoun topicalization in the Middle English period (which must demonstrate the proposed loss of FM). All examples of topicalization after this point are predicted to be examples of TAB, and thus to fit with the claim that Modern English allows topicalization only of (contrastively) accented objects. .. The puzzle Speyer’s proposal for English is revisited in Stevens (2010) and Stevens and Light (2012), which test the hypothesis that English lost noncontrastive topicalization following the Old English period. What they find, however, is that in contexts in which topicalization remains prosodically well-formed (following Speyer’s analysis), although personal pronoun topicalization declines as Speyer originally observed, demonstrative pronoun topicalization remains relatively stable across the history of English and into the Modern British English period. The behaviour of demonstrative pronouns presents a serious challenge to Speyer’s analysis. They appear to be capable of topicalizing without a contrastive interpretation in every stage of English; example (4), taken from the Penn Parsed Corpus of Modern British English (Kroch et al. 2010), shows a topicalized demonstrative pronoun in a context which affords no plausible contrastive interpretation according to a traditional analysis. By all appearances, topicalization of this type remains felicitous in contemporary English. (4) and, on receiving my answer, brought in a pint bottle with a large label on its side — Fine Crab-apple Cider. This he opened . . . (READE-1863,224.589–90) Demonstrative pronoun topicalization is surprisingly common: in Early Modern English for example, demonstrative pronouns topicalize more often than they remain in situ (a fact revisited in Section 12.5). What’s more, the end of the Old English period does not signal the end of noncontrastive topicalization as a general trend. If all noncontrastive topicalization was a result of FM, then Speyer’s hypothesis predicts a sudden spike in the proportion of topicalized objects which have a contrastive interpretation, around the time that

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i



Caitlin Light

FM is supposed to disappear. However, Stevens and Light find that the proportion of topicalized objects with a possible contrastive interpretation remains stable through every period of English, and perhaps more surprisingly, never approaches 100 per cent. Stevens and Light present a clear problem for Speyer, given his prediction that English had only (contrastively) accented topicalization following the Old English period. Demonstrative pronouns, in particular, appear (under our current assumptions) to represent a class of elements which continue to topicalize without a prosodic accent throughout the history of English and into the present day. But is this grounds to reject Speyer’s theory entirely? Certainly, given that it otherwise assigns an attractive and plausible narrative to the history of English topicalization, we would hope not. .. Proposal I will argue that this puzzle may be solved in such a way as to repair and maintain Speyer’s overall account of topicalization in English. The apparent problem is due to a flawed assumption about the information-structural status of demonstratives: namely, the assumption that demonstrative pronouns behave like unaccented, noncontrastive pronouns. In fact, my ultimate conclusion will be that in those cases where we would not traditionally analyse these demonstratives as contrastive, they do in fact signal a pragmatically contrastive interpretation. In fact, this is what fundamentally distinguishes demonstrative pronouns from personal pronouns on the pragmatic level. The solution to this puzzle will be found in a unified account of topicalization in Germanic. Differences in the topicalization patterns of individual Germanic languages may be explained by the availability or absence of Formal Movement, and on certain restrictions which may be imposed on the types of elements that undergo FM. However, as I will show, the availability of True A-Bar Movement is constant in all of the languages considered. The crucial observation for our puzzle will be that all demonstrative pronouns, even those not initially analysed as contrastive, are in fact eligible for topicalization via the information-structurally driven movement, TAB.

. Corpus data on syntax and information structure The work in this chapter will be based largely on evidence from parsed synchronic and diachronic corpora of various languages. Data of this sort presents a unique type of evidence which will seriously impact our understanding of the phenomenon. There is a certain limit to the amount we can learn about information structure by using artificially constructed example sentences. Context is crucial to understanding the properties of such phenomena, but if we are basing our conclusions entirely on ‘planned’ contexts, then we are necessarily limited to the contexts we can imagine might be relevant. This study, accordingly, will attempt to synthesize corpus data with speaker intuitions and existing experimental data on the related phenomena. If we can

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

The pragmatics of demonstratives in Germanic



find any patterns which hold across a range of data types, we can be certain they are something of significance. The corpus-based data mentioned in this chapter will be drawn from a number of sources. English data will be taken from the Penn–Helsinki Parsed Corpus of Early Modern English (PPCEME; Kroch et al. 2004). For other Germanic languages I use the the Icelandic Parsed Historical Corpus (IcePaHC; Wallenberg et al. 2011) and my own hand-parsed corpus of Early New High German (Light 2011), which consists of a fully parsed and hand-corrected section of Martin Luther’s first translation of the New Testament, called the Septembertestament (1522). This corpus contains roughly 100,000 words of text. I have used corpora for Icelandic and German which also adhere to the Penn annotation guidelines used for the PPCEME. This allows for more consistency in the comparison of the corpus results. The motivation for using these particular corpora also relates to their ability to form, together, a small parallel parsed corpus of New Testament translations. Samples of New Testament translations known to have been influenced by Luther’s New Testament are available in both the IcePaHC and the PPCEME. The result is a smaller subcorpus of parallel New Testament translations, aligned by influence, style, and time period, across English, Icelandic, and German. This corpus allows a more fine-grained comparison of these languages because they allow comparison of the syntactic choices made within individual verses. This is particularly useful in the study of information structure, because we may assume a relatively consistent context across the three translations. .. Coding for information-structural or pragmatic properties At this time, none of these corpora contain pragmatic or information-structural annotations. Once the relevant sentence types have been collected, I examine individual examples and code them by hand for certain properties which are relevant to the current study. A key notion for this work will be that of contrast. As has been established, I will eventually propose a broadening of the category of contrastive elements to include demonstrative pronouns as a group. However, I will not begin by assuming that these elements are contrastive. The idea, rather, is to outline the behaviour of those fronted elements which are unarguably contrastive. We can then build from there to see how the behaviour of demonstrative pronouns is related. In analysing the data, I will begin by identifying contrast according to a relatively traditional definition, and will attempt to impose strict guidelines to characterize any given example as contrastive. I will code constituents as having a plausible contrastive interpretation only if there exist entities in the discourse which are plausibly treated as forming a partially ordered set with the referent of the constituent in question. An example to demonstrate this is given in (5). Two types of food (bread and fish) are explicitly introduced as members of a set by a direct object in the preceding context, die funff brott vnd zween fisch (‘the five [loaves of] bread and two fish’). The narration continues to describe what Jesus does with each member of this set individually, thereby placing them in a plausibly contrastive relationship.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i



Caitlin Light

(5) Vnd er nam die funff brott vnd zween fisch ... and he.nom took the.acc five.acc bread.acc and two.acc fish.acc vnnd brach die brott, vnd gab sie den iungern ... and broke the.acc bread.acc and gave it.acc the.dat disciples.dat vnnd die zween fisch teylet er vnter sie alle and the.acc two.acc fish.acc split he.nom under them.acc all.acc ‘And he took the five loaves and the two fish . . . and broke the bread and gave it to the disciples, . . . and the two fish he split among them all.’ Perhaps against expectations, criteria of this sort make the coding of (plausible) contrast relatively straightforward. In any case where a set has not been explicitly established in the recent discourse, the example is assumed to lack a plausible contrastive interpretation according to a relatively uncontroversial definition of contrast.

. Types of topicalization Previous work on topicalization in a language like German has grappled with the contrast between cases such as (6), in which topicalization is associated with a marked (contrastive) interpretation on the topicalized object, and cases such as the one in (7), where the dative experiencer is topicalized with a default accent pattern and no marked interpretation. (6) Fast jeden Kollegen schätzt der Hans. almost every.acc colleague values the.nom Hans ‘Hans values nearly every colleague.’ (7) Dem Karl hat das Spiel gut gefallen. the.dat Karl has the.nom match well pleased ‘Karl liked the match very much.’ Many researchers have taken data of this sort as a reason to conclude that topicalization can affect various information-structural categories. The best that can be done in this case would be to list the types of categories that are observed to topicalize. Proposing a single construction or movement to cover the full breadth of categories eligible for topicalization is difficult to say the least, especially without constructing an analysis that is highly stipulative. By Frey’s theory, the dissimilarity is not surprising: we are actually observing two different phenomena. In cases where topicalization can be argued to be informationstructurally motivated, and is accompanied by a contrastive interpretation on the topicalized object, it is a case of TAB. Where this interpretation is lacking, the topicalization is assumed to be a case of FM, and to have no information-structural significance; it is a movement entirely driven by the need to satisfy the V2 constraint in the syntax. The Formal Movement hypothesis is attractive because it helps to explain why subjects occur in sentence-initial position by default,2 despite the fact that, under the 2 This is well-known, but trivially, is also confirmed by corpus data. The Septembertestament corpus confirms that while subjects appear before the finite verb in . of their total appearances (leaving only . of subjects in a low position), objects only topicalize . of the time.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

The pragmatics of demonstratives in Germanic



assumption that the finite verb moves as high as the C domain in German, this is not a subject position. Even more interestingly, some elements behave like subjects exclusively with respect to the V2 constraint: namely, those as in (7) above. It has been argued, notably by Sigurðsson (2004), that oblique experiencers in German do not have subject properties, which makes it surprising that they seem to occur sentence-initially by default, as subjects do. Frey (2006a) proposes instead that this can be explained if we assume that dative experiencers are base-generated higher than the subject. As FM targets the highest available constituent, these dative experiencer arguments would, by default, be prime candidates for unmarked topicalization. For arguments other than these to be eligible for Formal Movement, they need to be able to move leftward, to a position structurally higher than other arguments that may be targeted by FM, prior to topicalization. In German, the primary movement operation which allows reordering of arguments in the Middlefield is scrambling. The FM hypothesis predicts, therefore, that the availability of movement operations like scrambling affects the availability of arguments to topicalize without a contrastive accent or interpretation; in the same vein, it predicts that the absence of scrambling as an option will restrict the argument’s ability to undergo noncontrastive topicalization. .. Topicalization across Germanic Although Frey’s proposal is initially intended exclusively for German, it can be expanded to account for topicalization across Germanic. The details of this are discussed further in Light (2012), but can be summarized as follows: two mechanisms for topicalization occur in Germanic, related by the nature of the V2 constraint. Only one, TAB, is an information-structurally driven movement. Observed cross-linguistic variation in topicalization can be explained if we assume this to be the case. The apparent variation does not reflect an underlying difference in the syntactic properties of object topicalization across Germanic. Rather, the observed variation is due to independent differences in the syntax of these languages: specifically, whether the language is V2, and what scrambling-like operations it possesses, will determine the availability of FM of different categories. A general summary of cross-Germanic patterns of topicalization is described in Chocano (2012). Object topicalization (or, in Chocano’s terminology, ‘fronting’) is reported to occur with different information-structural categories in different languages. While English and Dutch front contrastive topics only, German is observed to front contrastive topics, ‘aboutness’-shift topics, and continued topics. What Chocano describes is actually evidence of variation in the availability of Formal Movement. German has clear FM of noncontrastive objects. Other V2 languages with a more restricted scrambling system show very little topicalization of this type, including Icelandic and Dutch (as elaborated in Light 2012). English, meanwhile, is not V2, and thus does not have FM at all. .. Testing the analysis As we consider the distinction between topicalization via FM and topicalization via TAB, pronouns are a particularly telling source of data. According to Frey’s hypothesis, the topicalization of unaccented elements should only be possible via FM, because

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i



Caitlin Light

TAB is specifically associated with a contrastive interpretation (and thus a contrastive accent) on the topicalized constituent. Pronouns are resistant to accent, being light elements that carry salient, discourseold information. Therefore, in the absence of a clear contrastive or focal interpretation, pronouns can be assumed to bear no prosodic accent. I will begin with the assumption that where no plausible pragmatic source of accent can be identified, topicalized pronouns present unambiguous examples of unaccented topicalization. If Frey’s hypothesis is correct, we predict that these examples must be plausible cases of topicalization by FM. Because FM is restricted to moving only the highest available constituent below the finite verb, this would require them to be able to appear sufficiently high in the structure to be targeted. I consider this hypothesis in German by looking in depth at the entire set of topicalized pronouns available in the Septembertestament corpus. The corpus includes fifty-one topicalized object pronouns: thirty-three accusative and eighteen dative. As I will show, the majority of these examples are of exactly the type predicted to occur if we assume FM. First, a number can be set aside as trivially consistent with Frey’s analysis. Thirteen (25.5) of the examples involve an unambiguous contrastive accent on the topicalized object pronoun. In these, the preceding context makes the contrastive interpretation clear: consider (8) as an example. These examples are likely to have topicalized via TAB. (8) Die wellt kan euch nicht hassen, mich aber the.nom world.nom can you.acc.pl not hate me.acc prt hasset sie hates it.nom ‘The world cannot hate you, but it hates me.’ (John 7:7) Of the remaining examples, fourteen are of exactly the type of oblique argument Frey predicts to behave like subjects with respect to FM. Eight are datives in a ditransitive passive construction. Another six (four dative and two accusative) are the oblique experiencers of psych verbs, such as those discussed in Section 12.4. These kinds of examples account for 27.5 of all topicalized object pronouns, and 66.7 of topicalized dative pronouns. Two of the remaining examples occur with a quantified subject, as shown in (9). (9) vnd euch wirtt nichts vnmuglich seyn and you.dat will nothing.nom impossible be ‘And nothing will be impossible to you.’

(Matthew 17:20)

Quantified subjects are plausibly quite low, making it reasonable to hypothesize that, prior to fronting, the pronominal objects in these clauses were in a higher position in the Middlefield. Judgements by native speakers of contemporary German confirm that the subordinate clause equivalent of these examples allows the object pronoun at the left edge of the Middlefield. We are left with twenty-two examples, 43.1 of the dataset. All of them are demonstrative pronouns which lack a plausible contrastive interpretation, based on the current methodology.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

The pragmatics of demonstratives in Germanic



. The special status of demonstratives Frey’s analysis predicts that FM is the only way to topicalize noncontrastive/unaccented elements. According to our current expectations, this should include the demonstrative pronouns without a ‘traditional’ contrastive interpretation found in the Septembertestament dataset. A prediction follows: the topicalization of these (apparently) noncontrastive demonstratives should only appear in languages that allow FM (that is, V2 languages). Additionally, languages with stricter limitations on the FM of objects, such as Icelandic and Dutch, should not topicalize demonstrative pronouns as frequently as German does. However, the fronting patterns of German, Icelandic, Dutch, and English are more similar than predicted. I will ultimately argue that demonstrative elements are, in fact, eligible for topicalizion via TAB. In this section I will present evidence that leads me to re-evaluate and reject the assumption that these elements are noncontrastive by default. Based on recent studies of the pragmatics of demonstrative pronouns in Germanic, I will suggest that the effect of these elements on the discourse structure relies fundamentally on the notion of alternatives. For this reason, I will revise our original assumption, and argue that demonstrative pronouns should be treated as pragmatically contrastive. It follows that its members may be fronted via TAB, and thus do not present a problem for the Formal Movement hypothesis. .. Topicalized demonstratives in the parallel New Testament corpora A fine-grained comparison of the fronting behaviour of demonstrative pronouns is possible in the parallel parsed corpora of Bible translations. William Tyndale’s bible translation of Early Modern English, published 1534, and Oddur Góttskálksson’s translation of Icelandic, published 1540, were both substantially influenced by the Luther version represented by the Septembertestament corpus. Contra to my initial prediction based on the FM hypothesis, all three languages have parallel examples of noncontrastive topicalized elements, such as those shown in (10) and (11). (10) a. disen this.acc wyr we.nom b. þennan this.acc vér we.nom

Jhesum hat Gott auff erweckt, des sind Jesus.acc has God.nom up awakened who.gen are alle zeugen. all.nom witnesses.nom Jesúm upp vakti Guð, hvers vottar Jesus.acc up woke God.nom who.gen witnesses.nom erum allir. are all.nom

c. This Jesus hath God raysyd vp, wher of we all are witnesses. (11)

(Acts 2:32)

a. Das saget er aber, zu deutten, wilchs todts er that.acc said he.nom prt to indicate which.gen death he.nom sterben wurde die would

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i



Caitlin Light

Table .. Topicalization of different DP types in the IcePaHC corpus

Topicalized Non-topicalized  Topicalized

Full DPs

Pronouns

Demonstratives

1000 13312 6.99

531 3961 11.82

131 209 38.53

Table .. Topicalization of different DP types in the PPCEME corpus

Topicalized Non-topicalized  Topicalized

Full DPs

Pronouns

Demonstratives

469 12491 3.62

39 3536 1.09

130 78 62.50

b. En þetta sagði hann teiknandi með hverjum dauða eð but this.acc said he.nom illustrating with which.dat death c hann skyldi deyja. he should die c. This sayde Iesus signifyinge what deeth he shuld dye. (John 12:33) This phenomenon cannot be dismissed as a translation effect. A consideration of demonstrative topicalization in the entire Icelandic and EME corpora confirms that demonstrative topicalization is commonplace in both languages. In Icelandic, demonstrative pronouns topicalize at a rate higher than any other argument type, as the data in Table 12.1 demonstrate. The same is true of Early Modern English, shown in Table 12.2. In fact, in EME, demonstrative pronouns topicalize more often than not, which is not true of any other DP type.

. Analysing demonstratives The data from these corpora provides justification to more closely investigate the notion of contrastivity that is relevant to topicalization. By digging deeper, we may be able to pinpoint what it is about demonstrative pronouns that causes these unexpected topicalization patterns. For this, I first look back to an analysis of Yiddish topicalization presented in Prince (1999). Prince presents a detailed proposal for the pragmatic motivation of object topicalization in English and Yiddish, based on evidence from naturally-occurring

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

The pragmatics of demonstratives in Germanic



examples. What is relevant for us is Prince’s characterization of contrastivity in her analysis. Her definition of topicalization in Yiddish is presented in (12). (12) Double Discourse Function of Yiddish Topicalization a. Yiddish Topicalization triggers an inference on the part of the hearer that the entity represented by the initial NP stands in a salient partially-ordered set (‘poset’) relation to some entity or entities already evoked in the discoursemodel. b. Yiddish Topicalization triggers an inference on the part of the hearer that the proposition is to be structured into a focus and a focus-frame as follows. First, if the entity evoked by the leftmost NP represents an element of some salient set, make that set-relation explicit. Then, in all cases, the open proposition resulting from the replacement of the tonically stressed constituent (in the clause!) with a variable is taken to represent information that is known or at least plausible to the hearer at that point in the discourse, the tonically stressed constituent representing the instantiation of the variable and the new information in the discourse. (Prince 1999: 14) A crucial point for us is contained in (12a): the requirement that the topicalized entity is placed in contrast specifically to entities that have been evoked in the discourse. Prince’s account of ‘Topicalization’ therefore hinges on a certain pragmatic notion of contrastivity, relying on contextually salient sets. Furthermore, one of the effects of topicalization outlined in (12a) is to bring that set relation to the attention of the hearer. Prince does not address how our apparently noncontrastive demonstratives fit into this analysis. However, she was aware of the potential problem: elsewhere, she proposed that demonstrative pronoun topicalization may be treated as selection from a singleton set (Anthony Kroch, p.c.). Under her analysis, these elements are not contrasted with other elements in the set, but may be thought of as contrasted in some sense with all non-members of the set. Thus, the entity referred to by the demonstrative would be the entirety of the salient set evoked in the discourse-model. There are obvious problems with this proposal as it stands. It seriously changes our idea of contrastivity: as C. Roberts (1996[2012]: 48) points out, ‘It is generally assumed that alternative sets, like operators’ domains, are nonsingleton, nonempty.’ Allowing singleton sets into the definition of contrastivity makes the notion dangerously close to vacuous. However, I believe that there are some crucial truths in the discourse function of topicalization as Prince (1999) sketched it, which may be expanded past Yiddish to provide insight into the phenomenon in Germanic more generally. By taking this as a starting point for investigating the behaviour of demonstrative pronouns, I hope to gain insight into how they might fit into this notion of contextually established alternatives as Prince envisioned it. .. Demonstrative pronouns and reference resolution Bosch et al. (2003) describe a distinction between personal and demonstrative pronouns in German. Their initial observation is that in discourse fragments like (13), the choice of a personal or demonstrative pronoun in the second sentence will affect the

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i



Caitlin Light

meaning of that sentence: the personal pronoun will prefer the subject of the preceding sentence, Paul, as its antecedent, while the demonstrative will prefer the available nonsubject antecedent, Peter. (13) Pauli wollte mit Peterk laufen gehen. Aber Paul.nom wanted with Peter.dat run go but / derk ] war erkältet. [eri he.pro.nom he.dem.nom was caught-cold ‘Paul wanted to go running with Peter. But he had a cold.’ Furthermore, Bosch et al. (2007) show that speakers demonstrate a bias when evaluating pronouns in postverbal (non-topicalized) position. Postverbal pronouns were overwhelmingly judged as more natural when realized as personal pronouns, and less natural as demonstrative pronouns, suggesting that demonstrative pronouns are most natural in the preverbal (topicalized) position. The authors conclude that while the grammatical status of the antecedent may be a related factor, it is fundamentally information structure that determines the choice between a personal and a demonstrative pronoun. Bosch and Umbach (2007) pursue this claim further. They demonstrate that the properties of the antecedent, and particularly the grammatical status of the antecedent, are not the deciding factor in personal and demonstrative pronoun usage. Both pronouns and demonstratives can be used without any antecedent whatsoever, which means that defining the properties of these elements based solely on the properties of their antecedents is not sufficient. They also demonstrate that, if the context is manipulated, the demonstrative can happily pick up a subject antecedent, confirming that grammatical status of the antecedent is not the core concern. das weiß? Peterk hat es ihmi (14) Woher Karli where-from Karl.nom that.acc knows Peter.nom has it.acc him.dat / Erj,k ] war gerade hier. gesagt. [Derk he.dem.nom he.pro.nom was recently here said ‘How does Karl know? Peter told him. He was just here.’ In addition, if the right context is supplied, the demonstrative may clearly be dispreferred to refer to a nonsubject antecedent, even when no other antecedent is available in the discourse. In (15), the use of the demonstrative is simply odd, while the use of the personal pronoun is perfectly acceptable. das weiß? Peter hat es ihri (15) Woher Mariai where-from Maria.nom that.acc knows Peter.nom has it.acc her.dat / Siei ] war gerade hier. gesagt. [? Diei she.dem.nom she.pro.nom was recently here said ‘How does Maria know? Peter told her. She was just here.’ Hinterwimmer (2015) further explores the information-structural properties of demonstrative pronouns in German. He argues, following Bosch and Umbach (2007), that the central factor in determining the referential properties of demon-

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

The pragmatics of demonstratives in Germanic



strative pronouns is not the grammatical role of the antecedent, but rather its information-structural role: crucially, demonstratives may not refer to a topical entity. Hinterwimmer hinges his analysis on the notion of aboutness topic as defined by Reinhart (1981). But in fact, as Schwarz (2015) points out, it is not so simple: aboutness topics can be referred to in contexts like (16). Here, the referent Maria is contextually identified as an aboutness topic by the phrase was ist mit Maria? (‘what about Maria?’). However, the context also places Maria in contrast with another salient entity, Anne. Thus, in this context, we have a possible referent who is both topical (in the aboutness sense) and contrastive. (16) a. Die meisten Leute haben Harry Geschenke mitgebracht. Anne hat ihm zum Beispiel ein Bild geschenkt. ‘Most people brought Harry presents. For example, Anne gave him a picture.’ b. Und was ist mit Maria? Was hat sie Harry gegeben? ‘And what about Maria? What did she give Harry?’ (i) #Dem hat sie ein Hemd gegeben. him.dem.dat has she.nom a.acc shirt.acc given ‘She gave him a shirt.’ (ii) Die hat ihm ein Hemd gegeben. she.dem.nom has him.dat a.acc shirt.acc given ‘She gave him a shirt.’ (iii) Sie hat ihm ein Hemd gegeben. she.nom has him.dat a.acc shirt.acc given ‘She gave him a shirt.’ The demonstrative pronoun may refer to Maria without issue in this case (note that it cannot refer to Harry, for comparison). If the contrastive accent is removed from sie (‘she’) in the preceding sentence, then the demonstrative is no longer grammatical. Schwarz suggests that something more nuanced than topicality is involved with the reference resolution of demonstratives. Schwarz offers a counterproposal in the framework of situation semantics. I believe it can be most clearly described, and further developed, using the discourse tree (or d-tree) model set out in Büring (2003). .. Contrast and the d-tree Büring (2003) proposes a discourse-oriented model to represent the theory of contrastive topicality (and focus) developed earlier in Büring (1994, 1997). Where the previous model was exclusively semantic, defining contrastive topics in terms of sets of sets of propositions, Büring’s later work puts the bulk of the load into the pragmatics. The discourse structure proposed is similar to models proposed by Klein and von Sutterheim (1987); van Kuppevelt (1991, 1995, 1996); C. Roberts (1996 [2012]). Büring models discourse as a discourse tree (or d-tree), whose nodes are composed of broad questions under discussion (QUD), subquestions which contribute to the answer of the broader QUD, and answers to subquestions, in a structure like the one shown in Figure 12.1.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i



Caitlin Light discourse question subq

subq

answer

answer

question

subq

subq

subsubq

subsubq

answer

answer

…

answer

Figure .. Structure of the discourse tree Source: Büring 2003.

What did the pop stars wear? What did the female pop stars wear?

What did the male pop stars wear?

The femaleCT pop stars wore caftansF.

…

Figure .. Implicit moves in the d-tree Source: Büring 2003.

Every node in a d-tree is called a move (Carlson 1983), and represents an utterance in the discourse, whether an assertion or question. Crucially, Büring assumes that each move is a syntactic phrase marker representing a declarative or interrogative sentence. Büring also allows for the possibility of implicit moves. For example, in the following discourse (labelled as a partial topic in Büring 1997), the second speaker answers only a (potential) subquestion of the question asked: (17) Q: What did the pop stars wear? A: The femaleCT pop stars wore caftansF . Example (17) can be represented by a discourse tree like the one in Figure 12.2, in which the explicit moves are given in bold. The subquestion which can be seen between the explicit moves is thus treated as an implicit move in the discourse. Within a d-tree, a question and the subquestion(s) belonging to it are cumulatively called a strategy. Contrastive topics, in this system, can be analysed as a tool to indicate the strategy being used: for example, in Figure 12.2, contrastive marking on the female pop stars is used to indicate that the speaker is answering a broader QUD (What did the pop stars wear?) by addressing individual subquestions addressing subsets of the set of pop stars. Using Büring’s (2003) discourse model, and drawing on the situation-semantic analysis in Schwarz (2015), I analyse demonstrative pronouns as follows: (18) Demonstrative pronoun reference: The referent of a demonstrative may not appear in all the possible answers of a strategy. Consider, for example, the example in (19). An intuitive assumption is that the QUD being addressed in this context is ‘Who examined the patient?’ The low subject of the

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

The pragmatics of demonstratives in Germanic



first sentence, der Chefarzt, is taken to be in focus, as the answer to this question; as a result, it cannot possibly be in every possible answer of the strategy, because it is in itself one possible answer (one possible selection from the focus set). In comparison, the object den Patienten is necessarily represented in every possible answer to the strategy, because it is represented in the QUD itself. Chefarztk . Derk (19) Den Patienteni untersucht der the.acc patient.acc examined the.nom head-doctor.nom he.dem.nom ist nämlich Herzspezialist. is namely heart-specialist.nom ‘The head doctor examined the patient. He is a heart specialist.’ The possible answers are restricted by the sorts of discourse constraints which have been discussed extensively in the pragmatics literature, and are in the same spirit as the cooperative principle proposed by Grice (1975). For example, Büring defines a version of Grice’s principle of Relevance relative to the QUD: For any move M, the question under discussion is the move M immediately dominating it. For a move to be relevant, it must answer or at least address this question under discussion. I will resort to the rather vague formulation that A is an answer to Q if A shifts the probabilistic weights among the propositions denoted by Q. (Buring 2003: 517)

Thus, a well-formed strategy will allow only those answers which are relevant to the immediate QUD. Such a constraint helps to limit the set of possible answers within a strategy. Within these possible answers, the demonstrative pronoun’s referent may be represented in only some subset. This type of discourse function is described by Gast (2010) as subinformativity. A subinformative move in the d-tree is one which addresses only some part of the current QUD. As Gast (2010: 24–5) points out, ‘the most important property of sentences with contrastive topics is that they are in some way “subinformative” relative to some strategy they are contained in, insofar as they leave open questions.’ One may ask what it means to suggest that demonstrative pronouns signal subinformativity in the discourse structure. I will argue that the demonstrative pronoun may be used to amend the existing QUD. Recall that in the case of implicit moves, the existence of questions or subquestions may be signalled by the use of a contrastive topic. As Büring (1997) points out, the most extreme version of this is what he calls purely implicational contrastive topics, as in (20). (20) Q: Did your wife kiss other men? A: [My]CT wife [didn’t]F kiss other men. Here, a relatively straightforward, explicit QUD is answered by an utterance containing a contrastive topic. It is clear that the answerer in this example is implying that the questioner should worry about somebody else’s wife (perhaps his own). The specific function of the contrastive topic is to signal to the hearer that there are additional subquestions which were not assumed by the original QUD but which ought to be added to the discourse structure. The contrastive topic is being used to revise the original QUD.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i



Caitlin Light

The demonstrative pronoun is capable of a similar effect. Let us reconsider an example discussed in previous sections, (13), repeated in (21). As I have discussed previously, the choice of a personal or demonstrative pronoun in the second sentence will determine the preferred referent of that pronoun. (21) Pauli wollte mit Peterk laufen gehen. Aber Paul.pro.nom wanted with Peter.dat run go but / derk ] war erkältet. [eri he.pro.nom he.dem.nom was caught-cold ‘Paul wanted to go running with Peter. But he had a cold.’ What happens if we build further on this context? Consider (22), which adds an additional sentence to the context given in (21). There is some preference for the personal pronoun in the third clause (er) to refer to Paul.3 wollte mit Peterk laufen gehen. Aber derk war (22) Pauli Paul.nom wanted with Peter.dat run go but he.dem.nom was war sehr enttäuscht. erkältet. Eri caught-cold he.nom was very disappointed ‘Paul wanted to go running with Peter. But he had a cold. He was very disappointed.’ Adjusting the context to more clearly favour Paul confirms this judgement.4 wollte mit Peterk laufen gehen. Aber derk war (23) Pauli Paul.nom wanted with Peter.dat run go but he.dem.nom was ging sowieso laufen. erkältet. Eri caught-cold he.nom went anyway run ‘Paul wanted to go running with Peter. But he had a cold. He went running anyway.’ However, if the context is adjusted to favour Peter, the judgement flips. (24) Pauli wollte mit Peterk laufen gehen. Aber derk war Paul.nom wanted with Peter.dat run go but he.dem.nom was hatte hohes Fieber. erkältet. Erk caught-cold he.nom had high fever ‘Paul wanted to go running with Peter. But he had a cold. He had a high fever.’ It appears that the use of the demonstrative pronoun in the second sentence allows either Peter or Paul to be interpreted as the referent of a personal pronoun in the continuing discourse. Context becomes the deciding factor after this point. Compare the parallel cases, in which the second clause contains a personal pronoun instead of a demonstrative pronoun. Now, the personal pronoun in the third clause

3 However, I have also been told by one speaker that this context can allow the third pronoun to refer to either Peter or Paul fairly easily. 4 In this case, one speaker finds the third clause simply infelicitous; interestingly, however, this speaker agrees with the judgement reported for (), making the reason for rejecting this example unclear.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

The pragmatics of demonstratives in Germanic



will refer to Paul even when context would strongly disfavour Paul as the referent, as in (26). (25) Pauli wollte mit Peterk laufen gehen. Aber eri war Paul.nom wanted with Peter.dat run go but he.nom was war sehr enttäuscht. erkältet. Eri caught-cold he.nom was very disappointed ‘Paul wanted to go running with Peter. But he had a cold. He was very disappointed.’ (26) Pauli wollte mit Peterk laufen gehen. Aber eri war Paul.nom wanted with Peter.dat run go but he.nom was ging sowieso laufen. erkältet. Eri caught-cold he.nom went anyway run ‘Paul wanted to go running with Peter. But he had a cold. He went running anyway.’ wollte mit Peterk laufen gehen. Aber eri war (27) Pauli Paul.nom wanted with Peter.dat run go but he.nom was hatte hohes Fieber. erkältet. Eri caught-cold he.nom had high fever ‘Paul wanted to go running with Peter. But he had a cold. He had a high fever.’ I propose that the discourse structure becomes more complex following the use of the demonstrative pronoun. Given the initial sentence of the context, hearers initially expect that we are pursuing a QUD roughly equivalent to the question What can you tell me about Paul? The entity Paul is consequently part of every question and potential subquestion of the assumed strategy. If the speaker proceeds using the personal pronoun, then this strategy is accepted as correct. However, the speaker may also use the demonstrative pronoun in the second sentence. By convention, this move is expected to refer to an entity not represented in every possible answer to the given strategy, which in this case would be Peter. The use of the demonstrative pronoun is, in effect, a signal that the strategy should be altered from the one assumed based on the first sentence. The speaker signals that answers to a QUD like What can you tell me about Peter? will also be relevant. If this analysis is approaching truth, a question invites itself: what happens if a demonstrative pronoun is used when only one referent is available? Consider, for instance, example (28).5 (28) Ich habe heute Peteri getroffen. Eri / Deri/? war I.nom have today Peter.acc met he.nom he.dem.nom was leider sehr in Eile. unfortunately very in hurry ‘Today I met Peter. Unfortunately he was in a great hurry.’

5

Thanks to an anonymous reviewer for suggesting this example.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i



Caitlin Light

Unsurprisingly, using the personal pronoun in (28) is unproblematic, and refers easily to Peter. If the demonstrative pronoun is used instead, however, the situation seems slightly trickier. When asked, a native speaker of German reported that it was possible to refer to Peter with the demonstrative pronoun, but in this case it suggests that there is ‘more to follow’—‘it’s a set-up for the “real” story you’re going to tell me,’ in the words of the consultant. Remarkably, this sounds like a neat, intuitive description of the proposal that the demonstrative pronoun leads to a revision of the QUD. I propose that it signals a switch to a discourse strategy that focuses on Peter, rather than on the first-person narrator.6 Ultimately, the demonstrative pronoun and the contrastive topic may both be used as conventionalized signals of a more complex discourse structure. Each may signal a relatively specific revision of the assumed QUD, and thereby demonstrate to the hearer how the information currently being conveyed is relevant to the discourse. Because both contrastive topics and demonstrative pronouns influence the discourse by drawing attention to the current QUD or sub-QUD and by introducing the idea of alternatives which contribute partial answers to a broader QUD, both fit into Gast’s notion of subinformativity, and both are similarly contrastive. Framed more intuitively, the pragmatic function of demonstrative pronouns can be conceptualized like this: demonstrative pronouns exist as a marked alternative to the default pronoun set, personal pronouns.7 Demonstrative pronouns are used to refer to the unexpected referent (hence the intuition that demonstratives refer to nontopical referents). By nature, unexpected answers presuppose expected answers, and thus a set of alternatives (expected and unexpected) is necessarily established.8 When we say

6 The consultant also suggested that der in this case may refer to another, additional referent mentioned previously in the discourse, who perhaps hadn’t been rushing (therefore placing Peter in a traditionally contrastive relationship with this hypothetical referent). In other words, the listener is actively seeking some way to interpret the demonstrative pronoun contrastively, even if the provided context is insufficient to support it. 7 In discussing the shared roles of different pronoun sets, it is important to note the possibility of overlap. An anonymous reviewer has pointed out that ‘most “constrastive” jobs can be done by strong/stressed (contrastively focused) pronouns.’ This is certainly the case, and fits my intuition that different pronoun sets are adapted to discourse strategies as needed. For example, in English, we would certainly expect personal pronouns with human referents to more frequently serve this purpose, since demonstrative pronouns are disallowed in these contexts; similarly, in German we expect demonstrative pronouns to more often serve the role of constrastivity when pronouns like es (‘it’) are not capable of bearing stress. Elsewhere, however, it would not be surprising to see some overlap between the roles of demonstratives and stressed pronouns in expressing pragmatic contrast, or other ‘marked’ interpretations. 8 An anonymous reviewer raises the question of how this analysis would deal with complex demonstratives being used with a descriptive intention, as described in King (): if the demonstrative performs to evoke a set of alternates in the discourse, then how can this take place if the demonstrative does not refer to any specific entity, but rather to whichever satisfies an established property? The point is appreciated, and suggests to me an interesting avenue for future exploration. On an immediate glance, topicalized complex demonstratives in the corpora are generally not of the type that suggests descriptive intention, but rather perceptual intention. It may be that complex demonstratives with descriptive intention should be considered separately as pertains to their discourse function; this requires more investigation.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

The pragmatics of demonstratives in Germanic



that a demonstrative pronoun signals ‘marked’ or ‘unexpected’ information, what we really mean is that it signals a revision of the original QUD or discourse strategy. Like contrastive topics, demonstrative pronouns are conventionally used to draw attention to the moves available in a given discourse tree, and to signal the strategy that the speaker has chosen. This hereby gives us the common ground shared by traditionally recognized contrastive elements and demonstrative pronouns which would not be traditionally identified as ‘contrastive.’

. Conclusion To conclude, I revisit the historical puzzle that began this investigation. Recall that Speyer (2010) presented a unified account of topicalization across the history of English, which hinged on the assumption that only TAB, or information-structurally driven topicalization, was available following the Old English period. Stevens (2010) and Stevens and Light (2012), however, found a stable rate of demonstrative pronoun topicalization across the history of English, which appeared to challenge Speyer’s analysis under the assumption that noncontrastive demonstratives are pragmatically parallel to unaccented, noncontrastive personal pronouns. We now find that, in fact, these observations are just as expected given the fact that demonstrative pronouns are inherently contrastive in their pragmatic function. As a result, they are naturally eligible to topicalize via TAB.9 I conclude that, if all Germanic languages are considered, all have TAB in common. As hypothesized, contrastivity is the key trigger for this topicalization strategy; what was necessary was merely a re-evaluation of what counts as ‘contrastive’ for information-structurally driven topicalization. In exploring this possibility, we are led to a clearer understanding of the behaviour of demonstrative pronouns in discourse. Ultimately, I find that a detailed and thorough consideration of the syntax–pragmatics interface leads to insight into both crucial details of the grammar, and how these issues relate to language in use.

Acknowledgements Thanks to Anthony Kroch, Robin Clark, Florian Schwarz, Joel Wallenberg, and a variety of past and present Penn Linguistics students for their help and support in preparing this research. Additional thanks to Beatrice Santorini and Tatjana Scheffler for providing German judgements. 9 Note that while I confirm Speyer’s claim with respect to TAB, I do not confirm his further claim that Old English must have had Formal Movement. This hinges on the fact that Old English was not V of the type required for FM to be possible (see for example Pintzuk ; Fischer et al. ; Haeberli ). The appearance of clauses which are superficially ‘verb-second’ is not sufficient to motivate the existence of Formal Movement in Old English, and we do not expect topicalization via FM otherwise. I agree with Speyer that the higher rate of pronoun fronting in Old English must be associated with a different phenomenon. I differ from him in claiming that the phenomenon in question is not Formal Movement, but leftward scrambling of the pronominal object, which was possible in Old English but subsequently restricted (cf. Wallenberg ). A more detailed discussion of this issue is beyond the scope of the current chapter.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

 Persistence as a diagnostic of grammatical status: The case of Middle English negation A A R O N E C AY A N D M E R E D I T H TA M M I N G A

. Introduction The field of diachronic syntax encompasses the analyses of both historical grammatical structures and the processes by which they change. Processes of change are in many cases easily measured by changes in surface patterns, but without access to native speakers and their grammaticality judgements, theoretical analysis of the syntactic structures of past language stages is at a disadvantage. This analysis is impeded by our inability to collect new data probing crucial minimal pairs or rarely-occurring contexts, as well as by our own lack of intuitions about the variety. As diachronic syntacticians, then, we are pressed to be resourceful and creative in our analytic approaches. One method by which researchers have made headway on the problem of analysing underlying structural unity (or disunity) across surface strings in historical data is by application of the Constant Rate Hypothesis (CRH; Kroch 1989b). In this approach, quantitative data on rates of change across different syntactic contexts is used to infer grammatical analyses. Although the CRH has been applied successfully to a range of cases, it is not the only logically possible solution to the methodological dilemma at hand. In this chapter we propose an independent source of quantitative evidence about historical grammatical analyses: the observation of repetitiveness in variant choice, or persistence. Persistence in corpus data is generally understood to reflect the psycholinguistic mechanism of structural priming, whereby reuse of a recently-processed syntactic structure is facilitated. The reasoning underlying this type of evidence is quite distinct from the application of the CRH, thus allowing a single corpus to generate multiple strains of quantitative evidence. We present a case study, namely the change in negation in Middle English, for which CRH evidence has been extensively investigated without yielding a definitive conclusion. Our argument is that persistence evidence favours one extant analysis over its competitor, thereby demonstrating the utility of persistence as a diagnostic tool that can be used in tandem with other approaches. Micro-change and Macro-change in Diachronic Syntax. First edition. Éric Mathieu and Robert Truswell (eds). This chapter © Aaron Ecay and Meredith Tamminga 2017. First published 2017 by Oxford University Press.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

Persistence as a diagnostic of grammatical status



The chapter is organized as follows. In Section 13.2, we discuss the literature on corpus persistence and describe the assumptions necessary to link written historical data to psycholinguistic processes. In Section 13.3, we introduce the syntactic change that forms our case study, the replacement of the sentence negator ne with not by way of a period where both co-occur. We outline two competing accounts for the structures at play during the period of co-occurence; one account due to Frisch (1997) and the other to Wallage (2008). Section 13.4 first makes explicit the predictions that these two accounts make about the expected persistence behaviour of the negators, then tests these predictions against quantitative data from the PPCME2 (Kroch and Taylor 2000). Finally, in Section 13.5 we discuss the implications of our results and the potential for more widespread application of the methodology we propose here.

. Background: Persistence In quantitative studies of language variation, it is generally assumed that each token is an independent observation of a random variable. While this is a useful simplifying assumption, it is widely known not to hold as strictly as a researcher might hope (for an early discussion, see Sankoff and Laberge 1978; Paolillo 2011 offers a more general critique of independence assumptions in quantitative linguistics). Persistence is the tendency for a recently-used variant of a linguistic variable to be used again—each token is to some extent dependent on the token or tokens before it. Such repetitiveness in variant choice is of interest in its own right, but in this chapter we focus on its use as a tool for understanding the relationships between grammatical strings. By taking persistence as a dependent variable that is sensitive to the structural identity of sequential constructions in written corpora, it is possible to determine whether those constructions are related or not. This section briefly outlines the assumptions necessary to make such a claim and the previous research supporting those assumptions. Early demonstrations of the persistence effect come from spoken corpora. The earliest such study, to our knowledge, is Sankoff and Laberge (1978), which documents a persistence effect on three variables in Montreal French: the pronominal alternations between on and tu/vous for general indefinite human reference, on and ils for exclusive indefinite reference, and nous and on for first person plural. The authors show that in sequences of tokens, speakers switch from one option to the other only about one-third as often as the null hypothesis of total statistical independence would predict. Another influential early demonstration of such an effect comes from Poplack (1980, 1984), two studies of the factors conditioning the variable deletion of inflectional /s/ and /n/ in Puerto Rican Spanish. In these papers, Poplack shows that preceding /s/-deletion within the noun phrase favours further deletion, while preceding retentions disfavour deletion. In her terms, ‘One marker leads to more, but zeros lead to zeros’ (Poplack 1980: 378). Poplack and Tagliamonte, in their study of verbal /s/ in nonstandard English, similarly demonstrate that ‘once a zero is used . . . another zero is most likely’ (1989: 70). One of the earliest corpus studies to identify a persistence effect for a syntactic variable is Weiner and Labov (1983). This study of the constraints governing selection of the active or passive voice, which the authors treat as functional alternants, finds

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i



Aaron Ecay and Meredith Tamminga

that while the external factors of style, sex, class, ethnicity, and age have small-toinsignificant effects on the selection of the active or the passive, the internal effects of information structure (given vs. new) and parallel surface structure (persistence) are large and significant. They conclude that ‘the ordering of surface syntax across clauses is the predominant linguistic influence on this choice’ (Weiner and Labov 1983: 52). This conclusion holds even when excluding from the analysis clause pairs with coreferential subjects. The same variable is revisited in more detail by Estival (1985). More recent corpus-based studies of syntactic persistence include Gries (2005), in which the variables under consideration are the dative alternation and particle placement, and Szmrecsanyi (2006) on future marking, particle placement, complementation, comparatives, and genitives in English dialects. The Gries study is particularly informative because he finds comparable results using written and spoken data sources, thus bolstering the viability of persistence studies in the context of historical syntactic research. The corpus persistence literature is widely understood to be intimately linked with the experimental structural priming literature, in which speakers who have heard a particular syntactic construction are more likely to use that construction in a picture description task. Pickering and Branigan (1999: 136) define structural priming as ‘the phenomenon whereby the act of processing an utterance with a particular form facilitates processing a subsequent utterance with the same or a related form’. The pioneering paper in this domain, Bock (1986), was motivated by the corpus results from Estival (1985) and Weiner and Labov (1983) described above. In Bock’s original methodology, which heavily influenced subsequent studies, participants are exposed to sentences and then asked to describe pictures. The syntactic structure the participant chooses for the description (passive or active, double object or prepositional object) is the experimental target and is dependent on the structures in sentences that the participant is exposed to as primes. Pickering and Ferreira (2008) identify sentence recall, written sentence completion, and spoken sentence completion as additional methodologies that have subsequently been added to the structural priming toolbox. They also provide an overview of the wide range of results on the robustness and sensitivity of structural priming. The experimental structural priming literature has made progress by exploiting priming relationships to answer questions of structural identity. The general mode of reasoning is that experimental observations of priming relationships between different syntactic structures are taken as evidence for the nature of the mental representation of the underlying linguistic objects. Branigan et al. (1995) lay out the argument explicitly: If the processing of a stimulus affects the processing of another stimulus, then the two stimuli must be related along a dimension that is relevant to the cognitive system. Under certain circumstances, we can conclude that they are represented in a related manner. If the relationship between two stimuli is syntactic, then we can use this relationship as a way of understanding what syntactic information is represented, and how that information can interact with other information.’ (Branigan et al. 1995: 490)

Pickering and Branigan also argue this point when they ‘claim that syntactic priming taps into knowledge of language itself, and as such can inform linguistic theories

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

Persistence as a diagnostic of grammatical status



1.00

Rate

0.75

Target lexical transitive

0.50 0.25 0.00 none

lexical Prime

transitive

Figure .. Results from Estival (1985), showing that English lexical and transformational passives prime themselves, but not each other, in conversational speech

that are concerned with accounting for knowledge of language’ (1999: 140). They specifically advocate the use of priming studies to determine relationships between sentences. Taking the argument one step further, Pickering and Ferreira (2008: 429) suggest that in the face of overwhelming evidence for structural priming in many constructions, a failure to observe priming for a specific construction may indicate that the putative prime and target are not representationally related. A handful of studies have taken up these types of grammatical questions. Estival (1985) asks whether the persistence effect holds across different types of passives or whether it shows a within-type restriction. The two types of passives she considers are transformational passives (‘John is believed to have left’) and lexical passives (‘John is interested in music’—the examples are hers). She finds that each type of passive makes it more likely that a passive of its type will be used again, but that the effect does not extend to passives of the other type; this is illustrated in Figure 13.1.1 One early experimental demonstration that priming reflects structural identity beyond simple surface repetition comes from Bock and Loebell (1990). They show through the use of a picture description task that prime sentences with PPs introduced by to facilitate the choice of the prepositional structure over the double-object one in subjects’ subsequent production of ditransitive sentences. This relationship holds regardless of whether the to-phrase in the prime sentence introduces a beneficiary (‘The wealthy widow gave her Mercedes to the church’) or a location (‘The wealthy widow drove her Mercedes to the church’). The same priming is not observed, on the other hand, when the prime contains to functioning as an infinitival marker (‘Susan brought a book to study’). In a more recent paper, Ferreira (2003) investigates the sensitivity of the optional complementizer that to priming by other functions of that. Using a sentence recall task, he shows that when the prime that heads a CP which is the object of a verb as in (1), it boosts the rate at which subjects choose to use that in their productions in just

1 Estival’s study was carried out at a time when the nature of these different passives was in question, but her result is in line with the modern consensus that the two types of passives are structurally distinct (see Embick  for discussion).

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i



Aaron Ecay and Meredith Tamminga

that context (where it is optional). But when the prime is a determiner as in (2) or the head of a noun’s complement as in (3), no such priming effect results.2 (1)

The company ensured that the farm was covered for two million dollars.

(2) The company insured that farm for two million dollars. (3) The theory that penguins built the igloos was completely false. The core methodological premises of this chapter, then, are twofold: that persistence effects observed in written corpora reflect priming effects in the language production of the time period; and that repetition reveals grammatical sameness while lack of repetition indicates grammatical difference. In the context of historical syntax, where we have no access to native speaker judgements or experimental behaviour, there is a need for as many independent sources of evidence as possible. When different sources of evidence for a theoretical analysis converge, it considerably strengthens our confidence in that analysis. We make the case that persistence can be one of these sources of evidence using the case study of a change in sentence negation in Middle English.

. Background: Negation In the Middle English period, there is a change in the exponence of sentence negation (represented as the category Neg in our syntactic diagrams). The negator ne, which was inherited from Old English, is eventually lost, being replaced by not, originally an adverb of emphatic negation.3 During the period when this change is ongoing, there are a large number of sentences which appear with both ne and not, as in the following example. (4) he ne shal nouÇt decieue him (Early Prose Psalter, 161:131:11, from Frisch 1997) The progress of this change is charted in Figure 13.2. .. Frisch () The literature contains two different analyses of the grammatical underpinnings of this change. The first is due to Frisch (1997), who analyses the change as stemming from competition between (just) two grammars. The first grammar, inherited from

2 A reviewer points out that many syntactic theories treat the thats in () and () as the same lexical item. If these theories are taken to be true (and we see no compelling reason not to do so), the lack of priming effects between sentences like () and () needs to be explained. Ferreira’s data indicate that lexical and categorial identity is a necessary but not sufficient condition for syntactic priming. A contextual factor is also at play—that alternates with ∅ in (), but is obligatory in (). Precisely characterizing the contextual conditions on priming effects in terms of grammatical structures is an open question for psycholinguistic research, and clearly has implications for the generality of our proposed corpus-based diagnostic. However in the context of this paper, where we compare two analyses of the same construction rather than looking at the behaviour of lexically identical material across constructions, the issue does not arise. 3 Not could also be used as a DP, with approximately the same meaning as modern nothing. In modern standard English this use of the word is spelled naught or nought.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

Persistence as a diagnostic of grammatical status



Proportion neg. decl.

1.00 Type 0.75

ne not ne…not

0.50 N

0.25

750 1500

0.00 1200

1300 Year

1400

1500

Figure .. The change from ne to not in Middle English, as measured in negative declarative sentences in the PPCME2 Note: Sentences with both ne and not are a separate category from sentences with just one or the other negator. The sizes of the circles represent the total number of negative declarative tokens in a time period (the denominator of the proportion), so triplets of points in the same year are identically sized. See Section 13.4.1 for details on how the dataset was collected.

OE, contains an entry for ne as the head of NegP.4 The second, innovative, grammar contains not as the specifier of NegP. In both grammars, the single negator contributes the semantics of sentence negation to the derivation. Because one grammar contains an entry for a head and the other for a specifier, there is no reason why the negator from both grammars could not be merged in a single derivation (this would create a situation with some parallels to the phenomenon of code-switching, though Frisch does not invoke that term). This is precisely the situation that, on Frisch’s account, causes ne . . . not sentences to surface. Since both lexical entries have the semantics of negation, it might be feared that merging both would result in double negation (in the logical sense, i.e. ¬¬p ≡ p). Frisch accounts for this by putting forth an Economy of Projection principle, taken from Speas (1994). Under his account, it is the presence in the derivation of NegP (the maximal projection of Neg0 ) which contributes negative semantics. The presence of either ne or not is a sufficient condition for the projection of NegP, but neither is necessary; both together introduce only one NegP. The structures of the different options are presented in Figure 13.3. The evidence that Frisch musters for his analysis is as follows. First, he must distinguish between the use of not as sentence negation and as an (emphatic) sentence adverb. This is not always possible for a given token; however, it is clear that occurrences of not before a tensed verb must be sentence adverbs. The adverb never has a similar distribution to sentence-adverb not, and occurs preverbally 16 per cent of the time in Frisch’s measurement. Thus, Frisch infers a similar rate of preverbal position4 Consistent with the Borer–Chomsky Conjecture (Baker ), we are using the term ‘grammar’ to mean a list of lexical entries for functional categories. Because ne and not are not mutually exclusive, it is possible to activate both in the structure of a single clause. This lack of mutual exclusivity differs from classical cases of grammar competition as discussed by Kroch (b). It is clear from the developments in ME that not replaces ne, but this competition is not necessitated by the structure of UG. Instead it may be driven by pragmatic concerns; see Ahern and Clark (to appear) for discussion.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i



Aaron Ecay and Meredith Tamminga TP T

TP

NegP[+Neg]

T

Neg

VP

XP

ne

…

not

TP

NegP[+Neg] Neg

T

NegP[+Neg] Neg

XP

Neg

VP

ne

...

not

Neg

VP

0

...

Figure .. The analysis of three stages in the evolution of English negation, from Frisch (1997) Note: The notation [+Neg] indicates where the semantics of sentence negation is introduced—in all cases, at the NegP level.

ing for not, and thus that the number of sentence adverb not occurrences in a given time period is the number of preverbal occurrences divided by 0.16 (or equivalently, multiplied by 6.25). Frisch argues that there is no single grammatical change underlying the shift from ne to not. By the CRH (Kroch 1989b) the trajectories of different surface realizations of a single change will be the same. Unless exogenous complications present themselves, the natural spread of a language change through a population will proceed along an S-shaped curve in probability space, which is equivalent to a straight line in logittransformed space. In Frisch’s data, the decline of ne and the rise of not proceed with different slopes in logit space, leading him to conclude that the two phenomena are each linked to different underlying changes. He also argues that co-occurrences of ne and not arise from the combination of two independent lexical insertions. In order to do this, he demonstrates in his data that across various time periods the rate of occurrence of ne . . . not is roughly equal to the product of the rate of occurrence of ne and the rate of occurrence of not (in both cases, the relevant measurement is of all the occurrences of these negators, whether alone or in combination with the other). In other words, insertion of either negator is governed by its own binary random process (effectively, a coin flip). Seeing both negators in the same sentence is the result of both coins happening to turn up heads at once, rather than a face of a special three-sided coin. .. Wallage () The second analysis of the change comes from Wallage (2008) and is based on typologies of Jespersen’s Cycle (JC; Jespersen 1917). Wallage holds that ne, ne . . . not, and not are each stages of an instantiation of JC in English, and that all have a separate grammatical identity. In ne . . . not constructions, the ne that appears does not have negative force, and is rather a manifestation of negative concord (or a similar process driven by formal agreement between negative features and Neg0 ). The various structures needed for this analysis are illustrated in Figure 13.4.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

Persistence as a diagnostic of grammatical status TP T

TP NegP

T



TP NegP

T Neg

Neg

VP

XP

ne[+Neg]

…

not[+Neg] Neg ne

NegP Neg

XP

VP ...

not[+Neg] Neg 0

VP ...

Figure .. The analysis of three stages in the evolution of English negation, from Wallage (2008) Note: The notation [+Neg] indicates where the semantics of sentence negation is introduced. An arrow indicates negative concord (or a similar agreement relationship).

Wallage provides several pieces of evidence arguing against Frisch’s account. For example, Wallage shows that the distribution of ne appearing by itself differs between main and (several different types of) subordinate clauses. On the other hand, the distribution of ne . . . not is identical across clause types. This distinction militates against the hypothesis that there is a unified process generating ne when it appears with and without not. Wallage goes on to show that the loss of ne by itself in main and subordinate clauses occurs with the same trajectory (logistic regression slope), thus providing evidence that ne by itself shares an underlying cause in main and subordinate clauses, which differs from the cause of ne . . . not. The second part of Wallage’s argument, that the ne of ne . . . not constructions is generated by negative concord, is based on the comparison of different types of ‘redundant ne’ in ME. The two types of redundant ne are (A) that licensed by a higher negative (negating a verb of doubt) as in (5) and (B) that licensed by a verb of prohibition or denial as in (6). (5) ne doute the nat that alle thinges ne ben don aryght neg doubt you not that all things neg are done rightfully ‘Do not doubt that all things are done rightfully.’ (Chaucer’s Boethius; Wallage 2008: (25a)) (6) Iesus hire þo for-bed þat heo attryne ne sceolde his hond ne Jesus her though forbade that she bind neg ought his hands nor his fet his feet ‘though Jesus forbade her to bind his hands or his feet’ (Passion 581; Wallage 2008: (24b)) Further examples of the second type can also be found in Early Modern English as in (7), whereas the first type is lost during the ME period. (7) You may deny that you were not the meane of my Lord Hastings late imprisonment (Shakespeare, Richard III)

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i



Aaron Ecay and Meredith Tamminga

Wallage claims that the ne that appears jointly with not is another instantiation of type A redundant ne—that is, licensed by the presence of a Neg projection.5 This claim, insofar as it links the behaviour of sentential negation to other relevant linguistic phenomena, strengthens Wallage’s overall picture. .. Summary There is a fundamental disagreement between Frisch’s account and Wallage’s about how to analyse the change in ME negation in terms of grammatical structures. This disagreement boils down to the question of whether there are two or three atomic objects interacting in speaker–learners’ minds during the change. Each position is supported by some amount of quantitative evidence analysed according to the CRH paradigm. The corpus of ME is strictly limited in size, and Frisch’s and Wallage’s analyses have already incorporated most of it. The amount of further progress that CRHbased approaches can make is thus also quite limited. We propose instead to inject a new methodology into the debate. Priming is a quantitative diagnostic which, like the CRH, is attuned to the grammatical status of constructions. An appropriatelyconstructed examination of priming effects (or the lack thereof) in the available ME corpus will shed new light on the question of which of these two analyses is correct.

. Results .. Background The dataset used in this investigation was derived from the Penn Parsed Corpus of Middle English, version 2 (PPCME2; Kroch and Taylor 2000).6 All tokens of negative declarative clauses were extracted from the corpus, and matched into consecutive pairs. We placed no limit on the number of affirmative clauses intervening between members of a pair. However, the intervening occurrence of a negative clause other than a declarative (such as a question) would prevent two negative declaratives from being joined into a pair.7 For the bulk of the study, we focus on the period from 1250 to 1350, which is the approximate middle of the change when all three surface variants are roughly equiprobable. This maximizes the opportunity to find unequal variants in paired contact with each other, and thus the greatest opportunity to observe prim-

5 Wallage’s presentation of the proposal is not without some degree of hedging: ‘Hence we might try to link change in the distribution of redundant ne to the change which happens in sentential negation contexts. The effect of change in both contexts is the same: restriction of ne to contexts in which another negative [i.e. not in the sentential negation case—AE&MT] is present’ (Wallage : ). This proposal generates a CRH prediction which is not tested. 6 Specifically, Penn’s copy of the working files as of  June  were used, incorporating various fixes since the last release. Layamon’s Brut, a poetic text not included in any release, was excluded. 7 We also excluded the following constructions, which introduce tokens of ne or not which are not part of the envelope of variation: contracted ne as in nis = ne is; ne in concord with none or never; not only and similar constructions; probable cases of not as constituent negation of verbs and adverbs. The interested reader is referred to column  of the queries/coding.c file in the associated GitHub repository (https:// github.com/aecay/digs15-negative-priming) for specific implementational details of these exclusions.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

Persistence as a diagnostic of grammatical status



Table .. Counts in each prime/target position of our primary negation dataset Target

Prime

ne

not

both

Sum

ne not both

124 20 69

13 156 41

58 39 74

195 215 184

Sum

213

210

171

594

Note: The dataset comprises PPCME2 texts from 1250 to 1350 (inclusive), with specific characteristics as described in Section 13.4.1.

ing effects. The corpus contains 594 target–prime pairs during this time period. The counts within each prime/target condition are given in Table 13.1. In examining the data, we looked for evidence of priming effects (or the lack thereof) by examining each prime condition separately. These are the divisions along the x-axis of our plots. Within each condition, we compared the rate of realization of the possible targets. If the rate of realization of each target type is equal to the others within a single prime condition, there is no effect of the prime type on the realization of the target—that is, there is no priming. On the other hand, if there is a difference in target realization proportions, this constitutes a priming effect.8 .. Two-atom model The predictions of Frisch’s two-atom model are shown in Figure 13.7(a). Under the independence assumptions of this model, we expect that uses of ne alone will facilitate following ne (alone or with not), and similarly for not alone. We also predict that tokens of both negators together will have the same effect as ne alone on following use of ne, and similarly for not. This prediction is not borne out completely, as illustrated in Figure 13.5. We expect the middle dark grey bar (measuring the priming effect of ne . . . not on a ne target) to be as tall as the middle light grey one, and the middle light grey bar (ne . . . not on a not target) to be as tall as the right-hand one. The binomial confidence intervals, represented by the error bars, should not be taken as definitive, since there are non-independence effects (such as that generated by repeatedly sampling a small population of speakers) which are not accounted for by the simple underlying model. Nonetheless, they may be taken as a rough guide to the degree to which the point estimates in the bars are uncertain based on the size of the sample subtending them. Note that in the present case, the confidence intervals in the crucial cases discussed above are entirely non-overlapping. 8 A reviewer points out that priming effects might arise because of heterogeneities in the levels of different negation strategies on a text-by-text basis. This is indeed a concern, though it cannot explain our most important result, which is a difference in the fulfilment of the priming predictions of the two different analyses considered.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i



Aaron Ecay and Meredith Tamminga

Percent target outcome

1.00 0.75 Target contains ne contains not

0.50 0.25 0.00 ne alone ne…not not alone Prime

Figure .. The predictions of the two-atom model measured in the corpus data Note: The error bars give 95 confidence intervals according to the binomial distribution.

Percent target outcome

1.00 0.75 Target ne alone not alone ne…not

0.50 0.25 0.00 ne alone

ne…not Prime

not alone

Figure .. The predictions of the three-atom model measured in the corpus data Note: The error bars give 95 confidence intervals according to the binomial distribution.

.. Three-atom model If the three-atom model is correct, then we predict that each kind of negation should facilitate itself, and not any of the other forms. This prediction too is only partially borne out. In Figure 13.6, the right-hand case, where the prime is not, exhibits the expected behaviour, whereby it strongly primes a following not but the other two types of negation have low and (most importantly) equal rates. On the other hand, in the other two cases, there is some degree of cross-priming between ne and ne . . . not, in the sense that, in the ne or ne . . . not prime cases, the bar for the opposite target from this set is higher than the bar for not alone. However, a degree of cross-priming can be explained by appealing to the notion that some of the surface ne . . . not tokens are in fact tokens where ne is the negator, and not

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

Persistence as a diagnostic of grammatical status

ne

NE

ne

ne...not

NE

ne...not

not

NOT

NE–NOT

not

(a) Two-atom model.



NOT

(b) Three-atom model.

Figure .. Priming predictions and actual results of the two models Note: Each diagram shows the predicted priming relationships between surface forms (left) and grammatical objects (right). Solid lines indicate priming relationships which are both predicted and observed. Dashed lines indicate relationships which are not predicted, but are nonetheless observed in the corpus. Finally, solid lines with cut marks indicate relationships which are predicted but not observed.

TP T

TP NegP

Neg ne[+Neg]

T

NegP

VP

Neg

…

ne[+Neg]

VP AdvP

VP

not (/never, etc.)

...

Figure .. Structure with ne as sentence negation in combination with not or another reinforcing adverb Note: The arrow indicates a priming relationship between two tokens.

is a sentence adverb. Such a structure is illustrated in Figure 13.8. In these cases, the negator ne facilitates itself and emphatic not is additionally either added or subtracted. It is possible to test this fix, using the divide-by-0.16 method from Frisch (1997) to calculate the rate of ne . . . not tokens which contain adverbial not. For ne . . . not targets, the test is exact: we discount the number of observed ne . . . not tokens by 16. For ne . . . not primes, we cannot assume that the distribution of adverbial not is consistent across target categories. However, we can set an upper bound on the discount by assuming that all adverbial not primes (i.e. negator ne primes) precede ne by itself. The (unknowable) true amount which must be corrected is less than or equal to the output of such a calculation.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i



Aaron Ecay and Meredith Tamminga 1.00

Percent

0.75 Target ne alone not alone ne…not

0.50 0.25 0.00 ne alone

ne…not Prime

not alone

Figure .. The three-way priming results, after applying Frisch’s divide-by-0.16 adjustment for adverbial not Note: The black marks indicate the original height of the bars directly affected by the patch. (Because the bars in this graph are constrained to sum to 100, the other bars’ heights are affected as well.)

1.00

Percent

0.75 Target ne alone not alone ne…not

0.50 0.25 0.00 ne alone

ne…not Prime

not alone

Figure .. Priming effects on ME negation in the period 1350–1400 Note: The error bars give 95 confidence intervals according to the binomial distribution. Note that invariant texts are excluded from the figure.

The results of applying this analysis can be seen in Figure 13.9, which presents a picture much closer to the predicted situation. (The remaining discrepancy is the relatively small amount by which the height of the leftmost light grey bar exceeds the leftmost medium grey one, when they are predicted to be equal in height.)9 9 The curious reader may wonder whether this same adjustment will improve the picture for the twoatom model depicted in Figure .. It is in fact of equivocal benefit: in the crucial ne . . . not prime condition it raises the not + ne . . . not bar while lowering the ne + ne . . . not bar. Thus, the existence of this patch does not increase our confidence in the two-atom model. For reasons of clarity, we do not present the result here, although the code is available in the associated GitHub repository in the function two.way.patch.graph in the file scripts/digs-proceedings-graphs.R.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

Persistence as a diagnostic of grammatical status



Another piece of evidence in favour of the three-atom model comes from the later period of the change (1350–1400; N = 1617), and is depicted in Figure 13.10. In this period, we see that ne facilitates not more strongly than ne . . . not does. That is to say, the leftmost medium-grey bar is taller than the middle one. The two-atom model assumes that there is a relationship of structural identity between ne . . . not primes and not targets, of the sort that should give rise to priming. There is no such relationship between ne primes and not targets. Thus, the two-atom model predicts that priming should raise the height of the middle medium-grey bar above that of the left-hand one, contrary to fact. The three-atom model on the other hand has no problem with this data; it predicts (accurately) that the bar for targets identical to the prime should be higher than the same-colour bar for the other two possible primes.

. Conclusion In this chapter, we have presented corpus priming data which are inconsistent with the two-atom model and provide tentative support for the three-atom model proposed by Wallage (2008). We note that this evidence by itself does not dispose of the question of which of the two models best describes the historical data. The total picture must also include the CRH-based quantitative evidence discussed by Frisch (1997) and Wallage (2008). Broadening our view beyond the ME negation case study, we observe that the CRH is important because it provides a link between frequency data attested in historical corpora and the mental representations that underlie language and language change. We have here argued that persistence data constitute another, independent source of linkage between these two domains. The investigation of persistence in historical corpora (used in combination with other techniques) can contribute to a body of convergent evidence about syntactic history, and thus to conclusions more robust than those confined to any single methodological framework. Furthermore, the existence in historical corpora of patterns that can be interpreted as priming, in line with studies that uncover priming in purely synchronic textual corpora, lends confidence that the subject of historical corpus investigation has an underlying psychological reality.

Acknowledgements We would like to thank the following people for their assistance: The compilers of the PPCME2, Beatrice Santorini, Anthony Kroch, Phillip Wallage, our fellow graduate students at Penn, the audiences of DiGS15 and PLC38, and an anonymous reviewer of this manuscript. All remaining imperfections are attributable only to the authors. Further, we point out that the data and code used in this analysis, including the LATEX source code for this article, is available on GitHub: https://github.com/aecay/digs15negative-priming.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

 The origins of the Romance analytic passive: Evidence from word order L I E V E N DA N C K A E RT

. Introduction: Passives from Latin to Romance In Classical Latin (informally defined as the period from c. 100 bc to 200 ad), verbal inflectional paradigms consist of both synthetic and analytic forms. The latter are always a combination of a past participle and a BE-auxiliary. Passive verbs are synthetic in the so-called infectum tenses (present (1), imperfect, and future tense), but analytic in the perfectum (perfect (2), pluperfect, and future perfect). Exactly the same holds for (semi-)deponents (i.e. predicates with passive morphology but no passive semantics/argument structure, see e.g. Embick 2000). (1)

si non ips-e amic-us per se ama-tur tot-o if not self-nom friend-nom by refl.acc love-pass.prs.3sg whole-abl pector-e heart-abl ‘if the friend himself is not wholeheartedly loved for his own sake’ (Cicero, De legibus, 1.49)

(2) Iason a Mede-a Vener-is impulsu amat-us Iason.nom by Medea-abl Venus-gen instigation.abl loved-nom.m.sg est. be.prs.3sg ‘Jason was loved by Medea at the instigation of Venus.’ (Hyginus, Fzabulae, 22.4) In (2) there is an apparent mismatch between the tense of the auxiliary est (which in isolation is a present tense), and the tense of the entire analytic expression amatus est, which is a (perfective) past tense. As will be elaborated on shortly, from the earliest Latin texts onwards alternative perfective BE-periphrases of the type amatus fui (with a perfective auxiliary) are available, which are not characterized by this ‘tense mismatch’. This innovative pattern becomes more frequent over time, in apparent competition with the older amatus est perfects. This development can be considered the Micro-change and Macro-change in Diachronic Syntax. First edition. Éric Mathieu and Robert Truswell (eds). This chapter © Lieven Danckaert 2017. First published 2017 by Oxford University Press.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

The origins of the Romance analytic passive: Evidence from word order



Table .. Passives in Latin and Romance (indicatives)

Present (infectum) Past (perfectum)

(Classical) Latin

Romance (Standard Italian)

amor ‘I am loved’ synthetic amatus sum ‘I was loved’ analytic, tense mismatch

sono amato ‘I am loved’ analytic, no tense mismatch fui amato ‘I was loved’ analytic, no tense mismatch

first in a series of innovations that gave rise to the formation of the Romance passive voice paradigm, which is different from the (Classical) Latin one in two important respects. First, none of the Romance passive periphrases display the ‘tense mismatch’ just introduced. Second, all synthetic passives have been replaced by analytic expressions consisting of a past participle and one or more auxiliaries. The basic picture is summarized in Table 14.1, with examples from Standard Modern Italian taken to exemplify the whole of present day Romance (with obvious simplifications). Here I take the Italian simple past (passato remoto, which I assume to be the direct descendant of (Late) Latin amatus fui-type periphrases) as the prototype of a Romance passive past tense, as it is clearly older than (and now often replaced by) other periphrastic structures such as sono stato amato ‘I was/have been loved’ (the socalled passato prossimo), of which there are no traces before 600 ad (i.e. the period that I am concerned with here).1 In any event, there is good evidence that the change from amatus sum to fui amato (which does away with the tense mismatch, cf. the bottom row in Table 14.1) started very early. In contrast, the decline of synthetic passives sets in much later, analytic formations with nonperfective semantics only being attested very sporadically even in the very latest Latin texts (cf. Section 14.2.1.2). It is commonly assumed that once the new analytic perfects (amatus fui, no tense mismatch) were firmly established, the old (amatus est, tense mismatch) perfects underwent a semantic change (from denoting past tense to present tense) and were so to speak ‘recycled’ as the new analytic present passives. In other words, amatus est is taken to be the source of sono amato, as the endpoint of a series of changes which started much earlier. The aim of this chapter is to modify this picture, with key evidence coming from a number of novel observations on word order in Late Latin.

. The development of Latin BE-periphrases: The state of the art .. Some background I will start by giving an overview of a number of developments that affected the Latin voice system, as they are documented in the textual records. I will restrict myself to 1 The full empirical landscape of (Old) Romance BE-periphrases is of course much more complicated (see Ledgeway – and Telve  for relevant discussion). For one thing, many of the structures that were prevalent in Late Latin, such as future perfects (amatus fuerit) and pluperfects (amatus fuerat) do not seem to survive until all that much later than  ad.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i



Lieven Danckaert

those changes which are relevant to the main point at issue, viz. the ones summarized in Table 14.1. I will not be concerned with other Late Latin developments, such as the use of fio ‘become’ as a passive auxiliary, or the emergence of passive-like structures involving the reflexive pronoun se (for recent discussion of these and related issues, see Cennamo 2001, 2005 and J. Adams 2013: 674–724). ... Perfectum tenses: The disappearance of the tense mismatch As hinted at above, some of the developments that differentiate the Classical (and actually also Archaic) Latin voice system and the corresponding Romance paradigms can be traced back to the earliest documented stages of Latin. More specifically, from Plautus (c. 200 bc) onwards, the old amatus est periphrases, characterized by the above-mentioned tense mismatch, appear alongside apparently synonymous structures lacking this mismatch. Consider, for instance, the following minimal pair: the example in (3) features a (deponent) future perfect with a ‘simple future’ auxiliary (ero, i.e. a plain future), whereas in (4) we see the perfective form fuero (a future perfect). (3) donec persecut-us uolp-em er-o until tracked.down-nom.m.sg fox-acc be-fut.1sg ‘until I have tracked down the fox’ (Plautus, Miles gloriosus, 269) (4) Si ego min-am non ult-us fu-er-o probe if I.nom mina-acc not avenged-nom.m.sg be-futprf-1sg well ‘if I will not have taken proper revenge for the mina [a sum of money, LD]’ (Plautus, Poenulus, 1280) In what follows, I will refer to the older structures as ‘E-periphrases’ and to the innovative pattern as ‘F-periphrases’ (after the first letter of (most of) the auxiliaries in the relevant structures). The basic paradigm is given in Table 14.2 (illustrated with passives only, but exactly the same facts hold for deponents). The development from E- to F-periphrases is discussed in among others Leumann (1921); Haverling (2008, 2010); de Melo (2012); and Danckaert (2016). As detailed in Danckaert (2016), throughout the lifespan of the Latin language,

Table .. E- and F-periphrases in passives: the perfectum E-periphrases (old): tense mismatch

F-periphrases (new): no tense mismatch

Infinitive

(perfect)

amatus esse

amatus fuisse

Indicative

perfect pluperfect future perfect

amatus sum amatus eram amatus ero

amatus fui amatus fueram amatus fuero

Subjunctive

perfect pluperfect

amatus sim amatus essem

amatus fuerim amatus fuissem

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

The origins of the Romance analytic passive: Evidence from word order



F-periphrases are the overall minority pattern, but they can be shown to gain ground in the course of time. In certain environments (especially future perfects, and to a lesser extent pluperfects), Late Latin F-periphrases even outnumber their E-counterparts. ... Infectum tenses: From synthetic to analytic It is widely accepted that the changes affecting the infectum tenses take place later than the shift from E to F in the perfectum (for general discussion, see Muller 1924; Herman 2002; de Melo 2012). The data summarized in Table 14.3 show a new set of BE-periphrases (rightmost column), which are apparently formally identical to the old E-periphrases in the perfectum (cf. the E-paradigm in Table 14.2). It needs to be added that most of the forms in the rightmost column of Table 14.3 are actually hypothetical, as they are not attested as such in any Latin text. Only a few convincing examples of clearly verbal (i.e. nonadjectival) BE-periphrases which do not express any notion of perfectivity or temporal anteriority can be found (Väänänen 1981: 130; de Melo 2012). One of the earliest attestations is from Egeria’s Itinerarium (written around 385 ad): (5) [. . . ] tant-us rugitu-s et mugitu-s tot-ius popul-i so.big-nom roaring-nom and bellowing-nom entire-gen people-gen est cum fletu, [ut forsitan porro ad ciuitat-em be.prs.3sg with crying.abl that perhaps further to city-acc gemitu-s popul-i omni-s audit-us si-t]. moaning-nom people-gen entire-gen heard-nom.m.sg be.prs.sbjv-3sg ‘Such was the roaring and bellowing, moaning and crying of the entire people, that their outbursts were perhaps heard in the city.’ (Itinerarium Egeriae 36.3) In this example, the event described in the matrix clause (people moaning and crying) and the hearing event in the embedded clause (cf. auditus sit) are best understood as taking place (quasi-)simultaneously, as we are arguably dealing with a case of direct

Table .. Passives in the infectum Synthetic (old)

Analytic, no tense mismatch (new)

Infinitive

(perfect)

amari

amatus esse

Indicative

perfect pluperfect future perfect

amor amabar amabor

amatus sum amatus eram amatus ero

Subjunctive

perfect pluperfect

amer amarer

amatus sim amatus essem

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i



Lieven Danckaert

auditory perception.2 Therefore, the expression auditus sit in the embedded clause is in all likelihood a genuine present subjunctive. Elsewhere in Egeria’s text, we find a very comparable passage describing essentially the same situation, but here the subjunctive in the embedded clause appears as the single synthetic verb audiantur (i.e. what one would get in Classical Latin), which unambiguously is a present subjunctive: (6) Disputant-e autem episcop-o singul-a et discussing-abl.m.sg prt bishop-abl single-acc.n.pl and narrant-e tant-e uoc-es sunt collauda nt-ium, telling-abl.m.sg so.many-nom voices-nom be.prs.3pl praising-gen.m.pl [ut porro foras ecclesi-a audi-antur that further.away outside.of church-abl hear-pass.prs.sbjv.3pl uoc-es eorum]. voices-nom they.gen ‘But when the bishop was discussing and narrating these items, so many voices arose of people praising him, that they could be heard outside of the church.’ (Itinerarium Egeriae 47.2) In sum, as the interpretation of the example in (5) also does not favour an adjectival (stative or resultative) reading of the expression auditus sit, we can safely conclude that this is a genuine example of the new, ‘Romance-type’ passive. .. The standard account There is a broad consensus that the disappearance of the synthetic passive does not only happen later than the shift from E to F in the perfectum, but that there is also a causal relation between the earlier and the later development. More particularly, the standard assumption is that once the F-periphrases were well established, the entire voice paradigm was analogically levelled, yielding only periphrases of the ‘no tense mismatch’ type (on the role of analogy in this process, see among others Clackson and Horrocks 2007: 280 and de Melo 2012). In what follows, I will assume that this is correct. In addition, however, most authors assume that the formal between the old E-periphrases (amatus est) and the new analytic present tense passives (sono amato) is no coincidence. The relation between the two patterns is typically explained in the following terms. First, it is pointed out that E-periphrases are often ambiguous between an adjectival reading (with a present tense copula and a predicative adjective) and a purely verbal one. The former qualify as present tenses, whereas the latter are perfects. Next, once F-periphrases had become the standard way to express passive perfects, the old E-periphrases only kept their nonpast (adjectival) meaning. The development from an adjectival present tense passive towards a verbal present passive is then

2 In Classical Latin, the combination of a present subjunctive of BE with a past participle (which is formally identical to what we see in ()) is also allowed in resultative ut-clauses, but there it qualifies as a perfect subjunctive and indicates anteriority with respect to the event in the matrix clause.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

The origins of the Romance analytic passive: Evidence from word order



assumed to be a small step. A scenario along these lines is, for instance, outlined in Hewson (1997):3 The passive perfect laud¯atus est ‘he has been praised’ is made up of the perfect participle (passive) and the present (Infectum) auxiliary of the verb esse ‘to be’, so that in analytic terms the whole formation was the equivalent of ‘he is praised’, where . . . we see the confusion between the perfect and the passive. . . . [A] perfect passive with a present auxiliary may be interpreted as an ordinary present passive. (Hewson 1997: 315)

Under this, at first glance simple and elegant scenario, one and the same structure (viz. the combination of a present tense BE-auxiliary and a past participle) changes meaning. As it stands, this account is not a priori implausible, although it would remain to be made precise how exactly the meaning of an adjectival passive can be extended so as to express (present tense) verbal passives too. In the remainder of this chapter, I will develop a slightly different proposal, which does not assume a direct connection between amatus sum and sono amato. Instead, the latter type will be argued to be an entirely new formation, whose origins can be traced back to Late Latin.

. Some overlooked evidence: Word order One important aspect of the development of Latin BE-periphrases which until now has not received any attention in the literature is word order. In this section, I will present a series of new generalizations on this topic, which I will claim shed some interesting light on the developments just outlined. In order to document word order preferences in BE-periphrases in the history of Latin, I have analysed a corpus of Latin prose texts, starting with Cato (at about 160 bc) and ending with Gregory of Tours (end of the sixth century ad).4

3 This standard account assuming a direct connection between Latin perfects and the Romance present tenses comes in a number of versions, sometimes differing in the details. Compare the following quotes:

The central immediate cause of the shift . . . was the need to communicate temporal information. Within the analytic tenses, ambiguities were already present in C[lassical] L[atin], e.g. the dual meaning of the p. ptc. + ESSE: amatus est ‘he is loved, beloved’ with the p. ptc. functioning as an adjective, and ‘he has been loved’, a perfect passive expression. . . . It was the ambiguity of tense within a single expression that motivated the change. (Winters 1984: 450–1) In Italian, Latin amor is replaced by sono amato, from amatus sum, which has acquired present meaning. (de Melo 2012: 84) Other proponents of (variants of) this ‘standard account’ include Ernout and Thomas (: –); Harris (: ); Väänänen (: –); Ledgeway (–: ); and Clackson and Horrocks (: ). 4 The following works were taken up in the corpus (for each author all non-fragmentary prose texts were considered, unless mentioned otherwise): Cato ( bc), Cicero (selection of prose works,  bc), Caesar ( bc), Varro ( bc), Sallust ( bc), Hyginus (De astronomia,  bc), Vitruvius ( ad), Livy ( ad), Celsus ( ad), Seneca (selection of prose works,  ad), Columella ( ad), Petronius ( ad), Frontinus ( ad), Quintilian ( ad), Pliny the Younger ( ad), Tacitus ( ad), Suetonius ( ad), Gaius ( ad), Tertullian (Aduersus Marcionem –,  ad), Cyprian (Epistulae –, except , , , , , , and ,  ad), Palladius ( ad), Itinerarium Egeriae ( ad), Jerome (Epistulae – except , –,  ad), Augustine (Sermones –, –A, –E, –A,  ad), Gesta Conlationis Carthaginiensis ( ad), Vegetius ( ad), Cassius Felix ( ad), Victor of Vita ( ad), Pompeius

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i



Lieven Danckaert

.. A surprising discrepancy As is well-known, Latin auxiliaries and their nonfinite (infinitival or participial) complements can be linearized in either order (see among others Ledgeway 2012: 234–5 and J. Adams 2013: 821–38, and references cited there). Such is, for instance, the case in clauses with a modal auxiliary (like possum ‘be able’) and a dependent infinitive: (7) ut omni-s contio audi-re pos-se-t so.that whole-nom gathering.nom hear-prs.inf be.able-ipfv.sbjv-3sg ‘so that the entire gathering may be able to hear’ (Cicero, Pro Cluentio 134) (8) ut nemo [. . . ] uoc-em acerb-am atque inimic-am so.that nobody.nom word-acc bitter-acc and hostile-acc bon-is pos-se-t audi-re good-dat.m.pl be.able-ipfv.sbjv-3sg hear-prs.inf ‘so that nobody could hear any bitter or hostile word uttered against the good citizens’ (Cicero, Post reditum in senatu 26) In contrast, the present-day Romance languages are characterized by a strictly headinitial TP. Interestingly, corpus data show that the shift towards this new system was already going on in Latin. Clauses with the modal possum constitute a very clear case. As can be seen in Figure 14.1, despite there being a fair amount of synchronic variation throughout the entire period, a steady rise of the head-initial order can be observed, as witnessed by the trajectory of both the straight and the smoothed regression line.5 A similar (albeit milder) development can be observed in clauses with the modal auxiliary debeo ‘have to’ (cf. Figure 14.2), suggesting that the tendency observed in Figure 14.1 is not accidental. The crucial observation is that the alternation between the orders amatus est and est amatus shows a very different picture. As shown in Figure 14.3 (which only contains data on E-periphrases), no rise of the head-initial order can be observed. Instead, without a single exception all Late Latin authors heavily favour the head-final order amatus est.6 Maurus ( ad), Caesarius of Arles (Sermones –,  ad), Anthimus ( ad), Iordanes ( ad), Itinerarium Antonini Placentini ( ad), Gregory of Tours (Historiae,  ad). 5 For the sake of simplicity, in this and the following graphs I am lumping together cases where the auxiliary and the nonfinite verb are string-adjacent with cases where they are not. Nothing crucial hinges on this (see also note ). In addition, each time I only report on data from the authors/texts listed in footnote  which contain at least twenty tokens of a given auxiliary complemented by a nonfinite verb. 6 Note that the observed effect is not to be ascribed to increasing frequencies of leftward movement of the past participle (yielding a structure in which the nonfinite verb and the auxiliary are not (structurally) adjacent), as a reviewer suggests. Such an operation does indeed exist, but it clearly is a minority pattern: in only , out of a total of , head-final E-periphrases, the past participle and the auxiliary are not linearly adjacent (cases where PaPa and Aux are only separated by negation are not taken into account, for reasons discussed in Danckaert , ch.; to appear). In addition, the frequency of leftward participle movement—if the non-contiguous patterns are indeed to be analysed in these terms—declines over time: out of , head-final E-periphrases dating from before  ad,  (.) tokens are discontinuous. For the period after  ad, the corresponding figures are  instances (.) out of a total of ,.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

The origins of the Romance analytic passive: Evidence from word order



100 29 80 7 60

12

19

6 11 10

40 2 20

1

4 5 9

21

15 16 14 17 13

20

26 23 25 24

22

27 28

31 30

32

18

8

3 0 0

200

400

600

Figure .. Relative frequency (in percentages) of the order ‘possum–infinitive’ over time Case labels: 1 = Cato, 2 = Cicero, 3 = Caesar, 4 = Varro, 5 = Sallust, 6 = Hyginus, 7 = Vitruvius, 8 = Livy, 9 = Celsus, 10 = Seneca, 11 = Columella, 12 = Petronius, 13 = Frontinus, 14 = Quintilian, 15 = Pliny, 16 = Tacitus, 17 = Suetonius, 18 = Gaius, 19 = Tertullian, 20 = Cyprian, 21 = Palladius, 22 = Itinerarium Egeriae, 23 = Jerome, 24 = Augustine, 25 = Gesta Conlationis Carthaginiensis, 26 = Vegetius, 27 = Cassius Felix, 28 = Victor of Vita, 29 = Pompeius Maurus, 30 = Caesarius of Arles, 31 = Iordanes, 32 = Gregory of Tours.

As can be deduced from the irregular trajectory of the smoothed regression line, there is probably no justification for saying that there is a real decline of the head-initial order. Rather, what we seem to witness is a loss of word order flexibility, resulting in a ‘fossilization’ of the head-final pattern. In any event, the E-periphrases clearly do not behave like the modals, and they certainly do not display the type of behaviour one expects from a structure that evolves towards the Romance languages. Crucially, these unexpected word order preferences are not a general property of BE-auxiliaries. As is shown in Figure 14.4, the final graph of this section, F-periphrases follow a diachronic path which is very different from that of their E-type counterparts. Although the F-periphrases do not seem to behave exactly like the modals (there appears to be a rise of the head-initial order, but variation is considerable, and the upward trend is rather mild), there is a clear contrast with the E-periphrases. Crucially, here we do not observe any Late Latin preference for the order ‘past participle–BE’.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i



Lieven Danckaert 100

80

19 60

15

3

14 4

40

1

5

21

12

7

20

16 17

10 9

2

20

18

13

8

11

6 0 0

100

200

300

400

500

600

Figure .. Relative frequency (in percentages) of the order ‘debeo–infinitive’ over time Case labels: 1 = Cicero, 2 = Caesar, 3 = Varro, 4 = Vitruvius, 5 = Livy, 6 = Celsus , 7 = Seneca, 8 = Columella, 9 = Quintilian , 10 = Pliny, 11 = Gaius, 12 = Tertullian , 13 = Cyprian, 14 = Palladius, 15 = Jerome, 16 = Augustine, 17 = Gesta Conlationis Carthaginiensis, 18 = Vegetius, 19 = Pompeius Maurus, 20 = Caesarius of Arles, 21 = Gregory of Tours.

.. The origins of the ‘no tense mismatch’ passive: A simpler alternative Recall that the standard account on the origins of the Romance voice paradigm posits that Latin amatus est perfects live on in the Romance languages as present tense passives. Under this scenario, one would expect the relevant structure to take part in a change that otherwise affects the entire language system, namely the shifts towards a head-initial TP. Given that this is clearly not the case, the facts reviewed in the previous section constitute a major problem for the standard account, as the observed word order tendencies in E-periphrases remain entirely unexplained. As an alternative, I would like to suggest a very simple solution which is only slightly different from what is standardly assumed, but which avoids the problem just sketched. Recall that previous studies strongly emphasized the role of analogical levelling in bringing about a passive paradigm which is entirely of the F-type (‘no tense mismatch’, see Clackson and Horrocks 2007; de Melo 2012). Assuming this to be correct, I propose that when they are being replaced by F-periphrases, Late Latin perfective E-periphrases do not change meaning but slowly disappear from

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

The origins of the Romance analytic passive: Evidence from word order



100

80

5

6

60 3 40

2 1

13 10

20

11

7

0

8

4 0

9

12 14

16 15

100

18

20 22

19

17 200

24 25 23

31

26 28 29 30 27

21 300

400

500

600

Figure .. Relative frequency (in percentages) of the order ‘BE–past participle’ in passive and deponent E-periphrases over time Case labels: 1 = Cicero, 2 = Caesar, 3 = Varro, 4 = Sallust, 5 = Hyginus, 6 = Vitruvius, 7 = Livy, 8 = Celsus, 9 = Seneca, 10 = Columella, 11 = Petronius, 12 = Frontinus, 13 = Quintilian, 14 = Pliny, 15 = Tacitus, 16 = Suetonius, 17 = Gaius, 18 = Tertullian, 19 = Cyprian, 20 = Palladius, 21 = Itinerarium Egeriae, 22 = Jerome, 23 = Augustine, 24 = Gesta Conlationis Carthaginiensis, 25 = Vegetius, 26 = Cassius Felix, 27 = Victor of Vita, 28 = Pompeius Maurus, 29 = Caesarius of Arles, 30 = Iordanes, 31 = Gregory of Tours.

the language. The innovative, ‘no tense mismatch’ Romance passives are then a new formation, created by analogy with the perfective F-periphrases, which as we have seen exhibit ‘normal’ Late Latin word order behaviour. This solution has at least two advantages over the standard account. First, it avoids the word order problem: the newly created present tense passives can be expected to be equally likely to be headinitial as their perfective counterparts (i.e. the older F-periphrases). Second, there is no need to explain the shift (extension) from an adjectival present tense passive (like amatus sum ‘I am popular’) towards a verbal present tense passive of the type sono amato ‘I am (being) loved’. In all accounts known to me, the change from ‘past– verbal/present–adjectival’ to ‘present–verbal/present–adjectival’ is simply asserted to have taken place, but no details of the putative transition are ever made explicit, let alone formalized. Under the present approach, no such explanation is required, as there is no semantic change to begin with.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i



Lieven Danckaert 100

20

80

24 18 2

60

12

7

21

16 19

40

3

1

5 8

13

10

22 14

20 11

6 4

0 0

23

25

17 15

9

100

200

300

400

500

600

Figure .. Relative frequency (in percentages) of the order ‘BE–past participle’ in passive and deponent F-periphrases over time Case labels: 1 = Cicero, 2 = Hyginus, 3 = Vitruvius, 4 = Livy, 5 = Celsus, 6 = Seneca, 7 = Columella, 8 = Quintilian, 9 = Pliny, 10 = Suetonius, 11 = Gaius, 12 = Tertullian, 13 = Cyprian, 14 = Palladius, 15 = Itinerarium Egeriae, 16 = Jerome, 17 = Augustine, 18 = Gesta Conlationis Carthaginiensis, 19 = Vegetius, 20 = Cassius Felix, 21 = Victor of Vita, 22 = Pompeius Maurus, 23 = Caesarius of Arles, 24 = Iordanes, 25 = Gregory of Tours.

.. A note on deponents and Romance unaccusatives Before moving on, I will first say a couple of words about present-day Romance expressions like Italian sono venuto ‘I have come’. Such (perfective) past tenses of unaccusatives feature a BE-auxiliary which in isolation would be a present tense, and thus clearly display the tense mismatch discussed above. It has been proposed that these structures are derived from Latin deponent E-periphrases (see for instance Ledgeway 1997–9: 115–16). Given the obvious formal similarities between expressions like Latin natus sum ‘I was born’ and mortuus sum ‘I died (I am dead)’ on the one hand, and Italian sono nato and sono morto on the other, this hypothesis certainly has some intuitive appeal. However, there are also a number of problems with this idea. First of all, an obvious difference between Latin deponents and Romance unaccusatives is that unlike the latter, Latin deponents constitute a semantically very heterogeneous class of verbs, consisting of unaccusatives like morior ‘die’ and proficiscor ‘depart, leave’, unergatives like iocor ‘jest, joke’ and epulor ‘dine, feast’, and clear-cut transitives like adipiscor

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

The origins of the Romance analytic passive: Evidence from word order



100

80

60

40

20

0 0

100

200 Deponent

300

400

500

600

Passive

Figure .. Relative frequency (in percentages) of deponent and passive F-periphrases (as compared to E-periphrases) over time

‘obtain’ and consequor ‘follow, pursue’.7 In other words, a direct transfer from Latin deponent unaccusatives (like natus sum) to Romance unaccusatives (sono nato) is of course straightforward enough, but both (i) the extension of the BE-perfect to other unaccusatives (like sono venuto, for which there is no Latin deponent counterpart) and (ii) the exclusion of all other predicate types (i.e. all Latin unergative and transitive deponents) would still have to be accounted for. Second, a line of reasoning very similar to the one developed above concerning the diachrony of Latin passives can be applied to deponents, as in the period under investigation (c. 100 bc–600 ad), the diachronic trajectory of these two types of BE-periphrases is virtually identical. For instance, when we look at the proportion of F-periphrases in passives and deponents (as shown in Figure 14.5), we see that the change from E to F affects passives and deponents in equal measure.8 7 This is of course not to say that no generalizations can be made about the historical roots of many deponents, which often derive from mediopassives (Flobert ; Gianollo ). However, this clearly does not offer a satisfactory synchronic description of the relevant facts: whether or not a given verb is deponent (at a particular stage of the language) is a purely lexical idiosyncracy. 8 Note that the fact that the frequency of F-periphrases increases over time does not obviously appear in Figure .. As detailed in Danckaert (), the rise of the F-pattern is strongest in future perfects and (to

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i



Lieven Danckaert

From this dataset we can conclude that Late Latin analytic deponents take part in the same shift from E to F, in more or less exactly the same way as their passive counterparts (see once again Danckaert 2016 for more detailed discussion of exactly this point). This can be considered evidence against an account that assumes direct continuity from (certain) Latin unaccusative deponents to Romance unaccusative BEperfects. More specifically, what the (Late) Latin data suggest is that analytic deponents like natus sum do not remain stable over time, but rather (slowly) change to the type natus fui ( fui natus). In addition, if it is correct to think that the change from E to F in the perfectum, on the one hand, and the change from synthetic to analytic in the infectum, on the other, are part of one larger integrated development (viz. the reorganization of the entire Latin passive voice paradigm, cf. Section 14.2), we predict that if the diachronic trajectory of passives and deponent analytic verb forms is not differentiated, Late Latin deponent E-periphrases should also display a preference for the order ‘PaPa–Aux’. As shown in Figure 14.6, this is indeed the case.9 Assuming that the Late Latin preference for the ‘PaPa–Aux’ order is correlated with the fact E-periphrases were dying out (and thus did not take part in an otherwise general change towards a head-initial T-node), these facts also suggest that Latin natus sum might actually not be that closely related to Italian sono nato, despite formal similarities. To conclude, given the above considerations, and extending the logic developed in Section 14.3.2, one can reasonably hypothesize that Romance periphrases of the sono nato/sono venuto type are also new formations (whose origins are arguably related to the rise of analytic HAVE-perfects, cf. Ledgeway 1997–9). One only needs to add the proviso that the proposed new perfects probably came into being (much) later than the new present tense passives (which are attested as early as the fourth century ad, cf. Section 14.2.1.2 above).

. Towards an analysis of the word order facts In the remainder of this chapter I will offer some additional discussion of the unexpected (and hitherto unnoticed) Late Latin preference for head-final E-periphrases. For reasons of space, here I will only provide a summary of the account. A detailed analysis can be found in Danckaert (2015). In a nutshell, I take it that the word order facts shown in Figure 14.3 are to be understood in terms of a PF constraint on the placement of the auxiliary, which I take to have acquired clitic-like properties. More specifically, I assume a lexical split between

a lesser extent) pluperfects, but these two categories are largely outnumbered by plain perfects, where the relevant change is much slower. Also, note that the apparently anomalously high frequency of F-periphrases in one data point in Late Latin (viz. Cassius Felix at  ad) is related to the fact that most BE-periphrases in this text happen to be future perfects. 9 Only E-periphrases were included in this dataset, and only data from authors/texts with at least ten deponent tokens.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

The origins of the Romance analytic passive: Evidence from word order



100

80

60

40

20

0 0

100

200 Deponent

300

400

500

600

Passive

Figure .. Relative frequency (in percentages) of the order ‘Aux–PaPa’ in E-periphrases over time, passives and deponents compared

a strong (phonologically independent) and a weak variant of BE.10 The strong variant, which is the same as in earlier stages of the language, appears in contexts which are preserved in Romance, viz. copular clauses (including adjectival passives of various kinds) and verbal F-periphrases.11 In contrast, in E-periphrases we get the weak version, which displays the unexpected word order behaviour described in Section 14.3.12 10 Note that the phonological weakening of BE which I am about to analyse is different from (i) BEcontraction of the type described in Pezzini (), which only affects forms of esse with a vocalic onset and which is restricted to Archaic and early Classical Latin, as well as from (ii) certain (putative) ‘second position’ phenomena which have been analysed in terms of BE-cliticization (J. Adams ). 11 I remain agnostic as to whether copular and auxiliary BE are themselves separate lexical entries. 12 It is possible that weak BE also occurs in other environments which do not survive in Romance, like deontic constructions involving BE and a gerundive, as in (i).

(i) prius instrument-a praepara-nd-a sunt before.comp.adv tools-nom prepare-gdv-nom.n.pl be.prs.3pl ‘first the tools have to be prepared.’ (Columella, De agricultura ..) I leave it to future research to verify whether this structure also favours the head-final order, which would be compatible with the idea that it involves weak BE.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i



Lieven Danckaert

.. Weak BE, and prosodic restrictions on word order The key ingredient of the proposal is that weak BE is not a genuine clitic (which BE can be in English and in many Slavic languages), but rather a category which is somewhere in between a full lexical item and a true clitic. This proposal is clearly reminiscent of accounts where structurally deficient variants of full lexical items are said to come in (at least) two versions, called Xmax and X0 clitics in Halpern and Fontana (1994), and weak and clitic items in Cardinaletti and Starke (1999). In both of these analyses, however, the difference between weak elements and true clitics is argued to involve a difference in phrase structure: the former are taken to be (reduced) phrases, whereas the latter are considered syntactic heads. In the present context, I will assume that both manifestations of Latin BE are heads (as they are obviously ‘verbal’), but that the lexical entry corresponding to them is slightly different.13 Concretely, I take it that whereas both are identical in terms of semantic feature composition, they differ as to their phonological realization. Assuming a standard version of the prosodic hierarchy (see Selkirk 1980, 1984; Nespor and Vogel 2007; and much related literature) I will take it that prosodic words— which are themselves composed of one or more feet (F)—are grouped together to form phonological phrases (ϕ) (I am abstracting away from clitic groups). Next, I will assume that strong BE always constitutes an independent prosodic word (ω) of its own, but that weak BE can never project a prosodic word of its own. Instead, I propose that the latter is a ‘stray’ metrical foot which can only survive if it occurs in an extrametrical position at the right edge of a phonological phrase. Consider the three structures in (9), where ‘F’ represents weak BE.

φ

(9) a. *

F

ω

φ

b. * ω

ω

F

φ

c. ω

ω

ω

Assuming that extrametrical material can only occur at the right edge of phonological domains (Hyde 2011: 1045–7), it follows that the only context in which weak BE can be incorporated in a well-formed phonological phrase is (9c), where the extrametrical status of the final foot is indicated by means of angle brackets. This phonological ‘escape hatch’ for weak BE is not available in the other two structures diagrammed in (9), where the weak element occurs at the beginning (9a) or in the middle (9b) of a phonological phrase. As elaborated on in Danckaert (2015), this analysis effectively derives the abundance of Late Latin ‘past participle–BE’ patterns, as only auxiliaries in head-final T-projections have a chance of ending up at the end of a phonological phrase.

13 Alternatively, one could assume that all lexical items are actually phrasal, and that there is a difference between two or more types of Latin BE which corresponds to a difference in size (shape). An account along these lines could be worked out in the framework of nanosyntax (Starke ).

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

The origins of the Romance analytic passive: Evidence from word order



.. No X0 -incorporation (‘true cliticization’): Placement of negation in VP–Aux clauses An obvious alternative to the account just sketched is to say that the Late Latin preference for head-final BE-periphrases involves encliticization through syntactic head movement. Under this approach, the auxiliary would be an affix in T, and the lexical verb would incorporate to it. The relevant structure would look as in (10), which does indeed feature the desired linear order ‘PaPa–Aux’ (assuming the incorporating head to left-adjoin to its host).

TP

(10)

T T0

VP

V0

T0

PaPa

BE

V V0

Assuming head movement to take place in a strictly local fashion, this account predicts the lexical verb and the auxiliary to show ‘clustering’ effects, in the sense that they would either be strictly adjacent, or separated from one another only by heads which are base-generated below T, and which incorporate into V before the latter head moves to T. However, there are good reasons to assume that Late Latin participles and BE-auxiliaries are not actually structurally contiguous, as a head movement account would predict. Evidence for this comes from placement of negation in clauses with a head-final BE-periphrasis. The basic facts concerning placement of negation in Latin are as follows. First of all, there is an exceptionless generalization to the effect that the canonical marker of sentential negation non ‘not’ always precedes the hierarchically highest verb in the clause. This generalization can be considered to follow from (i) the fact that non is first merged higher than T, and (ii) independent locality principles which make sure that the verb in T0 cannot move across negation (say the Head Movement Constraint, assuming non to be a head). Second, although there is a clear tendency for non and the highest verb to be linearly adjacent (as they are in (11)), at all stages of the language non can also appear in more leftward positions (see Kühner and Stegmann 1966, vol. 2.1: 818–19 for additional discussion). For instance, in (12) we find the string ‘Neg–PaPa–Aux’. (11)

pro qu-o frument-o pecuni-a omni-s solut-a for which-abl corn-abl money-nom all-nom paid-nom non est not be.prs.3sg ‘In exchange for this corn not the entire sum of money was paid.’ (Cicero, Diuinatio in Q. Caecilium 32)

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i



Lieven Danckaert

(12) illud si incid-it, hoc non sequut-um this.nom.n.sg if befall-prs.3sg that.nom.n.sg not followed-nom.n.sg est be.prs.3sg ‘if the former happens without the latter following’ (Celsus, De medicina 2.17.6) Here I will not be concerned with the proper analysis of the syntax of sentential negation in Latin, which is dealt with in Danckaert (2017): I will simply assume that the data in (11) and (12) reflect the availability of multiple NegPs in the higher functional field (Zanuttini 1997; Cinque 1999: 120–6). Instead, I will focus on one particular prediction one makes when assuming a structure like (10) for Late Latin BE-periphrases: if the latter are indeed derived by means of V-to-T movement, we expect that in VP– Aux clauses, the order ‘Neg–PaPa–Aux’ should become more frequent over time, at the expense of the alternative order ‘PaPa–Neg–Aux’, which arguably does not involve head movement of V across Neg (which would violate the Head Movement Constraint), but rather (remnant) VP movement past negation (of the type discussed in Danckaert 2012, 2014, 2017). Corpus data suggest that this prediction is not borne out. As shown in Table 14.4, negated BE-periphrases of the type amatus non est remain productive until well after c. 150 ad (cf. the shaded cells), which is when the order ‘PaPa–Aux’ seems to have fossilized.

Table .. Frequency of the orders ‘PaPa–Neg–Aux’ and ‘Neg– PaPa–Aux’ in BE-periphrases of the E-type (only authors/texts with at least four negated E-periphrases included) Author

Cicero Varro Livy Celsus Seneca Columella Quintilian Suetonius Gaius Tertullian Cyprian Augustine Pompeius Maurus Gregory of Tours

Date

60 45 5 30 50 55 95 120 170 210 255 400 500 590

bc bc ad ad ad ad ad ad ad ad ad ad ad ad

PaPa–Neg–Aux (amatus non est)

Neg–PaPa–Aux (non amatus est)

62 4 5 18 6 7 16 3 30 8 6 11 1 2

16 2 13 3 6 2 3 2 0 1 3 11 3 2

(79.49) (66.67) (27.78) (85.71) (50) (77.78) (84.21) (60) (100) (88.89) (66.67) (50) (25) (50)

(20.51) (33.33) (72.22) (14.29) (50) (22.22) (15.79) (40) (0) (11.11) (33.33) (50) (75) (50)

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

The origins of the Romance analytic passive: Evidence from word order



Despite the small number of tokens in some of the later authors,14 it seems safe to conclude that the fixation of the order ‘PaPa–Aux’ in Late Latin E-periphrases does not go hand in hand with a requirement for the participle and the auxiliary to be strictly linearly adjacent, witness the continuing availability of the order ‘PaPa–Neg–Aux’.

. An interesting parallel: The genesis of the Romance synthetic future I will conclude this chapter by discussing an interesting parallel with the development of BE-periphrases in the history of Latin, namely the oft-discussed rise of the Romance future and conditional. As is well-known, these structures are the result of a grammaticalization process whose input is a (Late) Latin periphrastic structure involving an infinitive and the verb habeo ‘have’ (see among others Thielmann 1885; Rohlfs 1922; Valesio 1968, 1970; Coleman 1971, 1976; Clancy 1975; Pinkster 1985; J. Adams 1991, 2013: 625–73; I. Roberts 1992; Bourova 2005; Bourova and Tasmowski 2007). The first known attestation of the Romance synthetic future is given in (13), which according to Stimm (1977) can be dated at c. 550–600 ad (see also Ledgeway 2012: 136–7). (13) qui ill-a pussedir-auit uiu-a[t] usqui who.nom.m.sg that-acc possess-fut.3sg live-sbjv.3sg until ann-us mili in d[e]-o years-acc.m.pl thousand in god-abl.m.sg ‘May he who shall possess this live for a thousand years in God.’ (inscription on a Merovingian buckle found in Ladoix-Serrigny (France, Côte-d’Or), cf. Stimm 1977) In this example, the future tense marker -auit is still bisyllabic, and thus quite transparently related to Latin habet. In (14), on the other hand, from around 650, the old synthetic future dabo ‘you will give’ appears alongside its synonymous successor daras, in which the future tense marker is now monosyllabic (cf. -as, cf. Latin habes). (14) et ill-e responde-ba-t: ‘non da-b-o’. Iustinian-us and he-nom answer-ipfv-3sg not give-fut-1sg Iustinianus-nom dic-eba-t: ‘dar-as’. say-ipfv-3sg give-fut.2sg ‘And he answered: “I will not give”. // Iustinianus said: “you will give”.’ (Fredegarius scholasticus, Scripta Rerum Merovingorum 2.2.62 (p.85 l.32)) The form daras corresponds to Latin dare habes (lit. ‘give you.have’), but the HAVEverb clearly has changed both meaning and syntactic/phonological shape. Interestingly, there is good evidence to assume that early on, the future marker was not yet a clear-cut verbal affix. Thus I. Roberts (1992):

14 As it happens, word order in Late Latin negated clauses with BE-periphrases is much more complicated and cannot be done justice to here: see Danckaert (, : –) for full discussion.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i



Lieven Danckaert

It is well-known that Old Spanish (OSp) and conservative European Portuguese (EP) have future and conditional markers which are more clitic-like than affix-like. . . . [T]hey are halfway between being auxiliaries and being affixes. (I. Roberts 1992: 230)

Evidence for the non-affixal status of the grammaticalized future marker comes from the availability of mesoclisis, a phenomenon whereby a pronominal clitic intervenes between the lexical verb and the future marker. This pattern was available in many Old Romance varieties (see for instance Old Spanish (15)), and survives until today in varieties of European Portuguese (16). (15) Senora, — dixo el — dezir=lo hedes a-l rey? Old Spanish lady said he tell=it you.will to-the king ‘Lady, he said, will you tell it to the king?’ (Libro del caballero Zifar 124, cf. Lema and Rivero 1989: 250, (21a)) (16) Ele vê=la=a. he see.inf=cl.acc.f.3sg=fut.3sg ‘He will see her.’

European Portuguese (cf. Duarte and Matos 2000: 117, (5))

In whatever way they are to be analysed, these interpolation phenomena are of course reminiscent of the (Late Latin) ‘PaPa–Neg–Aux’ pattern: in both cases, the ‘light’ auxiliary follows its lexical complement, without there being any strong adjacency requirement. It is therefore tempting to conjecture that the grammaticalization process that gave rise to the synthetic future went through a ‘weak HABEO’ stage, in which the future marker was subject to the same constraint as the one governing placement of weak BE, as proposed in the previous section. However, the parallel between the passive/deponent ‘PaPa–BE’ sequences and the new synthetic future is clearly only partial. First of all, only in the case of phonological weakening of HAVE is there any process of grammaticalization (meaning change) involved. Second, if the analysis proposed in Section 14.3 is on the right track, E-periphrases died out, whereas the HAVE-futures are still today very much alive in many Romance varieties. Related to this, the comparison between the two developments also suggests that there is no principled (causal) relation between phonological weakening of Late Latin BE and the ultimate disappearance of E-periphrases. Rather, the latter development is in all likelihood first and foremost a matter of competition with (perfective) F-periphrases (whose origins remain to be better understood).

. Conclusion The goal of this chapter was to reconsider the development of Romance present tense passives of the type sono amato ‘I am loved’. I have argued that these are not derived from Classical Latin analytic perfects, contrary to what is standardly assumed. Instead, I have proposed that the Romance pattern is a new formation, which goes back to at least the fourth century ad. The main advantage of this new approach is that it is compatible with the observation that Late Latin amatus est perfects show a very strong preference for the otherwise declining head-final order, no trace of which can be found in Romance BE-periphrases.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

The origins of the Romance analytic passive: Evidence from word order



Acknowledgements The research reported on in this paper was funded by postdoctoral grants of the ‘Bijzonder Onderzoeksfonds’ (BOF) of Ghent University (grant no. BOF11/PDO/042) and of the ‘Fonds voor Wetenschappelijk Onderzoek’ (FWO) (grant no. FWO13/ PDO/024).

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

 Reconciling syntactic and post-syntactic complementizer agreement S A R A H G . C O U RT N E Y

. Introduction The occurrence of morphological agreement features hosted on complementizers— known as Complementizer Agreement (CA)—is a cross-linguistically rare but robust pattern found in several language families. The pattern is well-documented in dialects of several Germanic languages (i.e. dialects of Dutch and German, particularly Flemish and Bavarian) and in several Bantu and Mande languages. This chapter focuses on the Germanic CA pattern in which C agrees with the embedded subject.1 In this pattern φ-features of the embedded clause subject appear on the clause-linking complementizer, as seen in (1) below. (1)

a. Kpeinzen dan-k (ik) morgen goan. I-think that-I I tomorrow go ‘I think that I’ll go tomorrow.’ b. Kpeinzen da-j (gie) morgen goat. I-think that-you you tomorrow go ‘I think that you’ll go tomorrow.’ c. Kvinden dan die boeken te diere zyn. I-find that-pl the books too expensive are ‘I find those books too expensive.’ (West Flemish, Haegeman 1992)

Due to both the typological rarity of the phenomenon, and the theoretical interest generated by extra agreement relationships within CP and the role such 1 Other types of CA exist. For instance, several Bantu languages allow CA with the matrix subject. See Diercks () and Carstens () for a treatment of the Bantu patterns, Carstens () for a comparison of the Germanic and Bantu patterns, and Courtney (in preparation) for a comparison of the diachronic development of the two.

Micro-change and Macro-change in Diachronic Syntax. First edition. Éric Mathieu and Robert Truswell (eds). This chapter © Sarah G. Courtney 2017. First published 2017 by Oxford University Press.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

Reconciling syntactic and post-syntactic complementizer agreement



agreement may play in clarifying or complicating the relationship between CP and TP, CA has attracted attention from multiple branches of linguistics. Two primary accounts exist for CA. In the pure syntax accounts (e.g. Carstens 2003; van Koppen 2006; Haegeman and van Koppen 2012), CA is the output of an Agree relation between a uφ Probe (either generated at C0 or shared with T0 ; see Section 15.3) and the embedded subject as a φ-bearing goal. In the post-syntactic accounts (e.g. Ackema and Neeleman 2004; Fuß 2007; Miyagawa 2009; Zwart 2012), CA is not the output of a syntactic Agree relation, but instead of a feature-copying process licensed under adjacency or membership within the same phonological phrase that allows uφ features at C0 to be valued outside the narrow syntax. Both of these accounts have attempted to present a single, unified explanation of CA cross-linguistically—or at least throughout Germanic. This chapter reconciles these two seemingly contradictory analyses, arguing that they are not competing models of the same cross-linguistic structure, but instead analyses of related but distinct CA structures that exist in microvariation within the CA dialect areas. I also argue that the variation within CA grammars should be placed within an overall diachronic analysis of CA as a process of grammaticalization, with each CA-generating grammar existing as a stable point on the cline of grammaticalization between analogy, cliticization, reanalysis, and a new Agree relation between C0 and the embedded subject (pure syntactic CA). This chapter is structured as follows. Sections 15.2 and 15.3 review the competing theories of CA in the literature, presenting the data that support each, and showing the contradictions generated with data that is not well-matched to the given theory. In Section 15.4 I show that CA is not a unitary phenomenon but a situation of microvariation, and examine the type of inter- and intra-speaker variation that can give rise to the CA data available in the literature. Section 15.5 positions the CA grammars identified in Section 15.4 within the cline of grammaticalization.

. Complementizer agreement as a PF interface phenomenon The cases of CA that are the most widespread are also the most straightforward to analyse under either a syntactic or post-syntactic framework. These are standard cases of declarative embedding where the complementizer appears immediately adjacent to the embedded subject. I will address these universal CA patterns before turning to specific types and their variation. Although different languages show different patterns of CA, the most complete paradigm and the most famous examples come from West Flemish. Example (1) is repeated from the introduction. (1)

a. Kpeinzen dan-k (ik) morgen goan. I-think that-I I tomorrow go ‘I think that I’ll go tomorrow.’ b. Kpeinzen da-j (gie) morgen goat. I-think that-you you tomorrow go ‘I think that you’ll go tomorrow.’

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i



Sarah G. Courtney c. Kvinden dan die boeken te diere zyn. I-find that-pl the books too expensive are ‘I find those books too expensive.’ (West Flemish, Haegeman 1992)

A full paradigm of agreement is shown with pronominal subjects,2 while nominal subjects show agreement in number, as in (2). (2) a. Kpeinzen da Valère morgen goat. I-think that Valère tomorrow go ‘I think that Valère will go tomorrow.’ b. Kpeinzen da-n Valère en Pol morgen goan. I-think that-pl Valère and Pol tomorrow go ‘I think that Valère and Pol will go tomorrow.’ (West Flemish, Haegeman 1992) Patterns similar to those found in West Flemish—clitics aside—are also found in other Germanic languages. Data from Zwart (1993) shows distinctions between singular and plural first person in South Hollandic as in (3), between a second person singular form and a default in Groningen and Frisian as in (4) and (5), and between second person singular and plural and a default form in Munich Bavarian (6) and Luxemburgish (7).3

2 West Flemish has a full set of agreeing clitics (i) and singular/plural agreement in sentences with nominal subjects.

(i) a. da-n *(=k) ik werk-en that-1sg 1sg I work-1sg b. da-t *(=j) gie werk-t that-2sg 2sg you work-2sg c. da-t *(=j) ij werk-t that-3sg 3sg.m he work-3sg d. da-t (=ze) zie werk-t that-3sg 3sg.f she work-3sg e. da-t (=t) tet werk-t that-3sg 3sg.n it work-3sg f. da-n (=me) wunder werk-en that-1pl 1pl we work-1pl g. da-t *(=j) gunder werk-t that-2pl 2pl you.pl work-2pl h. da-n (=ze) zunder werk-en that-3pl 3pl they work-3pl

(Shlonsky 1994)

The clitic pronoun is optional in all but the sg, sg/pl, and sg.m examples. See Bayer () for an exploration of the relationship between CA and pro-drop. 3 Several accounts exist for the prevalence of second person CA compared with the rarity of non-second person morphology on C0 . Weiß () and Fuß () both attribute the pattern to phonological changes applying only to second person clitic pronouns that made the reanalysis as an agreement marker possible for those persons but not others. A syntactic account is given by Bayer (). This restriction falls outside the scope of this paper and the restriction of CA to certain persons in some dialects will not receive a thorough accounting here. I direct interested readers to the works cited in this footnote.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

Reconciling syntactic and post-syntactic complementizer agreement (3) a. dat ik kom that I come ‘that I come’ b. date we komme that-pl we come-pl ‘that we come’



(South Hollandic, Zwart 1993)

(4) a. of ik kom whether I come ‘whether I come’ b. of-s toe koms whether-2sg you come-2sg ‘whether you come’ (5) a. datst (do) jûn komst that-2sg you tonight come-2sg ‘that you come tonight’ b. dat (er) jûn komt that he tonight come-3sg ‘that he comes tonight’

(Groningen, Zwart 1993)

(Frisian, Zwart 1993)

(6) a. damid ich komm so.that I come ‘so that I come’ b. damidsd kommsd so.that-2sg come-2sg ‘so that you come’ c. damidds kommds so.that-2pl come-2pl ‘so that you come’

(Munich Bavarian, Zwart 1993)

(7) a. ob ech wëll whether I want ‘whether I want’ b. ob s du wëlls whether 2sg you want-2sg ‘whether you want’ c. datt e mir wëllen that pl we want-pl ‘that we want’

(Luxemburgish, Zwart 1993)

In some CA dialects, CA is blocked when a nonsubject intervenes between the subject and the complementizer.4 4 Van Koppen () refers to the data here—and other data from Peter Vermeulen, who gave this example—as being from De Panne Dutch, a dialect of West Flemish. Ackema and Neeleman characterize it as an example of ‘Flemish’. I will use van Koppen’s more specific description.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i



Sarah G. Courtney

(8) a. da / dan zunder op den warmste dag van ’t jaar tegen that that-3pl they on the hottest day of the year against under wil gewerkt en their will worked have ‘that they have worked against their will on the hottest day of the year’ b. da / *dan op den warmste dag van ’t jaar zunder tegen that that-3pl on the hottest day of the year they against under wil gewerkt en their will worked have ‘that on the hottest day of the year they have worked against their will’ (De Panne Dutch, Peter Vermeulen in Ackema and Neeleman 2004) Since such phrases do not serve as barriers for verbal agreement, examples like those in (8) bolster the argument that CA is triggered by adjacency and not by a traditional Agree relationship. Post-syntactic accounts of CA are based on phonological copying of features licensed through adjacency (Ackema and Neeleman 2004; Fuß 2005; Miyagawa 2009) which is more easily interrupted by phonological additions, or scrambling of the lower clause. PF accounts rely heavily on data such as (8), and are thus specifically tailored for dialects where intervention effects occur. Ackema and Neeleman (2004) include West Flemish CA in the set of constructions that they believe are best explained by feature checking at PF. According to this account, uφ-features at C find and take on valued φ-features within the same prosodic domain. Miyagawa (2009) makes a similar proposal, allowing CA to occur at PF as a type of concord based on string adjacency. In Miyagawa (2009) this suggestion is motivated by a desire to defend a central assumption that C and T share a single φ-probe between them, generated at C and valued upon its inheritance by T. For Ackema and Neeleman (2004) a PF analysis is designed to account for data from dialects of Flemish where CA is optional but obligatorily absent when adverbial material intervenes between C and the subject, as in the dialect of (8). For Ackema and Neeleman’s model to predict a dialect where CA was obligatory— that is, one in which a complementizer taking a default form and not agreeing with the subject (as in (9b)) is ungrammatical—such a dialect would have to have a restriction on material occurring between C and the subject. In fact, they report that such speakers who have an absolute restriction on such intervention do exist (Ackema and Neeleman 2004: 240). However, there also exist speakers for whom CA is obligatory and no restriction or intervention occurs. For example, the West Flemish speakers described by Haegeman and van Koppen (2012) do not allow the non-agreeing complementizer even in cases of intervention. (9) a. . . . da-n / ?*dat toen juste men twee broers kwamen. that-pl that then just my two brothers came ‘. . . that my two brothers came just then.’ b. . . . da-n / ?*dat juste ip dienen moment men twee broers that-pl that just at that moment my two brothers kwamen. came ‘. . . that my two brothers came just at that moment.’ (West Flemish, Haegeman and van Koppen 2012) i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

Reconciling syntactic and post-syntactic complementizer agreement



For these speakers, CA is obligatory and not subject to intervention effects, which undermines an adjacency-based analysis. More will be said about the grammar of such speakers in Section 15.3. The post-syntactic CA account from Fuß (2005) is based on a dissociated morpheme at C that has uφ-features. Fuß treats the valuation of Agr on the dissociated morpheme as ‘parasitic on the presence of an Agr-morpheme that has been valued in the syntax’ (2005: 109). This morpheme would take its features directly from the Agr-morpheme of the verbal agreement on which it was parasitic. Fuß specifically addresses the issue of adjacency effects—drawing on data from Bavarian and East Netherlandic dialects that exhibit a blocking effect on CA when an adverb intervenes between C and the subject—and argues that his account provides a simpler model of such cases than do the narrow syntax accounts. One further issue encountered by a PF account is that it is more difficult to rule out object agreement at C. Objects, even when scrambled into a position between C and the subject, do not trigger agreement. (10) a.

Kpeinzen da-n / *dat zelfs men broers zuknen boek niet lezen. I.think that-pl that even my brothers such.a book not read

b. ??Kpeinzen da-n zuknen boek zelfs men broers niet lezen. I.think that-pl such.a book even my brothers not read c. *Kpeinzen dat zuknen boek zelfs men broers niet lezen. I.think that such.a book even my brothers not read ‘I think that even my brothers do not read such a book.’ (West Flemish, Haegeman and van Koppen 2012) This remains true even in dialects in which CA is obligatory. Examples (9) and (10) are from the same dialect of West Flemish, in which CA is obligatory and not subject to intervention effects when the intervener is non-φ-bearing as in (9). However, a scrambled object presents a problem for such a dialect. The closest φ-goal is the object, but agreement between C and the object is completely ungrammatical as in (10c)— worse in fact than (10b), where the agreement skips the closest goal and agrees with the subject. When such post-syntactic accounts are extended to account for data where CA occurs outside of direct adjacency, or across prosodic domains, the account results in an overpowered phonological interface. In order for a fully post-syntactic version of CA to account for the data in (9) and to rule out the agreement between the complementizer and a scrambled object in (10), the post-syntactic agreement process needs to contain some level of syntactic sensitivity to subjecthood and φ-features, and an A vs. A-bar distinction. Otherwise, the φ-features of an object should serve as an appropriate goal for CA. It is unclear how such notions could be included in the phonology.5

5 One account that I have not addressed here is that of Zwart (), who proposes that CA is not part of the narrow syntax because it is part of a process of synchronic analogy, based on the process of analogy that gave rise to the modern pattern of CA. De Vogelaer and van der Auwera () document the changes that allowed the agreement morphemes once associated with verbs to appear on complementizers. Zwart () imports the diachronic account wholesale into the synchronic domain, arguing essentially that the

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i



Sarah G. Courtney

. Complementizer agreement in the narrow syntax Analyses that place CA within the narrow syntax have an advantage for accounting for cases in which CA is obligatory and scrambling is not ruled out. While PF accounts would need to incorporate a mechanism by which PF can access information from the syntax module (e.g. φ-features, an A vs. A-bar distinction), a syntactic account has a built-in mechanism for ruling out object CA (as in (10c) above) and for accounting for the non-intervention of PP (as in (9b) above). If CA is treated as the output of a syntactic Agree relation, then we must assume it to be the product of a uφ-probe finding the closest available φ-bearing goal. Such a probe will naturally ignore a PP or adverb that is not φ-bearing. Agreement with an object will also be ruled out. Object scrambling is standardly accepted to be due to A-bar movement; the φ-features of the object should be inert, having been ‘used up’ in the previous phase, and should not trigger CA.6 Analyses that treat CA as a part of the narrow syntax fall into two categories: those treating CA as the result of a C-to-T inheritance or similar linkage allowing the φ-features properly linked to T (or C) to be shared with the other (probe-less) head, and those that propose that each head has its own φ-probe. I will show that neither of these analyses are adequate for all of the data. Cases of simple CA ((1)–(3))—where the agreement features of the tensed verb and complementizer match—have been used to bolster a case for a single φ-probe copied or split between the two heads and movement of features proceeding head-tohead leading to the same features at T0 and C0 (see e.g. Zwart 1997; Watanabe 2000). In this account the probe controlling CA exists in the ‘canonical’ uφ position at T0 and its features are realized on C0 thanks to head-to-head movement, as illustrated in Figure 15.1. CP C0

head movement

TP T0 uφ

VP φ Agree

Figure .. Feature-sharing between T and C via head movement

analogical change has given rise to no real changes in the narrow syntax. While the analogy account is diachronically well-motivated, once the analogical change has occurred, speakers are no longer beholden to the source construction. It is unclear what sort of analogical process can exist in the synchronic grammar to mimic the type of analogy that occurs in language change. This issue will be returned to briefly in Section .. 6 Carstens () leverages the reusability of φ-features to account for cross-linguistic differences in object agreement at C. Indo-European languages do not have reusable φ-features, while Bantu languages may reuse the Gender feature, leading to object CA in several Bantu languages.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

Reconciling syntactic and post-syntactic complementizer agreement



Following Chomsky’s (2008) account of feature inheritance, the relationship was inverted to allow for the φ-probe—still unitary—to be generated at C0 . In this case the probe is passed on to T0 and then probes from this position, resulting in the canonical agreement between tensed verb and subject, and φ-features on C0 are realized through head-to-head movement, as above. Since T0 and C0 under these accounts host a single probe, the agreement features expressed on the complementizer and the verb would be expected to match. However, this is not always the case.7 A syntactic account in which C0 contains an independent φ-probe deals more elegantly with cases of C0 and T0 mismatch. Such accounts are needed to explain some of the data described in the following paragraphs. When the embedded subject is coordinated, some dialects allow complementizer agreement to be with the first conjunct only rather than with the entire coordinated subject. Others show CA with the features of the entire coordinated subject. Van Koppen (2007) found two different patterns represented in Dutch dialects. (11)

kenne treffe. Ich dink de-s doow en ich os I think that-2sg [you.sg and I]1pl each.other-1pl can-pl meet ‘I think that you and I can meet each other.’ (Tegelen Dutch, van Koppen 2007)

(12) Oa-n Bart en Liesje nie ipletn. . . If-3pl Bart and Lisa not watch.out. . . ‘If Bart and Lisa don’t watch out. . . ’

(Tielt Dutch, van Koppen 2007)

This variation does not appear to be parametric, and the two types may co-occur in the same speech community. Bavarian German allows both patterns of agreement. (13) a. daß-sd du und d’Maria an Hauptpreis gwunna hab-ds that-2sg [you.sg and the Maria]2pl the first.prize won have-2pl ‘that you and Maria have won the first prize’ b. daß-ds du und d’Maria an Hauptpreis gwunna hab-ds that-2pl [you.sg and the Maria]2pl the first.prize won have-2pl ‘that you and Maria have won the first prize’ (Bavarian, van Koppen 2007) Some Flemish speakers also allow, under certain circumstances, CA with the possessor of the embedded subject rather than the subject itself. (14) . . . omda-n die venten toen juste gebeld een because-pl those guys then just phoned have.pl ‘. . . because those guys called just then’

7 In fact, even without a φ-feature mismatch, some C-to-T inheritance accounts predict problems for CA. Obata and Epstein (), for instance, argue that the single probe, once passed down to T0 , should have no ability to realize φ-features at C0 except in languages where φ-features are reusable, which is not the case in Indo-European languages. They propose that due to this restriction all Germanic CA must be post-syntactic. However, there are empirical reasons to believe this is not the case. See Haegeman and van Koppen () for a more thorough argument against this approach.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i



Sarah G. Courtney

(15) . . . omda-n die venten toen juste underen computer kapot was because-pl those guys then just their computer broken was ‘. . . because those guys’ computer broke just then’ (West Flemish, Haegeman and van Koppen 2012) This pattern appears much more constrained than CA generally, requiring the presence of a focused adjunct to the left of the subject, with the possessor alone occurring between the complementizer and the adjunct. The external possessor construction itself—allowing the possessor to be separated from the possessed NP by an adjunct— is a rare construction, subject to constraints on animacy and an unclear distribution across dialects. Haegeman and Danckaert (2011) found the construction to be accepted by some—not all—Flemish speakers, and did not find the construction to be limited to a specific region (although Flemish speakers rejecting the construction reported that it sounded like a West Flemish construction to them). The construction is also limited to cases where the possessor is animate and alive at the time the utterance refers to (Haegeman and Danckaert 2011). Buelens and D’Hulster (2014) found the construction to be more common in West Flemish, but neither unheard of outside that region nor universally accepted within it. The mismatches seen in (13) and (15) above are unexpected under a framework where φ-features are shared between T and C—or under a theory of post-syntactic feature copying in which φ-features are copied from T0 onto C0 . Van Koppen (2007) proposes that the uφ-features of C0 and T0 find different goals in such cases because of the movement of the conjoined phrase into subject position and past T0 . The copy that T’s uφ-probe finds is a ‘reduced copy’, lacking internal structure, while C’s uφ-probe encounters a fully realized coordination structure and takes the closest goal within that structure—the first conjunct. A similar account of external possessor CA is given by Haegeman and van Koppen (2012). They propose that the presence of a focalized temporal adjunct occupying Spec,FocP forces the generation of an additional, higher projection to host the possessor. In its new position at the top of the middle field, it serves as the closest goal for a uφ-probe in C0 . This analysis relies on independent probes—even though they find copies (or elements within copies) of the same goal.8 The expected output of feature sharing would be either identity of the agreement expressed on the complementizer and the verb due to the same features being moved head-to-head or the same element being given multiple spell-out, or the omission of agreement at one or the other. If, for instance, the uφ-probe was passed on unvalued from C to T, and only valued at T, no φ-expression should be expected at C. However, CA does occur, even in contexts where it does not bear the same features as verbal agreement. 8 Bahloul and Harbert () describe similar agreement facts in both Arabic and Welsh in which a verb only agrees with the first conjunct of a postverbal subject. They propose that this is a process of agreement through government where the first conjunct is the nearest c-commanded element. Spec–head agreement with preverbal conjoined subjects in the same languages agrees with the combined φ-features of the entire phrase. This theory accounts for the agreement shown in (). Verbal agreement in T selects the φ-features of its conjoined goal (pl) while the φ-features of C0 are valued by the absolute closest goal and nearest ccommanded element: the first conjunct (sg). Such a structure does not support the idea that the processes are related or that either C or T has its φ-features valued by the other.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

Reconciling syntactic and post-syntactic complementizer agreement



The various accounts reviewed in the first two sections explain a wide variety of CA patterns, but no single model provides a simple explanation for all the data at once. In some cases a post-syntactic account is better suited, and in other cases a syntactic one fits best. However, rather than a theoretical deficiency, this situation may be indicative of dialect variation.

. Complementizer agreement and microvariation Complementizer agreement is most common in regional, ‘nonstandard’ linguistic varieties, and its expression varies greatly within and across speech communities. This has made a single theoretical account elusive. This variation appears even across closely related dialects, and speakers in neighbouring towns or regions may exhibit slightly different grammars with respect to CA (e.g. variation can be found between speakers of highly localized varieties of both Flemish and Bavarian). This section will argue that rather than choosing between the previous syntactic and post-syntactic accounts given for CA, multiple accounts should be seen as nonmutually exclusive. In fact, different grammars underlie the different CA patterns, and attempts to unify the theoretical accounts have led to an assumption of synchronic and diachronic uniformity—a universal CA structure that accounts for the phenomenon cross-linguistically—that has eclipsed the variation present in the phenomenon and the existence of multiple stable stages of a diachronic cline between post-syntactic and narrow syntactic CA grammars. The fact that CA is most often found in regional spoken varieties can mean both that subtle local differences are preserved—a factor that can be of great interest to linguists—and can simultaneously make the construction difficult to study. Some of the variation in CA between speakers may in fact be due to code-switching between dialects with a CA pattern and those without—or between two different CA dialects. This may account for some of the constructions rated ‘marginal’ by speakers, or where speakers appear able to accept conflicting patterns. As pointed out by an anonymous reviewer, the marginality of the constructions with a non-agreeing complementizer in (9) and the acceptability of them in (8) could be the product of code-switching. Speakers who would not otherwise allow material to occur between C and the subject may accept the construction as an output of standard Dutch grammar, where it is grammatical. Thus, this intra-speaker variation may be the product of a single speaker having multiple grammars, including one that does not contain CA at all in this context. Some patterns related to CA are highly restricted even within the relatively small dialect area in which they are found. Buelens and D’Hulster (2014) look specifically at external possessor agreement, the pattern seen in (15). They find that this pattern is more accepted by West and East Flemish speakers than by Brabants and Antwerp Flemish speakers. However, they find that the construction is somewhat marginal even in dialects where it does occur.9 9 In fact, some degree of intra-speaker variability may also be at play here. The speakers who accept the external possessor pattern appear to find it somewhat acceptable rather than wholly acceptable, so it remains unknown what patterns such speakers may be able to produce and use in conversation.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i



Sarah G. Courtney

Not all dialectal or idiolectal variation can lead us to a definitive choice between a syntactic and a post-syntactic account for CA in a given grammar. The aforementioned variations in external possessor agreement, for instance, do not provide a sufficient clue. Buelens and D’Hulster (2014) provide a syntactic account in which the possessor is raised into a higher position on the clausal spine than the possessee, in order to satisfy case assignment.10 However, since the external possessor in its raised position is the closest φ-bearing goal to the complementizer, an account based on linear adjacency or membership within a limited prosodic domain cannot be ruled out, and the data are not diagnostic. Likewise, the core cases in (1)–(3) are easily accounted for by all of the analyses above. Only a handful of rare sentence types distinguish between the grammars. That is to say, a post-syntactic, PF-feature-copying account and a uφ-probe-based account handle the majority of CA cases equivalently well, but they do not handle the outlier cases—coordinated subjects, intervention effects, and external possessor agreement— with equal success. These cases have been leveraged as diagnostic, to support either syntactic or PF accounts of CA as a whole. However, they can instead be viewed as indicative of grammatical variation showing the range of variability of CA and the different analyses needed to account for these variations. For example, the dialects of Flemish reported in Ackema and Neeleman (2004)— including data from inside the West Flemish dialect area—would be candidates for a post-syntactic analysis, while the West Flemish dialect reported in Haegeman and van Koppen (2012) would not. The ability, and in fact necessity, of agreement in Haegeman and van Koppen’s data regardless of phonological adjacency between C and the subject suggests a strictly syntactic grammar. The data in (8), however, suggest that phonological adjacency is key in triggering agreement for De Panne speakers, and that non-agreeing complementizers are much more acceptable to them than to other West Flemish speakers. In the examples in (9) and (10), non-agreeing complementizers are ungrammatical. Such data contrast with those in (8) and demonstrate a clear dialect distinction. While the data in (8) could be generated by PF feature checking— with a default non-agreeing complementizer used when no subject φ-features are available in a string-adjacent position—the data in (9) and (10) require a syntactic explanation. In these dialects, a derivation in which C’s φ-features are not valued crashes. First conjunct agreement and external possessor agreement provide less straightforward contrasts. For instance, the dialect contrast drawn by Haegeman and van Koppen (2012), shown in (11) and (12), repeated below, provides a good example of the type of microvariation possible in CA constructions. (11)

kenne treffe. Ich dink de-s doow en ich os I think that-2sg [you.sg and I]1pl each.other-1pl can-pl meet ‘I think that you and I can meet each other.’ (Tegelen Dutch, van Koppen 2007)

10

See Buelens and D’Hulster () for a thorough account.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

Reconciling syntactic and post-syntactic complementizer agreement (12) Oa-n Bart en Liesje nie ipletn . . . If-3pl Bart and Lisa not watch.out . . . ‘If Bart and Lisa don’t watch out . . . ’



(Tielt Dutch, van Koppen 2007)

Although geographically close, the two groups of speakers show a distinction that may be indicative of significant underlying grammatical differences. It is tempting to apply a syntactic account to the data in (12) and a PF account to that in (11). However, a PF account cannot be ruled out in any case, and van Koppen (2007) accounts neatly for first conjunct agreement within the narrow syntax by proposing that once the coordinated subject has raised into the lower clause’s periphery, the first conjunct and its parent node (the full conjoined subject) are equally close goals for C’s φ-probe. However, whether the difference between the two dialects runs as deep as a PF versus narrow syntax distinction, or whether it is merely a distinction between two related syntactic processes, a grammatical divergence exists between the two neighbouring dialects. External possessor agreement provides a similar example of microvariation. The construction itself is extremely limited, even within the West Flemish dialect area, with not all speakers allowing CA to occur with a nonsubject at all. These cases are important for their very inconsistency. As seen in (13) for Bavarian, it is not necessary for each dialect to have a single grammar associated with CA. In addition to the variation found within Flemish dialects, Bavarian also shows some small but crucial differences in its CA pattern. Gruber (2008) shows that, unlike the Bavarian dialects detailed in Bayer (1984) and Fuß (2005), the Gmunden dialect spoken in upper Austria shows a CA pattern that is not sensitive to adjacency. Contrast (16) from Gmunden, in which an entire adverbial phrase intervenes between C and the subject, with (17), in which an adverb occurring between C and the subject in another Bavarian dialect blocks CA.11 (16) Waun-st beim ärgstn Regn in Gmunden du oiwei ausse geh If-2sg at worst rain in Gmunden you always out go mua-st, daun kaun I da a net höfn. must-2sg, then can I you also not help ‘If you always have to go to Gmunden during the worst rain, then I cannot help you either.’ (Gmunden, Gruber 2008) (17) *obwoi-st woartscheints du ins Kino ganga bist although-2sg probably you to-the movies gone are ‘although you probably went to the movies’ (Bavarian, Günther Grewendorf, from Fuß 2005)

11 Van Koppen (to appear) notes a similar dialect split in both Flemish and Bavarian, and links it with the presence of double agreement (DA)—a pattern in which verbal agreement is expressed with a different ending in subject–verb inversion contexts. I have not addressed double agreement at all in this chapter, and direct interested readers to van Koppen’s work. However, the possible connection between DA and CA— especially as they relate to the diachronic relationships between CA grammars and the potential for the extension of CA patterns—is something that certainly deserves further attention.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i



Sarah G. Courtney

The distinction between Gmunden Upper Austrian and the Bavarian of Fuß (2005) is the same as that discussed above between the West Flemish dialects studied by Haegeman and van Koppen (2012) and the De Panne Dutch reported in Ackema and Neeleman (2004). The Gmunden dialect, along with several West Flemish dialects, shows a pattern best captured with a syntactic approach, as a uφ probe will be able to look past non-φ-bearing adjunct phrases to the next appropriate goal. The Bavarian and De Panne data, however, suggest a post-syntactic CA grammar in which the separation of the complementizer and the subject blocks CA.

. Complementizer agreement and grammaticalization The desire for theoretical unity in CA accounts has impeded study of the diachronic relationships between the post-syntactic and syntactic analyses presented in Sections 15.2 and 15.3, respectively. As argued in Section 15.4, rather than competing explanations for a single universal CA pattern, these analyses describe distinct grammars that yield distinct versions of CA. In fact, the grammars, far from being incompatible, can be viewed as stages on a cline of grammaticalization, with the post-syntactic CA grammars representing independently stable intermediate stages and the narrow syntax CA grammars representing a fully grammaticalized CA. The creation of a new independent φ-probe at a previously φ-less head is a natural outcome of such a process of grammaticalization—a natural ‘simplest assumption’ for a speaker confronted with agreement morphology. The rise of CA is generally accepted to have been a process of analogical change in which the pattern of agreement found on verbs was extended to apply to complementizers. De Vogelaer and van der Auwera (2010) provide an explanation of the development of CA in Dutch dialects, from the prior existing pattern of agreement between a verb and a subject enclitic in sentences with postverbal subjects. (18) Gaa-n=ze morgen naar Gent? go-3pl=they tomorrow to Ghent ‘Are they going to Ghent tomorrow?’ (19) Naar Brussels ga=me. to Brussels go=we. ‘We go to Brussels.’

(Dutch, de Vogelaer and van der Auwera 2010)

The similarity between the sequence verb–weak pronoun and Comp–weak pronoun as in (20) and (21) below resulted in speakers assuming a similar structure for CP and TP and extending the pattern of agreement from verbs to complementizers. (20) Morgen zal hij het boek lezen. tomorrow will he that book read ‘Tomorrow he will read that book.’ (21) Ik geloof dat hij het boek morgen zal lezen. I believe that he that book tomorrow will read ‘I believe that he will read that book tomorrow.’ (Dutch, de Vogelaer and van der Auwera 2010)

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

Reconciling syntactic and post-syntactic complementizer agreement



While de Vogelaer and van der Auwera (2010) draw a contrast between an analogically derived C and one that contains its own probe, I contend that these two states of affairs are in fact two different stages in the development of syntactic CA. De Vogelaer and van der Auwera (2010), much like Zwart (2012), argue for a synchronic distinction between an account based on analogy—extending the verbal pattern to the complementizer—and one in which C0 has its own φ-features. They predict that the analogical pattern will be less regular and more dependent on the similarity in context between the source environment and the CA environment. Although there are dialects where CA is limited to those contexts most closely resembling the original analogical extension (e.g. dialects in which CA is limited to contexts with strict C–subject adjacency, or those where CA is only with pronominal subjects), there are also those where the pattern appears more regular, and is obligatory even in nonadjacency contexts and with DP subjects. Therefore, it seems that de Vogelaer and van der Auwera’s analogical account needs to be extended one more step to encompass the pure syntax grammars of CA.12 The full reanalysis should proceed as follows. Prior to the analogical change, the complementizer itself bears no φ-features. The complementizer is merely adjacent to the embedded clause subject, which bears inherent φ-features. When this subject is reanalysed as an enclitic on the complementizer—a rather standard progression from weak pronoun to clitic predicted by Fuß (2005), among others—the path toward syntactic change begins. This first step is a phonological reduction, and no deep syntactic change needs to have occurred to generate the sentence in (22). (22) Ze zegg-en da-n=ze naar Brussel gaa-n. they say-3pl that-3pl=they to Brussels go-3pl ‘They say that they are going to Brussels.’ The analysis in Figure 15.2 is a possible underlying structure for (22), with the clitic occupying the embedded subject position. While the presence of agreement morphology on the complementizer shows that some form of Agree is taking place, this could be a PF phenomenon based on copying the features of the weak pronoun. However, the ambiguous surface structure necessary to generate a syntactic reanalysis now exists. Following Roberts’ (2010a) analysis of clitics as bundles of φ-features attracted to the cliticization site by a requirement to value a φ-probe, the structure in Figure 15.3 can be proposed as an alternative underlying structure for (22). Under Roberts’ framework, the same type of uφ probe that would find the agreement features of the subject would also serve as a trigger for movement of the subject. The subject is in this case a clitic pronoun which would undergo head movement to C0 . So, as the pronoun in Spec,TP is reanalysed as a clitic phonologically, its syntactic role 12 There are additional, theoretical reasons to be wary of an account of CA as analogy that does not allow for the development of new uφ probes and new syntactic relationships. Morphosyntactic changes are often—if not always—fed by ambiguous surface strings that the learner may reinterpret as having a different underlying structure. Putting a limit on the endpoint of such reanalysis would rule out a crucial and wellknown avenue of syntactic change. There also remains the problem of defining analogy as a synchronic process (see footnote ).

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i



Sarah G. Courtney TP ze zeggen

CP

da-n

TP ze naar Brussel gaa-n

Figure .. A possible syntactic structure for the sentence in (22) TP ze zeggen

CP

da-n=ze

TP

naar Brussel gaa-n

Figure .. An alternative syntactic structure for the sentence in (22)

also undergoes change. A full pronoun moved to Spec,TP by phrasal movement is reanalysed as a cliticized pronoun. Due to its linear position following C and preceding T, it becomes an enclitic on C, bolstering the learner’s assumption of uφ at C0 as the explanation for the Agreement features on the complementizer, since both agreement and head movement can be triggered by such a probe. With the structure in Figure 15.3, C0 hosts an unvalued φ-probe, which must search for a goal, finds the embedded subject, and raises its φ-features to C resulting in both Agreement and cliticization. The existence of this uφ-probe at C frees the phenomenon of CA from the strict adjacency of a complementizer with a weak pronoun. A uφ-probe at C may locate φ-features on any available embedded subject, not just a pronominal one. This last stage of the change has taken place where CA is the most widespread and obligatory, such as in the grammars of speakers who require CA with full DP subjects and across intervening phonological material. This brings us to the present situation in dialects that have been proposed to have CA in the narrow syntax. However, rather than just steps along the way to a syntactic, teleological endpoint, other patterns of CA form stable grammars in and of themselves. Fuß (2005) also proposes a specific pattern for the rise of CA at PF. Fuß’s pro-

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

Reconciling syntactic and post-syntactic complementizer agreement



Stage 1 CP

TP Spec,TP VP pronoun

Figure .. Movement of a pronoun to Spec,TP Stage 2 CP cl

TP VP pronoun

Figure .. Cliticization of a pronoun to C Stage 3 CP

TP Agree

VP pronoun

Figure .. Reanalysis of a pronominal clitic as an output of agreement

posal for the rise of CA looks similar to that for the rise of verbal agreement up until the point of the syntactic reanalysis itself. First, movement of a pronoun to Spec,TP (Figure 15.4) is reanalysed as movement of a clitic, and cliticization to C (Figure 15.5). Next, the clitic pronoun is analysed as the output of an agreement relation (Figure 15.6). However, here the pattern for CP and TP are supposed to behave differently. A clitic at TP will be reanalysed as agreement realized through a probe– goal relation, that is, a probe will be assumed to exist at T0 . A clitic at CP, however, will be reanalysed as a dissociated agreement morpheme ‘licensed under structural adjacency with Agr-on-T’ (Fuß 2005: 170).

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i



Sarah G. Courtney

Fuß gives an analysis that terminates at the modern post-syntactic structure but which is largely parallel to the early stages of the syntactic reanalysis account— founded in de Vogelaer and van der Auwera (2010) and extended in the present chapter—for the pure syntax pattern. The parallel stages are completed in all CA dialects. It is the final stage of reanalysis that distinguishes those dialects that have a pure syntax-based CA from those that have a post-syntactic (more restricted) CA pattern. The remainder of this section will look at the role that ambiguous surface strings and rare sentence types may play in the development of different CA grammars. The grammars laid out in Sections 15.2 and 15.3 are hard to distinguish and the majority of cases of CA found in all relevant dialects could be generated by either a post-syntactic or narrow syntactic account. Although cases of nonidentity between the φ-features of C and T could serve as cues for a learner to assign an independent φ-probe in C0 , and examples where non-agreeing complementizers replace agreeing ones in nonadjacent contexts (e.g. the data in example (8)) could bias a learner towards a PF approach, both of these cases are uncommon and some learners may never encounter them during acquisition. So how do the two dialect types remain distinct? The data available is rife with idiolectal variation, patterns judged by speakers to be marginal—not fully grammatical nor fully unacceptable—and apparently unconstrained optionality, all of which suggest that perhaps the multiple CA-generating grammars that exist are not remaining distinct at all, but are in fact coexisting as microvariation or even multiple grammars within individual speakers. In the absence of specific and rare data, a learner could either assume that CA happens through a normal Agree relation, or through a PF process. The Agree relation, however, is already a part of the learner’s grammar, and is used to generate T-agreement. The reuse of an already-learned grammatical operation seems more likely than the assumption of a new one, absent empirical reasons to rule it out. In fact, the post-syntactic, intervention-effect grammar could easily give way to a puresyntax, intervention-effect-free grammar through straightforward syntactic change. Only very specific data—sentences that contain an intervener and do not exhibit CA— can force speakers towards a post-syntactic analysis. A pure syntax approach, in which C is assumed to contain a uφ probe, is simpler from a learner’s perspective. It is possible that the variation that we see in CA is in fact change in progress for some speakers or speech communities. The PF, adjacency-dependent version of CA could undergo reanalysis yielding a more regular, pure syntax CA. Similarly, highly restricted patterns of CA (such as those limited to pronominals or to certain persons), could also be generalized into a fuller pattern that looks more like a typical realization of Agree. However, highly limited Agreement patterns could also cease to be realized if insufficient data is available to learners to include them in the grammar: CA patterns realized on only one or two persons could as easily disappear entirely as expand to a regularized pattern. Language change will always be in the direction of regularity, and an irregularly expressed pattern will tend to expand or die out (i.e. the lack of agreement will be the pattern that expands). The situation in many (if not most or all) dialects described in the existing literature is one of stable variation. A sufficient number of learners are successfully acquiring

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

Reconciling syntactic and post-syntactic complementizer agreement



the variety of CA phenomena that several patterns are attested. Some learners may be acquiring more than one CA grammar, yielding optionality in the surface structure and competition between grammars within a speaker’s I-language. The development of a pure syntactic grammar for CA is by no means the teleological endpoint of a necessary series of changes. However, it is certainly not an unexpected type of development from the data available to learners of many CA dialects, and it is a change that falls nicely into line with recent work on grammaticalization and directionality in syntactic change. The reanalysis of a φ-feature bundle—first an inherently valued one (pronoun), then a clitic, then an unvalued φ-feature that takes its value from an adjacent valued φ-feature, and finally a full φ-probe—is an expected pattern, following both Roberts and Roussou’s (2003) and van Gelderen’s (2009b) expectations of syntactic change. According to Roberts and Roussou, reanalysis happens ‘up’ the cline of functionality, and morphemes are reanalysed into higher positions in the tree. This is certainly the case of all CA reanalyses, including those that do not occur in the narrow syntax. The PF φ-features proposed by Fuß (2005); Ackema and Neeleman (2004); and Miyagawa (2009) exist at C, and have been reanalysed into that position from origins at T (Fuß 2005). Furthermore, van Gelderen’s (2009b) account of reanalysis is that the standard direction of change is one in which interpretable features will be minimized in favour of uninterpretable features: goals tend to be reanalysed as probes, as seen in the final stage of the syntactic reanalysis, represented in Figures 15.5 and 15.6 above. One of the primary objections to generating a new uφ probe at C0 is that the existence of such a probe is problematic for theories that seek to limit the occurrence of φ checking to TP, although it is accepted by many syntacticians working on CA (see Section 15.3). Among those who claim that C cannot be a φ position in its own right, Zwart (2012) in particular argues that learners ought not to be able to acquire a grammar that includes uφ at C0 , but this makes predictions that are not borne out in the data—that is, the existence of grammars that cannot be captured by post-syntactic accounts—and artificially constrains the type of language change that can occur. If we are to rule out the innovation of φ-features at C0 , we must assume either that learners are never faced with a dataset that would cause them to attribute a φ-probe to that position, or that learners somehow ‘know’ that C0 is an inappropriate place for such structure. As the first assumption is clearly incorrect in the face of the CA data, the second assumption remains as the last defence against φ-probes being generated by language change for those who wish to avoid it for theoretical reasons. What would such a restriction mean for language change? Diachronically, a grammar that contains knowledge of what kind of probes can occur where should be resistant to the type of changes that van Gelderen (2009b) finds to be widespread and basic to the nature of language change. This would preclude grammaticalization and van Gelderen’s linguistic cycle almost entirely.

. Conclusion I have argued that the patterns of CA found in Germanic dialects are not the output of one cross-linguistic underlying structure, but are in fact several related patterns

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i



Sarah G. Courtney

generated by different grammars in microvariation. The variants of CA range from post-syntactic patterns which are highly restricted in their occurrence, to obligatory agreements generated by purely syntactic processes. I argue that these patterns exist at several points on the cline of grammaticalization, and represent different stages of diachronic development. The pure syntax pattern found in dialects where CA is the most obligatory and least restricted is the pattern predicted to arise over time as the terminus of grammaticalization. However, several different CA grammars remain stable at earlier stages of the grammaticalization process. The fact that the pattern varies even within a small dialect region and may be subject to significant idiolectal variation makes the pattern a rich field to further study microvariation and the role of rare sentence types on syntactic acquisition.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

 On the grammaticalization of temporal–aspectual heads: The case of German versprechen ‘promise’ ŁUKASZ JĘDRZEJOWSKI

. Introduction: Versprechen in present-day German In present-day German the predicate versprechen ‘promise’ can be used in two different ways. When it occurs as a speech act verb, the speaker commits himself to perform p. Usually, it takes three arguments, merges as a lexical verb, and exhibits subject control (henceforth: v1): (1)

v1 [Ich]i verspreche [Ihnen]j , [inf proi/*j heute freundlich zu sein]. I promise.1sg you.dat today nice to be.inf ‘I promise you to be nice today.’ (DeReKo, Braunschweiger Zeitung, 14/9/2005)

On the other hand, versprechen can also be employed as a functional verb expressing a temporal–aspectual attitude of the speaker towards what is embedded1 (henceforth: v2). In this case, it is a subject-to-subject raising verb and its theta grid is associated with two arguments. A corpus example is given in (2).

1 Drohen ‘threaten’ behaves in a similar way, allowing both readings. However, in this chapter I will not elaborate on the development of drohen. Cross-linguistically, both predicates have usually been analysed together as an inseparable pair—see Rooryck () and Verhagen (, ) for Dutch dreigen and beloven, Traugott (, ) for English threaten and promise, Askedal (); Colomo (); Reis (, ); and Wurmbrand () for German drohen and versprechen, Cornillie (, ) for Spanish amenazar and prometer, and so forth. However, there are many reasons why threaten and promise should be examined separately, i.e. each on its own. In Polish, for instance, only grozić ‘threaten’ can occur as a functional verb. Obiecać ‘promise’, in contrast, never grammaticalized into a temporal–aspectual head, indicating that it can only be used as a lexical verb corresponding to v’s meaning in German. In addition to that, as demonstrated in Jędrzejowski (), although drohen and versprechen pattern similarly, they emerged out of two different structures exemplifying two different development paths.

Micro-change and Macro-change in Diachronic Syntax. First edition. Éric Mathieu and Robert Truswell (eds). This chapter © Łukasz Jędrzejowski 2017. First published 2017 by Oxford University Press.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i



Łukasz Jędrzejowski

v2 [inf ti ein Erfolg zu werden]. (2) [Das Buch]i verspricht a success to become.inf the book promise.3sg ‘The book promises to be a success.’ (DeReKo, Mannheimer Morgen, 7/8/2004) Following the standard analysis of subject-to-subject raising, das Buch ‘the book’ in (2) is base-generated in the embedded infinitive clause and then A-moved to the matrix subject position leaving a trace, ti , in the dependent clause. In addition to infinitival complements, both v1 and v2 can embed DP complements as well:2 (3) Der Machthaber verspricht [DP Reformen]. v1 the strongman promise.3sg reforms ‘The strongman promises reforms.’ (DeReKo, Nürnberger Nachrichten, 11/1/2012) (4) Das Wetter verspricht [DP schnelles Winterende]. v2 the weather promise.3sg quick winter.end ‘The weather promises a quick winter’s end.’ (DeReKo, Burgenländische Volkszeitung, 1/3/2012) In the following, I show that although v1 and v2 can embed the same kinds of complements, they occupy two different syntactic positions. Whereas v1 merges as a lexical verbal head, v2 is, in my opinion, base-generated in a functional projection (= FP) below CP and above VP. The main aim of this chapter is twofold. First, I illustrate that v1 has grammaticalized into v2 and provide diachronic evidence underpinning the view that grammaticalization is upward and leftward in the syntactic structure (cf. Roberts and Roussou 1999, 2003; I. Roberts 2010b). Following von Fintel’s (1995) approach, I claim that no semantic bleaching was involved in the development of v2, indicating that functional meanings are permutation-invariant and that functional verbal heads are more complex in their meaning than their lexical counterparts. Second, I propose a new account of how v2 came into being. Contrary to Heine and Miyashita (2004), I argue that the subject-to-subject raising use of v2 emerged out of the pattern v2 + DP, and not out of an infinitive control embedded under v1, as exemplified in (1) above. 2

Diewald and Smirnova (: , (b)) judge the following example as ungrammatical:

(i) *Das Wetter verspricht Verbesserung the weather promise.3sg improvement Intended meaning: ‘The weather promises improvement.’ Note, however, that we can find such corpus examples: (ii) Das Wetter versprach Besserung. the weather promise.3sg.pst improvement (DeReKo, St. Galler Tagblatt, 4/8/1999) In addition, they do not elaborate on the fact that v also allows a dative DP in such contexts, although they register one such example (see Diewald and Smirnova : , ()): (iii) Das Wetter verspricht uns Besserung. the weather promise.3sg us.dat improvement

(DeReKo, Mannheimer Morgen, 26/5/2003)

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

On the grammaticalization of temporal–aspectual heads



This chapter is organized as follows. Section 16.2 highlights selected differences between both uses of versprechen and accounts for where these differences come from. Section 16.3 summarizes previous analyses treating v2 either as an epistemic or as an evidential verbal head. I show that none of these analyses can, in a satisfactory way, account for the distributional properties of v2 and conclude that v2 ought to be associated with a functional projection placed above the verbal phrase. Reis (2007: 39) states that ‘we know as yet far too little about the history of German drohen2/versprechen2 in general’. Thus, this is the main aim of Section 16.4: to offer a new account of how v2 evolved3 and to explore theoretical implications of how functional verbal heads emerge in general. Finally, Section 16.5 summarizes the main results.

. Selected differences between v and v What v1 and v2 have in common is that they can both embed the same complement types, that is, infinitives and DPs. On the other hand, v1 and v2 differ in many respects. In what follows, I point out some of the striking differences between both uses of versprechen, focusing mainly on infinitive complements providing a structural explanation of where these differences come from. .. Imperative mood In contrast to v1, v2 cannot occur in the imperative mood: (5) Versprich mir, ihm das Fliegen beizubringen. v1 promise.2sg.imp me.dat him.dat the flying to.teach.inf ‘Promise me to teach him how to fly.’ (DeReKo, Mannheimer Morgen, 3/11/2011) (6) *Du furchtbares Wetter, versprich schön zu werden. you terrible weather promise.2sg.imp nice to become.inf Intended meaning: ‘You terrible weather, promise to be nice.’

v2

.. Argument structure Whereas the theta grid of v1 can consist of three arguments, as illustrated in (1) above and repeated here for convenience as (7), v2 can take only two arguments; compare (8), the slightly modified example given in (2): (7) [Ich]i verspreche [Ihnen]j , [inf proi/*j heute freundlich zu sein]. today nice to be.inf I promise.1sg you.dat ‘I promise you to be nice today.’

v1

3 Diewald and Smirnova () analyse v as an evidential verb and claim that ‘there has been no explicit discussion on evidentiality as a grammatical category of German, nor is there a comprehensive study on the expression of that category in present-day German or on its diachronic development’ (: ). ‘As to the diachronic development of evidential expressions, it must be stated that in the literature on the history of German and especially on grammaticalization phenomena, the notion of evidentiality is virtually absent’ (: ). What this means is that diachronic studies on the development of v are still missing.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i



Łukasz Jędrzejowski

uns [inf ti ein Erfolg zu werden]. (8) *[Das Buch]i verspricht a success to become.inf the book promise.3sg us.dat Intended meaning: ‘The book promises us to be a success.’

v2

In (8), the indirect object uns ‘us’, marked for the dative case, makes the derivation crash. Note, however, that this restriction does not hold, when v2 selects for a DP complement: (9) Das Wetter verspricht uns [DP Besserung]. the weather promise.3sg us.dat improvement ‘The weather promises us improvement.’ (DeReKo, Mannheimer Morgen, 24/5/2003) I return to this issue in the diachronic discussion in Section 16.4. .. Control shift The ‘defective’ argument structure of v2, used as a subject-to-subject raising verb, goes along with another difference. If v2 cannot license a DP marked for the dative case in infinitive complements, it automatically disallows control shift, in contrast with v1: (10) [Maria]i verspricht [Peter]k [inf prok zur Party gehen v1 M. promise.3sg P.dat to.the party go.inf zu dürfen]. to may.inf ‘Mary promises Peter to be allowed to go to the party.’ (Polinsky 2013: 583, (31a)) (11)

*[Das Wetter]i verspricht [uns]k [inf ti prok schön werden v2 the weather promise.3sg us.dat nice become.inf zu dürfen]. to may.inf Intended meaning: ‘The weather promises us that we will be allowed to be nice.’

.. Indirect speech complements Another difference concerns the types of complements v1 and v2 can take. As we have already seen in this section, both v1 and v2 can realize their internal arguments either as DPs or as infinitives. Remarkably, only v1 admits indirect speech complements: (12) Die Frau hatte ihr versprochen, sie werde v1 the woman have.3sg.pst her.dat promise.ptcp she will.3sg.sbjv dank der Behandlung ihr Glück finden. due the treatment her happiness find.inf ‘The woman promised her that she would find her happiness due to this treatment.’ (DeReKo, St. Galler Tagesblatt, 6/9/2000)

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

On the grammaticalization of temporal–aspectual heads (13) *Das Wetter verspricht, es werde schön. the weather promise.3sg it become.3sg.sbjv nice Intended meaning: ‘The weather promises that it would be nice.’

 v2

In both (12) and (13) the dependent finite clause has declarative force. The embedded verbs, the future auxiliary werde ‘will’ in (12) and its lexical cognate werde ‘become’ in (13), underwent V-to-C movement and are marked for the subjunctive mood. .. Finite dass-clauses Similarly, complement clauses headed by the complementizer dass4 ‘that’ are embeddable only under v1: (14) Präsident Böttger versprach, dass das Fest im nächsten v2 president B. promise.3sg.pst that the festival in.the next Jahr wieder stattfinden wird. year again take.place.inf will.3sg ‘President Böttger promised that the festival will take place in the next year again.’ (DeReKo, Niederösterreichische Nachrichten, 23/8/2012) (15) *Das Wetter versprach, dass es schön wird. the weather promise.3sg.pst that it nice become.3sg Intended meaning: ‘The weather promised that it would be nice.’

v2

If versprechen merges with a human subject and embeds an infinitive, two readings occur:5 (16) Paul verspricht ein guter Arzt zu werden. P. promise.3sg a good doctor to become.inf ‘Paul promises to be a good doctor.’

OK

v1/OK v2

(Reis 2005: 125, (1b))

However, as soon as we turn the infinitive clause into a finite dass-clause, versprechen can only be interpreted as v1: 4 Interestingly, this restriction does not hold for drohen ‘threaten’, when used as a functional verb (see also footnote ). Klein (: , (a–c)) illustrates that it can select for CPs headed by the complementizer dass ‘that’ and that CPs can even be topicalized (see also Heine and Miyashita : ):

(i) Es droht (seit langem), dass Hans gewinnt. it threaten.3sg since long that H. win.3sg ‘Hans threatens to win (for a long time).’ (ii) Seit langem droht, dass Hans gewinnt. since long threaten.3sg that H. win.3sg (iii) Dass Hans gewinnt, droht seit langem. that H. win.3sg threaten.3sg since long 5 This observation sharply contrasts with what Diewald and Smirnova (: ) say about the pattern v + zu-infinitive: ‘Evidential versprechen . . . never occurs with animate subjects.’ At a later point, they weaken this statement in the following way: ‘The subject is mostly inanimate’ (: ).

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i



Łukasz Jędrzejowski

(17) Paul verspricht, dass er ein guter Arzt werden wird. P. promise.3sg that he a good doctor become.inf will.3sg ‘Paul promises that he will be a good doctor.’

OK

v1/*v2

Of course, it does not mean that versprechen cannot merge with nonhuman subjects and, simultaneously, embed finite dass-clauses. As the following case illustrates, such corpus examples exist: (18) Das Wetter verspricht, dass der zwölfte Spieltag the weather promise.3sg that the twelfth matchday stattfinden wird. take.place.inf will.3sg ‘The weather promises that the twelfth matchday will take place.’ (DeReKo, Rhein-Zeitung, 30/10/1999) Due to the nonhuman subject das Wetter ‘the weather’, one might get the impression that versprechen is to be analysed as v2 embedding a finite dass-clause. This would contradict our statement made above though. As Simon Blum (pers. comm.), a native speaker of German, pointed out to me, however, versprechen in (18) should be treated as v1 and not as v2; the subject das Wetter, in turn, has been personified. .. Direct speech complements Whereas v1 admits direct speech complements, their combination with v2 yields illformed results: (19) Messi verspricht: Wir können es schaffen! v1 M. promise.3sg we can.1pl it make.inf ‘Messi is promising: We can do it!’ (DeReKo, Nürnberger Zeitung, 1/5/2013) (20) *Der Tag verspricht: Ich werde schön! the day promise.3sg I become.1sg nice Intended meaning: ‘The day is promising: I will be nice!’

v2

.. Distinct adverbial modifications Although both v1 and v2 can embed infinitive clauses and although there seem to be no differences between the infinitives on the surface, they differ with respect to their structural size. While infinitives embedded under v1 are CPs (see Haider 2009: 272–353 and Sternefeld 2008: 567–9, 625–33 for independent evidence), the size of infinitives selected by v2 is smaller. This difference follows from the fact that an infinitive-embedding predicate and its CP-complement can be separately modified by two distinct temporal adverbials and negations. v2 clearly disallows such constellations: (21) Er hat ihr gestern versprochen, [CP das Auto morgen the car tomorrow he have.3sg her.dat yesterday promise.ptcp abzuholen]. to.come.for.inf ‘Yesterday, he promised her to come for the car tomorrow.’

v1

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

On the grammaticalization of temporal–aspectual heads (22) *Der Tag versprach gestern, [FP heute schön zu werden]. the day promise.3sg.pst yesterday today nice to become.inf Intended meaning: ‘Yesterday, the day promised to be nice tomorrow.’

 v2

(23) Sie hat ihm nicht versprochen, [CP nicht mehr zu rauchen]. v1 she have.3sg him.dat neg promise.ptcp neg more to smoke.inf ‘She didn’t promise him not to smoke anymore.’ (24) *Der Tag versprach nicht, [FP nicht schön zu werden]. the day promise.3sg.pst neg neg nice to become.inf Intended meaning: ‘The day didn’t promise to not become nice.’

v2

.. Extraposition Finally, as opposed to v1, v2 disallows extraposition of embedded infinitives: (25) wenn ich ihnen verspreche, [CP heute freundlich zu sein] if I you.dat promise.1sg today nice to be.inf ‘if I promise to you to be nice today’

v1

(26) *wenn verspricht [FP der Abend interessant zu werden] v2 if promise.3sg the evening interesting to become.inf Intended meaning: ‘if the evening promises to be nice.’ (Reis 2005: 142, (40c)) This difference is not surprising. As observed already by Bech (1955/7), the majority of German control infinitives can be extraposed, while verbal phrases headed, for example, by auxiliary verbs cannot. Note that the example given in (27) is not a counterexample to (26). (27) weil das Wetter verspricht, heiter zu werden v2 because the weather promise.3sg fair to become.inf ‘because the weather promises to be fair.’ (Rosen 1992: 279) What (27) instantiates is the so-called Third Construction, that is, an amalgamation of arguments of the matrix and the embedded clause (see den Besten and Rutten 1989; Reis 2005, 2007; and Wöllstein-Leisten 2001 for more details). If we take v2 to be a subject-to-subject raising predicate when an infinitive clause is embedded, we are forced to assume that das Wetter ‘the weather’ in (27), that is, the matrix subject, has A-moved from the embedded clause into the matrix subject position.6 With that said, 6

Meurers (: , ()) notices the following example.

(i) obwohl heute verspricht ein wunderschöner Tag zu werden even.though today promise.3sg a wonderful day to become.inf ‘even though today looks like it will develop into a wonderful day.’ The example given in (i) can theoretically have two interpretations. On the one hand, heute ‘today’ can be analysed as a subject base-generated in the infinitive clause and A-moved to a higher position in the matrix clause (cf. ()). On the other hand, if we interpret heute as a temporal adverb, we are forced to assume that ein wunderschöner Tag ‘a wonderful day’ is the subject appearing in the right-adjacent part of the clause. Meurers () discusses this example to show that v does not have to form verbal clusters in subordinate environments. However, as Simon Blum and Malika Reetz, native speakers of German, point out to me,

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i



Łukasz Jędrzejowski

Table .. Selected differences between the lexical and functional uses of versprechen v1 (V-head)

v2 (F-head)

       

– – – – – – – –

1. Imperative form 2. Dative argument 3. Control shift 4. Indirect speech complements 5. Finite dass-clauses 6. Direct speech complements 7. Distinct adverbial modifications 8. Extraposition

CP C0

FP F0 ν2

VP V0 ν1

XP

Figure .. Base positions of v1 and v2 in present-day German

we end up having a pooling of arguments of both the matrix and the dependent clause, meaning that there is no strict clause boundary with an extraposed infinitive clause. (28) weil [das Wetter]i verspricht, [fp ti heiter zu werden]

A-movement of the embedded subject In comparison to v1, v2 appears to be ‘defective’ in many respects. Table 16.1 gives an overview of the extent to which v2 differs from v1. In this connection, we need to ask where these differences come from. I argue that they can be accounted for with respect to which structural positions v1 and v2 occupy. Accordingly, I assume v1 to be base-generated as a lexical verbal head within the verbal phrase (= VP), while v2 is associated with a functional projection (= FP) placed below CP and above VP, as in Figure 16.1.

the latter interpretation does not obtain; they judge it either as ‘not good’ or as ‘bad’. But even if one could find similar corpus examples, it would not mean that the raising analysis should be ruled out. One could still assume the subject to A-move within the embedded clause, but not necessarily into a selected matrix position (see also Reis :  for drohen ‘threaten’).

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

On the grammaticalization of temporal–aspectual heads



The differences between v1 and v2 pointed out in this section follow from their different syntactic positions; the ‘defectiveness’—in the sense claimed by Roberts and Roussou (2003)—of v2 is the result of being a functional head. Accordingly, I assume, following Platzack and Rosengren (1998), that imperative verbs are equipped with a feature [imp] which indicates the imperative mood and which is attracted by a formal feature in the Force head in the left periphery of the clause. Consequently, I argue that v2, by virtue of its functional head status, lacks [imp], and is therefore incapable of moving to or agreeing with the Force head. v1 a as lexical verb, in turn, is not restricted in such a way. This accounts for why (6) is ungrammatical. Likewise, we can explain the (im)possibility of embedding indirect speech complements. Because v1 can be used as a speech verb, as exemplified in (12), it allows subjunctive morphology on the verb in the dependent verb-second clause. As functional heads cannot be used as speech verbs, it straightforwardly follows that (13) must be infelicitous. Similarly, both finite dass-clauses and direct complements are compatible only with v1, that is, with the lexical use of versprechen, and not with v2 in its grammatical function. The structural height correlates with the argument structure of v1 and v2, as well. When used as a subject control verb, versprechen can license a DP argument marked for the dative case (cf. Ihnen ‘you’ in (1)). If, on the other hand, it occurs as a subject-to-subject raising verb, the dative argument intervenes and blocks A-movement of the embedded subject. To be more accurate, das Buch ‘the book’ in (8) is base-generated in the embedded infinitive clause and is attracted by the finite matrix verb. However, locality constraints prevent it from moving into the matrix subject position, as uns ‘us’, another DP, intervenes. Nor is this dative argument itself eligible for agreement or movement, beacuse it already has its features checked by some other probe (see Bruening 2014; Chomsky 2000; Hartman 2011 for a recent discussion on this topic). This accounts for why uns ‘us’ in (8) crashes the derivation. Hence, if dative phrases are disallowed in the subject-to-subject raising use of v2, no control shift can occur. Finally, being a functional verbal head occupying a syntactic position above the lexical VP level means that no sentential complements corresponding to full clauses (= CPs) can be selected. If so, we expect v2 to disallow two distinct adverbial modifications and extraposition, as both properties are hallmarks of CP-hood (cf. Haider 2009: 272–353). It follows that v2’s complements must be smaller in size. In the next section, I outline selected approaches which try to explain the functional ‘defectiveness’ of v2 in terms of the structural position it occupies. As it will turn out, two of these approaches fail to account for the differences between v1 and v2.

. Versprechen as a functional verbal head In this section, I briefly discuss three approaches dealing with the functional use of versprechen. The main aim is to demonstrate that v2 cannot be considered to be either an epistemic modal verb or an evidential predicate. As convincingly argued by Reis (2005, 2007), v2 should be treated as a temporal–aspectual predicate. In what follows, I adopt this view, elaborate on Reis’ account, discuss new corpus data supporting her analysis, and, finally, show that v2 occupies a lower functional position in the syntactic tree.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i



Łukasz Jędrzejowski AuxP Aux

¬Θ ModP ¬Θ

Mod

vP/AspP VP

Aux0 ν2

Mod0

v0/Asp0

Figure .. v2 as Aux head and its merge position according to Wurmbrand (2001)

.. Versprechen as an epistemic verb Typologically, v2 and its counterparts in other languages have been analysed as epistemic predicates expressing a subjective judgement on the part of the speaker, cf. Lubańska (2010) and Traugott (1993, 1997) for English, Verhagen (2000, 2006) for Dutch, de Lima (2006) for Portuguese, or Gunkel (2000) and Wurmbrand (2001) for German. According to Wurmbrand (2001: 144), v2 merges as a functional Aux head above the Aspect (= AspP) and Modality (= ModP) Phrases, as shown in Figure 16.2. Wurmbrand (1999, 2001) argues that modal verbs, in particular those taking an epistemic modal base, are subject-to-subject raising verbs merged in AuxP. By definition subject-to-subject raising verbs do not assign theta-roles to their subjects, and do not impose selectional restrictions on A-elements raised from the embedded clause (=¬ in Figure 16.2). Arguments provided for this structural height come from the observation that epistemic modal verbs outscope TP and that they cannot be embedded either under auxiliary verbs or under modal verbs narrowed down by a nonepistemic conversational background (e.g. circumstantial, bouletic, teleological, etc.). Analogously, Wurmbrand (2001) assumes v2 to be an epistemic predicate and examines its subject-to-subject raising verb properties, proposing AuxP as its merge position. Following this line of reasoning, there are five infinitive-embedding predicates in present-day German having subject-to-subject raising status, selecting zu-infinitives, and merging as Aux-heads. They are listed in Table 16.2. Such a high position implies that v2 and the other four predicates listed above are not allowed to be embedded under auxiliary verbs (see also Askedal 1997: 13; Cornillie 2005: 403; Diewald and Smirnova 2010: 118, (28b) for a similar view on v2) and that they cannot occur in a nonfinite form, either as an infinitive or as a participle perfect: [R]aising constructions (i.e., infinitives combining with the verbs begin, threaten, used to, seem, promise) cannot be considered here since these constructions cannot occur in compound tenses in German and hence cannot occur in the perfective construction for independent reasons. (Wurmbrand 2001: 163) Under our analysis, the raising versions of these predicates are generated in Aux, whereas the control versions are main verbs. Evidence for this claim comes from the loss of the raising interpretation when these verbs are embedded under auxiliaries or modals. (Wurmbrand 2001: 209)

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

On the grammaticalization of temporal–aspectual heads



Table .. Subject-to-subject raising verbs in presentday German according to Wurmbrand (2001) Verb

Category

beginnen ‘begin’ drohen ‘threaten’ pflegen ‘use(d) to’ scheinen ‘seem’ versprechen ‘promise’

Aux/lexical restructuring verb Aux/(reduced) non-restructuring verb Aux Aux Aux/(reduced) non-restructuring verb

Empirically, these assumptions cannot be upheld. First, v2 can occur in the scope of the perfect/pluperfect auxiliary haben ‘have’, as the following two examples illustrate: (29) Er ist gut, dieser Maxwell. Und trotzdem noch nicht ganz so he be.3sg good this M. and nevertheless yet neg totally so herausragend, wie er es zu werden versprochen hat. outstanding as he it to become.inf promise.ptcp have.3sg ‘He is good, this Maxwell. Though, he is still not that outstanding, as he promised to be.’ (DeReKo, Zürcher Tagesanzeiger, 21/9/1998, in Colomo 2011: 261, (370c)) (30) Was ein sehenswertes Derby der Handball-Landesliga zu what a worth.seeing derby of.the handball.sixth.division to werden versprochen hatte, verkam in den Augen become.inf promise.ptcp have.3sg.pst go.bad.3sg.pst in the eyes manches Zeugen zur Lachnummer. some spectators to.the laughing.stock ‘What promised to be a worth seeing handball match has been perceived as a laughing stock, according to some spectators.’ (DeReKo, Rhein-Zeitung, 28/11/2000, in Colomo 2011: 260, (370a)) Second, cross-linguistically modal verbs are assumed to occupy two distinct syntactic positions. If they receive a non-epistemic interpretation, they are interpreted as Mod heads merging below TP. If, on the other hand, modal verbs are used as epistemic or evidential verbs, they outscope TP: (31) Moodevidential P > Modepistemic P > TP > Modnon-epistemic P > VP The hierarchy in (31) is in accordance with the rigid hierarchy of functional projections proposed in Cinque (1999, 2006) and advocated in Butler (2004: 138–75, 2006). I illustrate now how temporality interacts with these two modal flavours (non-epistemic vs. epistemic) of modal verbs, taking müssen ‘must’ as an example: (32) Er hat das Zimmer aufräumen müssen. he have.3sg the room clean.up.inf must.inf ‘He had to clean up the room.’

T(Past) > Mod

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i



Łukasz Jędrzejowski

(33) Er wird das Zimmer aufräumen müssen. he will.3sg the room clean.up.inf must.inf ‘He will have to clean up the room.’ (34) Er muss das Zimmer aufgeräumt haben.7 he must.3sg the room clean.up.ptcp have.inf ‘He must have cleaned up the room.’

T(Future) > Mod

Mod > T(Past)

We can deduce from (32)–(34) that syntactically neither the past perfect auxiliary haben nor the future auxiliary werden can take scope over epistemic modality. If they co-occur with modal verbs, they disambiguate their interpretation and only a nonepistemic reading is available. If a modal verb is evaluated against an epistemic modal base, as müssen in (34), the speech act time and the epistemic evaluation time collapse, even though the event time itself is rooted in the past. Accordingly, the embedded proposition falls under the scope of epistemic modality (see also Hacquard 2006, 2010, who provides semantic arguments for the hierarchy given in (31)). Given this line of reasoning, if v2 is taken to be an epistemic verb, it is expected to behave as müssen in (34) does. However, this prediction is not borne out: (35) Diese niedrigen Preise resultieren aus einer Fehleinschätzung der these low prices result.3pl.pst from a miscalculation of.the Ernte 2010, die eine sehr Gute zu werden versprach. harvest 2010 which a very good to become.inf promise.3sg.pst ‘These low prices resulted from a miscalculation of the harvest from 2010, which promised to be a good one.’ (DeReKo, Niederösterreichische Nachrichten, 17/3/2011) (36) Ich habe mich für diesen Chor gemeldet, weil es sehr I have.1sg me.acc for this choir register.ptcp because it very abwechslungsreich zu werden versprach. diversified to become.inf promise.3sg.pst ‘I have registered myself for this choir because it promised to be very diversified.’ (DeReKo, St. Galler Tagblatt, 17/1/2009) The epistemic evaluation time does not outscope the event time the way müssen in (34) does. This can be inferred from the fact that in (35) and (36) v2 bears the past tense morphology. The present tense morphology on v2 leads to ill-formed results: (35 )

*Diese niedrigen Preise resultieren aus einer Fehleinschätzung der Ernte 2010, die eine sehr Gute geworden zu sein verspricht.

7 I do not elaborate here on cases in which the propositional event is rooted in the future tense and the modal gets a deontic interpretation, even though it is accompanied by the past participle, as given in (i).

(i) Du must dein Zimmer bis heute Abend aufgeräumt haben. you must.2sg your room until today evening clean.up.ptcp have.inf ‘You have until this evening to clean up your room.’

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

On the grammaticalization of temporal–aspectual heads



(36 ) *Ich habe mich für diesen Chor gemeldet, weil es sehr abwechslungsreich geworden zu sein verspricht. Third, it is well-known that epistemic modal verbs do not usually occur in root questions (see Maché 2013: 291–7 for a detailed discussion). No such restrictions hold for v2: (37) Verspricht / *Musstepist das Spiel überhaupt interessant zu werden? promise must.3sg the game at.all interesting to become.inf a. ‘Does the game promise to become interesting at all?’ b. *‘Must the game be becoming interesting at all?’ (Reis 2005: 130, (12b)) (38) Verspricht vielleicht der Koffer ein trockenes Plätzchen zu sein? promise.3sg maybe the suitcase a dry little.place to be.inf ‘Does this suitcase promise to be a dry little place?’ (DeReKo, Nürnberger Zeitung, 8/12/2009) Based on what we have seen so far, we can conclude that v2 cannot be analysed as a verb with an epistemic interpretation. In contrast to epistemic modal verbs, v2 is allowed to occur under the scope of the perfect auxiliary haben ‘have’, it cannot bear present tense morphology in past contexts, and it can occur in root questions (for more differences between epistemic modal verbs and v2, see in particular Reis 2005, 2007). On top of that, AuxP presupposes the finiteness of the verbal head. Due to the fact that v2 does occur in nonfinite environments (cf. (29) and (30) above), AuxP cannot be the target position for v2. Keeping these restrictions in mind, I will additionally illustrate below that v2 can occur as an infinitive as well, indicating that AuxP must be ruled out altogether. .. Versprechen as an evidential verb Another attempt to account for the ‘defectiveness’ of v2 relates to works in which v2 has been treated as an evidential verb, see Colomo (2011) and Diewald and Smirnova (2010, 2011) for German, Cornillie (2005, 2007) for Spanish, as well as Rooryck (2001) for data from several languages. Evidentiality is usually associated with the C domain in the cartographic approach proposed in Cinque (1999, 2006) (see (39)) and defined as a grammatical category encoding the source of information, and in a broader sense, marking the speaker’s attitude towards his/her knowledge of reality (cf. Aikhenvald 2004: 1–21 for a discussion of different definitions). (39) [frankly Moodspeech act [fortunately Moodevaluative [allegedly Moodevidential [probably Modepistemic [once T(Past) [then T(Future) [perhaps Moodirrealis [. . . ]]]]]]]] To the best of my knowledge, Rooryck (2001: 132–3) is the first in the generative tradition to treat raising verbs as evidential verbs. In his view, seem and appear in English, passivized forms of selected verbs in Dutch, English, French, and Latin, and Dutch beloven ‘promise’ and dreigen ‘threaten’ and their French counterparts (promettre, menacer) have an evidential interpretation. With respect to the last two verbs, Rooryck (2001: 133) additionally points out that ‘they acquire a meaning that is a combination of future tense and a positive or negative evaluation’. According to his analysis,

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i



Łukasz Jędrzejowski

all of these evidential predicates either covertly raise to Moodevidential 0 or enter an Agree relation with it. As v2 behaves like Dutch beloven and French promettre, it is reasonable to analyse v2 as an evidential verb, too. This is exactly what Colomo (2011) and Diewald and Smirnova (2010, 2011) do, although they do not refer to the work by Rooryck (2001). What is particularly interesting about the hierarchy in (39) in the light of the discussion above is that structurally evidentiality outscopes epistemic modality. This leads to the conclusion—as in Section 16.3.1 above—that this position is too high for v2 and its equivalents in other languages. First, it would again presuppose the absence of nonfinite forms, irrespective of whether v2 would target C0 or Fin0 in the elaborated structure of the left periphery proposed in Rizzi (1997). As the data given in (29) and (30) show, this is not the case though (see also (41) and (42) below). Second, similar observations can be made as to the scope relationships between v2 and auxiliary verbs. In other words, if v2 were attracted by Moodevidential 0 , it would not be allowed to sit in the scope of temporal operators. Again, this view cannot be upheld. Third, Colomo (2011: 235) claims that v2 can be replaced by scheinen ‘seem’, that is, by a classical evidential predicate: (40) a. der Auftakt einer Saison, die auch für Mika Kojonkoski the prelude of.a season which also for M. K. schwieriger zu werden verspricht more.difficult to become.inf promise.3sg ‘the prelude of a season, which promises to become more difficult for Mika Kojonkoski’ (DeReKo, Die Presse, 10/11/1998; in Colomo 2011: 235, (325a)) b. der Auftakt einer Saison, die auch für Mika Kojonkoski schwieriger zu werden scheint (Colomo 2011: 235, (325b)) However, Colomo’s (2011) claim appears to be an overgeneralization that contradicts the empirical data. As (41) and (42) demonstrate, v2 cannot always be replaced by scheinen. As soon as v2 is embedded under evidential scheinen, the replacement possibility is lost: (41) a. Grosse Erwartungen hatten die Anleger gehabt, weil das big expectations have.3pl.pst the investors have.ptcp because the Konzept von Planet Hollywood so viel zu versprechen schien. concept of P. H. so much to promise.inf seem.3sg.pst ‘The investors had big expectations because the concept of Planet Hollywood seemed to promise so much.’ (DeReKo, Zürcher Tagesanzeiger, 19/8/1999) b. *Grosse Erwartungen hatten die Anleger gehabt, weil das Konzept von Planet Hollywood so viel zu scheinen schien. (42) a. The Faint haben mit ihrem Stil eine Nische gefunden, die the F. have.3pl with their style a niche find.ptcp the Erfolg zu versprechen scheint. success to promise.inf seem.3sg ‘Due to their style, the Faint found a niche, which seems to promise to be successful.’ (DeReKo, Mannheimer Morgen, 27/12/2004) b. *The Faint haben mit ihrem Stil eine Nische gefunden, die Erfolg zu scheinen scheint.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

On the grammaticalization of temporal–aspectual heads



What (41) and (42) also indicate is that v2 can occur as an infinitive. This fact strongly suggests that v2 cannot be considered to be either an epistemic or an evidential verb. The structural height of both grammatical categories imposes restrctions that are too strong. As it turns out, no such restrictions are expected to occur if one assumes v2 occupies a functional position directly above the VP. .. Versprechen as a temporal–aspectual verb Finally, Bech (1955/7); Grosse (2005); and Reis (2005, 2007) analyse v2 as a temporal– aspectual predicate. I adopt this view here and claim that v2 externally merges as a functional verbal head (= F-head) directly above VP (see Figure 16.1). As for v2’s meaning, Bech (1955/7: 127) describes it as ungefähr sein ‘be about to’, nahe daran sein ‘be close to’. Reis (2007: 11) takes up Bech’s paraphrases and characterizes v2 as a temporal–aspectual expression ‘locating the realization of p in the imminent future’ (see also the quote from Rooryck 2001 in Section 16.3.2): In our case, the temporal moment is that drohen2/versprechen2 locate p in the future, the putative aspectual moment is that there is contact between the situation given at the speech time and the beginning of the event p, in other words: the reported situation contains clear signs for the imminence of p and/or the incipient state of p. (Reis 2007: 12)

In addition, Reis (2007: (32–3)) provides syntactic evidence showing that v2 behaves like classical phase predicates, for example beginnen ‘begin’ or aufhören ‘stop’, which take zu-infinitives, appear in nonfinite contexts, and allow the Third Construction: (43) Es wird bald zu regnen beginnen / aufhören. it will.3sg soon to rain.inf begin.inf stop.inf ‘It will soon start to rain/stop raining.’ (44) Es hatte zu regnen begonnen / aufgehört. it have.3sg.pst to rain.inf begin.ptcp stop.ptcp ‘It had started/stopped raining.’ (45) Als der Ballon begann / aufhörte dramatisch zu sinken when the balloon start.3sg.pst stop.3sg.pst dramatically to sink.inf ‘when the balloon started/stopped sinking dramatically’ She concludes that ‘drohen2/versprechen2 form a natural class together with the phase verbs in practically all important respects’ (Reis 2007: 32). Based on these observations, I take v2 to be an overt expression of what has been referred to in the literature as prospective aspect. Frawley (1992) defines prospective aspect as follows: Languages encode not only the beginning and end of an event, but also a point just prior to the beginning of an event. This is known as prospective aspect, though it is sometimes called the intensive because it indicates that an event is expected. The prospective denotes that an event is about to take place or is on the verge of obtaining. (Frawley 1992: 322)

and discusses an example from Nkore-Kiga (Niger–Congo), in which the morpheme ka signals the prospective status of embedded events. What is interesting with respect to the prospective aspect, as Frawley (1992) points out, is the fact that it is associated with perfective events. Note that this is in accordance with what

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i



Łukasz Jędrzejowski

Reis (2007: 14) observes for v2: it does not (easily) combine with stative complements; instead v2 favours eventive complements. Comrie (1976) states that many languages possess overt expressions encoding a prospective meaning, but ‘in some languages it is difficult to find exact equivalents without going into long periphrases’ (1976: 64). He mentions the English expressions to be going to, to be about to, or to be on the point of as being typical prospective aspect markers. Again, their semantics is similar to how Bech (1955/7) and Reis (2007) paraphrase the meaning of v2. Cinque (1999: 99) mentions the English adverbs almost, nearly, and imminently as well as their analogues in other languages, and refers to Kobon (Trans-New Guinea) and Gungbe (Niger– Congo), in which there are dedicated expressions encoding the prospective aspect. Cinque (2006: 44–5) points out that the restructuring verbs cuider (lit. ‘believe’) in Middle French as well as prendere (lit. ‘catch’) in colloquial Italian can be analysed as overt markers of the prospective aspect, as well. What this boils down to is that a single functional category can be encoded by means of different lexical expressions. Following this line of reasoning, I assume versprechen to have grammaticalized into one of the prospective aspect expressions in German. If we take v2 to be merged as a prospective aspect marker placed directly above the VP level, we encounter no problems with regard to its syntax. According to the universal hierarchy of functional projections proposed by Cinque (1999, 2006), the prospective aspect is associated with a lower structural position: (46) [frankly Moodspeech act [fortunately Moodevaluative [allegedly Moodevidential [probably Modepistemic [once T(Past) [then T(Future) [perhaps Moodirrealis [necessarily Modnecessity [possibly Modpossibility [usually Asphabitual [again Asprepetitive(I) [often Aspfrequentative(I) [intentionally Modvolitional [quickly Aspcelerative(I) [already T(Anterior) [no longer Aspterminative [still Aspcontinuative [always Aspperfect(?) [just Aspretrospective [soon Aspproximative [briefly Aspdurative [characteristically(?) Aspgeneric/progressive [almost Aspprospective [completely AspSgCompletive(I) [tutto AspPlCompletive [well Voice [fast/early Aspcelerative(II) [again Asprepetitive(II) [often Aspfrequentative(II) [completely AspSgCompletive(II) ]]]]]]]]]]]]]]]]]]]]]]]]]]]]]] This position, in turn, imposes no syntactic restrictions on the use of v2. It is allowed, for example, to occur both as a past participle ((29)–(30)) and as an infinitive ((41)– (42)). On the other hand, as Aspprospective is a functional projection, v2 is expected to be inherently deficient. Hence, the ‘defectiveness’ of v2 described in Section 16.2 follows from being a functional head. One more question remains to be asked: How did v2 evolve? As I show in Section 16.4, the circumstances surrounding its emergence instantiate a canonical grammaticalization process. In what follows, I account for v2’s development based on the syntactic approach to grammaticalization proposed by Roberts and Roussou (1999, 2003) and I. Roberts (2010b).

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

On the grammaticalization of temporal–aspectual heads



. Reanalysis Cross-linguistically and diachronically, not much attention has been paid to v2 embedding a DP complement. Most studies are concerned with the development of v2 in combination with infinitive complements. Traugott (1993, 1997), for instance, claims that the raising use of promise in English developed out of a control use (cf. (1)). Heine and Miyashita (2004) make similar observations with respect to v2 and propose the development path outlined in Table 16.3 (slightly modified by the author). I challenge this view and argue for a different diachronic scenario. More precisely, I claim that the raising use of v2 with an infinitive came into being due to the pattern v2 + DP, which is absent in the model proposed by Heine and Miyashita (2004), and furthermore that a control infinitive associated with v1 is not a necessary precondition for the development of v2 at all. Grimm and Grimm (1854–1961: 1452) observe, in addition, that the pattern v1 + infinitive is a recent development in the history of German (‘die verbindung mit vorausgehendem infinitivsatz ist jünger’). Accordingly, versprechen has grammaticalized into a functional verbal head selecting DP complements and thus quantifying over a set of objects. I argue that this process happened in Early New High German (1350–1650), as the following example shows.8 , 9

Table .. Grammaticalization of v2 according to Heine and Miyashita (2004) Stage

Context

Meaning

I

(a) human subjects (b) actions that are intended and controlled

Maria verspricht, Paul zu helfen ‘Maria promises to help Paul’

II

(a) human subjects (b) actions that are not necessarily intended and controlled

Klaus verspricht ein guter Arzt zu werden ‘Klaus promises to be a good doctor.’

III

(a) inanimate subjects (b) actions that are not intended and controlled

Sein Buch verspricht ein Erfolg zu werden ‘His book promises to be a success.’

8 Interestingly, Diewald and Smirnova (: ) claim to have analysed the data extracted from the Bonner Frühneuhochdeutschkorpus too. It is not clear to me, however, why they do not mention this example. 9 In the oldest stages of German, i.e. in Old High German (–) and Middle High German (– ), v does not occur. Versprechen is used as a negative verb with the meaning deny in most contexts. Through time, particularly in Middle High German, versprechen started allowing a positive interpretation. Grimm and Grimm (–) attribute this change to the prefix ver- having the potential to encode both negative and positive meanings.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i



Łukasz Jędrzejowski

(47) Das Schaff war von den Wellen dergestalt bezwungen/ daß the ship be.3sg.pst by the waves to.such.an.extent defeat.ptcp that wir keine andere Hoffnung hatten/ als etliche Stein-Klippen welche we no other hope have.1pl than several stone.cliffs which im Meere [DP einige Sicherheit] wider das Wasser/ nicht aber some safety against the water neg but in.the sea wider den Hunger versprachen. against the starvation promise.3pl.pst ‘The ship was hit by the waves to such an extent that we had no other hope than some stone cliffs that promised safety against the water, but not against the starvation.’ (BFK, Christian Weise, 1684, Jugendlust, 25–7, Upper Saxon) In (47), versprechen embeds the DP complement einige Sicherheit ‘some safety’. Its subject, etliche Stein-Klippen ‘several stone cliffs’, is equipped with the feature bundle [−agentive], [−human], [−animate], triggering a v2 reading and, simultaneously, blocking the possibility of forming an imperative clause (cf. (6) above). Adelung (1793–1801: Theil 4: 1144–5) noticed three similar examples: (48) Seine Fähigkeiten versprechen einen großen Mann. his abilities promise.3pl a big man ‘His abilities promise a big man.’ (49) Diese Witterung verspricht eine reiche Ernte. this weather promise.3sg a rich harvest ‘This weather promises a rich harvest.’ (50) Der Anschein verspricht nicht viel. the appearance promise.3sg neg much ‘Appearances do not promise much.’ Grimm and Grimm (1854–1961: 1459) point out that this meaning occurred around 1700 and the example provided in (47) confirms this observation: die sichere aussicht auf etwas eröffnen; die gewisse hoffnung erwecken, dasz etwas eintreten werde, aber ohne die einwirkung des versprechenden. die vorstellung des sprechens tritt zurück. der von der handlung betroffene gegenstand ist gewöhnlich eine sache . . . . dieser gebrauch hat sich erst in der jüngeren sprache herausgebildet; er erscheint um 1700, verbreitet sich dann aber sehr schnell in der schriftsprache. [to have the safe prospect of something; to raise the certain hope that something will happen, but without the influence on what is referred to. The act of speaking does not take place. The object of the action is usually a thing . . . . This use evolved first in the recent language period; it appears around 1700, gains currency in the literary language very quickly. My translation, ŁJ] (Grimm and Grimm 1854–1961: 1459)

Diewald and Smirnova (2010: 287) claim that this pattern occurred first in the eighteenth century and discuss the following case: (51) Gleich der Anfang der Fahrt versprach wenig Gutes. soon the beginning of.the journey promise.3sg.pst little good ‘The beginning of the journey promised little good.’

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

On the grammaticalization of temporal–aspectual heads



As illustrated in (47), this pattern was already present in the seventeenth century. In addition, Diewald and Smirnova (2010) argue that this usage of versprechen differs from how v2 is used with infinitive complements. According to them, only the pattern v2 + infinitive encodes evidentiality, while the pattern v2 + DP, as illustrated in the examples (47)–(52), is semantically linked to v1. They interpret this use of versprechen as an initial stage in the grammaticalization process. If v2 had developed its functional meaning in combination with infinitives, then the examples of v2 given in (47)–(52) are not expected to occur. Diewald and Smirnova (2010) do not really specify the meaning of versprechen attested in the cases in (47)–(52). They argue for a non-evidential status of versprechen, but they make no clear statement as to how to interpret it. On the one hand, they analyse versprechen occurring in the examples (47)–(52) as a lexical verb with non-agentive semantics. On the other hand, they paraphrase it as ‘something positive (desirable) is about to happen’ (2010: 207) and state that in this use versprechen ‘describes a whole situation which is seen as being likely to occur in the future’ (2012: 208). These paraphrases, however, do not have a lot in common with evidentiality; rather they correspond to the meaning of the prospective aspect, in particular when they translate versprechen as be about to. Grimm and Grimm (1854–1961), on the other hand, analyse versprechen in both cases as a single verb, define it as cited above and list various kinds of complements which versprechen can embed in this meaning, in particular DPs and infinitive complements which allow a raising analysis. I follow the latter view and show below how this happened. In this connection, there are several arguments to reject the approach taken by Heine and Miyashita (2004). First, as no diachronic data are discussed, this account is highly speculative and it remains totally unknown when v2 first occurred in the history of German. Second, it is a mystery what role the pattern v2 + DP plays in the whole development path and, what appears to be more crucial, to which stage it should belong. As pointed out by Kiss (2005), this pattern cannot be ignored, as v2 occurs with DP complements more frequently than with infinitives. Third, in GerManC, a corpus from 1650–1800, v2 still occurs only with DP complements: (52) ein Wirthshaus, das [. . . ] uns viel Vergnügen verspricht10 a tavern that us.dat much pleasure promise.3sg ‘a tavern that promises much pleasure.’ (GerManC, 1798) I could not find any examples illustrating the use of v2 with infinitives in GerManC. In total, I extracted 104 examples with versprechen. In three cases, versprechen is to be interpreted as v2 embedding DP complements. The subjects (Wirthshaus ‘tavern’, Beschluß ‘resolution’, Welt ‘world’) are of the type [−agentive], [−human], and [−animate]. In all other cases, as detailed in Table 16.4, versprechen is used either as v1 or as a substantive/participle (= others). In twenty-eight examples (= 27) versprechen embeds an infinitive clause. None of these examples, though, allows a v2 interpretation; even ambiguous cases in which

10 Note that if v occurs with DP complements, it can also license dative phrases, as uns ‘us’ in (). This is possible due to the fact that no infinitive is embedded and no A-movement takes place. Hence, dative phrases cannot intervene.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i



Łukasz Jędrzejowski

Table .. The use of versprechen in the GerManC corpus Pattern

versprechen + DP

human subject nonhuman subject

v1 + infinitive v2 + infinitive v1 + verb-second clause v1 + dass-clause v1 + direct speech complement other uses (substantive, participle) In total

Number of Examples

Percentage

31 3 28 0 1 4 2 35 104

30 1 27 0 1 4 2 33 100

Table .. The use of versprechen in the BFK corpus Pattern

versprechen + DP

human subject nonhuman subject

v1 + infinitive v2 + infinitive v1 + verb-second clause v1 + dass-clause v1 + direct speech complement other uses (substantive, participle) In total

Number of Examples

Percentage

23 1 13 0 1 1 0 26 65

35.5 1.5 20 0 1.5 1.5 0 40 100

versprechen could be interpreted either as v1 or as v2 do not occur (cf. (16)). A similar situation can be observed with respect to BFK, an Early New High German corpus (see Table 16.5). Only one example was attested with v2 embedding a DP complement. It is given in (47) above. As far as infinitive complements are concerned, I found thirteen examples. One of them deserves particular attention: (53) Hat dir auch die vergangene zeit jemahlen versprochen, have.3sg you.dat also the bygone time ever promise.ptcp wider herumb#zu kommen, wann du sie bedörffen möchtest? again come.back.inf if you her.acc need.inf want.3sg ‘Did the bygone time ever promise you to come back if you would need it again?’ (BFK, Hans Michael Moscherosch, 1650, 22–4, Alsatian)

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

On the grammaticalization of temporal–aspectual heads



In this case, versprechen should, in my view, be analysed as v1, similar to the example given in (18). The subject die vergangene zeit ‘the bygone time’ has been personified and the question as such can be interpreted as an ironic question. A raising analysis, which would be expected under a v2 reading, must be excluded, because the dative phrase dir ‘you’ blocks the A-movement of the embedded subject (see Section 16.2 for more details). Both Table 16.4 and Table 16.5 provide empirical evidence that v2 did not develop out of the pattern v1 + infinitive. If that were the case, we would expect v2 to occur with infinitives earlier; in particular, ambiguous contexts that are necessary for a grammaticalization process would be expected to be available, as well. However, neither BFK nor GerManC contain such examples. All forty-one cases clearly point to the use of versprechen as v1. Fourth, Diewald and Smirnova (2010, 2011) demonstrate that the first occurrences of v2 with infinitives are attested from the eighteenth century onwards: the infinitive constructions with drohen and versprechen are the youngest members of the today’s evidential paradigm in German. Their development towards evidential markers did not start until the 18th century. For this reason, the data from the earlier periods of German could not be used to trace back the exact developmental path of these two constructions. (Diewald and Smirnova 2010: 31) Our own diachronic analysis starts with the situation at the beginning of the 18th century and looks for those changes which most probably led to the establishment of the construction versprechen & zu-infinitive as an evidential construction. (Diewald and Smirnova 2010: 230)

My own empirical data confirm this observation. In the corpora GOE (1772–1827) and KHZ (1737–1877), I found thirty-one clear unambiguous cases in which an infinitive clause is embedded and in which versprechen can be interpreted only as v2 (see Table 16.6 for GOE, and Table 16.7 for KHZ). Examples (54) and (55) are a case in point.

Table .. The use of versprechen in the GOE corpus Pattern

versprechen + DP

human subject nonhuman subject

v1 + infinitive v2 + infinitive v1 + verb-second clause v1 + dass-clause v1 + direct speech complement other uses (substantive, participle) In total

Number of Examples

Percentage

124 23 123 19 4 15 1 53 362

34 6 34 5 1 5 1 14 100

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i



Łukasz Jędrzejowski

Table .. The use of versprechen in the KHZ corpus Pattern

versprechen + DP

human subject nonhuman subject

v1 + infinitive v2 + infinitive v1 + verb-second clause v1 + dass-clause v1 + direct speech complement other uses (substantive, participle) In total

Number of Examples

Percentage

107 32 59 12 4 8 0 59 281

38 11 21 4 2 3 0 21 100

Stage 1

Stage 2

Stage 3

ν1+DP

ν2+DP

ν2+infinitive

Figure .. The development of versprechen as a subject raising verb

(54) ein starker Nordwind versprach, alles Gewölk völlig zu a strong north.wind promise.3sg.pst all clouds totally to vertreiben dispel.inf ‘A strong north wind promises to dispel all clouds totally.’ (GOE, Italienische Reise, 1813–16) (55) das Leben verspricht außergewöhnlich rege zu werden the life promise.3sg extraordinarily active to become.inf ‘The life promises to become an extraordinarily active one.’ (KHZ, Deutsche=Auswanderer=Zeitung 24, 23/3/1852) Based on what could be stated so far, I propose the developmental path in Figure 16.3 for versprechen. I argue that versprechen is a lexical verb, v1, grammaticalized into v2 and that this process happened due to a metaphorical extension of the predicate versprechen. In other words, both factors allowed versprechen to develop a functional, temporal– aspectual meaning. In Early New High German (1350–1650), versprechen underwent grammaticalization developing into v2 and allowing for the selection of DP complements (=Stage 2). In order to be classified as inherent v2, it has to be able to appear with subjects equipped with the features [−agentive], [−human], [−animate]. In this

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

On the grammaticalization of temporal–aspectual heads FP F0

FP F0 versprechen

VP V0 versprechen



DP

DP

Figure .. The grammaticalization of versprechen

use, v2 quantifies over a set of objects.11 Over time, v2 started to quantify over a set of events/situations embedding infinitive clauses (=Stage 3). What is crucial about the development of versprechen from a lexical to functional verbal head is the transition from Stage 1 to Stage 2, which can be modelled as in Figure 16.4. In Stage 1, versprechen is base-generated as a lexical verbal head. It raises to FP when an ambiguous reading occurs. The movement to the higher position is triggered by one or more formal features located in the functional projection. Functional categories, in turn, ‘are those which attract other material, that is they are the categories which act as triggers for movement’ (Roberts and Roussou 2003: 6). Following Eckardt (2006, 2010), I assume that grammaticalization entails changes in the syntactic structure of a sentence and due to the fact that syntactic structure guides semantic composition, it is expected that the compositional meaning of the sentence needs to change, too. This is what we can observe with respect to the development of versprechen. There is a metaphorical component that allows the concept of ‘promise’ to extend to [−human] or [−animate] subjects. Let us concentrate on the example given in (47). If we talk about stone cliffs promising safety, we conceptualize stone cliffs approaching on the horizon as an animate subject, which—as Eckardt (2010: 2690) delineates in a similar fashion for German drohen ‘threaten’—‘volitionally causes the emotional impression that we feel’. Uttering the sentence in (47) in an appropriate context, the speaker makes the hearer consider a more plausible new structural analysis, in particular a structure in which the interpretation of some animate promising agent is infelicitous in the given context. The hearer sees the possibility for a different reading of the utterance and hypothesizes a second possible analysis (the so called Principle of Avoiding Pragmatic Overload in terms of Eckardt 2006). In doing so, the hearer is not restricted with regard to the interpretation of the subject and, simultaneously, allowed to extend to the use of versprechen with [−human] or [−animate] subjects. A similar reanalysis can be observed in (52). As a tavern cannot make a promise, the hearer is forced to interpret versprechen in another way. Even if the speaker could still have had a v1 interpretation in (52), and what seems to be a plausible reading due to the dative phrase uns

11 Of course, it is possible for v to occur with subjects of the type [+human] and/or [+animate]. In such a case, an ambiguous reading usually arises (v vs. v). This step can be seen as an intermediate step between Stage  and Stage  (cf. ()).

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i



Łukasz Jędrzejowski

‘us’, the hearer may have interpreted versprechen as v2, that is, as a temporal–aspectual verb. In other words, a more concrete meaning has been reinterpreted in a discrete way to yield a more abstract meaning (see also Roberts and Roussou 2003: 220 for a similar reanalysis of the Greek verb thelo). The abstract meaning of v2 or the semantic shift of versprechen, in turn, came about due to a metaphorical extension. Stage 2 is achieved as soon as versprechen externally merges in FP and no movement is involved. When versprechen is used functionally, no movement takes place. Hence, the lexical item, v1, that formerly realized a lower head (= V0 ) has become the realization of a higher functional head (=F0 ). What versprechen instantiates is a change from the Move to the Merge option (= F*Move > F*Merge ), or, in the markedness-based theory advocated in Roberts and Roussou (1999, 2003), a change from a marked to a less marked form.

. Conclusion The main objective of this chapter has been to show that versprechen can be used in two different ways in present-day German: either as a lexical verbal head or as a functional verbal head. I have discussed several syntactic differences between both predicates and argued that versprechen in its grammatical function cannot be analysed either as an epistemic or as an evidential predicate. Instead, I have proposed, based on Reis (2005, 2007), to treat it as a functional verbal head occupying a functional projection placed directly above the VP level. As for its diachrony, I have illustrated that versprechen grammaticalized in Early New High German (1350–1650) and first started by selecting DP complements which quantified over a set of objects. As time went on, the pattern v2 + DP paved the way for infinitives allowing v2 to quantify over a set of events and situations as well as giving rise to the subject raising verb use.

Primary sources BFK: Das Bonner Frühneuhochdeutschkorpus (https://korpora.zim.uni-duisburg-essen.de/ Fnhd/). DeReKo: Das Deutsche Referenzkorpus (http://www.ids-mannheim.de/cosmas2/). GerManC: GerManC Korpus, Historische Texte aus der Zeit von 1650 bis 1800. GOE: Goethe-Korpus, 1772–1827. KHZ: Mannheimer Korpus Historischer Zeitungen, 1737–1877.

Acknowledgements I wish to express my gratitude to Éric Mathieu, Rob Truswell, and two anonymous reviewers who provided many useful suggestions which considerably improved the readability of this chapter. Earlier versions of this work were presented at the conferences Diachronic Generative Syntax 15 at the University of Ottawa (August 2013) and Sprachwandel und seine Reflexe im Neuhochdeutschen at the University of Siegen (September 2012) as well as at the workshop Systematic Semantic Change at the

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

On the grammaticalization of temporal–aspectual heads



University of Texas at Austin (April 2013). I thank the audiences as well as Simon Blum, Ulrike Demske, Ashwini Deo, Elly van Gelderen, Anthony Kroch, Roumyana Pancheva, Malika Reetz, and David Willis for their insightful comments and thoughtprovoking questions. Many thanks also go to Heidi Klockmann for proofreading the manuscript. Obviously, all errors and inadequacies are my own.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

References Abraham, Werner (1997). The interdependence of case, aspect, and referentiality in the history of German: The case of the genitive. In Parameters of Morphosyntactic Change (ed. A. van Kemenade and N. Vincent), pp. 29–61. Cambridge University Press, Cambridge. Acedo Matellán, Victor (2010). Argument Structure and the Syntax–Lexicon Interface. Ph.D. thesis, Universitat de Barcelona. Acedo Matellán, Victor and Mateu, Jaume (2013). Satellite-framed Latin vs. verb-framed Romance: A syntactic approach. Probus, 25, 227–65. Ackema, Peter and Neeleman, Ad (2004). Beyond Morphology: Interface Conditions on Word Formation. Oxford University Press, Oxford. Adams, James (1991). Some neglected evidence for Latin habeo with infinitive: The order of the constituents. Transactions of the Philological Society, 89, 131–96. Adams, James (1994). Wackernagel’s Law and the Placement of the Copula esse in Classical Latin. Philological Society, Cambridge. Adams, James (2013). Social Variation and the Latin Language. Cambridge University Press, Cambridge. Adams, Marianne (1987). From Old French to the theory of pro-drop. Natural Language & Linguistic Theory, 5, 1–32. Adams, Marianne (1989). Verb second effects in Medieval French. In Studies in Romance Linguistics: Selected Papers from the Seventeenth Linguistic Symposium on Romance Languages (ed. C. Kirschner and J. DeCesaris), pp. 1–33. John Benjamins, Amsterdam. Adelung, Johann Christoph (1793–1801). Grammatisch-kritisches Wörterbuch der Hochdeutschen Mundart, mit beständiger Vergleichung der übrigen Mundarten, besonders aber der Oberdeutschen. Piechler, Wien. Ahern, Christopher and Clark, Robin (to appear). Conflict, cheap talk, and Jespersen’s Cycle. Semantics and Pragmatics. Aikhenvald, Alexandra (2004). Evidentiality. Oxford University Press, Oxford. Aissen, Judith (2003). Differential object marking: Iconicity vs. economy. Natural Language & Linguistic Theory, 21, 435–83. Albright, Adam (2009). Modeling analogy as probabilistic grammar. In Analogy in Grammar: Form and Acquisition, (ed. J. Blevins and J. Blevins) pp. 185–213. Oxford University Press, Oxford. Albright, Adam and Hayes, Bruce (2003). Rules vs. analogy in English past tenses: A computational/experimental study. Cognition 90, 119–61. Albu, Mihai (2006). Quantitative Analyses of Typological Data. Ph.D. thesis, Universität Leipzig. Aldridge, Edith (2012). PPs and applicatives in late Archaic Chinese. Studies in Chinese Linguistics, 33, 139–64. Alexiadou, Artemis, Anagnostopoulou, Elena, and Sevdali, Christina (2013). Patterns of dative–nominative alternations. In NELS 41: Proceedings of the 41st Annual Meeting of the North East Linguistic Society (ed. Y. Fainleib, N. LaCara, and Y. Park), pp. 15–28. GSLA, Amherst, MA.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

References



Alexiadou, Artemis, Anagnostopoulou, Elena, and Sevdali, Christina (2014). Opaque and transparent datives, and how they behave in passives. Journal of Comparative Germanic Linguistics, 17, 1–34. Allen, Cynthia (1980). Topics in Diachronic English Syntax. Garland, New York. Altmann, Gabriel, von Buttlar, H., Rott, W., and Strauß, U. (1983). A law of change in language. In Historical Linguistics (ed. B. Brainard), pp. 104–15. Studienverlag Brockmeyer, Bochum. Andersen, Henning (1973). Abductive and deductive change. Language, 49, 765–93. Anderson, Stephen and Lightfoot, David (2002). The Language Organ: Linguistics as Cognitive Physiology. Cambridge University Press, Cambridge. Andersson, Lars-Gunnar (1975). Form and Function of Subordinate Clauses. Ph.D. thesis, Göteborgs Universitet. Arad, Maya (1998). VP Structure and the Syntax–Lexicon Interface. Ph.D. thesis, University College London. Aristar, Anthony (1991). On diachronic sources and synchronic patterns: An investigation into the origin of linguistic universals. Language, 67, 1–33. Askedal, John Ole (1997). Drohen und versprechen als sogennante ‘Modalitätsverben’ in der deutschen Gegenwartssprache. Deutsch als Fremdsprache, 34, 12–19. Baayen, R. Harald (2008). Analyzing Linguistic Data: A Practical Introduction to Statistics using R. Cambridge University Press, Cambridge. Babko-Malaya, Olga (1999). Zero Morphology: A Study of Aspect, Argument Structure and Case. Ph.D. thesis, Rutgers University. Bahloul, Maher and Harbert, Wayne (1993). Agreement asymmetries in Arabic. In Proceedings of the Eleventh West Coast Conference on Formal Linguistics (ed. J. Mead), pp. 15–31. CSLI, Stanford, CA. Bailey, Charles-James (1973). Variation and Linguistic Theory. Center for Applied Linguistics, Arlington, VA. Baker, Mark (1988). Incorporation: A Theory of Grammatical Function Changing. University of Chicago Press, Chicago. Baker, Mark (1998). The macroparameter in a microparametric world. In The Limits of Syntactic Variation (ed T. Biberauer), pp. 351–73. John Benjamins, Amsterdam. Baker, Mark (2012). On the relationship of object agreement and accusative case: Evidence from Amharic. Linguistic Inquiry, 43, 255–74. Baker, Mark (2014). Types of crosslinguistic variation in case assignment. In Linguistic Variation in the Minimalist Framework (ed. M. C. Picallo), pp. 36–60. Oxford University Press, Oxford. Baker, Mark and Vinokurova, Nadya (2010). Two modalities of Case assignment: Case in Sakha. Natural Language & Linguistic Theory, 28, 593–642. Bakker, Stéphanie (2009). The Noun Phrase in Ancient Greek. A Functional Analysis of the Order and Articulation of NP Constituents in Herodotus. Brill, Leiden. Baron, Naomi (1977). Language Acquisition and Historical Change. North Holland, Amsterdam. Bassano, Dominique (1996). Functional and formal constraints on the emergence of epistemic modality: A longitudinal study on French. First Language, 16, 77–113. Batllori, Montserrat, Iglésias, Narcís, and Martins, Ana Maria (2005). Sintaxi dels clítics pronominals en catalá medieval. Caplletra, 38, 137–77.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i



References

Baxter, William and Sagart, Laurent (2014). The Baxter–Sagart Old Chinese reconstruction, version 1.1 (20 September 2014). http://ocbaxtersagart.lsait.lsa.umich.edu/. Accessed 21 June 2015. Bayer, Josef (1984). COMP in Bavarian syntax. The Linguistic Review, 3, 209–74. Beavers, John, Levin, Beth, and Tham, Shiao Wei (2010). Typology of motion expressions revisited. Journal of Linguistics, 46, 331–77. Bech, Gunnar (1955/7). Studien über das deutsche Verbum infinitum. Munksgaard, Kopenhagen. Bech, Kristin and Salvesen, Christine Meklenborg (2014). Preverbal word order in Old English and Old French. In Information Structure and Syntactic Change in Germanic and Romance languages (ed. K. Bech and K. Eide), pp. 233–70. John Benjamins, Amsterdam. Benincà, Paola (2006). A detailed map of the left periphery of Medieval Romance. In Crosslinguistic Research in Syntax and Semantics: Negation, Tense and Clausal Architecture (ed. R. Zanuttini, H. Campos, E. Herburger, and P. Portner), pp. 53–86. Georgetown University Press, Washington, DC. Benincà, Paola and Poletto, Cecilia (2004). Topic, focus, and V2: Defining the CP sublayers. In The Cartography of Syntactic Structures, Volume 2: The Structure of CP and IP (ed. L. Rizzi), pp. 52–75. Oxford University Press, Oxford. Benzécri, Jean-Paul (1973). L’analyse des données. Tome 2: L’analyse des correspondances. Dunod, Paris. Bereczki, Gábor (1983). A Volga–Káma-vidék nyelveinek areális kapcsolatai [Areal contacts of the languages of the Volga–Kama area]. In Areális nyelvészeti tanulmányok (ed. J. Balázs), pp. 207–37. Tankönyvkiadó, Budapest. Bereczki, Gábor (1993). The character and the scale of Turkic influence on the structure of Finno-Ugric languages. In Comparative–Historical Linguistics: Indo-European and FinnoUgric (ed. B. Brogyányi and R. Lipp), pp. 501–19. John Benjamins, Amsterdam. Berwick, Robert and Chomsky, Noam (2011). The biolinguistic program: The current state of its evolution and development. In The Biolinguistic Enterprise: New Perspectives on the Evolution and Nature of the Human Language Faculty (ed. A. M. Di Sciullo and C. Boeckx), pp. 19–41. Oxford University Press, Oxford. Biberauer, Theresa (2002). Reconsidering embedded verb second: How ‘real’ is this phenomenon? Cambridge Occasional Papers in Linguistics, 8, 25–60. Biberauer, Theresa, Holmberg, Anders, and Roberts, Ian (2014). A syntactic universal and its consequences. Linguistic Inquiry, 45, 169–225. Blass, Friedrich and Debrunner, Albert (1961). A Greek Grammar of the New Testament and Other Christian Literature. University of Chicago Press, Chicago. Bloom, Lois (1991). Language Development from Two to Three. Cambridge University Press, Cambridge. Bloom, Lois (2000). The intentionality model of word learning: How to learn a word, any word. In Becoming a Word Learner: A Debate on Lexical Acquisition (ed. R. Golinkoff and K. HirshPasek), pp. 19–50. Oxford University Press, Oxford. Blythe, Richard and Croft, William (2012). S-curves and the mechanisms of propagation in language change. Language, 88, 269–304. Boas, Hans and Sag, Ivan (2012). Sign-Based Construction Grammar. CSLI, Stanford, CA. Bock, Kathryn (1986). Syntactic persistence in language production. Cognitive Psychology, 18, 355–87.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

References



Bock, Kathryn and Loebell, Helga (1990). Framing sentences. Cognition, 35, 1–39. Borer, Hagit (1984). Parametric Syntax: Case Studies in Semitic and Romance Languages. Foris, Dordrecht. Borer, Hagit (2004). The grammar machine. In The Unaccusativity Puzzle (ed. A. Alexiadou, E. Anagnostopoulou, and M. Everaert), pp. 288–331. Oxford University Press, Oxford. Borer, Hagit (2005a). Structuring Sense. Volume 1: In Name Only. Oxford University Press, Oxford. Borer, Hagit (2005b). Structuring Sense. Volume 2: The Normal Course of Events. Oxford University Press, Oxford. Borik, Olga (2002). Aspect and Reference Time. LOT, Utrecht. Bosch, Peter and Umbach, Carla (2007). Reference determination for demonstrative pronouns. In Intersentential Pronominal Reference in Child and Adult Language (ed. D. Bittner and N. Gagarina), pp. 39–51. Zentrum für Allgemeine Sprachwissenschaft, Berlin. Bosch, Peter, Katz, Graham, and Umbach, Carla (2007). The non-subject bias of German demonstrative pronouns. In Anaphors in Text: Cognitive, Formal and Applied Approaches to Anaphoric Reference (ed. M. Schwarz-Friesel, M. Consten, and M. Knees), pp. 145–64. John Benjamins, Amsterdam. Bosch, Peter, Rozario, Tom, and Zhao, Yufan (2003). Demonstrative pronouns and personal pronouns. German der vs. er. In Proceedings of the EACL 2003 Workshop on the Computational Treatment of Anaphora, pp. 61–8. Bourova, Viara (2005). A la recherche du ‘conditionnel latin’: les constructions ‘infinitif + forme de habere’ examinées à partir d’un corpus électronique. In Romanistische Korpuslinguistik: Korpora und diachrone Sprachwissenschaft (ed. C. Pusch, J. Kabatek, and W. Raible), pp. 303–16. Gunter Narr Verlag, Tübingen. Bourova, Viara and Tasmowski, Liliane (2007). La préhistoire des futurs romans: ordre des constituants et sémantique. In Études sémantiques et pragmatiques sur le temps, l’aspect et la modalité (ed. L. de Saussure, J. Moeschler, and G. Puskás), pp. 25–41. Rodopi, Amsterdam. Bowerman, Melissa (1985). What shapes children’s grammars? In The Crosslinguistic Study of Language Acquisition, Volume 2: Theoretical Issues (ed. D. Slobin), pp. 1257–319. Lawrence Erlbaum, Hillsdale, NJ. Bowern, Claire (2005). Syntactic change and syntactic borrowing in generative grammar. In Principles of Syntactic Reconstruction (ed. G. Ferraresi and M. Goldbach), pp. 187–216. John Benjamins, Amsterdam. Branigan, Holly, Pickering, Martin, Liversedge, Simon, Stewart, Andrew, and Urbach, Thomas (1995). Syntactic priming: Investigating the mental representation of language. Journal of Psycholinguistic Research, 24, 489–506. Bresnan, Joan (2001). Lexical-Functional Syntax. Blackwell, Oxford. Bresnan, Joan and Grimshaw, Jane (1978). The syntax of free relatives in English. Linguistic Inquiry, 9, 331–91. Brinton, Laurel (1988). The Development of English Aspectual Systems. Cambridge University Press, Cambridge. Brown, Roger (1973). A First Language: The Early Stages. Harvard University Press, Cambridge, MA. Bruening, Benjamin (2014). Defects of defective intervention. Linguistic Inquiry 45, 707–19.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i



References

Buelens, Lisa and D’Hulster, Tijs (2014). On the edge of acceptability: Arguments for the syntactic dependence of the Flemish external possessor on the possessee DP. Phrasis, 50, 51–93. Bungenstab, Edith (1933). Der Genitiv beim Verbum und sein Ersatz im Laufe der englischen Sprachgeschichte. Eschenhagen, Breslau. Buridant, Claude (2000). Grammaire nouvelle de l’ancien français. Sedes, Paris. Büring, Daniel (1994). Topic. In Focus & Natural Language Processing, Volume 2 (ed. P. Bosch and R. van der Sandt), pp. 271–80. IBM, Heidelberg. Büring, Daniel (1997). The Meaning of Topic and Focus: The 59th Street Bridge Accent. Routledge, London. Büring, Daniel. (2003). On D-trees, beans, and B-accents. Linguistics and Philosophy, 26, 511–45. Burnett, Heather and Tremblay, Mireille (2009). Variable-behaviour Ps and the location of PATH in Old French. In Romance Languages and Linguistic Theory: Selected Papers from ‘Going Romance’ Amsterdam 2007 (ed. E. Aboh, E. van der Linden, J. Quer, and P. Sleeman), pp. 25–50. John Benjamins, Amsterdam. Burnett, Heather and Troberg, Michelle (2014). The diachronic semantics of resultative secondary predication in French. In Empirical Issues in Syntax and Semantics 10 (ed. C. Piñón), pp. 37–54. Available at http://www.cssp.cnrs.fr/eiss10/. Butler, Jonny (2004). Phase Structure, Phrase Structure, and Quantification. Ph.D. thesis, University of York. Butler, Jonny (2006). The structure of modality and temporality (or, towards deriving something like a Cinque hierarchy). Linguistic Variation Yearbook, 6, 161–201. Bybee, Joan and Slobin, Dan (1982a). Rules and schemas in the development and use of the English past tense. Language, 58, 265–89. Bybee, Joan and Slobin, Dan (1982b). Why small children cannot change language on their own: Evidence from the past tense. In Papers from the Fifth International Conference on Historical Linguistics, Galway, April 6–10 1981 (ed. A. Ahlqvist), pp. 29–37. John Benjamins, Amsterdam. Bybee, Joan, Perkins, Revere, and Pagliuca, William (1994). The Evolution of Grammar: Tense, Aspect, and Modality in the Languages of the World. University of Chicago Press, Chicago. Campbell, Richard (1989). The Grammatical Structure of Verbal Predicates. Ph.D. thesis, University of California, Los Angeles. Canale, William (1978). Word Order Change in Old English: Base Reanalysis in Generative Grammar. Ph.D. thesis, McGill University. Cardinaletti, Anna (2003). Stylistic fronting in Italian. In Grammatik i Focus / Grammar in Focus. Festschrift for Christer Platzack, Volume 2 (ed. L.-O. Delsing, C. Falk, G. Josefsson, and H. A. Sigurðsson), pp. 47–55. Wallin and Dalholm, Lund. Cardinaletti, Anna and Roberts, Ian (2002). Clause structure and X-second. In The Cartography of Syntactic Structures, Volume 1: Functional Structure in DP and IP (ed. G. Cinque), pp. 123–66. Oxford University Press, Oxford. Cardinaletti, Anna and Starke, Michal (1999). The typology of structural deficiency: A case study of the three grammatical classes. In Clitics in the languages of Europe (ed. H. van Riemsdijk), pp. 145–233. Mouton de Gruyter, Berlin. Carlson, Lauri (1983). Dialog Games: An Approach to Discourse Analysis. Reidel, Dordrecht.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

References



Carstens, Vicki (2003). Rethinking complementizer agreement: Agree with a case-checked goal. Linguistic Inquiry, 34, 393–412. Carstens, Vicki (2005). Agree and EPP in Bantu. Natural Language & Linguistic Theory, 23, 219–79. Cekaite, Asta, Blum-Kulka, Shoshana, Grøver, Vibeke, and Teubal, Eva (2014). Children’s Peer Talk: Learning from Each Other. Cambridge University Press, Cambridge. Cennamo, Michela (2001). On the reorganization of voice distinctions and grammatical relations in Late Latin. In De lingua latina novae quaestiones (ed. C. Moussy, J. Dangel, M. Fruyt, L. Sznajder, and L. Nadjo), pp. 51–65. Peeters, Louvain-la-Neuve. Cennamo, Michela (2005). Passive auxiliaries in Late Latin. In Latin et langues romanes: études de linguistique offertes à József Herman à l’occasion de son 80e` me anniversaire (ed. S. Kiss, L. Mondin, and G. Salvi), pp. 177–94. Max Niemeyer, Tübingen. Chen, Jidong and Shirai, Yasuhiro (2015). The acquisition of relative clauses in spontaneous child speech in Mandarin Chinese. Journal of Child Language, 42, 394–422. Chocano, Gema (2012). On the fronting of non-contrastive topics in Germanic. In Of Grammar, Words, and Verses: In Honor of Carlos Piera (ed. E. Torrego), pp. 143–69. John Benjamins, Amsterdam. Chomsky, Noam (1957). Syntactic Structures. Mouton, The Hague. Chomsky, Noam (1965). Aspects of the Theory of Syntax. MIT Press, Cambridge, MA. Chomsky, Noam (1977). On wh-movement. In Formal Syntax (ed. P. Culicover, T. Wasow, and A. Akmajian), pp. 71–132. Academic Press, New York. Chomsky, Noam (1981). Lectures on Government and Binding. Foris, Dordrecht. Chomsky, Noam (1986). Knowledge of Language: Its Nature, Origins, and Use. Praeger, New York. Chomsky, Noam (1995). The Minimalist Program. MIT Press, Cambridge, MA. Chomsky, Noam (2000). Minimalist inquiries: The framework. In Step by Step: Essays on Minimalist Syntax in Honor of Howard Lasnik (ed. R. Martin, D. Michaels, and J. Uriagereka), pp. 89–115. MIT Press, Cambridge, MA. Chomsky, Noam (2008). On phases. In Foundational Issues in Linguistic Theory: Essays in Honor of Jean-Roger Vergnaud (ed. R. Freidin, C. Otero, and M. L. Zubizarreta), pp. 133–66. MIT Press, Cambridge, MA. Cinque, Guglielmo (1999). Adverbs and Functional Heads: A Cross-linguistic Perspective. Oxford University Press, Oxford. Cinque, Guglielmo (2006). Restructuring and Functional Heads. Oxford University Press, Oxford. Clackson, James (2007). Indo-European Linguistics: An Introduction. Cambridge University Press, Cambridge. Clackson, James and Horrocks, Geoffrey (2007). The Blackwell History of the Latin Language. Blackwell, Oxford. Clancy, Patricia (1975). The rise of the synthetic future tense in Romance: A critique of some recent hypotheses (Benveniste, Valesio, Butler). Romance Philology, 28, 545–54. Clark, Eve (1977). Strategies and the mapping problem in first language acquisition. In Language Learning and Thought (ed. J. Macnamara), pp. 147–68. Academic Press, New York. Clark, Eve (1988). On the logic of contrast. Journal of Child Language, 15, 317–35. Clark, Eve (1993). The Lexicon in Acquisition. Cambridge University Press, Cambridge. Clark, Eve and Carpenter, Kathie (1989). The notion of source in language acquisition. Language, 65, 1–30.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i



References

Clark, Robin and Roberts, Ian (1993). A computational model of language learnability and language change. Linguistic Inquiry, 24, 299–345. Claudi, U. (1993). Die Stellung von Verb und Objekt in Niger–Kongo-Sprachen: Ein Beitrag zur Rekonstruktion historischer Syntax. Institüt für Afrikanistik, Universität zu Köln. Coleman, Robert (1971). The origin and development of Latin habeo + infinitive. Classical Quarterly n.s., 21, 215–32. Coleman, Robert (1976). Further observations on habeo + infinitive as an exponent of futurity. Classical Quarterly n.s., 26, 151–9. Colomo, Katarina (2011). Modalität im Verbalkomplex. Halbmodalverben und Modalitätsverben im System statusregierender Verbklassen. Ph.D. thesis, Ruhr-Universität Bochum. Comparetti, Domenico and Vitelli, Girolamo (1906–15). Papiri Fiorentini: documenti pubblici e privati dell’ et à romana e byzantina (Papiri Greco–Egizii). Hoepli, Milan. 3 volumes. Comrie, Bernard (1976). Aspect. Cambridge University Press, Cambridge. Comrie, Bernard (1989). Language Universals and Linguistic Typology. University of Chicago Press, Chicago. Comrie, Bernard (1998). Rethinking the typology of relative clauses. Language Design, 1, 59–86. Cooper, Adam and Georgala, Effi (2012). Dative loss and its replacement in the history of Greek. In Historical Linguistics 2009. Selected Papers from the 19th International Conference on Historical Linguistics (ed. A. van Kemenade and N. de Haas), pp. 275–92. John Benjamins, Amsterdam. Cooper III, Guy (1998). Attic Greek Prose Syntax, vol. II. University of Michigan Press, Ann Arbor, MI. Copley, Bridget (2002). The Semantics of the Future. Ph.D. thesis, Massachusetts Institute of Technology. Cornillie, Bert (2005). A paradigmatic view of Spanish amenazar ‘to threaten’ and prometer ‘to promise’. Folia Linguistica, 39, 385–415. Cornillie, Bert (2007). Evidentiality and Epistemic Modality in (Semi-)Auxiliaries. De Gruyter, Berlin. Cournane, Ailís (2014). In search of L1 evidence for diachronic reanalysis: Mapping modal verbs. Language Acquisition, 21, 103–17. Cournane, Ailís (2015a). Modal Development: Input-divergent L1 Acquisition in the Direction of Diachronic Reanalysis. Ph.D. thesis, University of Toronto. Cournane, Ailís (2015b). Revisiting the epistemic gap: Evidence for a grammatical source. In BUCLD 39: Proceedings of the 39th Annual Boston University Conference on Language Development (ed. E. Grillo and K. Jepson), Somerville, MA. Cascadilla Press. Courtney, Sarah (in preparation). Variation and Diachrony in Complementizer Agreement. Ph.D. thesis, Cornell University. Cowper, Elizabeth (2003). Tense, mood and aspect: A feature-geometric approach. Ms., University of Toronto. Culicover, Peter and Jackendoff, Ray (2005). Simpler Syntax. Oxford University Press, Oxford. Cysouw, Michael (2003). Against implicational universals. Linguistic Typology, 7, 89–101. Cysouw, Michael (2005). Quantitative methods in typology. In Quantitative Linguistik: Ein internationales Handbuch (ed. R. Köhler, G. Altmann, and R. Piotrowski), pp. 554–78. De Gruyter, Berlin.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

References



Cysouw, Michael (2007). New approaches to cluster analysis of typological indices. In Exact Methods in the Study of Language and Text (ed. R. Köhler and P. Grzbek), pp. 61–76. Mouton de Gruyter, Berlin. Dagnac, Anne (2012). How do you double your C? Evidence from an oïl dialect. In Empirical Issues in Syntax and Semantics 9 (ed. C. Piñón), pp. 77–94. CNRS, Paris. Dahl, Östen (1985). Aspect. Cambridge University Press, Cambridge. Dahl, Östen and Karlsson, Fred (1976). Verbien aspektit ja objektin sijamerkintä. Sananjalka, 18, 28–52. Danckaert, Lieven (2012). Latin Embedded Clauses: The Left Periphery. John Benjamins, Amsterdam. Danckaert, Lieven (2014). The derivation of Classical Latin Aux-final clauses: Implications for the internal structure of the verb phrase. In Romance Languages and Linguistic Theory 2012 (ed. K. Lahousse and S. Marzo), pp. 141–59. John Benjamins, Amsterdam. Danckaert, Lieven (2015). Assessing phonological correlates of syntactic change: The case of Late Latin BE-periphrases. Ms., Universiteit Gent. Danckaert, Lieven (2016). Variation and change in Latin BE-periphrases: Empirical and methodological considerations. In Early and Late Latin (ed. J. Adams and N. Vincent). Cambridge University Press, Cambridge. Danckaert, Lieven (2017). The Development of Latin Clause Structure: A Study of the Extended Verb Phrase. Oxford University Press, Oxford. Danckaert, Lieven (to appear). The decline of Latin VOAux: Neg-incorporation and syntactic reanalysis. In Word Order Change (ed. A. M. Martins and A. Cardoso). Oxford University Press, Oxford. de Lima, José Pinte (2006). Zur Grammatikalisierung von dt. drohen und pg. ameaçar. In Portugiesisch kontrastiv gesehen und Anglizismen weltweit (ed. J. Schmidt-Radefeldt and R. Windisch), pp. 205–20. Peter Lang, Frankfurt am Main. de Melo, Wolfgang (2012). Kuryłowicz’s first ‘law of analogy’ and the development of passive periphrases in Latin. In Laws and Rules in Indo-European (ed. P. Probert and A. Willi), pp. 83–101. Oxford University Press, Oxford. de Villiers, Jill (2007). The interface of language and Theory of Mind. Lingua, 117, 1858–78. de Villiers, Jill, Garfield, Jay, Gernet-Girard, Harper, Roeper, Tom, and Speas, Margaret (2009). Evidentials in Tibetan: Acquisition, semantics, and cognitive development. New Directions for Child and Adolescent Development, 125, 29–47. de Vogelaer, Gunther and van der Auwera, Johan (2010). When typological rara generate rarissima: Analogical extension of verbal agreement in Dutch dialects. In Rara & Rarissima: Documenting the Fringes of Linguistic Diversity (ed. J. Wohlgemuth and M. Cysouw), pp. 47–74. Mouton de Gruyter, Berlin. De Vries, Mark (2002). The Syntax of Relativization. Ph.D. thesis, Universiteit van Amsterdam. De Vries, Mark (2006). The syntax of appositive relativization: On specifying coordination, false free relatives, and promotion. Linguistic Inquiry, 37, 229–70. De Vries, Mark (2012). Parenthetical main clauses—or not? On appositives and quasi-relatives. In Main Clause Phenomena: New Horizons (ed. L. Aelbrecht, L. Haegeman, and R. Nye), pp. 177–201. John Benjamins, Amsterdam. DeLancey, Scott (1997). Grammaticalization and the gradience of categories: Relator nouns and postpositions in Tibetan and Burmese. In Essays on Language Function and Language Type,

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i



References

Volume 2 (ed. J. Hammond, J. Haiman, and S. Thompson), pp. 151–205. Cambridge University Press, Cambridge. den Besten, Hans (1977). On the interaction of root transformations and lexical deletive verbs. Ms., University of Amsterdam. den Besten, Hans and Rutten, Jean (1989). On verb raising, extraposition, and free word order in Dutch. In Sentential Complementation and the Lexicon: Studies in Honour of Wim de Geest (ed. D. Jaspers, Y. Putseys, W. Klooster, and P. Seuren), pp. 41–56. Foris, Dordrecht. den Besten, Hans and Webelhuth, Gert (1987). Adjunction and remnant topicalization in the Germanic SOV-languages. Paper presented at GLOW, Venice, March/April 1987. den Besten, Hans and Webelhuth, Gert (1990). Stranding. In Scrambling and Barriers (ed. G. Grewendorf and W. Sternefeld), pp. 77–92. John Benjamins, Amsterdam. den Dikken, Marcel (2010). On the functional structure of locative and directional PPs. In The Cartography of Syntactic Structures, Volume 6: Mapping Spatial PPs (ed. G. Cinque and L. Rizzi), pp. 74–126. Oxford University Press, Oxford. Denison, David (1993). English Historical Syntax: Verbal Constructions. Longman, London. Derwing, Bruce and Baker, William (1986). Assessing morphological development. In Language Acquisition: Studies in First Language Development (2nd edn) (ed. P. Fletcher and M. Garman), pp. 326–38. Cambridge University Press, Cambridge. Diercks, Michael (2013). Indirect agree in Lukubusu complementizer agreement. Natural Language & Linguistic Theory, 31, 357–407. Diesing, Molly (1990). Verb movement and the subject position in Yiddish. Natural Language & Linguistic Theory, 8, 41–79. Diesing, Molly (1992). Indefinites. MIT Press, Cambridge, MA. Diessel, Holger (2004). The Acquisition of Complex Sentences. Cambridge University Press, Cambridge. Diessel, Holger (2011). Grammaticalization and language acquisition. In Handbook of Grammaticalization (ed. B. Heine and H. Norrog), pp. 130–41. Oxford University Press, Oxford. Diessel, Holger (2012). Language change and language acquisition. In English Historical Linguistics: An International Handbook, Volume 2 (ed. A. Bergs and L. Brinton), pp. 1599–613. Mouton de Gruyter, Berlin. Diewald, Gabriele and Smirnova, Elena (2010). Evidentiality in German: Linguistic Realization and Regularities in Grammaticalization. De Gruyter, Berlin. Diewald, Gabriele and Smirnova, Elena (2011). The German evidential constructions and their origins: A corpus based analysis. In In the Mood for Mood (ed. T. Mortelmans, J. Mortelmans, and W. de Mulder), pp. 81–100. Rodopi, Amsterdam. Djamouri, Redouane and Paul, Waltraud (2012). Deverbal postpositions in Chinese from a diachronic perspective. Paper presented at the Twentieth Annual Meeting of the International Association of Chinese Linguistics, Hongkong Polytechnic University, 29–31 July, 2012. Djamouri, Redouane, Paul, Waltraud, and Whitman, John (2013). Postpositions vs. prepositions in Mandarin Chinese: The articulation of disharmony. In Theoretical Approaches to Disharmonic Word Order (ed. T. Biberauer and M. Sheehan), pp. 74–105. Oxford University Press, Oxford. Donaldson, Bryan (2012a). Diachronic change in Old French clause structure: Effects on initial subordinate clauses. Ms., University of Texas at Austin.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

References



Donaldson, Bryan (2012b). Initial subordinate clauses in Old French: Syntactic variation and the clausal left periphery. Lingua, 122, 1021–46. Dryer, Matthew (1992). The Greenbergian word order correlations. Language, 68, 81–138. Dryer, Matthew (1998). Why statistical universals are better than absolute universals. In Proceedings of the Chicago Linguistic Society 33: The Panels, pp. 123–45. Dryer, Matthew (2011). The evidence for word order correlations. Linguistic Typology, 15, 335–80. Dryer, Matthew and Haspelmath, Martin (2013). The World Atlas of Language Structures online. Available online at http://wals.info. Accessed 24 April 2016. Duarte, Inês and Matos, Gabriela (2000). Romance clitics and the minimalist program. In Portuguese Syntax (ed. J. Costa), pp. 116–42. Oxford University Press, Oxford. Dufresne, Monique, Dupuis, Fernande, and Longtin, Catherine-Marie (2001). Un changement dans la diachronie du français: la perte de la préfixation aspectuelle en à-. Revue québécoise de linguistique, 29, 33–54. Dufresne, Monique, Dupuis, Fernande, and Tremblay, Mireille (2003). Preverbs and particles in Old French. In Yearbook of Morphology 2003 (ed. G. Booij and A. van Kemenade), pp. 30–60. Kluwer, Dordrecht. Dupuis, Fernande (1989). L’ expression du sujet dans les subordonnées en ancien français. Ph.D. thesis, Université de Montréal. E. Abaffy, Erzsébet (1991). Az igemód-és igeidő-rendszer [verbal mood and tense]. In L. Benkő (Ed.), In A magyar nyelv történeti nyelvtana I. A korai ómagyar kor és előzményei, (ed. L. Benkő), pp. 104–21. Akadémiai Kiadó, Budapest. Eckardt, Regine (2006). Meaning Change in Grammaticalization: An Enquiry into Semantic Reanalysis. Oxford University Press, Oxford. Eckardt, Regine (2010). Grammaticalization and semantic reanalysis. In Semantics. An International Handbook of Natural Language Meaning, Volume 3 (ed. C. Maienborn, K. von Heusinger, and P. Portner), pp. 2675–702. De Gruyter, Berlin. Egerland, Verner (2011). Fronting, background, focus: A comparative study of Sardinian and Icelandic. Working Papers in Scandinavian Syntax, 87, 103–35. É. Kiss, Katalin (2006). From the grammaticalization of viewpoint aspect to the grammaticalization of situation aspect. In Event Structure and the Left Periphery (ed. K. É. Kiss), pp. 129–58. Springer, Dordrecht. Embick, David (2000). Features, syntax and categories in the Latin perfect. Linguistic Inquiry, 31, 185–230. Embick, David (2004). On the structure of resultative participles in English. Linguistic Inquiry, 35, 355–92. Emonds, Joseph (1970). Root and Structure-preserving Transformations. Ph.D. thesis, Massachusetts Institute of Technology. Erdal, Marcel (2004). A Grammar of Old Turkic. Brill, Leiden. Ernout, Alfred and Thomas, François (1953). Syntaxe latine (2e édition, tirage revu et corrigé). Klincksieck, Paris. Estival, Dominique (1985). Syntactic priming of the passive in English. Text, 5, 7–21. Evers, Arnold (1981). Verb-second movement rules. Wiener Linguistische Gazette, 26, 15–34. Eyþórsson, Þórhallur (1995). Verbal Syntax in the Early Germanic Languages. Ph.D. thesis, Cornell University.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i



References

Fanselow, Gisbert (2003). Münchhausen-style head movement and the analysis of verb-second. In Syntax at Sunset: Head Movement and Syntactic Theory (ed. A. Mahajan), pp. 40–76. UCLA Working Papers in Linguistics, Los Angeles, CA. Ferreira, Victor (2003). The persistence of optional complementizer production: Why saying “that” is not saying “that” at all. Journal of Memory and Language, 48, 379–98. Fischer, Olga and van der Wurff, Wim (2006). Syntax. In A History of the English Language (ed. R. Hogg and D. Denison), pp. 109–98. Cambridge University Press, Cambridge. Fischer, Olga, van Kemenade, Ans, Koopman, Willem, and van der Wurff, Wim (2000). The Syntax of Early English. Cambridge University Press, Cambridge. Fischer, Susann (2010). Word Order Change as a Source of Grammaticalization. John Benjamins, Amsterdam. Fischer, Susann (2014). Revisiting stylistic fronting in Old Spanish. In Left Sentence Peripheries in Spanish: Diachronic, Variationist and Comparative Perspectives (ed. A. Dufter and A. Octavio de Toledo), pp. 53–76. John Benjamins, Amsterdam. Fischer, Susann and Alexiadou, Artemis (2001). On stylistic fronting: Germanic vs. Romance. Working Papers in Scandinavian Syntax, 68, 117–45. Fisher, Ronald (1922). On the interpretation of χ 2 from contingency tables, and the calculation of P. Journal of the Royal Statistical Society, 85, 87–94. Fleischmann, Suzanne (1989). Temporal distance: A basic linguistic metaphor. Studies in Language, 13, 1–50. Flobert, Pierre (1975). Les verbes déponents latins des origines à Charlemagne. Les Belles Lettres, Paris. Folli, Raffaella and Ramchand, Gillian (2005). Prepositions and results in Italian and English: An analysis from event decomposition. In Perspectives on Aspect (ed. H. Verkuyl, A. van Hout, and H. de Swart), pp. 81–105. Springer, Dordrecht. Fontana, Josep (1993). Phrase Structure and the Syntax of Clitics in the History of Spanish. Ph.D. thesis, University of Pennsylvania. Foulet, Lucien (1958). Petite syntaxe de l’ancien français (3rd edn). Champion, Paris. Franco, Irene (2009). Verbs, Subjects and Stylistic Fronting. A Comparative Analysis of the Interaction of CP Properties with Verb Movement and Subject Positions in Icelandic and Old Italian. Ph.D. thesis, Università di Siena. Franco, Irene (2012). Verbal stylistic fronting in Old Florentine. Ms., Leiden University. Frascarelli, Mara and Hinterhölzl, Roland (2007). Types of topics in German and Italian. In On Information Structure, Meaning and Form: Generalizations across Languages (ed. K. Schwabe and S. Winkler), pp. 87–116. John Benjamins, Amsterdam. Frawley, William (1992). Linguistic Semantics. Lawrence Erlbaum Associates, Hillsdale, NJ. Frey, Werner (2004). The grammar–pragmatics interface and the German prefield. Sprache und Pragmatik, 52, 1–39. Frey, Werner (2006a). Contrast and movement to the German prefield. In The Architecture of Focus (ed. V. Molnár and S. Winkler), pp. 235–64. Mouton de Gruyter, Berlin. Frey, Werner (2006b). How to get an object-es into the German prefield. In Form, Structure, and Grammar: A Festschrift Presented to Günther Grewendorf on the Occasion of his 60th Birthday (ed. P. Brandt and E. Fuß), pp. 159–85. Mouton de Gruyter, Berlin. Frisch, Stefan (1997). The change in negation in Middle English: A NEGP licensing account. Lingua, 101, 21–64.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

References



Fritz, Gerd (1997). Deutsche Modalverben 1609: Nicht-epistemische Verwendungsweisen. In Untersuchungen zur semantischen Entwicklungsgeschichte der Modalverben im Deutschen (ed. G. Fritz and T. Gloning), pp. 249–305. De Gruyter, Berlin. Fruehwald, Josef and Wallenberg, Joel (2013). Optionality is stable variation in competing grammars. Paper presented at 25th Scandinavian Conference of Linguistics, Formal Ways of Analyzing Variation (FWAV) Workshop. Fuß, Eric. (2003). On the historical core of V2 in Germanic. Nordic Journal of Linguistics 26, 195–231. Fuß, Eric. (2005). The Rise of Agreement. John Benjamins, Amsterdam. Fuß, Eric. (2007). Cyclic spell-out and the nature of post-syntactic operations. Linguistic Analysis, 33, 267–302. Gast, Volker (2010). Contrastive topics and distributed foci as instances of sub-informativity. In Comparative and Contrastive Studies of Information Structure (ed. C. Breul and E. Göbbel), pp. 15–50. John Benjamins, Amsterdam. Gathercole, Virginia and Thomas, Enlli (2009). Bilingual first-language development: Dominant language takeover, threatened minority language take-up. Bilingualism: Language and Cognition, 12, 213–37. Gazdar, Gerald, Klein, Ewan, Pullum, Geoffrey, and Sag, Ivan (1985). Generalized Phrase Structure Grammar. Blackwell, Oxford. Gergel, Remus (2009). Rather: On a modal cycle. In Cyclical Change (ed. E. van Gelderen), pp. 243–64. John Benjamins, Amsterdam. Gerőcs, Mátyás (2011). Aspektus és igekötő. A mondat belső időszerkezete a kései ómagyarban [The internal temporal organization of the sentence in Old Hungarian]. In Nyelvelmélet es diakrónia (ed. K. É. Kiss and A. Hegedűs), pp. 153–70. Szent István Társulat, Budapest– Piliscsaba. Gianollo, Chiara (2010). I verbi deponenti latini e l’unità della flessione in -r. In Incontri Triestini di Filologia Classica VIII (2008–2009), Trieste, pp. 23–49. Edizioni Università di Trieste. Gildea, Daniel and Jurafsky, Dan (1996). Learning bias and phonological rule induction. Computational Linguistics, 22, 497–530. Gildea, Spike (1998). On Reconstructing Grammar: Comparative Cariban Morphosyntax. Oxford University Press, Oxford. Givón, Talmy (1975). Serial verbs and syntactic change: Niger–Congo. In Word Order and Word Order Change (ed. C. Li), pp. 47–112. University of Texas Press, Austin, TX. Givón, Talmy (2009). The Genesis of Syntactic Complexity. John Benjamins, Amsterdam. Glaser, Elvira (2000). Der bestimmte Artikel in den althochdeutschen Glossen. In Zur Geschichte der Nominalgruppe im älteren Deutsch: Festschrift für Paul Valentin. Akten des Pariser Kolloquiums (ed. Y. Desportes), pp. 187–212. Carl Winter, Heidelberg. Gleitman, Lila (1990). The structural sources of verb meanings. Language Acquisition, 1, 3–55. Gleitman, Lila, Cassidy, Kimberly, Papafragou, Anna, Nappa, Rebecca, and Trueswell, John (2005). Hard words. Journal of Language Learning and Development, 1, 23–64. Godefroy, Frédéric (1880). Dictionnaire de l’ancienne langue française et de tous ses dialectes du IXe au XVe siécle. F. Vieweg, Paris. Goldberg, Adele (1995). Constructions: A Construction Grammar Approach to Argument Structure. University of Chicago Press, Chicago, IL.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i



References

Green, Georgia (1976). Main clause phenomena in subordinate clauses. Language, 52, 382–97. Greenberg, Joseph (1963). Some universals of grammar with particular reference to the order of meaningful elements. In Universals of Language (ed. J. Greenberg), pp. 73–113. MIT Press, Cambridge, MA. Grice, Herbert Paul (1975). Logic and conversation. In Syntax and Semantics 3: Speech Acts (ed. P. Cole and J. Morgan), pp. 41–58. Academic Press, New York. Gries, Stefan (2005). Syntactic priming: A corpus-based approach. Journal of Psycholinguistic Research, 34, 365–99. Grimm, Jackob and Grimm, Wilhelm (1854–1961). Deutsches Wörterbuch. Verlag von S. Hirzel, Leipzig. Ed. by E. Wülcker, R. Meiszner, M. Leopold, C. Wesle, and researchers of Deutsches Wörterbuch zu Berlin. Grosse, Julia (2005). Zu Kohärenz und Kontrolle in infiniten Konstruktionen des Deutschen. Tectum Verlag, Marburg. Gruber, Bettina (2008). Complementiser agreement—new evidence from the Upper Austrian variant of Gmunden. Master’s thesis, Universität Wien. Guardiano, Cristina (2011). Parametric changes in the history of the Greek article. In Grammatical Change: Origins, Nature, Outcomes (ed. D. Jonas, J. Whitman, and A. Garrett), pp. 179–97. Oxford University Press, Oxford. Guardiano, Cristina (2013). The Greek definite article across time. Studies in Greek Linguistics, 33, 76–91. Gunkel, Lutz (2000). Selektion verbaler Komplemente. Zur Syntax der Halbmodal- und Phasenverben. In Deutsche Grammatik in Theorie und Praxis (ed. R. Thieroff, M. Tamrat, N. Fuhrhop, and O. Teuber), pp. 111–21. Max Niemeyer, Tübingen. Guo, Jiansheng (1994). Social Interaction, Meaning, and Grammatical Form: Children’s Development and Use of Modal Auxiliaries in Mandarin Chinese. Ph.D. thesis, University of California, Berkeley. Haader, Lea (2004). Nyelvi szempontú megjegyzések a müncheni emlékhez [Linguistic remarks on the Munich relic]. Magyar Nyelv, 101, 161–78. Hacquard, Valentine (2006). Aspects of Modality. Ph.D. thesis, Massachusetts Institute of Technology. Hacquard, Valentine (2010). On the event relativity of modal auxiliaries. Natural Language Semantics, 18, 79–114. Hacquard, Valentine (2013). On the grammatical category of modality. In Proceedings of the 19th Amsterdam Colloquium (ed. M. Aloni, M. Franke, and F. Roelofsen), pp. 19–26. Hacquard, Valentine and Cournane, Ailís (2015). Themes and variations in the expression of modality. Paper presented at the 46th Annual Meeting of the Northeastern Linguistics Society (NELS 46). Concordia University, Montréal, QC. Haeberli, Eric (1999). Features, Categories and the Syntax of A-positions. Ph.D. thesis, Université de Genève. Haeberli, Eric (2000). Adjuncts and the syntax of subjects in Old and Middle English. In Diachronic Syntax: Models and Mechanisms (ed. S. Pintzuk, G. Tsoulas, and A. Warner), pp. 109–131. Oxford University Press, Oxford. Haeberli, Eric (2002). Inflectional morphology and the loss of V2 in English. In Syntactic Effects of Morphological Change (ed. D. Lightfoot), pp. 88–106. Oxford University Press, Oxford.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

References



Haeberli, Eric (2005). Clause type asymmetries in Old English and the syntax of verb movement. In Grammaticalization and Parametric Variation (ed. M. Batllori, M.-L. Hernanz, C. Picallo, and F. Roca), pp. 267–83. Oxford University Press, Oxford. Haegeman, Liliane (1992). Theory and Description in Generative Syntax: A Case Study in West Flemish. Cambridge University Press, Cambridge. Haegeman, Liliane and Danckaert, Lieven (2011). Multiple subjects in Flemish: The external possessor. To appear in Belfast Working Papers in Linguistics, http://www.gist.ugent.be/ file/226. Haegeman, Liliane and van Koppen, Marjo (2012). Complementizer agreement and the relation between C0 and T0 . Linguistic Inquiry, 43, 441–54. Haider, Hubert (2009). The Syntax of German. Cambridge University Press, Cambridge. Hale, Mark (1998). Diachronic syntax. Syntax, 1, 1–18. Halle, Morris (1962). Phonology in generative grammar. Word, 18, 54–72. Halle, Morris (1964). On the bases of phonology. In The Structure of Language (ed. J. Katz and J. Fodor), pp. 324–33. Prentice-Hall, Englewood Cliffs, NJ. Halpern, Aaron and Fontana, Josip (1994). X0 and Xmax clitics. In West Coast Conference in Formal Linguistics 12, pp. 251–66. Hare, Mary and Elman, Jeffrey (1995). Learning and morphological change. Cognition, 56, 61–98. Harris, Alice and Campbell, Lyle (1995). Historical Syntax in Crosslinguistic Perspective. Cambridge University Press, Cambridge. Harris, Martin (1978). The Evolution of French Syntax: A Comparative Approach. Longman, London. Hartman, Jeremy (2011). (Non-)intervention in A-movement: Some cross-constructional and cross-linguistic considerations. Linguistic Variation, 11, 121–48. Haverling, Gerd (2000). On Sco-verbs, Prefixes and Semantic Functions: A Study in the Development of Prefixed and Unprefixed Verbs from Early to Late Latin. Acta Universitatis Gothoburgensis, Göteborg. Haverling, Gerd (2003). On prefixes and actionality in Classical and Late Latin. Acta Linguistica Hungarica, 50, 113–35. Haverling, Gerd (2008). On the development of actionality, tense and viewpoint from Early to Late Latin. In Interdependence of Diachronic and Synchronic Analyses (ed. F. Josephson and I. Söhrman), pp. 73–104. John Benjamins, Amsterdam. Haverling, Gerd (2010). Actionality, tense and viewpoint. In New Perspectives on Historical Latin Syntax 2. Constituent Syntax: Adverbial Phrases, Adverbs, Mood, Tense (ed. P. Baldi and P. Cuzzolin), pp. 277–523. Mouton de Gruyter, Berlin. Hawkins, John (1995). A Performance Theory of Order and Constituency. Cambridge University Press, Cambridge. Hayashi, Chikio (1952). On the prediction of phenomena from qualitative data and the quantification of qualitative data from the mathematico-statistical point of view. Annals of the Institute of Statistical Mathematics, 3, 69–98. Hegedűs, Attila and Papp, Lajos (1991). Középkori leveleink 1541-ig. Tankönyvkiadó, Budapest. Heiden, Serge, Guillot, Céline, Lavrentiev, Alexei, and Bertrand, Lauranne (2010). Base de français médiéval. Lyon: ENS. Heine, Bernd, Claudi, Ulrike, and Hünnemeyer, Frederike (1991). Grammaticalization. A Conceptual Framework. Chicago: University of Chicago Press.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i



References

Heine, Bernd and Kuteva, Tania (2002). World Lexicon of Grammaticalization. Cambridge University Press, Cambridge. Heine, Bernd and Miyashita, Hiroyuki (2004). Drohen und versprechen—zur Genese von funktionalen Kategorien. Neue Beiträge zur Germanistik, 3, 9–33. Heine, Bernd and Miyashita, Hiroyuki (2008). Accounting for a functional category: German drohen ‘threaten’. Language Sciences, 30, 53–101. Heizmann, Tanja (2006). Acquisition of deontic and epistemic readings of must and müssen. In UMOP 34: Current Issues in Language Acquisition (ed. T. Heizmann). GLSA, Amherst, MA. Hendery, Rachel (2012). Relative Clauses in Time and Space. John Benjamins, Amsterdam. Herman, József (2002). La disparition du passif synthétique latin: nouvel essai sur l’écrit et le parlé en latin mérovingien. Estudis romànics, 24, 31–46. Hewson, John (1997). From Latin to Modern Romance. In Tense and Aspect in Indo-European Languages (ed. J. Hewson and V. Bubeník), pp. 314–30. John Benjamins, Amsterdam. Hickmann, Maya and Robert, Stéphane (2006). Introduction. Space, language, and cognition: Some new challenges. In Space in Languages: Linguistic Systems and Cognitive Categories (ed. M. Hickmann and S. Robert), pp. 1–15. John Benjamins, Amsterdam. Hinterhölzl, Roland and Petrova, Svetlana (2009). From V1 to V2 in West Germanic. Lingua, 120, 315–28. Hinterwimmer, Stefan (2015). A unified account of the properties of German demonstrative pronouns. In Special Volume for the Pronoun Workshop at NELS 40 (ed. P. Grosz, P. PatelGrosz, and I. Yanovich), pp. 61–107. GLSA, Amherst, MA. Hirschbühler, Paul (1995). Null subjects in V1 embedded clauses in Philippe de Vigneulles’ Cent nouvelles nouvelles. In Clause Structure and Language Change (ed. A. Battye and I. Roberts), pp. 257–91. Oxford University Press, Oxford. Hirschbühler, Paul and Junker, Marie-Odile (1988). Remarques sur les sujets nuls en subordonnées en ancien et en moyen français. Revue québécoise de linguistique théorique et appliquée, 7, 63–84. Holmberg, Anders (2006). Stylistic fronting. In The Blackwell Companion to Syntax, Volume 4 (ed. M. Everaert and H. van Riemsdijk), pp. 532–65. Blackwell, Oxford. Holmberg, Anders (2015). Verb second. In Syntax—Theory and Analysis. An International Handbook, volume 1 (ed. T. Kiss and A. Alexiadou), pp. 342–83. Mouton De Gruyter, Berlin. Honti, László (2001). A magyar igekötő: nyelvünk kései jóvevénye? [Is the Hungarian verbal particle a latecomer in our language?] In Újabb tanulmányok a strukturális magyar nyelvtan és a nyelvtörténet köréből (ed. M. Bakró Nagy, Z. Bánréti, and K. É. Kiss), pp. 357–67. Osiris, Budapest. Hooper, Joan and Thompson, Sandra (1973). On the applicability of root transformations. Linguistic Inquiry, 4, 465–97. Hopper, Paul and Thompson, Sandra (1980). Transitivity in grammar and discourse. Language, 56, 251–99. Hopper, Paul and Traugott, Elizabeth (1993). Grammaticalization. Cambridge University Press, Cambridge. Horrocks, Geoffrey (2010). Greek: A History of the Language and its Speakers (2nd edn). Blackwell, Oxford. Hrafnbjargarson, Gunnar Hrafn (2004). Stylistic fronting. Studia Linguistica, 58, 88–134.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

References



Hrafnbjargarson, Gunnar Hrafn and Wiklund, Anna-Lena (2009). General embedded V2: Icelandic A, B, C, etc. Working Papers in Scandinavian Syntax, 84, 21–51. Huddleston, Rodney and Pullum, Geoffrey (2002). The Cambridge Grammar of the English Language. Cambridge University Press, Cambridge. Hudson, Richard (2007). Language Networks: Towards a New Word Grammar. Oxford University Press, Oxford. Humbert, Jean (1930). La disparition du datif en grec (du Ier au Xe siècle). Édouard Champion, Paris. Hyde, Brett (2011). Extrametricality and non-finality. In The Blackwell Companion to Phonology (ed. M. van Oostendorp, C. Ewen, E. Hume, and K. Rice), pp. 1027–51. Wiley-Blackwell, Oxford. Iacobini, Claudio (2009). The role of dialects in the emergence of Italian phrasal verbs. Morphology, 19, 15–44. Iacobini, Claudio (2012). Grammaticalization and innovation in the encoding of motion events. Folia Linguistica, 46, 359–85. Iacobini, Claudio and Fagard, Benjamin (2011). A diachronic approach to variation and change in the typology of motion expression. A case study: From Latin to Romance. Cahiers de Faits de Langue, 3, 151–72. Iacobini, Claudio and Masini, Francesca (2006). The emergence of verb–particle constructions in Italian: Locative and actional meanings. Morphology, 16, 155–88. Ingham, Richard (2012). A derivational approach to negative polarity item licensing in Old French. In Research on Old French: The State of the Art (ed. D. Arteaga), pp. 261–81. Springer, Dordrecht. Jackendoff, Ray (1972). Semantics in Generative Grammar. MIT Press, Cambridge, MA. Jean, Latronne Antoine, de Presle, Wladimir Brunet, and Egger, Émile (1866). Notices et textes des papyrus grecs du Musée du Louvre et de la Bibliothèque Impériale. Imprimerie impériale, Paris. Jędrzejowski, Łukasz. (2015). Subjektanhebungsverben im Deutschen. Ihre Entstehung, Entwicklung und Komplemente. Ph.D. thesis, Universität Potsdam. Jespersen, Otto (1917). Negation in English and other Languages. Høst, Copenhagen. Jespersen, Otto (1924). The Philosophy of Grammar. Allen and Unwin, London. Jespersen, Otto (1949). A Modern English Grammar on Historical Principles. Part 7: Syntax. Ejnar Munksgaard, Copenhagen. Johnson, Daniel Ezra (2007). Stability and Change along a Dialect Boundary: The Low Vowels of Southeastern New England. Ph.D. thesis, University of Pennsylvania. Julien, Marit (2007). Embedded V2 in Norwegian and Swedish. Working Papers in Scandinavian Syntax, 80, 103–61. Julien, Marit (2009). Embedded clauses with main clause word order in Mainland Scandinavian. Ms., Lund University, available at http://ling.auf.net/lingBuzz/000475. Karttunen, Lauri (1971). Some observations on factivity. Papers in Linguistics, 4, 55–69. Kaufman, Daniel (2009). Austronesian nominalism and its consequences: A Tagalog case study. Theoretical Linguistics, 35, 1–49. Kayne, Richard (1994). The Antisymmetry of Syntax. MIT Press, Cambridge, MA. Keenan, Edward and Comrie, Bernard (1977). Noun phrase accessibility and Universal Grammar. Linguistic Inquiry, 8, 63–99.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i



References

Keenan, Edward and Hawkins, Sarah (1987). The psychological validity of the accessibility hierarchy. In Universal Grammar: 15 Essays (ed. E. Keenan), pp. 60–85. Croon Helm, London. Kermack, William Ogilvy and McKendrick, Anderson Gray (1927). A contribution to the mathematical theory of epidemics. Proceedings of the Royal Society of London, Series A, 115, 700–21. Kerswill, Paul (1996). Children, adolescents, and language change. Language Variation and Change, 8, 177–202. Keyser, Samuel Jay (1975). A partial history of the relative clause in English. In University of Massachusetts Occasional Papers in Linguistics vol.1: Papers in the History and Structure of English (ed. J. Grimshaw), pp. 1–33. University of Massachusetts, Amherst, MA. Kiefer, Ferenc (2010). Areal–typological aspects of word-formation. In Variation and Change in Morphology (ed. F. Rainer, W. Dressler, D. Kastovsky, and H. C. Luschützky), pp. 129–48. John Benjamins, Amsterdam. King, Jeffrey (2001). Complex Demonstratives: A Quantificational Account. MIT Press, Cambridge. Kiparsky, Paul (1974). Remarks on analogical change. In Proceedings of the First International Congress of Historical Linguistics (ed. C. Jones and J. Anderson), pp. 257–76. North Holland, Amsterdam. Kiparsky, Paul (1998). Partitive case and aspect. In The Projection of Arguments (ed. M. Butt and W. Geuder), pp. 265–307. CSLI, Stanford, CA. Kirby, Simon (1996). Function, Selection and Innateness: The Emergence of Language Universals. Ph.D. thesis, University of Edinburgh. Kiss, Tibor (2005). Subjektselektion bei Infinitiven. In Der Infinitiv im Deutschen (ed. J.-F. Marillier and C. Rozier), pp. 115–32. Stauffenburg, Tübingen. Klein, Katarina (2009). Semi-modal variation. In Describing and Modeling Variation in Grammar (ed. A. Dufter, J. Fleischer, and G. Seiler), pp. 297–324. De Gruyter, Berlin. Klein, Wolfgang and von Sutterheim, Christine (1987). Quaestio und referentielle Bewegung in Erzählungen. Linguistische Berichte, 109, 163–83. Koopman, Hilda (2000). Prepositions, postpositions, circumpositions, and particles: The structure of Dutch PPs. In The Syntax of Specifiers and Heads: Collected Essays of Hilda J. Koopman (ed. H. Koopman), pp. 204–60. Routledge, London. Koopman, Willem (1995). Verb-final main clauses in Old English prose. Studia Neophilologica, 67, 129–44. Kopecka, Anetta (2006). The semantic structure of motion verbs in French: Typological perspectives. In Space in Languages: Linguistic Systems and Cognitive Categories (ed. M. Hickmann and S. Robert), pp. 83–101. John Benjamins, Amsterdam. Kopecka, Anetta (2009). Continuity and change in the representation of motion events in French. In Crosslinguistic Approaches to the Psychology of Language (ed. J. Guo, E. Lieven, N. Budwig, S. Ervin-Tripp, K. Nakamura, and S. Özçaliskan), pp. 415–26. Psychology Press, New York. Kopecka, Anetta (in press). From a satellite- to a verb-framed pattern: A typological shift in French. In Variation and Change in Adpositions of Movement (ed. H. Cuyckens, W. De Mulder, and T. Mortelmans). John Benjamins, Amsterdam. Koster, Jan (1994). Predicate incorporation and the word order of Dutch. In Paths towards Universal Grammar: Studies in Honor of Richard S. Kayne (ed. G. Cinque, J. Koster, J.-Y. Pollock, L. Rizzi, and R. Zanuttini), pp. 255–76. Oxford University Press, Oxford.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

References



Kratzer, Angelika (1977). What ‘must’ and ‘can’ must and can mean. Linguistics and Philosophy, 1, 337–55. Kratzer, Angelika (1981). The notional category of modality. In Words, Worlds, and Contexts (ed. H. Elkmeyer and H. Rieser), pp. 38–74. De Gruyter, Berlin. Kratzer, Angelika (2012). Modals and Conditionals: New and Revised Perspectives. Oxford University Press, Oxford. Kroch, Anthony (1989a). Function and grammar in the history of English: Periphrastic do. In Language Change and Variation (ed. R. Fasold and D. Schiffrin), pp. 133–72. John Benjamins, Amsterdam. Kroch, Anthony (1989b). Reflexes of grammar in patterns of language change. Language Variation and Change, 1, 199–244. Kroch, Anthony (1994). Morphosyntactic variation. In CLS 30: Papers from the 30th Regional Meeting of the Chicago Linguistic Society, pp. 180–201. Kroch, Anthony and Santorini, Beatrice (2012). Penn supplement to the MCVF corpus of historical French. Department of Linguistics, University of Pennsylvania. http://www. ling.upenn.edu/hist-corpora. Kroch, Anthony, Santorini, Beatrice, and Delfs, Lauren (2004). The Penn-Helsinki Parsed Corpus of Early Modern English (PPCEME). Department of Linguistics, University of Pennsylvania. http://www.ling.upenn.edu/hist-corpora/PPCEME-RELEASE-1/. Kroch, Anthony, Santorini, Beatrice, and Diertani, Ariel (2010). The Penn–Helsinki Parsed Corpus of Modern British English (PPCMBE). Department of Linguistics, University of Pennsylvania. http://www.ling.upenn.edu/hist-corpora/PPCMBE-RELEASE-1/. Kroch, Anthony and Taylor, Ann (2000). The Penn–Helsinki Parsed Corpus of Middle English, 2nd edition (PPCME2). Department of Linguistics, University of Pennsylvania. http://www.ling.upenn.edu/hist-corpora/PPCME2-RELEASE-2/. Kuczaj, Stan (1977). Old and new forms, old and new meanings: The form–function hypothesis revisited. Paper presented at the meeting of the Society for Research on Child Development, New Orleans. Kühner, Raphael and Gerth, Bernhard (1898). Ausführliche Grammatik der griechischen Sprache. Zweiter Teil: Satzlehre. Hahn, Hannover. Kühner, Raphael and Stegmann, Karl (1966). Ausführliche Grammatik der lateinischen Sprache. Hahn, Hannover. Labelle, Marie (2007). Clausal architecture in early Old French. Lingua, 117, 289–316. Labelle, Marie (2013). Participle fronting and clause structure in Old and Middle French. Presentation given at LSRL 43, CUNY Graduate Center, New York, 17–19 April, 2013. Labelle, Marie and Hirschbühler, Paul (2005). Changes in clausal organization and the position of clitics in Old French. In Grammaticalization and Parametric Change (ed. M. Batllori and F. Roca), pp. 60–71. Oxford University Press, Oxford. Labov, William (2001). Principles of Linguistic Change, Volume 2: Social Factors. Blackwell, Oxford. Labov, William (2007). Transmission and diffusion. Language, 83, 344–87. Labov, William (2012). What is to be learned: The community as the focus of social cognition. Review of Cognitive Linguistics, 10, 265–93. Lavidas, Nikolaos (2009). Transitivity Alternations in Diachrony: Changes in Argument Structure and Voice Morphology. Cambridge Scholars Publishing, Newcastle upon Tyne.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i



References

Lavidas, Nikolaos (2016). Development of transitivity in a language without/with object clitics: English vs. Greek (a diachronic contrastive study). In Selected Papers from the 21st International Symposium on Theoretical and Applied Linguistics (ed. M. Mattheoudakis and K. Nicolaidis), pp. 186–210. Aristotle University of Thessaloniki. Lavidas, Nikolaos (to appear). Non-canonical case marking, Voice, and Case theory: Evidence from the diachrony of Greek. In Diachronic Typology of Voice and Valency-changing Categories (ed. S. Kittilä and L. Kulikov). John Benjamins, Amsterdam. Lavidas, Nikolaos and Papangeli, Dimitra (2007). Deponency in the diachrony of Greek. In Deponency and Morphological Mismatches (ed. M. Baerman, G. Corbett, D. Brown, and A. Hippisley), pp. 97–126. Oxford University Press, Oxford. LeCoq, Albert (2011). Chuastuanift, ein Sündenbekenntnis der manichäischen Auditores gefunden in Turfan. De Gruyter, Berlin. Ledgeway, Adam (1997–9). I tempi sovraccomposti nel napoletano antico. L’Italia dialettale, 60, 105–24. Ledgeway, Adam (2012). From Latin to Romance: Morphosyntactic Typology and Change. Oxford University Press, Oxford. Legendre, Géraldine and Sorace, Antonella (2003). Auxiliaires et intransitivité en français et dans les langues romanes. In Les langues romanes. Problèmes de la phrase simple (ed. D. Godard), pp. 185–233. CNRS éditions, Paris. Leiss, Elisabeth (1994). Die Entstehung des Artikels im Deutschen. Sprachwissenschaft, 19, 307–19. Leiss, Elisabeth (2000). Artikel und Aspekt. Die grammatischen Muster von Definitheit. De Gruyter, Berlin. Lema, José and Rivero, María-Luisa (1989). Types of verbal movement in Old Spanish: Modals, futures and perfects. Probus, 3, 237–78. Leumann, Manu (1921). Part. perf. pass. mit fui im späteren Latein. Glotta, 11, 192–4. Levey, Stephen (2006). The sociolinguistic distribution of discourse marker LIKE in preadolescent speech. Multilingua, 25, 413–41. Li, Zhuanxu and Yang, Yingxin (1999). Baizu wenhua daguan [Overview of Bai culture]. Yunnan Minzu Chubanshe, Kunming. Light, Caitlin (2011). Parsed corpus of Early New High German. http://enhgcorpus.wikispaces. com/. Light, Caitlin (2012). The Syntax and Pragmatics of Fronting in Germanic. Ph.D. thesis, University of Pennsylvania. Lightfoot, David (1979). Principles of Diachronic Syntax. Cambridge University Press, Cambridge. Lightfoot, David (1999). The Development of Language: Acquisition, Change and Evolution. Blackwell, Oxford. Lightfoot, David (2002). Myths and the prehistory of grammars. Journal of Linguistics, 38, 113–36. Lightfoot, David (2006). How New Languages Emerge. Cambridge University Press, Cambridge. Longobardi, Giuseppe (2001). Formal syntax, diachronic minimalism, and etymology: The history of French chez. Linguistic Inquiry, 32, 275–302. Lord, Carol (1973). Serial verbs in transition. Studies in African Linguistics, 4, 269–96. Lubańska, Maja (2010). Grammaticalisation and generative theory: The case of epistemic promise. Belgrade English Language & Literature Studies, 2, 93–111.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

References



Luraghi, Silvia (2003). On the Meaning of Prepositions and Cases. John Benjamins, Amsterdam. Luraghi, Silvia (2009). The internal structure of prepositional phrases and the notion of government. In Form and Function in Language Research: Papers in Honour of Christian Lehmann (ed. J. Helmbrecht, Y. Nishina, Y.-M. Shin, S. Skopeteas, and E. Verhoeven), pp. 231–54. Mouton de Gruyter, Berlin. Luraghi, Silvia (2010a). Adverbials. In A New Historical Syntax of Latin (ed. P. Baldi and P. Cuzzolin), pp. 1–75. Mouton de Gruyter, Berlin. Luraghi, Silvia (2010b). The extension of the transitive construction in Ancient Greek. Acta Linguistica Hafniensia: International Journal of Linguistics, 42, 60–74. McFadden, Thomas (2002). The rise of the to-dative in Middle English. In Syntactic Effects of Morphological Change (ed. D. Lightfoot), pp. 107–23. Oxford University Press, Oxford. Maché, Jakob (2013). On Black Magic: How Epistemic Modifiers Emerge. Ph.D. thesis, Freie Universität zu Berlin. Maling, Joan (1980). Inversion in embedded clauses in Modern Icelandic. Íslenskt mál og almenn málfræi, 2, 175–93. Reprinted in Modern Icelandic Syntax (ed. J. Maling and A. Zaenen), pp. 71–91. Academic Press, San Diego. Malthus, Thomas (1826). An Essay on the Principles of Population, or a View of its Past and Present Effects on Human Happiness; with an Inquiry into our Prospects Respecting the Future Removal or Mitigation of the Evils which it Occasions (6th edn). John Murray, London. Manolessou, Io (2000). Greek Noun Phrase Structure: A Study in Syntactic Evolution. Ph.D. thesis, University of Cambridge. Manolessou, Io (2001). The evolution of the demonstrative system in Greek. Journal of Greek Linguistics, 2, 119–48. Marantz, Alec (1997). No escape from syntax: Don’t try morphological analysis in the privacy of your own lexicon. In University of Pennsylvania Working Papers in Linguistics (ed. A. Dimitriadis, L. Siegel, C. Surek-Clark, and A. Williams), pp. 201–25. Marchello-Nizia, Cristiane (2002). Prépositions françaises en diachronie: une catégorie en question. Lingvisticae Investigationes, 25, 205–21. Markman, Ellen and Wachtel, Gwyn (1988). Children’s use of mutual exclusivity to constrain the meanings of words. Cognitive Psychology, 20, 121–57. Martin, Robert (2001). Le préfixe a-/ad- en moyen français. Romania, 119, 289–322. Martineau, France, Hirschbühler, Paul, Kroch, Anthony, and Morin, Yves Charles (2010). Corpus MCVF annoté syntaxiquement. Département de français, Université d’Ottawa. http://continent.uottawa.ca/fr/corpus/corpusmcvf/structure/. Martins, Ana Maria (2005). Clitic placement, VP-ellipsis, and scrambling in Romance. In Grammaticalization and Parametric Variation (ed. M. Batllori, M.-L. Hernanz, C. Picallo, and F. Roca), pp. 175–93. Oxford University Press, Oxford. Martins, Ana Maria (2011). Scrambling and information focus in Old and Contemporary Portuguese. Catalan Journal of Linguistics, 10, 133–58. Mathieu, Éric (2006a). Quirky subjects in Old French. Studia Linguistica, 60, 282–312. Mathieu, Éric (2006b). Stylistic fronting in Old French. Probus, 18, 219–66. Mathieu, Éric (2009). On the Germanic properties of Old French. In Historical Syntax and Linguistic Theory (ed. P. Crisma and G. Longobardi), pp. 344–57. Oxford University Press, Oxford.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i



References

Mathieu, Éric (2013). The left periphery in Old French. In Research in Old French: The State of the Art (ed. D. Arteaga), pp. 327–50. Springer, Dordrecht. Matisoff, James (1972). Lahu nominalization, relativization, and genitivization. In Syntax and Semantics, Volume 1 (ed. J. Kimball), pp. 237–57. Seminar Press, New York. Maxwell, Daniel (1979). Strategies of relativization and NP accessibility. Language, 55, 352–71. Maxwell, Dan (1982). Implications of NP accessibility for diachronic syntax. Folia Linguistica Historica, 3, 135–52. Meillet, Antoine (1908). Introduction à l’étude comparative des langues Indo-Européennes. Hachette, Paris. Meillet, Antoine (1912). L’ évolution des formes grammaticales. In Linguistique Historique et Linguistique Générale (ed. A. Meillet), pp. 130–58. Honoré Champion, Paris. Meisel, Jurgen, Elsig, Martin, and Rinke, Esther (2013). Language Acquisition and Change: A Morphosyntactic Perspective. Edinburgh University Press, Edinburgh. Meurers, Detmar (2000). Lexical Generalizations in the Syntax of German Non-finite Constructions. Ph.D. thesis, Universität Tübingen. Michelioudakis, Dimitris and Sitaridou, Ioanna (2012). Syntactic microvariation: Dative constructions in Greek. In Variation in Datives: A Micro-Comparative Perspective (ed. B. Fernández and R. Etxepare), pp. 212–55. Oxford University Press, Oxford. Mikkelsen, Line (2010). On what comes first in a verb-second language. Ms., University of California, Berkeley. Mitchell, Bruce (1985). Old English Grammar, Volumes 1–2. Clarendon Press, Oxford. Miyagawa, Shigeru (2009). Why Agree? Why Move? MIT Press, Cambridge, MA. Mohr, Sabine (2009). V2 as a single-edge phenomenon. In Selected Papers from the 2006 Cyprus Syntaxfest (ed. K. Grohmann and P. Panagiotidis), pp. 141–58. Cambridge Scholars Publishing, Newcastle upon Tyne. Molnár, Valeria (2010). Stylistic fronting and discourse. Tampa Papers in Linguistics, 1, 30–61. Moore, Chris, Pure, Kiran, and Furrow, David (1990). Children’s understanding of the modal expression of speaker certainty and uncertainty and its relation to the development of a representational theory of mind. Child Development, 61, 722–30. Moser, Amalia (2008). The changing relationship of tense and aspect in Greek. Sprachtypologie und Universalienforschung / Language Typology and Universals (STUF), 61, 5–18. Moser, Amalia (2009). Apopsi ke hronos stin istoria tis Elinikis [Aspect and Tense in the History of Greek]. University of Athens. Moser, Amalia (2014). From Aktionsart to aspect: Grammaticalization and subjectification in Greek. Acta Linguistica Hafniensia: International Journal of Linguistics, 46, 64–84. Müller, Gereon (2004). Verb-second as vP-first. Journal of Comparative Germanic Linguistics, 7, 179–234. Muller, Henri François (1924). The passive voice in Vulgar Latin. The Romanic Review, 15, 68–93. Mustanoja, Tauno (1960). A Middle English Syntax. Part 1: Parts of Speech. Société néophilologique, Helsinki. Næss, Åshild (2004). What markedness marks: The markedness problem with direct objects. Lingua, 114, 1186–212.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

References



Napoli, Maria (2006). Aspect and Actionality in Homeric Greek: A Contrastive Analysis. Francoangeli, Milano. Napoli, Maria (2009). Aspects of definiteness in Greek. Studies in Language, 33, 569–611. Naro, Anthony and Lemle, Miriam (1976). Syntactic diffusion. In Papers from the Parasession on Diachronic Syntax (ed. S. Steever, C. Walker, and S. Mufwene), pp. 221–39. Chicago Linguistic Society, Chicago. Nespor, Marina and Vogel, Irene (2007). Prosodic Phonology (with a New Foreword). Mouton de Gruyter, Berlin. Newmeyer, Frederick (2005). Possible and Probable Languages. Oxford University Press, Oxford. Nikolaeva, Irina (1999). Ostyak. LINCOM Europas, München. Obata, Miki and Epstein, Samuel David (2011). Feature-splitting internal Merge: Improper movement, intervention, and the A/A distinction. Syntax, 14, 122–47. O’Neill, Daniela and Atance, Cristina (2000). ‘Maybe my daddy give me a big piano’: The development of children’s use of modals to express uncertainty. First Language, 20, 29–52. Osgood, Charles and Sebeok, Thomas (1954). Psycholinguistics: A Survey of Theory and Research Problems. Waverly Press, Baltimore, MD. Ota, Tatsuo (1958). Chugokugo rekishi bunpo [An Historical Grammar of Chinese]. Konan Shoin, Tokyo. Ott, Dennis (2009). Stylistic fronting as remnant movement. Working Papers in Scandinavian Syntax, 83, 141–78. Paoli, Sandra (2007). The fine structure of the left periphery: COMPs and subjects. Evidence from Romance. Lingua, 117, 1057–79. Paolillo, John (2011). Independence claims in linguistics. Language Variation and Change, 23, 257–74. Papafragou, Anna (1998). The acquisition of modality: Implications for theories of semantic representation. Mind & Language, 13, 370–99. Paul, Hermann (1920). Prinzipien der Sprachgeschichte. Max Niemeyer, Halle. Paul, Waltraud (2015). New Perspectives on Chinese Linguistics. De Gruyter, Berlin. Pérez-Leroux, Ana Teresa, Castilla-Earls, Anny, Bejar, Susana, and Massam, Diane (2012). Elmo’s sister’s ball: The development of nominal recursion in children. Language Acquisition, 19, 301–11. Perkins, Michael (1983). Modal Expressions in English. Frances Pinter, London. Pesetsky, David and Torrego, Esther (2004). Tense, case and the nature of syntactic categories. In The Syntax of Time (ed. J. Guéron and J. Lecarme), pp. 495–538. MIT Press, Cambridge, MA. Pezzini, Giuseppe (2011). Contraction of est in Latin. Transactions of the Philological Society, 109, 327–43. Philippi, Julia (1997). The rise of the article in Germanic languages. In Parameters of Morphosyntactic Change (ed. A. van Kemenade and N. Vincent), pp. 62–93. Cambridge University Press, Cambridge. Pickering, Martin and Branigan, Holly (1999). Syntactic priming in language production. Trends in Cognitive Science, 3, 136–41. Pickering, Martin and Ferreira, Victor (2008). Structural priming: A critical review. Psycholinguistic Bulletin, 124, 427–59. Pinkster, Harm (1985). The development of future tense auxiliaries in Latin. Glotta, 63, 186–208.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i



References

Pintzuk, Susan (1991). Phrase Structures in Competition: Variation and Change in Old English Word Order. Ph.D. thesis, University of Pennsylvania. Published as Pintzuk (1999). Pintzuk, Susan (1993). Verb seconding in Old English: Verb movement to Infl. The Linguistic Review, 10, 5–35. Pintzuk, Susan (1999). Phrase Structures in Competition: Variation and Change in Old English Word Order. Garland, New York. Pintzuk, Susan and Haeberli, Eric (2008). Structural variation in Old English root clauses. Language Variation and Change, 20, 367–407. Piqueira, José Roberto (2010). Rumor propagation model: An equilibrium study. Mathematical Problems in Engineering. Article ID 631357. Platzack, Christer (1987). The Scandinavian languages and the null-subject parameter. Natural Language & Linguistic Theory, 5, 377–401. Platzack, Christer and Rosengren, Inger (1998). On the subject of imperatives: A minimalist account of the imperative clause. Journal of Comparative Germanic Linguistics, 1, 177–224. Poletto, Cecilia (2002). The left periphery of V2 Rhaetoromance dialects: A new view on V2 and V3. In Syntactic Microvariation (ed. S. Barbiers, L. Cornips, and S. van der Kleij), pp. 214–42. Meertens Institute Electronic Publications in Linguistics, Amsterdam. Poletto, Cecilia (2005). Diachronic variation in Romance. Handout of a presentation at EGG, Wrocław. Available at http://egg.auf.net/05/docs/handouts/Cecilia/eggdiachronicpoletto.pdf. Poletto, Cecilia (2006). Parallel phases: A study on the high and low left periphery of Old Italian. In Phases of Interpretation (ed. M. Frascarelli), pp. 261–95. Mouton de Gruyter, Berlin. Polinsky, Maria (2013). Raising and control. In The Cambridge Handbook of Generative Syntax (ed. M. den Dikken), pp. 577–606. Cambridge University Press, Cambridge. Pollock, Jean-Yves (1989). Verb movement, Universal Grammar, and the structure of IP. Linguistic Inquiry, 20, 365–424. Polo, Chiara (2002). Double objects and morphological triggers for syntactic case. In Syntactic Effects of Morphological Change (ed. D. Lightfoot), pp. 124–42. Oxford University Press, Oxford. Poplack, Shana (1980). Deletion and disambiguation in Puerto Rican Spanish. Language, 56, 371–85. Poplack, Shana (1984). Variable concord and sentential plural marking in Puerto Rican Spanish. Hispanic Review, 52, 205–22. Poplack, Shana and Tagliamonte, Sali (1989). There’s no tense like the present: Verbal -s inflection in early Black English. Language Variation and Change, 1, 47–84. Postma, Gertjan (2010). The impact of failed changes. In Continuity and Change in Grammar (ed. A. Breitbarth, C. Lucas, S. Watts, and D. Willis), pp. 269–302. John Benjamins, Amsterdam. Postma, Gertjan (2011). Language contact and linguistic complexity—The rise of the reflexive pronoun zich in a 15th century Netherlands’ border dialect. In Grammatical Change: Origins, Nature, Outcomes (ed. D. Jonas, J. Whitman, and A. Garrett), pp. 139–59. Oxford University Press, Oxford. Pourcel, Stéphanie and Kopecka, Anetta (2005). Motion expression in French: Typological intricacies. Durham & Newcastle Working Papers in Linguistics, 11, 139–53.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

References



Prince, Ellen (1999). How not to mark topics: ‘Topicalization’ in English and Yiddish. Ms., University of Pennsylvania. Radwan, Sonja (2011). L’ordre des mots en ancien français: La syntaxe et la structure informationnelle de la zone préverbale. Master’s thesis, Universitetet i Oslo. Rappaport Hovav, Malka and Levin, Beth (2008). The English dative alternation: The case for verb sensitivity. Journal of Linguistics, 44, 129–67. Reinhart, Tanya (1981). A second COMP position. In Theory of Markedness in Generative Grammar (ed. A. Belletti, L. Brandi, and L. Rizzi), pp. 517–57. Scuola Normale Superiore di Pisa. Reis, Marga (2005). Zur Grammatik der sog. ‘Halbmodale’ drohen / versprechen + Infinitiv. In Deutsche Syntax: Empirie und Theorie. Symposium in Göteborg 13.–15. Mai 2004 (ed. F. J. d’Avis), pp. 125–45. Acta Universitatis Gothoburgensis, Göteborg. Reis, Marga (2007). Modals, so-called semi-modals, and grammaticalization in German. Interdisciplinary Journal for Germanic Linguistics and Semiotic Analysis, 12, 1–57. Ribeiro, Ilza and Torres Morais, Maria (2012). Doubling -que embedded constructions in Old Portuguese: A diachronic perspective. In Parameter Theory and Linguistic Change (ed. C. Galves, S. Cyrino, R. Lopes, F. Sandalo, and J. Avelar), pp. 97–116. Oxford University Press, Oxford. Richter, Frank and Sailer, Manfred (1998). Complementizers and finite verbs in German sentence structure. In Proceedings of the ESSLLI-98 Workshop on Current Topics in Constraintbased Theories of Germanic Syntax (ed. T. Kiss and D. Meurers), pp. 133–48. Universität des Saarlandes, Saarbrücken. Rickard, Peter (1962). The word-order object–verb–subject in Medieval French. Transactions of the Philological Society, 61, 1–39. Rinke, Esther and Meisel, Jürgen (2009). Subject inversion in Old French: Syntax and information structure. In Proceedings of the Workshop ‘Null-subjects, expletives, and locatives in Romance’ (ed. G. Kaiser and E.-M. Remberger), pp. 93–130. Fachbereich Sprachwissenschaft, Konstanz. Ritter, Elizabeth and Rosen, Sara Thomas (1998). Delimiting events in syntax. In The Projection of Arguments (ed. W. Geuder and M. Butt), pp. 135–64. CSLI, Stanford, CA. Rivero, María-Luisa (1992). Clitic and NP climbing in Old Spanish. In Current Studies in Spanish Linguistics (ed. H. Campos and F. Martínez-Gil), pp. 241–82. Georgetown University Press, Washington, DC. Rizzi, Luigi (1990). Relativized Minimality. MIT Press, Cambridge, MA. Rizzi, Luigi (1997). The fine structure of the left periphery. In Elements of Grammar: A Handbook of Generative Syntax (ed. L. Haegeman), pp. 281–337. Kluwer, Dordrecht. Rizzi, Luigi (2001). On the position INT(ERROGATIVE) in the left periphery of the clause. In Current Studies in Italian Syntax (ed. G. Cinque and G. Salvi), pp. 287–96. Elsevier, Amsterdam. Rizzi, Luigi (2006). On the form of chains: Criterial positions and ECP effects. In WHMovement: Moving on (ed. L. Cheng and N. Corver), pp. 97–134. MIT Press, Cambridge, MA. Rizzi, Luigi and Shlonsky, Ur (2007). Strategies of subject extraction. In Interfaces + Recursion = Language? Chomsky’s Minimalism and the View from Syntax–Semantics (ed. U. Sauerland and H.-M. Gärtner), pp. 115–60. De Gruyter, Berlin. Roberts, Craige (1996). Information structure in discourse: Towards an integrated formal theory of pragmatics. In OSU Working Papers in Linguistics 49: Papers in Semantics (ed. J.-H. Yoon

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i



References

and A. Kathol), pp. 91–136. Revised version published (2012) in Semantics and Pragmatics 5, article 6, pp. 1–69. Page numbers refer to the 2012 version. Roberts, Ian (1985). Agreement parameters and the development of English modal auxiliaries. Natural Language & Linguistic Theory, 3, 21–58. Roberts, Ian (1992). A formal account of grammaticalisation in the history of Romance futures. Folia Linguistica Historica, 13, 219–58. Roberts, Ian (1993). Verbs and Diachronic Syntax: A Comparative History of English and French. Kluwer, Dordrecht. Roberts, Ian (1996). Remarks on the Old English C-system and the diachrony of V2. In Language Change and Generative Grammar (ed. E. Brandner and G. Ferraresi), pp. 154–64. Westdeutscher Verlag, Opladen. Roberts, Ian (2004). The C-system in Brythonic Celtic languages, V2, and the EPP. In The Cartography of Syntactic Structures, Volume 2: The Structure of CP and IP (ed. L. Rizzi), pp. 297–328. Oxford University Press, Oxford. Roberts, Ian (2010a). Agreement and Head Movement: Clitics, Incorporation, and Defective Goals. MIT Press, Cambridge, MA. Roberts, Ian (2010b). Grammaticalization, the clausal hierarchy and semantic bleaching. In Gradience, Gradualness and Grammaticalization (ed. E. C. Traugott and G. Trousdale), pp. 45–73. John Benjamins, Amsterdam. Roberts, Ian and Roussou, Anna (1999). A formal approach to ‘grammaticalization’. Linguistics, 37, 1011–41. Roberts, Ian and Roussou, Anna (2003). Syntactic Change: A Minimalist Approach to Grammaticalization. Cambridge University Press, Cambridge. Rögnvaldsson, Eiríkur and Thráinsson, Höskuldur (1990). On Icelandic word order once more. In Modern Icelandic Syntax (ed. J. Maling and A. Zaenen), pp. 3–40. Academic Press, San Diego. Rohlfs, Gerhard (1922). Das romanische habeo-Futurum und Konditionalis. Archivum Romanicum, 6, 105–52. Romaine, Suzanne (1980). The relative clause marker in Scots English: Diffusion, complexity, and style as dimensions of syntactic change. Language in Society, 9, 221–47. Romaine, Suzanne (1982). Socio-historical Linguistics: Its Status and Methodology. Cambridge University Press, Cambridge. Romaine, Suzanne (1984). Towards a typology of relative clause formation in Germanic. In Historical Syntax (ed. J. Fisiak), pp. 437–70. De Gruyter, Berlin. Rooryck, Johan (2001). Evidentiality. Glot International, 5, 125–33, 161–8. Rosen, Haiim (1992). Die Komposita mit co(n)- in funktioneller und vergleichender Sicht. In Latein und Indogermanisch (ed. O. Panagl and T. Krisch), pp. 357–67. Institut für Sprachwissenschaft, Innsbruck. Rosengren, Inger (1992). Zum Problem der kohärenten Verben im Deutschen. In Biologische und soziale Grundlagen der Sprache. Interdisziplinäres Symposium des Wissenschaftsbereiches Germanistik der Friedrich-Schiller-Universität Jena, 17.–19. Oktober 1989 (ed. Peter Suchsland), pp. 265–97. Max Niemeyer, Tübingen. Rouveret, Alain (2004). Les clitiques pronominaux et la périphérie gauche en ancien français. Bulletin de la Société de linguistique de Paris, XCIX, 181–237. Russell-Gebbett, Paul (1965). Mediaeval Catalan Linguistic Texts. Dolphin Book Company, Oxford.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

References



Sablayrolles, Jean-François (2000). La néologie en français contemporain. Examen du concept et analyse de productions néologiques récentes. Honoré Champion, Paris. Sag, Ivan (2012). Sign-Based Construction Grammar: An informal synopsis. In Sign-Based Construction Grammar (ed. H. Boas and I. Sag), pp. 69–202. CSLI, Stanford, CA. Salvesen, Christine Meklenborg (2011). Stylistic fronting and remnant movement in Old French. In Romance Languages and Linguistic Theory: Selected Papers from ‘Going Romance’ Nice 2009 (ed. J. Bern, H. Jacobs, and T. Scheer), pp. 323–42. John Benjamins, Amsterdam. Salvesen, Christine Meklenborg (2013). Topics and the left periphery: A comparison of Old French and Modern Germanic. In In Search of Universal Grammar. From Old Norse to Zoque (ed. T. Lohndal), pp. 131–71. John Benjamins, Amsterdam. Salvesen, Christine Meklenborg (2014a). Le complémenteur que et la périphérie gauche. Analyse diachronique. Syntaxe et Sémantique, 15, 47–80. Salvesen, Christine Meklenborg (2014b). Licensing pro in Old French. Ms., Universitetet i Oslo. Salvesen, Christine Meklenborg and Bech, Kristin (2014). Postverbal subjects in Old English and Old French. Oslo Studies in Language, 6, 201–27. Sándor, Klára (2011). Nyelvrokonság és hunhagyomány. Typotex, Budapest. Sankoff, David and Laberge, Suzanne (1978). Statistical dependence among successive occurrences of a variable in discourse. In Linguistic Variation: Models and Methods (ed. D. Sankoff), pp. 119–26. Academic Press, New York. Sankoff, Gillian and Blondeau, Hélène (2008). Language change across the lifespan: /r/ in Montreal French. Language, 83, 560–88. Santorini, Beatrice (1994). Some similarities and differences between Icelandic and Yiddish. In Verb movement (ed. D. Lightfoot and N. Hornstein), pp. 87–106. Cambridge University Press, Cambridge. Sanz, Montserrat (2000). Events and Predication: A New Approach to Syntactic Processing in English and Spanish. John Benjamins, Amsterdam. Sárosi, Zsófia (2003). Morfématörténet. Az ősmagyar kor [The history of morphology: The Proto-Hungarian period]. In Magyar nyelvtörténet (ed. J. Kiss and F. Pusztai), pp. 129–72. Osiris, Budapest. Schmidtke-Bode, Karsten (2009). Going-to-V and gonna-V in child language: A quantitative approach to constructional development. Cognitive Linguistics, 20, 509–38. Schøsler, Lene and van Reenen, Pieter (2000). The pragmatic functions of the Old French particles AINZ, APRES, DONC, LORS, OR, PUIS, and SI. In Textual Parameters in Older Languages (ed. L. Schøsler, S. Herring, and P. van Reenen), pp. 59–105. John Benjamins, Amsterdam. Schwartz, Bonnie and Vikner, Sten (1989). All verb second clauses are CPs. Working Papers in Scandinavian Syntax, 43, 27–49. Schwartz, Bonnie and Vikner, Sten (1996). The verb always leaves IP in V2 clauses. In Parameters and Functional Heads (ed. A. Belletti and L. Rizzi), pp. 11–63. Oxford University Press, Oxford. Schwarz, Florian (2015). Topics & situations—Comments on Hinterwimmer. In Special Volume for the Pronoun Workshop at NELS 40 (ed. P. Grosz, P. Patel-Grosz, and I. Yanovich), pp. 109– 26. GLSA, Amherst, MA. Scott, John (1929). Plural verbs with neuter plural subjects in Homer. American Journal of Philology, 50, 71–6.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i



References

Selkirk, Elizabeth (1980). Prosodic domains in phonology: Sanskrit revisited. In Juncture: a Collection of Original Papers (ed. M. Aronoff and M.-L. Kean), pp. 107–29. Anma Libri, Saratoga, CA. Selkirk, Elizabeth (1984). Phonology and Syntax: The Relation between Sound and Structure. MIT Press, Cambridge, MA. Sells, Peter (2001). Structure, Alignment and Optimality in Swedish. CSLI Publications, Stanford, CA. Senghas, Ann and Coppola, Marie (2001). Children creating language: How Nicaraguan Sign Language acquired a spatial grammar. Psychological Science, 12, 323–8. Shepherd, Susan (1982). From deontic to epistemic: An analysis of modals in the history of English, creoles and language acquisition. In Papers from the 5th International Conference on Historical Linguistics (ed. A. Ahlqvist), pp. 316–23. John Benjamins, Amsterdam. Shlonsky, Ur (1994). Agreement in Comp. The Linguistic Review, 11, 351–75. Sigurðsson, Halldór Ármann (2004). Icelandic non-nominative subjects: Facts and implications. In Non-nominative Subjects, Volume 2 (ed. P. Bhaskararao and K. V. Subbarao), pp. 137–59. John Benjamins, Amsterdam. Sigurðsson, Halldór Ármann (2010). On EPP effects. Studia Linguistica, 64, 159–89. Sihler, Andrew (1995). New Comparative Grammar of Greek and Latin. Oxford University Press, Oxford. Slobin, Dan (1985). Crosslinguistic evidence for the language-making capacity. In The Crosslinguistic Study of Language Acquisition, Volume 2 (ed. D. Slobin), pp. 1157–249. Lawrence Erlbaum Associates, Hillsdale, NJ. Slobin, Dan (2004). The many ways to search for a frog: Linguistic typology and the expression of motion events. In Relating Events in Narrative: Typological and Contextual Perspectives (ed. S. Strömqvist and L. Verhoeven), pp. 219–57. Lawrence Erlbaum Associates, Mahawah, NJ. Smith, Carlota (1991). The Parameter of Aspect. Kluwer, Dordrecht. Snyder, William (2007). Child Language: The Parametric Approach. Oxford University Press, Oxford. Son, Minjeong (2007). Directionality and resultativity: The cross-linguistic correlation revisited. Tromsø Working Papers on Language and Linguistics: Nordlyd, 34, 126–64. Son, Minjeong and Svenonius, Peter (2008). Microparameters of cross-linguistic variation: Directed motion and resultatives. In West Coast Conference on Formal Linguistics (WCCFL) 27, pp. 338–96. Sørensen, Thorvald (1948). A method of establishing groups of equal amplitude in plant sociology based on similarity of species and its application to analyses of the vegetation on Danish commons. Biologiske Skrifter, 5, 1–34. Speas, Margaret (1994). Null arguments in a theory of economy of projection. University of Massachussetts Occasional Papers in Linguistics, 17, 179–208. Speyer, Augustin (2008). Topicalization and Clash Avoidance: On the Interaction of Prosody and Syntax in the History of English with a few Glimpses at German. Ph.D. thesis, University of Pennsylvania. Speyer, Augustin (2010). Topicalization and Stress Clash Avoidance in the History of English. Mouton de Gruyter, Berlin. Srivastav, Veneeta (1991). The syntax and semantics of correlatives. Natural Language & Linguistic Theory, 9, 637–86.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

References



Starke, Michal (2014). Towards elegant parameters: Language variation reduces to the size of lexically-stored trees. In Linguistic Variation in the Minimalist Framework (ed. M. Picallo), pp. 140–52. Oxford University Press, Oxford. Starosta, Stanley, Pawley, Andrew, and Reid, Lawrence (1982). The evolution of focus in Austronesian. In Papers from the Third International Conference on Austronesian Linguistics, Volume 2: Tracking the Travellers (ed. A. Halim, L. Carrington, and S. Wurm), pp. 145–70. Steedman, Mark (2000). The Syntactic Process. MIT Press, Cambridge, MA. Steiner, B. Devan (2014). The Evolution of Information Structure and Verb Second in the History of French. Ph.D. thesis, Indiana University. Stephany, Ursula (1979). Modality. In Language Acquisition: Studies in First Language Development (ed. P. Fletcher and M. Garman), pp. 375–400. Cambridge University Press, Cambridge. Sternefeld, Wolfgang (2008). Syntax. Eine morphologisch motivierte generative Beschreibung des Deutschen (3. überarbeitete Auflage). Stauffenburg, Tübingen. Stevens, Jon (2010). A diachronic look at topicalization and contrast in English. Ms., University of Pennsylvania. Stevens, Jon and Light, Caitlin (2012). The pragmatics of direct object fronting in historical English. Talk presented at the 36th annual Penn Linguistics Colloquium (PLC 36), University of Pennsylvania, 24 March, 2012. Stimm, Helmut (1977). Ein neuer früher Beleg des synthetischen Futurs. In Festgabe für Julius Wilhelm zum 80. Geburtstag (ed. H. Laitenberger), pp. 40–52. Steiner, Wiesbaden. Streiner, David (2012). Unicorns do exist: A tutorial on ‘proving’ the null hypothesis. In A Guide for the Statistically Perplexed: Selected Readings for Clinical Researchers (ed. D. Streiner), pp. 211–25. University of Toronto Press, Toronto, ON. Sun, Chaofen (1996). Word-order Change and Grammaticalization in the History of Chinese. Stanford University Press, Stanford, CA. Svenonius, Peter (2010). Spatial P in English. In The Cartography of Syntactic Structures, Volume 6: Mapping Spatial PPs (ed. G. Cinque and L. Rizzi), pp. 127–60. Oxford University Press, Oxford. Sweetser, Eve (1990). From Etymology to Pragmatics: Metaphorical and Cultural Aspects of Semantic Structure. Cambridge University Press, Cambridge. Szegő, Júlia (1988). Ismeretlen moldvai nótafák. Európa Könyvkiadó, Budapest. Szmrecsanyi, Benedikt (2006). Morphosyntactic Persistence in Spoken English: A Corpus Study at the Intersection of Variationist Sociolinguistics, Psycholinguistics, and Discourse Analysis. Mouton de Gruyter, Berlin. Tagliamonte, Sali and D’Arcy, Alexandra (2004). He’s like, she’s like: The quotative system in Canadian youth. Journal of Sociolinguistics, 8, 493–514. Tagliamonte, Sali and D’Arcy, Alexandra (2007). The modals of obligation/necessity in Canadian perspective. English World-Wide, 28, 47–87. Tailleur, Sandrine (2012). The French Wh-Interrogative System: Est-ce que, Clefting? Ph.D. thesis, University of Toronto. Talmy, Leonard (1975). Semantics and syntax of motion. In Syntax and Semantics 4 (ed. J. Kimball), pp. 181–238. Academic Press, New York. Talmy, Leonard (1985). Lexicalization patterns: Semantic structure in lexical forms. In Language Typology and Syntactic Description, Volume 3 (ed. T. Shopen), pp. 57–149. Cambridge University Press, Cambridge. Talmy, Leonard (2000). Toward a Cognitive Semantics. MIT Press, Cambridge, MA.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i



References

Taraldsen, Knut Tarald (2001). Subject extraction, the distribution of expletives, and stylistic inversion. In Subject Inversion in Romance and the Theory of Universal Grammar (ed. A. Hulk and J.-Y. Pollock), pp. 163–81. Oxford University Press, Oxford. Taylor, Ann, Nurmi, Arja, Warner, Anthony, Pintzuk, Susan, and Nevalainen, Terttu (2006). The York–Helsinki Parsed Corpus of Early English Correspondence (PCEEC). Department of Language and Linguistic Science, University of York. http://wwwusers.york.ac.uk/∼lang22/PCEEC-manual/. Taylor, Ann, Warner, Anthony, Pintzuk, Susan, and Beths, Frank (2003). The York– Toronto–Helsinki Parsed Corpus of Old English prose (YCOE). Department of Language and Linguistic Science, University of York. http://www-users.york.ac.uk/∼lang22/YCOE/ YcoeHome.htm. Telve, Stefano (2005). Sul trapassato remoto: valori sintattici, aspettuali e semantici dei tipi fui stato e fui stato amato in italiano antico e moderno. Lingua e Stile, 40, 263–94. Tenny, Carol (1987). Grammaticalizing Aspect and Affectedness. Ph.D. thesis, Massachusetts Institute of Technology. Tesnière, Lucien (1959). Eléments de syntaxe structurale. Klincksieck, Paris. Thielmann, Philipp (1885). Habere mit dem Infinitiv und die Enstehung des romanischen Futurums. Archiv für lateinische Lexikographie und Grammatik, 2, 48–89, 157–202. Thurneysen, Rudolf (1892). Zur Stellung des Verbums im Altfranzösischen. Zeitschrift für romanische Philologie, 16, 289–307. Tollan, Rebecca (2013). Modals in West Germanic: Syntax, semantics and diachronic change. Ms., University of Toronto. Tomblin, J. Bruce, Records, Nancy, Buckwalter, Paula, Zhang, Xuyang, Smith, Elaine, and O’Brien, Marlea (1997). Prevalence of specific language impairment in kindergarten children. Journal of Speech, Language, and Hearing Research, 40, 1245–60. Trapp, Erich (1965). Der Dativ und der Ersatz seiner Funktionen in der byzantinischen Vulgärdichtung bis zur Mitte des 15. Jahrhunderts. Jahrbuch der Österreichischen Byzantinischen Gesellschaft, 14, 21–34. Traugott, Elizabeth Closs (1989). On the rise of epistemic meanings in English: An example of subjectification in semantic change. Language, 65, 31–55. Traugott, Elizabeth Closs (1993). The conflict promises/threatens to escalate into war. In Proceedings of the Nineteenth Annual Meeting of the Berkeley Linguistics Society: General Session and Para-Session on Semantic Typology and Semantic Universals (ed. J. Guenter, B. Kaiser, and C. Zoll), pp. 348–58. Berkeley Linguistics Society, Berkeley, CA. Traugott, Elizabeth Closs (1997). Subjectification and the development of epistemic meaning: The case of ‘promise’ and ‘threaten’. In Modality in Germanic Languages: Historical and Comparative Perspectives (ed. T. Swan and O. J. Westvik), pp. 185–210. De Gruyter, Berlin. Traugott, Elizabeth Closs (2006). Historical aspects of modality. In The Expression of Modality (ed. W. Frawley), pp. 107–40. De Gruyter, Berlin. Traugott, Elizabeth Closs and Dasher, Richard (2005). Regularity in Semantic Change. Cambridge University Press, Cambridge. Travis, Lisa (1984). Parameters and Effects of Word Order Variation. Ph.D. thesis, Massachusetts Institute of Technology. Travis, Lisa (1991). Parameters of phrase structure and V2 phenomena. In Principles and Parameters in Comparative Grammar (ed. R. Freidin), pp. 339–64. MIT Press, Cambridge, MA.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

References



Tremblay, Mireille, Dupuis, Fernande, and Dufresne, Monique (2003). Les prépositions dans l’histoire du français: transitivité, grammaticalisation et lexicalisation. Verbum, 25, 549–62. Troberg, Michelle (2011). Directed motion in Medieval French. In Romance Linguistics 2010: Selected Papers from the 40th Linguistics Symposium on Romance Languages (ed. J. Herschensohn), pp. 117–34. John Benjamins, Amsterdam. Troberg, Michelle and Burnett, Heather (2014). Le prédicat résultatif adjectival en français médiéval. Linguisticae Investigationes, 37, 156–80. Truswell, Robert and Gisborne, Nikolas (2015). Quantificational variability and the genesis of English headed wh-relatives. In Proceedings of Sinn und Bedeutung 19 (ed. E. Csipak and H. Zeijlstra), pp. 639–56. Tsimpli, Ianthi Maria and Papadopoulou, Despina (2006). Aspect and argument realization: A study on antecedentless null objects in Greek. Lingua, 116, 1595–615. Tsunoda, Tasaku, Ueda, Sumie, and Itoh, Yoshiaki (1995). Adpositions in word order typology. Linguistics, 33, 741–61. Väänänen, Veikko (1981). Introduction au latin vulgaire (3e édition revue et augmentée). Klincksieck, Paris. Valesio, Paolo (1968). The Romance synthetic future pattern and its first attestations. Lingua, 20, 113–61, 279–306. Valesio, Paolo (1970). The synthetic future again: Phonology and morpho-syntax. Lingua, 24, 181–93. van der Auwera, Johan and Ammann, Andreas (2011). Overlap between situational and epistemic modal marking. In The World Atlas of Language Structures Online (ed. M. Dryer and M. Haspelmath). Max Planck Digital Library, Munich. http://wals.info/chapter/76. van der Wurff, Wim (1990). Diffusion and Reanalysis in Syntax. Ph.D. thesis, Universiteit Leiden. van Driem, George (1986). A Grammar of Limbu. Mouton de Gruyter, Berlin. van Gelderen, Elly (1997). Structures of tense and aspect. Linguistic Analysis, 27, 138–65. van Gelderen, Elly (2000). A History of English Reflexive Pronouns. John Benjamins, Amsterdam. van Gelderen, Elly (2004). Grammaticalization as Economy. John Benjamins, Amsterdam. van Gelderen, Elly (2009a). Cyclical Change. John Benjamins, Amsterdam. van Gelderen, Elly (2009b). Feature economy in the linguistic cycle. In Historical Syntax and Linguistic Theory (ed. P. Crisma and G. Longobardi), pp. 93–109. Oxford University Press, Oxford. van Gelderen, Elly (2011a). The Linguistic Cycle. Oxford University Press, Oxford. van Gelderen, Elly (2011b). Valency changes in the history of English. Journal of Historical Linguistics, 1, 106–43. van Kemenade, Ans (1987). Syntactic Case and Morphological Case in the History of English. Foris, Dordrecht. van Kemenade, Ans (1997). V2 and embedded topicalization in Old and Middle English. In Parameters of Morphosyntactic Change (ed. A. van Kemenade and N. Vincent), pp. 326–52. Cambridge University Press, Cambridge. van Kemenade, Ans and Miličev, Tanja (2012). Syntax and discourse in Old and Middle English word order. In Grammatical Change: Origins, Nature, Outcomes (ed. D. Jonas, J. Whitman, and A. Garrett), pp. 239–54. Oxford University Press, Oxford.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i



References

van Kemenade, Ans and Westergaard, Marit (2012). Syntax and information structure: Verb second variation in Middle English. In Information Structure and Syntactic Change in the History of English (ed. A. Meurman-Solin, M. J. López-Couso, and B. Los), pp. 87–118. Oxford University Press, Oxford. van Koert, Margreet, Koeneman, Olaf, Weerman, Fred, and Hulk, Aafke (2015). The quantificational asymmetry: A comparative look. Lingua, 155, 140–53. van Koppen, Marjo (2006). A new view on first conjunct agreement: Evidence from Dutch dialects. In Studies on Agreement (ed. J. Costa and M. C. Figueiredo Silva), pp. 121–41. John Benjamins, Amsterdam. van Koppen, Marjo (2007). Agreement with coordinated subjects: A comparative perspective. Linguistic Variation Yearbook, 7, 121–61. van Koppen, Marjo (to appear). Complementizer agreement. In The Blackwell Companion to Syntax (ed. M. Everaert and H. van Riemsdijk). Wiley-Blackwell, Oxford. van Kuppevelt, Jan (1991). Topic en Comment: Expliciete en Impliciete Vraagstellingen in Discourse. Ph.D. thesis, Katholieke Universiteit te Nijmegen. van Kuppevelt, Jan (1995). Discourse structure, topicality, and questioning. Linguistics, 31, 109–47. van Kuppevelt, Jan (1996). Inferring from topics. Linguistics and Philosophy, 19, 393–443. Vance, Barbara (1989). The evolution of pro-drop in Medieval French. In Studies in Romance Linguistics (ed. C. Kirschner and J. DeCesaris), pp. 413–41. John Benjamins, Amsterdam. Vance, Barbara (1997). Syntactic Change in Medieval French: Verb Second and Null Subjects. Kluwer, Dordrecht. Vance, Barbara, Donaldson, Bryan, and Steiner, B. Devan (2010). V2 loss in Old French and Old Occitan: The role of fronted clauses. In Romance linguistics 2009: Selected Papers from the 39th Linguistic Symposium on Romance Languages (ed. S. Colina, A. Olarrea, and A. M. Carvalho), pp. 301–20. John Benjamins, Amsterdam. Vanelli, Laura, Renzi, Lorenzo, and Benincà, Paola (1986). Typologie des pronoms sujets dans les langues romanes. In Actes du XVIIème Congrès International de Linguistique et de Philologie Romanes, Volume 3: Linguistique descriptive: phonétique, morphologie et lexique, Aix-enProvence, pp. 161–76. Université de Provence. Verhagen, Arie (2000). ‘The girl that promised to become something’: An exploration into diachronic subjectification in Dutch. In The Berkeley Conference on Dutch Linguistics 1997: The Dutch Language and the Millennium (ed. T. Shannon and J. Snapper), pp. 197–208. University Press of America, Lannham, MD. Verhagen, Arie (2006). Dreigt polysemie uit de hand te lopen? Een oefening in semantische categorisering. Jaarboek voor Neerlandistiek, 24, 159–68. Verhulst, Pierre-François (1845). Recherches mathématiques sur la loi d’accroissement de la population [Mathematical research into the law of population growth increase]. Nouveaux Mémoires de l’Académie Royale des Sciences et Belles-Lettres de Bruxelles, 18, 1–42. Vikner, Sten (1995). Verb Movement and Expletive Subjects in the Germanic Languages. Oxford University Press, Oxford. Villa-García, Julio (2012). Characterizing medial and low complementizers in Spanish: Recomplementation que and ‘jussive/optative’ que. In Current Formal Aspects of Spanish Syntax and Semantics (ed. M. González-Rivera and S. Sessarego), pp. 198–229. Cambridge Scholars Publishing, Newcastle upon Tyne.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

References



von Fintel, Kai (1995). The formal semantics of grammaticalization. In Proceedings of the 25th North East Linguistics Society (NELS) (ed. J. Beckman), pp. 175–90. GLSA, Amherst, MA. Walkden, George (2009). The comparative method in syntactic reconstruction. Master’s thesis, University of Cambridge. Walkden, George (2012). Syntactic Reconstruction and Proto-Germanic. Ph.D. thesis, University of Cambridge. Walkden, George (2014). Syntactic Reconstruction and Proto-Germanic. Oxford University Press, Oxford. Wallage, Phillip (2008). Jespersen’s Cycle in Middle English: Parametric variation and grammatical competition. Lingua, 118, 643–74. Wallenberg, Joel (2009). Antisymmetry and the Conservation of C-command: Scrambling and Phrase Structure in Synchronic and Diachronic Perspective. Ph.D. thesis, University of Pennsylvania. Wallenberg, Joel, Ingason, Anton Karl, Sigurðsson, Einar Freyr, and Rögnvaldsson, Eiríkur (2011). Icelandic Parsed Historical Corpus (IcePaHC). http://www.linguist.is/ icelandic_treebank. Ward, Jr, Joe (1963). Hierarchical grouping to optimize an objective function. Journal of the American Statistical Association, 58, 236–44. Warner, Anthony (1993). English Auxiliaries: Structure and History. Cambridge University Press, Cambridge. Warner, Anthony (2006). Variation and the interpretation of change in periphrastic do. In The Handbook of the History of English (ed. A. van Kemenade and B. Los), pp. 45–91. Blackwell, Oxford. Watanabe, Akira (2000). Feature copying and binding: Evidence from complementizer agreement and switch reference. Syntax, 3, 159–81. Weerman, Fred (1993). The diachronic consequences of first and second language acquisition: The change from OV to VO. Linguistics, 31, 903–31. Weerman, Fred (2011). Diachronic change: Early versus late acquisition. Bilingualism: Language and Cognition, 14, 149–51. Weiner, E. Judith and Labov, William (1983). Constraints on the agentless passive. Journal of Linguistics, 19, 29–58. Weinreich, Uriel, Labov, William, and Herzog, Marvin (1968). Empirical foundations for a theory of language change. In Directions for Historical Linguistics (ed. W. Lehmann and Y. Malkiel), pp. 95–195. University of Texas Press, Austin, TX. Weiß, Helmut (2012). The diachrony of complementizer agreement. Paper presented at Generative Initiatives in Syntactic Theory 6: Complementizer Agreement Workshop, 17 October, 2012, Universiteit Gent. http://www.gist.ugent.be/file/366. Wells, Gordon (1979). Learning and using the auxiliary verbs in English. In Cognitive Development: Language and Thinking from Birth to Adolescence (ed. V. Lee), pp. 250–70. Croom Helm, London. Wessely, Carl (1895). Corpus Papyrorum Raineri Archeducis Austriae 1: Griechische Texte. Vindobonae, Vienna. Westergaard, Marit (2005). Norwegian child language and the history of English: The interaction of syntax and information structure in the development of word order. In Contexts— Historical, Social, Linguistic: Studies in Celebration of Toril Swan (ed. K. McCafferty, T. Bull, and K. Killie), pp. 293–310. Peter Lang, Bern.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i



References

Wexler, Kenneth (1998). Very early parameter setting and the unique checking constraint: A new explanation of the optional infinitive stage. Lingua, 106, 23–79. Whitman, John (2001). Relabeling. In Diachronic Syntax: Models and Mechanisms (ed. S. Pintzuk, G. Tsoulas, and A. Warner), pp. 220–40. Oxford University Press, Oxford. Whitman, John (2008). The classification of constituent order generalizations and diachronic explanation. In Linguistic Universals and Language Change (ed. J. Good), pp. 233–52. Oxford University Press, Oxford. Whitman, John and Paul, Waltraud (2005). Reanalysis and conservancy of structure in Chinese. In Grammaticalization and Parametric Variation (ed. M. Batllori, M.-L. Hernanz, C. Picallo, and F. Roca), pp. 82–94. Oxford University Press, Oxford. Widmer, Paul (2013). Agreement. In Encyclopedia of Ancient Greek Language and Linguistics, Leiden. Brill. http://referenceworks.brillonline.com/entries/encyclopedia-of-ancient-greeklanguage-and-linguistics/agreement-COM_00000014. Winters, Margaret (1984). Steps toward the Romance passive inferrable from the Itinerarium Egeriae. Romance Philology, 37, 445–54. Wöllstein-Leisten, Angelika (2001). Die Syntax der Dritten Konstruktion. Eine repräsentationelle Analyse zur Monosententialität von ‘zu’-Infinitiven im Deutschen. Stauffenburg, Tübingen. Wood, Jim (2011). Stylistic fronting in spoken Icelandic relatives. Nordic Journal of Linguistics, 34, 29–60. Wright, Jonathan (2009). Hkongso grammar sketch. Master’s thesis, Graduate Institute of Applied Linguistics, Dallas, TX. Wurmbrand, Susanne (1999). Modal verbs must be raising verbs. In Proceedings of the 18th West Coast Conference on Formal Linguistics (WCCFL 18) (ed. S. Bird, A. Carnie, J. Haugen, and P. Norquest), pp. 599–612. Cascadilla Press, Somerville, MA. Wurmbrand, Susanne (2001). Infinitives: Restructuring and Clause Structure. De Gruyter, Berlin. Xydopoulos, George and Tsangalidis, Anastasios (2007). I rimatiki apopsi stin eliniki ke i shesis E, R, S [Aspect in Greek and the E, R, S relations]. Studies in Greek Linguistics, 27, 322–36. Yang, Charles (2002). Knowledge and Learning in Natural Language. Oxford University Press, Oxford. Yap, Foong ha, Hårsta-Grunow, Karen, and Wrona, Janick (2011). Nominalization in Asian Languages: Diachronic and Typological Perspectives. John Benjamins, Amsterdam. Zanuttini, Raffaella (1997). Negation and Clausal Structure: A Comparative Study of Romance Languages. Oxford University Press, Oxford. Zaring, Laurie (2011). On the nature of OV and VO order in Old French. Lingua, 121, 1831–52. Ziegeler, Debra (1997). Retention in ontogenetic and diachronic grammaticalization. Cognitive Linguistics, 8, 207–41. Zwart, C. Jan-Wouter (1991). Clitics in Dutch: Evidence for the position of INFL. Groninger Arbeiten zur germanistischen Linguistik, 33, 71–92. Zwart, C. Jan-Wouter (1993). Dutch Syntax: A Minimalist Approach. Ph.D. thesis, Rijksuniversiteit Groningen. Zwart, C. Jan-Wouter (1997). Morphosyntax of Verb Movement: A Minimalist Approach to the Syntax of Dutch. Kluwer, Dordrecht. Zwart, C. Jan-Wouter (2012). Agreement without agree. Paper presented at Generative Initiatives in Syntactic Theory 6: Complementizer Agreement Workshop, 17 October 2012, Universiteit Gent. http://www.let.rug.nl/zwart/docs/gent2012.pdf.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

Index abstraction 6, 61, 67, 73, 278 accent 184–6, 188–91, 201 accessibility hierarchy 25–6, 31–42 accomplishment 69, 109 achievement 63 acquisition 1, 3–5, 10–23, 40, 41, 43, 45, 46, 59, 63, 67–8, 73, 76–9, 101, 118, 121, 122, 210, 249–50, 252–4, see also first language, second language first language 10–20, 23, 78, 89 imperfect learning 9–13, 16, 17, 22–4, 76–8, 97, see also extension, reanalysis Language Acquisition Device 78 peer-to-peer 11–13, 24 second language 63, 78 activity 109, 111 actuation 13, 32, 35, 40, 82–7, 92, 95 adjacency 184, 237, 240, 241, 246–53 adolescent 12, 78 adposition 14, 16, 44–50, 56, 58, 59, see also postposition, preposition adult 10–12, 17, 79, 89, 91, 97 as innovator 10, 17, 77–9 adverb 31, 32, 35–7, 39, 58, 110, 113, 127, 145, 146, 151–2, 154–7, 160, 162, 171, 206–8, 210, 213, 214, 229, 240–2, 247, 260–3, 270, see also negation, adverbial adverbial clause 46–8, 171, 172 affectedness 127–30, 136, 139 agent 136, 143, 277 agreement 7, 9, 11, 19, 21, 28, 62, 63, 65–7, 73, 126, 129, 140–4, 147, 148, 153, 236–54, 263 complementizer 9, 236–54 first conjunct 243–4, 246–7 goal 237, 241, 242, 244, 246–8, 250, 251, 253 possessor 66–7, 73, 243–7 probe 237, 240, 242–4, 246–53, 263 Amis 45 analogy 17, 220, 225, 237, 241, 242, 249 analogical extension 6, 41, 61, 67, 70, 73, 248, 249 analogical levelling 220, 224 analytic 8, 216, 217, 219–21, 228, 234 animacy 9, 97, 138, 141, 244, 259, 271–7 applicative 46, 48

appositive 36, 149, see also relative clause, appositive Arabic 126, 244 Armenian 55 article 126–8, 132–4, 137, 141–3 definite 125, 127, 128, 130–2, 136, 139, see also definiteness indefinite 132, 139, see also indefiniteness aspect 7, 48, 61–5, 67, 68, 70–3, 107, 109, 125–30, 132–43, 255, 263, 264, 269, 276, 278 aspectual morpheme 48, 65, 67–73, 106–16, 123, 127–30, 136, 255, 270 durative 48, 134, 270 imperfective 62–4, 67–9, 71, 73, 106, 110, 126, 127, 132–4, 138, 139, 142, 222, 233 inner 61, 68–73, 104, 106–9, 112, 113, 115, 121–3, 125–6, 129, 131, 133–6, 139, 141, 143, 144 outer 6, 61–3, 68, 71–3, 125, 131, 133–6, 138–41, 144 perfective 48, 62–5, 67, 68, 71, 105, 126, 127, 129, 130, 132–6, 138–9, 142, 143, 216, 218, 224–6, 234, 264, 269 prospective 269–70, 273 assertion 149, 173–4, 196 Austronesian languages 48, see also Amis, Tongan auxiliary 5, 8, 14, 19–21, 62, 64, 67, 68, 71–3, 108, 111, 115, 121, 149, 216–18, 220–34, 259, 261, 264–8 background 149, 150, 264 Bai 45 Bantu languages 236, 242 bare nominal 20, 126, 131, 132 bare verb 20, 69–71 BE-periphrasis 216–34 benefactive 46, 136 bleaching 121, 123, 256 bleeding 101 Borer–Chomsky Conjecture 1, 207 bridge verb 145, 173 Burmese 48 Cariban languages 48 cartography 267

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i



Index

case 6–7, 30, 32–3, 35, 40, 54, 56, 57, 97, 99, 103, 105, 107, 125–32, 135–44, 174, 246 ablative 105, 106, 216, 219, 220, 231, 233 accusative 6, 7, 28, 30, 62–4, 68, 69, 71, 97, 105, 106, 126–32, 134, 136–7, 139–44, 169, 177, 183, 188, 190–2, 194, 195, 197, 199, 216, 218–20, 222, 233, 234, 266, 274 dative 6, 30, 63, 69, 71, 94, 96–9, 101, 103, 105, 128, 132, 136, 137, 139, 141, 188–90, 192, 194, 195, 198, 199, 204, 222, 255–8, 260–3, 273–5, 277 genitive 30–2, 37, 43–6, 48, 49, 53, 54, 56–60, 73, 105, 127–30, 132, 134, 136–7, 139, 141–3, 150, 151, 156, 159, 165, 166, 177–9, 191, 204, 216, 219, 220 inherent 125, 128, 129, 135–6, 140, 143, 144 instrumental 71 nominative 28, 30, 67, 70, 73, 105, 106, 114, 127, 132–4, 137, 141–3, 169, 178, 183, 188, 190–2, 194, 195, 197–9, 216, 218–20, 222, 229, 231–3 oblique 31, 37, 56, 137, 189, 190 partitive 6, 126–8 structural 6, 125, 129, 131, 135, 136, 141, 142 Catalan, Old 106, 153, 157, 159 categorial grammar 26 change of state 108, 136 child 10–14, 16–24, 31, 45, 67, 68, 77, 78, 88, 89, 121, 122 as innovator 10–14, 16–24, 63, 77, 78, 97 preadolescent 11–13 Chinese 43, 45–9, 60, 126, see also Mandarin, Sino-Tibetan languages Late Old 46 Old 46 Chuvash 64, 65 clausal complementation 5, 9, 11, 18–21, 99, 134, 145–8, 150, 152–6, 162, 165–70, 172–81, 190, 209, 219, 220, 236, 237, 243, 249, 250, 256–64, 266, 270, 275 direct speech 260, 262, 263, 274–6 embedded question 163 indirect speech 258, 262, 263 of factive verbs (Class D) 173–6, 178, 180 of other verbs (Class C) 173–6, 178, 180 of semifactive verbs (Class E) 173–6, 178, 181 of strong assertive verb (Class A) 173–8, 181 of verbs of volition (Class V) 175, 176, 179, 181 of weak assertive verb (Class B) 173–7, 181

that-clause 168, 205–6, 259–60, 262, 263, 273–5 clitic 19, 45, 65, 137, 153, 178, 180, 228–31, 234, 237, 238, 248–51, 253 cluster analysis 5, 43, 49, 51, 53, 55–60 code-switching 207, 245 comparative 204 competition 33, 40, 41, 85, 207, 216, 234 grammar competition 4, 6, 40, 91, 92, 101, 118, 206, 207, 253 complementizer 27, 28, 30, 153, 164, 169–73, 179–81, 205, 236–52, 259, see also agreement, complementizer relative complementizer 27–9, 29, 32, 33, 41 complex verb 5, 64–6, 68, 71–3 complexity 13, 14, 18–23, 28, 31, 32, 199, 200, 256 conditional 62, 64, 65, 70, 233, 234 connectivity 28, 30 Constant Rate Hypothesis 4, 6, 8, 76, 83–5, 92–6, 100–3, 118, 202, 208, 210, 215 construction grammar 26, 87 contact 5, 6, 11–13, 17, 23, 32, 45, 61, 64, 66, 68, 70, 72–91 contrast 8, 161, 182, 185–93, 195–6, 200, 201, see also focus, contrastive; topic, contrastive control 9, 48, 255, 256, 258, 261–4, 271 coordination 39, 150, 243, 244, 246, 247, see also agreement, first conjunct copula 45, 62, 65–8, 73, 176, 220, 229 corpus 8, 23, 27, 29, 32, 33, 37–9, 64, 109–14, 116, 117, 121, 122, 146, 148, 150, 156, 168, 174, 175, 179, 185–8, 190–2, 200, 202–4, 206, 207, 210–13, 215, 221, 222, 232, 255–60, 262, 263, 265–8, 272–6, 278 creole 13 critical period 12 cycle 2–4, 11, 18, 22, 23, 253 Jespersen’s Cycle 4, 208 modal 14–18 Danish 174 Middle 146 Old 146 defectiveness 258, 262, 263, 267, 270 definiteness 36, 67, 69, 125–32, 135, 136, 138, 139, 143, 159, see also article, definite demonstrative 5, 8, 26, 30, 33, 40, 49, 52, 57–9, 125, 128, 130–2, 136, 178, 182–3, 185–7, 190–201 dependency grammar 26 deponent 139, 140, 142, 216, 218, 225–9, 234

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

Index differential equation 76, 84–91, 93, 95 diffusion 4, 6, 10–13, 16, 17, 23–6, 32, 37–41, 68, 78, 85, 86, 95, 124, 208 directed change 9, 13, 19, 26, 252, 253 directional 106–10, 112, 113, 115–17, 119–23 ditransitive 6, 102, 103, 190, 205 dative alternation 204, 205 do-support 83, 87, 95 Dutch 9, 111, 126, 169, 189, 191, 236–50, 255, 264, 267, 268 Flemish 236–41, 243–8 Middle 80, 87 E-language 76, 83, 91, 92 economy 21, 207 English 7, 14–21, 25–30, 32, 35, 36, 39, 41, 45, 55, 62, 78, 83, 94–9, 101–3, 106, 120, 125–30, 132, 133, 135, 136, 143, 173–5, 182–7, 189, 191, 192, 196, 197, 200, 201, 203–6, 208, 209, 230, 255, 264, 267, 270, 271 African American Vernacular 22 Early Modern 8, 129–30, 185, 187, 191–2, 209 Middle 6–8, 19, 25–8, 33–4, 38, 97–9, 101, 103, 128–30, 136, 185, 202, 206–10 Early Middle 30, 33–41, 97, 101, 127, 129, 130 Old 7, 19, 22, 26–30, 35–40, 80, 97, 99, 127–30, 136, 139, 157, 168–86, 201, 206 Scots, Middle 32 erosion 21, 97, 109, 121, 249 event 11, 14, 15, 22, 62, 63, 68, 71, 105, 108, 109, 113, 114, 117, 122–4, 126, 127, 129, 219, 220, 266, 269, 270, 277, 278 evidential 12, 257, 259, 263, 265, 267–70, 273, 275, 278 experiencer 136, 188–90 extension 5, 11–13, 15–19, 22, 23, 35, 39, 61, 83, 137, 221, 225, 227, 247, 248, 276–8, see also analogy, analogical extension extrametricality 230 extraposition 36, 39, 261–3 failed change 6, 75–6, 87, 90–6, 98–101, 103, see also transient state feature inheritance 240, 242–3 figure–ground 105, 107, 119 Fin 147, 161–4, 166, 179–81, 268 final-over-final constraint 44 Finnish 16, 55, 126, 127 Finno-Ugric languages 29, see also Finnish, Hungarian, Khanty, Komi, Mansi,



Mari, Udmurt, Ugric languages, Uralic languages Florentine, Old 158 focus 70, 147–50, 161–6, 193, 195–7, 244 contrastive 148, 151, 162, 165, 200, see also contrast information 149, 150 Force 161, 179, 180, 263 French 6–8, 16, 28, 32, 55, 101, 104, 107–9, 115, 116, 118, 123, 129, 146, 154, 267, 268 Medieval 6, 104, 106, 109–23, 145–66 Middle French 116, 145, 155, 156, 166, 270 Old French 7, 116, 145–6, 148, 152–9, 168–9, 171–81 Montreal dialect 203 frequency 3, 6, 32, 34, 72, 92, 113, 116, 122, 142, 157, 176, 183, 184, 191, 200, 215, 216, 222–9, 232, 273 Frisian 238, 239 functional head 44, 46, 48, 118, 124, 126, 143, 148, 163, 164, 171, 207, 232, 256, 257, 262–5, 269, 270, 277–8 Georgian 55 German 7–9, 15, 28, 55, 72, 80, 126, 162, 169, 173, 179, 181, 183, 185, 187–91, 193–5, 197–201, 236, 255–78 Austrian 247–8 Bavarian 236, 238–9, 241, 243, 247 Early New High 8, 187, 188, 271–2, 274, 276, 278 Middle High 15, 271 Middle Low 80 Old High 7, 15, 127, 128, 271 Germanic languages 7, 8, 168–70, 182–3, 186, 187, 189, 191, 193, 201, 236–8, 243, 253 North 7, 169, 173, 179, 181, see also Danish, Icelandic, Norse (Old), Norwegian, Swedish West 16, see also Dutch, English, Frisian, German, Luxemburgish, Yiddish gerund 66–8, 73, 229 goal 36, 97, 104, 105, 107–12, 115–23, 139 grammaticalization 2, 5, 9, 15, 22–3, 48, 119, 132, 135, 233, 234, 237, 248–57, 270, 271, 273, 275–8 Greek 7, 16, 55, 125–6, 129–44, 278 Ancient 125–6, 131–44 Koine 125, 132–4, 137, 143 Post-Koine 6, 125, 126, 131, 132, 134, 135, 137, 139–44 Gungbe 270

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i



Index

heavy NP shift 98, 99 Hebrew 55 heritage language 11, 13 Hindi 55 Hkongso 45 Hungarian 5, 6, 55, 61–73 Middle 69, 70, 72 Old 61–72 Proto-Hungarian 61, 66–8 I-language 12, 76, 83, 83–253 Icelandic 8, 146, 148, 151–4, 158, 166, 169–70, 181, 183, 187, 189, 191–2 Old 157 imperative 69, 71, 95, 257, 262, 263, 272 imperfect 216 impersonal predicate 176 incorporation 48, 105, 119–21, 230, 231 indefiniteness 30, 36, 40, 67, 126, 159, 203, see also article, indefinite Indo-Aryan languages 29, see also Hindi Indo-European languages 5, 25–6, 29, 30, 33, 41, 55, 131, 133, 136, 139, 242, 243, see also Armenian, Germanic languages, Greek, Indo-Aryan languages, Irish, Latvian, Persian, Romance languages, Slavic languages, Welsh Proto-Indo-European 25–6, 29, 41, 141 infinitive 18–21, 122, 131, 132, 134, 143, 146, 150, 154, 155, 158, 159, 165, 205, 218, 219, 222–4, 233, 234, 255–78, see also clausal complementation, infinitival information structure 171, 182, 186–9, 194, 195, 201, 204, see also contrast, focus, subinformativity, topic new information 149, 193, 204 old information 190 innovation 4, 6, 11–24, 33, 78, 85, 89, 90, 104, 111–12, 115, 123, 207, 216, 218, 225, see also adult, as innovator; child, as innovator interrogative 5, 95, 165, 196, 210, 267, see also clausal complementation, embedded question wh-question 83, 98, 164 intervention 148, 151, 163, 184, 234, 239–41, 246, 247, 250, 252, 263, 273 intransitive 136 inversion 155, 183, 247 Irish 55 Italian 157, 217, 221, 226, 228, 270 Old 106, 159 Japanese 55, 126

Khalkha 55 Khanty 63, 66 Kobon 270 Komi 65 Latin 6, 8, 32, 61, 63–4, 104–10, 112–13, 216–33, 267 Classical 104–7, 109, 111, 113, 216–20, 222, 229, 231–2, 234 Early 104–7, 109, 115, 116, 118, 119, 216, 218, 221, 229 Late 8, 63, 109, 113, 119, 121, 217–25, 228, 230–4 Latvian 55 left periphery 2, 8, 39, 180–1, 247, 263, 268, see also Fin, Force left-dislocation 147, 161 Leftward Stylistic Displacement 7, 145–67, 173, see also Stylistic Fronting LSDLeft 146, 153, 154, 155, 157, 161, 165–6 LSDRight 146, 154–7, 161–7 Limbu 126 Lithuanian 72 locative 34–5, 46, 104, 106, 107, 110, 112, 115, 118–20, 205 Logical Form 15, 18 logistic function 75–6, 81, 83–7, 91–6, 100, 101, 103, 209 Luxemburgish 238–9 Mandarin 16, 45, 46, see also Chinese Mande languages 236 manner 105–9, 111–12, 118, 120–3, 174 Mansi 63 mapping problem 11, 13–23 Mari 65 measure 127–9, 136 Merge 2, 16, 119, 138, 169, 176, 179–81, 231, 255, 256, 259, 264, 269, 278 mesoclisis 234 minimalism 2, 9, 26, 44 modal 12–16, 18–22, 176, 222, 223, 264–6, see also cycle, modal epistemic 5, 14–18, 257, 263–8, 270 polysemous 14–16, 18 root 5, 14–18, 229, 266 morphology 7, 11, 12, 17, 31, 32, 41, 46, 49, 56–8, 61, 68, 72, 97, 99, 106, 119, 133, 134, 136, 140–3, 216, 236, 238, 248, 263, 266, 267, see also agreement movement 2, 7, 8, 28, 98, 99, 138, 145, 147, 148, 157, 159–61, 163, 165, 183, 185, 188, 189, 242, 244, 249–51, 263, 277, 278,

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

Index see also heavy NP shift, scrambling, Stylistic Fronting, topicalization A 148, 163, 164, 242 Formal Movement 185, 186, 188–91, 201 True A-bar Movement 162, 185, 186, 188–91, 201 head movement 7, 107, 119, 146–8, 152–4, 157, 161–2, 168, 170–2, 176, 178, 180–1, 189, 222, 231, 232, 242–4, 249–50, 259 Head Movement Constraint 231, 232 raising 143, 255–8, 261–5, 267, 271, 273, 275, 276, 278 VP-movement 7, 148, 157–61, 165, 173 remnant 157–62, 167, 232 multiple correspondence analysis 43, 49–56, 58–9 mutual exclusivity 41, 207, 245 negation 8, 38, 39, 54, 57, 70, 83, 95, 121, 143, 148, 151, 152, 156, 158, 162, 164, 166, 171, 173, 180, 202–3, 206–15, 222, 231–4, 260–1, 265, 272 adverbial 151–2, 162, 207, 212, 213 negative concord 208–9 Nigaraguan Sign Language 13 Niger–Congo languages 48, see also Bantu languages, Gungbe, Mande languages, Nkore-Kiga Nkore-Kiga 269 nominalization 48, 49, 60, 143 Norse, Old 127 Norwegian 169, 174 noun, relational 47–9 null object 69, 132 null subject 45, 46, 146, 148, 153–5, 163, 176, 238 optionality 48, 172, 180, 203, 205–7, 238, 240, 252, 253 parameter 1, 5, 43, 49, 50, 56–8, 77–9, 89, 91, 126, 129, 169, 181, 243 head parameter 44, 49, 59, 78 parenthetical 5, 39, 149 participle 66, 106, 122, 137, 146, 149, 150, 152–64, 231, 216–34, 234, 258, 260, 261, 264–6, 268–70, 272–6 particle 15, 28, 48, 54, 63, 68–71, 131, 132, 134, 141, 142, 151, 158, 172, 190, 191, 204, 220 verbal 6, 61, 68–73, 104–6, 109, 112–18, 120–1, 123, see also resultative, particle passive 8, 69, 71, 135, 136, 143, 176, 178, 190, 203–5, 216–21, 224–9, 234, 267 adjectival 205, 220–1, 225, 229



path 6, 7, 14, 97, 104–13, 115–23 pathway 14, 17, 19, 21 acquisition 5, 11, 16, 19 diachronic 3–5, 9, 11, 13–14, 18, 223, 255, 271, 273, 275, 276 oppositional 5, 13, 19–23 parallel 5, 13–18 patient 136 perfect 21, 62–5, 67, 68, 71, 73, 129, 133, 134, 216–21, 224, 227, 228, 234, 264–7 peripheral rule 77–9, 89, 91, 92 Persian 55, 126 persistence 8, 202–6, 215 phase 147–8, 242 Phonological Form 119, 144, 228, 237, 240–2, 246–7, 249, 250, 252, 253 phonological phrase 237, 240–1, 246 piedpiping 28, 30 place 6, 7, 112, 113, 119 pluperfect 216–19, 228, 265 Portuguese 147, 264 European 234 Old 157, 180 possessor 37, 67, 71, 97, 136, 244, 246, see also agreement, possessor; case, genitive postposition 44–8, 52, 57, 60 preposition 5, 7, 28–32, 37, 44–7, 52, 56, 97, 98, 103–8, 110–22, 162, 204, 205, 242 prestige 12 presupposition 200, 267, 268 primary linguistic data 13, 79, 80 priming 8, 202, 204–6, 210–15 Principles & Parameters 83 processing 17, 25, 31–3, 35, 204 productivity 16–19, 22–4, 46, 72, 107–10, 115, 116, 118, 123, 130, 142, 150, 178, 179, 181, 232 prosodic hierarchy 230 quantifier 12, 14–16, 18, 153, 154, 190, 271, 277, 278 scope 264–8 quotative 12 reanalysis 3, 5, 6, 11, 15, 19–23, 25, 29, 35–9, 41, 45–9, 60, 61, 66, 67, 70, 72, 73, 92, 97, 104, 107, 109, 111, 116, 118–24, 128, 132, 135, 137, 140, 144, 237, 238, 249–53, 271, 277, 278 recipient 6, 97–103, 136 reflexive 80, 87, 88, 90, 108, 114, 146, 150–2, 156, 165, 216, 218 regularization 11, 17, 252

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i



Index

relative clause 27–40, 45, 56–8, 99, 131, 154, 164, 181, see also complementizer, relative; relative specifier adjoined 29 correlative 26, 29 appositive 32, 36, 149 dependent 5, 25–41 demonstrative 181 free 26–7, 29, 30, 35–7, 40, 41 relative specifier 25–34, 40, 41 renewal 18, 22 resultative 6, 104–9, 114–18, 122, 123, 220 particle 61, 68, 70, 73, see also particle, verbal resumption 32, 33, 56 Romance languages 6–8, 101, 103, 106, 107, 145, 176, 216–18, 220–9, 233–4, see also Catalan, Florentine, French, Italian, Latin, Portuguese, Spanish Medieval 145, 157 Old 152, 217, 234 Romani 72 root phenomena 7, 173–4, 178–80, 209, 267, see also topicalization Russian 55, 126, 127, 132 S-curve 6, 34, 81–5, 94–6, 100, 101, 118, 208 Sakha 143 satellite-framed 6, 104–18, 123 scrambling 7, 8, 98, 157, 159–61, 167, 189, 201, 240–2 semantic change 5, 11, 13–19, 21–4, 26, 109, 113, 121, 123, 124, 217, 220–1, 224–5, 233, 234, 278, see also bleaching serial verb construction 45, 48 Sino-Tibetan languages 45, see also Bai, Chinese, Hkongso Slavic languages 6, 61, 70, 72, 126, 230, see also Russian Spanish 55, 101, 255, 267 Old 153, 157, 158, 234 Puerto Rican 203 Specific Language Impairment 13 specificity 126, 131 spell-out 244 Sprachbund 72 stability 75, 76, 88, 184–6, 201, 237, 245, 248, 250, 252, 254 state 22, 114, 133, 269 result 62, 69, 71, 105, 107, 108, 110, 118, 123 Stylistic Fronting 7, 146–54, 158, 160, 166, see also Leftward Stylistic Displacement subinformativity 8, 182, 197, 200

subject condition 148, 153–7 subjunctive 21, 36, 37, 68, 71, 175, 178, 179, 218–20, 222, 233, 258, 259, 263 Swedish 174 synthetic 97, 99, 216, 217, 219, 220, 228, 233, 234 tense 6, 14, 19–20, 61–8, 71–3, 207, 216, 221, 242–3 aorist 133 complex 6, 61–4, 66–8, 71–3, 264 future 8, 14, 17, 21, 22, 133, 134, 204, 216–19, 227, 228, 233, 234, 259, 266, 267, 270 infectum 8, 216, 217, 219, 221, 228, see also imperfect; tense, future; tense, present mismatch 216–20, 224–6 nonpast 6, 61–3, 72, 73, 220 past 6, 19, 61–9, 71–3, 126, 127, 134, 142, 216, 217, 225, 226, 244, 256, 258, 259, 261, 265–70, 272, 276, see also participle, past perfectum 8, 216–20, 228, see also perfect, pluperfect present 21, 62, 67, 68, 71, 73, 127, 133, 143, 216, 217, 219–22, 224–6, 228, 229, 231, 232, 234, 266, 267 Thai 55 thematic role 97, 129, 255, 257, 264, see also agent, benefactive, goal, patient, possessor, recipient, theme theme 6, 46, 69, 97–103, 105, 129, 139 Theory of Mind 18 third construction 261, 269 Tibetan 48 Tongan 32 topic 131, 147, 149, 161, 163–6, 189, 195, 196, see also topicalization aboutness 189, 195 contrastive 182, 189, 195–7, 200, 201, see also contrast Top+P 147–9, 153, 160 topicalization 98, 158, 165, 169, 172, 173, 182–6, 188–94, 200–1, 259 transient state 6, 76, 80, 75–80, 83–91, see also failed change transitive 46–8, 70, 114, 129, 130, 135–7, 139, 140, 143, 226, 227 transmission 38, 76–8, see also diffusion Turkic languages 6, 64–7, see also Chuvash, Sakha, Turkish Old Turkic 61, 64–8 West 64–5

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 10/6/2017, SPi i

i

Index Turkish 16, 126 typology, statistical 43, 49 Udmurt 65 Ugric languages 66, 68, see also Finno-Ugric languages Proto-Ugric 68 unaccusative 70, 111, 176, 226–8 unergative 70, 111, 226, 227 Universal Grammar 43, 44, 59, 77–80, 207 Uralic languages 6, 61, 63–6, 72, see also Finno-Ugric languages variation 1, 3, 8, 9, 13, 17, 32, 34, 44, 48, 72, 85, 91–3, 95, 99, 101, 107, 118, 121–3, 171, 189, 193, 203–4, 210, 222, 223, 229, 230, 237, 243, 245–7, 252–4 verb-framed 6, 104–5, 107–9, 115–23 Vietnamese 55 voice 7, 136, 139, 140, 142, 143, 217, 218, 220, 224, 228, 270, see also passive



active 63, 140–4, 203, 204 mediopassive 6, 126, 136, 139–44, 227 Welsh 244 word order 5–8, 43–59, 78, 98–9, 123, 145, 148, 154, 158–62, 164, 171, 172, 180, 204, 216, 217, 221–34 crosscategorial generalization 5, 43–5, 59–60 NP-internal 43–5, 48, 49, 53, 54, 56–9 verb-first 127, 172 verb-second 7–8, 145–7, 150, 153, 158, 161, 162, 166, 168–73, 176, 179–84, 188, 189, 191, 201, 263, 274–6 asymmetric 7, 145, 168–73 embedded 7, 147, 162, 168–70, 172–81 verb-third 7–8, 146, 170–2, 183 Yiddish 72, 169, 170, 181, 192–3

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 13/6/2017, SPi i

i

OX F O R D S T U D I E S I N D IAC H R O N IC A N D H I S T O R I C A L L I N G U I S T I C S general editors: Adam Ledgeway and Ian Roberts, University of Cambridge advisory editors: Cynthia Allen, Australian National University; Ricardo Bermúdez-Otero, University of Manchester; Theresa Biberauer, University of Cambridge; Charlotte Galves, University of Campinas; Geoff Horrocks, University of Cambridge; Paul Kiparsky, Stanford University; Anthony Kroch, University of Pennsylvania; David Lightfoot, Georgetown University; Giuseppe Longobardi, University of York; George Walkden, University of Konstanz; David Willis, University of Cambridge Published 1 From Latin to Romance Morphosyntactic Typology and Change Adam Ledgeway 2 Parameter Theory and Linguistic Change Edited by Charlotte Galves, Sonia Cyrino, Ruth Lopes, Filomena Sandalo, and Juanito Avelar 3 Case in Semitic Roles, Relations, and Reconstruction Rebecca Hasselbach 4 The Boundaries of Pure Morphology Diachronic and Synchronic Perspectives Edited by Silvio Cruschina, Martin Maiden, and John Charles Smith 5 The History of Negation in the Languages of Europe and the Mediterranean Volume I: Case Studies Edited by David Willis, Christopher Lucas, and Anne Breitbarth 6 Constructionalization and Constructional Changes Elizabeth Traugott and Graeme Trousdale 7 Word Order in Old Italian Cecilia Poletto 8 Diachrony and Dialects Grammatical Change in the Dialects of Italy Edited by Paola Benincà, Adam Ledgeway, and Nigel Vincent 9 Discourse and Pragmatic Markers from Latin to the Romance Languages Edited by Chiara Ghezzi and Piera Molinelli

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 13/6/2017, SPi i

i

10 Vowel Length from Latin to Romance Michele Loporcaro 11 The Evolution of Functional Left Peripheries in Hungarian Syntax Edited by Katalin É. Kiss 12 Syntactic Reconstruction and Proto-Germanic George Walkden 13 The History of Low German Negation Anne Breitbarth 14 Arabic Indefinites, Interrogatives, and Negators A Linguistic History of Western Dialects David Wilmsen 15 Syntax over Time Lexical, Morphological, and Information-Structural Interactions Edited by Theresa Biberauer and George Walkden 16 Syllable and Segment in Latin Ranjan Sen 17 Participles in Rigvedic Sanskrit The Syntax and Semantics of Adjectival Verb Forms John J. Lowe 18 Verb Movement and Clause Structure in Old Romanian Virginia Hill and Gabriela Alboiu 19 The Syntax of Old Romanian Edited by Gabriela Pană Dindelegan 20 Grammaticalization and the Rise of Configurationality in Indo-Aryan Uta Reinöhl 21 The Rise and Fall of Ergativity in Aramaic Cycles of Alignment Change Eleanor Coghill 22 Portuguese Relative Clauses in Synchrony and Diachrony Adriana Cardoso 23 Micro-change and Macro-change in Diachronic Syntax Edited by Éric Mathieu and Robert Truswell

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, 13/6/2017, SPi i

i

In preparation Negation and Nonveridicality in the History of Greek Katerina Chatzopoulou Morphological Borrowing Francesco Gardani Nominal Expressions and Language Change From Early Latin to Modern Romance Giuliana Giusti Grammaticalization from a Typological Perspective Edited by Bernd Heine, Heiko Narrog, and Prashant Pardeshi The Historical Dialectology of Arabic: Linguistic and Sociolinguistic Approaches Edited by Clive Holes A Study in Grammatical Change The Modern Greek Weak Subject Pronoun τ oς and its Implications for Language Change and Structure Brian D. Joseph Gender from Latin to Romance Michele Loporcaro Reconstructing Pre-Islamic Arabic Dialects Alexander Magidow Word Order Change Edited by Anna Maria Martins and Adriana Cardoso Quantitative Historical Linguistics Barbara McGillivray and Gard Jenset Grammaticalization from a Typological Perspective Heiko Narrog and Bernd Heine Syntactic Change and Stability Joel Wallenberg The History of Negation in the Languages of Europe and the Mediterranean Volume II: Patterns and Processes Edited by David Willis, Christopher Lucas, and Anne Breitbarth

i

i i

i