The Rise of Agreement: A Formal Approach to the Syntax And Grammaticalization of Verbal Inflection (Linguistik Aktuell Linguistics Today)

The Rise of Agreement Linguistik Aktuell/Linguistics Today Linguistik Aktuell/Linguistics Today (LA) provides a pla...

Author: Eric Fuss

63 downloads 851 Views 4MB Size Report

This content was uploaded by our users and we assume good faith they have the permission to share this book. If you own the copyright to this book and it is wrongfully on our website, we offer a simple DMCA procedure to remove your content from our site. Start by pressing the button below!

Report copyright / DMCA form

DOWNLOAD PDF

The Rise of Agreement

Linguistik Aktuell/Linguistics Today Linguistik Aktuell/Linguistics Today (LA) provides a platform for original monograph studies into synchronic and diachronic linguistics. Studies in LA confront empirical and theoretical problems as these are currently discussed in syntax, semantics, morphology, phonology, and systematic pragmatics with the aim to establish robust empirical generalizations within a universalistic perspective.

Series Editors Werner Abraham

Elly van Gelderen

University of Vienna

Arizona State University

Advisory Editorial Board Cedric Boeckx

Ian Roberts

Harvard University

Cambridge University

Guglielmo Cinque

Ken Saﬁr

University of Venice

Rutgers University, New Brunswick NJ

Günther Grewendorf

Lisa deMena Travis

J.W. Goethe-University, Frankfurt

McGill University

Liliane Haegeman

Sten Vikner

University of Lille, France

University of Aarhus

Hubert Haider

C. Jan-Wouter Zwart

University of Salzburg

University of Groningen

Christer Platzack University of Lund

Volume 81 The Rise of Agreement: A formal approach to the syntax and grammaticalization of verbal inﬂection by Eric Fuß

The Rise of Agreement A formal approach to the syntax and grammaticalization of verbal inﬂection

Eric Fuß Johann-Wolfgang-Goethe University

John Benjamins Publishing Company Amsterdam/Philadelphia

8

TM

The paper used in this publication meets the minimum requirements of American National Standard for Information Sciences – Permanence of Paper for Printed Library Materials, ansi z39.48-1984.

Library of Congress Cataloging-in-Publication Data Fuß, Eric. The rise of agreement : a formal approach to the syntax and grammaticalization of verbal inﬂection / Eric Fuß. p. cm. (Linguistik Aktuell/Linguistics Today, issn 0166–0829 ; v. 81) Includes bibliographical references and index. 1. Grammar, Comparative and general--Agreement. 2. Grammar, Comparative and general--Verb. 3. Grammar, Comparative and general-Grammaticalization. 4. Historical linguistics. 5. Grammar, Comparative and general--Inﬂection. I. Title. II. Series. P299.A35 F87 2005 415 22--dc22 isbn 90 272 2805 1 (Hb; alk. paper)

2005049336

© 2005 – John Benjamins B.V. No part of this book may be reproduced in any form, by print, photoprint, microﬁlm, or any other means, without written permission from the publisher. John Benjamins Publishing Co. · P.O. Box 36224 · 1020 me Amsterdam · The Netherlands John Benjamins North America · P.O. Box 27519 · Philadelphia pa 19118-0519 · usa

JB[v.20020404] Prn:18/07/2005; 14:55

F: LA81CO.tex / p.1 (46-100)

Table of contents

Acknowledgements

ix

Notes for the reader and list of abbreviations

xi

Chapter 1 Introduction 1.1 The grammaticalization of verbal agreement markers 2 1.2 Previous accounts 5 1.2.1 NP-detachment (Givón 1976) 6 1.2.1.1 The suffixing preference 7 1.2.1.2 The pioneering role of 1st and 2nd person 9 1.2.1.3 Incompatibility between verbal agreement and overt subjects 10 1.2.2 Accessibility theory (Ariel 2000) 12 1.2.3 Reanalysis of focus shells (Simpson & Wu 2002) 13 1.2.4 Section summary 15 1.3 Outline of the Book 16

1

Chapter 2 Theoretical preliminaries 2.1 Introduction 23 2.2 Minimalist syntax 23 2.3 Distributed Morphology 28 2.4 Language change: A generative perspective 2.5 Grammaticalization 40 2.6 Summary 49

23

33

Chapter 3 The structural design of agreement 3.1 Introduction 55 3.2 The phrase-structural representation of agreement 56 3.2.1 Conceptual arguments 58 3.2.2 Evidence from word order facts 59 3.2.3 The morphological realization of agreement 61

55

JB[v.20020404] Prn:18/07/2005; 14:55



F: LA81CO.tex / p.2 (100-140)

Table of contents

3.3 3.4 3.5 3.6

The structural relation involved in feature matching 67 Syntactic approaches to word formation 75 Interim summary: The design of agreement 80 Toward a realizational theory of agreement 81 3.6.1 The structural representation of agreement morphemes 81 3.6.2 Feature matching under closest c-command 83 3.6.3 Agreement and word formation 89 3.6.4 Section summary 93 3.7 Multiple agreement: Inflected complementizers in Germanic 94 3.7.1 Previous accounts of complementizer agreement 98 3.7.1.1 Infl-to-C movement 98 3.7.1.2 AgrP within CP 101 3.7.1.3 Multiple Agree 104 3.7.2 A Late Insertion account of complementizer agreement 105 3.8 Conclusion 114 Chapter 4 The transition from pronoun to inflectional marker 4.1 Introduction 129 4.2 Telling apart clitics and agreement markers 130 4.2.1 Syntactic criteria 130 4.2.2 Morphological criteria 135 4.3 Syntactic preconditions for the rise of agreement 139 4.4 Paths toward agreement I: Infl-oriented clitics 143 4.5 Paths toward agreement II: C-oriented clitics 146 4.6 Summary 151

129

Chapter 5 The reanalysis of C-oriented clitics 5.1 Introduction 157 5.2 Bavarian 158 5.2.1 The diachronic development of Agr-on-C in Bavarian 162 5.2.1.1 2sg /-st/ 162 5.2.1.2 2pl /-ts/ 164 5.2.1.3 1pl /-ma/ in Lower Bavarian 165 5.2.2 Clitics, V2, and the rise of agreement 167 5.2.3 Developments in other German varieties 171 5.2.3.1 Cimbrian 172 5.2.3.2 Walser dialects 175 5.2.4 Section summary 177 5.3 Rhaeto-Romance 179 5.3.1 A grammatical sketch of the Swiss Rhaeto-Romance dialects

157

179

JB[v.20020404] Prn:18/07/2005; 14:55

F: LA81CO.tex / p.3 (140-184)

Table of contents 

5.3.1.1 Word order: V2 and SVO 180 5.3.1.2 The distribution of pronominal forms and pro-drop 183 5.3.2 Earlier grammaticalization processes affecting enclitic pronouns 185 5.3.2.1 1st person forms 185 5.3.2.2 2nd person forms 187 5.3.3 Clitic doubling and the rise of agreement 190 5.3.4 The reanalysis of emphatic doubling structures 193 5.3.4.1 Syntactic restrictions on Agr-on-C 199 5.3.5 Section summary 201 5.4 Reanalysis of C-oriented clitics in non-V2 languages 203 5.4.1 Uto-Aztecan 203 5.4.2 Mongolian 209 5.5 Conclusion 214 Chapter 6 Morphological blocking and the rise of agreement 6.1 Introduction 229 6.2 The Blocking Principle 231 6.2.1 The rise of new verbal agreement endings in Bavarian 232 6.2.1.1 The diachrony of 2sg /-st/ 235 6.2.2 Morphological blocking versus analogical leveling 237 6.2.3 On the status of morphological doublets 242 6.2.4 Section summary 246 6.3 The pioneering role of 1st and 2nd person 247 6.4 French 249 6.5 Northern Italian dialects 256 6.5.1 Piattino 257 6.5.2 Vicentino 260 6.5.3 Section summary 268 6.6 Rhaeto-Romance 269 6.7 Language loss and the grammaticalization of agreement markers 275 6.8 Grammaticalization and multiple agreement in Skou 279 6.8.1 Strategies of agreement marking in present-day Skou 279 6.8.2 The historical origin of multiple agreement marking in Skou 282 6.9 Conclusion 283

229

Chapter 7 Concluding summary

299

References

305

Index

325

JB[v.20020404] Prn:2/08/2005; 9:33

F: LA81AC.tex / p.1 (44-118)

Acknowledgements

This book is a revised version of my doctoral dissertation, submitted to the University of Frankfurt in November 2004 and defended in February 2005. I would like to use this opportunity to express my gratitude to a number of people who have contributed in various ways to the completion of this work. First, I would like to thank my thesis supervisor Günther Grewendorf for his confidence and support over the years. He originally attracted me to theoretical linguistics and without him, I probably would never have discovered the intellectual challenge and pleasures of studying human language. Likewise, I am grateful to my second supervisor Ian Roberts, whose work on diachronic syntax and in particular the phenomenon of grammaticalization provided the initial stimulus for this book. His ideas had a profound influence on my work and I benefited greatly from his suggestions and comments on earlier versions of this book. I would also like to thank the other members of my thesis committee, Helen Leuninger, Jörg Meibauer, and in particular Ede Zimmermann, who readily agreed to act as a third supervisor when it became necessary. Furthermore, I would like to express my gratitude to a number of people who provided comments, support and advice and reassured me that the rise of agreement is a topic worth writing a book about: Artemis Alexiadou, Mira Ariel, Ermenegildo Bidese, Theresa Biberauer, Adriana Castagna, Greville Corbett, Mark Donohue, Hans-Martin Gärtner, Germen de Haan, Eric Haeberli, Liliane Haegeman, Katharina Hartmann, Ruth Kempson, Anthony Kroch, Tom McFadden, Cian and Melanie McLoughlin, William Morris, Peter Öhl, Albert Ortmann, Cecilia Poletto, Henk van Riemsdijk, Joachim Sabel, Wolfgang Schulze, Peter Sells, Halldór Sigurðsson, Andrew Simpson, Markus Steinbach, Carola Trips, Helmut Weiß, Gunther De Vogelaer, Hedde Zeijlstra, and Jochen Zeller. Special thanks go to my colleagues, both linguist and non-linguist, who accompanied the race to finish this work and who provided constant support and encouragement during the whole of the last year: Patrick Brandt, Ortrud Bruchelt, Cécile Meier, and Shin-Sook Kim. Furthermore, I am particularly grateful to Werner Abraham and Elly van Gelderen, who not only provided detailed comments on this work, but also gave me the opportunity to have it published in the series Linguistik Aktuell. To be sure, this name list is not exhaustive, and I apologize to those who I should have mentioned here, but failed to do so. At this point, I would also like to give my warmest thanks to my parents and my grandma who always supported and encouraged me without a second thought in everything I have ever done.

JB[v.20020404] Prn:2/08/2005; 9:33



F: LA81AC.tex / p.2 (118-122)

Acknowledgements

Finally, I think I would never have completed this book it if it had not been for the three A’s in my life, Anne, Adrian and Anton, whose love, support and friendship provided constant comfort and enabled me to carry on when things didn’t look that bright. This work is dedicated to you.

JB[v.20020404] Prn:18/07/2005; 14:57

F: LA81NO.tex / p.1 (46-185)

Notes for the reader and list of abbreviations

In the glosses to the linguistic examples cited in this work, the following notational conventions are employed to distinguish between different kinds of bound (inflectional) markers: discrete inflectional formatives are separated by a hyphen, whereas a clitic boundary is marked by ‘=’. Fused marking, that is, instances where a given form realizes more than a single (inflectional) feature/meaning, is indicated by a dot separating the individual features/meanings. Further abbreviations used in the glosses to the linguistic examples are listed below. abs absolutive acc accusative agr agreement agro object agreement agrs subject agreement aor aorist ap antipassive asp aspect caus causative cl noun class clit clitic comp complementizer cond conditional dat dative emph emphatic particle erg ergative fem feminine foc focus (marker) fut future tense imperf imperfect indic indicative infl inflection masc masculine modal modal verb modprt modal particle

neg negation neut neuter nmlz nominalization nom nominative object object (marker) part participle pass passive past past tense perf perfect pl plural poss possessive pron pronoun prt particle prev preverb prog progressive pstprt past participle pstwit past witnessed pres present tense qprt question particle qmarker question marker refl reflexive pronoun sg singular subjnc subjunctive t tense wh wh-pronoun/wh-question marker

JB[v.20020404] Prn:18/07/2005; 15:02

F: LA8101.tex / p.1 (45-120)

Chapter 1

Introduction

The goal of this book is to explore the diachronic development of subject-verb agreement, focusing on the reanalysis of pronominal clitics as inflectional formatives. It is shown that pronominal clitics can evolve into agreement morphology in quite a variety of different syntactic environments, contra common beliefs prevalent in the literature on this topic (cf. e.g. Givón 1976; Siewierska 1999). However, it is a central claim of this study that the apparent diversity is only superficial whereas a closer inspection reveals a well-defined set of underlying syntactic and morphological conditions which are operative during language acquisition and shape the grammaticalization process in question across languages. These abstract conditions are shown to derive from the way verbal agreement is established in natural languages by a combination of syntactic mechanisms which value the content of agreement heads and morphological principles governing the spell-out of the resulting feature combinations. In this scenario, the grammaticalization of inflectional morphology is taken to reflect a division of labor between syntax and morphology in which the former imposes a set of necessary conditions on the reanalysis of pronominal elements while the ultimate trigger of this change is morphological in nature. More specifically, it is demonstrated that the reanalysis of subject clitics is triggered by morphological blocking effects which operate during language acquisition and ensure that the most specific potential exponent of a given inflectional head is selected and stored in the Lexicon. This hypothesis is motivated by the observation that cross-linguistically, the grammaticalization process under discussion either establishes agreement in languages that previously lacked agreement or serves to repair a defective paradigm (cf. Chapter 6 for details): (1) New verbal agreement morphology arises historically only for those slots of the agreement paradigm where the existing verbal inflection is nondistinctive. The theoretical framework adopted in this book combines the recent version of minimalist syntax devised in Chomsky (2000, 2001a, 2001b) with a Late Insertion approach to morphology (Distributed Morphology). Thus, it is assumed that syntactic heads are provided with phonological exponents post-syntactically (cf. e.g. Halle & Marantz 1993, 1994; Noyer 1997). This introductory chapter is organized as follows. Section 1.1 illustrates the basic properties of the diachronic phenomenon under consideration. Section 1.2 provides a critical discussion of previous analyses of the grammaticalization of verbal agreement ranging from the seminal paper by Talmy Givón (1976) to a recent approach to

JB[v.20020404] Prn:18/07/2005; 15:02



F: LA8101.tex / p.2 (120-207)

The Rise of Agreement

the topic by Simpson and Wu (2002). In Section 1.3, I give an outline of this book, sketching out the main theoretical proposals, both synchronic and diachronic, that are developed in the chapters to come. Before introducing the subject matter of this study in some more detail, let me establish some basic descriptive terminology that will be used throughout this work (basically following Corbett 1998, 2003). The element which determines agreement is referred to as the controller. The element which changes its form as a result of the agreement relation is called the target. The properties in which controller and target exhibit covariance (i.e., agreement) are called agreement features. The phonological realization of agreement features is called the marker or exponent of the relevant agreement features. Finally, the agreement relation is typically restricted to a certain local syntactic environment, which is referred to as the domain of the agreement relation.1

. The grammaticalization of verbal agreement markers It is a long-standing observation in historical linguistics that verbal agreement morphology develops from (originally independent) personal pronouns. This diachronic relation was first established for the person and number endings of the early IndoEuropean languages by 19th century grammarians such as Bopp (1816), Müller (1875), or Brugmann (1916) when early work in historical linguistics sought to establish the sounds and the lexical inventory of a (hypothetical) common ancestor of the Indic, Germanic and Romance language families, now commonly referred to as Proto-Indo-European (PIE). In the 20th century, the basic insight has been elaborated in the work of Indo-Europeanists such as Meillet (1912), Kuryłowicz (1964), Watkins (1969), and Szemerényi (1989) (cf. W. Lehmann 1993 for an overview). The diachronic relation between agreement morphemes and personal pronouns in Indo-European can be detected from a superficial inspection of the shape of pronouns and agreement morphemes.2 Table 1 lists the forms that are usually reconstructed for the set of personal pronouns (nominative and accusative) and the primary and secondary agreement endings for the present active (and aorist), taken from Szemerényi (1989: 228, 248).3 Table 1. Reconstructed pronouns and agreement suffixes of Proto-Indo-European.

1sg 2sg 3sg 1pl 2pl 3pl

Nominative

Pronouns Accusative

Verbal agreement Primary Secondary

*eg(h)om, *eg¯o *t¯u *tu – *wei, *nsmés *y¯us, *usmés/uswes –

*(e)me, *m¯e, *m¯em *twe/*te, *tw¯e/*t¯e *tw¯em/*t¯em – *nes/*nos, *n¯es/*n¯os,*nsme *wes/*wos; *w¯es/*w¯os *usme, *uswes –

*-mi *-si *-ti *-mes *-tes *-nti

*-m *-s (<*-t) *-t *-mes *-tes *-nt

JB[v.20020404] Prn:18/07/2005; 15:02

F: LA8101.tex / p.3 (207-277)

Chapter 1. Introduction

Table 2. Pronouns and absolutive/ergative agreement in Basque.

1sg 2sg 1pl 2pl

Pronouns absolutive ergative

Absolutive agreement ‘go’

Ergative agreement ‘have’

ni su gu súe-k

n-u s-us g-us s-us-e

d-itu-t d-itu-su d-itu-gu d-itu-sue

ni-k su-k gu-k súe-k

‘I have them’ ‘you have them’ ‘we have them’ ‘you have them’

The similarity between personal pronouns and agreement suffixes is particularly clear in the 1st person forms (see Szemerényi 1989: 229 for the claim that 1sg *-m developed from the nominative form *eg(h)om consisting of a particle eg(h)- which was prefixed to the pronominal stem -om). At first sight, no such similarity holds between the respective 2nd person forms. However, it is possible to construct a pronominal origin for the 2nd person agreement endings as well (cf. e.g. Szemerényi 1989: 228ff., 359ff. for details).4 Similar observations can be made in present-day languages such as Basque where the verbal affixes signaling agreement with (absolutive and ergative) 1st and 2nd person arguments bear a clear resemblance with the respective pronouns (cf. e.g. Arregi 1999).5 As can be seen from Table 2, the prefixes of the absolutive paradigm are identical with the onset of the corresponding absolutive pronouns. Apart from the form for 1sg, the suffixes marking agreement with ergative subjects are even more similar to the relevant pronouns, since they preserve even the vowel of the 2nd person and 1pl pronouns. The study of non-European languages and language families led to the discovery that such similarities can be observed cross-linguistically, leading to the insight that in general, pronouns are the primary source of bound person/number markers (cf. e.g. Humboldt 1822, 1836; Meinhof 1936; Greenberg 1966b, 1978; Givón 1971, 1976; Lehmann 1988, 2002; Hopper & Traugott 1993; Corbett 1995; Siewierska 1999, 2004). For example, Moscati et al. (1969) note that in the Semitic languages, there are systematic similarities between agreement markers and personal pronouns.6 Similar observations can be made in various Bantu languages such as Swahili (Givón 1976) or Chichewa (Bresnan & Mchombo 1987). Andersen (1988) demonstrates that in Pari, a Western Nilotic language spoken in the Sudan, the pronominal origin of the proclitic agreement markers is still transparent in the present-day language. The Caucasian languages represent a particularly interesting case since the rise of new (person) agreement markers goes hand in hand with a change from an absolutive/ergative to nominative/accusative alignment (cf. e.g. Harris 1994; Schulze 1998). Poppe (1954), (1960), and Comrie (1980) observe that a number of Mongolian languages (including Buryat, Kalmyk, Dagur and Moghol) have grammaticalized verbal agreement suffixes from pronouns quite recently.7 This seems also to be true of several Tibeto-Burman languages (such as Mao Naga, Tangut, Mikir, and Sgaw Karen, cf. LaPolla 1994; Siewierska 2004). Burenhult (2002) shows that the preverbal subject agreement markers found in the Mon-Khmer language Jahai clearly resemble the relevant pronominal



JB[v.20020404] Prn:18/07/2005; 15:02



F: LA8101.tex / p.4 (277-337)

The Rise of Agreement

forms. The change from clitic pronouns to agreement markers can also be observed in many Australian (Hale 1973; Dixon 1980, 2002; Nordlinger 1998) and Papuan languages (Donohue 2002). In the Austronesian family, relevant examples come from Kisar (Blood 1992), a set of languages spoken on Sulawesi (Taratán, Uma, and Tukang Besi, cf. Cysouw 2003b), and the Oceanic language Asumboa (mentioned in Siewierska 2004). Steele (1977, 1995) claims that second position clitics have developed into prefixal agreement markers in a number of Uto-Aztecan languages. Related developments can be observed in many other native American languages (cf. Mithun 1991 for an overview and a set of interesting generalizations). It is usually assumed that the historical development of verbal agreement markers from previously free pronouns universally follows the grammaticalization path sketched in (2) (cf. e.g. Lehmann 1988: 59ff., 2002; Corbett 1995; Hopper & Traugott 1993; Siewierska 2004: 262).8 (2) independent pronoun → weak pronoun → clitic pronoun → affixal (agglutinative) agreement marker → fused agreement marker → Ø Due to phonological erosion, a formerly independent referential pronoun turns into a phonologically weak clitic which needs to attach to a lexical host (typically the finite verb). At some point, this pronominal element undergoes even further reduction, loses its referential potential and is reanalyzed as an obligatory part of the verbal inflection which redundantly marks the featural content of an argument DP.9 Subsequently, the resulting agglutinative agreement marker may be subject to further phonological erosion and undergo fusion with other inflectional markers such as tense before it eventually disappears completely, restoring the stage that marked the beginning of the grammaticalization path in (2). Note that the cyclic character of this change is a typical trait of grammaticalization processes (the most prominent being Jespersen’s cycle). The last step of this historical development, that is, the loss of agreement marking, can be observed in many Germanic languages, most notably English and Mainland Scandinavian. The observation that the rise of agreement markers follows a universal historical pathway raises the question of whether it is possible to attribute the continuing reduction of pronouns to deeper factors. The present work adheres to the idea that the facts described by (2) are to be explained by the assumption that grammaticalization universally involves the reanalysis of lexical material as the phonological realization of functional categories/heads (Roberts & Roussou 2003). By assumption, the latter are inherently prosodically defective. Thus, phonological deficiency is viewed as a precondition for the reanalysis as a functional head. In addition, due to the inherent prosodical weakness of functional categories, further phonological reduction is promoted after the reanalysis has taken place (see Chapter 2, Section 4 for discussion). In many cases, the investigation of the historical processes in question is complicated by the fact that the agreement markers are of considerable antiquity. Thus, the relevant changes often predate the earliest written records of a language. This is particularly clear in the case of a reconstructed proto-language such as Proto-Indo-

JB[v.20020404] Prn:18/07/2005; 15:02

F: LA8101.tex / p.5 (337-391)

Chapter 1. Introduction

European, but also holds for cases such as Basque where the historical relation between pronouns and agreement markers is still transparent in the present-day language (in contrast to e.g. the Germanic family where the inherited agreement formatives bear no resemblance to the present-day pronouns). Therefore, instances where the relevant processes took place quite recently or can actually be witnessed as an ongoing process are of special interest. Accordingly, the data sample discussed in this book comes primarily from languages where the relevant changes are either still underway or have taken place during the recorded history of the language, focusing on non-standard varieties of Romance and Germanic. A well-known case in point is Non-Standard French, where subject clitics have developed into a new form of verbal agreement marking (cf. Kuen 1957; Wartburg 1970; Ashby 1977; Harris 1978; Lambrecht 1981; Auger 1993, 1994; Gerlach 2002, among others). The historical development of subject clitics into markers of verbal agreement is also attested in various Northern Italian and Rhaeto-Romance dialects (cf. e.g. Meyer-Lübke 1894; Rohlfs 1949; Spiess 1956; Kuen 1957; Renzi & Vanelli 1983; Rizzi 1986; Linder 1987; Brandi & Cordin 1989; Haiman 1991; Haiman & Benincà 1992; Poletto 1995, 1997, 1999, 2000; Roberts & Roussou 2003; Tortora 2003). Similar developments can be observed in a number of Germanic dialects, in particular Bavarian, where subject enclitics which attach to the finite verb in inversion contexts have evolved into new forms of (verbal) inflection (for Bavarian cf. Pfalz 1918; Lessiak 1963; Altmann 1984; Bayer 1984; Kollmer 1987; Wiesinger 1989; Weiß 1998, among others).10 Summing up, it appears that there is some consensus concerning the general path that is taken by the diachronic development of verbal agreement morphology from former pronouns. However, nothing has been said so far about syntactic aspects and possible triggers of the change in question. These questions are dealt with in the next section which considers a set of previous explanations for the grammaticalization of agreement markers.

. Previous accounts In this section, I briefly discuss three scenarios which have been proposed in the literature to explain the reanalysis of pronominal elements as verbal agreement markers. The major part of this section is devoted to the most influential of these proposals, the so-called NP-detachment hypothesis offered in Givón (1976). In addition, I discuss an approach in terms of Accessibility theory (Ariel 2000) and a recent proposal by Simpson and Wu (2002) which attributes the grammaticalization of agreement markers to the reanalysis of focus constructions as agreement configurations. Since the analysis developed in this book is partially based on the model of grammaticalization envisaged in Roberts and Roussou (2003), their account of the development of agreement markers will be dealt with in some more detail in Chapter 2 which introduces the key assumptions I will make use of.



JB[v.20020404] Prn:18/07/2005; 15:02



F: LA8101.tex / p.6 (391-461)

The Rise of Agreement

.. NP-detachment (Givón 1976) Current thinking about syntactic aspects of the historical development of predicateargument agreement is perhaps most substantially influenced by the work of Talmy Givón, most notably Givón (1971), (1976). Accordingly, Givón’s hypotheses are discussed in some more detail here. Givón claims that cross-linguistically, agreement markers arise historically from the reanalysis of resumptive pronouns which relate anaphorically to a left or right dislocated topic (so-called “NP-detachment”): [Joe]i , hei scored a goal. → topic pronoun b. Hei scored a goal, [Joe]i . → pronoun topic

(3) a.

Joe he-scored a goal. subject agr he-scored a goal, Joe. agr subject

This change goes hand in hand with another major reanalysis. Due to an over-use of the stylistically marked sentences involving NP-detachment (Givón 1976: 154 calls this process ‘de-marking’), the former dislocation structures are reinterpreted as the ‘neutral syntax’, with the former topic becoming the new subject. In other words, Givón assumes that the diachronic source of grammatical agreement is some form of ‘topicagreement’.11 The affixation of the former (free) resumptive pronoun to the verb is attributed to the fact that unstressed pronouns often cliticize to the verb, which facilitates a reanalysis as part of the verbal inflection. A similar process is taken to be the source of object agreement, with topicalized objects being reanalyzed as residing in their base (or case) position and resumptive clitics reanalyzed as markers of object agreement. An often discussed example of this scenario comes from Non-Standard French, where clitics have developed into obligatory elements which can be doubled by full phrases for reasons of emphasis (cf. e.g. Wartburg 1970; Ashby 1977). The relevant sentences preferably receive a basic, non-dislocated interpretation. Therefore, the subject clitics are probably better analyzed as clitic/prefixal agreement markers on the verb (see Chapter 6 for more arguments in favor of this analysis): (4) a.

(Moi) je porte la table. me clit.1sg carry the table ‘I carry the table.’ #‘As for me, I carry the table.’ b. Moi *(je) porte la table. me clit.1sg carry the table

(Gerlach 2002: 224)

The relation between the above examples and their diachronic origin as topic left dislocation structures is still obvious: on the surface, they are identical to the corresponding topic left dislocation construction of standard French. We can therefore conclude that topic left dislocation structures are apparently a viable syntactic environment for the historical development of subject-verb agreement. This scenario predicts that across languages, the realization of agreement is connected to the topicality of the agreement

JB[v.20020404] Prn:18/07/2005; 15:02

F: LA8101.tex / p.7 (461-548)

Chapter 1. Introduction

controller (since only topics can evolve into agreement controllers). And indeed, it appears that we do find traces of this particular historical origin of agreement markers. For example, if it is assumed that of all the verb’s arguments, the subject/agent is the most likely topic in a given utterance (cf. Givón 1976), Givón’s analysis provides an explanation for the observation that cross-linguistically, subject agreement is the most frequent instance of syntactic agreement. Furthermore, the fact that in many languages (Givón mentions Mandarin, Kinyarwanda and Malagasy as examples), subjects must be either definite or generic is taken to follow from the origin of these subjects as former topics, which are generally subject to similar restrictions (either definite or generic, but never referential-indefinite; Givón stresses that even in languages that tolerate indefinite subjects such as English, subjects tend to be definite and referential). In addition, it appears that in many languages, the possibility of object agreement is determined by the discourse status of the object: in general, object agreement is more likely to be triggered by topicalized DPs while focused objects often fail to control agreement (cf. Siewierska 2004). However, at a closer look, it appears that Givón’s analysis is not compatible with another set of cross-linguistic generalizations on the rise and realization of agreement. This suggests that there must exist alternative scenarios where pronouns can evolve into agreement markers, in contrast to Givón’s strong claim that the reanalysis of resumptive pronouns in NP-detachment constructions constitutes the only diachronic path to grammatical agreement (cf. e.g. Siewierska 1999: 240 for a related statement): One overriding theme – and claim – of this paper is that verb agreement paradigms always arise from anaphoric pronoun paradigms. (Givón 1976: 180)

... The suffixing preference Another claim put forward in Givón (1971, 1976) is that the position of agreement affixes relative to the verb stem reflects the position of the pronominal elements prior to the reanalysis as agreement markers. Given that subjects precede the verb in the vast majority of the world’s languages (i.e., in both SVO and SOV word order types), this should lead us to expect that subject agreement is much more frequently realized by prefixes than by suffixes. However, it has often been observed that cross-linguistically, there is a general tendency for inflectional markers to be realized as suffixes on the verb, sometimes referred to as the suffixing preference (cf. Sapir 1921; Greenberg 1963; Cutler, Hawkins, & Gilligan 1985; Hawkins & Gilligan 1988; Bybee et al. 1990; Siewierska & Bakker 1996; Julien 2002).12 Table 3 illustrates that the effects of the suffixing preference can also be observed if only agreement markers are taken into account (combining figures from Tables 2, 4, 7 and 8 from Bybee et al. 1990).13 From Table 3, we can see that in all word order types, suffixation is the most frequent morphological means for marking agreement on the verb. At a closer look, it becomes clear that the cross-linguistic suffixing preference is due to a set of independent factors. First, in SOV languages, the vast majority of agreement markers is realized as bound postverbal forms: prefixal agreement markers are found in only six



JB[v.20020404] Prn:18/07/2005; 15:02



F: LA8101.tex / p.8 (548-591)

The Rise of Agreement

Table 3. Position of agreement markers in 71 languages, by word order. Position of Agr

V-initial

Word order V-medial

V-final

Total

Agr V Agr+V V Agr V+Agr

0 11 (19%) 12 (20%) 36 (61%)

27 (10%) 103 (37%) 10 (4%) 137 (49%)

10 (4%) 80 (31%) 0 171 (65%)

37 (6%) 194 (32%) 22 (4%) 344 (58%)

of the 32 SOV languages examined by Bybee and her colleagues. Second, instead of exhibiting the mirror image of SOV languages, SVO languages employ roughly as many postverbal agreement markers as preverbal agreement markers, whereas verb-initial languages seem to pattern with SOV languages. In particular, there are many SVO languages that show only suffixal instead of prefixal verbal agreement markers (e.g. the Romance family).14 These findings contradict the prediction that all languages where subjects precede the verb should develop prefixal agreement morphology. Moreover, note that the very existence of agreement suffixes is problematic, given the fact that unstressed clitic pronouns, which are generally taken to be the source of agreement affixes, occur very rarely in post-verbal position (apart from verb-initial languages and inversion contexts). In contrast, they often tend to occupy a preverbal position rather high up in the clause, to the left of Infl (as in Romance) or in the C-domain (e.g., the Wackernagel clitics found in many Slavic languages). The problem raised by SOV languages is also discussed by Givón (1976: 180ff.) in connection with the suffixal agreement markers of (Proto-) Indo-European, which is usually reconstructed as an SOV language. Givón suggests that the agreement suffixes arose as a by-product of two major reanalyses that took place in early stages of IndoEuropean and started from a periphrastic (perfective) participial-stative construction which consisted of a non-agreeing participle followed by an inflected auxiliary that carried a prefixal agreement marker (originally arising from the reanalysis of a pronoun): (5) subjecti ... [[ participle ] agri +aux] In the course of time, the auxiliary cliticized to the participle. As a result, the prefixal agreement marker was ‘sandwiched’ between the verb and its host. Eventually, the inflected auxiliary was reanalyzed as suffixal agreement, leading from a periphrastic to a synthetic verb form. In an additional reanalysis, the passive construction turned into an active past tense form. Givón mentions other examples where this complex sequence of change events has apparently taken place (the suffixal perfect conjugation of Semitic and certain past tense forms in Aramaic). Furthermore, as is well-known, a somewhat similar development can be observed in Romance, where an inflected form of habere cliticized onto the non-finite main verb, giving rise to the synthetic future (Meillet 1912; Benveniste 1968; Roberts 1993b; Roberts & Roussou 2003). Thus, we cannot a priori exclude the possibility that this development took place in early IndoEuropean (still, there are open questions, e.g. concerning the origin of the agreement

JB[v.20020404] Prn:18/07/2005; 15:02

F: LA8101.tex / p.9 (591-644)

Chapter 1. Introduction

prefixes on the auxiliary). However, since the person endings are already present in the earliest records and are reconstructed for Proto-Indo-European as well, a closer understanding of the changes that actually gave rise to them presumably cannot be achieved and any account must remain speculative.15 Crucially, however, it is rather obvious that the development envisaged by Givón cannot account for all instances of suffixal agreement in SOV languages. Moreover, it cannot account for suffixal agreement in SVO languages (at least without further assumptions), since the auxiliary precedes the main verb in periphrastic constructions. Therefore, we must explore alternative paths that may lead to suffixal agreement in order to account for its predominance across languages (cf. Chapter 4 and Chapter 5 below for some discussion).16

... The pioneering role of 1st and 2nd person If it is assumed that verbal agreement marking develops from the reanalysis of resumptive pronouns in NP-detachment structures, we predict that new agreement markers should develop initially for 3rd person, given that the vast majority of left- or rightdislocated elements are 3rd person DPs. Accordingly, the development of 1st and 2nd person agreement should result from a later development, presumably driven by analogy.17 However, this prediction is not borne out by the facts. Across the world’s languages, we can observe that the manifestation of verbal agreement is sensitive to person distinctions. More specifically, there are many languages where verbs agree with 1st and 2nd person subjects, but fail to do so with 3rd person subjects (cf. e.g. Bybee 1985; Mithun 1991; Cysouw 2003a; Siewierska 2004). Relevant examples come from languages such as Turkish (no verbal agreement for 3sg, Kornfilt 1990), the Nilotic language Turkana (Dimmendaal 1983), the Mongolian language Buryat (Poppe 1960; Comrie 1980), Pashto (no number agreement with 3rd person subjects, MacKenzie 1990), the Australian language Wambaya (no 3sg form, Nordlinger 1998), a number of Tibeto-Burman languages (LaPolla 1992), and many native languages of North America (e.g. the Yuman and Siouan families, Mithun 1991).18 The observation that verbal agreement marking for 1st and 2nd person subjects is much more common than for 3rd person subjects is also corroborated by quantitative studies. Bybee (1985) notes that 54% of the languages in her sample which manifest agreement on the verb do not mark agreement with 3rd person arguments. Moreover, in those languages which exhibit a full paradigm of person agreement markers, it can often be shown that the grammaticalization of 1st and 2nd person agreement markers predated the grammaticalization of 3rd person forms, in contrast to what is predicted by Givón’s analysis. Good examples of this chronological difference are provided by many native American languages (cf. Mithun 1991 for an overview).19 We can therefore conclude that the generalization in (6) describes a characteristic of the grammaticalization of bound person agreement markers across languages.20 (6) Sequence of the development of person agreement marking 1st and 2nd person markers become bound to the verb before 3rd person markers.



JB[v.20020404] Prn:18/07/2005; 15:02



F: LA8101.tex / p.10 (644-697)

The Rise of Agreement

Similar person restrictions can be observed in languages where the grammaticalization of new verbal agreement markers took place quite recently and which are discussed in more detail in this book such as Bavarian, Non-Standard French, and various Swiss Rhaeto-Romance dialects. Therefore, we can conclude that we must look for alternative syntactic scenarios apart from NP-detachment which may license the diachronic development of verbal agreement markers, at least for those languages that have failed to develop 3rd person agreement markers historically.

... Incompatibility between verbal agreement and overt subjects Some languages exhibit curious restrictions on the co-occurrence of overt subjects and full-fledged verbal agreement. A well-known instance are the Celtic languages. In Welsh, for example, the verb agrees in person and number with pronominal subjects, but fails to do so with full subject DPs (cf. e.g. Roberts 1999b; Siewierska 2004). (7) Welsh a. Gwel-sant (hwy) y ferch. see-3pl.past they the girl ‘They saw the girl.’ b. *Gwel-sant y plant y ferch. see-3pl.past the children the girl ‘The children saw the girl.’ c. Gwel-odd y bachgen/bechgyn y ferch. see-3sg.past the boy/boys the girl ‘The boy/boys saw the girl.’ (Siewierska 2004: 152) The examples in (7) show that the manifestation of verbal agreement depends on the choice of subject. While the verb agrees in person and number with a pronominal subject, as in (7a), it must carry default 3sg agreement if the subject is realized by a full DP, as shown in (7b, c). Similar facts can be observed in Modern Irish where even the co-occurrence of pronominal subjects and full-fledged verbal agreement is prohibited:21 (8) Irish a. *Chuirf-inn mé isteach ar an phost sin. put-cond.1sg I in on the job that b. Chuirf-inn isteach ar an phost sin. put-cond.1sg in on the job that ‘I would apply for that job.’ (McCloskey & Hale 1984: 491) According to McCloskey and Hale (1984), in sentences with rich agreement, the subject position is occupied by a referential pro which is accessible for other grammatical processes such as coordination (note that in Irish, the verb agrees only with the first of two conjoined DPs):22

JB[v.20020404] Prn:18/07/2005; 15:02

F: LA8101.tex / p.11 (697-771)

Chapter 1. Introduction

(9) Dá mbeinn-se pro agus tusa ann. if be-cond.1sg and you there ‘If you and I were there.’ A similar phenomenon can be observed in the Austronesian language Chamorro (Chung 1998: 30f.). In Chamorro, the presence of person agreement on the verb requires that a pronominal subject must be realized as pro. In (10), the verb carries merely number agreement and an overt pronoun may be optionally present. In contrast, the presence of an affix which signals both person and number agreement requires that the (pronominal) subject be realized as pro, as shown in (11): (10) Ginin man-mañ-áñakki (siha) kamuti käda dia. imperf pl-ap-steal.prog they sweet.potato every day ‘They used to steal sweet potatoes every day.’ (11) Ha-fahan pro/*gui’ i lepblu. 3sg-buy he the book ‘He bought the book.’ As noted by Ariel (2000), the distribution of agreement and overt subjects exhibited by languages such as Irish, Welsh or Chamorro is not expected on the assumption that verbal agreement develops from NP-detachment, since this reanalysis necessarily involves the presence of an overt subject DP which developed from a former dislocated topic that regularly co-occurred with a resumptive pronoun. For languages where the presence of verbal agreement is incompatible with the presence of overt subjects, we are therefore led to conclude that the relevant agreement morphology does not originate from resumptive pronouns that relate to a dislocated topic, but must have evolved in another context. Summing up, the previous discussion has shown that the strong claim advocated by Givón (NP-detachment is the only context where verbal agreement can evolve) is contradicted by a set of cross-linguistic observations on the realization of subjectverb agreement. First, it has been argued that Givón’s analysis is not compatible with the so-called ‘suffixing preference’, the fact that in many languages, person/number agreement is preferably realized by verbal suffixes, independent of other properties such as word order. A second problem has been shown to arise from the observation that in many instances, 1st and 2nd person agreement markers evolve prior to verbal affixes signaling agreement with 3rd person subjects. Finally, I have argued that Givón’s hypothesis is challenged by the existence of languages where full verbal agreement in person and number is not compatible with overt subjects. These problems are taken to indicate that there must exist other diachronic paths that lead to predicateargument agreement. In the following, I discuss two alternative proposals concerning the historical development of agreement markers.



JB[v.20020404] Prn:18/07/2005; 15:02



F: LA8101.tex / p.12 (771-834)

The Rise of Agreement

.. Accessibility theory (Ariel 2000) The pioneering role of 1st and 2nd person has inspired an alternative approach to the rise of agreement which is based on Accessibility theory (Ariel 1998, 2000). The basic idea behind Accessibility theory (Ariel 1990) is that the speaker chooses between different types of referring expressions such as full DPs, pronouns, clitics, or agreement markers (called ‘accessibility markers’) according to the salience (referred to as ‘mental accessibility’) of the referred element in a given discourse context. In other words, there is by assumption a close relationship between the salience/accessibility and the morphological encoding of a discourse referent. The higher the accessibility of a discourse referent, the higher the accessibility marker chosen. Lower accessibility markers such as proper names or full DPs are more informative and are typically chosen when the referred element is introduced for the first time into the discourse or has a low degree of salience. Markers of higher accessibility include (weak) pronouns, clitics, agreement markers and zero (i.e., pro). These are less informative and often phonologically reduced or deficient. Based on these assumptions, Ariel (2000) claims that the development of agreement markers from pronouns is not merely the result of phonological reduction, but also driven by the speaker’s desire to use appropriate (reduced) markers to refer to highly accessible/salient discourse referents. Note that in contrast to the NPdetachment analysis, an approach in terms of Accessibility theory does not require the presence of an additional (topicalized) DP. Rather, pronouns may develop ‘directly’ into verbal inflections: (12) a. i. You scored two goals. ii. You-scored two goals. iii. pro 2sg-scored two goals. b. You 2sg-scored two goals.

(free pronoun) (clitic pronoun) (agreement marker) (addition of an overt subject)

The transition from free pronoun to bound agreement marker illustrated in (12a.i– iii) is attributed to the pressure to develop an appropriate morphological encoding of highly accessible discourse referents. According to Ariel, an independent argument may then be added in contexts involving lower accessibility where the reduced marker is not appropriate anymore, giving rise to new instances of grammatical agreement with the reduced form losing its referential potential. This account predicts that agreement markers should develop initially for 1st and 2nd person arguments since these are always highly accessible as necessary discourse participants. In contrast, 3rd persons are non-participants and therefore not as highly accessible as the speaker and hearer. Accordingly, agreement for 3rd person forms should develop later, if at all. Similarly, the requirement that reduced markers develop only for highly accessible referents hinders the development of object agreement, since objects are inherently less accessible referents than subjects (cf. Siewierska 2004: 265f.). Thus, an approach in terms of Accessibility theory avoids the problems arising in connection with the special role of 1st and 2nd person forms. However, it is associ-

JB[v.20020404] Prn:18/07/2005; 15:02

F: LA8101.tex / p.13 (834-849)

Chapter 1. Introduction

ated with a different set of limitations since the development of 3rd person agreement and object agreement seems to fall out of the scope of this proposal. Furthermore, it does not provide a satisfying explanation for the presence of an overt controller in an agreement configuration, that is, the development from stage (12a.iii) to (12b): if verbal agreement is assumed to arise historically as a marker of high accessibility, then the presence of an additional overt subject – typically a marker of low accessibility – is not expected. Ariel’s suggestion that the presence of an overt controller may result from the need to introduce an intermediate stage of accessibility marking (which combines high and low accessibility markers) undermines the whole concept of accessibility marking (the further reduction of the high accessibility marker in question is accompanied by the introduction of a lower accessibility marker). In addition, the assumption that the reduced marker loses its referential potential in this development is contradicted by pro-drop languages such as Italian where overt subjects can co-occur with verbal agreement (note that Ariel assumes that agreement is referential in prodrop languages). Thus, we can conclude that Accessibility theory does not provide a satisfying account of the historical development of agreement either. See Chapter 6 for an analysis of the pioneering role of 1st and 2nd person which does not make use of functionalist notions such as discourse salience/accessibility.

.. Reanalysis of focus shells (Simpson & Wu 2002) In a recent paper, Simpson and Wu (2002) suggest that in general, syntactic concord/agreement develops out of focus constructions which have lost their emphatic function. In most instances (e.g., the development of negative concord), this corresponds to the following phrase-structural configurations, where (13) is reanalyzed as (14). (13)

XP1 X1’ X10

FocP spec

Foc’ Foc0

ZP



JB[v.20020404] Prn:18/07/2005; 15:02



F: LA8101.tex / p.14 (849-918)

The Rise of Agreement

(14)

XP1 X1’ X10

AgrP spec

Agr’ Agr0

ZP

By assumption, FocPs are originally selected by some higher functional head. The morphological material contained in SpecFocP (or Foc0 ) serves to repeat and thereby emphasize the semantic content of the selecting functional head. Over time, the focus interpretation of the selected FocP decays, and the FocP is reanalyzed as an AgrP, the content of which merely redundantly signals the feature content of the selecting head. Later on, the content of the AgrP may come to be reanalyzed as the ‘real’, formerly selecting functional category, giving rise to Jespersen’s cycle effects. The rise of subject-verb agreement is attributed to a related scenario. The initial stage of this change involves a subject pronoun originally base-generated in SpecνP which turns into a clitic and is reanalyzed as an instantiation of ν0 . For reasons of emphasis, the clitic may select a FocP that contains a reinforcing full DP (in SpecFocP) with matching φ-features. In turn, the additional full DP may raise up to the structural subject position (SpecTP) while the clitic (which is in need of a lexical host) attracts the finite verb to ν0 . When this construction loses its stylistic force, the clitic is eventually reanalyzed as a mere agreement morpheme that is part of the finite verb, with the formerly reinforcing subject becoming the real subject. The final change involves a transition from (15a) to (15b). (15) a. [TP Johni [T’ [νP [ν’ V+heclit [FocP ti ... [VP ... tV ...]]]]]] b. [TP Johni [T’ [νP ti [ν’ V+heAgr [VP ... tV ...]]]]] This analysis is related to the traditional idea that verbal agreement develops from a doubling configuration where a full DP/pronoun is added to reinforce a weak or clitic subject pronoun which cannot bear stress or convey emphasis. Eventually, the clitic is reanalyzed as an agreement marker and the originally reinforcing DP becomes the true subject (cf. e.g. Kuen 1957; Wartburg 1970; Ashby 1977 on the history of the Romance languages, in particular French). Simpson and Wu’s formal rendition of this intuition is appealing since it provides us with a unified account of the historical development of agreement/concord phenomena. In addition, Simpson and Wu (p. 308) point out that their proposal can account for the suffixing preference if it is assumed that the verb will always left-adjoin to the light verb containing the former clitic (cf. Kayne 1994). However, note that according to this analysis, new subject-verb agreement morphology develops quite low in the clause structure. This raises two problems. First, the relevant functional head, ν0 , is standardly associated with object agreement, not

JB[v.20020404] Prn:18/07/2005; 15:02

F: LA8101.tex / p.15 (918-966)

Chapter 1. Introduction

with subject agreement (cf. e.g. Chomsky 1995). Second, in most languages, we can observe that clitic pronouns tend to occur in a rather high position, either adjacent to C or to the left of T/Infl. Again, this observation conflicts with the idea that clitics located in ν0 are reanalyzed as agreement markers. Furthermore, it has been observed that in a number of languages, the realization of agreement is subject to restrictions connected with the topic/focus distinction. In general, it appears that topical arguments are likely to trigger agreement while focused ones fail to do so (cf. Siewierska 2004). If these restrictions are taken to reflect the context where verbal agreement developed historically (which is quite plausible), this asymmetry is exactly the opposite of what is predicted by Simpson and Wu’s proposal. Thus, at least there must exist alternative syntactic environments apart from the reanalysis of focus constructions where subject-verb agreement may develop historically.

.. Section summary In this section, I have discussed three previous approaches to the development of agreement from pronominal clitics, arguing that each scenario accounts only for a subset of the relevant phenomena. More specifically, it has been demonstrated that an analysis in terms of NP-detachment can account for the development of new agreement markers in languages such as Non-Standard French, but is at odds with the suffixing preference, the pioneering role of 1st and 2nd person forms, and the incompatibility between overt subjects and full-fledged agreement exhibited by some Celtic languages, for example. In contrast, an Accessibility-based account fares better with respect to the special role of 1st and 2nd person, but requires special assumptions to account for 3rd person agreement and the presence of an overt subject in agreement configurations. Finally, the assumption that verbal agreement evolves out of former focus constructions covers instances where the reinforcing full DP/pronoun in clitic doubling configurations is reanalyzed as the subject of the clause, but falsely predicts that cross-linguistically, agreement is more likely to be triggered by focal than by topical constituents (apart from other, more technical problems). Thus, it appears that NP-detachment, ‘direct’ reanalysis of pronouns (i.e., Accessibility theory) or reanalysis of focus shells are possible pathways towards verbal agreement, but crucially, none of the relevant approaches can claim to constitute the only historical scenario where agreement markers may develop from pronouns. In this book, I claim that the search for such a unique scenario is in fact misguided. Rather, it is argued that the reanalysis of pronominal elements is subject to a set of abstract syntactic and morphological conditions which may be fulfilled in a variety of different syntactic environments (cf. Chapters 4 and 5).



JB[v.20020404] Prn:18/07/2005; 15:02



F: LA8101.tex / p.16 (966-1015)

The Rise of Agreement

. Outline of the Book This book aims at developing a formal account of the grammaticalization of verbal agreement morphology, focusing on the reanalysis of (C-oriented) subject clitics as inflectional markers.23 Note that this work presupposes familiarity with the Principles and Parameters approach to syntax, that is, the theoretical apparatus laid out for example in Chomsky (1981, 1982, 1986a, 1986b), and Chomsky and Lasnik (1993), including X-bar theory and properties of the individual modules of Government and Binding theory. Relevant theoretical concepts which are of more recent origin are introduced and discussed in Chapter 2, including key properties of minimalist syntax and Distributed Morphology. In the following, I provide an overview over the organization of the book. Chapter 2 introduces the basic theoretical assumptions adopted in this work. As already hinted at above, a diachronic investigation into the historical development of agreement morphology touches on several disciplines of linguistics, involving syntax, morphology and the theory of language change in equal parts. Accordingly, I provide an outline of the current version of minimalist syntax (Chomsky 2000, 2001a, 2001b), an introduction into Distributed Morphology, and an overview of generative approaches to language change, focusing on the phenomenon of grammaticalization which is analyzed as a historical process that provides new phonological exponents for a set of (universally present) functional heads (cf. Roberts & Roussou 2003). As already mentioned, an analysis of the diachronic development of verbal agreement necessarily builds on a theory of the syntactic and morphological mechanisms which establish agreement in natural languages. In Chapter 3, I therefore develop a synchronic approach to predicate-argument agreement which is based on the following assumptions. First, agreement morphemes do not head their own projection in the syntax. Rather, they are parasitic on other functional heads with which they may combine (via adjunction) either before or after the syntactic derivation. Agr-heads present in the syntactic derivation are valued by an appropriate controller under closest c-command (i.e., Agree), while post-syntactically inserted Agr-heads (so-called dissociated Agr-morphemes, cf. Embick 1997) are licensed under structural adjacency with an Agr-morpheme that has been valued in the syntax. Based on the latter mechanism, a novel analysis of complementizer agreement in Germanic is proposed, arguing that this particular form of multiple agreement results from the post-syntactic insertion of a dissociated Agr-morpheme onto C. Chapter 4 explores the diachronic consequences of this particular approach to verbal agreement. In particular, I propose a set of restrictions on the reanalysis of clitic pronouns as exponents of agreement morphemes which derive from the theory of verbal agreement proposed in the previous chapter. Furthermore, it is claimed that the reanalysis of subject clitics may extend the pro-drop properties of a language by creating new instances of referential pro. Alternatively, subject clitics may be reanalyzed as agreement markers in clitic doubling structures, with the relevant θ-role assigned to the former double. In addition, I argue that the grammaticalization process in question

JB[v.20020404] Prn:18/07/2005; 15:02

F: LA8101.tex / p.17 (1015-1071)

Chapter 1. Introduction

must give rise to structural simplification, in the sense that the structures resulting from the reanalysis of clitics must involve less complex (syntactic) derivations. Based on these assumptions, a set of historical scenarios is presented where the reanalysis of clitic pronouns as agreement markers may be triggered. In Chapter 5, I take a closer look at a subset of these scenarios, focusing on the reanalysis of C-oriented clitics. It is shown that the latter change is shaped by independent syntactic properties of the languages under consideration, arguing that the V2 property facilitated the development of new agreement markers in German and Rhaeto-Romance varieties. In addition, I discuss a set of Uto-Aztecan and Mongolian languages which lack the V2 property but still exhibit a historical development in which C-oriented clitics turned into verbal agreement markers. Furthermore, it is demonstrated that the grammaticalization of new agreement formatives often proceeds via an initial stage where a clitic is reanalyzed as a dissociated agreement morpheme which is inserted and valued in the morphological component, giving rise to instances of multiple agreement (such as complementizer agreement in Germanic). In Chapter 6, I discuss a set of morphological factors that shape the grammaticalization of agreement markers, focusing on the observation that across languages, the development of new agreement formatives is triggered only in contexts where the existing verbal agreement morphology is non-distinctive. This observation is accounted for by the assumption that the learner scans the input for the most specific exponent of a given inflectional morpheme. More technically speaking, it is argued that the acquisition of inflectional morphology is guided by blocking effects which ensure that new verbal agreement formatives are more specific (i.e., realize a greater subset of the φ-features contained in the relevant Agr-head) than existing morphology. Relevant examples come from a number of German varieties, Non-Standard French, and various Northern Italian and Rhaeto-Romance dialects. In addition, I take a look at American Russian and the Papuan language Skou. Moreover, the account proposed in this section is shown to provide a new explanation for the pioneering role of 1st and 2nd person in the development of predicate-argument agreement if it is assumed that 3rd person forms are inherently underspecified for person features.

Notes . Note that the choice of the terms controller and target already implies that the structural relation between these two elements is asymmetrical in nature, in the sense that the target changes its morphological form in order to match (inherent) feature specifications of the controller (cf. Lehmann 1993; Corbett 1998; Chomsky 2002, among many others). Furthermore, there is a clear intuition that the semantic interpretation of agreement features is linked to properties of the controller, rather than to properties of the target. In current syntactic theorizing, this asymmetry is generally modeled by the notion that agreement features are interpretable (i.e., legible by grammar-external modules such as the conceptual system) on nouns, but uninterpretable on verbs, adjectives etc. (Chomsky 1995, 2000, 2001a, 2001b). As Chomsky (2002: 112) puts it: “It looks as if there is some real truth to the traditional idea that verbs agree with nouns and not



JB[v.20020404] Prn:18/07/2005; 15:02



F: LA8101.tex / p.18 (1071-1152)

The Rise of Agreement

conversely. The thing that is agreeing, presumably the verb, the adjective, the article, and so on, they all seem to have uninterpretable features, features that are not independently interpreted by the outside systems.” . A more indirect indication of the diachronic origin of verbal agreement markers is the fact that cross-linguistically, agreement morphemes mark a subset of the grammatical distinctions marked by personal pronouns (cf. Barlow 1992; Lehmann 1993; Corbett 1998; Ariel 2000). The set of φ-features marked by verbal agreement is usually assumed to be confined to [person], [number], [gender] and [noun class] (the set of languages which exhibit agreement in noun class includes Bantu languages such as Swahili or Chichewa, cf. Givón 1976; Bresnan & Mchombo 1987; Krifka 1995, and many Caucasian languages, cf. e.g. Schulze 1998; Harris & Smeets 2003). . The primary endings appear in the present whereas the secondary endings are used in the aorist and imperfect (e.g. in Sanskrit and the earliest Greek records). Historically, the primary endings are derived via affixation of a deictic -i to the secondary endings. In other words, in contrast to the somewhat misleading terminology, the ‘secondary’, unextended endings are in fact older than the ‘primary’, extended endings (Szemerényi 1989; W. Lehmann 1993). . Note that the 3rd person forms did not develop from personal pronouns (as becomes clear from table 1, no 3rd person pronouns are reconstructed for PIE). Traditionally, it is assumed that the 3sg ending -t developed from the demonstrative *to, whereas the 3pl suffix *-nt is related to the participial (i.e. originally derivational) affix *-nt- (Szemerényi 1989: 361). . Arregi (1999) claims that the pronouns and agreement markers of Basque are not only historically, but also synchronically related. For example, he decomposes the (absolutive) 2pl pronoun sue into a set of formatives including s- ‘2nd person’, -u- ‘nominal stem’ and -e ‘plural’. Under this analysis, the similarity between pronouns and agreement becomes even more intimate. Arregi then suggests that pronouns and verbal agreement actually involve the same lexical items, which are underspecified for the kind of stem (verbal or nominal) they attach to. . Recently, Ariel (1998, 2000) has observed that clitics are currently evolving into a new agreement paradigm in the future tense of Colloquial Hebrew. . In Dagur Mongolian, for example, the agreement suffixes are still fully identical to the pronominal forms (cf. Martin 1961). . Note that (2) actually conflates two distinctions. On the one hand, it represents the different morphological stages which can be identified in this grammaticalization process (such as free vs. bound form). On the other hand, it includes information about the changing syntacticosemantic properties of the element in question, that is, the transition from a referential pronoun (with a deictic and/or anaphoric potential) to an agreement marker which does not have any referential potential on its own, but merely reflects the φ-set of an argument. . The fact that there are a few languages (mostly Creoles) which apparently exhibit non-bound agreement markers suggests that the phonological reduction involved in this process does not necessarily result in bound agreement formatives (cf. Bybee et al. 1990; Julien 2002). . An instance of the change under discussion which has received some attention in historical studies on Germanic is the development of the 2sg ending -st, which is usually attributed to the cliticization of the 2sg pronoun -t(hu) to the inherited 2sg suffix -s in inversion contexts (for Old High German cf. e.g. Brinkmann 1931; Braune 1950; Paul 1952; Lühr 1984; for Old English e.g. Wright 1925). . See Bresnan and Mchombo (1987) for an application of Givón’s proposals in an LFG analysis of verbal agreement marking in Chichewa (Bantu). According to Bresnan and Mchombo,

JB[v.20020404] Prn:18/07/2005; 15:02

F: LA8101.tex / p.19 (1152-1206)

Chapter 1. Introduction

Chichewa exhibits two types of agreement which developed historically via the incorporation of resumptive pronouns. First, there is ‘true’ grammatical agreement where an NP bears an argument relation to the verb, while its φ-features are redundantly expressed by verbal affixes. Second, there is the possibility of so-called ‘anaphoric agreement’ where the apparent Agrmorpheme is an incorporated pronominal argument, while the coreferential DP is analyzed as an adjunct that occupies an A’-position (either topic or focus). By assumption, anaphoric agreement preserves the historical origin of these constructions more faithfully than grammatical agreement which is analyzed as a more recent development. Based on this distinction, it is argued that in Chichewa, the subject agreement marker can function either as anaphoric or as grammatical agreement, whereas the object marker can solely be used for anaphoric agreement, which leads to a set of subject/object asymmetries in this language. . Note that the term suffixing preference actually involves two separate generalizations: first, of all morphological means to realize verbal inflection, suffixing is the most frequent one. Second, postverbal inflectional markers show a much greater tendency to appear as bound elements (on the verb) than preverbal markers. . Note that Table 3 distinguishes between free (e.g. Agr V) and bound (Agr+V) forms. The figures refer to individual agreement markers in the languages under consideration (i.e., 1sg, 2sg etc. are counted separately). . Furthermore, it becomes clear that there is a general tendency for agreement markers to be affixed to the verb. Apart from the verb-initial languages in Bybee et al.’s sample where all preverbal markers are bound to the verb, this tendency is generally stronger for postverbal markers (particularly in SOV languages where it is 100%). . In fact most researchers agree today that even the exact origin of the number markers in Indo-European (which came into existence via a later development) eludes explanation, although it seems to be clear that they bear some relation to the oblique forms of the personal pronouns (see Table 1 above) (cf. e.g. W. Lehmann 1993). . Based on a survey involving 402 languages, Siewierska (2004: 165) claims that the position of person agreement affixes in fact does not reflect the suffixing preference as clearly as other inflectional markers: “The existence of a preference for suffixes over prefixes receives support from the ordering of tense, aspect and modality affixes and from the existence of languages which are exclusively suffixing as well as from the lack of languages which are exclusively prefixing (see, e.g., Hawkins & Gilligan 1988). However, it finds only very weak support from the location of person agreement affixes. Among the languages in the sample the markers of a[gent] agreement, p[atient] agreement and possessor agreement are marginally more often suffixes than prefixes, but the difference is only of 1 to 3 per cent.” However, even if agreement markers are not primarily suffixing, there is still a mismatch between the number of languages where the source of agreement (subject clitics) precedes the verb and the number of languages where agreement is realized by suffixes, which is not expected under Givón’s assumptions. . Moreover, note that in many languages left dislocation of 1st and 2nd person forms is extremely marked, if not ungrammatical: (i) ??I, I will read the book on clitics. Similar observations hold with respect to left dislocation of personal pronouns in general, which again seems to indicate that in these contexts, verbal agreement could only develop via analogy.



JB[v.20020404] Prn:18/07/2005; 15:02



F: LA8101.tex / p.20 (1206-1283)

The Rise of Agreement

(ii) *Er, der hat das Buch gestohlen. he pron has the book stolen ‘As for him, he stole the book.’ . Note that many languages lack 3rd person personal pronouns. Instead, 3rd persons are referred to by full DPs, demonstratives (which may be a historical source for pronouns and verbal agreement, cf. e.g. Siewierska 2004), or by nothing at all. In particular the use of demonstratives instead of explicit 3rd person personal pronouns is quite widespread across the languages of the world. Examples include Hindi/Urdu (Kachru 1990), Korean (Shin-Sook Kim, personal communication), Mongolian (Poppe 1954, 1960), Pashto (MacKenzie 1990), Sanskrit (W. Lehmann 1993), and many native American languages (Mithun 1991). Given these facts, it is likely that at least in some instances, the lack of 3rd person agreement can be attributed to the lack of an appropriate lexical source for the grammaticalization process in question. . Mithun (1991) observes that in the Athabaskan languages (spoken e.g. in Alaska, Western Canada, Oregon, and California), 1st and 2nd person subject prefixes occur very close to the verb stem, while 3rd person subject markers occupy a position at the left edge of the verbal inflection, separated from the 1st and 2nd person markers by mode and aspect prefixes. This fact is then taken to indicate that the 1st and 2nd person markers were grammaticalized earlier than the 3rd person markers. A similar point can be made with respect to number marking in these languages. In general, person and number markers are fused forms in the case of 1st and 2nd person subjects, which indicates that they are of considerable antiquity. In contrast, number marking for 3rd person subjects occupies a separate slot in the inflectional make-up of the verb or has often barely been grammaticalized. Particularly revealing is the shape of transitive pronominal affixes in a number of languages where affixes for agent and patient are adjacent (e.g. Lakhota, Caddo, Maricopa). In general, affixes that mark 1st and 2nd persons show a high degree of fusion, in contrast to those including 3rd persons (Mithun 1991: 87). In other words, a 1st person subject acting on a 2nd person object is marked by a fused 1st/2nd person marker, whereas a 1st person acting on a 3rd person often involves two separate person markers. Again, the advanced phonological reduction of 1st/2nd affixes can be taken to indicate their special antiquity. . Note that the facts described by (6) may be blurred by the subsequent erosion of the verbal agreement system, which may lead to such exceptional cases like English, where only 3rd person subjects are marked on the verb (via the 3sg.pres.indic ending -s). See Chapter 6 for a discussion of further apparent exceptions where the development of 3rd person markers seem to take place parallel to or even predate the development of 1st and 2nd person markers. . In Modern Irish, an overt subject requires the use of the so-called analytic form of the verb, which is unmarked for person and number, whereas the fully inflected ‘synthetic’ form is not allowed in these contexts. Note that rich person-number marking on verbs is more or less restricted to the conditional, whereas other moods/tenses show less agreement morphology (cf. McCloskey & Hale 1984: 492). . Other possible analyses include incorporation of a subject pronoun into the verb (as discussed by McClosky & Hale 1984: 527f.) or checking of the EPP by rich verbal agreement with pronominal properties (cf. Alexiadou & Anagnostopoulou 1998). . Alternative sources of agreement morphology are not taken into consideration or only briefly commented on in this book (cf. Siewierska 2004 for an overview). These may include the reanalysis of other inflectional affixes as agreement markers (cf. Chafe 1977 on the 3rd person subject prefixes in Iroquoian), the incorporation of inflected auxiliaries into the main verb

JB[v.20020404] Prn:18/07/2005; 15:02

F: LA8101.tex / p.21 (1283-1288)

Chapter 1. Introduction

(cf. Hetzron 1976 on Cushitic, Haas 1977 on Muskogean, Roberts 1993b, Roberts and Roussou 2003 on Romance) or analogical changes in which agreement markers originally confined to a certain tense/mood or verb class gradually gain a wider distribution (cf. e.g. Lühr 1984 who claims that Germanic bi-st 2sg ‘be’ developed in analogy to the relevant agreement ending found on the preterite-presents). Note that at least the latter two scenarios have nothing to say about the origin of the agreement morphology which is transferred to other verbs (either via fusion or analogy). However, it is in fact quite likely that this inflection is the result of a previous process in which pronouns were reanalyzed as agreement markers. Hence, it appears that at least some of the alternative scenarios can be traced back to the reanalysis of pronouns as well.



JB[v.20020404] Prn:2/08/2005; 9:38

F: LA8102.tex / p.1 (45-121)

Chapter 2

Theoretical preliminaries

. Introduction This chapter serves to lay the theoretical foundations for the synchronic and diachronic analyses that are developed in the chapters to come. Since agreement is a phenomenon that involves syntax and morphology in equal parts, I will give an outline of both the syntactic and morphological mechanisms that participate in the realization of predicate-argument agreement and the way these mechanisms interact with each other. In addition, a formal account of the historical rise of agreement has to be founded on a theory of language change and in particular the phenomenon of grammaticalization. Accordingly, this chapter introduces generative approaches to language change (and variation) and grammaticalization in some more detail. The chapter is organized as follows. Section 2.2 presents an outline of the current version of minimalist syntax devised in Chomsky (2000, 2001a, 2001b). Section 2.3 provides an introduction into Distributed Morphology (Halle & Marantz 1993), the morphological framework adopted in this work. In Section 2.4, I give an outline of the generative approach to syntactic change in terms of a change in parameter values. Finally, Section 2.5 focuses on the phenomenon of grammaticalization, arguing that this kind of change provides new phonological exponents for a closed class of functional heads (Roberts & Roussou 2003).

. Minimalist syntax This section serves to introduce the core properties of the syntactic framework adopted in this work (the version of minimalism proposed in Chomsky 2000, 2001a, 2001b), focusing on the aspects of the theory that are relevant for the implementation of agreement. Perhaps the most obvious difference between minimalism and previous versions of the Principles and Parameters approach concerns the elimination of the representational levels D-structure and S-structure. Compare (1a), the so-called T-model (or Y-model) assumed in work such as Chomsky (1981), (1986a), with (1b), the grammar model proposed in Chomsky (1993) and subsequent work:

JB[v.20020404] Prn:2/08/2005; 9:38



F: LA8102.tex / p.2 (121-183)

The Rise of Agreement

(1) a.

Lexicon D-structure Move-á S-structure Move-á PF LF

b.

Lexicon syntactic derivation Spell-out PF LF

The minimalist grammar model in (1b) involves only two (instead of four) levels of representation (see Section 3 for a slightly revised form of (1b)). Accordingly, principles and constraints that formerly applied to the levels D-structure or S-structure are reformulated as conditions on derivations or as wellformedness conditions on the structure of the remaining levels of representation PF and LF which constitute the interface levels to other cognitive systems (‘bare output conditions’). Syntactic structures are built by the operations Merge and Move. The basic structure-building operation Merge forms larger units by combining two elements X, Y either drawn from the lexicon or constructed previously during the syntactic derivation. The syntactic category of the resulting structure is determined by properties of either X or Y. The second operation that forms new phrase-markers is syntactic movement, that is, the operation Move. Following Chomsky (1995), the operation Move can be defined as follows: (2) Given the phrase marker Σ with syntactic objects K and α, Move targets K, raises α, and merges α with K to form the new category γ with the constituents α, K. Thus, the basic structure-building operation Merge is also part of the more complex operation Move. In minimalist syntax, it is generally assumed that the syntactic derivation is morphologically driven. More precisely, syntactic operations (Move) are triggered by the need to eliminate certain morphological features which cannot be interpreted at the interfaces to other cognitive systems (PF, LF). Failure to delete noninterpretable features prior to the interface level where they are visible causes the syntactic derivation to crash. These features are also referred to as formal features, since they are associated with formal properties of grammar such as tense, case and agreement. In earlier versions of the minimalist program (Chomsky 1993, 1995), it is assumed that functional heads host formal features such as [Nominative], [Past], and φ-features (e.g. [person], [number], [gender]) which are deleted by entering into a checking relation with identical features on substantial lexical categories such as N, V or A. The latter are combined with inflectional affixes in the lexicon and are inserted fully inflected. Checking relations are established by either head movement that adjoins a lower head to a higher functional head or XP-movement into the specifier of a functional head. Overt movement is triggered by ‘strong’ non-interpretable features which are visible at PF, whereas ‘weak’ non-interpretable features can be eliminated by covert movement operations that take place between Spell-out and LF. Economy

JB[v.20020404] Prn:2/08/2005; 9:38

F: LA8102.tex / p.3 (183-235)

Chapter 2. Theoretical preliminaries

principles such as ‘Procrastinate’ ensure that movement and feature checking must be postponed to the covert syntax if possible. In general, considerations of derivational economy play an important role in the implementation of syntactic operations. For example, in contrast to the nature of the operation Move α which applies freely (if not ruled out by independent principles), syntactic movement is considered a Last Resort operation that may be triggered only by the need to check off non-interpretable features. Locality is attributed to the workings of economy principles as well. Conditions such as Shortest Move/the Minimal Link Condition or Merge over Move require that at any given point in the syntactic derivation, the computational system must carry out the most economic operation, preferring less complex over more complex derivations. In Chomsky (1993, 1995), syntactic agreement is established via a specifier-head configuration between an argument (the agreement controller) and the predicate (the agreement target). Thus, (overt) agreement checking is parasitic on A-movement to the specifier of a functional head (Chomsky 1993: AgrS/AgrO; Chomsky 1995: T/ν) which also attracts the inflected verb. By assumption, overt movement is triggered by ‘strong’ categorial features (e.g. [*D], [*V], where the star indicates feature strength) which are part of the set of formal features of a given functional head and must be eliminated prior to Spell-out.1 The minimalist analysis of agreement is completely revised in Chomsky (2000, 2001a, 2001b), where a rather different model of the computational system and its basic components is developed.2 First, on this view, the syntactic operations which are necessary to check off (or, rather, value) the set of non-interpretable/non-valued φ-features located in T/ν are triggered by the φ-set itself (and not by the presence of non-interpretable categorial [D]/[N] or [V] features on T/ν). Second, feature checking/valuing is dissociated from (overt) DP-movement into the specifier of the relevant functional head. Instead, the set of non-interpretable φ-features located in T/ν may access the interpretable φ-set of an argument which stays in situ. The latter operation is called Agree “which establishes a relation (agreement, Case checking) between an LI [lexical item] α and a feature F in some restricted search space (its domain)” (Chomsky 2000: 101). A feature set that starts an Agree operation is referred to as a probe. A probe seeks to establish a relation with another set of ‘matching’ features which is called the goal. In the case at hand, the unvalued φ-set in T/ν acts as a probe that seeks a set of matching interpretable φ-features (the goal) with which it can establish agreement. Similar to Chomsky (1993, 1995), φ-features are assumed to be interpretable on DP/NP, but uninterpretable on T/ν. More precisely, Chomsky (2000, 2001a) proposes that the value of the φ-features on T/ν is not specified. When the set of unvalued φ-features succeeds in establishing an Agree relation with the φ-set of a nominal argument, the value of the latter is assigned to the previously unvalued φ-set. Afterwards, the uninterpretable φ-set on T/ν is marked for deletion (ultimate deletion is postponed to the point where the derivation reaches Spell-out). The core properties of the probe/goal mechanism can be summarized as follows (Chomsky 2000: 122):



JB[v.20020404] Prn:2/08/2005; 9:38



F: LA8102.tex / p.4 (235-295)

The Rise of Agreement

(3) Matching is a relation that holds of a probe P and a goal G. Not every matching pair induces Agree. To do so, G must (at least) be in the domain D(P) of P and satisfy locality conditions. The simplest assumptions for the probe-goal system are: a. Matching is feature identity. b. D(P) is the sister of P. c. Locality reduces to closest c-command. In this framework, the significance of structural Case is considerably diminished. By assumption, Case is not part of the feature make-up of the probes T/ν. Therefore, it cannot trigger checking operations, since Agree requires feature matching between probe and goal. Still, Chomsky assumes that there is an intricate relation between structural Case and φ-feature checking. More precisely, an uninterpretable Case feature renders the goal active, that is, accessible for the operations Agree and Move. The establishment of an Agree relation then leads to the erasure of uninterpretable features in both the probe (φ) and the goal (Case).3 In other words, an uninterpretable φ-set (the probe) can only be marked for deletion by a goal with an unchecked Case feature. Accordingly, a phrase with a checked/deleted Case feature cannot undergo further A-movement and is “frozen in place” (Chomsky 2000: 123).4 Thus, we must add the following statement to the characterization of the probe/goal mechanism in (3) above (Chomsky 2000: 123): (4) The operations Agree and Move require a goal that is both local and active. By assumption, the activating Case feature can only be deleted by a probe with a complete set of φ-features. For example, the defective T head (Tdef ) of non-control infinitivals hosts only a [person] feature. Movement of the subject to SpecTP deletes the uninterpretable [person] feature, but does not lead to deletion of the subject’s structural Case feature. Accordingly, the subject of a raising infinitival can be targeted by further syntactic operations (Agree and Move triggered by a probe in the matrix clause). In a similar vein, the reduced feature content of an expletive (presumably only [person] as well) fails to value and delete the uninterpretable φ-set of nondefective T which may then establish long-distance agreement with an associate DP (in situ) via Agree: (5) There were [three men] declared guilty. Agree Overt phrasal movement is parasitic on a previously established Agree relation. It is triggered by the requirement that a given functional head must project a specifier. This is implemented by the assumption that functional heads may host an EPP feature. The EPP feature is assumed to be universally present on T, while its presence on C (whmovement) and ν (object shift) is subject to parametrization. Thus, Move is a complex operation which consists of the basic operations Agree and Merge:

JB[v.20020404] Prn:2/08/2005; 9:38

F: LA8102.tex / p.5 (295-357)

Chapter 2. Theoretical preliminaries

(6) Move of β, targeting α, has three components (Chomsky 2000: 135): a. A probe P in the label L of α locates the closest matching G in its domain. b. A feature G’ of the label containing G selects a phrase β as a candidate for ‘pied-piping’. c. β is merged to a category K. As in previous work (Chomsky 1995), it is assumed that the more basic operation Merge preempts Move (Merge over Move). In addition to the reformulation of syntactic movement and feature checking, Chomsky (2000) devises a new concept of locality that is based on the hypothesis that the syntactic computation proceeds via certain derivational stages which are called phases. By assumption, CP and νP constitute phases, but not TP or VP.5 This yields a strong form of cyclicity: neither may the head of a phase α trigger operations after α is completed, nor can operations triggered by higher heads access elements inside a completed phase α apart from its head and specifiers. These conditions can be stated as follows (Chomsky 2000: 107f.): (7) The head of a phase is ‘inert’ after the phase is completed, triggering no further operations. (8) Phase Impenetrability Condition (PIC) In phase α with head H, the domain of H is not accessible to operations outside α, only H and its edge are accessible to such operations. Accordingly, material contained in a phase α must move to the left edge of α before it can enter into an Agree or Move operation triggered by a higher probe. For example, the PIC requires that A’-movement must target the edge of every phase (CP, νP) that lies between the foot and the head of the resulting movement chain. From the PIC, it follows that the left edge of a phase is accessible to syntactic operations until the next higher phase is completed. Furthermore, phases are associated with Spell-out. That, is at every phase level, the complement of the phase head is sent to the interface levels PF and LF after the phase has been completed (so-called ‘cyclic Spell-out’). The phase head and its specifiers are sent to the interfaces with the complement of the next higher phase head. For example, the VP complement of ν is subject to Spell-out after νP has been completed, while ν and its specifiers are spelled-out together with the TP complement of the next higher C head. This entails that there are no ‘covert’ syntactic operations that take place after Spell-out (i.e., there is no LF-movement). Rather, all syntactic operations (Merge, Agree, Move) must take place prior to Spell-out, independent of whether they have visible effects at PF or not. To summarize, in the present section I gave an outline of the syntactic framework adopted in this book. Importantly, an analysis of agreement in terms of Agree no longer requires the creation of a spec-head configuration during the syntactic derivation. In Chapter 3, I will present further evidence that this approach is on the right track, drawing on evidence from long-distance agreement in the Nakh-Dagestanian language Tsez. Note that the basic assumptions laid out in the present section will be



JB[v.20020404] Prn:2/08/2005; 9:38



F: LA8102.tex / p.6 (357-400)

The Rise of Agreement

slightly modified in the chapters to come. In Chapter 3 (Section 3.6.2) it is argued that Case and agreement licensing may be dissociated (i.e., checked off by different functional heads). This requires some modifications. First, Case checking does not necessarily require the existence of an unvalued φ-set in the probe. Second, a DP with a Case feature marked for deletion may still be accessible for a further Agree operation that values the uninterpretable φ-features of a higher Agr-head (see below for discussion). In the introduction to this chapter, I noted that agreement is a phenomenon on the borderline between syntax and morphology. Accordingly, the next section presents a set of basic assumptions on the role and workings of morphology in the realization of agreement and on the nature of the interface between syntax and morphology.

. Distributed Morphology The analysis of agreement presented in this book is framed in a realizational model of grammar where word building operations are distributed over several components of grammar (Distributed Morphology, henceforth DM; Halle & Marantz 1993, 1994). DM assumes that the morphological component operates post-syntactically, interpreting the output of the syntactic derivation. In other words, the morphological component mediates between the syntactic and the phonological modules of grammar. Accordingly, the structural design of the grammar looks like (9).6 (9)

Lexicon (morphosyntactic/semantic features) Syntactic derivation Spell-out Morphology LF PF

In this model of grammar, the syntactic operations Merge and Move operate on bundles of morphosyntactic features that constitute syntactic terminal nodes (i.e., heads). The syntactic terminal nodes are referred to as morphemes.7 In the morphological component, the terminal nodes are associated with phonological exponents in a process called Vocabulary Insertion. The idea that phonological content is added after syntax is also known as Late Insertion.8 The information that links phonological exponents with morphosyntactic features (i.e., insertion contexts) is stored in individual Vocabulary items. For example, the English verbal inflection 3sg.pres.indic. /-z/ is associated with the following Vocabulary item (which can be read as an insertion rule: “the phonological exponent /-z/ is inserted in the context [3, sg, pres., indic.]”):

JB[v.20020404] Prn:2/08/2005; 9:38

F: LA8102.tex / p.7 (400-470)

Chapter 2. Theoretical preliminaries

(10) [3, sg, pres., indic.] ↔ /-z/ The insertion procedure requires that the feature specification of the Vocabulary item is nondistinct from the features of the insertion site (i.e., a certain morpheme). Usually, this requirement is met by several items, which then enter into a competition. The item that realizes the greatest subset of features is chosen for insertion. In the example at hand, the availability of the Vocabulary item in (10) blocks the insertion of the less specified exponent -Ø, which is found in all other contexts, representing the ‘elsewhere’ case (cf. Kiparsky 1973; Aronoff 1976; Anderson 1986 for the use of elsewhere principles in phonology/morphology). Thus, the paradigm exhibited in (11) is captured by positing merely two Vocabulary items (listed in (12)) for the realization of Infl in English. (11) a. I/you/we/you-pl/they sleep. b. He/she/it sleep-s. c. *He/she/it sleep. (12) a. [3, sg, pres., indic.] ↔ /-z/ b. elsewhere ↔ -Ø This approach implies that Vocabulary items may be underspecified for the feature complexes they realize. In (12), the exponent -Ø is maximally underspecified since it is not linked to any feature specification at all (the ‘elsewhere’ case). The basic properties of the insertion procedure are captured by Halle’s (1997: 428) Subset Principle (not to be confused with the Subset Principle of Wexler & Manzini 1987):9 (13) The Subset Principle The phonological exponent of a Vocabulary item is inserted into a morpheme in the terminal string if the item matches all of a subset of the grammatical features specified in the terminal morpheme. Insertion does not take place if the Vocabulary item contains features not present in the morpheme. Where several Vocabulary items meet the conditions for insertion, the item matching the greatest number of features specified in the terminal morpheme must be chosen. Most work in DM acknowledges a systematic difference between lexical and functional categories (Halle 1990: ‘concrete’ vs. ‘abstract’ morphemes; Harley & Noyer 1999: ‘l-morphemes’ vs. ‘f-morphemes’). Functional morphemes constitute a closed class and contain features relevant for the syntactic computation (e.g., [plural], [+past], [+wh]). In addition, they presumably carry semantic features associated with a certain set of logical meanings (cf. Section 2.5 below). A property specific to functional morphemes is that their spell-out is deterministic, that is, the feature content of a given functional morpheme serves to determine a unique phonological realization. Thus, the insertion of Vocabulary items realizing functional morphemes is guided by the Subset Principle which ensures that the most specific candidate wins out over its competitors.



JB[v.20020404] Prn:2/08/2005; 9:38



F: LA8102.tex / p.8 (470-531)

The Rise of Agreement

In contrast, the spell-out of open class lexical morphemes is non-deterministic and not constrained by a competition between compatible Vocabulary items. Thus, in a context (i.e., a certain lexical morpheme) which requires the insertion of a nominal head, any Vocabulary item that is compatible with this specification can be chosen for insertion (dog, cat, house, ball etc.). In other words, for lexical morphemes, there is a choice in spell-out which does not exist for functional morphemes and which I refer to as free-choice spell-out.10 This fact is reflected by the shape of insertion rules/Vocabulary items for lexical categories which are identical for all members of a given lexical category, cf.11 (14) a. [DP D [NP ___ ]] ↔ /dog/ b. [DP D [NP ___ ]] ↔ /kæt/ etc. Prior to Vocabulary Insertion, a set of morphological operations may apply to the output of the syntactic component, changing the content and (hierarchical) structure of morphemes. The most important of these are the insertion of so-called dissociated morphemes, (morphological) Merger or Fusion, Fission and Impoverishment. In the following, these mechanisms are only briefly introduced; they are discussed in more detail when they have a concrete bearing on the issues dealt with in this book. The constituent structure of morphemes derived in the syntax can be modified by the post-syntactic insertion of (functional) dissociated morphemes which may attach to other functional morphemes. Following Embick (1997), these morphemes are called ‘dissociated’, since they are not present in the syntactic derivation and merely reflect properties expressed by structural configurations in the syntax proper. Within DM, this mechanism is often used to account for case and agreement phenomena. For example, Marantz (1992), Halle and Marantz (1993), Halle (1997) analyze subjectverb agreement in terms of the post-syntactic adjunction of an [Agr] morpheme to T (cf. Embick 1997 for a detailed discussion of the insertion of dissociated morphemes). See Chapter 3 for an analysis of complementizer agreement in Germanic in terms of dissociated Agr-morphemes. In Chapters 4–6, it is argued that dissociated Agr-morphemes play an important role in the historical rise of new agreement morphology as an intermediate step on the grammaticalization path from clitics to syntactic agreement markers. Another morphological operation that may change the constituent structure generated by the syntax is Morphological Merger of terminal nodes that do not form a constituent in the syntactic output (cf. Marantz 1984, 1988). For our purposes, the most important instance of Merger are cases of apparent syntactic lowering, that is, syntactic heads that are not joined together via head movement, but still are spelled out as a unit. A prominent example of this type of Merger is affix-hopping in English, the post-syntactic affixation of Tense/Agr to the main verb, which on standard assumptions fails to undergo overt head movement in English (cf. Bobaljik 1995, 2002, 2003).12

JB[v.20020404] Prn:2/08/2005; 9:38

F: LA8102.tex / p.9 (531-582)

Chapter 2. Theoretical preliminaries

A related operation is Fusion which may create a mismatch between the number of underlying morphemes and the number of inserted Vocabulary items, in the sense that two (or more) syntactic nodes are fused into a single terminal node which is then realized by a single phonological exponent. Again, the verbal inflection of English is a case in point: it is commonly assumed that Agr and T fuse into a single morpheme prior to Vocabulary Insertion (Halle & Marantz 1993; Halle 1997). Instances of Fusion are traditionally referred as cumulative exponence and may involve quite a number of different inflectional features as in the verbal inflections of Latin (cf. Matthews 1991: 233): in a 1st person singular present indicative active form like am-¯o ‘I love’, the single exponent -¯o serves to simultaneously realize five inflectional features, namely person, number, tense, mood, and voice.13 Similar to Fusion, Fission (Halle & Marantz 1993; Halle 1997; Noyer 1997) also disturbs the isomorphism between syntax and morphology, albeit with contrary effects: while Fusion leads to the amalgamation of two separate syntactic terminals, Fission results in a situation where a single syntactic terminal node is realized by more than one Vocabulary item. The concept of Fission is related to the notion that the insertion of Vocabulary items discharges the inflectional features present in the morpheme (Noyer 1997). In standard cases, the insertion procedure stops after a phonological exponent is inserted, even if this exponent discharges only a subset of the inflectional features present in the morpheme. However, when a morpheme is marked for undergoing Fission, the inflectional features that are not discharged (or ‘matched’, Halle 1997) by the first insertion operation are copied into an additional morpheme which is generated by the insertion procedure. Subsequently, this additional morpheme is then itself subject to Vocabulary Insertion. Typical examples of Fission come from languages where agreement is marked by a combination of prefixes and suffixes, as in many Afro-Asiatic languages (cf. Noyer 1997 for extensive discussion).14 This is illustrated by the following example from Tamazight Berber where the agreement morpheme “splits into three positions of exponence” (Noyer 1997: 89) which are realized by successive Fission (of a single Agr-morpheme) and insertion of the Vocabulary items in (16): (15) t-dawa-n-t 2-cure-pl-fem ‘you (pl, fem) cured’ (Noyer 1997: 89) (16) a. [2] ↔ /t-/ b. [pl] ↔ /-n/ c. [fem] ↔ /-t/ Due to the fact that morphemes created by Fission contain only a subset of the features contained in the original morpheme, Fission often results in the insertion of less marked or ‘elsewhere’ Vocabulary items, that is, “Fission extends the domain of less marked exponents” (Halle 1997: 432). The insertion of less marked exponents can also follow from context-sensitive Impoverishment rules which delete morphosyntactic features from morphemes prior to



JB[v.20020404] Prn:2/08/2005; 9:38



F: LA8102.tex / p.10 (582-642)

The Rise of Agreement

Vocabulary Insertion (cf. Bonet 1991 for the original proposal). As a result, Vocabulary items that require the presence of those features cannot be inserted and a less specified exponent must be used to realize the morpheme affected by Impoverishment (which Halle & Marantz 1994 call the “Retreat to the General Case”). An illustrative example of the workings of Impoverishment is provided by Sauerland (1996). It is a well-known fact that the Germanic languages show different paradigms for the inflection of (attributive) adjectives. The choice of the weak inflectional ending on adjectives is determined by the presence of inflection on the determiner: in cases where a definite determiner is present, a weak ending must appear on the adjective (de-r gut-e Mann ‘the good man-nom’, de-m gut-en Mann ‘the good man-dat’). However, if no determiner is present, the strong adjectival inflection must be chosen (mit gut-em Gewissen ‘with good conscience-dat’). In the presence of an indefinite article, a mixed paradigm is observed – the strong adjectival inflection appears if the article shows no agreement/inflection (ein gut-er Mann ‘a good mannom’), while the weak adjectival inflection shows up if the indefinite article carries agreement/inflection (ein-em gut-en Mann ‘a good man-dat’). Interestingly, Sauerland observes that the endings used in the weak paradigm are in fact a subset of the strong endings, namely the least marked (or default) endings of the strong inflection. He therefore concludes that the weak paradigm is the result of an impoverishment rule that deletes inflectional features in the presence of an inflected determiner (in most languages [gender] is deleted; the picture is more complex in languages with rich inflection such as German and Icelandic, see Sauerland 1996 for details). This in turn leads to the insertion of less specified forms that constitute the most unmarked, underspecified Vocabulary items of the strong paradigm.15 From the above discussion, it has become clear that DM posits a strongly syntactic approach to the formation of words: morphosyntactic primitives are combined via the syntactic operations Merge and Move and then (eventually) spelled-out by phonological exponents. This model seems to entail (or at least favor) a very complex clause structure where there is ideally a one-to-one correspondence between morphosyntactic features and terminal nodes (cf. e.g. Sigurðsson 2000, 2001, 2002; Poletto 2000 for the idea that features such as person and number head separate projections), that is, there are separate projecting nodes for individual inflectional categories such as person, number, gender etc. However, the idea that the syntactic component manipulates morphosyntactic primitives does not necessarily require that each individual feature heads its own projection (cf. Harley & Noyer 1999). First, the possibility to insert (dissociated) morphemes post-syntactically entails that not every morpheme (and thus feature) enters the syntactic computation as a projecting head. Second, if we agree on the existence of a purely morphological operation such as Fission, then it follows that the syntax must operate (at least sometimes) on bundles of morphosyntactic features which can then be split up into several morphemes by post-syntactic morphological operations. Note that Fission may give the (false) impression that this split of inflectional features/heads is located in the syntax. Thus, we can maintain our assumption that there is a small core inventory of functional nodes (C, T, ν, D) which is universally

JB[v.20020404] Prn:2/08/2005; 9:38

F: LA8102.tex / p.11 (642-709)

Chapter 2. Theoretical preliminaries

present in each human language. Wider arrays of functional categories are possible (such as e.g. split-CP, Rizzi 1997), but only triggered if there is relevant evidence in the input data during language acquisition. Still, morphological operations such as Fission and the insertion of dissociated morphemes may give rise to the (wrong) impression of a richer system of functional heads present in the syntax. After establishing the main traits of the theoretical framework adopted in this book, the remainder of this chapter presents a set of assumptions on the theory of language change (and variation) and in particular the phenomenon of grammaticalization that lay the foundations for the diachronic analyses developed in the Chapters 4–6.

. Language change: A generative perspective In the last twenty years, the investigation of diachronic phenomena has developed into a productive and well-established sub-discipline of generative linguistics (for an overview of more recent work cf. the volumes edited by Battye & Roberts 1995; Beckman 1995; van Kemenade & Vincent 1997; Pintzuk, Tsoulas, & Warner 2000, Lightfoot 2002). Following the pioneering work of scholars such as David Lightfoot, Anthony Kroch and Ian Roberts, there is by now general agreement on the notion that the study of language change can provide important insights into the properties of Universal Grammar that cannot be gained from a purely synchronic perspective. This conviction stems from the insight that the triggers for language change are located in the workings of language acquisition, that is, the interaction between innate principles of grammar and the linguistic evidence the child encounters, reviving ideas that were already present in the work of neo-grammarians such as Hermann Paul (e.g., Paul 1880). From this point of view, language change reflects the “limits to attainable grammars” (Lightfoot 1991: 172), in the sense that language change reveals how certain grammarspecific choices which give rise to individual languages are selected on the basis of the evidence available to the learner. In the Principles and Parameters framework, (syntactic) variation between individual languages is attributed to different settings for a limited number of parameters that are associated with invariable principles of Universal Grammar (UG) (Chomsky 1981, 1986a). The task of acquiring a given grammar then consists of filling in the gaps left open by the principles of UG, that is, detecting the parameter settings which are reflected by the linguistic input the child is confronted with. The stimulus that serves to set a given parameter one way or other is usually called trigger experience or cue (see Lightfoot 1991, 1999 for discussion). Apart from accounting for synchronic parametric differences between different individual languages (such as e.g. present-day English and Italian), this approach can also be applied to diachronic parametric differences between different historical stages of a single language (such as e.g. Old English and present-day English). Accordingly, syntactic change is often conceived of as a change that affects the parameter settings for a given language.16



JB[v.20020404] Prn:2/08/2005; 9:38



F: LA8102.tex / p.12 (709-754)

The Rise of Agreement

Due to the fact that a single parameter typically determines a whole complex of (syntactic) properties, a single parametric change manifests itself in a variety of complex effects and at times dramatic changes at the syntactic surface. In other words, parametric change is often abrupt and ‘catastrophic’ in nature (Lightfoot 1991, 1999). An instructive example of the workings of parametric change is given in Roberts (1993a) who shows that in the history of French, three apparently independent changes (the loss of V2, null subjects, and so-called ‘simple inversion’ in interrogatives) can actually be analyzed as reflexes of a single parametric change that affected the mechanism of nominative case assignment. Furthermore, Roberts links the syntactic change in question with morphophonological changes that led to the loss of distinctive inflectional endings. Thus, morphological properties of individual lexical items are assumed to play an important causal role for setting off changes that affect the syntactic component of the grammar (see below for details). This assumption is in line with recent approaches to parametrization which take parameters to be lexical, that is, they are associated with properties of individual lexical items (the Lexical Parametrization Hypothesis, cf. Borer 1984; Wexler & Manzini 1987). In the late 1980s, the Lexical Parametrization Hypothesis converged with another theoretical development within the framework of Principles and Parameters theory that aimed at implementing the traditional distinction between lexical and grammatical categories directly into the structure of the clause. Following work by, for example, Fukui (1986), Chomsky (1986b), and Abney (1987), grammatical categories like determiners, conjunctions and inflections were associated with a closed class of functional heads that project their featural content in accordance with universal principles of phrase structure (X-bar theory or Bare Phrase Structure). Initially, in the verbal/clausal domain only two functional heads were distinguished: C (conjunctions, clause type, subordination) and Infl (verbal inflection, finiteness, nominative case assignment). In later work by Pollock (1989), Belletti (1990), Chomsky (1991), the category Infl is split further into distinctive AgrS (subject agreement, nominative case assignment), T (tense, finiteness), and AgrO (object agreement, accusative case assignment) heads. In the 1990s, this line of research has led to a sudden increase of the number of functional categories (cf. e.g. Cinque 1999). By now, most researchers agree on the existence of a universal inventory of core functional categories that consists of the elements C (clause type, subordination), T (tense, subject-verb agreement, nominative assignment), ν (voice, transitivity, accusative assignment, object agreement) and D (nominal inflection, definiteness) (cf. Chomsky 1995, 2000, 2001a, 2001b). Note, however, that many scholars today argue for the existence of more elaborate systems that include a fine-grained structure of inflectional/functional heads in the Infl and C domain (cf. Rizzi 1997; Cinque 1999, among others). At least with respect to the set of core functional categories it is assumed that they are universally present in natural languages as basic ‘building blocks’ or ‘skeleton’ of clause structure and that they trigger syntactic operations in order to license their (abstract) morphological content (cf. Chomsky 1991, 1995, 2000). On these assumptions, it seems natural to analyze parametric variation between individual languages

JB[v.20020404] Prn:2/08/2005; 9:38

F: LA8102.tex / p.13 (754-822)

Chapter 2. Theoretical preliminaries

in terms of varying (morphological) feature specifications of a (universally) given set of functional categories (cf. e.g. Chomsky 1991, 1995; Ouhalla 1991): If substantive elements (verbs, nouns, etc.) are drawn from an invariant universal vocabulary, then only functional elements will be parametrized. (Chomsky 1991: 419)

In this approach to parametrization, the computational system (CHL ), that is, the mechanisms that build up syntactic hierarchical structures by combining individual lexical items, is universal and therefore cross-linguistically and diachronically invariable (Chomsky 2000, 2001a, 2001b). Thus, strictly speaking, there is no such thing as ‘syntactic change’: the properties of the syntactic component of grammar remain constant over time (cf. Hale 1996, 1998; Longobardi 2001; Keenan 2002). Apparent ‘syntactic change’ (and synchronic differences between languages) results from changes affecting the feature content of a closed class of functional categories such as C, T, ν and D (e.g., via phonological erosion or grammaticalization processes). It is expected that a single change in the featural properties of these core functional categories results in a complex of (at times dramatic) distinct changes on the syntactic surface, which is a hallmark of parametric change (cf. Lightfoot 1979, 1991, 1999). Moreover, if overt inflectional morphology is taken to reflect the (abstract) feature content of functional categories, it is possible to construe a correlation between parametric change and morphological change, that is, to provide a principled explanation for the traditional observation that changes affecting the inflectional morphology of a given language often go hand in hand with syntactic change (cf. e.g. Paul 1880; Sapir 1921). In the generative literature on language change, this line of research has proven to be very productive, covering for example the historical impact of the loss of verbal inflection on the availability of verb movement (for the history of English cf. Roberts 1993a; for the Scandinavian languages cf. Platzack 1988 on Swedish) and pro-drop (for the history of French cf. Roberts 1993a; Vance 1997; for Swedish Falk 1993; for English Allen 1995; Haeberli 1999), or the relation between the loss of nominal inflections (i.e., case) and changes affecting argument order and the rise of ECM constructions (for the history of English cf. Lightfoot 1979, 1991, 1999; van Kemenade 1987; Roberts 1997; Kiparsky 1997; Haeberli 1999). However, while the loss of inflectional morphology (and its syntactic consequences) has received quite some attention, the related question of how inflectional morphology arises historically has remained largely unadressed in generative works, although the latter constitutes a major field of interest in traditional and typological approaches to language change (see Section 2.5 for discussion). Still, the generative approach to language change has not gone unchallenged. In particular the concept of parametric change seems to be in conflict with two widelyheld views on language change and acquisition. First, it is a traditional observation that language change is a gradual process. A given change may go on over many centuries before it is completed (cf. e.g. the change from OV to VO in the history of English, Lightfoot 1979, 1991, 1999; Kemenade 1987; Roberts 1993a; Pintzuk 1999; Trips 2002;



JB[v.20020404] Prn:2/08/2005; 9:38



F: LA8102.tex / p.14 (822-885)

The Rise of Agreement

Fuß & Trips 2002). At first sight, this trait of language change does not seem to be compatible with the abrupt character of parametric change. Another problem arises from the assumption widely held in generative linguistics that language acquisition proceeds in a flawless fashion. In other words, most researchers (working in the generative paradigm) agree that children always succeed in acquiring the target grammar that generates the linguistic data they are exposed to, even if this data is apparently flawed and insufficient (sometimes called the “the logical problem of language acquisition”, cf. Chomsky 1986a for discussion). This leads to the “logical problem of language change” as Niyogi and Berwick (1998) choose to call it: After all, if all children successfully attain the grammars of their parents and they continue to do this generation after generation, then the linguistic composition of every generation would look exactly like the linguistic composition of the previous generation and languages would not change with time. Yet they do. (Niyogi & Berwick 1998: 192f.)

Thus, more has to be said to reconcile the idea of parameter change with (i) the apparent gradualness of language change and (ii) the apparent perfection of language acquisition. To answer the latter problem, it is usually assumed that for some reason, the trigger experience that resulted in a given parameter setting in the parent grammar has become obscure or ambiguous in the output of this grammar (cf. Lightfoot 1979, 1991) due to factors such as language contact, (morpho-) phonological erosion or syntactic reanalyses that blur the evidence for certain parametric choices in the linguistic input. Of course, this raises a number of further questions concerning the nature of the trigger experience (see below for some discussion). Note furthermore that ambiguity of trigger experience is only a necessary condition for a change to take place. It is still left unclear what actually motivates parametric change. Here, a widespread line of thinking assumes that principles of economy take over and ultimately decide if the trigger experience contained in the input is truly ambiguous (Clark & Roberts 1993; Roberts 1993b; Roberts & Roussou 2003, among others). In general, these economy considerations come in two varieties: first, some researchers assume that there are marked and unmarked (or default) parameter values and that the learner assigns a given parameter the unmarked value if no decision can be made based on the evidence available in the input (cf. Wexler & Culicover 1980; Berwick 1985 for discussion). A related idea lies behind the Subset Principle (Berwick 1985: 37; Wexler & Manzini 1987: 61):17 (17) The Subset Principle (O’Grady 1997: 283) The acquisition device selects the most restrictive parametric value consistent with experience. Thus, if there are two possible settings A and B for a given parameter, and A generates a subset of the sentences generated by B, the learner will acquire the more restrictive setting A in the absence of clear trigger experience. In this sense, A can be viewed as the unmarked setting.

JB[v.20020404] Prn:2/08/2005; 9:38

F: LA8102.tex / p.15 (885-925)

Chapter 2. Theoretical preliminaries

Alternatively, the decision in question is assumed to be sensitive to the notion of derivational/representational economy (e.g., a ‘Least Effort Strategy’, Clark & Roberts 1993), in the sense that the learner assigns a given input string the most economical representation/derivation that is compatible with the data. If the manifestation of a given marked parameter value is ambiguous in the input, the learner will choose the parameter value that guarantees the most economical representation/derivation. These considerations are usually employed to account for the diachronic loss of movement operations (if the data is compatible with a non-movement analysis), as for example the loss of V-to-I or object movement (for Case licensing) in the history of English (cf. e.g. Clark & Roberts 1993; Roberts 1993a, 1995, 1997, 1999a). Most recently, Roberts and Roussou (2003) argue that the kind of structural simplification associated with the loss of movement operations is a key feature of grammaticalization processes, in the sense that an element previously moved to the vicinity of a functional head X is reanalyzed as the exponent of X (see this chapter, Section 2.5 and Chapters 4 and 5 for discussion). A different approach is presented in Lightfoot (1999) who argues for a ‘cue-based’ theory of language acquisition, originally developed by Dresher and Kaye (1990) and Dresher (1999) for the acquisition of phonological properties such as stress patterns. The basic assumption is that UG contains not only a set of parameters, but also specifies for each parameter a cue that serves to switch the parameter one way or other (cf. Fodor 1998 for a related approach). According to Lightfoot, a cue is not directly present in the input, but rather part of the abstract mental representations/structures derived by parsing the input (i.e., a piece of ‘I-language’ in the sense of Chomsky 1986a). For example, Lightfoot assumes that the cue for the V2 property is SpecCP [XP], XP an arbitrary phrasal category that occupies SpecCP. This form of abstract trigger experience is contained in a (incomplete) structural description of an input string, but not in the input itself. In other words, a cue can only be detected if the learner has assigned a given input string a structure. If the learner detects a cue that is attested robustly in these (initially incomplete) parses, this will activate a given parameter or syntactic operation in the learner’s grammar.18 In other words, cues are ‘points of variation’ the absence/presence of which determines differences between individual grammars. Language change results either if a given syntactic operation fails to be cued or if it starts to be cued, in contrast to the target grammar. Thus, instead of assuming the existence of binary parameters, it is perhaps more appropriate to speak of ‘basic’ or ‘extended’ parametric choices, where the cues for the latter are present only in certain grammars. From this point of view, language change is a contingent and unpredictable process that depends on the availability of certain cues in the linguistic input the learner is exposed to. Often, a given parametric choice is associated with a set of cues that may be syntactic or morphological in nature. For example, it is usually assumed that the cue for V-to-I movement consists of (i) syntactic evidence that the verb has moved out of VP (i.e., the configuration Infl [Vfin ]) and (ii) morphological evidence, namely the presence of rich verbal agreement morphology.19 In the context of the present study, then, the question arises which cues are involved in the acquisition of agreement morphol-



JB[v.20020404] Prn:2/08/2005; 9:38



F: LA8102.tex / p.16 (925-995)

The Rise of Agreement

ogy. In Chapter 4 below, I will discuss a set of morphological and syntactic signals that help the learner to decide whether a given element should be categorized as a piece of verbal inflection or a pronominal clitic. Crucially, the expression of cues interacts in intriguing ways with core syntactic properties of grammar. Thus, we should expect that the categorial reanalysis of a pronoun may at times provoke catastrophic changes in other parts of the grammar (see Chapters 4 and 5 for discussion and relevant examples). This is in line with Lightfoot’s tenet that the learner does not try to match the input (in contrast to most other approaches, cf. e.g. Clark & Roberts 1993; Gibson & Wexler 1994), but rather scans structures for cues that trigger certain parametric choices, “without regard to the final result” (Lightfoot 1999: 149), that is, whether the new grammar will differ from the target grammar or not. The latter aspect fits in well with the view that language change is an abrupt process that may at times lead to catastrophic consequences, but seems to be at odds with the often noted apparently gradual nature of language change, which brings us back to the first problem noted above. It has been argued that the often gradual character of language change represents a serious problem for the concept of parametric change which predicts that languages should change rather abruptly (cf. Hopper & Traugott 1993; Harris & Campbell 1995 for discussion). In what follows, however, it is shown that the impression of gradualness arises from two factors which traditional approaches failed to recognize and that we can therefore maintain the concept of parametric change. First, I demonstrate that the impression of gradualness disappears under a sufficiently constrained notion of ‘grammar’ (and ‘grammar change’) as the proper object of linguistic study. Second, it is claimed (following the pioneering work by Kroch 1989) that language change typically proceeds via a stage of ‘internal bilingualism’, where generations of speakers have command over more than one internalized grammar giving rise to a degree of linguistic variation which gradually disappears when one grammar wins out over the other over time. In generative linguistics, there is general agreement that the object of fruitful linguistic study is a special notion of ‘language’, namely the linguistic knowledge of an idealized speaker/hearer, cf. Chomsky (1965: 3) for a classic statement: Linguistic theory is concerned primarily with an ideal speaker-listener, in a completely homogenous speech-community, who knows its language perfectly and is unaffected by such grammatically irrelevant conditions as memory limitations, distractions, shifts of attention and interest, and errors (random or characteristic) in applying his knowledge of the language in actual performance.

On this conception of ‘language’, the proper object of linguistic study is generally referred to as linguistic competence or simply grammar. Accordingly, the proper object of the linguistic study of language change must be defined as a change between (individual) grammars, which I will refer to as ‘grammar change’ (cf. Hale 1996; Lightfoot 1999). It is fairly clear that in this sense, language change is necessarily an abrupt phenomenon, namely a clearly identifiable difference between a grammar A and a

JB[v.20020404] Prn:2/08/2005; 9:38

F: LA8102.tex / p.17 (995-1049)

Chapter 2. Theoretical preliminaries

grammar B. Hence, the impression of gradualness is in fact a misperception that arises from confusing the actual change process with the (highly gradual) diffusion of this change within a population/linguistic community.20 The actual change is not a sociolinguistic phenomenon, but rather the result of cognitive processes that determine the process of language acquisition, resulting in a grammar in the mind of the individual speaker that differs from the target grammar. Only under this restricted interpretation of language change as grammar change we can hope that the inspection of language change can reveal something about the structure and the workings of the language faculty/UG. Accordingly, this study focuses on instances of grammar change, that is, given the subject matter of this work, cases where the categorial status of a pronominal element changes in the ‘new’ grammar acquired by the learner. The issue at hand is also taken up in the work of Anthony Kroch, who tries to reconcile a formal approach in terms of parametric change with the apparent gradualness of language change. It is a well-known observation that language change usually proceeds via a (intermediate) stage where old and new (i.e., changed) linguistic forms co-exist side by side, leading to a degree of variation which is not encountered in ‘stable’ linguistic communities. Furthermore, it has been noted that the loss of the old form and its eventual replacement by the new form is a gradual process which at times may extend over many centuries.21 In a series of publications (cf. e.g. Kroch 1989, 1994, 2001), Kroch has developed a formal account of these observations which is based on the Principles and Parameters framework. Here, the notion of grammar competition represents the core concept of an integrated theory of language change and variation. The basic idea of Kroch’s approach is that parametric change must always proceed via a stage where the speaker (or, a generation of speakers) of a language X has access to more than one internalized grammar (sometimes referred to as an instance of internal diglossia). The grammars in question may differ in a number of parametrical choices, giving rise to a wider range of linguistic variation. However, blocking effects imposed by UG (see Kroch 1994 for details; cf. Aronoff 1976 on blocking effects on morphological doublets) restrict the co-existence of grammars that differ only minimally with respect to a set of parameter doublets (i.e., co-existing competing values for a single parameter), thereby warranting that one grammar will eventually win out over its competitors, which completes the change in question. We can therefore conclude that the apparent gradualness of language change does not impose a challenge on the concept of parametric change. Rather, change between grammars is necessarily abrupt but may involve a stage where speakers have command over more than one internalized grammar, giving rise to an unusual degree of linguistic variation, and, over time, the impression that language change is gradual when one parametrical choice wins out over the other. The next section discusses the phenomenon of grammaticalization, the prime example of an allegedly gradual change, and shows how this diachronic process can be subsumed under the model of language change outlined in the present section.



JB[v.20020404] Prn:2/08/2005; 9:38



F: LA8102.tex / p.18 (1049-1102)

The Rise of Agreement

. Grammaticalization It is an important insight of 19th century linguists such as Bopp (1816), Schlegel (1818), Humboldt (1822), (1836), Müller (1875), Paul (1880), Gabelentz (1891), or Brugmann (1916) that grammatical categories like determiners, inflections, conjunctions or auxiliaries evolve historically from formerly (free) substantial lexical categories such as nouns or verbs. The term grammaticalization is first coined in the work of Meillet (1912, “L’evolution des formes grammaticales”, republished 1965: 133) to describe this type of change. A more recent characterization of grammaticalization defines it as [...] the process whereby lexical items and constructions come in certain linguistic contexts to serve grammatical functions, and, once grammaticalized, continue to develop new grammatical functions. Thus nouns and verbs may change over time into grammatical elements such as case markers, sentence connectives, and auxiliaries. (Hopper & Traugott 1993: i)

From the 1980s on, the study of grammaticalization processes has become a main focus of descriptive diachronic/typological linguistics, leading to a wealth of new data and a set of generalizations on the course and defining properties of grammaticalization (major publications include Heine & Reh 1984; Traugott & Heine 1991; Heine, Claudi, & Hünemeyer 1991; Hopper & Traugott 1993; Ramat & Hopper 1998; Heine & Kuteva 2002; Wischer & Diewald 2002; Lehmann 2002). Recently, the issue is taken up in generative work as well, cf. for instance Roberts (1993b), von Fintel (1995), Newmeyer (1998, 2001), Roberts and Roussou (1999, 2003), Lightfoot (1999), Abraham (2003), (2004), and van Gelderen (2004a, b). Traditional approaches stress that grammaticalization is a highly gradual process, changing slowly but surely the syntactic category of a lexical item along universal historical paths, called grammaticalization clines or paths (cf. Hopper & Traugott 1993; Heine & Kuteva 2002; Lehmann 2002 for overviews). A typical example is the cline that characterizes the development of agreement markers (see Chapter 1, example (2), repeated here for convenience): (18) independent pronoun → weak pronoun → clitic pronoun → affixal (agglutinative) agreement marker → fused agreement marker → Ø Often, different stages of a grammaticalization cline are taken to represent a continuum between, for example, pronouns and agreement markers.22 Accordingly, it is often claimed that syntactic categories are not discrete units, but rather fuzzy theoretical entities. Note that these properties conflict with standard generative assumptions and the notion of grammar (or parametric) change introduced in the previous section. In particular, the gradual character of grammaticalization processes and the apparent existence of universal pathways of change seem to be incompatible with the idea that grammar change proceeds in the non-deterministic, ‘bumpy’ fashion envisaged in the work of David Lightfoot, for example. In what follows, however, we will see

JB[v.20020404] Prn:2/08/2005; 9:38

F: LA8102.tex / p.19 (1102-1175)

Chapter 2. Theoretical preliminaries

that it is possible to reconcile these apparently contradictory views if certain theoretical assumptions on language change in general and grammaticalization processes in particular are adopted (cf. Hale 1996; Lightfoot 1999). This study adheres to the idea that syntactic categories represent discrete units. Accordingly, the apparent gradual nature of grammaticalization processes is considered a superficial impression that results from the factors discussed in the previous section, that is, mixing up the notions of grammar change and the (sociolinguistic) diffusion of a change and failing to recognize the role of grammar competition in grammar change. As noted above, the existence of more than one internalized grammar may generate an unusual degree of variation in the E-language of an individual. If this type of variation concerns the syntactic category of a given element (e.g., a clitic in grammar A and an agreement marker in grammar B), then this will create the superficial impression of a highly gradual change process involving indiscrete or fuzzy category boundaries. The study of grammaticalization processes has led to the identification of another set of properties which are characteristic of this type of change such as gradual loss of phonological and semantic content, a development from free into bound forms, an increasingly fixed position etc. In this book, I maintain that these seemingly distinct properties or processes can receive a uniform formal explanation on the assumption that grammaticalization is to be analyzed as the reanalysis of substantial lexical elements as exponents of (higher) functional categories/heads (basically following von Fintel 1995; Roberts & Roussou 1999, 2003). This change may also affect functional elements, which then turn into other functional elements, as in the case of clitics (presumably realizations of D0 ) that turn into agreement markers (cf. Roberts & Roussou 2003). Below, it is shown that on this assumption, many characteristics of grammaticalization processes can be directly attributed to properties of functional categories. For example, if phonological/prosodic deficiency is a universal property of functional categories, then a given lexical element must undergo a certain degree of phonological reduction before it can be reanalyzed as the realization of a functional head.23 In the following, this is illustrated with a selection of typical properties of grammaticalization processes (cf. Hopper & Traugott 1993; Lehmann 2002 for detailed overviews). It is a well-known observation that grammaticalization processes are characterized by the gradual loss of phonological and semantic substance, often referred to as phonological attrition and semantic bleaching, respectively. The reduction of phonological substance can be characterized as the loss of segments or of marked phonological features (cf. the reduction of Latin ille to French le). The process of phonological reduction is usually accompanied by a loss of prosodic independence which proceeds along the following pathway (cf. Hopper & Traugott 1993: 7) (19) content item > grammatical word > clitic > inflectional affix > phoneme > Ø It has often been observed that functional elements such as complementizers, determiners, or auxiliaries (and, of course, bound elements such as inflections) are prosodically/phonologically deficient, since they cannot bear stress and are preferably monosyllabic/moraic elements in contrast to lexical categories (cf. Kenstowicz 1994



JB[v.20020404] Prn:2/08/2005; 9:38



F: LA8102.tex / p.20 (1175-1229)

The Rise of Agreement

on English; Vogel 1999 on Italian; Roberts & Roussou 2003).24 Thus, the constant loss of phonological content is expected on the assumption that grammaticalization is to be analyzed as the transition from lexical elements to functional categories. It seems plausible to assume that the absence of stress (via an intermediate clitic stage) and a reduced segmental make up is a necessary precondition for the reanalysis as a functional head to take place (cf. Roberts & Roussou 2003 for discussion). Moreover, after the reanalysis is completed, it is predicted that the element is prone to undergo further phonological reduction due to the deficient phonological nature of functional categories, accounting for the pathway described by (19). More or less directly related to the general property of phonological attrition is the notion of coalescence which refers to the fact that grammaticalization typically involves the development of bound forms from formerly free forms, cf. (19). Again, this observation can be attributed to the general phonological deficiency of functional categories, which are often realized by affixes or clitic elements which require the presence of a (lexical) host they can attach to. Note that the existence of free inflectional markers (as in many Creole languages for example) indicates that the development of bound forms is not a necessary consequence of the reanalysis of lexical material as functional heads, but rather a tendency, caused by the phonological deficiency of functional categories. Whereas phonological attrition is a rather straightforward process, the exact meaning of the notion ‘semantic bleaching’ cannot be pinned down that easily. Intuitively, it designates the erosion of substantial lexical meaning. For example, nouns evolving into nominalizing affixes (or, in our case, pronouns evolving into agreement markers) lose their referential (and descriptive) capacity (such as German -heit < Old High German heit, Gothic haidus ‘person, nature, form, rank’), while full verbs evolving into auxiliaries or tense markers lose their thematic (and predicational) properties (cf. Old English sceal, full verb ‘subject has to pay an amount of money or has to return something to somebody’ > ‘(root) modal’ > ‘future marker’, Lehmann 2002: 114). Following proposals by von Fintel (1995) and Roberts & Roussou (2003), these seemingly distinct semantic changes can be subsumed under a unified account that is based on the following two insights: First, the loss of semantic content characteristic of grammaticalization does not affect the semantic properties of a given lexical item in a random fashion. Rather, the semantic erosion always concerns a certain subset of semantic properties: verbs lose their argument structure (or, their thematic properties), nouns their referential (and descriptive) content, prepositions their capacity to designate spatial relations etc. This can be subsumed under the notion that grammaticalization involves the loss of ‘predicative’ or ‘descriptive’ content (see Roberts & Roussou 2003 for examples and extensive discussion). Second, another set of semantic properties is preserved or added to the formerly substantial lexical category as a result of the grammaticalization process. This set of semantic properties can be termed ‘logical content’. Examples include the retention of modal content in the development of modals/auxiliaries from formerly full verbs (cf. the case of Old English sceal noted above) or the addition of the feature [±definiteness] in the case of determiners

JB[v.20020404] Prn:2/08/2005; 9:38

F: LA8102.tex / p.21 (1229-1297)

Chapter 2. Theoretical preliminaries

evolving from nouns or numerals. Thus, the semantic changes typical of grammaticalization are not adequately described by the very broad term ‘semantic bleaching’, since they show a set of clearly structured properties and, even more important, the retention or even addition of logical semantic content. According to von Fintel (1995), the distinctive feature that differentiates logical from predicative or descriptive content is ‘permutation invariance’, which can be characterized as follows: The intuition is that logicality means being insensitive to specific facts about the world. For example, the quantifier all expresses a purely mathematical relationship between two sets of individuals (the subset relation). Its semantics would not be affected if we switched a couple of individuals while keeping the cardinality of the two sets constant. There couldn’t be a logical item all blonde because it would be sensitive to more than numerical relations. (von Fintel 1995: 179)

More generally, logical content is associated with quantificational properties. Interestingly, quantificational properties can be shown to be a characteristic of functional categories, cf. the following passage taken from Roberts and Roussou (2003: 223): Many functional elements clearly have logical meanings, in a sense that seems very close to that defined above. This is clearly true of quantificational elements in DP (occupying D, Q or Num). It is true also of modal elements, to the extent that these quantify over possible worlds. It is also true of negation, as this just defines a complement relation between sets. It may be true of Tense and Aspect, to the extent that these notions can be construed as quantification over times or events. Alternatively, Tense may be an ordering predicate (Stowell 1996), another kind of logical relation (see above). Complementizers, to the extent that they may be factive or irrealis, are connected to modality.25

In addition, functional meanings have high semantic types, in contrast to lexical meanings, which have low semantic types. To take an example from von Fintel (1995: 184), an adjective (of the type <<e,t>,<e,t>>) must assume a more complex semantic type if it develops into a determiner (<<e,t>,<<e,t>,t>>). Again, these observations suggest that an analysis of grammaticalization as the reanalysis of lexical elements as functional categories is on the right track. If functional categories are characterized by (and limited to) a certain kind of semantic content (i.e., logical meanings and high semantic types, cf. von Fintel 1995), then it is expected that the transition from lexical to functional category requires the loss of those semantic properties that are not compatible with the universal make up of functional categories, that is, non-logical (i.e., predicative or descriptive) content, and the retention or addition of logical content. Exactly these types of semantic changes are a characteristic trait of grammaticalization processes.26 As von Fintel (1995: 184) puts it: When a lexical item wants to become a functional item, it needs to become permutation-invariant, it needs to shed any reference to particular entities, properties, or situations in the world. The item will also have to assume a high semantic type. Lastly, it will have to conform to whatever semantic universals apply to the particular functional category it wants to belong to.



JB[v.20020404] Prn:2/08/2005; 9:38



F: LA8102.tex / p.22 (1297-1345)

The Rise of Agreement

Apart from the loss of thematic properties, grammaticalization also results in increasingly restricted selectional properties (i.e., a simplified subcategorization frame), which typically involves a ‘decrease of structural scope’ (Lehmann 2002), that is, a development from syntactic selection towards morphological selection. Traditionally, this characteristic is referred to as condensation. A good example of this process is the development of the Romance synthetic future, which started as a main verb (habere) that took a non-finite clause as its complement, passed through a stage as a futureindicating auxiliary selecting a VP and finally became an inflectional suffix with its (morphological) subcategorization frame reduced to a verb stem. Roberts and Roussou (1999) argue that this development can also be attributed to the universal profile of functional categories, on the assumption that only functional categories can select for lexical categories, whereas lexical categories cannot select for other lexical categories. In other words, it seems that grammaticalization always proceeds in an upwards fashion, that is, exponents of hierarchically lower lexical categories are reanalyzed as exponents of higher functional heads. According to Roberts and Roussou (2003), this is a key property of grammaticalization processes in general: “Successive upwards reanalysis along the functional hierarchy is thus how we define grammaticalization paths.” (p. 202). The idea that categorial reanalysis always leads to a new exponent of higher functional head is captured by the following structural configuration (Roberts & Roussou 2003: 200), where a lower element Y turns into the realization of a higher functional head X: (20)

XP Y=X

... YP Y

... (YP does not have to be the complement of X)

Roberts and Roussou (2003) explain this particular trait of grammaticalization processes by assuming that grammaticalization is typically an instance of a Move > Merge reanalysis, in which an element moved into the vicinity of a functional head F is reanalyzed as the phonological realization of F (i.e., a movement dependency is converted into an instance of Merge).27 To take a concrete example, the development of verbal agreement markers is analyzed as a two-step process in which a pronominal element previously moved to SpecAgrSP is first reanalyzed as a subject clitic which is merged as an instantiation of AgrS (Roberts & Roussou 2003: 178):28 (21) [AgrSP DPi [AgrS’ V ] [TP . . . [VP ti ... →[AgrSP [AgrS’ D+V ] ... In a subsequent change, D may undergo further reduction and develop into an agreement affix which is part of V: (22) [AgrSP [AgrS’ D+Vj ]... tj ...] → [AgrSP [AgrS’ Vj ]... tj ... ]

JB[v.20020404] Prn:2/08/2005; 9:38

F: LA8102.tex / p.23 (1345-1404)

Chapter 2. Theoretical preliminaries

In the resulting structure, the content of AgrS is licensed by moving the inflected verb to this position (in Roberts and Roussou’s terms, this change involves a transition from AgrS*Merge+Move to AgrS*Move ). While I agree with the gist of this particular approach to grammaticalization, there are still several points where the present work deviates from Roberts and Roussou’s specific proposals. For example, I do not share the assumption that subject clitics are merged as agreement heads, since this assumption implies that inflectional heads may bear a θ-role (Roberts & Roussou 2003: 183), a consequence which I do not find particularly attractive. Instead, I adopt the common idea that subject clitics receive their θ-role in the VP (or, rather SpecνP) and move up into the inflectional domain in order to be licensed. Furthermore, for reasons that will become clear in Chapter 3, I assume that agreement morphemes do not head their own projection in the syntax. Moreover, in Chapters 4 and 5, it is argued that there are cases which contradict the claim that the grammaticalization of agreement markers always proceeds in an upwards fashion. More specifically, it is shown that in a set of German and Rhaeto-Romance dialects, new agreement markers develop first as the realization of an agreement morpheme confined to the C-domain and spread later to lower verbal positions (i.e., T0 ). Obligatorification and fixation are two other traits of grammaticalization processes that can be directly related to syntactic properties of functional categories. The term obligatorification refers to the observation that lexical material subject to grammaticalization becomes more and more obligatory in a certain syntactic context. For example, in the often documented development from demonstratives to determiners (as in Germanic), the demonstrative is initially restricted to contexts where it adds an indexical meaning to the noun it modifies. In contrast, the resulting determiners are always obligatory.29 This follows if functional categories are analyzed as building blocks of syntax which are universally present (either with phonological content or zero). Thus, if D0 has a PF-realization /α/, then /α/ will show up in all NPs/DPs since D0 is a necessary component of a nominal expression (see von Fintel 1995 for a semantic motivation of the universal presence of D0 in noun phrases). The term fixation refers to the fact that grammaticalized elements tend to occur in a fixed position. This can be neatly illustrated with the development of agreement markers. Whereas free pronouns can occupy a variety of different positions (IP-internal, fronted to clause-initial position etc.), clitics are usually confined to certain structural positions (e.g., the Wackernagel position) or certain hosts (most often the finite verb), but still may at times switch positions inside the clitic complex. Finally, agreement markers typically show no variation at all, occupying fixed positions/slots in the verbal inflection. Again, this is expected on the assumptions (i) that the order/hierarchical position of functional categories is universally fixed (Kayne 1994; Cinque 1999; Chomsky 2000) and (ii) that the relative order of inflectional affixes is determined by the hierarchical structure of head complexes generated by successivecyclic adjunction of (inflectional) functional heads (i.e., the Mirror Principle, Baker 1988). In other words, if grammaticalization involves the reanalysis of lexical elements



JB[v.20020404] Prn:2/08/2005; 9:38



F: LA8102.tex / p.24 (1404-1466)

The Rise of Agreement

as exponents of functional heads, it is expected that this change leads to a more and more fixed position of the grammaticalized material. Another property of grammaticalization which can be related to universal morphological differences between lexical and functional categories is paradigmatization, the development of an open class lexical element into an element that belongs to a closed class of functional words or is integrated into a morphological paradigm typically exhibiting a high degree of formal, functional and semantic homogeneity. In recent generative approaches to the interface between syntax and morphology, it is usually assumed that paradigmaticity is associated with functional categories. The relevant difference between lexical and functional categories becomes particularly clear in Late Insertion models such as Distributed Morphology (cf. Section 2.3 above). In contrast to substantial lexical heads (or l-morphemes), which are subject to free-choice spell-out, different Vocabulary items ‘belonging to the same paradigm’ (note that the notion of ‘paradigm’ is only a descriptive artifact here) compete for the insertion into functional heads (or f -morphemes; see Section 2.3 above for details). Items ‘belonging to the same paradigm’ form a natural class due to the fact that they serve to realize an identical subset of inflectional features. The item realizing the greatest number of inflectional features is chosen for insertion. On these assumptions, the notion of paradigm is directly related to the insertion procedure for functional heads which differs significantly from the way Vocabulary Insertion proceeds in the case of lexical heads. Thus, increasing paradigmaticity is an expected outcome of the reanalysis of lexical as functional categories (cf. Chapter 6 for some discussion). In brief, it seems that an analysis of grammaticalization as the transition from lexical material to functional categories is on the right track since it provides a unified explanation for basic characteristics of this type of change, attributing them to the universal profile of functional categories. However, the notion ‘transition from lexical material to functional categories’ is in fact rather vague. Some clarifications are in order, concerning, for example, whether it is really correct to say that a lexical category X changes into a functional category Y or whether grammaticalization can create ‘new’ functional categories absent in the target grammar. It is fairly clear that the grammaticalization of a given lexical element as a functional head does not lead to the disappearance of this lexical element. For example, the development of the Romance synthetic future from inflected forms of habere was not accompanied by the loss of the verb habere. Similarly, the grammaticalization of the complementizer dass/that from demonstratives in Germanic did not lead to the loss of the demonstratives. Instead, the lexical sources continue to exist in the presentday languages. Therefore, the often encountered statement that a lexical category X becomes a functional head is actually misleading. For instance, in the case of the Germanic declarative complementizer, the lexical item that [D0 ] did not turn into a lexical item that [C0 ]. Rather, a new lexical item that [C0 ] came into existence, with that [D0 ] still part of the lexicon. This innovation was presumably first restricted to syntactic contexts where the relevant evidence was ambiguous, that is, which were compatible with an analysis either as complementizer or as demonstrative. In all other contexts,

JB[v.20020404] Prn:2/08/2005; 9:38

F: LA8102.tex / p.25 (1466-1538)

Chapter 2. Theoretical preliminaries

that was still perceived as a demonstrative. Note that these considerations seem to suggest that grammaticalization can in fact create new functional categories. In turn, it is argued that this is not the case, at least with respect to the core functional categories C, T, ν and D. In current generative approaches to clause structure there is general agreement upon the existence of a set of core functional categories, which, following Chomsky (1995, 2000), are usually taken to be C, T, ν and D (see above). More precisely, it is assumed that these categories represent the skeleton of clause structure, which is universally present in all natural languages (cf. Rizzi 1997 and in particular Cinque 1999 for more elaborate arrays of functional categories). Support for this assumption comes from the consideration that functional categories are needed as a ‘glue’ to combine lexical categories in natural languages. Morphosyntactically, this ‘glue’ is represented by inflections, case-endings and complementizers. Semantically, functional categories are required to interpret combinations of lexical items with low semantic types: “[...] semantically we need functional meanings as a kind of glue holding together the lowtype meanings of content morphemes.” (von Fintel 1995: 183). These considerations have an important impact on formal theories of grammaticalization (note that this has already been hinted at above in the discussion of the characteristics of grammaticalization processes): if the existence of a set of (core) functional categories is an innate property, readily given by UG, then grammaticalization must be conceived of as the (diachronic) lexicalization of a universal inventory of functional categories/heads.30 If combined with the assumptions on the syntax-morphology interface laid out in Section 2.3 above, this basic idea can be made more precise. Recall that the morphological component operates post-syntactically, that is, lexical insertion takes place post-syntactically, realizing syntactic terminal nodes, instead of feeding the syntactic derivation as traditionally assumed. Then, the term grammaticalization describes a change where a certain set of phonological features that serves to realize a lexical head in the input string generated by the target grammar is reanalyzed by the learner as the spell-out of a (adjacent) functional head. In other words, there is no change in syntactic category at all (in the sense that ‘category change’ means a change from, say, V to T), but – at least in the lexicon – only a change in the set of phonological features associated with a given functional head. In combination with the views adopted directly above, then, grammaticalization does not create new functional categories; rather, it provides new lexical realizations for a universal inventory of functional categories/heads (cf. also Roberts & Roussou 2003 on this aspect). Thus, grammaticalization clines describe a historical development where a certain set of phonological features serves to spell-out different syntactic heads in distinct (diachronically related) grammars, each time reflecting the different properties of the underlying syntactic terminals. However, grammaticalization does not fully reduce to the redistribution of phonological matrices. For various reasons, it may lead to – at time severe – syntactic changes. For example, syntactic change may result from a reanalysis of the syntactic structure involved: if a (auxiliary) verb moving to a functional head is reanalyzed as a piece of



JB[v.20020404] Prn:2/08/2005; 9:38



F: LA8102.tex / p.26 (1538-1602)

The Rise of Agreement

inflection associated with this functional head, then this reanalysis will lead to the loss of verb movement in all contexts where the reanalysis takes place (cf. Roberts 1993b on the synthetic future in Romance; Roberts & Roussou 2003 for a variety of further examples). In addition, on the assumption that the set of universally present functional categories includes only C, T, ν and D, while wider arrays may occur in individual languages, grammaticalization may enrich the functional inventory via adding ‘peripheral’ categories to the set of functional categories present in the grammar (cf. e.g. Wu 2004 on the grammaticalization of a functional category Aspect in the history of Chinese). Finally, grammaticalization may lead to the addition of lexical meanings to a given functional head, in the sense “that the meaning of a lexical category is composed with a functional meaning to yield a new, more complex functional meaning.” (von Fintel 1995: 184). For example, the reanalysis of an adjective ‘many’ as a D-head creates a new functional head (a quantifier) which combines the existential force of the formerly phonologically empty determiner with the plural meaning of ‘many’ (cf. von Fintel 1995: 185): (23) a.

b.

DP D $

Þ

NP AP

N

DP D

NP

$ ° many

many

Summing up, this section has shown that an analysis of grammaticalization as the transition from lexical elements into exponents of (formerly present, but phonologically empty) functional categories proves to be a promising avenue of research. It has been demonstrated that this basic assumption provides a unified explanation for a number of surface characteristics of grammaticalization phenomena, attributing the latter to universal properties of a closed class of functional categories. In addition, we have seen that the apparently gradual character of grammaticalization processes does not represent a special property of this type of language change that sets it apart from other types of changes and requires non-standard assumptions such as ‘fuzzy’ syntactic categories. Rather, the notions of grammar change and grammar competition are sufficient to reconcile the gradual nature of grammaticalization with standard generative assumptions on language change and the profile of syntactic categories. Finally, making use of a Late Insertion model of grammar, I have argued for an analysis of grammaticalization in terms of a redistribution of phonological features which nevertheless may affect the syntax of a given language via adding new (peripheral) functional categories, enriching the semantic content of existing categories or causing the loss of movement operations. In Chapters 4–6 it is shown how the diachronic rise of agreement morphology (and its syntactic effects) can be integrated into this overall picture of grammaticalization.

JB[v.20020404] Prn:2/08/2005; 9:38

F: LA8102.tex / p.27 (1602-1626)

Chapter 2. Theoretical preliminaries

. Summary In this chapter, I have introduced the main theoretical notions that I will assume throughout this book. In Section 2.2 I provided an introduction to the most recent version of minimalist syntax, the probe/goal system developed in Chomsky (2000, 2001a, 2001b), which makes available a new approach to agreement in which the uninterpretable φ-set of T/ν (the probe) accesses an agreement controller (the goal) under closest c-command (via the operation Agree). Agreement can therefore be established independent of (overt) NP-movement into the specifier of the relevant functional head. Section 2.3 gave an outline of Distributed Morphology and the ensuing model of the syntax-morphology interface where syntactic terminal nodes are provided with phonological exponents after the syntactic derivation. Section 2.4 illustrated the generative perspective on language change, according to which differences between historical stages of a single language are to be analyzed parallel to parametric variation between different individual languages. Hence, syntactic change is interpreted as an abrupt change in parameter settings (associated with a closed class of functional categories) that is triggered during language acquisition. In addition, it was shown that this model can be reconciled with the apparent gradualness of language change if certain assumptions on the proper object of linguistic study (‘grammar’ instead of ‘language’) and the course of parametric change (involving grammar competition) are adopted. Finally, in Section 2.5, I argued that grammaticalization has to be analyzed as a historical process where (formerly phonologically empty) functional categories are provided with a phonological realization via a reanalysis of exponents previously realizing lexical heads. It was demonstrated that this model provides a unified explanation for a number of surface characteristics of grammaticalization phenomena, attributing the latter to universal properties of functional categories. In addition, I gave an outline of how the historical processes in question can be modeled in a realizational model of grammar like Distributed Morphology. In the next chapter, I will develop a formal approach to the realization of verbal agreement (and complementizer inflection) which is based on the theoretical assumptions laid out above.

Notes . If the relevant categorial features are weak, the inflectional features of the verb and the argument are eliminated by covert movement of the relevant formal feature bundles (cf. Chomsky 1995). . One motivation for this revision are instances of long-distance agreement in Icelandic where the matrix verb agrees in number with a nominative DP that is contained in an embedded lower clause, while the matrix subject position is occupied by a ‘quirky’ dative subject; cf. the following example taken from Sigurðsson (1996: 30):



JB[v.20020404] Prn:2/08/2005; 9:38



F: LA8102.tex / p.28 (1626-1697)

The Rise of Agreement

(i)

Mér virtust [ þær vinna vel]. me.dat seem-pl they.nom work well ‘To me, they seem to work well.’

. Similar to uninterpretable φ-features, structural Case is analyzed as an undifferentiated (or unvalued) feature the manifestation of which depends on properties of the probe: finite T ‘assigns’ nominative, ν assigns accusative and the T head of control infinitives assigns null case. . Still, a DP with a deleted Case feature is visible to the computation and may block the establishment of an Agree or Move operation that targets a lower goal c-commanded by the DP in question, giving rise to so-called defective intervention effects. Defective intervention effects are only triggered by the head of an A-chain. In contrast, “A-movement traces are ‘invisible’ to the probe-associate relation” (Chomsky 2000: 131). . Chomsky (2000, 2001a) further claims that unaccusative and passive νP do not constitute a phase either. However, see Legate (2003) for a set of arguments that unaccusative and passive νPs are in fact phases as well. . Thus, DM is an ‘anti-lexicalist’ model, in the sense that no complete lexical elements (fully specified for semantic, morphosyntactic and phonological features) are inserted into the syntactic derivation (cf. Marantz 1997 for discussion). Instead, the information that is traditionally assumed to be part of a single lexical entry is divided into separate lexical entries that are part of distinct subparts of the lexicon: one part of the lexicon contains abstract morphemes, that is, only bundles of semantic and morphosyntactic features which are subject to ‘early’, that is presyntactic insertion (such as e.g. C[+wh] , T[+past] etc.). Another part contains the set of Vocabulary items which are inserted post-syntactically and provide phonological realizations of the abstract morphemes. Finally, there is the so-called Encyclopedia which contains the non-linguistic (idiomatic) information associated with lexical items (e.g. dog: ‘four legs, canine, pet, sometimes bites etc.’). . Some authors make use of the notion abstract morpheme to refer to syntactic terminals in order to avoid confusion with the traditional usage of the term ‘morpheme’ as part-of-speech. Note that throughout this book, the notion morpheme is used in the technical sense introduced in the main text, that is, as referring to syntactic heads/terminal nodes. . Most current morphological theories accept the notion that the phonological form of a given affix should be separated from its morphosyntactic function. Thus, they adopt (in some form or other) the Separation Hypothesis (Beard 1988, 1995; cf. Stump 1998; Borer 1998 for some discussion), according to which grammatical operations (which manipulate the constituency and feature content of lexical items) are discrete from the morphological or phonological operations that control the spell-out of words. Accordingly, it is widely held that affixes or other forms of morphological marking (such as Umlaut or the modification of tone or accent) are merely the exponents (Matthews 1991) of abstract morphological features such as [future], [plural], [1st person]. Note that this distinction follows automatically from the overall structure of DM. . Note that the Subset Principle alone is not sufficient to determine the winner of the competition if two Vocabulary items realize the same number of inflectional features. For such cases, two different solutions are proposed in the literature. The winner is either simply stipulated by an extrinsic rule ordering (Halle & Marantz 1993) or follows from a universal hierarchy of morphosyntactic features (cf. e.g. Noyer 1997; see Harley 1994; Harley & Ritter 2002 for the use of structured feature geometries instead of simple hierarchies).

JB[v.20020404] Prn:2/08/2005; 9:38

F: LA8102.tex / p.29 (1697-1764)

Chapter 2. Theoretical preliminaries . The systematic differences between lexical and functional morphemes can perhaps be employed to explain certain characteristics of grammaticalization processes which are usually assumed to involve a transition from lexical to functional categories, that is, from lexical to functional morphemes in the approach outlined here. See Section 2.5 below for the analysis of grammaticalization phenomena in a generative setting. . Note that this exposition is somewhat simplifying. Current work in DM (cf. e.g. Marantz 1997; Embick 1997; Harley & Noyer 1999) usually assumes that traditional category labels such as N, V, A have no significance in the grammar. Instead, the category of a given element is determined by the syntactic configuration in which it appears. More specifically, it is assumed that instead of traditional lexical categories, the lexicon contains category-neutral roots. The morphosyntactic category (if this is a valid notion at all) of a given root is then determined by the closest c-commanding functional morpheme. For example, the traditional category ‘noun’ corresponds to a root that is locally c-commanded and licensed by a D-head, the category ‘verb’ to a root locally c-commanded by ν (or T) etc. On this view, functional morphemes assume a category-defining role. Importantly, one and the same root may show up as a ‘verb’ or a ‘noun’ depending on the syntactic context where it is merged. This is then reflected by the shape of the Vocabulary items for roots, where ‘LP’ can be read as ‘Lexical Phrase’: (i)

a. [DP D [LP ___ ]] ↔ /dog/ b. [DP D [LP ___ ]] ↔ /kæt/ etc.

However, as long as it does not affect the argument made in the text I will continue to use the familiar notion of different lexical categories. . Another set of phenomena which is frequently analyzed in terms of Morphological Merger is the positioning of clitics, in particular second position clitics of the Wackernagel type. See for example Schütze (1994) on second position clitics in Serbo-Croatian. . According to Matthews (1991: 180) it is important to distinguish between genuine cumulative exponence and two other phenomena that may give rise to similar surface forms. The latter include fused marking (or so-called portmanteau forms), where regular phonological processes (sandhi) lead to a fusion of adjacent (simple) exponents similar to French au < à le. Another phenomenon that may give rise to the impression of apparent cumulative exponence is so-called overlapping exponence: for example, in Latin, voice and subject agreement are realized by a single formative in 2nd person plural forms: -tis in laudatis ‘you praise’ and -min¯ı in laudamin¯ı ‘you are praised’. However, in 3rd person plural forms, separate affixes for voice and agreement can be discerned: compare laudant ‘they praise’ with laudantur ‘they are praised’ (-nt ‘3pl’, -ur ‘passive’) (cf. Stump 1998: 31f.). This contrasts with cases of genuine cumulative exponence, where the exponents of two or more inflectional features coincide completely. . See Arregi (1999) for an analysis of the person and number inflection of Basque and Halle (1997) for an analysis of the Latin noun declension in terms of Fission. . Note that impoverishment rules may perhaps also be used to account for a number of apparently syntactic anti-agreement effects. For example, the absence of verbal agreement in the context of wh-subject extraction, which can be obsrved in a number of Northern Italian dialects (cf. e.g. Brandi & Cordin 1989), may result from an impoverishment rule that deletes agreement features in the presence of a wh-feature, leading to the insertion of the default 3sg ending. . Note that this does not amount to say that there are no changes affecting the syntax of a given language apart from parametric change. For example, there might be changes that have



JB[v.20020404] Prn:2/08/2005; 9:38



F: LA8102.tex / p.30 (1764-1828)

The Rise of Agreement

to do with the frequency of a given syntactic construction (such as, say, V1 structures). Importantly for our purposes, however, it is only parametric changes that are of theoretical interest and significance, in the sense that only they can reveal something about the structure and the workings of the language faculty/UG, see Lightfoot (1991, 1999) and Hale (1996) for discussion. . Note that an approach in terms of default settings is not always compatible with the Subset Principle. For example, Hyams (1986) claims that the default setting for the pro-drop parameter is [+pro-drop], which is clearly in conflict with the Subset Principle, since [+pro-drop] permits a set of grammatical sentences that includes the set generated by the setting [–pro-drop] (cf. O’Grady 1997 for discussion). . Concerning the question of when a given cue is ‘attested robustly’ factors such as ‘unambiguous expression’ and frequency play a role. In the case of V2, for instance, only those utterances that can only be analyzed as SpecCP [XP] are taken to express the cue in question. Furthermore, it has been observed that in the Germanic V2 languages, only 30% of main clauses exhibit an arbitrary XP in clause-initial position, while 70% are subject-initial. Lightfoot (1999) is led to conclude that these 30% of nonsubjects occupying SpecCP represent the threshold that is sufficient to count as a ‘robust’ instantiation of SpecCP [XP]. . See Bobaljik (2003) and Haeberli (2004) on the relative significance of syntactic and morphological cues. Haeberli argues that the impact of morphological or syntactic cues may differ from parameter to parameter (for similar considerations cf. Anderson 1980). Interestingly, he claims that mismatches between morphology and syntax (e.g., a grammar with syntactic evidence for verb movement but without rich verbal inflection) triggers grammar competition, that is, the existence of a second internalized grammar where the mismatch is resolved (cf. Kroch 1989). Over time, the latter more harmonic (or economic) variant may win out over the older grammar. . According to Hale (1996: 16) the difference between change and diffusion can be characterized as follows: “In the case of change, there has been imperfect transmission of some feature of the grammar. The acquirer’s input sources had features X, Y, and Z and the acquirer constructed a grammar which had features X, Y and W. The difference (W instead of Z) represents a change (Z>W). In the case of diffusion, the acquirer had input sources with features X, Y, Z and other input sources (or another input source) with features A, Y, W and constructed a grammar with features X, Y, and W. Note that there is no imperfect transmission of the relevant features: the child had feature W in his or her input sources and constructed a grammar with feature W. (Similarly for X and Y).” . Apart from the transition from OV to VO order, the rise of do-support and the loss of verb movement are two of the best studied examples of this kind of long-term change in the history of English (cf. Ellegård 1953; Lightfoot 1979, 1991, 1999; Kemenade 1987; Kroch 1989; Roberts 1993a; Pintzuk 1999, among many others). . Furthermore, it is generally assumed that grammaticalization is a unidirectional process (cf. Hopper & Traugott 1993; Lehmann 2002 for discussion). Thus, if an element has entered a grammaticalization cline it must continue along the individual stages of the cline but cannot go back, from e.g. affix to clitic or pronoun. However, it has been pointed out that this strong claim is confronted with counterexamples in which a former affix developed into a clitic. For example, in Estonian, (clitic) question particles and emphatic particles appear to have developed from former affixes (Campbell 1991). In a similar vein, the possessive clitic ‘s can be shown to have evolved out of a former genitive suffix in the history of English (Allen 1997).

JB[v.20020404] Prn:2/08/2005; 9:38

F: LA8102.tex / p.31 (1828-1892)

Chapter 2. Theoretical preliminaries . Note that if grammaticalization reflects universal properties of functional heads in the way sketched above, then the study of this type of diachronic change can reveal important insights into the universal profile of functional categories, which is a major research interest in current grammar models. See Roberts and Roussou (2003: Ch. 5.3) for discussion. . Note, moreover, that it is generally assumed that many functional heads (such as T in English or D in languages without overt determiners such as Tagalog) are phonologically empty which is the endpoint of phonological deficiency. . Note that on these assumptions, the category ‘agreement’ seemingly does not to qualify as a functional category in the same sense as D, T or C, since it is apparently not quantificational. See Chapter 3 for discussion of the phrase-structural status of agreement. . Note that von Fintel (1995) assumes that the transition from lexical content morpheme to functional morpheme usually involves an intermediate stage, that is, a content morpheme that carries logical meanings and is of high semantic type (examples include adjectives such as mere, former or propositional attitude verbs like believe or deny). . The account of grammaticalization proposed by Roberts and Roussou (2003) is based on the idea that a functional category F which requires PF realization (labeled F*) can satisfy this requirement either via Move (F*Move ) or Merge (F*Merge ). Grammaticalization processes then usually involve a (initial) change from F*Move > F*Merge . . Roberts and Roussou (2003) claim that this kind of change took place in previous stages of the Northern Italian dialect Veneto, drawing on data from Poletto (1995, 2000). Since this dialect already exhibits a paradigm of verbal agreement suffixes, the reanalysis in (25) apparently leads to multiple realization of a single set of agreement features, possibly an instance of (illicit) multiple exponence. To avoid this unwelcome result (and for various other reasons), Roberts and Roussou assume that AgrSP splits into separate person (PersP) and number (NumP) projections, basically following proposals by Poletto (2000) and Manzini and Savoia (2002). The pronoun is then reanalyzed as the realization of Pers0 while the verb occupies Num0 . Under these assumptions, the reanalysis is captured by the following structural change (p. 182): (i)

[PersP DPi [Pers V ] [NumP . . . [VP ti ... → [PersP [Pers D [NumP [Num V ] ...

. Apart from their incompatibility with prenominal genitive possessors in some languages such as German or English, cf. *the Peter’s book. . Of course, this raises the question of which categories are included in the set of functional categories determined by UG. I adopt the view that properties of UG determine what kinds of functional categories may exist in natural languages, that is, there is an innate ‘definition’ of what is a possible functional category. Howeves, in contrast to work by, for example, Cinque (1999), it is assumed that the set of (theoretically possible) functional categories compatible with this definition does not have an extension in every human language. Instead, only the core functional categories C, T, ν, D are taken to be universally present, whereas the acquisition of a wider array of functional categories is dependent on relevant evidence in the input (for related views cf. Thráinsson 1996; Bobaljik & Thráinsson 1998; Haeberli 2002; Bobaljik 2003).



JB[v.20020404] Prn:2/08/2005; 9:49

F: LA8103.tex / p.1 (45-125)

Chapter 3

The structural design of agreement

. Introduction An investigation of the historical emergence of agreement morphology necessarily builds on a theory of agreement in general. Therefore, the purpose of this chapter is to develop a formal account of predicate-argument agreement. Which are the key issues a theory of agreement must address? Even a very elementary inspection reveals that agreement differs fundamentally from other inflectional categories like Tense and Aspect. The latter require merely a single (morpho-) syntactic operation which serves to license the relevant inflectional features by establishing a structural relation between a verbal head and a higher functional head. This structural relation is usually modeled in terms of head movement. The resulting head complex is then interpreted by the morphological component giving rise to an inflected verb. In contrast, the realization of agreement is a more complex process: in addition to the combination of verb stem and inflectional affix, the feature content of the agreement marker has to be determined via feature matching with the φ-set of a nominal argument. We therefore have to acknowledge the existence of two separate structural relations: a relation between two heads (e.g., verb movement leading to affixation of an agreement marker to the verb) and feature matching between an agreement head (the agreement target) and an argument’s φ-set (the agreement controller).1 Accordingly, a formal model of syntactic agreement must provide an answer to the following fundamental questions: (i) How does the verb combine with the agreement affix (concerning both syntax and morphology)? (ii) How is feature matching between a certain argument and the agreement affix established? A third issue which is intimately related to these questions concerns the phrase-structural representation of agreement, that is, the question of whether agreement features project a separate functional agreement phrase (AgrP) in the syntax. The central proposals of this chapter can be summarized as follows. First, it is argued that agreement morphemes do not head their own projection in the structure of the clause. Rather, they attach parasitically to other functional heads the content of which is interpretable at the interfaces. Second, the feature content of agreement morphemes is not identified under spec-head agreement. Instead, the relevant structural relation is assumed to be closest c-command (i.e., the Agree relation, Chomsky 2000). Third, it is claimed that agreement morphemes are provided with phonological content post-syntactically (Late Insertion) and that the particular properties of

JB[v.20020404] Prn:2/08/2005; 9:49



F: LA8103.tex / p.2 (125-183)

The Rise of Agreement

the placement of agreement markers follow from the way complex head adjunction structures are linearized in a autonomous morphological component, Morphological Structure. The chapter is organized as follows. Section 3.2 focuses on the phrase-structural representation of agreement. More specifically, it addresses the question of whether agreement heads project a separate phrasal category in the structure of the clause. Section 3.2.1 reviews a number of conceptual arguments that bear on this issue. In Section 3.2.2, I discuss relevant evidence from word order facts. Section 3.2.3 deals with the morphological realization of agreement, giving particular attention to the position of agreement markers relative to the verb stem and other inflectional markers. Section 3.3 discusses the structural relation under which predicate-argument agreement is established, focusing on the question of whether an approach that makes use of the Agree mechanism recently proposed by Chomsky (2000) is to be preferred over the standard analysis in terms of spec-head agreement. It is argued that decisive evidence in favor of an Agree-type approach comes from long-distance agreement in the Nakh-Dagestanian language Tsez. Section 3.4 deals with previous approaches to the linearization of inflectional markers, arguing that the order of inflectional affixes cannot be determined by the syntax alone, contra recent proposals by Julien (2002). Section 3.5 briefly summarizes the empirical and conceptual findings reached so far. In Section 3.6, I develop the approach to predicate-argument agreement advocated in this book. Section 3.6.1 argues that agreement heads/morphemes are base-generated as adjuncts to other functional heads. Section 3.6.2 presents an analysis where feature matching between a controller and an agreement target is established under closest ccommand (i.e., Agree). Under the assumption that Agr-morphemes may in principle attach to a variety of functional heads, this analysis predicts that Case and agreement can be dissociated. It is shown that this prediction is borne out by agreement facts in Georgian and participle agreement in French. Section 3.6.3 develops a formal account of the spell-out of inflectional markers, focusing on the creation of linear order. Section 3.6.4 summarizes the basic traits of the analysis developed so far. Section 3.7 argues that we must acknowledge the existence of a further mechanism that may give rise to agreement phenomena, drawing on evidence from complementizer agreement in Germanic. Section 3.7.1 provides a critical review of previous analyses of the phenomenon in question, arguing that a purely syntactic account is not tenable. In Section 3.7.2, I develop a new analysis of complementizer agreement that attributes the existence of inflection in the C-domain to the post-syntactic insertion of Agr-morphemes. Section 3.8 provides a summary of the issues discussed in this chapter.

. The phrase-structural representation of agreement As already noted in Chapter 2 above, in the pre-minimalist stage of the Principles and Parameters framework the standard approach to subject-verb agreement posited a spec-head configuration between the subject and the functional Infl node which

JB[v.20020404] Prn:2/08/2005; 9:49

F: LA8103.tex / p.3 (183-218)

Chapter 3. The structural design of agreement

contains the inflectional features normally associated with the verb (i.e., tense, mood, and agreement features) and mediates the subject-predicate relation (Chomsky 1981, 1986b). In later work, it was reasoned that the features/affixes which were previously assumed to be part of Infl should head their own projections in the syntax (cf. e.g. Pollock 1989; Kayne 1989; Belletti 1990; Chomsky 1991), leading to a more articulate conception of clause structure with separate projections for T(ense) and subject/object agreement (AgrSP and AgrOP respectively): (1)

CP C’ C

AgrSP AgrS’ AgrS

TP T’ T

AgrOP AgrO’ AgrO

VP

In this approach, feature matching between an argument’s φ-set and the φ-set of an agreement morpheme is attributed to a spec-head relation between these two elements, that is, the argument DP moves to SpecAgrSP/SpecAgrOP to license the feature content of AgrS/AgrO. In addition, the spec-head relation in question is assumed to result in Case assignment, in the sense that agreement projections provide the structural configuration in which the Case features carried by V (accusative) and T (nominative) are assigned to nominal arguments under spec-head agreement (which necessitates movement of T and V to the relevant Agr-heads). In line with the morphosyntactic analyses of the time, it is assumed that the verb stem moves to AgrS/AgrO to pick up the relevant inflectional morphemes (Ouhalla 1991; Chomsky 1991, among others). Following Baker (1985, 1988), the successive head movement operations that combine the verb with its inflections are mirrored by the order of inflectional affixes, with lower functional morphemes being realized closer to the verb stem than higher ones (the so-called Mirror Principle, see below). Furthermore, it is sometimes assumed that the hierarchical order of functional projections may be subject to parametrization (Laka 1990; Ouhalla 1991), since in some languages like German or Turkish tense morphology is closer to the verb stem than agreement, while it is the other way around in languages such as Berber.



JB[v.20020404] Prn:2/08/2005; 9:49



F: LA8103.tex / p.4 (218-254)

The Rise of Agreement

However, this approach to agreement phenomena has not gone unchallenged. The following section reviews a set of arguments that have been put forward against the existence of separate syntactic agreement projections. These arguments can be divided into three types: (i) evidence from word order facts; (ii) evidence from the morphological realization of agreement; (iii) conceptual arguments.

.. Conceptual arguments Already early on it was pointed out that the assumption of separate agreement projections raises conceptual problems due to the fact that the nature of Agr differs fundamentally from that of other inflectional categories. For example, Speas (1991: 411) notes that typical functional categories such as Tense, Aspect or Negation “are not dependent on some other constituent for their interpretation”, whereas the content of Agr is determined in relation to some other constituent with which Agr shares φ-features. In other words, Agr lacks an independent meaning and is more accurately characterized as a relational instead of a (separate) functional category (Iatridou 1990; Speas 1991; Mitchell 1994; Chung 1998). This line of reasoning is taken up in more formal terms in Chomsky (1995: 349ff.), who argues that agreement projections should be eliminated from the clause structure since they lack features that receive an interpretation at the interface levels PF and LF. Chomsky points out that agreement projections are present only for theory-internal reasons, namely to provide the structural configuration in which the feature content of T, V (Case, non-interpretable φ-features) is checked against the feature content of nominal arguments. As a substitute for an AgrP-based theory of Case and agreement checking, he proposes that the relevant non-interpretable features are added to the ‘core’ functional categories T and ν, which contain sets of features interpretable at the interfaces. Accordingly, Case and the predicate’s non-interpretable φ-features are checked in an additional specifier of these categories.2 Thus, the removal of AgrPs leads to a theory where heads may project multiple specifiers (see Ura 2000, 2001 for discussion). The following phrase marker illustrates the resulting clause structure, including the A-/head movement operations that result in Case and agreement checking.

JB[v.20020404] Prn:2/08/2005; 9:49

F: LA8103.tex / p.5 (254-303)

Chapter 3. The structural design of agreement

(2)

CP C’ C

TP subj.

T’

[T[íV[í]]T] íP í’

obj.

í’

tsubj. tn

VP tV

tobj.

Apart from these conceptual considerations, there is also empirical evidence indicating that the idea of separate agreement projections may be flawed. The relevant data are discussed in the following sections.

.. Evidence from word order facts One of the original arguments for splitting the IP into separate tense and agreement projections comes from word order facts in French infinitives. Pollock (1989) argues that the diverging placement properties of non-finite auxiliaries and main verbs relative to sentence negation and certain classes of adverbs should be taken to indicate that there are two different landing sites for verb movement in the inflectional domain. His argument goes as follows. In infinitives, the auxiliaries avoir ‘have’ and être ‘be’ can appear on either side of the negation pas. (3) a.

Ne pas être heureux est une condition pour écrire des romans. neg neg be happy is a prerequisite for write novels ‘Not to be happy is a prerequisite for writing novels.’ b. N’être pas heureux est une condition pour écrire des romans. (Pollock 1989: 373)

Pollock therefore concludes that non-finite auxiliaries can optionally undergo movement to Infl/T (which is assumed to be higher than NegP) in French. In contrast, non-finite main verbs can appear to the left or to the right of VP-adverbs but must follow the negation pas:



JB[v.20020404] Prn:2/08/2005; 9:49



F: LA8103.tex / p.6 (303-361)

The Rise of Agreement

(4) a.

Complètement perdre la tête pour les belles étudiantes, c’est completely lose the head for the pretty students that is dangereux! dangerous ‘To completely lose one’s head over pretty students is dangerous.’ b. Perdre complètement la tête pour les belles étudiantes, c’est dangereux! (Pollock 1989: 377f.)

(5) a.

Ne pas sembler heureux est une condition pour écrire neg neg seem happy is a prerequisite for write des romans. novels ‘Not to seem happy is a prerequisite for writing novels.’ b. *Ne sembler pas heureux est une condition pour écrire des romans. (Pollock 1989: 374)

On the assumption that VP-adverbs mark the left edge of the VP, these examples are taken to indicate that non-finite main verbs can undergo optional movement to an inflectional head that is located between NegP and VP. Pollock identifies this head as Agr, heading a separate AgrP. Moreover, he claims that both non-finite main verbs and auxiliaries can undergo ‘short’ verb movement to this position, cf. (6) Être souvent triste... be often sad ‘To often be sad...’ (Iatridou 1990: 562) Hence, according to Pollock the structure of the clause is as follows: (7) [CP C’ [ C [TP [T’ T [AgrP [Agr’ Agr [VP [V’ V ]]]]]]]] However, shortly after Pollock’s paper was published, Iatridou (1990) presented a set of French data that challenges the assumption that there is a separate AgrP between NegP and VP. First, she discusses the pair of examples in (8). (8) a.

Pierre a à peine vu Marie. Pierre has hardly seen Marie b. Pierre a vu à peine Marie. (Iatridou 1990: 563)

(8a) follows straightforwardly from Pollock’s analysis: the finite auxiliary moves to T, while the non-finite main verb stays behind in situ (to the right of the VP-adverb). However, Iatridou points out that (8b) creates a problem for Pollock’s account. Note that both the finite auxiliary and the non-finite main verb precede the VP-adverb. According to Pollock’s proposals, this configuration implies that both verbs have moved over the adverb, which occupies a fixed position. However, if verb movement has to obey the Head Movement Constraint (Travis 1984), then the finite auxiliary must move through Agr before it reaches T. But then, there is no head position left where the participle can move to. Pollock tries to handle the problematic example by assuming

JB[v.20020404] Prn:2/08/2005; 9:49

F: LA8103.tex / p.7 (361-435)

Chapter 3. The structural design of agreement

that the structure of sentences like (8b) is actually as shown in (9), where the participle moves to the head of a separate AgrP that is located between VPaux and VPmain . To ensure that participle movement lands in a position to the left of à peine, he assumes that the adverb is adjoined to VPmain . (9) ... [AgrP [Agr’ Agr [VPaux [Vaux’ Vaux [AgrP [Agr’ Agr[VPmain à peine [VPmain [Vmain’ Vmain ]]]]]]]]] Iatridou notes that this proposal partially weakens Pollock’s argument for the existence of AgrP, since it is now possible to base-generate auxiliaries in the correct surface order in examples such as (6) (which Pollock analyzes as resulting from ‘short’ movement of the auxiliary). Even more problematic is the following data set, which includes not one, but two VP-adverbs: (10) a.

Souvent mal faire ses devoirs, c’est stupide. frequently badly make one’s homework that is stupid ‘To frequently do one’s homework badly is stupid.’ b. Faire souvent mal ses devoirs... c. Souvent faire mal ses devoirs... (Iatridou 1990: 567)

The crucial example here is (10c). Whereas (10a) and (10b) can be analyzed as the result of non-movement and ‘short’ verb movement, respectively, it is not possible to account for (10c) under Pollock’s assumptions: if both adverbs are adjoined to the VP, there is simply no head position available where the non-finite main verb can move to. Iatridou concludes that Pollock’s arguments in favor of a separate AgrP do not go through and that the French data in question is better handled by a more flexible placement of adverbs. In addition, she suggests an alternative analysis of agreement which is based on the idea that predicate-argument agreement does not involve a separate projecting Agr-node, but merely reflects a certain structural (spec-head) relation between two elements.

.. The morphological realization of agreement Some of the original conceptual motivations for splitting the Infl node into a number of functional heads concerned the interface between syntax and morphology. First, the fact that the Infl node contained different sets of morphosyntactic features (such as Tense, Agr, etc.) was deemed conceptually unattractive. Second, in line with the Mirror Principle proposed by Baker (1985), it was generally assumed that syntax and morphology are basically isomorphic, that is, ideally, each separate inflectional affix should correspond to a separate functional head.3 (11) The Mirror Principle Morphological derivations must directly reflect syntactic derivations (and vice versa).



JB[v.20020404] Prn:2/08/2005; 9:49



F: LA8103.tex / p.8 (435-499)

The Rise of Agreement

Thus, if a given functional head is apparently associated with more than a single set of inflectional features, this is attributed to head movement operations which yield head adjunction structures (in line with the Head Movement Constraint, Travis 1984). One important consequence of the Mirror Principle is that one should be able to ascertain the order of embedding of functional categories simply by inspecting the order of inflectional markers on the verb stem in a given language. While this line of research has proven to be very fruitful with respect to functional categories such as Tense, Aspect and Negation (cf. e.g. Laka 1990; Ouhalla 1990, 1991; Baker 1996; Zanuttini 1997, and most recently Julien 2002), the evidence coming from the placement of agreement markers is less straightforward and often rather problematic for the hypothesis that there is a separate functional agreement projection (cf. Laka 1990; Speas 1991; Mitchell 1994; Spencer 1997; Julien 2002). This section discusses a set of morphosyntactic properties of agreement which indicate that there is a fundamental difference between agreement and other inflectional categories such as tense, aspect, or negation, for which it is generally assumed that they head their own projection in the syntax. The first difference concerns the distribution of inflectional markers. Functional categories like tense, aspect, negation, or complementizers usually have a unique expression in any given clause, that is, they generally occur (i) only once and (ii) in a fixed position.4 The following facts demonstrate that agreement may behave completely differently. First, we find (redundant) multiple realizations of one and the same agreement marker on various verbal elements. Second, agreement markers may appear in different positions dependent on the (morpho-) syntactic context. In addition, agreement may be signaled by a set of discontinuous markers scattered over the clause. Finally, agreement may be realized by a combination of these options. Multiple agreement on different (verbal) heads is found in languages such as Swahili where subject agreement (=agreement in noun class) is repeated on every element of the verbal complex: Swahili (12) wa-toto wa-li-kuwa wa-me-ki-soma ki-tabu cl2-children cl2-past-be cl2-perf-cl7-read cl7-book ‘The children had read the book.’ (Krifka 1995: 1416) Other languages with multiple subject agreement include Sudanic languages like Mbay (Central Sudanic, AgrS on verb and tense marker/aux) or Lango (Nilotic, AgrS on verb and negation): Mbay (13) m-¯ m-nd¿ 6g¯o ¯ a ¯ 1sg-fut 1sg-buy ‘I will buy.’ (Keegan 1997: 70) Lango (14) án à-pé à-wótò kàmpálà I 1sg-neg 1sg-go.perf Kampala ‘I didn’t go to Kampala.’ (Noonan 1992: 142)

JB[v.20020404] Prn:2/08/2005; 9:49

F: LA8103.tex / p.9 (499-586)

Chapter 3. The structural design of agreement

Another possibility is ‘traveling agreement’, where the agreement markers attach to different verbal elements dependent on the syntactic context. For example, in the MonKhmer language Sengoi, subject agreement is expressed on the tense marker/auxiliary in the future tense, but on the verb in the past tense:5 Sengoi (15) a. Ke ki-ha muit. 3sg 3sg-fut enter ‘She will enter.’ b. Guru ajeh ya ki-lei. teacher that past 3sg-come ‘The teacher came.’ (Means et al. 1986) A combination of multiple agreement and agreement scattering is found in languages such as Serbo-Croat, where the auxiliary inflects for person and number, while the main verb shows agreement for gender and number of the subject (note that the pronoun may be dropped): (16) on-i su došl-i 3-masc.pl be.3.pl came-masc.pl ‘They came.’ (Corbett 2000: 244) An even more complex system is exhibited by the Finnic language Livonian, where each element of the verbal complex agrees with a specific subset of the subject’s φ-features (negation: person; aspectual auxiliaries: person and number; participle: number): Table 1. Multiple agreement in Livonian (Mitchell 1994: 114). 1sg 2sg 3sg 1pl 2pl 3pl

neg-agrs(person)

‘be’-agrs(person & number)

‘give’-part-agrs(number)

ä-b ä-d ä-b ä-b ä-d ä-b

ùo-Ø ùo-Ø ùo-Ø ùo-m ùo-tt ùo-tt

and-ên-Ø and-ên-Ø and-ên-Ø and-ên-D and-ên-D and-ên-D

Under the assumption that the presence of agreement morphology should be taken as evidence for a separate AgrP (in line with the Mirror Principle), these facts require the existence of multiple agreement phrases to account for subject agreement alone. As noted above, this contrasts significantly with the behavior of functional heads such as T or Asp, which typically occupy a single fixed position; the latter neither spread to other positions nor are they realized in different positions dependent on the syntactic context. Alternatively, one might resort to the assumption that some of the different realizations of agreement are to be attributed to agreement morphemes/features that are generated outside of AgrP. Yet, the latter solution undermines the whole idea that agreement morphology corresponds to functional agreement nodes that head their own projection (cf. Mitchell 1994: 114).



JB[v.20020404] Prn:2/08/2005; 9:49



F: LA8103.tex / p.10 (586-786)

The Rise of Agreement

Further support for the claim that agreement does not head a separate projection in the syntax comes from the fact that cross-linguistically, the ordering properties of agreement markers differ significantly from those of other inflectional markers such as tense or aspect. The following discussion is based on empirical findings of Julien (2002). Based on an extensive typological study covering 530 languages, Julien discusses the relation between word order and the realization and position of tense and aspect markers. She shows that these inflectional categories show only a very limited amount of cross-linguistic variation with respect to their relative ordering. More specifically, she claims that the following generalizations hold across languages (p. 235): (17) Attested sequences of inflectional markers a. (S) T-Asp-V (O) b. (S) T-V-Asp (O) c. (S) (O) V-Asp-T (O) d. S Asp-V O T (18) Non-attested sequences a. *(S) Asp-T-V (O) b. *(S) Asp-V-T (O) c. *(S) V-T-Asp (O) Thus, it appears that aspect markers are always closer to the verb stem than tense markers (if both follow or precede the verb stem). Julien argues that this observation directly follows from the Mirror Principle if it is assumed that there is a universal hierarchy of functional projections (cf. e.g. Cinque 1999) in which T is obligatorily generated above Asp. Interestingly, Julien shows that with respect to their placement properties, agreement markers are much less well-behaved than tense and aspect markers. Table 2 presents Julien’s findings concerning the relative position of verb stem, tense (T) and subject agreement marker (AgrS) in her language sample.6 Even a cursory look at Table 2 reveals that there is a vast amount of variation concerning the position and realization of agreement markers across languages. For example, we can see that all six theoretically possible linearizations of V, T and AgrS are attested.7 The overall picture gets even more complex if object agreement marking is taken into consideration as well, cf. Table 3.8 A closer look at Table 3 reveals that of the 24 theoretically possible orderings (abstracting away from the complications created by portmanteau forms), only four are not attested in Julien’s sample: *AgrO-T-AgrS-V, *T-AgrO-V-AgrS, *AgrO-T-V-AgrS, and *AgrS-V-AgrO-T.9 Thus, it appears that there is a significant difference between functional categories such as T and Asp on the one hand and Agr on the other. While the relative order of tense and aspect markers is rather fixed (reflecting the fact that T and Asp universally occur in a fixed hierarchical relation), apparently almost anything goes with respect to agreement markers. If this is again taken to reflect the hierarchical position of Agr, then it follows either (i) that the hierarchical position of agreement

JB[v.20020404] Prn:2/08/2005; 9:49

F: LA8103.tex / p.11 (786-807)

Chapter 3. The structural design of agreement

Table 2. Relative order of verb stem, tense marker, and AgrS marker, by word order (see Table 5.2 of Julien 2002: 249). Morpheme order 1. 2. 3. 4. 5. 6. 7. 8. 9. 10. 11. 12. 13. 14. 15. 16.

AgrS/T V AgrS/T+V V+AgrS/T V+T+AgrS V+AgrS+T V+AgrS T AgrS+T+V AgrS T V AgrS+T V AgrS+V+T AgrS+V T T+V+AgrS T V+AgrS T+AgrS+V T+AgrS V T AgrS+V

V-initial 1 3 4 3 1 4

4 4 5 1

Word order V-medial V-final 4 4 17

7 6 5 29 1 2 1 5 2 3

2 3 17 42 10 2 5 1 28 5 2 3 1

Uncertain 1 2 8 3

7 1 1

Total 2 7 26 64 16 3 14 6 5 57 5 9 1 9 3 3

projections is subject to parametrization across languages (in contrast to T or Asp), basically following proposals by Ouhalla (1991), or (ii) that the phrase-structural properties of Agr differ fundamentally from those of other functional categories such as T or Asp. Taken together with the other arguments put forward in this section, it appears that the second option in fact represents the more accurate choice. I am therefore led to conclude that Agr-nodes do not head their own projection in the syntax, in contrast to ‘substantial’ functional categories such as C, T or Asp, which are projecting syntactic heads, characterized by a cross-linguistically uniform order of embedding (this conclusion is also reached by Julien 2002: 6). Rather, Agr-morphemes may attach parasitically to other ‘substantial’ functional categories. This explains the multitude of serialization patterns if it is further assumed that languages may differ with respect to the functional head the Agr-morpheme attaches to. A radical implementation of this insight is that agreement heads are actually not part of the syntax at all, as claimed by Anderson (1992), Marantz (1992), and Halle and Marantz (1993), for example. Instead, these authors propose that agreement morphemes are added to the structure post-syntactically at Morphological Structure. According to this view, only ‘substantial’ functional categories such as T, Neg or Asp are represented in the syntax, while agreement is purely morphological in nature, merely reflecting structural relations established in the syntactic derivation (see Section 3.6 below for some discussion and a more moderate proposal according to which at least canonical instances of agreement are represented in the syntax, but parasitic on other functional heads).10



JB[v.20020404] Prn:2/08/2005; 9:49



F: LA8103.tex / p.12 (807-807)

The Rise of Agreement

Table 3. Relative order of verb root, tense marker, AgrS marker, and AgrO marker, by word order (see Table 5.3, Julien 2002: 253). Morpheme order V-initial 1. 2. 3. 4. 5. 6. 7. 8. 9. 10. 11. 12. 13. 14. 15. 16. 17. 18. 19. 20. 21. 22. 23. 24. 25. 26. 27.

AgrS/O-V-T T-V-AgrS/O V-T-AgrS/O V-AgrS/O-T T-AgrS-AgrO-V AgrS-T-AgrO-V AgrS-AgrO-T-V AgrS-AgrO-V-T T-AgrO-AgrS-V AgrO-AgrS-T-V AgrO-AgrS-V-T T-V-AgrS-AgrO V-T-AgrS-AgrO V-AgrS-T-AgrO V-AgrS-AgrO-T T-V-AgrO-AgrS V-T-AgrO-AgrS V-AgrO-T-AgrS V-AgrO-AgrS-T V-AgrO-T/AgrS T-AgrS-V-AgrO AgrS/T-V-AgrO AgrS-T-V-AgrO AgrS-V-T-AgrO AgrO-V-T-AgrS AgrO-V-AgrS-T AgrO-V-T/AgrS

1 1 1 1 1 1

1

Word order V-medial V-final 8 3 4 2 1 1 3 1 1 1 4 1

11 3 1 1 2 1 7 1 2 6

Uncertain

Total

2

19 3 7 2 3 4 1 11 2 2 8 2 9 2 1 3 9 6 4 1 4 2 3 6 7 4 2

1

1

1 1

6 1 1

1

2 2

1 2 2 3

2 2 2 3

1

6 5 2 1 1 2

1 1 1 1

2 7 4 1

This view on agreement markers is further supported by the fact that agreement markers are preferably bound elements (cf. Chapter 1). (19) lists the sequences that are theoretically possible, but not attested in Julien’s sample (focusing on AgrS only): (19) a. b. c. d. e. f. g. h. i.

*V AgrS/T *V T AgrS *V T+AgrS *V+T AgrS *V AgrS T *V AgrS+T *AgrS T+V *AgrS V T *AgrS V+T

JB[v.20020404] Prn:2/08/2005; 9:49

F: LA8103.tex / p.13 (807-869)

Chapter 3. The structural design of agreement

j. *T V AgrS k. *T+V AgrS l. *T AgrS V Note that the vast majority of the non-attested patterns (9 out of 12) involve a free AgrS marker. In fact, there are only six genera in Julien’s sample that exhibit free AgrS markers. All the relevant languages are of the type SVO (mostly Creoles), with free subject agreement and tense markers that both precede the verb (i.e., AgrS T V): (20) Generalization: The distribution of free AgrS markers A free AgrS marker is only possible in an SVO language in which the sequence of inflectional markers is AgrS T V. On the assumption that Agr heads its own projection, this finding is rather surprising. Rather, we would expect Agr to occur as frequently as a non-bound element as other ‘substantial’ functional categories like tense auxiliaries, aspect markers or complementizers. In contrast, if agreement does not head its own projection but attaches parasitically to other functional heads, it is predicted that Agr is preferably realized as an affix that attaches either to its functional host or to a lexical head moved to the functional head complex in question (a similar point is made with respect to T+Agr portmanteau forms in Speas 1991).11

. The structural relation involved in feature matching The second ingredient of the agreement relation concerns the structural configuration in which feature matching between an argument and a predicate can be established. The following proposals are found in the literature: (i) the spec-head configuration (Koopman 1992, 1996, 2003; Sportiche 1998; Kayne 1989, 1994; Chomsky 1991, 1993, 1995; Watanabe 1996; Belletti 2001; Julien 2002); (ii) government (Polinsky & Potsdam 2001); (iii) local c-command (i.e., the Agree relation, Chomsky 2000; Carstens 2003; Bejar 2003); (iv) a combination of some of these proposals (cf. e.g. Chomsky 1986b; Chung 1998; Bobaljik & Wurmbrand 2004). This section focuses on the approaches that are prevalent in current syntactic theorizing, that is, the spec-head relation and an analysis of agreement in terms of the Agree relation proposed in Chomsky (2000, 2001a, 2001b). It is shown that facts from so-called long-distance agreement (henceforth LDA) favor an Agree-type analysis over the traditional approach in terms of spec-head agreement. It is commonly assumed that the agreement relation is strictly local, in the sense that the controller and the target must be clause-mates. In an analysis based on the spec-head configuration, this condition follows automatically from the standard assumptions (i) that agreement involves A-specifiers only (similar to case assignment) and (ii) that A-movement is typically clause-bounded, that is, A-movement may not cross a CP boundary.12 Note that this kind of locality holds for both subject and object



JB[v.20020404] Prn:2/08/2005; 9:49



F: LA8103.tex / p.14 (869-904)

The Rise of Agreement

agreement. In both cases, feature matching between an argument and the predicate is licensed by moving the argument to the closest A-specifier (following Chomsky 1995, SpecTP for subject agreement and SpecνP for object agreement), which creates a checking configuration between the interpretable φ-set of the argument and the corresponding non-interpretable features of the predicate. In contrast, the locality conditions imposed by the operation Agree are somewhat different, due to the fact that the agreement domain extends to the left edge of a lower (strong) phase. Hence, material at the left edge of an embedded CP is accessible for the establishment of (object) agreement as well.13 Thus, the two approaches to agreement give rise to a set of slightly different agreement domains, which are illustrated in (21): (21) Agreement domains a. Spec-head Subject agreement

[CP C [TP spec T [íP spec í [VP V ... [CP spec C [TP... ]]]]]]

Object agreement

b. Agree Subject agreement

[CP C [TP spec T [íP spec í [VP V ... [CP spec C [TP... ]]]]]]

Object agreement

In standard cases of subject or object agreement, the empirical coverage of both approaches is identical (at least roughly, cf. Koopman 2003 for some discussion). To decide which analysis is to be preferred over the other, we must therefore look for cases where the subtle difference between (21a) and (21b) with respect to locality has an effect on the realization of agreement. This section shows that the locality conditions imposed by the spec-head relation are violated in instances of long-distance agreement (LDA) where a matrix verb can apparently access a controller that is part of an embedded clause.14 Therefore, the existence of LDA constitutes a problem for an analysis of agreement that makes use of the spec-head relation (cf. e.g. Polinsky & Potsdam 2001; Bruening 2001; Bobaljik & Wurmbrand 2004). However, before we can decide whether the extension of the agreement domain made available by the workings of the operation Agree is sufficient to capture the problematic data, we must take a closer look at the phenomenon in question (the following discussion is based on work by Polinsky & Potsdam 2001; Polinsky 2003; Bobaljik & Wurmbrand 2004). Genuine LDA involves a matrix verb (the target) that selects for a clausal complement and agrees with an argument (the controller) contained in the embedded

JB[v.20020404] Prn:2/08/2005; 9:49

F: LA8103.tex / p.15 (904-965)

Chapter 3. The structural design of agreement

clause.15 The following definition of the phenomenon in question is given in Polinsky (2003: 282): (22) Long-distance agreement Agreement between a controller and a target that are in distinct clauses. Languages which reportedly exhibit this form of long-distance agreement are Tsez (Polinsky & Potsdam 2001) and the Algonquian languages Pasamaquoddy (Bruening 2001) and Innu-aimûn (Branigan & MacKenzie 2002). The following discussion will focus on the relevant facts in the Nakh-Dagestanian language Tsez, which is spoken in the Caucasus region and exhibits basic SOV order (cf. Polinsky & Comrie 1999; Polinsky & Potsdam 2001; Comrie 2003; Polinsky 2003). Verbal agreement in Tsez is organized in an ergative-absolutive fashion, that is, the verb agrees in gender and number with its absolutive argument (i.e., either the direct object of a transitive or the subject of an intransitive verb):16 In the following examples, Roman numerals indicate gender distinctions. (23) a.

kid y-ik’i-x girl.abs.II II-go-pres ‘The girl goes.’ b. už-¯a bikori b-exu-r-si. boy-erg snake.abs.III III-die-caus-pstwit ‘The boy killed the snake.’ (Comrie 2003: 323)

However, in cases where the matrix verb selects a clausal argument, it may agree either with the whole embedded clause or with the absolutive argument of the embedded clause. This state of affairs is illustrated by the following pair of examples: (24) a.

eni-r [už-¯a magalu b-¯ac’-ru-łi] mother-dat boy-erg bread.abs.III III-eat-pstprt-nmlz-abs r-iy-xo. IV-know-pres ‘The mother knows that the boy ate the bread.’ b. eni-r [už-¯a magalu b-¯ac’-ru-łi] mother-dat boy-erg bread.abs.III III-eat-pstprt-nmlz-abs b-iy-xo. III-know-pres ‘The mother knows that the boy ate the bread.’ (Polinsky & Potsdam 2001: 584)

In (24a), the matrix verb riyxo ‘knows’ agrees in gender with its complement clause, as can be seen from the fact that the verb is inflected for default gender class IV agreement, which is realized by the prefix r-. However, in (24b), the matrix verb carries the agreement prefix b-, which serves to signal agreement with a gender class III absolutive argument. In the example at hand, the only available controller of this particular agreement marker is the absolutive-marked direct object of the embedded clause, magalu



JB[v.20020404] Prn:2/08/2005; 9:49



F: LA8103.tex / p.16 (965-1044)

The Rise of Agreement

‘bread’, which also controls gender class III agreement on the embedded verb. In other words, (24b) exemplifies a case of LDA where the controller is apparently separated from its target by a clause-boundary. Polinsky and Potsdam show that LDA is only possible when the controller is the primary topic of the embedded clause (in other words, the presence of LDA signals that the controller serves as the topic of the embedded clause). Thus, LDA is not possible if the embedded absolutive argument carries the clitic focus marker -kin, cf. (25), or if there is another topic in the embedded clause. Example (26a) shows that the temporal adverb Auł ‘yesterday’ does not block LDA when it occurs in situ. However, LDA with the embedded absolutive DP is ruled out when the temporal adverb is topicalized within the embedded clause, as in (26b):17 (25) *eni-r [t’ek-kin y-igu y¯ał-ru-łi] y-iy-xo. mother-dat book.II.abs-foc II-good be-pstprt-nmlz II-know-pres ‘The mother knows that the BOOK is good.’ (Polinsky & Potsdam 2001: 611) eni-r [už-¯a Auł magalu b-¯ac’rul-łi] mother-dat boy-erg yesterday bread.III.abs III-eat-pstprt-nmlz r/b-iy-xo. IV/III-know-pres ‘The mother knows that the boy ate bread yesterday.’ b. eni-r [Auł už-¯a magalu b-¯ac’rul-łi] mother-dat yesterday boy-erg bread.III.abs III-eat-pstprt-nmlz r/*b-iy-xo. IV/III-know-pres ‘The mother knows that the boy ate bread yesterday.’

(26) a.

Polinsky and Potsdam deduce from these facts that the embedded absolutive must undergo overt or covert movement into the left periphery of the embedded clause to be accessible for LDA. Crucially, it can be shown that neither restructuring nor overt raising into the matrix clause is involved. The following pair of examples demonstrates that the whole embedded clause including the controller ka>at ‘letter’, which triggers long-distance agreement on the matrix verb, can undergo fronting to clause-initial position: (27) a.

[kidb¯a ka>at t’¯at’rułi] y-iyx užir. girl letter.II.abs read II-knows boy ‘The boy knows that the girl has read the letter.’ b. užir y-iyx [kidb¯a ka>at t’¯at’rułi]. boy II-knows girl letter.II.abs read ‘The boy knows that the girl has read the letter.’ (Polinsky 2003: 302)

Therefore, it is quite clear that the controller has not undergone overt raising into the matrix clause. Furthermore, in contrast to restructuring contexts, each clause constitutes a separate domain for the scope of negation:

JB[v.20020404] Prn:2/08/2005; 9:49

F: LA8103.tex / p.17 (1044-1105)

Chapter 3. The structural design of agreement

(28) a.

[kidb¯a ka>at t’et’r-änˇc’i-ru-łi] y-iyx užir. girl letter.II.abs read-neg-pstprt-nmlz II-knows boy ‘The boy knows that the girl has not read the letter.’ b. [kidb¯a ka>at t’¯at’rułi] y-iy-x-¯anu užir. girl letter.II.abs read II-knows-pres-neg boy ‘The boy does not know that the girl has read the letter.’ (Polinsky 2003: 303)

These facts indicate that the construction in which long-distance agreement is observed is a genuine biclausal structure. Moreover, the controller remains in the embedded clause, at least in the overt syntax (see Polinsky & Potsdam 2001 for more arguments involving scrambling, binding, and scope interactions). This seems to suggest that the facts of LDA in Tsez cannot be accounted for if agreement is modeled in terms of an overt spec-head configuration. Yet, we must rule out the possibility that the controller undergoes covert raising into the matrix clause, thereby meeting the ‘normal’ locality condition on verbal agreement at LF. Polinsky and Potsdam (2001) provide several arguments against this analysis and other conceivable alternatives that situate the controller in the matrix clause. One such argument concerns the interpretation of embedded quantified expressions, the scopal properties of which are taken to indicate that the lower absolutive DP remains in the embedded clause throughout the entire syntactic derivation (Polinsky & Potsdam 2001: 618f.). In Tsez, quantifiers that are clause-mates may give rise to scope ambiguities. In the following mono-clausal examples, either the subject or the object may take wide scope over the other (X > Y reads as ‘X has scope over Y’):18 šibaw >’w ay-¯a sis k’et’u han-si. every dog-erg one cat-abs bite-pstwit ‘Every dog bit a cat.’ (∀dog > ∃cat, ∃cat > ∀dog) b. sida >’w ay-¯a šibaw k’et’u han-si. one-obl dog-erg every cat-abs bite-pstwit ‘A dog bit every cat.’ (∃dog > ∀cat, ∀cat > ∃dog)

(29) a.

No such ambiguities arise when one of the two quantified DPs is contained in an embedded clause. In other words, “quantifiers may not take scope out of their own clause” (Polinsky & Potsdam 2001: 618): (30) a.

sis uˇciteler [šibaw uži ø-ik’ixosi-łi] r-iyxo. one teacher every boy-I.abs I-go-nmlz IV-know ‘Some teacher knows that every boy is going.’ b. ok: ‘Some teacher is such that he knows that every boy is going.’ (∃teacher > ∀boy) c. *‘Every boy is such that some teacher knows that he is going.’ (*∀boy > ∃teacher)



JB[v.20020404] Prn:2/08/2005; 9:49



F: LA8103.tex / p.18 (1105-1157)

The Rise of Agreement

Now, if it is assumed that the embedded absolutive DP ends up in the matrix clause at some point in the derivation (e.g., by covert movement) in order to be able to control agreement on the matrix verb, then we should expect that this DP may take wide scope over another quantified element of the matrix clause. Crucially, the expectation is not borne out by the facts: (31) a.

sis uˇciteler [šibaw uži ø-ik’ixosi-łi] ø-iyxo. one teacher every boy-I.abs I-go-nmlz I-know ‘Some teacher knows that every boy is going.’ b. ok: ‘Some teacher is such that he knows that every boy is going.’ (∃teacher > ∀boy) c. *‘Every boy is such that some teacher knows that he is going.’ (*∀boy > ∃teacher)

In (31a), the matrix verb shows gender class agreement with the embedded absolutive DP šibaw uži ‘every boy’ (class I). Nevertheless, the sentence has the same interpretation as (30a), that is, šibaw uži ‘every boy’ cannot take scope over the quantified element sis uˇciteler ‘one teacher’ in the matrix clause. Therefore, we can conclude that šibaw uži ‘every boy’ must be located in the embedded clause at the level of representation where quantifier scope is computed. More important for our purposes, long-distance agreement cannot be licensed by raising of the embedded absolutive DP in the covert syntax. These findings can be summarized as follows: (32) Long-distance agreement in Tsez a. The structure is biclausal and the target is contained in the higher clause. b. The controller, which carries absolutive case, remains in the embedded clause at all levels of representation. Polinsky and Potsdam (2001) conclude that the form of LDA found in Tsez cannot be accounted for by an analysis in terms of spec-head agreement. Instead, they argue for an analysis where the agreement relation closely resembles head government, allowing the matrix verb to enter into an agreement relationship with a controller that undergoes covert A’-movement into the left periphery of the embedded clause. Still, Koopman (2003) claims that these findings can be reconciled with the idea that agreement universally involves the spec-head configuration if an analysis in terms of clausal pied-piping is adopted. Koopman assumes that an element that occupies the left periphery of an embedded clause can pied-pipe the whole clause to establish feature checking with a higher head under spec-head agreement (for related proposals cf. Koopman & Szabolcsi 2000; Müller 2004; cf. Simpson & Bhattacharya 2003 for a similar analysis of wh-in situ in Bangla). On this assumption, LDA between a matrix verb and an embedded DP is analyzed as resulting from moving the whole embedded clause containing the controller into the specifier of an AgrP of the matrix clause:

JB[v.20020404] Prn:2/08/2005; 9:49

F: LA8103.tex / p.19 (1157-1210)

Chapter 3. The structural design of agreement

(33)

...

AgrP Agr’

CP DP

Agr

C’ C

IP

... V tCP

...

Koopman claims that in this configuration, feature checking between Agr and the φset of the embedded absolutive DP is possible if it is assumed that spec-head agreement between the DP and C leads to the presence of φ-features in C0 /CP via feature percolation. In turn, the resulting φ-feature set on C0 /CP participates in feature checking with Agr under spec-head agreement. Note that this approach involves a number of non-standard assumptions. For example, analyses that make use of clausal pied-piping are generally concerned with A’-dependencies,19 for which it is commonly assumed that movement serves to satisfy properties of both the moved element (scope) and the attracting head (clause type/whfeature). Thus, percolation of, for example, wh-features can be said to be motivated by the need to satisfy scopal properties of an in-situ wh-phrase. This contrasts with the percolation of A-related features involved in Koopman’s analysis of LDA in Tsez, where no property of the controller is involved: the φ-set of the absolutive DP is interpretable and therefore need not be checked. Moreover, the absolutive DP has already participated in Case and agreement checking in the embedded clause, so there is no need for Case checking with the matrix verb either. Thus, there is no motivation for feature percolation to occur. In addition, it appears that a spec-head relation that licenses the presence of φ-features on a given head (which Koopman calls ‘percolation’) requires that the relevant specifier is an A-position (cf. e.g. Shlonsky 1994 on complementizer agreement in Germanic). Again, this is not the case in the LDA examples at hand. To conclude, it is very doubtful that the grammar makes available an operation like Apied-piping. Instead, it seems that the licensing of A-related features such as Case and agreement may not lead to pied-piping of more material than the minimal phrase that contains the attracted features. Apart from these conceptual problems, a spec-head analysis of LDA can also be shown to suffer from empirical shortcomings. Koopman explicitly assumes that agreement always results from overt movement processes, that is, from a spec-head configuration created in the overt syntax (which may be obscured by subsequent movement operations). On this assumption, however, a spec-head analysis of LDA cases such as (24b), repeated here as (34), faces a problem.



JB[v.20020404] Prn:2/08/2005; 9:49



F: LA8103.tex / p.20 (1210-1265)

The Rise of Agreement

(34) eni-r [už-¯a magalu b-¯ac’-ru-łi] mother-dat boy-erg bread.abs.III III-eat-pstprt-nmlz-abs b-iy-xo. III-know-pres ‘The mother knows that the boy ate the bread.’ In (34), the leftmost element of the embedded clause is not the absolutive DP magalu, but rather the ergative DP už-¯a. Therefore, Koopman’s analysis in terms of feature percolation and spec-head agreement leads us to expect that the matrix verb can only agree with the embedded ergative DP, since it occupies a position to the left of the absolutive DP (which should lead to the presence of the ergative DP’s φ-features in C if the relevant position is SpecCP). Still, it is the absolutive DP which triggers agreement on the matrix verb and not the ergative one. Since Koopman assumes that agreement relations must be established in the overt syntax, this fact cannot be attributed to agreement checking at LF involving covert movement of the absolutive DP into the left periphery of the embedded clause. Moreover, it is not possible that the order ergativeabsolutive found in the embedded clause is created by further movement operations after agreement checking has taken place: under standard assumptions about cyclicity, no movement operations can take place in the embedded clause after this clause itself has undergone movement (in the case at hand, to SpecAgrP of the matrix clause). Therefore, it seems that an analysis in terms of spec-head agreement is not capable of explaining the form of LDA exhibited by Tsez. However, before we can turn to a conclusion, it remains to be shown that an account in terms of Agree can handle the data set in question more adequately. Recall that Polinsky and Potsdam (2001) propose an analysis where the agreement relation closely resembles head government, allowing the matrix verb to enter into an agreement relationship with a controller located in the left periphery of the embedded clause. Of course, this is already reminiscent of the Agree mechanism and the concept of locality devised in Chomsky (2000, 2001a), where only material located at the left edge of a phase (a stage in the syntactic derivation represented by e.g. a finite CP) is accessible to syntactic operations triggered by functional heads higher up in the phrase marker (the Phase Impenetrability Condition). Thus, in principle an analysis in terms of Agree seems to be well-equipped to handle LDA in Tsez. Furthermore, as expected in the framework of Chomsky (2000), movement to the left edge of a phase may feed operations triggered by higher functional heads (in the case at hand, agreement), and LDA is subject to intervention effects created by other elements topicalized in the embedded clause. Still, there are some open questions, the most obvious having to do with the fact that the operation that moves the controller of LDA into the left periphery of the embedded clause apparently can take place in the covert syntax (see above). This implies that agreement is established by covert operations as well. This has an important consequence for the overall model of grammar. On the assumption that the content of agreement morphemes must be valued in the syntax before they can be subject to Vocabulary Insertion (Late Insertion), the evidence from Tsez suggests that the morphological component must operate after

JB[v.20020404] Prn:2/08/2005; 9:49

F: LA8103.tex / p.21 (1265-1347)

Chapter 3. The structural design of agreement

all syntactic movement operations have taken place, including covert movement (cf. Bobaljik 2002; Bobaljik & Wurmbrand 2004). Thus, we arrive at a model of grammar in which the point of Spell-out represents the interface to both Morphological Structure and LF, with apparent ‘covert’ movement resulting from assigning a phonological matrix to the base position of a chain (sometimes called ‘Spell-out at LF’, cf. Bobaljik 1995; Groat & O’Neill 1996). Summing up, this section has shown that the phenomenon of ‘genuine’ LDA which can be observed in languages like Tsez constitutes a serious problem for the ‘traditional’ approach to agreement in terms of a spec-head configuration. In addition, we have seen that the slight extension of the agreement domain following from the workings of the Agree mechanism enables a matrix verb to agree with an element in the left periphery of its complement clause. This provides us with a rather elegant explanation of the problematic data, lending further support to the claim that the structural relation which agreement phenomena are based on is rather closest c-command (i.e., Agree) than the spec-head configuration.20

. Syntactic approaches to word formation This section deals with the relation between syntactic structure and morpheme order (i.e., the order of phonological exponents). As already noted above, the most influential approach to this topic (in the Principles and Parameters framework) is based on the Mirror Principle (Baker 1985, 1988), repeated here for convenience: (35) The Mirror Principle Morphological derivations must directly reflect syntactic derivations (and vice versa). On the assumption that the verb combines with its inflectional affixes (which are generated in functional heads) via successive head movement operations that must respect the Head Movement Constraint, the Mirror Principle ensures that the order of inflectional markers on the verb stem reflects the hierarchical properties of a head adjunction structure derived in the syntax. For example, if the verb moves first to Asp and then to T, the morphological exponent of Asp must be closer to the verb stem than the exponent of T (if both affixes are on the same side of the stem). However, the assumption that the formation of complex words (here: the inflected verb) takes place in the syntactic component (via head movement) faces the problem that verbs can also carry inflection in languages that apparently lack the operation of verb movement. Here, the principal example is English, for which it is standardly assumed that the finite verb does not leave the VP. If adverbs like often are taken to mark the left edge of the VP, then the following contrast indicates that the verb does not move into the inflectional domain in the overt syntax (cf. Emonds 1978; Pollock 1989): (36) a. Floyd often eats bananas. b. *Floyd eats often bananas.



JB[v.20020404] Prn:2/08/2005; 9:49



F: LA8103.tex / p.22 (1347-1386)

The Rise of Agreement

Still, the verb clearly carries inflection, either the 3sg.pres.indicative /-z/ or the past tense marker /-d/. To obviate this problem, Chomsky (1993) proposes a lexicalist alternative which does not any longer make use of the assumption that the verb undergoes head movement to pick up inflectional markers in higher functional heads. Instead, inflectional markers are added to the verb prior to lexical insertion. Subsequently, the verb enters the syntactic derivation fully inflected and undergoes syntactic movement to functional heads to check whether it carries the appropriate inflectional features. The problem posed by examples such as (36) is now accounted for by the assumption that the relevant checking operations may be postponed to the covert syntactic component. Thus, it is assumed that in English, the verb undergoes invisible LF-movement to check off tense and agreement features in higher functional heads. The overt/covert distinction is implemented by the assumption that the V-feature that triggers verb movement may be either strong or weak (recall that only strong features require overt movement prior to Spell-out). This serves to account for the empirical facts, but gives up the idea that there is a correlation between syntactic operations and the order of inflectional affixes on the verb: if checking of inflectional features may take place after Spell-out, it is rather clear that the relevant movement operations cannot have any effect on the shape of the inflected verb. Thus, a checking model as envisaged by Chomsky (1993) makes no predictions at all with respect to verbal morphology, since word formation takes place prior to syntax and inflectional features may be checked by either overt or covert movement. There is no causal relation between the absence of head movement and the absence of inflectional morphology: a verb spelled-out in situ may still carry inflection.21 Recently, Julien (2002) has proposed a radical alternative to previous accounts of word formation, claiming that the order of inflectional markers and the suffix/prefix distinction are completely determined by the syntactic component (which might be called a Radical Mirror Principle; similar ideas are already present in Kayne 1994). Following Kayne (1994), she assumes that hierarchical relations created in the syntactic component exhaustively determine word order. More specifically, PF (or MS) maps asymmetric c-command relations into linear order (the Linear Correspondence Axiom, LCA). In addition, Julien claims that this approach carries over to word-internal ordering properties of stems and inflectional markers. Moreover, properties such as being a prefix or a suffix are taken to merely reflect the hierarchical position of a given terminal node in the syntactic component. For example, it is claimed that head movement is always left-adjunction, leading uniformly to a realization of the target head as a suffix (since by assumption, the target is asymmetrically c-commanded by the adjoined head): (37)

T V

uniformly maps into

V+T

T

Apart from head movement, Julien recognizes two other configurations that can give rise to affixation of inflectional material. Crucially, in these configurations, the el-

JB[v.20020404] Prn:2/08/2005; 9:49

F: LA8103.tex / p.23 (1386-1450)

Chapter 3. The structural design of agreement

ements forming a complex word do not constitute a single X0 -unit. Instead, Julien assumes that separate syntactic heads may be interpreted by MS/PF as one grammatical word if they are immediately adjacent to each other (adopting a Late Insertion model). Accordingly, a configuration where the head of XP and the head of a phrase YP moved to SpecXP are adjacent can lead to suffixation of the head of XP to the head of YP (cf. Kayne 1994 for this proposal concerning word formation in strict SOV languages). This is illustrated in (38), where SpecTP is filled by a VP (remnant). On the assumption that the complements of V have undergone further licensing movement operations, V and T are adjacent. In this configuration, then, V+T can be realized as a single word via Late Insertion. (38)

TP T’

VP ...

T

V

tVP

Julien claims that structures generated by so-called ‘roll-up’ movement, where VP moves into SpecνP, νP into SpecAspP, AspP into SpecTP etc., are the “the principal and perhaps the only means of deriving complex words” (p. 101) in SOV languages. Note that by assumption, both head movement and the configuration in (38) give rise to suffixing. What about prefixes? Here, Julien assumes that two heads which do not form a X0 -complex can still be realized as one grammatical word if they are linearly adjacent due to evacuation movement of intervening material. Thus, in a configuration such as (39), T may be realized as a prefix on V (moved to ν) if nothing intervenes between these heads. As a consequence, prefixation is always the outcome of non-movement.22 (39)

TP T’ íP

T

í’

... V+í

VP

Julien assumes that the structural configurations in (37)–(39) may give rise to either free or bound inflectional markers. In other words, the syntax has nothing to say about whether markers are free or bound.23 One of the most interesting aspects of Julien’s proposal is that it links the order of inflectional markers to general syntactic properties (word order) of a given language. For example, it has often been noted that SOV languages show predominantly suffixal inflection (cf. Sapir 1921; Greenberg 1963; Bybee 1985; Bybee et al. 1990; Siewierska



JB[v.20020404] Prn:2/08/2005; 9:49



F: LA8103.tex / p.24 (1450-1475)

The Rise of Agreement

& Bakker 1996, and many others). In Julien’s framework, this correlation is accounted for by assuming that strict SOV languages such as Japanese always have the structure in (38) and lack the option of head movement (cf. Kayne 1994). In other words, it appears that Julien’s account makes clear predictions with respect to which affix sequences A,B,C should be possible if a language has a certain set of syntactic properties X,Y,Z (a language with verb movement must show suffixes, prefixation of functional material implies the absence of verb movement etc.). Combined with the assumption that functional categories such as T or Asp occur in a universal order of embedding (e.g., T is always generated above Asp), this leads to a rigid, syntactically constrained theory of how X0 -complexes are linearized, which is at first sight conceptually attractive. However, similar to the related approach to word order (cf. Kayne 1994 and subsequent work), such an account requires quite an amount of heavy syntactic machinery (e.g., massive roll-up and remnant movement). In particular, the shift of explanatory burden from morphology to syntax often leads to highly complex syntactic structures and sometimes contra-intuitive derivations simply to account for a certain sequence of inflectional markers. For example, as already mentioned in Note 22, English must be analyzed as having overt verb movement into the inflectional domain in order to account for its suffixal verbal inflection, despite massive counterevidence (cf. Emonds 1978; Pollock 1989; Chomsky 1991, 1993; Halle & Marantz 1993; Roberts 1993a; Rohrbacher 1994; Bobaljik 1995, among others). Another general problem arises in connection with the assumption that the majority of SOV orders is generated by successive roll-up movement of complements into the specifier of their selecting heads. Since the specifiers of the inflectional domain are required to host constituents subject to roll-up movement (i.e., νP moves to SpecAspP, AspP to SpecTP etc.), there are no positions left for argument licensing (Julien follows Kayne 1994 in assuming that every head projects only a single specifier). Accordingly, Case and agreement checking must take place VP-internally, contra standard assumptions.24 Thus it appears that a purely syntactic approach to morpheme ordering gives rise to a number of serious conceptual problems. Moreover, it can be shown that the specific analysis in question is also empirically inadequate, predicting the existence of some sequences that are not existent and excluding others that are attested across languages. Concerning the former point, recall that Julien (p. 235) notes that the following sequences of inflectional markers are not attested across languages: (40) Non-attested sequences a. *(S) Asp-T-V (O) b. *(S) Asp-V-T (O) c. *(S) V-T-Asp (O) However, note that under Julien’s assumptions, even if T is universally generated above Asp, at least the order Asp-T-V can be derived rather easily by first moving the Asphead to T, giving rise to suffixation (i.e., Asp+T). Subsequently, this complex may attach to the verb (in situ) if nothing intervenes, leading to the sequence Asp-T-V:

JB[v.20020404] Prn:2/08/2005; 9:49

F: LA8103.tex / p.25 (1475-1535)

Chapter 3. The structural design of agreement

(41)

TP T’ T Asp

AspP T

tAsp

VP ...

V’ V

...

Thus, Julien’s theory is not as restrictive as it appears at first sight, due to the diversity of (i) possible syntactic derivations and (ii) configurations feeding word formation her theory allows for. Below, it is shown that a much more conservative theory of morpheme ordering is equally successful in restricting the amount of possible orders. Another empirical problem has to do with the realization of agreement. As noted above, Julien assumes that agreement features are added to other functional heads (Julien specifically mentions Fin, T and heads lower than T such as Asp or Voice). However, she is not very specific about the exact representation of agreement features. That is, it remains unclear whether agreement features are merely added to the feature set of a given functional head or if they are base-generated as separate heads adjoined to other functional heads (as proposed in this book, see Section 3.6.1 below). Importantly, both options can be shown to be rather unwieldy within the theoretical approach developed by Julien. Assume that Agr-features/morphemes are added to a functional head which is associated with morphological content as well, such as T. If the Agr-features are merely added to the feature content of T, the question arises of what determines the linear order of the phonological exponents of Agr and T. It is fairly clear that the linear relation in question cannot be determined by the syntax, since the set of features contained in T (e.g., [+past] and [3sg]) does not correspond to a syntactic structure involving c-command relations. Accordingly, the position of Agr (i.e., its realization as prefix/suffix) must be determined by the morphological component alone, presumably via lexical properties of the phonological exponent of Agr. This, however, would undermine the whole approach devised by Julien (cf. p. 255, for example: “It is a main thesis of the present work, however, that the position of an element in the surface order is never directly encoded in its lexical entry.”). However, if the order of T and Agr added to T must follow from a structural configuration, then the only option left in Julien’s framework is left-adjunction of Agr to T, that is, the structure in (42), where Agr must be realized to the left of T (or any other potential host). (42)

T Agr

T



JB[v.20020404] Prn:2/08/2005; 9:49



F: LA8103.tex / p.26 (1535-1592)

The Rise of Agreement

Yet, this can be shown to give rise to serious complications for Julien’s approach. More specifically, all instances where AgrS occurs to the right of T cannot be analyzed any longer as resulting from the presence of Agr-features on T. This order is found in 79 genera in Julien’s language sample, including the most frequent sequence V+T+AgrS, which is found in the Germanic, Celtic and Romance families, for example. In the case of Germanic and Romance, then, this requires that (i) AgrS is added to a functional head above T (under Julien’s assumptions presumably Fin) and (ii) that the finite verb undergoes overt movement to this head (roll-up movement is confined to ‘genuine’ OV languages). However, at least in embedded clauses of the Germanic VOlanguages (apart from Yiddish and Icelandic), it is rather obvious that the verb does not move that far up in the clause, but rather stays behind to the right of negation, cf. the following examples from Danish (which is representative for the other Mainland Scandinavian languages including Faroese):25 (43) a. *Jeg spurgte hvorfor Peter havde ikke læst den. I asked why Peter had not read it b. Jeg spurgte hvorfor Peter ikke havde læst den. I asked why Peter not had read it ‘I asked why Peter didn’t read it.’ (Vikner 1995: 145) Thus, even with the proviso that AgrS features may be located in a number of different functional heads, Julien’s system is not capable of deriving the ordering patterns found in Germanic in a satisfactory way. This state of affairs results from the fact that Julien’s idea that affix order is completely determined by the syntax is too strict. The failure to account for the linearization of Agr-morphemes casts serious doubts on the overall enterprise of providing a purely syntactic analysis of the linear order of inflectional markers. Note, however, that at least some of the problems of Julien’s system might be remedied if Agr-features added to, say, T might be spelled out to the left or to the right of T. Then, however, the claim that affix order is exhaustively determined by the syntax must be dropped. In Section 3.6.3 below, I will outline an alternative theory of the linear order of inflectional markers which is based on similar assumptions as Julien’s approach, but is more flexible, adopting the more conservative assumption that linear order is only partially determined by the syntactic component, which leaves open certain points of variation which are resolved in the post-syntactic morphological component.

. Interim summary: The design of agreement So far, the discussion in this chapter has provided us with the following insights into the structural design of agreement. First, it has been shown that the properties of agreement differ significantly from those of functional categories such as T, Neg or Asp. This has been taken to suggest that agreement features/morphemes do not head their own projection in the syntax. Rather, they are parasitic on other functional heads (cf. Ia-

JB[v.20020404] Prn:2/08/2005; 9:49

F: LA8103.tex / p.27 (1592-1660)

Chapter 3. The structural design of agreement

tridou 1990; Speas 1991; Marantz 1992; Halle & Marantz 1993; van Gelderen 1993; Mitchell 1994; Chomsky 1995, 2000; Halle 1997; Julien 2002). Second, Section 3.3 has provided further arguments (apart from those originally put forward in Chomsky 2000; cf. Chapter 2 above) that the structural configuration in which the feature content of Agr-morphemes is licensed/valued is closest c-command (i.e., the Agree relation) rather than spec-head agreement. Third, we have seen that it is presumably not possible to establish a one-to-one relation between the output of the syntactic component and the ordering of inflectional markers on the verb stem. Rather, it appears that the syntax constrains the set of possible affix sequences, but that lexical information (such as the prefix/suffix distinction) matters as well when the output of the syntactic component is interpreted at Morphological Structure.

. Toward a realizational theory of agreement In this section, I will give an outline of a theory of agreement which is based on the insights reached above. The approach presented here is framed in a realizational model of grammar, where morphology operates after syntax, interpreting the output of the syntactic component (Distributed Morphology; cf. Chapter 2 above). The section is organized as follows. First, Section 3.6.1 addresses the question of how agreement should be represented in the grammar if we abandon the idea that agreement heads project phrasal categories. Section 3.6.2 deals with the structural relation under which feature matching between a controller and an agreement morpheme is achieved. Finally, Section 3.6.3 discusses aspects of word formation in the model of agreement envisaged in this book, focusing on how syntactic structures are mapped into linear sequences of phonological exponents.

.. The structural representation of agreement morphemes If agreement features/morphemes do not head their own projection in the syntax, the question arises of how they are to be represented in the structure instead. As noted above, two major lines of reasoning can be discerned in the literature: agreement features/morphemes are added to other functional heads either before or after the syntactic component. According to the latter line of thinking, agreement is a purely morphological phenomenon and agreement heads are completely absent from the syntactic component. Instead, they are added post-syntactically at Morphological Structure as adjuncts to ‘substantial’ functional categories such as T, Neg or Asp which are represented in the syntax, while agreement merely reflects structural relations established in the syntactic derivation (Marantz 1992; Halle & Marantz 1993; Halle 1997).26 This mechanism is sometimes referred to as the insertion of dissociated morphemes (Embick 1997). In Distributed Morphology, this morphological operation is often used to account for case and agreement phenomena. For example, Marantz



JB[v.20020404] Prn:2/08/2005; 9:49



F: LA8103.tex / p.28 (1660-1688)

The Rise of Agreement

(1992), Halle and Marantz (1993), and Halle (1997) analyze subject-verb agreement in terms of the post-syntactic adjunction of an Agr-morpheme to T. However, I hesitate to adopt this mechanism for canonical instances of agreement for the following reasons. First, an analysis where agreement is generally attributed to the insertion of dissociated morphemes requires that MS has powerful syntax-like mechanisms at its disposal, which are necessary for detecting the correct agreement controller, to value Agr-morphemes via copy operations etc. That is, this analysis seems to establish a syntax after the real syntax, which is conceptually unattractive. Second, as shown above in Section 3.3, long-distance agreement phenomena are sensitive to intricate syntactic restrictions such as intervention effects created by topicalized constituents. It is rather doubtful that these restrictions can be handled by morphological mechanisms alone in a satisfactory way. Still, it seems that at least for some instances of multiple agreement, where agreement is realized in more than one place in a given clause, an approach in terms of a post-syntactic insertion of agreement morphemes might be on the right track (see Section 3.7.2 below for an analysis of complementizer agreement in Germanic which makes use of this mechanism). I therefore adopt a somewhat less radical alternative according to which (syntactic) agreement features/morphemes are present in the syntax, but parasitic on other functional heads (Iatridou 1990; Speas 1991; van Gelderen 1993; Mitchell 1994; Chomsky 1995, 2000; Julien 2002). More precisely, I assume that agreement morphemes are part of the numeration, but do not head their own projection in the syntax. Instead, agreement morphemes are merged with other ‘substantial’ functional heads before the latter are combined with a phrasal complement.27 ‘Canonical’ subject-verb agreement is then the result of an agreement morpheme attached to T (henceforth Agr-on-T). That is, the following X0 -complex consisting of an Agr-morpheme adjoined to T is merged with νP in the syntax: (44)

T T

Agr

As far as the syntactic derivation is concerned, the assumption that discrete, albeit non-projecting Agr-morphemes are adjoined to other functional heads does not differ substantially from the idea that agreement features are merely added to the feature content of other functional heads (Chomsky 1995, 2000). At first sight, the latter approach seems to be preferable on conceptual grounds, since it completely removes Agr-heads (and adjunction structures such as (44)) from the syntactic derivation. Still, I choose to represent agreement via separate syntactic heads/morphemes, since this facilitates a more straightforward morphological processing of agreement morphemes, in particular with respect to the creation of linear order (see Section 3.6.3).28 By assumption, any of the core functional categories C, T, ν or D (which may represent a wider array of functional categories) is in principle capable of hosting agreement morphemes (cf. Julien 2002 for some discussion and examples where subject agreement is analyzed as resulting from an Agr-morpheme attached to Fin, Asp or Voice). The actual presence of an agreement morpheme is then dependent on whether

JB[v.20020404] Prn:2/08/2005; 9:49

F: LA8103.tex / p.29 (1688-1724)

Chapter 3. The structural design of agreement

the learner can detect the presence of an Agr-morpheme on a certain functional head on the basis of the evidence available to him/her.

.. Feature matching under closest c-command In Section 3.3 above, it has already been pointed out that the structural relation underlying predicate-argument agreement should be modeled in terms of closest ccommand between an agreement morpheme and an appropriate goal, that is, the operation Agree proposed by Chomsky (2000). In this section, the workings of Agree in establishing feature matching in canonical instances of subject and object agreement are briefly illustrated. Recall that by assumption, subject-verb agreement involves the presence of an agreement morpheme adjoined to T, while object agreement results from an agreement morpheme added to ν. If syntactic trees are built up in a bottomup fashion, object agreement is checked after ν has been merged with its complement VP which contains the object (note that the following phrase markers abstract away from head movement). (45)

íP í’ í’

DPsubj. í Agr

VP í

V

DPobj.

Agree

In the structure (45), the head complex [ν Agr [ν]] may enter into an Agree relation with the feature set of the object DP (under closest c-command), which leads to feature matching between the interpretable φ-set of the object and the content of the Agr-morpheme, giving rise to object agreement (and Case licensing). Eventually, the object DP moves to a specifier of ν if the latter carries an EPP feature which triggers that operation. If Merge preempts Move, it is expected that the subject is introduced before object movement takes place (Chomsky 2000).29 Subsequently, the resulting structure is merged with T, which may host an agreement morpheme as well. Again, the head complex [T Agr [T]] initiates an Agree operation which targets the closest active DP with an appropriate feature content, leading to subject agreement and nominative licensing:



JB[v.20020404] Prn:2/08/2005; 9:49



F: LA8103.tex / p.30 (1724-1762)

The Rise of Agreement TP

(46)

T’

spec

íP

T Agr Agree

T

í’

DPobj.

í’

DPsubj. í Agr

VP í

V

tobj.

Note that the intervening object DP does not block the Agree relation between the head complex in T and the subject DP due to the fact that elements of the same minimal domain are equidistant to a higher probe (cf. e.g. Chomsky 2000: 122, 130). This analysis reflects the intuition that agreement should not be implemented via a projecting functional head (see Section 3.2 above). Rather, agreement is more accurately characterized as a relational category (Iatridou 1990; Speas 1991; Mitchell 1994; Chung 1998) the content of which is licensed by entering into an Agree relation with the interpretable φ-set of a DP argument. The theory of agreement outlined so far has certain consequences for the relation between Case and agreement. Recall that Chomsky (2000, 2001a) follows the standard assumption that Case and agreement licensing is correlated. In other words, object agreement is associated with the Agree relation initiated by ν, which also serves to license accusative Case, while subject agreement is associated with T in combination with nominative assignment/licensing. However, on the assumption that Agrmorphemes can in principle attach to a variety of functional heads apart from T or ν (see above), it should be possible for Case and agreement to be dissociated from each other (cf. Pesetsky & Torrego 2004 for related conclusions). In other words, it is expected that there are languages where Case is assigned by T and ν (or, rather by T and the head complex ν+V if it is assumed that the capability to assign accusative is associated with the lexical category V), while the agreement relation is mediated by other functional heads that host Agr-morphemes. In such languages, Case assignment should be independent of the realization of agreement. A first indication that something along these lines is on the right track comes from long-distance agreement in Tsez, repeated here for convenience:

JB[v.20020404] Prn:2/08/2005; 9:49

F: LA8103.tex / p.31 (1762-1837)

Chapter 3. The structural design of agreement

Tsez (47) a. eni-r [už-¯a magalu b-¯ac’-ru-łi] mother-dat boy-erg bread.abs.III III-eat-pstprt-nmlz-abs r-iy-xo. IV-know-pres ‘The mother knows that the boy ate the bread.’ b. eni-r [už-¯a magalu b-¯ac’-ru-łi] mother-dat boy-erg bread.abs.III III-eat-pstprt-nmlz-abs b-iy-xo. III-know-pres ‘The mother knows that the boy ate the bread.’ (Polinsky & Potsdam 2001: 584) As noted above, the matrix verb ‘know’ may choose to agree either with the whole complement clause, as in (47a), or with an absolutive DP that receives an interpretation as the topic of the embedded clause (and is therefore associated with the left edge of the embedded CP phase), cf. (47b). Note that in the latter case, the agreement controller magalu ‘bread’ has already undergone Case (absolutive) and agreement checking in the embedded clause. Thus, it appears that a DP with a checked Case feature may still be accessible for a further Agree operation that serves to value the uninterpretable φ-features of a predicate located in a higher phase, giving rise to a form of multiple agreement where a single argument repeatedly triggers agreement on different predicates. This state of affairs requires a modification of the assumption that an uninterpretable φ-set can only be valued by a DP with an activating unchecked structural Case feature. For example, one might pursue the idea that agreement checking is completely independent of Case checking (contra Chomsky 2000). Somewhat less radical, we may assume that the final disappearance of an activating Case feature that is marked for deletion may be postponed to a later stage of the derivation (cf. Pesetsky & Torrego 2001; Carstens 2003). For the time being, I adopt the latter proposal. For the sake of concreteness, I assume that in Tsez, an uninterpretable Case feature that is marked for deletion is finally deleted at the point where the relevant DP is sent to Spell-out. On the assumption that the left edge of a phase is spelled-out together with the complement of the next higher phase head, a Case-marked DP that occupies the left edge of an embedded CP phase remains accessible to further syntactic operations triggered by a probe in the next higher phase (i.e., the matrix νP).30 Further support for the claim that there is no direct relation between Case checking and agreement relations comes from Georgian (for details cf. Harris 1981; Hewitt 1996; see King 1994 for a similar conclusion with respect to Georgian). Georgian exhibits a split ergative system: in the present tense, future and other so-called ‘series I tenses’, Georgian exhibits nominative case on the subject and dative on the object.31 In contrast, in the so-called ‘series II’ tenses such as the aorist (also referred to as ‘simple past’) or the subjunctive, the subject carries ergative marking, while the (direct)



JB[v.20020404] Prn:2/08/2005; 9:49



F: LA8103.tex / p.32 (1837-1891)

The Rise of Agreement

object is marked with nominative (or absolutive) case (examples taken from Comrie 1978: 351f.):32 Georgian – Present (48) a. Student-i midi-s. student-nom go.pres.3sg ‘The student goes.’ b. Student-i ceril-s c’er-s. student-nom letter-dat write.pres.3sg ‘The student writes the letter.’ Georgian – Aorist (49) a. Student-i mivid-a. student-nom go.aor.3sg ‘The student went.’ b. Student-ma ceril-i da-c’er-a. student-erg letter-nom prev-write.aor.3sg ‘The student wrote the letter.’ Interestingly, the agreement pattern exhibited by the verb is not affected by the case marking changes in question. Agreement always works in a nom-acc fashion, that is, the ergative subject of a verb in the aorist (‘series II’) is marked on the verb in the same way as the nominative subject of a verb in the present tense (‘series I’). This is illustrated by the following examples where the verb agrees with the 1st person subject, independent of whether the latter is marked by ergative or nominative case (taken from Bejar 2003: 180).33 (50) me tseril-s v-c’er. I.nom letter-dat 1-write-pres ‘I am writing the/a letter.’ (51) me cign-eb-i da-v-c’er-e. I.erg book-pl-nom prev-1-write-indic.aor ‘I wrote the books.’ These facts suggest that there is no correlation between the realization of Case and the realization of agreement in Georgian (cf. Marantz 1992 for a similar conclusion, but a different analysis in which both case and agreement morphemes are added postsyntactically). Note that this is expected within the analysis of agreement proposed above where Case and agreement licensing may be associated with different functional heads (and where a deletion-marked Case feature may remain accessible to further operations under certain conditions). I will not attempt to sketch a possible analysis of the intricate agreement facts of Georgian, a topic the complexity of which justifies a monograph in itself (see Bejar 2003 for a recent analysis in terms of the probe/goal system).

JB[v.20020404] Prn:2/08/2005; 9:49

F: LA8103.tex / p.33 (1891-1954)

Chapter 3. The structural design of agreement

A similar point can be made with respect to participle agreement in French. Kayne (1989) shows that participle agreement in French is dependent on overt syntactic movement. The following examples illustrate that the participle is inflected for the φ-features of the object only if the latter undergoes overt movement to the left of the participle: (no movement) Marie a fait/*-e la robe. Marie has made/-fem the dress ‘Marie has made the dress.’ (NP-movement) b. La robe a été fait-e. the dress has been made-fem ‘The dress has been made.’ (clitic movement) c. Marie l’a fait-e. Marie it-has made-fem. ‘Marie has made it.’ (wh-movement) d. Combien de tables Paul a repeint-es? how-many of-the tables Paul has repainted-fem.pl ‘How many tables has Paul repainted?’

(52) a.

In Kayne’s (1989) analysis, participle agreement results from a spec-head relation between the object and the participle in the overt syntax. This relation is established as a by-product of independently triggered movement operations such as wh-movement, NP-movement or clitic-movement. In other words, the object passes through the relevant specifier on its way further up in the tree, triggering participle agreement. In contrast, objects that remain to the right of the participle fail to trigger agreement, since they do not enter into a spec-head relation with the participle. As pointed out by Koopman (2003), it seems that the gist of this elegant analysis cannot be captured by an approach that makes use of the Agree mechanism. Due to the fact that the object is always locally c-commanded by the participle, one would rather expect that participle agreement is always possible, independent of movement: “A post participle object in French certainly occurs in the same phase as the participle, and is locally c-commanded by the participle. It is therefore mysterious why the Probe of the participle must fail in this context.” (Koopman 2003: 6). In the following, however, it is shown that an analysis in terms of Agree is capable of capturing the correlation between overt object movement and participle agreement if it is assumed that accusative Case and object agreement licensing are associated with different functional heads in French, in line with the approach to agreement advocated in this book. Note that in the case of French, this appears to be a rather natural move, since it is fairly clear that the realization of object agreement is independent of Case assignment: the object receives Case independent of whether it triggers agreement on the participle. In other words, agreement is dependent on movement, whereas Case is not. This suggests that the relevant Case-assigning head is lower in the structure than Agr, as for example in the following structure in which accusative Case is licensed under Agree by ν, while participle agreement results from a separate Agree operation



JB[v.20020404] Prn:2/08/2005; 9:49



F: LA8103.tex / p.34 (1954-2003)

The Rise of Agreement

initiated by an agreement morpheme added to a higher functional head (abstracting away from head movement):34 (53) Aux

Part Part’ íP

Part Part

Agr Agree

í’

DPobj.

í’

DPsubj. í Agree

VP V

tobj.

participle

Following Pollock (1989) (among others), I assume that the participle moves into a position below the base position of the finite auxiliary to combine with its agreement morphology (note that this structure is necessary in all approaches that assume movement of the participle to a higher Agr head; otherwise the trace left by the auxiliary should always block participle movement due to the HMC). This head might be identified as Asp or some other functional head which occupies a low position in the inflectional field. For ease of exposition, I have labeled the relevant functional head simply ‘Part(iciple)’. Importantly, the structure (53) allows us to account for the correlation between overt object movement and participle agreement if the concept of locality developed in Chomsky (2000, 2001a) is adopted (see Chapter 2 above for details). Recall that the syntactic derivation proceeds in phases, which are identified as CP and νP. Locality of movement is guaranteed by the following condition, which ensures that operations triggered by higher functional heads can target only material at the left edge of the next lower phase, that is, they cannot look into the complement of the phase head. (54) Phase Impenetrability Condition (Chomsky 2000: 108) In phase α with head H, the domain of H is not accessible to operations outside α, only H and its edge are accessible to such operations. In other words, the Phase Impenetrability Condition requires that elements must move to the left edge of their phase to be accessible for movement operations triggered by

JB[v.20020404] Prn:2/08/2005; 9:49

F: LA8103.tex / p.35 (2003-2053)

Chapter 3. The structural design of agreement

higher functional heads. For example, wh-movement of the object to SpecCP requires that the object has previously moved to the left edge of νP. Under this concept of locality, it is now possible to account for the correlation between overt object movement and participle agreement in French in the following way. After the object has been Case-licensed by an Agree relation with ν, it remains in its base position inside VP. In this position, it cannot be targeted by an Agree relation started by Agr adjoined to Part0 , since Agr cannot look down into the complement of the phase head ν. Hence, no participle agreement is possible if the object stays in situ. This analysis implies that Case checking does not depend on the existence of an unvalued φ-set in the probe, in contrast to Chomsky (2000). For the sake of concreteness, I adopt the traditional idea that Case checking involves an uninterpretable Case feature on T (nominative) and V (accusative) which is checked against a matching uninterpretable feature on D (cf. e.g. Chomsky 1993; see Pesetsky & Torrego 2004 for the related claim that agreement – i.e., φ-feature checking – and case are independent phenomena). However, if the object undergoes overt movement, triggered by a [+wh] C-head, for example, it must pass through the left edge of νP for reasons of locality. In this intermediate position (the outer SpecνP in (53)), it is accessible for an Agree relation initiated by the head complex Agr+Part, giving rise to participle agreement. Again, if we continue to assume that Case renders a goal active in the sense of Chomsky (2000), this analysis requires that the Case feature marked for deletion is still accessible after movement to the left edge of νP.35 We can therefore conclude that an analysis in terms of Agree is capable of deriving the distribution of participle agreement in French, in contrast to the critique by Koopman (2003). Note that the explanation presented here becomes only available if we assume that Case and agreement licensing are associated with different functional heads, lending further support to the concept of agreement developed above.

.. Agreement and word formation In the framework adopted in this book, inflected verbs are not assembled in the lexicon, but rather built in the syntactic and/or morphological component and later realized by the insertion of phonological exponents (see Chapter 2 for details). Accordingly, an inflected verb can only be spelled-out if it is combined with its inflectional affixes (e.g., T, Agr, Asp, Mood etc.) prior to Vocabulary Insertion.36 In other words, the requirement that an inflectional affix must attach to a lexical host is treated as a morphophonological requirement that must be satisfied prior to PF (similar to the Stray Affix Filter, Baker 1985, 1988). This can be accomplished either in the syntactic component by overt head movement to a higher functional head or at MS by Morphological Merger which combines the verb root with its inflectional morphemes post-syntactically under (structural) adjacency, giving rise to the impression of apparent syntactic lowering (‘affix hopping’ as in the case of finite main verbs in English, cf. Halle & Marantz 1993; Bobaljik 1994, 1995; Embick & Noyer 2001; Baker 2002).37



JB[v.20020404] Prn:2/08/2005; 9:49



F: LA8103.tex / p.36 (2053-2098)

The Rise of Agreement

This section deals with the mechanisms that determine the spell-out of inflectional morphemes, focusing on the creation of linear order. Similar to Julien (2002), I assume that the linearization of morphemes is a function of the way syntactic structures are mapped into sequences of phonological exponents. More precisely, the linearization of syntactic structures is assumed to be part of the Spell-out procedure, that is, the mapping to PF. Following Halle and Marantz (1993), Bobaljik (1995), Embick and Noyer (2001), I assume that linear order is created by operations connected to Vocabulary Insertion that are part of a separate morphological component, Morphological Structure (MS). However, in contrast to Julien, I do not assume that the information provided by the syntactic output suffices to completely determine the order of inflectional markers on the verb. Instead, I will adopt the more conservative view that affix order is only partially determined by the syntactic structure. The idea that the insertion of phonological material takes place after syntax at a separate level of Morphological Structure (MS) suggests that the linearization of terminal nodes, that is, the phonological realization of sequences of words and affixes, is also a function of MS: by assumption, linear order is not a property which is directly encoded in syntactic representations. Instead, the output of the syntactic component contains only hierarchical information (i.e., c-command and dominance relations) which is interpreted by MS and subsequently mapped into certain linear orders, (cf. Bobaljik 2002; Chomsky 2005). For each pair of (sister) nodes [X,Y] in a binary branching structure, the syntactic component does not determine whether X precedes Y or vice versa. Instead, the assignment of precedence relations is a function of MS. More specifically, I assume that the mapping of syntactic structures into linear orderings is part of the operation of Vocabulary Insertion (cf. Embick & Noyer 2001: 562). (55) The Late Linearization Hypothesis The elements of a phrase-marker are linearized at Vocabulary Insertion. Still, the order of inflectional markers is taken to reflect the hierarchical relations in head adjunction structures created by syntactic movement. On the standard assumption that the morphological derivation must mirror the syntactic derivation (the Mirror Principle, Baker 1985, 1988), the phonological exponent of a lower functional head must be closer to the verb stem than the phonological exponents of higher functional heads.38 This follows if Vocabulary Insertion proceeds in a bottom-up fashion, as normally assumed in Distributed Morphology. That is, Vocabulary Insertion affects the verbal or nominal root before it affects functional heads the root adjoins to (socalled ‘root-out insertion’). For example, in the following head adjunction structure, Vocabulary Insertion targets first the verbal root. Subsequently, the phonological exponent of the closest functional head is inserted (here: Asp) and combines with the verb stem. Finally, the exponent of T is attached to the existing sequence of phonological exponents, ensuring that the exponent of T is the outermost element in the resulting word.

JB[v.20020404] Prn:2/08/2005; 9:49

F: LA8103.tex / p.37 (2098-2165)

Chapter 3. The structural design of agreement

(56)

T Asp V

T Asp

In this fashion, the effects described by the Mirror Principle follow from the way linear order is created at the point of the derivation where phonological exponents are inserted into the terminal elements of the syntactic structure. The ordering restrictions imposed by the hierarchical structure assembled in the syntactic component are supplemented by lexical properties of individual Vocabulary items to create the final ordering instructions which are sent to the articulatory system.39 The lexical properties in question determine the orientation of affixes (i.e., the prefix/suffix distinction). In other words, the linearization of a head adjunction structure depends on selectional properties of bound forms: prefixes select a host to their right, whereas suffixes require a host to their left. On these assumptions, a structure such as (56) may give rise to the sequences in (57), but excludes those in (58):40 (57) a. b. c. d.

V+Asp+T (uniformly suffixing) T+Asp+V (uniformly prefixing) T+V+Asp (T prefix, Asp suffix) Asp+V+T (T suffix, Asp prefix)

(58) a. *V+T+Asp b. *Asp+T+V The sequences in (57c–d) represent possible but typologically marked options, which are not very frequent across the world’s languages. This can be attributed to the assumption that there is normally a default orientation for inflectional affixes in a given grammar, that is, languages prefer to be uniformly prefixing or suffixing (cf. e.g. Bybee et al. 1990; Hale 1996).41 In contrast to the purely syntactic approach to affix order proposed in Julien (2002), the model outlined in this section requires no non-standard assumptions. Moreover, under the assumption that the order of embedding of functional categories is not subject to parametrization, the above model appears to be at least as restrictive as Julien’s much more complicated model. Recall, for example, that the (non-existing) sequence Asp-T-V can be derived in Julien’s approach as the result of Asp-to-T movement plus word formation with the V-head which stays in situ (see Section 3.4 above). In contrast, this order is ruled out in the model presented here without further stipulations: a structure like (56) can only be mapped into a linear sequence where the exponent of Asp is closer to the verb stem than the exponent of T. Let us now explore how agreement morphemes play out in this model. Above, it has been shown that cross-linguistically, the ordering properties of agreement markers differ considerably from those of other inflectional formatives. This was taken to suggest that agreement heads do not project a separate phrase, but are adjoined to other



JB[v.20020404] Prn:2/08/2005; 9:49



F: LA8103.tex / p.38 (2165-2212)

The Rise of Agreement

functional categories. The multitude of ordering possibilities observed with agreement markers were then attributed to the assumption that agreement morphemes can attach to a variety of functional heads. In what follows, it is argued that the special ordering possibilities of agreement markers may also result directly from the particular phrase-structural character of agreement morphemes as elements which adjoin to other functional heads. The basic idea is that the resulting head adjunction structures are compatible with different linear orderings and that a given grammar may choose between the possible outcomes. Compare the following two head adjunction structures, where the verb root combines with functional heads X and Y: Y

(59) a. X V

X

b. Y

X

X V

Y X

(59a) represents the structure created by successive head movement to the functional heads X and Y, while (59b) represents the structure resulting from verb movement to a functional head X that hosts a base-adjoined agreement morpheme Y. Above, we noted that a structure like (59a) rules out the linear orderings *V-Y-X and *X-Y-V, where the higher functional head is realized closer to the verb stem than the lower functional head. This was attributed to the idea that linearization (i.e., Vocabulary Insertion) proceeds in a bottom-up fashion, first affecting the lexical root (i.e., the most deeply embedded terminal node), then the next higher terminal element etc. In the case of a ‘normal’ head adjunction structure such as (59a), the choice of the next higher element is straightforward. Each terminal element corresponds to a unique head domain, which excludes the next higher head domain. This is indicated in the following phrase marker: Y

(60) X V

Y X

Following (Chomsky 1986: 9), exclusion is defined as follows: (61) Exclusion α excludes β if no segment of α dominates β. In the course of mapping syntactic structures into linear order, the insertion procedure moves upward in the phrase marker, processing one head domain after the other. The lexical root constitutes a separate linearization/head domain by default, since it is the starting point of Vocabulary Insertion.42

JB[v.20020404] Prn:2/08/2005; 9:49

F: LA8103.tex / p.39 (2212-2273)

Chapter 3. The structural design of agreement

Let us now address the question of whether (59b) imposes similar restrictions on the mapping into linear order. Again, Vocabulary Insertion starts at the lexical root. Subsequently, the insertion procedure has to determine the next terminal node that is targeted by Vocabulary Insertion. However, (59b) crucially differs from (59a) in that both X and Y are part of the same head domain, with Y adjoined to X. X

(62) X V

Y X

In other words, in contrast to (59a), the head domain of X does not exclude Y and the insertion procedure has in principle access to both terminal elements in the head domain. I assume that in this structure, insertion may target either X or Y as the next terminal node where a phonological exponent is inserted. Under the assumption that the choice between X and Y (during Vocabulary Insertion) is open to parametrization in structures like (59b), we derive the result that the kind of head adjunction structures created by the insertion of agreement heads may give rise to a greater variety of possible linear orderings across languages – even in a subset of languages that is uniform concerning the adjunction site of Agr (e.g., in which AgrS is always adjoined to T).

.. Section summary In this section, I have developed a realizational model of predicate-argument agreement. To account for the special distributional properties of agreement markers, it was proposed that agreement heads are base-generated as adjuncts to ‘substantial’ functional heads. Here, the usual choices are T for subject agreement and ν for object agreement. By assumption, however, agreement morphemes may attach to any other functional head, which is one reason for the multitude of ordering possibilities found across languages. Subsequently, I addressed the question of how feature matching between the controller’s φ-set and the agreement morpheme is established during the syntactic derivation. Following the basic insight reached in Section 3.3 above that an analysis in terms of the operation Agree is to be preferred over the traditional mechanism of spec-head agreement, I claimed that the feature content of agreement morphemes is identified under closest c-command with an appropriate goal (the interpretable φ-set of the DP that controls agreement). Under the assumption that agreement morphemes may attach to other functional heads apart from T and ν, this analysis predicts that Case and agreement licensing can be dissociated. This prediction is borne out in Georgian where the verb agrees with the external argument, regardless of whether the latter carries nominative or ergative case. Moreover, it was shown that participle agreement in French is also amenable



JB[v.20020404] Prn:2/08/2005; 9:49



F: LA8103.tex / p.40 (2273-2331)

The Rise of Agreement

to an analysis where AgrO is dissociated from the functional head that assigns accusative Case. The proposed analysis in terms of Agree also accounts for the fact that participle agreement requires overt object movement. The latter property was attributed to the Phase Impenetrability Condition, which forces movement of the object to the left edge of νP, raising the object into the agreement domain of the higher AgrO-morpheme. Finally, I addressed the question of how agreement markers are attached to the verb stem. I argued that the linearization of inflectional morphemes is only partially determined by hierarchical relations established in the syntax and that the ordering instructions sent to the articulatory system are also influenced by lexical properties of individual Vocabulary items (i.e., the prefix/suffix distinction). On the assumption that the mapping of syntactic structure into linear orderings is a function of Vocabulary Insertion, I derived the effects of the Mirror Principle by assuming that head adjunction structures are processed in a bottom-up fashion. Furthermore, it became clear that the workings of the mapping procedure introduce another possible source of cross-linguistic variation with respect to the linear placement of agreement markers apart from the possibility that Agr may attach to different functional heads.

. Multiple agreement: Inflected complementizers in Germanic Drawing on evidence from the phenomenon of complementizer agreement, this section argues that we must acknowledge the existence of another agreement mechanism in addition to the analysis of ‘canonical’ syntactic agreement (in terms of Agree) outlined in the previous section. More specifically, it is claimed that complementizer agreement (and presumably other instances of multiple agreement) results from the post-syntactic insertion of a so-called dissociated agreement morpheme at the level of Morphological Structure. A number of non-standard varieties of Germanic is characterized by the presence of inflection in the C-domain, usually referred to as complementizer agreement (cf. Bayer 1984; Altmann 1984; Weiß 1998, 2002 on Bavarian; Bennis & Haegeman 1984; Haegeman 1990, 1992; Shlonsky 1994; de Vogelaer et al. 2002 on (West) Flemish; de Haan & Weerman 1986; Hoekstra & Marácz 1989 on Frisian; Zwart 1993a, 1993b, 1997a on dialects of the eastern and southern Netherlands; Hoekstra & Smits 1999 for an overview). Thus, we can observe a peculiar form of multiple agreement where there is not only an agreement relation between the subject and the verb, but also between the subject and elements located in the CP domain (primarily complementizers). This kind of agreement is pretty uncommon across the world’s languages, but frequently exhibited by German and Dutch dialects, which share another typologically marked property, V2. Therefore, it is often assumed that the availability of inflected complementizers is linked to the V2 property. The phenomenon is illus-

JB[v.20020404] Prn:2/08/2005; 9:49

F: LA8103.tex / p.41 (2331-2369)

Chapter 3. The structural design of agreement

trated in (63) with examples from West Flemish (Bennis & Haegeman 1984; Haegeman 1990, 1992): West Flemish (63) a. da-n *(=k) ik werk-en that-1sg=clit.1sg I work-1sg ‘That I work’ b. da-t *(=j) gie werk-t that-2sg=clit.2sg you work-2sg c. da-t *(=j) ij werk-t that-3sg=clit.3sg.masc he work-3sg d. da-t (=ze) zie werk-t that-3sg=clit.3sg.fem she work-3sg e. da-t (=t) tet werk-t that-3sg=clit.3sg.neut it work-3sg f. da-n (=me) wunder werk-en that-1pl=clit.1pl we work-1pl g. da-t *(=j) gunder werk-t that-2pl=clit.2pl you.pl work-2pl h. da-n (=ze) zunder werk-en that-3pl=clit.3pl they work-3pl In the above examples, we can see that the complementizer carries an inflectional ending the shape of which varies in relation to the φ-features of the subject. While 1sg, 1pl, 3pl subjects trigger the ending -n, -t is found with 2sg, 3sg, 2pl subjects. In West Flemish, the form of complementizer agreement is identical to the verbal agreement endings. Note that the inflected complementizers are followed by subject clitics which are obligatory in some person/number combinations and optional in others (obligatory: 1sg, 2sg, 3sg.masc, 2pl; optional: 3sg.fem, 3sg.neut, 1pl, 3pl).43 West Flemish differs in two important ways from the other Germanic languages that exhibit inflected complementizers. First, West Flemish has a complete paradigm of complementizer agreement for all persons and numbers. Second, only in West-Flemish a clitic and a full subject pronoun can co-occur in addition to complementizer agreement, as shown by the examples in (63). The examples in (64) demonstrate that clitic doubling is impossible with full referential subject DPs and that the clitic is obligatory if no other subject is present (examples taken from Shlonsky 1994: 354). (64) a. *da-t=ze that-3sg=clit.3sg.fem ‘That Marie works’ b. da-t *(=ze) that-3sg=clit.3sg.fem ‘That she works’

Marie werk-t Marie work-3sg werk-t work-3sg



JB[v.20020404] Prn:2/08/2005; 9:49



F: LA8103.tex / p.42 (2369-2449)

The Rise of Agreement

In other German and Dutch dialects, complementizer agreement is generally restricted to certain person/number combinations. In most varieties of Bavarian, complementizer agreement is found only in 2nd person contexts, in the eastern Netherlands, it is restricted to 1pl, in the south to 1pl and 3pl, and in Frisian to 2sg (plus 2pl in some varieties, similar to the Brabants dialect of Dutch).44 Another kind of typological variation concerns the shape of the agreement affixes found on C. In the majority of varieties these are identical with the verbal agreement endings (cf. the West Flemish data in (63)). However, in a set of dialects spoken in the eastern Netherlands and Brabants, the shape of complementizer agreement differs from verbal agreement (Zwart 1993b; similar facts can be observed with 1pl in some Lower Bavarian dialects, see below). This is illustrated by the following example from an eastern dialect, where the 1pl ending on the complementizer is -e, while it is -t on the verb (in clause-final position):45 (65) datt-e wij speul-t that-1pl we play-1pl ‘That we play’ (Zwart 1993b: 253) Finally, in those varieties where complementizer agreement differs from verbal agreement, the former replaces the latter in inversion contexts (cf. e.g. Zwart 1993b: 254). Again, relevant examples come from the eastern Netherlands (similar examples are found in Brabants):46 (66) a.

Wij speul-t/*-e. we play-1pl ‘We play.’ b. Waar speul-e/*-t wij? where play-1pl we ‘Where do we play?” (Zwart 1993b: 254)

The examples in (66) show that the finite verb must carry regular verbal agreement (found e.g. on verbs in clause-final position) when it appears in a subject-initial main clause, as in (66a), whereas in inversion contexts, it displays the agreement ending normally found on complementizers, cf. (66b). Thus, the ultimate shape of the finite verb is determined by the syntactic position in which it occurs.47 In some varieties of Germanic, the presence of agreement in C licenses referential pro-drop. This is true of Frisian (2sg) and Bavarian (2sg, 2pl), for example: Frisian (67) a. Kom-st (do) jûn? come-2sg you tonight ‘Do you come tonight?’ b. dat-st (do) jûn kom-st that-2sg you tonight come-2sg ‘That you come tonight’ (Zwart 1993b: 256)

JB[v.20020404] Prn:2/08/2005; 9:49

F: LA8103.tex / p.43 (2449-2561)

Chapter 3. The structural design of agreement

Bavarian (68) a. Kumm-st pro noch Minga, dann muas-st pro me b’suacha. come-2sg to Munich then must-2sg me visit ‘If you come to Munich you must visit me.’ b. Kumm-ts pro noch Minga, dann müass-ts pro me b’suacha. come-2pl to Munich then must-2pl me visit ‘If you come to Munich you must visit me.’ (Bayer 1984: 211) (69) a.

ob-st pro noch Minga kumm-st whether-2sg to Munich come-2sg ‘...whether you come to Munich’ b. ob-ts pro noch Minga kumm-ts whether-2pl to Munich come-2pl ‘...whether you (pl) come to Munich’ (Bayer 1984: 240)

Hence, Frisian and Bavarian differ from West Flemish, which requires the presence of a subject clitic in the absence of a full pronoun/DP subject, as shown in (64b) above. Dialects spoken in the east and south of the Netherlands behave similar to West Flemish – the presence of complementizer agreement does not license pro-drop (examples taken from Zwart 1993b: 257): Southern varieties (70) a. Komm-e *(ze)? come-pl they ‘Do they come?’ b. ovv-e *(ze) komme whether-pl they come-pl ‘whether they come’ Eastern varieties (71) a. Speul-e *(we)? play-1pl we ‘Do we play?’ b. datt-e *(wij) speul-t that-1pl we play-1pl ‘That we play’ The descriptive findings reached so far are summarized in Table 4.48 The following section briefly reviews a selection of the most prominent theoretical approaches to the phenomenon of complementizer agreement in Germanic, introducing more data from other languages and highlighting a number of general issues that an adequate analysis has to address.



JB[v.20020404] Prn:2/08/2005; 9:49



F: LA8103.tex / p.44 (2561-2586)

The Rise of Agreement

Table 4. Properties of complementizer agreement in Germanic.

West Flemish Frisian

NL: Eastern dialects (Hellendoorn)

Paradigm

Identical with AgrS

AgrC replaces Additional clitic AgrC AgrS (inversion doubling: licenses contexts) AgrC+clit.+pron. pro-drop

complete defective (2sg; in some varieties also 2pl) defective (1pl)

yes yes

no no

yes no

no yes

no

yes

no

no

NL: Southern dialects

defective (1pl, 3pl)

yes

no

no

no

Bavarian

defective (2sg, 2pl; plus 1pl in some varieties)

yes (apart from 1pl)

no (apart from 1pl)

no

yes

.. Previous accounts of complementizer agreement The phenomenon of complementizer agreement is commonly attributed to a structural relation that holds between C and a lower inflectional head. This relation is often modeled in terms of syntactic head movement which provides C with inflectional features (e.g. Hoekstra & Marácz 1989: I-to-C; Zwart 1993b: AgrS-to-C). Alternatively, it has been proposed that the inflection found on the complementizer heads its own projection in the C-domain and is licensed in a specifier-head relation (Shlonsky 1994). In the following, it is shown that both analyses are confronted with a number of conceptual and empirical problems. In addition, I will examine whether the data in question can be more successfully accounted for by an analysis that makes use of the Agree mechanism, as proposed by Carstens (2003). Note that the following discussion is rather selective, focusing on the major lines of thought on this topic, while alternative approaches are only briefly commented on.

... Infl-to-C movement The ‘traditional’ approach to complementizer agreement attributes the inflection on C to a structural relation that holds between C and a lower inflectional head, which is usually modeled in terms of syntactic head movement.49 For example, Hoekstra and Marácz (1989) attribute the presence of agreement morphology on C to Infl-to-C movement. The problematic fact that agreement is realized on the finite verb as well is analyzed in terms of feature percolation from the C+Infl complex down to the verb. This general idea is adopted in slightly modified form in various publications of JanWouter Zwart (1993a, 1993b, 1997a, 2001), who argues that AgrS-to-C movement is

JB[v.20020404] Prn:2/08/2005; 9:49

F: LA8103.tex / p.45 (2586-2651)

Chapter 3. The structural design of agreement

triggered as a means to (case-) license the subject in SpecAgrSP in embedded clauses. In order to account for the fact that we find multiple agreement on both the verb and the complementizer, Zwart assumes (1993b: 267, fn. 21) that it is not the complete AgrS head that moves to C, but only an abstract agreement morpheme, which is involved in the formal licensing of the subject. In addition, he proposes that AgrS carries an additional (agreement) feature that engages in the licensing of the inflected verb (at LF), which stays in situ in the overt syntax.50 The latter assumption is supposed to ensure that complementizer agreement does not block the presence of agreement morphology on the verb.51 According to Zwart, further support for an analysis in terms of (general) AgrSto-C movement comes from the eastern dialects where the morphological realization of complementizer agreement differs from that of verbal agreement, repeated here for convenience: (72) a.

Wij speul-t/*-e. we play-1pl ‘We play.’ b. Waar speul-e/*-t wij? where play-1pl we ‘Where do we play?” (Zwart 1993b: 254)

Zwart argues that this asymmetry follows from the fact that the verb occupies different head positions in (72a) (AgrS) and (72b) (C). The fact that the finite verb apparently inflects for complementizer agreement in (72b) is analyzed as another overt morphological reflex of AgrS-to-C movement (recall that Zwart assumes that AgrS-to-C takes place in all embedded clauses and in main clauses with subject-verb inversion). At closer inspection, however, the phenomenon in question constitutes a serious problem for an analysis in terms of head movement. Relevant data come from Hellendoorn, an East Netherlandic dialect (Carstens 2003; Ackema & Neeleman 2004), where complementizer agreement can be observed with 1pl subjects, as shown in (73). Hellendoorn (73) datt-e wiej noar’t park loop-t that-1pl we to-the park walk-1pl ‘That we are walking to the park’ (Carstens 2003: 397) Similar to other dialects spoken in the eastern Netherlands, the realization of 1pl complementizer agreement (-e) differs from the relevant verbal agreement ending (-t). Furthermore, finite verbs in subject-verb inversion contexts exhibit the agreement morphology typical of complementizers (1pl only), whereas ‘normal’ verbal agreement is found on verbs in subject-initial main clauses: Hellendoorn (74) a. Volgens miej lop-e wiej noar’t park. according-to me walk-1pl we to-the park ‘According to me we are walking to the park.’



JB[v.20020404] Prn:2/08/2005; 9:49

F: LA8103.tex / p.46 (2651-2731)

 The Rise of Agreement

b. Wiej loop-t noar’t park. we walk-1pl to-the park ‘We are walking to the park.’ (Carstens 2003: 398) In other words, Hellendoorn seems to exhibit exactly the same behavior as the other eastern dialects discussed by Zwart. Interestingly, however, the presence of an adjunct that intervenes between C and the subject blocks the availability of complementizer agreement in Hellendoorn. This restriction holds for both main and embedded clauses, as can be seen from the examples in (75). Hellendoorn (75) a. dat/*datt-e [op den wärmsten dag van’t joar] wiej tegen that/that-1pl on the warmest day of-the year we against oonze wil ewärkt hebt. our will worked have ‘That on the warmest day of the year we have worked against our will’ b. Volgens miej loop-t/*lop-e [op den wärmsten dag van’t according-to me walk-1pl/walk-1pl on the warmest day of-the joar] ook wiej noar’t park. year also we to-the park ‘According to me we are also walking to the park on the warmest day of the year.’ (Carstens 2003: 398) On the assumption that complementizer agreement is the result of AgrS-to-C movement, its non-availability in V2-clauses such as (75b) where exactly this movement operation has taken place comes as a surprise. More generally, under standard assumptions, it is clear that the intervening PP should not block head movement (in both main and embedded contexts).52 Similar adjacency effects can be observed in the other Germanic varieties which exhibit complementizer agreement, cf. the following example from Bavarian, where complementizer agreement is impossible if an adverb intervenes between C and a full subject pronoun:53 Bavarian (76) *obwoi-st woartscheints du ins Kino ganga bist although-2sg probably you to-the movies gone are ‘although you probably went to the movies’ (Günther Grewendorf, p.c.) We can therefore conclude that an analysis in terms of head movement cannot adequately account for the phenomenon of complementizer agreement (see e.g. Carstens 2003: 401ff. for further arguments against Zwart’s analysis of complementizer agreement). In the next section, we will consider an alternative approach that is based on the assumption that agreement is universally licensed in a specifier-head configuration, attributing complementizer agreement to the presence of an AgrP in the CP domain.

JB[v.20020404] Prn:2/08/2005; 9:49

F: LA8103.tex / p.47 (2731-2767)

Chapter 3. The structural design of agreement

... AgrP within CP Shlonsky (1994) develops an account of complementizer agreement which is based on a split CP structure and makes use of the idea that agreement phenomena always involve a specifier-head relation between the controller and a syntactic agreement head (Belletti 1990, 2001; Chomsky 1993; Koopman 2003; cf. Roberts 1994 for a related approach). The analysis focuses on West Flemish, but is claimed to extend to the other Germanic languages that exhibit inflected complementizers. Shlonsky proposes the following structure for the left periphery of the clause, where AgrC is the host of the agreement features that give rise to inflected complementizers: (77) [CP [C’ C [AgrCP [AgrC’ AgrC [IP ... ]]]]] It is claimed that the content of AgrC is licensed either via insertion of a subject clitic in its specifier or by movement of a full nominal subject to SpecAgrCP, which by assumption is an A-position. Thus, the additional inflection in the C system is licensed in a separate agreement relation apart from the specifier-head configuration within IP (which serves to establish verbal agreement). Shlonsky further assumes that the φ-features located in AgrC attach to the complementizer via syntactic AgrC-toC movement. The analysis of an example such as (78) is illustrated by the phrase marker in (79). The full subject pronoun is located in its Case position SpecIP, where it licenses the set of φ-features located in I0 , giving rise to verbal agreement. The φfeatures of AgrC enter into a separate licensing relation with the subject clitic, which is base-generated in SpecAgrCP. (78) da-t (=ze) zie werk-t that-3sg=clit.3sg.fem she work-3sg (79)

CP C’ AgrCP

C da

Spec ze

AgrC’ AgrC -t

IP Spec zie

I’ VP

I

tV

werk-t

In the absence of a subject clitic, the subject (either a full DP or pronoun) is assumed to undergo A-movement to SpecAgrCP to license the φ-features in AgrC.54 The adjacency



JB[v.20020404] Prn:2/08/2005; 9:49

F: LA8103.tex / p.48 (2767-2828)

 The Rise of Agreement

effects observed above are taken to follow from the assumption that nothing can intervene between C (hosting the inflected complementizer) and the clitic/full pronoun in SpecAgrCP (Shlonsky rules out adjunction to AgrCP). However, this analysis faces the problem that the relevant spec-head relation between the subject and AgrC is never overtly realized, either because AgrC-to-C movement obscures the existence of a previous spec-head relation and the subject appears to the right of the inflected complementizer or because there is simply no subject overtly present, as for example in Bavarian, which exhibits pro-drop in 2nd person contexts (cf. (69) above, repeated here for convenience): (80) a.

ob-st pro noch Minga kumm-st whether-2sg to Munich come-2sg ‘...whether you come to Munich’ b. ob-ts pro noch Minga kumm-ts whether-2pl to Munich come-2pl ‘...whether you (pl) come to Munich’ (Bayer 1984: 240)

Another shortcoming of Shlonsky’s analysis concerns the hierarchical relation between the functional head that carries the agreement features and the lexical element where agreement is realized morphologically. It is standardly assumed that the lexical host of agreement morphology has to move up to a higher functional head to check agreement features. In other words, Agr is higher than its lexical host. However, in Shlonsky’s account, it’s the other way around (note that an analysis in terms of I-to-C movement is subject to the same criticism).55 In addition to these conceptual issues, Shlonsky’s analysis can be shown to be incompatible with data from other languages with inflected complementizers. Recall that Shlonsky claims that his account carries over to languages such as Frisian and Bavarian, which in contrast to West Flemish generally exclude the presence of a subject clitic in contexts with complementizer agreement. Thus, it is exactly these contexts that license pro-drop in Bavarian (Frisian behaves similarly, as noted above). This raises the question of how the content of AgrC can be licensed when the specifier of AgrCP is apparently not filled. Shlonsky is led to assume that in Bavarian (and Frisian), SpecAgrCP is occupied by a non-overt counterpart of the subject clitics of West Flemish, which serves to identify the φ-features present in AgrC in the contexts that license complementizer agreement (i.e., 2sg and 2pl). However, note that Bavarian has an elaborate system of subject clitics, exhibiting gaps only for the 2nd person nominative forms (in contrast to claims by Shlonsky (p. 362); cf. Altmann 1984; Bayer 1984; Weiß 1998; see Chapter 5 for details). Now, under Shlonsky’s analysis, it is totally unexpected that complementizer agreement is only found in those cases where no (overt) subject clitic exists. Instead, we should expect to find inflected complementizers rather in those instances where overt subject clitics exist, that is, 1sg/pl, 3sg/pl, contrary to the facts. Finally, the analysis proposed by Shlonsky makes wrong predictions for languages which show a reflex of Agr-on-C in main clauses as well. Shlonsky adopts an asymmet-

JB[v.20020404] Prn:2/08/2005; 9:49

F: LA8103.tex / p.49 (2828-2890)

Chapter 3. The structural design of agreement 

rical analysis of V2 (basically following Travis 1984; Zwart 1993a), according to which the finite verb moves to AgrC in subject-initial clauses and further to C in clauses with topicalized non-subjects. However, verb movement to AgrC (and further up) does not result in double affixation of AgrS and AgrC to the verb: (81) *Werk-t-t Marie? work-Agrs.3sg-Agrc.3sg Marie ‘Does Marie work?’ (Shlonsky 1994: 364) To rule out examples such as (81), Shlonsky stipulates – following Rizzi and Roberts (1989), who analyze head movement as a substitution operation – that “the contents of AgrC0 must be deleted to create a position into which the inflected verb may substitute” (p. 367). However, we have seen above in example (74) (repeated here for convenience) that in languages such as Hellendoorn (and other eastern dialects), there is a reflex of Agr-on-C in V2 clauses in the sense that the verb carries ‘complementizer agreement’ in inversion contexts, whereas it retains its normal agreement pattern in subject-initial clauses: Hellendoorn (82) a. Volgens miej lop-e wiej noar’t park. according-to me walk-1pl we to-the park ‘According to me we are walking to the park.’ b. Wiej loop-t noar’t park. we walk-1pl to-the park ‘We are walking to the park.’ These data create two problems for Shlonsky’s analysis. First, it is unclear why the content of AgrC is not deleted, but rather overwrites the content of AgrS in dialects such as Hellendoorn. Second, within the asymmetrical analysis of V2 adopted by Shlonsky, the contrast between (82a) and (82b) suggests that the φ-features giving rise to complementizer agreement are associated rather with C instead of AgrC, since the verb carries complementizer agreement only when it occupies the higher head (in inversion contexts). Summing up, we have seen that an analysis of complementizer agreement in terms of a separate specifier-head relation in the (split) CP domain can avoid some of the conceptual problems that arise in the ‘traditional’ head movement approach, for example by providing separate licensing configurations for distinct agreement phenomena. However, exactly the strong assumption that agreement always involves a specifierhead configuration necessitates a number of additional assumptions to account for the West Flemish data. Furthermore, it has been shown that this strong assumption leads to serious complications and wrong empirical predictions if data from other Germanic languages are taken into consideration. We can therefore conclude that complementizer agreement in Germanic should not be modeled in terms of a specifier-head relation, which corroborates our findings in Section 3.3 above concerning the structural relation involved in predicate-argument agreement. In the next section, we will

JB[v.20020404] Prn:2/08/2005; 9:49

F: LA8103.tex / p.50 (2890-2973)

 The Rise of Agreement

see whether the problems encountered so far can be resolved by an analysis that makes use of the Agree mechanism proposed in Chomsky (2000, 2001a).

... Multiple Agree Recently, Carstens (2003) has proposed an analysis of complementizer agreement in Germanic that makes use of the probe/goal mechanism developed in Chomsky (2000, 2001a, 2001b). Carstens assumes that the agreement morpheme on C (in her analysis: Fin) is valued under closest c-command (i.e., via the operation Agree) by the interpretable φ-features of the subject which occupies SpecTP.56 Before this Agree operation takes place, the subject’s φ-set has already been accessed by the uninterpretable φ-features of T (in its base position SpecνP, prior to movement to SpecTP), which serves to Case-license the subject and ensures that the finite verb carries agreement morphology. Note that this instance of ‘multiple Agree’ represents a deviation from the assumption that a category with an inactive (i.e., deletion-marked) Case feature should no longer be accessible to syntactic operations (Chomsky 2000, 2001a). That is, it should not be possible for the subject to enter into an additional probe-goal relation with the non-interpretable φ-features located in C after its Case feature has been deletion-marked by T. Accordingly, Carstens slightly relaxes Chomsky’s original proposal and claims that a Case feature that has been marked for deletion remains syntactically active until the next strong phase is reached (cf. Pesetsky & Torrego 2001 and Section 3.6.2 above for similar assumptions). This enables the subject to participate in multiple Agree relations, in the case at hand first with T and then with C. Again, it can be shown that this analysis does not account for a number of restrictions on complementizer agreement in a satisfactory way. The first set of problematic data has to do with the fact that the availability of complementizer agreement is subject to an adjacency requirement, as noted above. Recall that in Hellendoorn, for example, the presence of an adjunct which intervenes between C and the subject blocks the availability of complementizer agreement. This restriction holds for embedded clauses as well as inversion contexts (cf. (75) above, repeated here as (83)). Hellendoorn (83) a. dat/*datt-e [PP op den wärmsten dag van’t joar] wiej tegen that/that-1pl on the warmest day of-the year we against oonze wil ewärkt hebt. our will worked have ‘That on the warmest day of the year we have worked against our will’ b. Volgens miej loop-t/*lop-e [PP op den wärmsten dag according-to me walk-1pl/walk-1pl on the warmest day van’t joar] ook wiej noar’t park. of-the year also we to-the park ‘According to me we are also walking to the park on the warmest day of the year.’

JB[v.20020404] Prn:2/08/2005; 9:49

F: LA8103.tex / p.51 (2973-3016)

Chapter 3. The structural design of agreement 

Carstens attributes the absence of complementizer agreement in (83) to an intervention effect in the sense of Chomsky (2000, 2001a). More precisely, she claims that the adjoined adverbials in (83) bear an abstract Case feature that identifies the adverbial as a possible goal for the φ-set in C. As a consequence, the adverbial “disrupts closest ccommand of the subject by C0 ” (p. 398), thereby blocking the valuing and realization of complementizer agreement. However, apart from the fact that it is quite unusual to assume that PP adverbials like op den wärmsten dag van’t joar carry a Case feature, this analysis leads to a serious problem with respect to the Agree relation that serves to value the φ-set of T. Recall that Carstens assumes that the φ-set of T initiates an Agree relation with the subject in SpecνP. Now, under her analysis, we should expect that adverbials that intervene between T and the base position of the subject should give rise to the same kind of intervention effects that are taken to block Agree between C and the subject in SpecTP. Of course, this is not the case. Therefore, we can conclude that the analysis of Carstens (2003) does not provide an adequate account of the Hellendoorn data either. In the next section, I will present an alternative account of complementizer agreement which is based on the idea that the relevant phenomena should receive an explanation in terms of morphological operations carried out in the post-syntactic level of MS.

.. A Late Insertion account of complementizer agreement The fact that complementizer agreement is sensitive to an adjacency requirement suggests that this form of agreement is not a purely syntactic phenomenon, but at least partially determined by properties of PF (or MS). A similar conclusion is reached by Ackema and Neeleman (2004: Ch. 7), who develop an account of complementizer agreement which is based on the assumption that feature checking may also be accomplished in the phonological component of grammar (i.e., PF). More precisely, they propose the following PF feature checking rule, which applies if C and the subject occur in the same prosodic phrase (marked by braces):57 (84) Germanic complementizer agreement {[C (Prt) (Add) (Plr)] [D (Prt) (Add) (Plr)]} → {[C (Prti ) (Addj ) (Plrk )] [D (Prti ) (Addj ) (Plrk )]} (Ackema & Neeleman 2004: 241) The rule in (84) serves to identify the set of φ-features associated with C (Prt = Participant, Add = Addressee, Plr = Plural) with the relevant (interpretable) φ-features of the subject. The adjacency effect illustrated in (83) is then explained as follows: due to the presence of an intervening XP between C and the subject, rule (84) cannot apply, since the complementizer and the subject are in two different prosodic domains. Thus, an account in terms of PF feature checking seems to be able to capture the major empirical facts of the phenomenon under consideration and is therefore superior to the syntactic analyses discussed above. However, this analysis requires that PF is endowed with powerful syntax-like properties like identifying, checking and ma-

JB[v.20020404] Prn:2/08/2005; 9:49

F: LA8103.tex / p.52 (3016-3087)

 The Rise of Agreement

nipulating φ-features. This becomes particularly clear when we take a closer look at Ackema and Neeleman’s explanation of the fact that complementizer agreement is always with the subject and never with the object. Even in contexts where C forms a prosodic phrase with a scrambled object (excluding the subject), the complementizer cannot agree with the object and must appear in its non-agreeing form: (85) a.

dat [zulke boeken]i zelfs Jan ti niet leest that such books even Jan not reads ‘That even John does not read such books’ b. {dat zulke boeken} {zelfs Jan} {niet leest} (Ackema & Neeleman 2004: 242)

To prevent that the feature content of C is identified with the object’s φ-features, Ackema and Neeleman (p. 242) assume that PF feature checking can only access elements in A-positions. By assumption, then, the scrambled object in (85) is not visible for the relevant PF rule since it occupies an A’-position. However, this requires that PF is not only capable of identifying, checking and manipulating φ-features. In addition, PF must be sensitive to the A/A’-distinction. In other words, Ackema and Neeleman claim that information which is only legible in the syntactic module (types of syntactic positions) must also play a crucial role in PF-operations, contra standard assumptions. This establishes a syntax after the syntax proper, which is a rather dubious theoretical move. Next, it is shown that Ackema and Neeleman’s account also faces a serious empirical problem. More specifically, it appears that the evaluation of agreement features in the C-system does not involve a dependency between C and the subject’s φ-features.58 Rather, complementizer agreement is shown to be parasitic on the overt presence of an inflected verb at PF. The assumption that C carries its own set of non-interpretable φ-features, which initiates a checking relation with the subject (either in the syntax or at PF), predicts that the establishment of complementizer agreement is independent of the realization of verbal agreement. At least in Bavarian, however, this expectation is not borne out by the facts. Consider the comparative clauses in (86) (Bayer 1984: 269):59 (86) a.

D’Resl is gresser [als wia-st du bist]. the-Resl is taller than as-2sg you are ‘Resl is taller than you are.’ b. *D’Resl is gresser [als wia-st du]. the-Resl is taller than as-2sg you c. D’Resl is gresser [als wia du]. the-Resl is taller than as you

In comparatives, overt agreement on C leads to ungrammaticality if the finite verb is absent from the structure, cf. (86b). The sentence becomes acceptable when the complementizer bears no inflection, cf. (86c), an apparent violation of the generalization that complementizer agreement is obligatory in all contexts (for 2nd person), see above. This suggests that the possibility of complementizer agreement depends

JB[v.20020404] Prn:2/08/2005; 9:49

F: LA8103.tex / p.53 (3087-3144)

Chapter 3. The structural design of agreement 

on the presence of a finite verb within the same clause. In other words, it appears that Agr-on-C is parasitic on the presence of another already valued agreement morpheme, Agr-on-T. Crucially, these examples show that agreement between the complementizer and the subject cannot be implemented in terms of a checking relation between the set of φ-features in C and the lower subject. Otherwise one would expect examples such as (86b) to be grammatical (the φ-set of C should be able to enter into a checking relation with the subject’s φ-set). This conclusion holds for a syntactic analysis in terms of Agree (Carstens 2003) as well as for an account involving PF checking rules as proposed by Ackema and Neeleman (2004). Still, comparatives such as (86b, c) provide further evidence for the suggestion that complementizer agreement must operate post-syntactically. It is standardly assumed that comparatives are to be analyzed as the result of post-syntactic PF-operations that delete the inflected verb in the second clause, as shown in (87) (cf. Bresnan 1973; Lechner 2001).60 (87) D’Resl is gresser [als wia (*-st) du bist]. the-Resl is taller than as-2sg you (are) ‘Resl is taller than you are.’ Under this scenario, we would not expect any interaction with complementizer agreement if insertion and valuation of agreement morphemes were to take place in the syntax, since the finite verb would be present throughout the whole syntactic derivation, being subject to deletion only after the structure has been transmitted to Morphological Structure (and, finally, PF). In other words, it would remain a mystery why post-syntactic deletion of the finite verb affects the realization of Agr-on-C in these contexts in the way it does. In contrast, this interaction comes out much more naturally in a Late Insertion model where the insertion and valuation of Agr-morphemes on C take place after syntax as well, in the sense that the post-Spell-out operations that bring about agreement may be sensitive to other post-Spell-out operations such as deletion of the finite verb in examples like (86b, c). A similar argument can be made with respect to sluicing, that is, instances where an IP within a wh-CP is elided (cf. e.g. Ross 1969; Merchant 2001): (88) I woas dass-ts ihr a Madl gseng hoabts, owa I woas net I know that-2pl you a girl seen have-2pl but I know not wo (*-ts) ihr a Madl gseng hoabts. where-2pl (you a girl seen have-2pl) ‘I know that you have seen a girl, but I don’t know where.’ (Günther Grewendorf, p.c.) In examples such as (88), we can observe that complementizer agreement is not available after the lowest IP has been deleted at PF. Again, this sensitivity to post-syntactic processes is totally unexpected if it is assumed that complementizer agreement is established by syntactic operations. We can therefore conclude that certain properties of complementizer agreement in Germanic cannot be accounted for by a purely syntactic

JB[v.20020404] Prn:2/08/2005; 9:49

F: LA8103.tex / p.54 (3144-3179)

 The Rise of Agreement

analysis in terms of Agree. Instead, the sensitivity to properties such as linear adjacency or PF-deletion seems to indicate that the phenomenon of complementizer agreement is to be located in the post-syntactic components of grammar, that is, at MS or, more generally speaking, the mapping to PF. To account for these observations, I suggest a hybrid model of agreement in which agreement phenomena may result from either the pre-syntactic or the post-syntactic insertion of agreement morphemes. As proposed in Section 3.6 above, canonical subject agreement is then analyzed as the result of an agreement morpheme which enters the syntactic derivation as part of T0 and is valued by the operation Agree: (89) [CP... [TP T + Agr ... [íP subject ... ]]] Agree

In contrast, complementizer agreement is analyzed as resulting from a morphological operation, the post-syntactic insertion of a dissociated Agr-morpheme at the level of MS. Hence, I assume that the constituent structure of morphemes derived in the syntax can be modified by the post-syntactic insertion of (functional) morphemes which adjoin to syntactic terminal nodes (cf. Marantz 1992; Halle & Marantz 1993; Noyer 1997; Embick 1997). Following Embick (1997), these morphemes are called ‘dissociated’, since they are not present in the syntactic derivation and merely reflect properties expressed by structural configurations in the syntax proper.61 Accordingly, dissociated morphemes are not part of the numeration. The insertion process is illustrated by the following pair of phrase markers, where (90a) represents the output of the syntactic component and (90b) shows the structure resulting from the insertion of a dissociated Agr-morpheme to the C head, which is subsequently subject to Vocabulary Insertion (note that the Agr-morpheme on T is already present and valued in the syntax). (90) a.

CP C’ C

TP subj.

T’ VP

T T

tV

V

Agr T

JB[v.20020404] Prn:2/08/2005; 9:49

F: LA8103.tex / p.55 (3179-3230)

Chapter 3. The structural design of agreement  CP

b.

C’ C C

TP Agr

subj.

T’

(inserted at MS)

VP

T T

tV

V

Agr T

How can we ensure that the feature content of the post-syntactically inserted agreement morpheme on C matches the φ-features of the subject? By assumption, feature matching between Agr-on-C and the subject does not take place directly, but is mediated by another agreement morpheme that has been valued by a syntactic Agree relation. Recall that the Bavarian comparatives in (86)–(87) suggest that the licensing of Agr-on-C depends on the presence of (an already valued) Agr-on-T. To capture this observation, I assume that the feature content of the dissociated Agr-morpheme on C is determined by the feature content of an Agr-morpheme that has already been valued in the syntax, that is, Agr-on-T. More precisely, I propose that a dissociated morpheme which expresses agreement with an argument (here: the subject) is a copy of a relevant Agr-morpheme that has been valued by a syntactic Agree relation (here: Agr-on-T). This mechanism ensures feature identity between these different types of Agr-morphemes (which both reflect the φ-feature content of the same argument).62 Thus, we can formulate the following (informal) condition on the insertion of dissociated Agr-morphemes.63 (91) The insertion of a dissociated Agr-morpheme is parasitic on the presence of an Agr-morpheme that has been valued in the syntax. Accordingly, complementizer agreement is a primarily morphological process which is nevertheless sensitive to the syntactic environment. This account explains the restriction on complementizer agreement observed in Bavarian comparatives if we assume that the insertion of dissociated Agr-morphemes is a morphological operation that applies after the deletion of the inflected verb in comparatives (i.e., the X0 -structure formed by adjoining the verb to the T+Agr complex in (90)).64 In Distributed Morphology, rules operating at MS are usually subject to locality constraints. More specifically, operations such as Impoverishment or Morphological Merger may involve only structurally adjacent morphemes (cf. Halle & Marantz 1993 and Note 37 above). With respect to the insertion of dissociated Agr-morphemes, the relevant condition must capture the intuition that the relation between the syntactically valued Agr-morpheme and its late-inserted copy is sufficiently local. This can be

JB[v.20020404] Prn:2/08/2005; 9:49



F: LA8103.tex / p.56 (3230-3294)

The Rise of Agreement

achieved by the following condition on the insertion of dissociated Agr-morphemes (and the definition of structural adjacency in (93)).65 (92) Insertion of dissociated Agr-morphemes A dissociated Agr-morpheme can attach to a functional head X only if X is structurally adjacent to another functional head Y hosting an Agr-morpheme that has been valued in the syntax. (93) Structural adjacency A head X is structurally adjacent to a head Y iff (i) X c-commands Y (ii) There is no projecting syntactic head Z that (a) is c-commanded by X and (b) c-commands Y.66 In other words, (93) guarantees that a head X is structurally adjacent to the head Y of its complement. Hence, dissociated Agr-on-C can only be inserted if C is structurally adjacent to T, the latter hosting a valued Agr-morpheme. It is now possible to attribute the adjacency effects observed in Hellendoorn to the locality condition (92) if we assume that scrambled XPs intervening between C and TP are not adjoined to TP, but occupy the specifier of a separate projection (Haeberli 2002: AgrSP, Grewendorf 2004: TopP/FocP). Following proposals by Rizzi (1997) concerning the presence of TopP/FocP in the left periphery, I assume that this projection is present only if it serves to implement certain information-structural distinctions. Otherwise it is absent from the structure. In the structure (94), then, the (‘optionally’ projected) TopP hosting the PP op den wärmsten dag van’t joar disrupts structural adjacency between C and T. This blocks the insertion of a dissociated Agr-morpheme on C. As a consequence, complementizer agreement leads to ungrammaticality in contexts such as (94). (94) *[CP datt-e [TopP [op den wärmsten dag van’t joar] [TP wiej tegen that-1pl on the warmest day of-the year we against oonze wil ewärkt hebt]]] our will worked have ‘That on the warmest day of the year we have worked against our will’ How does this analysis account for the fact that in a certain set of dialects, we can observe a similar adjacency effect in matrix clauses?67 Recall that the presence of a scrambled XP prevents the exponent of Agr-on-C from replacing the verbal agreement ending in examples like (75b), repeated here as (95). Hellendoorn (95) Volgens miej loop-t/*lop-e [TopP [op den wärmsten dag according-to me walk-1pl/walk-1pl on the warmest day van’t joar] [TP ook wiej noar’t park]]. of-the year also we to-the park ‘According to me we are also walking to the park on the warmest day of the year.’

JB[v.20020404] Prn:2/08/2005; 9:49

F: LA8103.tex / p.57 (3294-3372)

Chapter 3. The structural design of agreement

In examples like (95), the syntactically evaluated Agr-morpheme adjoined to T is part of the complex C-head, due to V-to-T-to-C movement. So the question arises of whether the T+Agr complex is structurally adjacent to C in the relevant adjunction structure (96). This would generally license the insertion of a dissociated Agrmorpheme in V2 contexts, predicting that adjacency effects are absent in matrix clauses, contrary to the facts. (96)

C2 C1

T T V

Agr T

However, note that on standard assumptions, a category does not c-command material adjoined to it (Kayne 1994: 16; Chomsky 1995: 339f.).68 As a consequence, C1 does not c-command T in (96). The adjunction structure in (96) thus fails to meet the conditions for the insertion of a dissociated Agr-morpheme, since T is not structurally adjacent to C. It follows that in V2 contexts, the realization of complementizer agreement can only be licensed by structural adjacency (i.e., c-command) between the two-segment category [C2 , C1 ] and the (non-pronounced) lower copy of the T+Agr complex, which is disrupted by the TopP hosting the scrambled XP in (95). Interestingly, not all elements that intervene between C and an additional subject (or rather, the TP) block the realization of complementizer agreement (in contrast to claims by Carstens 2003: 401, Fn. 20: “My understanding is that agreement of C with a subject across any intervening material is impossible”). In Bavarian, which exhibits a similar adjacency requirement as Hellendoorn (as shown in (97)), modal particles such as aber, halt, ja and clitic object pronouns may intervene between Agr-on-C and TP/the subject (cf. e.g. Altmann 1984; Nübling 1992):69 (97) *obwoi-st woartscheints du ins Kino ganga bist although-2sg probably you to-the movies gone are ‘Although you probably went to the movies’ (Günther Grewendorf, p.c.) (98) dass-st oaba du ibaroi dabei bist that-2sg modprt you everywhere with-it are ‘That you really are involved everywhere’ (Altmann 1984: 205) (99) wia-sd-n du gseng hoast when-2sg-clit.3sg you seen have-2sg ‘When you saw him’ (Pfalz 1918: 231) Note that these examples do not represent counterexamples to the analysis proposed above if we assume that the structural position of clitics and modal particles differs from that of scrambled XPs, which move into a specifier position of a TopP or FocP



JB[v.20020404] Prn:2/08/2005; 9:49



F: LA8103.tex / p.58 (3372-3420)

The Rise of Agreement

intervening between C and TP. With respect to modal particles of Germanic, it is often assumed that they are base-generated as adjuncts (here: TP-adjuncts) (cf. e.g. Abraham 1995).70 Accordingly, they do not require the projection of a separate TopP or FocP and do not disrupt structural adjacency between C and TP. Concerning the placement of object clitics, I assume that their ultimate surface position is determined by late MS-processes such as Prosodic Inversion, which apply at the mapping to PF (cf. Bonet 1991; Halpern 1992; Schütze 1994). Therefore, they reach their surface position after the insertion and valuation of dissociated Agr-morphemes has been completed. Again, no interaction between these two processes is expected. In other words, the correct generalization seems to be that only scrambled XPs which undergo syntactic movement to a topic or focus position between C and TP block the realization of complementizer agreement, while base-generated adjuncts (i.e., modal particles) and object clitics that have undergone PF-movement do not disrupt the structural adjacency between C and T. This difference is expected on the analysis of complementizer agreement presented here, but difficult to account for in a purely syntactic approach to the phenomenon in question. Similar adjacency effects can be observed in connection with so-called agreement weakening in Standard Dutch (cf. e.g. Ackema & Neeleman 2003, 2004). As is wellknown, the 2sg agreement ending /-t/ is dropped (i.e., replaced by /-Ø/) in inversion contexts, resulting in a form homophonous to the 1st person singular: (100) a.

dat jij dagelijks met een hondje over straat loop-t that you daily with a doggy over street walk-2sg ‘That you walk with a doggy in the street every day’ b. Jij loop-t dagelijks met een hondje over straat. you walk-2sg daily with a doggy over street c. Dagelijks loop-Ø jij met een hondje over straat. daily walk you with a doggy over street (Ackema & Neeleman 2004: 193)

Similar to complementizer agreement, agreement weakening is blocked when a scrambled XP intervenes between the verb in C and the subject pronoun: (101) *Volgens mij ga-Ø [op de heetste dag van’t jaar] zelfs jij according-to me go on the hottest day of-the year even you naar het park. to the park (Ackema & Neeleman 2004: 196) Ackema and Neeleman (2003, 2004) develop an account of Dutch agreement weakening which is based on similar assumptions as their analysis of complementizer agreement (see above). More specifically, they propose an impoverishment rule which operates at PF and deletes the verb’s [Addressee] feature if the verb and the 2sg subject pronoun are part of the same prosodic phrase (marked by braces):

JB[v.20020404] Prn:2/08/2005; 9:49

F: LA8103.tex / p.59 (3420-3486)

Chapter 3. The structural design of agreement

(102) Dutch Agreement Weakening {[V Prt Add] [D Prt Add]} → {[V Add] [D Prt Add]} (Ackema & Neeleman 2004: 194) After impoverishment, a less specified exponent may be inserted into the relevant terminal node (/-Ø/, which is specified for [Participant] only), resulting in the weakening effect. The absence of agreement weakening in (101) is then attributed to the fact that the finite verb and the subject are no longer in the same prosodic domain if a scrambled XP intervenes between these two elements. As a result, the impoverishment rule in (102) cannot apply. Thus, according to Ackema and Neeleman, complementizer agreement and agreement weakening are effected by two different PF rules which are both sensitive to prosodic domains: complementizer agreement is attributed to a rule (see (84) above) which accomplishes PF feature checking, while agreement weakening is taken to be the outcome of a PF impoverishment rule. However, note that the mechanism proposed in this section, that is, the insertion of dissociated Agr-morphemes, makes available a single analysis of these superficially unrelated phenomena. More specifically, I assume that in Standard Dutch, a dissociated Agr-morpheme is added to the C-head in 2sg contexts. In other words, I claim that Standard Dutch exhibits a form of complementizer agreement as well, despite appearances. By assumption, the exponent of the relevant Agr-morpheme is phonetically empty (i.e., /-Ø/), in contrast to varieties with overtly inflected complementizers. As a result, the presence of Agr-on-C cannot be detected in embedded clauses of Standard Dutch. Still, its insertion does give rise to visible effects in the context of agreement weakening, where the exponent of Agr-on-C, /-Ø/, replaces the regular agreement ending 2sg /-t/ (recall that in a given X0 -complex, only the highest of two identical Agr-morphemes is spelled out, see Note 62). Similar to other instances of complementizer agreement, the adjacency effect illustrated in (101) can then be attributed to the fact that the scrambled XP disrupts structural adjacency between C and T, which prevents Agr-on-C from being licensed. To sum up, this section has argued that complementizer agreement in Germanic should be analyzed as resulting from the post-syntactic insertion of a dissociated Agrmorpheme, the feature content of which is identified under structural adjacency with another Agr-morpheme that has been valued in the syntactic derivation. We thus have to recognize the existence of a morphological mechanism giving rise to agreement phenomena, in addition to the syntactic licensing of Agr-morphemes discussed in Section 3.6. It seems likely that this analysis of complementizer agreement in Germanic carries over to other cases of multiple agreement, as for example in the following Swahili sentence (cf. (12) above, repeated here for convenience): (103) wa-toto wa-li-kuwa wa-me-ki-soma ki-tabu cl2-children cl2-past-be cl2-perf-cl7-read cl7-book ‘The children had read the book.’ (Krifka 1995: 1416)



JB[v.20020404] Prn:2/08/2005; 9:49



F: LA8103.tex / p.60 (3486-3553)

The Rise of Agreement

Similar to complementizer agreement in Germanic, the multiple occurrences of the agreement marker wa- (signaling class 2 noun class agreement) can be analyzed as resulting from the post-syntactic insertion of a dissociated Agr-morpheme, the content of which is identified in relation to a single Agr-morpheme that has been valued in the syntactic derivation under closest c-command with the DP wa-toto ‘the children’. At this point, however, I will leave the exact specifics of such an analysis open for future research (see Chapters 4–6 for the role of dissociated Agr-morphemes and multiple agreement in the grammaticalization of new agreement morphology).

. Conclusion In this chapter, I have outlined key features of a formal model of predicate-argument agreement. Needless to say, the exposition scratched only on the surface of the vast amount of diverse and intricate agreement phenomena found across the world’s languages, leaving aside topics such as anti-agreement, A’-agreement, context-sensitive agreement, or the role of morphological processes such as Impoverishment and Morphological Merger/Fusion in the realization of agreement, each of which would justify a monograph in itself (cf. e.g. Bobaljik 1995 on the realization of verbal inflection in Germanic; Noyer 1997 for an in-depth morphological analysis of complex agreement patterns found in a set of Afro-Asiatic and native American languages; Chung 1998 for A’-agreement in Chamorro; Bruening 2001 for long-distance agreement and A’-agreement in Algonquian; Bejar 2003; Rezac 2004 for an Agree-based analysis of a set of intricate agreement phenomena such as context-sensitive agreement or socalled agreement displacement in languages like Georgian, Algonquian, or Basque). Instead, I focused on the basic ingredients of the agreement relation, that is, the phrasestructural representation of agreement morphemes, the structural mechanisms that serve to identify their feature content and the way agreement markers attach to the verb stem. In Section 3.2, I presented a variety of arguments which suggest that agreement differs significantly from ‘substantial’ functional categories such as T, Neg or Asp, which are commonly assumed to receive an interpretation at the LF interface. These differences were taken to suggest that Agr does not head a separate projection in the syntax, but rather adjoins to other, ‘substantial’ functional heads. Section 3.3 dealt with the structural relation under which the feature content of these ‘parasitic’ Agrmorphemes is identified. Here I pointed out that evidence from long-distance agreement in Tsez suggests that the relevant structural configuration is closest c-command (i.e., the Agree relation) and not spec-head agreement. In Section 3.4, I focused on the question of how affix order is linked to syntactic structure, arguing that it is not possible to completely derive the order of inflectional markers from hierarchical relations created in the syntax, contra Julien (2002). In Section 3.6, the theoretical insights reached in the previous sections were worked out in some more detail. I proposed that agreement heads are inserted into

JB[v.20020404] Prn:2/08/2005; 9:49

F: LA8103.tex / p.61 (3553-3607)

Chapter 3. The structural design of agreement

the syntactic derivation as adjuncts to ‘substantial’ functional heads, with the ‘normal’ choices of T for subject agreement and ν for object agreement. In addition, agreement morphemes may attach to other functional heads, which is one reason for the multitude of ordering possibilities found across languages. Following the basic insight reached in Section 3.3, I claimed that feature matching between the controller’s φset and the agreement morpheme is established during the syntactic derivation via an Agree operation initiated by an Agr-head (or, rather, its non-valued φ-set). Under the assumption that Agr-morphemes may attach to other functional heads apart from T and ν, this analysis predicts that Case and agreement licensing can be dissociated, that is, implemented by different functional heads. This prediction was shown to be borne out by long-distance agreement in Tsez and intricate agreement patterns exhibited by Georgian where the verb agrees with the external argument, regardless of whether the latter carries nominative or ergative case. In addition, I argued that participle agreement in French is also amenable to an analysis in which AgrO is dissociated from the functional head that assigns accusative case. The observation that participle agreement requires overt object movement was then attributed to the Phase Impenetrability Condition, which forces movement of the object to the left edge of νP, raising the object into the Agree domain of the higher AgrO-morpheme. Next, I addressed the question of how agreement markers are attached to the verb stem, arguing that linear order is only partially determined by the hierarchical relations established during the syntactic derivation. More specifically, I adopted a more conservative approach, assuming that the ordering instructions sent to the articulatory system are influenced by lexical properties of individual Vocabulary items (i.e., the prefix/suffix distinction), in addition to structural restrictions commonly attributed to the Mirror Principle. Concerning the latter, I argued that its effects can be derived if head adjunction structures are processed in a bottom-up fashion during Vocabulary Insertion. As I pointed out, the workings of the mapping procedure introduce another source of variation with respect to the placement of agreement markers: due to the fact that an agreement morpheme is part of the head domain of the functional head X it adjoins to, the insertion procedure may target Agr either before or after X, giving rise to different linear orderings of X and Agr relative to the verb stem. In Section 3.7, I took a closer look at complementizer agreement in Germanic. It was shown that the realization of complementizer agreement is sensitive to factors such as adjacency and PF-deletion of the inflected verb, which creates a problem for purely syntactic approaches involving mechanisms such as head movement, spec-head agreement or Agree. Instead, these facts were taken to suggest an alternative account according to which complementizer agreement is instantiated post-syntactically (i.e., at MS) via the insertion of dissociated Agr-morphemes which adjoin to C. By assumption, the identification of the feature content of these post-syntactically inserted Agr-morphemes does not involve a direct structural relation with an interpretable φ-set. Rather, feature matching with the subject is mediated under structural adjacency with another agreement morpheme that has been valued by a syntactic Agree relation before.



JB[v.20020404] Prn:2/08/2005; 9:49



F: LA8103.tex / p.62 (3607-3650)

The Rise of Agreement

We thus arrive at a theory of syntactic agreement where feature matching between a controller and a target may result from two different underlying mechanisms: first, the surface realization of predicate-argument agreement may result from agreement morphemes that are added to other functional heads before the resulting complex head enters the syntactic derivation. The feature content of ‘syntactic’ agreement morphemes is then identified/valued via an Agree operation, which leads to feature matching with an appropriate set of interpretable φ-features under closest ccommand. In addition, agreement morphemes may be added after the syntax as copies of syntactically evaluated Agr-morphemes. The insertion of these so-called dissociated Agr-morphemes typically leads to instances of multiple agreement where agreement with a certain argument is realized in several places in a sentence. In the remainder of this book, it is argued that this mechanism plays an important role in the grammaticalization of new agreement markers, which often involves an intermediate stage of multiple agreement where the new agreement marker co-occurs with the old one before the latter is eventually replaced by the former. Note that apparently, the analysis developed in this chapter has not much to say about a number of the typological generalizations about agreement noted in Chapter 1 such as the suffixing preference or the observation that many languages lack verbal agreement markers for 3rd person arguments. Some of these questions will be taken up again in the next chapters, where it will be argued that a set of synchronic properties of predicate-argument agreement can be explained as a result of the way agreement markers arise historically.

Notes . Note that the discussion in this chapter focuses on approaches to verbal agreement that have been proposed within the range of Principles and Parameters theory. As a consequence, alternative approaches which make use, for example, of feature unification and which are prevalent in frameworks such as Generalized Phrase Structure Grammar (GPSG) or Head-driven Phrase Structure Grammar (HPSG) will not be commented on. . An additional SpecTP is only necessary for multiple subject constructions as for example transitive expletive constructions (TECs) in Icelandic (see e.g. Chomsky 1995: Ch. 4; Bobaljik & Jonas 1996 for discussion of TECs). . Note that the Mirror Principle was originally devised to account for the ordering of derivational morphemes. However, its implications for inflectional morphology were immediately recognized after the split IP structure had been introduced. . Of course, there are instances of different aspect and tense markers within a clause, as in examples with a future perfect like Joe will have left the party by two o’clock where each of the different tense markers/heads contributes some clearly identifiable piece of information to the interpretation of the clause. Thus, in contrast to what is shown below for agreement, it is likely that each tense marker corresponds to a discrete functional head (cf. e.g. Zeller 1994). However, note that there are very few languages which show the phenomenon of ‘tense agreement’, where one and the same tense formative appears in more than one place in a given sentence (Joachim

JB[v.20020404] Prn:2/08/2005; 9:49

F: LA8103.tex / p.63 (3650-3716)

Chapter 3. The structural design of agreement

Sabel informed me that in Malagasy, for example, tense morphology shows up on the verb and on certain adverbials). . A related phenomenon can be observed in Finnish, where subject agreement is normally realized as a verbal suffix which alternatively attaches to the negation. (i)

a.

b.

puhu-i-n. speak-past-1sg ‘I spoke.’ e-n puhu-nut. neg-1sg speak-past ‘I did not speak.’ (Mitchell 1994: 124)

. Table 2 is a slightly modified representation of Julien’s figures (her Table 5.2). Note that intervening markers of other categories are ignored. Bound markers are indicated by the notation ‘X+Y’. In addition, fused or portmanteau marking, where it is not possible to segment a given inflectional marker into separate tense and agreement markers, is indicated by a slash between the relevant categories (e.g., ‘AgrS/T’). Importantly, note that the numbers given in Table 2 refer to genera/families, not to individual languages. If the members of a genus/language family uniformly show a certain combination of word order and affix order, this is counted as a single instance of this combination. If a given language family is not uniform with respect to word and affix order, every single combination found in one genus is listed in the table. Therefore, the figures in the body of the table do not necessarily add up to the figures in the rightmost column. For example, the second type of affix combination, AgrS/T+V is found in seven genera. However, the figures given in the body of the table do not add up to seven, but to nine. Accordingly, some of the seven genera that exhibit the affix order AgrS/T+V must contain more than one type of word order. An empty cell indicates that no language in Julien’s sample exhibits the relevant combination of word and affix order. Note that Table 2 does not make reference to split agreement marking. For example, in languages such as Turkana or Cree, person marking employs prefixes whereas number is marked by suffixes; in Turkana (Dimmendaal 1983), the sequence of inflectional markers is apparently AgrSPerson+T+V+AgrSNumber. . Differences concerning the possibility of a certain order apparently only have to do with the bound/non-bound character of the inflectional markers. For example, there are 64 genera that exhibit V+T+AgrS, which is the most frequent pattern in Julien’s sample, but there is no language that shows V+T AgrS, that is, the same sequence of markers with a free instead of a bound postverbal subject agreement marker. . Intervening markers of other categories are ignored. Again, the figures in Table 3 refer to genera. Note that Table 3 indicates only the linear order of the inflectional affixes. No distinction is made between free and bound markers. The notation “AgrS/O” refers to portmanteau markers that realize both subject and object agreement. . Interestingly, Julien (p. 253) notes that the affix sequences expected on the basis of the clause structure proposed by Chomsky (1991), (1993) (AgrSP-TP-AgrOP-VP) are not very frequent cross-linguistically: AgrS-T-AgrO-V is found in only 4 genera and its suffixal mirror image VAgrO-T-AgrS is found in only 6 genera. In contrast, it appears that the tense marker preferably intervenes between the verb and suffixed AgrS and AgrO markers (cf. V-T-AgrS-AgrO, 9 genera; V-T-AgrO-AgrS, 9 genera, both mostly SOV). Furthermore, the often stated claim (cf. e.g. Speas 1991) that object agreement is closer to the verb stem than subject agreement appears to be only a (mild) tendency across languages: in Julien’s sample, AgrO is closer to the verb stem than AgrS in 42 genera, while it’s the other way around in 26 genera (note that these numbers refer only to



JB[v.20020404] Prn:2/08/2005; 9:49



F: LA8103.tex / p.64 (3716-3800)

The Rise of Agreement

languages without fused AgrS/AgrO markers and where AgrS and AgrO are located on the same side relative to the verb stem). . Haeberli (2000) develops a mixed approach that relates the presence of separate agreement projections to the richness of verbal agreement morphology. More precisely, he suggests that agreement features are contained in T but project a separate AgrSP (above TP) if (i) the verb carries separate tense and agreement suffixes (cf. Thráinsson 1996; Bobaljik & Thráinsson 1998; Bobaljik 2003), or (ii) if the agreement morphology signals more than a default/non-default two-way distinction. If these morphological conditions are not met (e.g., as in English, which exhibits only a two-way distinction where the default agreement ending is -Ø and the non-default is 3.sg.pres.indic /-z/), no separate AgrSP is projected and the agreement features contained in T are checked by the subject, which raises to SpecTP for EPP reasons. . Another piece of evidence which supports this view on agreement comes from research on agrammatic speech production. Friedmann and Grodzinsky (2000) study the linguistic abilities of a Hebrew speaking agrammatic aphasic. They show that in agrammatic speech production, tense is impaired, but agreement is intact. F&G conclude that these functional categories must be dissociated in the syntax. From the fact that properties associated with the C-system (e.g., the formation of wh-questions) are equally impaired, they further conclude that impairment of T implies impairment of other functional nodes above T. To be able to capture the different behavior of tense and agreement in agrammatic speech, F&G assume that TP dominates AgrSP in Hebrew. However, this set of assumptions fails to account for the fact that in languages like French, Italian or German agrammatic aphasics show impairment patterns that are similar to those found in Hebrew. In other words, F&G would predict that agreement is impaired in a language where AgrSP dominates TP. But this is obviously not the case. F&G speculate that these data can be handled if “agreement is checked in a mechanism different from that for tense. It may not have a node of itself, but it checks in one of the other checking points below T and is thus preserved.” (p. 99). These findings point into the same direction as the other empirical evidence considered so far: there is a fundamental phrase-structural difference between agreement and ‘core’ functional categories such as Tense. . Note that it is not necessary that the agreement controller originates in the same clause as the target. If no CP boundary intervenes between the matrix and the embedded clause, Amovement may feed agreement. Relevant examples include Raising in English, illustrated in (i) and so-called long passive in German (Höhle 1978; Haider 1986, 1993; Wurmbrand 2001), shown in (ii). (i)

[The books]i seem [ ti to be on sale].

(ii) a.

b.

c.

weil er [den Stuhl zu reparieren] versuchte since he the chair.acc to repair tried ‘Since he tried to repair the chair’ weil der Stuhl zu reparieren versucht wurde since the chair.nom to repair tried was ‘Since they tried to repair the chair’ weil [der Stuhl und der Tisch] zu reparieren versucht wurden since the chair and the table.nom to repair tried were ‘Since they tried to repair the chair and the table’

In (i), the thematic subject of the embedded non-finite clause has undergone A-movement into the matrix clause, from where it controls (number) agreement on the matrix verb. This possibil-

JB[v.20020404] Prn:2/08/2005; 9:49

F: LA8103.tex / p.65 (3800-3862)

Chapter 3. The structural design of agreement

ity is commonly attributed to the assumption that raising verbs either select a non-finite IP/TP (cf. e.g. Chomsky 2000) or delete the CP layer of the non-finite complement clause (Chomsky 1981). In a similar vein, long passive is only possible with infinitival complements of so-called ‘restructuring verbs’ such as versuchen ‘try’, vergessen ‘forget’ or beginnen ‘begin’, which give rise to clause union effects (see e.g. Bech 1955; Grewendorf 1987; Sabel 1996; Wurmbrand 2001). In restructuring infinitives, passivization of the matrix verb affects the case-assignment properties of the lower verb. Accordingly, the embedded object undergoes A-movement into the matrix clause for case assignment, triggering agreement on the matrix auxiliary. . In both approaches, the domain of subject agreement extends to material contained in the VP if there is no external argument (i.e., no νP) present, that is, in passives and unaccusatives. In a spec-head analysis, the object can undergo movement to SpecTP for case and agreement if no external argument intervenes (in SpecνP). In an Agree-type analysis, the absence of an external argument implies the absence of a νP phase boundary. As a consequence, an Agree operation initiated by T can access elements contained in the VP. Note that this raises the interesting question of whether material in an embedded SpecCP is accessible for subject agreement with passive/unaccusative matrix verbs. . Recall that Chomsky’s original motivation for the introduction of the Agree mechanism partially comes from instances of long-distance agreement in Icelandic, where a matrix verb may agree in number with a nominative DP that is contained in a lower clause (cf. Chapter 2 above). . Polinsky (2003) distinguishes between three types of (apparent) long-distance agreement. First, the impression of long-distance agreement might arise in a configuration where two distinct but coreferent arguments occur in the matrix and embedded clause, respectively. However, if the (local) agreement controller in the matrix clause happens to be phonetically empty, this may give the impression that the controller is in fact located in the embedded clause: (i)

He wants of (you)i [to throw the ball at youi ]. (Polinsky 2003: 283)

According to Polinsky, this form of ‘mediated locality’ is found in a number of Algonquian languages such as Fox, Cree, Blackfoot, or Pasamaquoddy. Another instance of only apparent long-distance agreement may result from raising or restructuring operations which bring the controller into the agreement domain of the matrix clause, similar to long passive in German. Again, this gives rise to an agreement configuration which respects the clause-boundedness constraint (examples taken from Polinsky 2003: 283): (ii) He wants youi [to be thrown the ball at ti ]. (iii) He wants-throw you the ball.

(subject-to-object raising) (clause union/restructuring)

Apart from the German long passive examples discussed above, similar phenomena can be observed in languages where subject-to-object raising/ECM feeds object agreement on the matrix verb. Polinsky mentions Imbabura Quechua, Kipsigis (Nilo-Saharan), Tsakhur (NakhDagestanian), and Hindi. Finally, there is the case of genuine long-distance agreement as exemplified by Tsez, in which the controller moves into the left periphery of the embedded clause from where it can trigger agreement on a higher target. See Polinsky (2003) for discussion and a set of diagnostics that help to distinguish between apparent and genuine long-distance agreement. It seems that all instances of long-distance agreement are linked to a topic interpretation of the embedded controller. Furthermore, some languages such as Pasamaquoddy (Bruening 2001) exhibit mediated long-distance agreement alongside genuine long-distance agreement



JB[v.20020404] Prn:2/08/2005; 9:49

F: LA8103.tex / p.66 (3862-3924)

 The Rise of Agreement

(with an element presumably in the left periphery of the embedded clause). This might be taken to indicate that historically, cases of genuine long-distance agreement developed from cases of ‘mediated locality’ via reanalysis. . Note that only vowel-initial verbs carry (prefixal) agreement markers. If the verb begins with a consonant, no agreement markers are observed. . Similar intervention effects are created by phrases marked by the topic particles -n(o) or -gon (Polinsky & Potsdam 2001: 636). Furthermore, the presence of a complementizer suffix on the embedded verb (instead of a nominalizing suffix) also prevents long-distance agreement, see Polinsky and Potsdam (2001: 635). . Ede Zimmermann pointed out to me that the particular choice of quantified expressions in (29) (∀ vs. ∃) actually does not justify any firm conclusions with respect to scope ambiguities, since one constellation (∃∀) implies the other (∀∃). Accordingly, ∃∀ does not necessarily constitute a separate reading (in the sense of a real scope ambiguity), but may be understood as a special case (of the constellation ∀∃), arising if the restriction of the existential quantifier happens to contain only a single element in a given discourse environment. In other words, more tests would actually be necessary to ascertain the presence of scope ambiguities in examples such as (29). . For example, Simpson and Bhattacharya (2003) assume that pied-piping of non-wh material in wh-questions is licensed by wh-feature percolation, which identifies constituents larger than wh-phrases as eligible for wh-movement. . Bobaljik and Wurmbrand (2004) argue for a hybrid approach where agreement may result either from Agree or from a spec-head relation created by A-movement. Data from German involving fronted VPs that contain a nominative are taken to indicate that the Agree mechanism is needed to account for the agreement and case facts found in this construction. On the other hand, anti-reconstruction effects found in the German long passive (in particular, the phenomenon of scope freezing) seem to require an analysis of case/agreement checking that necessarily involves A-movement into a specifier position of the case-assigning functional head, feeding agreement with the higher predicate. . To capture the gist of the Mirror Principle, it is sometimes assumed that the inflectional features on the verb come in a hierarchical order which reflects the order of checking relations (and therefore the hierarchy of functional projections). See Grewendorf (2002: 42) for some discussion. . As a result of this analysis, English must be analyzed as having overt verb movement, contra standard assumptions. An alternative analysis in terms of Morphological Merger under adjacency (Halle & Marantz 1993; Bobaljik 1995) is not available, since this is predicted to lead to prefixal inflection, contrary to the facts. . For this reason, Julien’s account does not provide a principled explanation for the fact that postposed inflectional markers show a strong tendency for being bound, in contrast to preposed markers (cf. Bybee et al. 1990 and Chapter 1 above). . In addition, similar problems arise due to Julien’s assumption that adverbs are merged as specifiers of functional heads (cf. Cinque 1999). Again, this conflicts with roll-up movement in SOV languages. Julien therefore assumes that merging of adverbs blocks further roll-up movement. This assumption leads to a variety of completely different derivations dependent on whether a given sentence contains certain adverbs or not. Moreover, it should not be possible that adverbs are merged in specifiers of functional heads that contribute to the verbal morphol-

JB[v.20020404] Prn:2/08/2005; 9:49

F: LA8103.tex / p.67 (3924-3998)

Chapter 3. The structural design of agreement

ogy; otherwise the presence of an adverb in the relevant specifier would block suffixation of the relevant functional head. . Note that Julien herself assumes that the finite verb in English targets a position very low in the IP domain, somewhere between a low Asp-head and Voice0 (in the system of Cinque 1999). . In a similar vein, Bobaljik (2005) argues that the realization of agreement is established by post-syntactic operations. More specifically, he claims that morphological agreement does not involve syntactic φ-feature checking/valuation, but is rather dependent on morphological case, a kind of information which is not available in the narrow syntax. See Anderson (1992) for related ideas, expressed in an alternative Late Insertion framework where inflection is not analyzed as the concatenation of discrete morphemes. Instead, complex words are the result of Word Formation Rules that interpret feature complexes generated in the syntactic component. . This gist of this analysis is reminiscent of the interarboreal head movement proposed in Bobaljik (1995), Bobaljik and Brown (1997), the only difference being that in the case of agreement morphemes, the complex head structures which enter the ongoing syntactic derivation are created by the operation Merge instead of Move. Note that in both instances, the creation of complex heads proceeds in line with the Extension Condition (Chomsky 1995). However, as pointed out to me by Ian Roberts, the idea that X0 -complexes may result from the operation Merge raises some non-trivial issues within the theory of Bare Phrase Structure (Chomsky 1995). . However, note that there is a slightly more complex alternative conceivable which perhaps combines the advantages of both approaches. More specifically, we may assume that agreement features added to functional heads (e.g., via Agree) are subject to Fission at the level of MS, creating a separate Agr-morpheme for the purposes of Vocabulary Insertion and linearization. This analysis removes Agr-heads from the syntactic computation, but maintains the possibility of separate Agr-morphemes at the level of MS. Furthermore, recall that Fission is a language specific process, which does not apply generally. This captures the observation that cross-linguistically, agreement marking often fuses with other inflectional affixes. However, for expository reasons, I will continue to represent Agr-morphemes as adjuncts to other functional heads. . For the present purposes, it does not matter whether the object moves to an outer specifier of νP after the subject has been merged or whether object movement takes place prior to the insertion of the subject. Alternatively, the object may be ‘tucked in’ in a lower specifier after the subject has been merged. See Chomsky (1995, 2000) and Richards (2001) for discussion. . This raises certain questions with respect to parametrization. It is doubtful that the deletion of uninterpretable features may wait until Spell-out in general. Otherwise we would expect that long-distance (object) agreement is a much more widespread phenomenon (contrary to the facts). Therefore, one might speculate that individual languages may differ with respect to the point at which features marked for deletion are ultimately deleted. I will leave that question open for future research. . In the history of Georgian, formerly distinct accusative and dative case endings have fallen together into a single morphological case ending -s which is commonly referred to as ‘dative’ in the literature on Georgian. . In Georgian, tense is marked by a combination of so-called ‘preverbs’ (abbreviated as prev) and fused Tense/Agr suffixes on the verb.



JB[v.20020404] Prn:2/08/2005; 9:49

F: LA8103.tex / p.68 (3998-4057)

 The Rise of Agreement . Note that this illustration is somewhat simplifying. Actually, the agreement patterns exhibited by Georgian are much more complex. In Georgian, the controller of a certain agreement slot on the verb may shift between subject and object dependent on the φ-features of the arguments in question, which Bejar (2003) calls context sensitive agreement. Cf. the following statement taken from Bejar (2003: 4): “a verb cross-references the person feature of its object, unless the object is 3rd person, in which case the person feature of the subject is cross-referenced. Number is preferentially that of the subject, unless the subject is singular, in which case the number of the object is cross-referenced, unless the object is 3rd person.” . I am indebted to Shin-Sook Kim for discussion and valuable suggestions concerning the analysis presented here. . Recall that by assumption, a deletion-marked Case feature remains is accessible to the syntactic derivation until the relevant DP is sent to the interfaces (in the case at hand, until completion of the next higher phase, that is, CP). . Thus, in contrast to proposals by Chomsky (1993), the option that the inflectional features of verbs are checked via covert raising is not available. . Marantz (1988: 261) defines the operation Morphological Merger as follows: (i)

“At any level of syntactic analysis (d-structure, s-structure, phonological structure), a relation between X and Y may be replaced by (expressed by) the affixation of the lexical head of X to the lexical head of Y.”

It is generally assumed that (structural) adjacency at MS is a necessary condition for Merger to take place (cf. Bobaljik 1994). Thus, in present-day English, T is affixed to VP-internal V at MS if nothing intervenes between these head positions; the presence of the negation not triggers do-support to pick up the inflectional morphology associated with T (cf. Chomsky 1957; Halle & Marantz 1993). Note that this analysis requires that VP-adjoined material such as adverbs does not count for the evaluation of adjacency at MS (cf. Bobaljik 1994, 1995, 2002 for concrete proposals). . Of course, these effects can only be detected if the relevant inflectional markers are located on the same side of the verb stem and if their order is not affected by other processes that take place at MS. . In addition, morphological operations such as Local Dislocation may reorder elements/heads which are adjacent and part of the same constituent at MS. Local Dislocation differs from Morphological Merger in that the latter takes place prior to linearization whereas the former applies to structures that have already been linearized (see Embick & Noyer 2001 for discussion). . One might speculate that the realization of word order (i.e., of exponents that constitute phonological words) proceeds along similar lines, resulting from the same mechanism. Note that this is actually expected in an approach such as DM which does not recognize a theoretically significant distinction between morphosyntactic features which are realized as bound or free markers, cf. Harley and Noyer (1999: 7): “Features which will eventually be realized as a subpart of a phonological word are treated no differently from features which will eventually be realized as an autonomous word.” That is, similar to the realization of affix sequences and the prefix/suffix distinction, word order (e.g., the VO/OV distinction) are presumably determined by a combination of lexical properties of certain Vocabulary items and hierarchical structures generated by the syntactic component, resulting in ordering instructions for PF. Note that the existence of an additional projection level that contains the specifier represents a complication which does not exist for the computation of affix order in head adjunction structures. Since this

JB[v.20020404] Prn:2/08/2005; 9:49

F: LA8103.tex / p.69 (4057-4129)

Chapter 3. The structural design of agreement

work is primarily concerned with the linear ordering of agreement markers, I will not further address the realization of word order here (cf. Bobaljik 2002: 213ff. for a concrete proposal). . Alternatively, it is conceivable that affixes select not only the direction where their host must appear, but also its category. Then, the marked sequences (57c–d) might follow from the fact that cross-linguistically, T preferably attaches to Asp. . A question arises with respect to lexical roots that are adjoined to other lexical roots, as for example in the case of morphological causatives (cf. e.g. Baker 1988). The model developed here predicts that in principle, insertion may either target the lower or the higher root first. As far as I can see, however, this creates no problems in the present approach, since it only affects the order of the two lexical roots, while higher functional heads are excluded by the V0 +V0 adjunction structure. Accordingly, the latter must be realized outside of the V0 +V0 complex, which captures the generalization that inflectional morphology must appear outside of derivational morphology changing grammatical functions (Baker 1988; Speas 1991). . Liliane Haegeman (personal communication) informed me that there is a strong preference to have clitic doubling in all person/number combinations, that is, the clitic is almost obligatory (if the full subject is a pronoun; otherwise no clitics are found). However, the clitic can be dropped somewhat more easily in 3sg.fem/neut, 1pl and 3pl contexts. According to Liliane Haegeman, the deletion of the clitic is presumably governed by phonological factors, in the sense that the clitic is always obligatorily present in the syntax, but can be deleted at PF if certain phonological conditions are met. A similar proposal is made by Shlonsky (p. 352f., fn. 4). In light of related phenomena discussed in Chapters 5 and 6, it seems to me that the (obligatory) subject clitics should be analyzed as additional agreement morphemes on C, the grammaticalization of which led to ‘repair’ the previously highly syncretic agreement paradigm. . To the best of my knowledge, these restrictions to certain person/number combinations are still left unexplained. Hoekstra and Smits (1999) claim that the distribution of complementizer agreement is governed by the following generalization: (i)

The Identity Generalization Complementizer agreement only occurs when the agreement ending of the inverted auxiliary in the present tense is identical to the agreement ending of the inverted auxiliary in the preterite.

Note, however, that even if (i) turns out to be correct, it is only a description of the distributional facts, and not an explanation. In Chapter 6 it is argued that at least in the case of Bavarian, the person/number restrictions ultimately receive a diachronic explanation. More precisely, it is shown that Agr-on-C is a residue of the grammaticalization of new verbal agreement endings in the history of Bavarian, which affected only 2nd person forms (plus 1pl in some dialects). . Zwart (1993b) claims that a similar difference can be observed for West Flemish. However, Liliane Haegeman pointed out to me that the endings are in fact morphologically identical and that the differences in question are most likely the result of phonological rules. . Similar facts can be observed in a subset of Lower Bavarian dialects, where complementizer agreement is also found with 1pl subjects. See Chapter 5 for details. . Note that this phenomenon provides further support for the hypothesis that syntactic terminals are supplied with phonological features after syntax (cf. Zwart 1997b). If the verb enters the derivation in fully inflected form (i.e., with phonological features), then the contextually determined change from speult to speule constitutes a problem that cannot be circumvented by base generating the verb as speule in (66b): due to the fact that the verb is first accessed by



JB[v.20020404] Prn:2/08/2005; 9:49

F: LA8103.tex / p.70 (4129-4199)

 The Rise of Agreement

the functional head(s) T/Agr, it is expected that this checking relation requires the form speult, which shows up when the verb is not moved to C. However, the change from speult to speule is easily captured in a Late Insertion model in which the morphological component interprets the representation generated by the syntactic component, assigning the head cluster V +T/Agr+C a spell-out that differs from the spell-out of V +T/Agr. . Frisian shows another restriction on complementizer agreement when the subordinate clause is embedded under a verb that optionally selects a V2 complement (commonly referred to as a ‘bridge verb’). Note that similar to the Scandinavian languages (but unlike German or Dutch), the sentential complements – including V2 clauses – of this set of verbs are always headed by a complementizer in Frisian. In these contexts, complementizer agreement is obligatory when the verb stays behind in final position, but excluded when the V2 option is chosen (de Haan & Weerman 1986; Zwart 1993b): (i)

Heit sei datst-o/*dat do soks net leauwe moast. dad said that-2sg=you/that you that not believe must-2sg ‘Dad said that you should not believe such things.’

(ii) Heit sei dat do/*datst-o moast soks net leauwe. dad said that you/that-2sg=you must-2sg that not believe . See den Besten (1982) for an early account of complementizer agreement in terms of a rule Move Tense and Bennis and Haegeman (1984) on West Flemish data. In an early analysis of complementizer agreement in Bavarian, Bayer (1984) develops an account that is based on the idea that in V2 languages, there is an abstract agreement relation between Comp, V/Infl and the subject leading to co-indexation of all three elements. In the case of 2nd person subjects in Bavarian, this form of agreement is overtly realized due to a linking rule that copies the φfeatures located in Infl to Comp. This overt manifestation of agreement serves to identify the referential content of the subject DP, thereby licensing an empty pronominal pro in the subject position. . Slightly modifying Zwart’s approach, Watanabe (2000) assumes that both T and V carry a separate set of φ-features to guarantee the presence of agreement morphology on the verb. . Note that Zwart’s account of multiple agreement on C and V raises of number of conceptual questions. First, the claim that a single functional head can carry multiple sets of agreement features is by no means a standard assumption and raises a number of questions. For example, it is well-known that the paradigms of complementizer agreement are often defective (see above). In contrast, verbal agreement paradigms tend to be much less defective (apart from a certain amount of syncretism). This asymmetry between the paradigms of verbal agreement and complementizer agreement is rather unexpected if both types of agreement features are represented by the same functional head. Furthermore, in the theoretical model adopted by Zwart (the form of Minimalism envisaged in Chomsky 1993, 1995), it is generally assumed that feature movement must affect the whole set of formal features contained in a given functional head. In other words, it should not be possible to move only a subset of the formal features contained in a given functional head, in contrast to what is assumed by Zwart. . This adjacency effect is also unexpected under the analysis proposed by Watanabe (2000), who assumes that complementizer agreement results from a two-step operation in which the subject’s φ-features are first copied onto T (as a result of Agree) and then accompany head movement of the T0 complex to C. Again, this analysis should lead us to expect that the verb always carries complementizer agreement in V2 clauses, contrary to the facts.

JB[v.20020404] Prn:2/08/2005; 9:49

F: LA8103.tex / p.71 (4199-4287)

Chapter 3. The structural design of agreement . A related adjacency effect can be observed in West Flemish: (i)

*da-t morgen zie werk-t that-3sg tomorrow she work-3sg ‘That she is working tomorrow’ (Shlonsky 1994: 360)

However, note that West Flemish generally requires strict adjacency between C and the subject (i.e., (i) is ruled out anyway, cf. Haeberli 2002, 2004). Thus, examples like (i) actually do not allow any firm conclusions concerning the sensitivity of complementizer agreement to adjacency effects in West Flemish. . For cases where there is only a subject clitic present (cf. (64b) above), Shlonsky assumes that SpecIP is filled by pro which is governed and Case-licensed by AgrC. . In addition, the assumption that SpecAgrCP is an A-position raises a number of further conceptual issues. First, there is no independent motivation for moving the full subject (in the absence of a clitic) to this position apart from the licensing needs of AgrC (i.e., the realization of agreement on the complementizer). Second, this operation conflicts with the generalization that a Case-marked NP cannot undergo further A-movement (see e.g. Chomsky 1981, 2000). . Note that this mechanism is somewhat reminiscent of the idea of Bayer (1984) that there exists an abstract agreement relation between Comp, V/Infl and the subject. . Ackema and Neeleman (2004: 235) stress that PF feature checking requires “post-head adjacency: a phrase BP whose features are to enter into a checking relation must immediately follow a head A that contains identical features [...] If another maximal projection intervenes ([AP A XP BP]), or if the phrase precedes the head ([AP BP A]), the two will not be part of the same φ [prosodic phrase], so that prosodic checking is impossible.” . Furthermore, the fact that complementizer agreement licenses pro-drop in varieties such as Bavarian and Frisian creates another serious problem for Ackema and Neeleman’s approach. More specifically, Ackema and Neeleman would have to assume that an empty subject, which is otherwise invisible at PF, is visible for the purposes of PF feature checking. . Note that similar facts hold for 1pl in a set of Lower Bavarian dialects (see Chapter 5 for some discussion), cf. (i)

De san g’scheider [(als) wia-ma mir san]. they are more-intelligent than as-1pl we are ‘They are more intelligent than we are.’

(ii) *De san g’scheider [(als) wia-ma mir]. they are more-intelligent than as-1pl we (iii) De san g’scheider [(als) wia mir]. they are more-intelligent than as we (Bayer 1984: 271) . Bresnan (1973) argues convincingly that comparatives are to be analyzed as the result of a (PF) deletion rule. Compare the following pairs of sentences: (i)

a. b.

(ii) a. b.

I’ve never seen a man taller than my father. I’ve never seen a taller man than my father. I’ve never seen a man taller than my mother. I’ve never seen a taller man than my mother.



JB[v.20020404] Prn:2/08/2005; 9:49

F: LA8103.tex / p.72 (4287-4355)

 The Rise of Agreement

While (ia) and (ib) can receive a more or less identical interpretation, (iia) and (iib) cannot. More specifically, the reading of (iib) is deviant, implying that the NP my mother refers to a man. Bresnan shows that this puzzling fact receives an elegant explanation if it is assumed that part of the than clause is deleted under identity with the head of the comparative. The difference in interpretation between (iia) and (iib) can then be attributed to the different shape of the antecedent (i.e., the head of the comparative) of the elided constituent. In (ia) and (iia), the antecedent of the deleted constituent consists of a predicative AP, as illustrated in (iii) (note that the special interpretation of comparative forms is commonly attributed to the presence of a degree phrase containing the -er suffix, labeled QP here. The than clause is then assumed to contain an identical degree phrase which encodes the degree – commonly labeled x – of a property ascribed to the subject): (iii) ... [AP [QP -er] tall][CP than [IP my father/mother [I’ is [AP [QP x ] tall]]]]] Accordingly, what is compared is how tall the subject of the than clause is. In contrast, the antecedent of the elided constituent in (ib), (iib) consists of a predicative NP: (iv) ... [NP [AP [QP -er] tall] a man][CP than[IP my father/mother[I’ is [NP [AP [QP x] tall] a man]]]]] In this case, we compare how tall a man the subject of the than clause is. As a result, we get a deviant reading with my mother as the subject of the than clause. . From the fact that dissociated morphemes are inserted after Spell-out, it follows that they are not interpreted at LF. . Above we have already noted that C-related agreement marking cannot be realized in addition to canonical verbal agreement in contexts where the verb has undergone movement to C. In other words, Agr-on-C cannot be realized in addition to Agr-on-T. Why is this form of ‘double agreement’ (i.e., *V+AgrT+AgrC) ruled out? Note that this question relates to all Germanic varieties with complementizer agreement. Following Carstens (2003), I assume that the impossibility of doubly inflected finite verbs results from a morphological condition ensuring that only the hierarchically highest Agr-morpheme is spelled out in a given head complex. Based on proposals of Kinyalolo (1991), Carstens (2003: 407) phrases the relevant condition as follows (where ‘inert’ means that the relevant Agr-morpheme is not pronounced): (i)

Morphological Economy In an adjoined structure, Agr on a lower head is inert iff its features are predictable from Agr on a higher head.

If it is assumed that complex heads are processed in a bottom-up fashion during Vocabulary Insertion (see Section 3.6.3 above), (i) can be implemented by a late MS-process which deletes the lower Agr-morpheme in a head complex. . The idea that complementizer agreement is parasitic on verbal agreement is further supported by the observation that across Germanic, there appear to exist no languages with complementizer agreement but without verbal agreement, while there are many languages that exhibit verbal agreement in the absence of complementizer agreement (Hoekstra & Smits 1999). Thus, it seems that cross-linguistically, the availability of complementizer agreement is dependent on the overt realization of verbal agreement morphology. . At first sight, it seems that this analysis is contradicted by those varieties (like Hellendoorn, or 1pl in Lower Bavarian) where the shape of Agr-on-C differs from that of Agr-on-T. However,

JB[v.20020404] Prn:2/08/2005; 9:49

F: LA8103.tex / p.73 (4355-4414)

Chapter 3. The structural design of agreement 

this phonological difference can be explained as the result of conditioned allomorphy which is sensitive to the syntactic environment (i.e., the adjunction site of the Agr-morpheme, C vs. T) or – as pointed out to me by Ian Roberts – the way the relevant agreement morpheme is valued (via Agree/head movement or post-syntactic insertion as a dissociated morpheme). This form of allomorphy is presumably introduced by the reanalysis of a subject clitic as a dissociated Agrmorpheme. As a result, the phonological exponent of the latter will in most cases differ from the existing verbal morphology due to its lexical source (the subject clitic), see Chapters 5 and 6 for details. . A similar theoretical notion of structural adjacency is proposed in Zeller (2001: 36). However, in contrast to the analysis proposed here, Zeller assumes that structural adjacency constitutes a local domain between two heads that is only syntactically relevant. . If clitics (and, possibly, proper names) are analyzed as D-heads, the restriction to “projecting syntactic heads” is necessary to warrant that a clitic in SpecTP (in the configuration [CP C [TP D [T’ T ]]]) does not interrupt the structural adjacency between C and T (Ian Roberts, personal communication). . Recall that in some of these dialects, the realization of complementizer agreement is confined to inversion contexts. This is illustrated by the following East Netherlandic examples ((65), repeated here as (i)), in which the verb carries the exponent of Agr-on-C (/-6/) only in inversion contexts (i.e., (ib)), while subject-initial clauses exhibit the regular agreement ending /-t/, cf. (ia): (i)

a.

b.

Wij speul-t/*-e. we play-1pl ‘We play.’ Waar speul-e/*-t wij? where play-1pl we ‘Where do we play?” (Zwart 1993b: 254)

To account for this difference, I tentatively follow Travis (1984) and Zwart (1997a) and assume that the dialects in question exhibit a structural difference between inversion contexts and subject-initial clauses, in the sense that only the former involve movement to the C-system. In contrast, subject-initial clauses result from a configuration where subject and verb occupy specifier and head of TP (sometimes called ‘asymmetric V2’), giving rise to regular agreement morphology on the verb. Furthermore, note that not all dialects with different exponents for Agr-on-C and Agr-on-T exhibit this contrast in matrix clauses. In Lower Bavarian, for example, the exponent of Agr-on-C (1pl /-ma/) replaces the regular agreement ending in subject-initial clauses as well (Weiß 2002: 9). This can then be taken to indicate that both subject-initial clauses and inversion structures involve movement to the C domain: (ii) a.

b.

Mia laff-ma/*laff-a hoam. we ran-1pl/ran-1pl home ‘We are running home.’ Mia gem-ma/*geng-an hoam. we go-1pl/go-1pl home ‘We are going home.’

. Cf. the following definition of c-command given in Chomsky (1995: 339). X and Y are taken to be “disconnected” if no segment of X contains Y:

JB[v.20020404] Prn:2/08/2005; 9:49

F: LA8103.tex / p.74 (4414-4448)

 The Rise of Agreement

(i)

C-command X c-commands Y if (a) every Z that dominates X dominates Y and (b) X and Y are disconnected.

. Similarly, object clitics may intervene between the subject and the inflected complementizer in West Flemish, which otherwise requires strict adjacency between C0 and the subject (Liliane Haegeman, personal communication): (i)

da-n ze Valère en Marie nie gezien een that-3pl her Valère and Marie niet seen have-3pl ‘That Valerie and Marie have not seen her’

This parallel between West Flemish and Bavarian suggests that the strict adjacency effect exhibited by West Flemish (see Note 53) might possibly be attributed to the existence of a full paradigm of complementizer agreement. As a result, C and T must be structurally adjacent in all contexts. . At first sight, the shape of many modal particles seems to suggest an analysis as X0 elements. However, the fact these elements do not block V-to-C movement in the Germanic V2 languages shows that they cannot be analyzed as syntactic heads. In addition, similar locality problems arise for XP-movement into the C-domain if they are analyzed as specifiers (cf. van Gelderen 2001 for discussion and further diachronic arguments against an analysis as specifiers). These facts can be taken to further support the idea that modal particles are base-generated in adjoined positions.

JB[v.20020404] Prn:3/08/2005; 10:20

F: LA8104.tex / p.1 (45-130)

Chapter 4

The transition from pronoun to inflectional marker

. Introduction As already mentioned, it is commonly assumed that pronouns are the primary historical source of subject-verb agreement morphology. However, the discussion in Chapter 1 has shown that our knowledge of syntactic aspects of this historical process is still rather limited. This chapter sets out to develop a theory of the syntactic environments where new verbal agreement markers can evolve via a reanalysis of subject clitics. In contrast to beliefs widespread in the literature, it is argued that this structural change is not confined to a unique syntactic environment. Rather, there is quite a variety of syntactic contexts that may feed the grammaticalization process in question. It is shown that the set of historical scenarios which may license a reanalysis of subject pronouns as agreement markers is defined by a small number of (necessary) syntactic conditions that the change in question must satisfy. These conditions derive from the theory of agreement developed in the previous chapter. Of course, a formal analysis of the transition from clitic pronouns to agreement markers must incorporate some diagnostics for telling apart clitics and agreement markers. Accordingly, the present chapter discusses a set of syntactic and morphological diagnostics to decide whether a given element has already developed into some form of agreement or should rather be analyzed as a clitic pronoun. Importantly, it is argued that these criteria not only provide a tool for linguistic analysis, but also can be conceived as cues for the learner during language acquisition. The chapter is organized as follows. Section 4.2 introduces a set of criteria for differentiating pronominal clitics and agreement markers. Section 4.2.1 is concerned with syntactic criteria, while Section 4.2.2 deals with morphological criteria (basically following Zwicky & Pullum 1983). In Section 4.3, I discuss a set of syntactic preconditions on the categorial reanalysis of pronominal clitics. Based on these conditions, I then introduce a set of historical scenarios where subject clitics can be reanalyzed as markers of verbal agreement. The presentation of these scenarios is organized along the common distinction between Infl-oriented (Section 4.4) and C-oriented clitics (Section 4.5). Section 4.6 summarizes the conclusions reached in this chapter.

JB[v.20020404] Prn:3/08/2005; 10:20

F: LA8104.tex / p.2 (130-182)

 The Rise of Agreement

. Telling apart clitics and agreement markers As noted in Chapter 1, the historical development of (bound) agreement markers necessarily proceeds via a stage where clitics are reanalyzed as inflectional affixes. For this reanalysis to take place, the properties of clitics and agreement markers must be sufficiently similar so that the learner can misinterpret the former as the latter. However, due to these similarities it is an equally difficult task for the linguist to decide whether a given clitic has already developed into an agreement marker, which is a necessary precondition for an adequate analysis of the diachronic development in question. Hence, both the learner and the linguist must look for cues that decide on the categorial status of the ‘clitic’ in the data at issue. This section sets out to develop a set of relevant diagnostics to assess the categorial status of clitics/agreement markers. In Section 4.2.1, I discuss a set of syntactic diagnostics while Section 4.2.2 focuses on morphological criteria that separate clitics from agreement markers.

.. Syntactic criteria Perhaps the most common syntactic criterion for distinguishing between clitics (and incorporated pronouns) and agreement markers is complementary distribution with a DP argument (cf. e.g. Corbett 2003 for some discussion): as a rule of thumb, genuine agreement markers may co-occur with a DP argument, while clitics and incorporated pronouns are arguments by themselves and may therefore not co-occur with an argument that receives the same θ-role. The way this diagnostic works can be illustrated with the following examples from Italian and the Carib language Macushi (examples taken from Corbett 2003: 186). In both Italian and Macushi, it is not necessary that the arguments of the verb (in Italian, this holds only for subjects) are realized overtly. The inflected verb alone is sufficient to create a well-formed sentence:1 Italian (1) Mangi-a. eat-3sg ‘He/she eats.’ Macushi (2) i-koneka-‘pî-i-ya. 3sg-make-past-3sg-erg ‘He made it.’ In Italian, the φ-features of the subject are expressed by the morpheme -a. In Macushi, the φ-features of the absolutive argument are expressed by the prefix i-, while the suffix -i marks the feature content of the ergative argument. Thus, at first sight, the person/number markers that appear on the finite verb in Italian and Macushi seem to behave similarly. However, the languages behave differently when overt arguments are

JB[v.20020404] Prn:3/08/2005; 10:20

F: LA8104.tex / p.3 (182-239)

Chapter 4. The transition from pronoun to inflectional marker

added. In Italian, the subject can be optionally realized as an overt DP (e.g., an overt subject pronoun), as in (3). Italian (3) Lui/Lei mangi-a. He/she eat-3sg ‘He/she eats.’ This is commonly taken to indicate that in Italian, the person/number marker is a true agreement suffix, which redundantly marks the person/number features of the subject, which may be pro (as e.g. in (1)). Now consider the following examples from Macushi. Macushi (4) t-ekîn era’ma-’pî paaka esa-’ya. refl-pet.abs see-past cow owner-erg ‘The owner of the cow saw his own pet.’ Macushi (5) *uurî-ya i-koneka-‘pî-i-ya I-erg 3sg-make-past-3sg-erg ‘He made it.’ (4) shows that the person/number markers disappear in the presence of overt DP arguments. Moreover, (5) shows that the co-occurrence of an overt DP argument and the corresponding person/number marker leads to ungrammaticality. The complementary distribution of person/number markers and overt DP arguments shows quite conclusively that the person/number markers in (2) are not instances of agreement, but rather clitic pronouns that have argument status and receive a θ-role from the verb. In this way, the distribution of person/number markers and overt DP arguments provides a first rationale for deciding whether a given element is a clitic or an agreement marker. Unfortunately, the picture is often less clear cut than in the Italian/Macushi case. On the one hand, there are languages such as Welsh, in which the presence of full-fledged person/number markers is compatible only with pronominal subjects (cf. Chapter 1 above). On the other hand, there are many languages where clitics generally may co-occur with another DP argument, giving rise to instances of clitic doubling, as in the following examples from French (subject clitic doubling), Spanish (object clitic doubling), and Italian (clitic left dislocation) (the clitic and the double are marked by underlining): French (6) Pierre il mange une pomme. Pierre he eats an apple ‘As for Pierre, he eats an apple.’



JB[v.20020404] Prn:3/08/2005; 10:20



F: LA8104.tex / p.4 (239-319)

The Rise of Agreement

Spanish (7) La llamaron a ella. her call-3pl to her ‘They called her.’ (Suñer 1988: 394) Italian molto. (8) Credo che il tuo libro, loro lo apprezzerebbero think-1sg that your book they it appreciate-subjnc much ‘I think that they would appreciate your book very much.’ (Rizzi 1997: 288) In general, instances of clitic doubling resemble the agreement relation in that there is a free phrasal element and a bound element which share φ-features. This abets a reanalysis in which the clitic is misinterpreted as a mere agreement marker, with the double turning into the ‘true’ argument.2 Accordingly, instances of doubling are particularly interesting in the context of this study. However, due to the similarities between doubling and agreement configurations, it is not an easy task to decide whether the clitic has already developed into an agreement marker. Hence, we must look for further criteria that can be used to decide on the categorial status of the ‘clitic’ in clitic doubling structures. In principle, these criteria may relate to (i) properties of the double, (ii) properties of the clitic, or (iii) the distribution of the doubling construction as a whole. Let me first consider some factors which relate to the distribution of the doubling construction as a whole. Recall that it is a characteristic of agreement that it is always obligatory, that is, the verb always agrees with the subject, independent of the context. In contrast, the doubling construction is often linked to certain stylistic purposes and therefore confined to certain contexts. For example, the speaker may choose to add a full DP for reasons of emphasis, to reinforce the prosodically weak clitic, which cannot bear stress, similar to examples where pro is replaced by an overt pronoun in languages such as Italian. As a consequence, the doubling construction is characteristically restricted to certain contexts where the speaker wishes to emphasize the argument expressed by the clitic. An example of this usage of the doubling construction is Standard French, where the clitic alone suffices to express the subject. Crucially, a full DP/pronoun can optionally be added to lay stress on the subject, cf. (9) (Moi) je porte la table. me clit.1sg carry the table ‘As for me, I carry the table.’ Therefore, doubling is an optional mechanism that is restricted to contexts where it serves certain communicative functions in French. A similar pattern can be observed in Standard Italian, where clitic doubling is obligatory only to resume a topic which has been moved into the left clausal periphery, as in (8) above. These contextual restrictions clearly show that the clitic element is a ‘true’ pronominal clitic that has not yet developed into an agreement marker. However, if doubling loses its stylistic force (presumably due to an over-use), it may gain a wider distribution and eventually become obligatory in all contexts. This can then be taken to indicate that the doubling configu-

JB[v.20020404] Prn:3/08/2005; 10:20

F: LA8104.tex / p.5 (319-375)

Chapter 4. The transition from pronoun to inflectional marker

ration has undergone a grammaticalization process and that the ‘clitic’ should actually be analyzed as an agreement marker. Well-known examples of this diachronic development are Non-Standard French, where subject clitic doubling is obligatory for 1sg, 2sg and 1pl forms (Lambrecht 1981; Auger 1993, 1994; Gerlach 2002), and Spanish, which exhibits obligatory object clitic doubling (cf. e.g. Suñer 1988). Accordingly, I will use the distribution of clitic doubling as a diagnostic for the categorial status of the clitic: optional doubling signals that the clitic is still a pronominal element, while obligatory doubling indicates that the clitic probably has been reanalyzed as an agreement marker (see Chapter 5 on a set of Swiss Rhaeto-Romance dialects).3 Another set of potential diagnostics has to do with properties of the double. Uriagereka (1995) notes that clitics usually receive an interpretation as definite/specific. Likewise, in clitic doubling constructions, the full nominal usually must be definite/specific. This observation can be turned into a diagnostic for agreement in cases where we have to decide whether a certain configuration is an instance of clitic doubling or predicate-argument agreement: if it is possible for the full nominal to be indefinite/non-specific, then it is rather clear that the relevant construction represents an agreement relation (cf. Brandi & Cordin 1989; Suñer 1992; Haiman 1991; Haiman & Benincà 1992 for similar considerations). For example, indefinite NPs such as ‘nobody’ cannot occur as topics in clitic left dislocation structures in Standard Italian, cf. (10) *Nessuno, lo conosco in questa cittá. nobody him know-1pl in this city ‘*Nobody, I know him in this city.’ (Brandi & Cordin 1989: 118) However, in a set of Northern Italian dialects this restriction does not hold any longer. The following examples show that indefinites such as ‘no one’ can co-occur with the clitic subject pronoun. This is commonly taken to suggest that the clitic is actually an agreement marker (Brandi & Cordin 1989: 118): (11) a.

Nisun l’ha dit niente. nobody he-has said nothing ‘Nobody said anything.’ b. Nessun gl’ha detto nulla. nobody he-has said nothing ‘Nobody said anything.’

(Trentino)

(Fiorentino)

Finally, we may look for properties of the clitic itself that give us a clue concerning its categorial status. A set of relevant syntactic criteria is discussed in Rizzi (1986), Brandi and Cordin (1989), Haiman (1991), Haiman and Benincà (1992), Suñer (1992), and Poletto (2000). First, it is sometimes suggested that anti-agreement or partial agreement are properties which are found solely with real agreement markers (e.g., in a number of Northern Italian dialects, Irish and Standard Arabic), but never with pronominal elements, which always show complete φ-feature identity with their antecedent (cf. Haiman & Benincà 1992). Under this assumption, anti-agreement be-



JB[v.20020404] Prn:3/08/2005; 10:20

F: LA8104.tex / p.6 (375-451)

 The Rise of Agreement

tween the clitic and its double can be taken to indicate that the clitic should be analyzed as an instance of agreement marking, as in the following examples from a variety of Northern Italian/Rhaeto-Romance dialects (taken from Haiman & Benincà 1992: 193): (Fassa) L e venju la vivano. he is come-masc.sg the witch.fem ‘There came the witch.’ (Moena) b. Chi e lo po i ozitegn? who is he then the Occitans-masc.pl ‘Who are the Occitans?’ (Badiot) c. Da doman vegn l oshore les vatges. of morning becomes he fed-masc.sg the cows-fem.pl ‘The cows are fed in the morning.’ (Fiorentino) d. Gl e venuto delle ragazze. he is come-masc.sg some girls-fem.pl ‘There came some girls.’ (Zeneyze) e. U vene a Katajning. he comes the Catherine ‘Catherine is coming.’ (Furlan) f. Al era una volta una fameja. he was one time a family-fem.sg ‘There was once a family.’

(12) a.

Another criterion relates to contexts with a subject gap, that is, subject relatives and conjoined sentences, where a coreferential subject is often deleted in the second conjunct. If the subject clitic is preserved in these contexts, so the argument goes, then it should rather be treated as an agreement marker, since agreement markers are obligatorily present in relative clauses and conjunction contexts. Again, the diagnostic can be illustrated with data from Northern Italian dialects (examples taken from Haiman 1991: 141):4 Relative clauses (13) a. dut ce che al e gno all that which he is mine ‘All that which is mine’ b. doj peres pitshuj ke i me prea l arberk two poor little-boys who they me ask the shelter ‘Two poor little boys who ask me for shelter’ Conjoined clauses (14) a. Al vent dut e al va lontan. he sells everything and he goes far away.

(Friulian)

(Fassa)

(Friulian)

JB[v.20020404] Prn:3/08/2005; 10:20

F: LA8104.tex / p.7 (451-508)

Chapter 4. The transition from pronoun to inflectional marker

b. Quant ch ‘al vigni a se e al cognosse when that-he came to himself and he recognized l predi. the priest (Haiman & Benincà 1992: 194)

(Friulian)

In other words, if we come upon a subject clitic in circumstances where we would normally expect a subject gap, as in (13) and (14), it seems likely that the clitic has actually turned into an agreement marker. The syntactic diagnostics discussed so far are summarized in (15).5 (15) Indications for an analysis of clitics as agreement markers a. The clitic is obligatory (doubling is not contextually restricted). b. The double in clitic doubling structures may be indefinite/non-specific. c. Anti-agreement between the clitic and the double. d. The clitic shows up in subject gap environments. In the next section, I will discuss a couple of morphological properties that may be used as further diagnostics for telling apart clitics and agreement markers.

.. Morphological criteria Zwicky and Pullum (1983) develop a set of morphosyntactic criteria which help to make a principled distinction between clitics and inflectional affixes. The criteria derive from the insight that “word-clitic combinability is largely governed by syntactic considerations. The conditions governing the combinability of stems with affixes are of quite a different sort: they are morphological and/or lexical in character, being concerned with the substructure of a finite set of words.” (p. 503). In this section, I will review some of their criteria and show how they can be applied to separate clitics from agreement markers. The first criterion proposed by Zwicky and Pullum relates to the well-known observation that affixes usually impose stricter selectional requirements with respect to their lexical host than clitics: (16) Clitics can exhibit a low degree of selection with respect to their hosts, while affixes exhibit a high degree of selection with respect to their stems. Zwicky and Pullum illustrate this criterion with differences between clitics and affixes in English. They observe that clitics such as ‘s ‘is’, ‘s ‘has’, and ‘ve ‘have’ “can attach to words of virtually any category” (p. 504): (17) a.

preposition: The person I was talking to’s going to be angry with me. b. verb: The ball you hit’s just broken my dining room window.



JB[v.20020404] Prn:3/08/2005; 10:20

F: LA8104.tex / p.8 (508-573)

 The Rise of Agreement

c.

adjective: Any answer not entirely right’s going to be marked as an error. d. adverb: The drive home tonight’s been really easy. In contrast, inflectional affixes in general display quite specific selectional properties (as to which element they attach to): in English, plural /-z/ attaches only to nouns, past /-d/ only to verbs, and the superlative /-est/ only to adjectives or adverbs. If applied to the grammaticalization of agreement markers from clitic pronouns, this criterion should lead us to expect that the change in question goes hand in hand with tighter restrictions on the class of elements the clitic/agreement marker can attach to. This development can be observed in a number of Uto-Aztecan languages (cf. Steele 1977). Many Uto-Aztecan languages exhibit a set of enclitic pronouns that move to second position and attach to the first word or constituent of the clause independent of its syntactic category: (18) hunwuti=pum se‘iwun. bear:object=clit.3pl are:shooting ‘They are shooting a bear.’ (Luiseño; Steele 1977: 539) (19) semati=ne napaha rarimea aré. nice=clit.1sg shirt gonna:buy probably ‘I am probably going to buy a nice shirt.’ (Tarahumara; Steele 1977: 554) (20) tuká=ne antónyta biˇcak. yesterday=clit.1sg Antonio saw ‘Yesterday, I saw Antonio.’ (Yaqui; Steele 1977: 554) Interestingly, in some languages such as Classical Aztec, the clitics uniformly attach to the left of the finite verb: (21) an-teeˇcLaso‘La. 2pl-love-us ‘You love us.’ (Classical Aztec; Steele 1977: 539) Accordingly, the criterion in (16) suggests that former clitics have developed into agreement prefixes on the verb in Classical Aztec (see Section 4.5 below for an outline of the relevant syntactic scenario and Chapter 5 for detailed discussion). The following criterion makes use of the observation that the morphophonological results of attaching a clitic to a certain lexical host are usually highly regular, that is, there are no cases where the host-clitic combination yields an unexpected form. In contrast, irregular forms, suppletion etc. are hallmarks of inflectional morphology. (22) Morphophonological idiosyncrasies are more characteristic of affixed words than of clitic groups. In connection with (22), Zwicky and Pullum (p. 505) note that this principle should also be taken to capture the fact that the attachment of clitics typically leaves the lexical

JB[v.20020404] Prn:3/08/2005; 10:20

F: LA8104.tex / p.9 (573-634)

Chapter 4. The transition from pronoun to inflectional marker 

host unaffected. Thus, if we can observe that an alleged clitic has an effect on the shape or morphological structure of its host, this should cast some doubt on its status as a clitic. Consider the following examples from the Northern Italian dialect Fiorentino (Brandi & Cordin 1989): (23) a.

Icché gl’hanno fatto? what clit.3pl=have-3pl done b. Icché ha(e)-gli-no fatto? what have-(clit.)3pl-3pl done (Brandi & Cordin 1989: 132)

In inversion contexts, the subject clitic may optionally intervene between the verb stem and the original agreement ending, as in (23b). Under the above assumptions, this possibility suggests that the 3pl form gli is no longer a clitic, but has developed into a marker of verbal agreement. In a similar vein, I will treat cases where an apparent clitic replaces the usual verbal agreement morphology as an indication that the clitic has in fact developed into an inflectional affix (see Chapter 5 for further examples from German and Swiss Rhaeto-Romance varieties). Another criterion concerns gaps in the paradigm, which are much more common with inflectional affixes than with pronominal elements (cf. e.g. Rizzi 1986): (24) Arbitrary gaps in the set of combinations are more characteristic of affixed words than of clitic groups. For example, Roberts and Roussou (2003: 180) argue that leveling of person/number distinctions in the pronoun paradigm of a set of Northern Italian dialects (Veneto) which show identical forms for 1sg, 1pl and 2pl promoted the reanalysis of these pronominal clitics as realizations of AgrS. Finally, Zwicky and Pullum propose two further criteria that follow from the assumption that cliticization applies after syntax (i.e., “no syntactic operations apply after syntax”, p. 504), which is also adopted in the present work (see Chapter 2 above): (25) Syntactic rules can affect affixed words, but cannot affect clitic groups. (26) Clitics can attach to material already containing clitics, but affixes cannot. Applied to cases like (23b) from Fiorentino, principle (26) further supports the above conclusion that the clitics in this dialects have evolved into inflectional affixes, since we would not expect an inflectional affix to be able to attach to the right of a verb-clitic combination. A good example of the application of (25) comes from differences with respect to the behavior of clitics in Standard and Non-Standard French.6 It is a well known fact that the subject clitics of Standard French are sensitive to syntactic context, that is, they precede the verb in declaratives, but must follow it in matrix questions with wh-ex-situ (Kayne 1975):

JB[v.20020404] Prn:3/08/2005; 10:20

F: LA8104.tex / p.10 (634-696)

 The Rise of Agreement

(27) a.

Tu attends qui? you await who b. Qui attends-tu? who await-you ‘Who are you waiting for?’

The kind of subject-verb inversion illustrated in (27b) is commonly analyzed as an instance of I-to-C movement of the finite verb which crosses the position of the subject clitic (Kayne 1984; Rizzi & Roberts 1989). That is, the subject clitics of Standard French are not affected by syntactic rules that affect their host, in the sense that they do not accompany verb movement to C. In contrast, the subject clitics of Non-Standard French fail to undergo subject-verb inversion in matrix questions, as can be seen from the example (28), taken from Lambrecht (1981: 6): (28) Où tu vas? where you go ‘Where are you going?’ In other words, the preverbal clitics of colloquial French must accompany the verb if the latter is affected by movement operations. This can be taken to suggest that they are already a fixed part of verbal morphology, namely prefixal agreement markers (for further arguments against an analysis as pronominal elements cf. Lambrecht 1981; Haiman 1991; Auger 1993, 1994; Friedemann 1995; Gerlach 2002). Finally, and related to the latter considerations, we can construe a last criterion which has to do with the observation that inflectional elements tend to occur in a fixed position, while clitics can be reordered with respect to other clitics: (29) Clitics can be reordered within the clitic complex, while affixes occupy fixed positions. Concerning the historical development of agreement markers, we can observe that free pronouns can occupy a variety of different positions (e.g., IP-internal, clauseinitial etc.), while clitics are usually confined to certain structural positions (e.g., the so-called Wackernagel position) or certain hosts (most often the finite verb), but still may switch positions inside the clitic complex. In contrast, agreement markers typically show no variation at all, occupying fixed positions/slots inside the inflectional domain of the verb. In this section, I have reviewed a number of diagnostics which may help to decide whether a given element should be analyzed as a clitic or rather an inflectional affix. The syntactic criteria developed in Section 4.2.1 pay special attention to the properties of clitic doubling configurations such as obligatory character, specificity of the double etc. In contrast, the morphological criteria outlined in 4.2.2 focus on properties of the clitic alone (basically following Zwicky & Pullum 1983), for example selectional attributes, morphophonological effects on the host and distribution with respect to other clitics and affixes. In the remainder of this book, the criteria developed in the present section will be used as diagnostics for the categorial status of pronominal clitics

JB[v.20020404] Prn:3/08/2005; 10:20

F: LA8104.tex / p.11 (696-758)

Chapter 4. The transition from pronoun to inflectional marker 

that are subject to a grammaticalization process which eventually gives rise to new verbal agreement markers. It is important to note that the above criteria not only constitute a tool for linguistic analysis. In addition, they may also be taken to reflect cues that signal the learner the categorial status of a clitic element during language acquisition (in the sense of Lightfoot 1999, see Chapter 2 above). For example, if the learner can identify clear contextual restrictions on the use of clitic doubling, he/she will presumably conclude that the element in question is actually a clitic and not an agreement marker. Conversely, obligatory presence of a clitic can be seen as a syntactic cue for agreement status, while effects on the morphophonological shape of the host act as morphological cues that equally signal agreement status. I will come back to this issue in connection with concrete examples in Chapters 5 and 6.

. Syntactic preconditions for the rise of agreement It is fairly clear that there is a number of general, theory-independent preconditions on the historical development of agreement markers from pronominal elements, including for example the availability of a series of non-stressed clitic pronouns (see Chapter 2 above), or a number of the properties discussed in the previous section, which might be associated with the notion of cues, in the sense of Lightfoot (1999). In addition, each restrictive theory of agreement imposes a set of further restrictions on the reanalysis in question, the exact nature of which depends on the set of assumptions that constitute the theory. In this section, I will discuss the implications of the theoretical approach to agreement outlined in Chapter 3 for the diachronic development of agreement markers from pronouns. It is shown that the theory of agreement proposed in Chapter 3 imposes certain restrictions on the categorial reanalysis of clitic pronouns, making a set of predictions with respect to the syntactic configurations where the change in question may take place. In addition, I discuss a set of further syntactic conditions imposed by θ-theory and derivational/representational economy, which shape the course of the reanalysis of clitic pronouns. In Chapter 3, I have argued for a theory of agreement that is based on the following assumptions: (i)

Agreement morphemes do not head their own projection in the syntax. Rather, they are parasitic on other functional heads. (ii) Agreement morphemes may combine with contentful functional categories such as C, T, or ν (and maybe others) in one of the following ways. First, the agreement morpheme may attach to its functional host prior to the insertion of that host into the syntactic derivation. Second, the agreement morpheme may be added postsyntactically as a dissociated Agr-morpheme, as in the case of complementizer agreement in Germanic. (iii) Canonical Agr-morphemes are valued by an appropriate controller under closest c-command (Agree) during the syntactic derivation. Dissociated Agr-morphemes

JB[v.20020404] Prn:3/08/2005; 10:20

F: LA8104.tex / p.12 (758-823)

 The Rise of Agreement

are valued under structural adjacency with an Agr-morpheme that has been licensed in the syntax. (iv) Bound Agr-morphemes must combine with their host prior to Vocabulary Insertion, either by syntactic verb movement or Morphological Merger under adjacency at MS. The assumptions (i)–(ii) actually do not impose any strong restrictions on the reanalysis of clitics as agreement heads. Compared with previous analyses in which agreement is assumed to occupy a unique, well-defined structural position in the structure of the clause, it appears that the present analysis is actually more liberal, since in principle agreement morphemes can come into existence as part of any contentful functional category if the learner can detect the presence of an Agr-morpheme on a certain functional head on the basis of the evidence available to him/her. This assumption is further exploited in Chapter 5, where it is argued that in V2 languages, new verbal agreement markers may develop via an initial stage where the learner posits an agreement morpheme attached to C. Let us now turn to the impact of the theoretical assumptions (iii)–(iv), which require that the feature content of Agr be valued and that Agr-morphemes combine with the verb prior to Vocabulary Insertion, respectively. The latter consideration correlates with the perhaps most obvious syntactic restriction on the reanalysis of pronouns as verbal agreement affixes: a clitic pronoun can only be reinterpreted as an inflectional affix if it is adjacent to the verb.7 (30) Adjacency requirement A clitic pronoun can be reanalyzed as a bound agreement affix on the verb only if the clitic is string-adjacent to the verb. Under the assumption that agreement morphemes are hosted by other functional heads, the intuition captured by (30) must be reformulated in more formal terms by saying that a pronominal clitic can only be reanalyzed as an agreement morpheme on a functional head X if X combines with the verb prior to Vocabulary Insertion. This can be accomplished either by verb movement to X in the overt syntax or by post-syntactic operations such as Morphological Merger, which require that X and the verb be adjacent at MS. Since this requirement concerns the process of word formation, that is, the creation of complex morphosyntactic units either in the syntactic component or at MS, I choose to call it the Word building constraint: (31) Word building constraint The reanalysis of a clitic adjacent to the verb as a (bound) verbal agreement marker on a functional head X requires that X combines with the verb prior to Vocabulary Insertion. In other words, one might say that the presence of the finite verb in a functional head X (as a result of overt movement) signals the learner that X is capable of hosting verbal agreement morphemes/features.8

JB[v.20020404] Prn:3/08/2005; 10:20

F: LA8104.tex / p.13 (823-894)

Chapter 4. The transition from pronoun to inflectional marker

In addition, the feature content of the newly created agreement morpheme must be licensed (or, ‘valued’, using the terminology of Chomsky 2000, 2001a) in a local structural configuration. If UG allows for two kinds of Agr-morphemes (i.e., syntactic and dissociated Agr-morphemes), which are subject to different licensing mechanisms (see Chapter 3), then the reanalysis of pronominal elements is restricted by the following condition: (32) Identification of feature content The reanalysis of a clitic pronoun is licit only if the resulting agreement morpheme is licensed (i) in the syntax by a local Agree relation with a matching set of interpretable φ-features, or (ii) at MS as a dissociated morpheme under structural adjacency with a syntactically licensed Agr-morpheme. Apart from the syntactic conditions that follow from the approach to agreement adopted in this work, the reanalysis of pronouns must arguably fulfill at least two further requirements imposed by θ-theory and general economy principles which shape the course of language change (see Chapter 2 above). A rather obvious condition on the categorial reanalysis of pronouns is imposed by θ-theory. Since a pronoun usually carries a thematic role, it must be ensured that the role in question can still be assigned when the pronoun is reanalyzed as an agreement marker and disappears from the set of arguments in a given clause: (33) Preservation of argument structure The reanalysis of a pronoun as an agreement marker must preserve the predicate’s argument structure. In other words, a reanalysis of a subject pronoun as an agreement marker can only take place if the relevant thematic role can be assigned to another element, either another overt DP (e.g., a former topic in a clitic left dislocation structure, cf. Givón 1976) or a covert pronominal element, that is, pro. This leads to some interesting issues concerning the interaction between the development of verbal agreement and the pro-drop property. More specifically, it seems that pro-drop may be either a precondition for or the result of the development of (new) verbal agreement marking. In other words, a pro-drop grammar facilitates the reanalysis of pronouns as agreement markers, since the learner has at his/her disposal an empty pronoun which may be inserted to receive the θ-role assigned to the pronoun in the target grammar. Conversely, the reanalysis of pronouns may introduce new pro-drop properties into a grammar that previously lacked pro-drop (for a related statement, cf. Roberts & Roussou 2003: 185f.). In generative approaches to language change it is often assumed that a reanalysis must conform to general economy principles, in the sense that the learner assigns a given input string the most simple or economic structure/derivation that is compatible with the evidence (cf. e.g. Clark & Roberts 1993). The preference for simplicity



JB[v.20020404] Prn:3/08/2005; 10:20

F: LA8104.tex / p.14 (894-945)

 The Rise of Agreement

may trigger a reanalysis of a given structure if the trigger for a more complex target grammar has been obscured by phonological erosion or independent change processes (see Chapter 2 above). In other words, if the evidence available to the learner is ambiguous, he/she will opt for the most economic structural option. Recently, Roberts and Roussou (2003) have argued that this kind of structural simplification is a key feature of grammaticalization, showing that grammaticalization processes generally lead to the loss of syntactic operations (either overt movement or an Agree relation between a higher and a lower head). The loss of movement operations fulfills the requirement of structural simplification rather straightforwardly: if a lexical element previously moved to a functional head is reanalyzed as being merged in this functional head position, the change in question instantiates a case where a rather ‘costly’ Move operation is replaced by the more basic operation Merge. This is exemplified by the following change which according to Poletto (1995, 2000) has taken place in a number of Northern Italian dialects such as Venetian. In (34), a pronominal element previously moved to SpecAgrSP is reanalyzed as an instantiation of AgrS, leading to the loss of DP-movement (Roberts & Roussou 2003: 178):9 (34) [AgrSP DPi [AgrS’ V ] [TP . . . [VP ti ... →[AgrSP [AgrS’ D+V ] ... However, the structural simplification associated with the loss of an Agree relation is less straightforward (e.g., it cannot be captured by counting nodes or syntactic objects present in the derivation). Roberts and Roussou (2003: 201) propose that all instances of structural simplification can be captured by the following simplicity metric (cf. Longobardi 2001: 294f. for a related proposal): (35) A structural representation R for a substring of input text S is simpler than an alternative representation R’ iff R contains fewer formal featuresyncretisms than R’. Feature syncretism is defined as “the presence of more than one formal feature in a given structural position” (Roberts & Roussou 2003: 201). For example, if a given functional head H triggers movement of an element X, this always implies the presence of an additional formal (EPP-) feature F in H that attracts X (in addition to the set of features that licenses Merge of H). If X is reanalyzed as being merged in the position of H, this leads to the disappearance of F and therefore to a structure with less feature syncretisms.10 This approach works in a parallel fashion if the reanalysis leads to the disappearance of an Agree-relation, implying the loss of the formal feature that originally triggered the Agree operation. In other words, the most economical structure is a structure “with the least occurrences of multiple features on single positions” (Roberts & Roussou 2003: 201), where ideally each syntactic head is spelled out by an element that has been merged as this head. Note that the rationale behind (35) is therefore quite similar to the traditional (structuralist) idea that ideally, there should be an one-to-one relation between underlying grammatical features and overt grammatical morphemes (the so-called agglutinative ideal, cf. Spencer 1991 for critical discussion; cf. Julien 2002 for a related proposal).

JB[v.20020404] Prn:3/08/2005; 10:20

F: LA8104.tex / p.15 (945-1007)

Chapter 4. The transition from pronoun to inflectional marker 

Concerning the development of agreement markers from pronominal elements, we should thus expect that the relevant reanalysis is only possible if the resulting structure involves less movement or Agree operations than the target structure. In general this seems to be true, for example in cases where a subject pronoun previously moved to SpecTP is reanalyzed as an agreement morpheme on T. Below, however, we will see that the reanalysis of pronouns often leads to structures that are morphologically more complex than the target structure, as in cases where a pronoun is reanalyzed as a dissociated Agr-morpheme that is inserted by a morphological operation at MS. The assumption that the reanalysis from pronouns to agreement markers is shaped by the syntactic conditions discussed so far predicts that the change under consideration can only take place in a certain set of syntactic environments where these restrictions are satisfied. In the following, I give an outline of (a selection of) the relevant syntactic configurations, focusing on the development of subject-verb agreement. The exposition is arranged according to the well-known distinction between Infl-oriented clitics (Section 4.4) and C-oriented clitics (4.5).11

. Paths toward agreement I: Infl-oriented clitics Let me begin with the configuration discussed by Givón (1976), that is, topic left dislocation structures where the topic becomes the new subject while the resumptive pronoun is reanalyzed as a subject agreement marker. Following Uriagereka (1995), Cecchetto (2000), Kayne (2002), and Grewendorf (2002), I assume that the resumptive pronoun and the dislocated topic are merged together as a big DP, undergoing movement to SpecTP for Case/EPP reasons. In a subsequent step, the topic further moves into the left clausal periphery.12 (36) [CP topicj ... [TP [DP tj pronoun]i [T’ T [νP ti ... → [CP [TP subjecti [T’ Agr+T [νP ti ... Structural simplification: loss of topic movement, loss of big DP Preservation of argument structure: θ-role previously assigned to big DP is now assigned to previous topic Identification of feature content: via Agree with the subject in SpecνP Word building constraint: verb is located in T (via movement) or adjacent to T at MS (enabling Morphological Merger) As already indicated in (36), the reanalysis in question satisfies the relevant constraints in a more or less straightforward way. Only the status of the Word building constraint

JB[v.20020404] Prn:3/08/2005; 10:20

F: LA8104.tex / p.16 (1007-1058)

 The Rise of Agreement

is somewhat unresolved in this structure and contingent on other properties of the grammar. That is, the language in question must already have had V-to-T movement prior to the reanalysis. Alternatively, T and the verb (which stays in situ) must be adjacent at MS. The latter option is arguably only available in an SVO grammar similar to English, but not in an SOV structure, where the object regularly intervenes between T and the (VP-internal) verb. Note that this predicts that the reanalysis in (36) should only be possible in (i) SVO languages or (ii) SOV languages with verb movement to head-initial T. Possible examples of this type of change include English- and Frenchbased Creoles (Givón 1976), a number of Bantu languages (Givón 1976; Bresnan & Mchombo 1987), and perhaps a number of Northern Italian Rhaeto-Romance dialects such as Friulian, Fassan, or Ampezzan (Haiman & Benincà 1992: 187ff.; but see the discussion around (37) below).13 In Chapter 1, I have argued that (36) does not constitute a viable source of 1st and 2nd person agreement, since in many languages, clitic left dislocation of 1st and 2nd person pronouns leads to highly marked, often ungrammatical results. Accordingly, we must look for a different historical source of 1st and 2nd person agreement. Here, I assume that verbal agreement markers may develop in a somewhat similar construction that also involves some form of clitic doubling (i.e., a big DP), where – for reasons of emphasis – a full DP double is added to reinforce a weak/clitic pronoun that cannot bear stress (cf. Simpson & Wu 2002 for a related idea): (37) [CP [TP [DP double [D’ pronoun]]i [T’ T [νP ti ... → [CP [TP subjecti [T’ Agr+T [νP ti ... Structural simplification: loss of big DP Preservation of argument structure: θ-role previously assigned to big DP is now assigned to previous double Identification of feature content: via Agree with the subject in SpecνP Word building constraint: verb is located in T (via movement) or adjacent to T at MS (enabling Morphological Merger) At first sight, this kind of change is very similar to the reanalysis of clitic left dislocation illustrated in (36). Still, it crucially differs from (36) in that it does not involve (left) dislocation of the double. Because of this difference, (37) may give rise to 1st and 2nd person agreement formatives, while (36) is restricted to 3rd person forms. Thus, it seems reasonable to suppose that languages which license the kind of doubling arising from a big DP structure may employ both strategies for the creation of new verbal agreement markers from former clitic pronouns. Relevant examples of (37) come from languages such as Non-Standard French, where 1sg, 2sg, and 1pl clitics are obligatory and are therefore presumably better analyzed as prefixal agreement markers (Auger 1993, 1994; Gerlach 2002), and a number of Rhaeto-Romance dialects spoken

JB[v.20020404] Prn:3/08/2005; 10:20

F: LA8104.tex / p.17 (1058-1122)

Chapter 4. The transition from pronoun to inflectional marker 

in Northern Italy, which exhibit obligatory clitic doubling with 1st and 2nd person forms (Haiman & Benincà 1992).14 Furthermore, it seems that with some slight modifications, (37) can also account for the rise of object agreement from clitic pronouns in the recent history of Spanish (cf. e.g. Suñer 1988).15 Another possible syntactic source of verbal agreement morphology, which in contrast to the previous scenarios does not involve an intermediate stage of clitic doubling, consists of the ‘direct’ reanalysis of Infl-oriented weak/clitic pronouns in languages that either were already pro-drop languages prior to the reanalysis or developed prodrop properties as a result of it. (38) [CP ... [TP pronouni [T’ T [νP ti ...→ [CP ... [TP [T’ Agr+T [νP pro ... Structural simplification: overt movement replaced by an Agree-relation Preservation of argument structure: θ-role previously assigned to overt pronoun is now assigned to pro Identification of feature content: via Agree with the subject (pro) in SpecνP Word building constraint: verb is located in T (via movement)16 Examples of this reanalysis come from a variety of Northern Italian dialects such as Venetian (Poletto 1995, 2000; Roberts & Roussou 2003), where pronominal elements moved to SpecAgrSP/SpecTP are reanalyzed as realizations of AgrS. As already mentioned in Note 9 above, the reanalysis described in (38) raises some questions with respect to the notion of structural simplification. Roberts and Roussou (2003) assume that the newly created AgrS-head receives the subject’s θ-role, which renders an additional pro unnecessary and ensures that the outcome of the change under consideration is simpler than the structure in the target grammar. However, I prefer to assume that the reanalysis in question leads to the result shown in (38), where a referential pro is added to the structure, which is more in line with standard assumptions concerning the assignment of θ-roles. The requirement of structural simplification is then met by replacing overt syntactic movement to SpecTP by an Agree relation between the T+Agr complex and the empty pronominal subject, which stays behind in its θ-position SpecνP. In other words, (38) satisfies the requirement that the output of the reanalysis should involve less feature syncretisms than the target structure (cf. (35) above) by the loss of an EPP feature in T. Assuming further that the EPP actually has a PF-flavor in the sense that it simply requires that SpecTP is filled by visible material at PF (cf. e.g. Holmberg 2000), it is actually expected that the reanalysis in (38) leads to the loss of an EPP feature in T: due to the fact that the resulting empty subject pronoun is not associated with phonological features, it cannot satisfy the EPP and is therefore banned from moving to SpecTP.17

JB[v.20020404] Prn:3/08/2005; 10:20

F: LA8104.tex / p.18 (1122-1178)

 The Rise of Agreement

Finally, new subject agreement markers may develop in VSO languages via a reanalysis of pronouns that follow the finite verb, which has undergone movement to a higher functional head. Under the assumption that in many VSO languages, the verb moves to an inflectional head such as T, while the subject occupies an intermediate position between T and νP/VP (simply labeled FP here; cf. e.g. McCloskey 1996, 1997 for the Celtic VSO languages, in particular Modern Irish) the relevant structural change can be described as follows:18 (39) [CP [TP [T’ V+T [FP pronouni [νP ti ... → [CP [TP [T’ V+T+Agr [FP ... [νP pro ... Structural simplification: overt movement replaced by an Agree-relation Preservation of argument structure: θ-role previously assigned to overt pronoun is now assigned to pro Identification of feature content: via Agree with the subject (pro) in SpecνP Word building constraint: verb is located in T (via movement) Similar to (38) above, a pronoun is ‘directly’ reanalyzed as a marker of verbal agreement, while the relevant θ-role is assigned to a phonologically empty pronoun that stays behind in SpecνP and is accessed by an Agree operation initiated by the T+Agr complex. Again, it is assumed that the requirement of structural simplification is met by replacing overt pronoun movement with an Agree relation (= loss of an EPP feature in T). The reanalysis in question has presumably taken place in the history of Welsh (cf. Roberts & Roussou 2003) and might be underway in Irish, cf. e.g. McCloskey and Hale (1984). Summing up, this section has introduced a set of syntactic environments that license the reanalysis of Infl-oriented clitics as (bound) agreement markers. It has been shown that the individual scenarios all comply with the set of syntactic conditions outlined in Section 4.3, albeit each in different ways. For example, the requirement of structural simplification may be satisfied by the loss of movement (as in (38) and (39)), the loss of structural complexity (as in (37)), or both, as in (36). Other points of variation concern the element which receives the θ-role previously assigned to the pronominal clitic. The vacant θ-role may be assigned to a full DP that originally served (i) to double the clitic in topic left dislocation structures (as in (36)) or (ii) to reinforce the clitic for reasons of emphasis, as in (37). Alternatively, the clitic may be directly reanalyzed as an agreement marker, with the subject θ-role being assigned to an empty pronoun (as in (38) and (39)). The discussion in this section has shown that Infl-oriented clitics are a viable source of new agreement morphology. However, it appears that the structural changes in question primarily involve head-initial structures. In the next section, it is argued

JB[v.20020404] Prn:3/08/2005; 10:20

F: LA8104.tex / p.19 (1178-1251)

Chapter 4. The transition from pronoun to inflectional marker 

that the reanalysis of C-oriented clitics provides alternative paths toward agreement which can be utilized to coin new inflectional markers in SOV languages.

. Paths toward agreement II: C-oriented clitics In the following, it is shown that languages with C-oriented clitics exhibit some interesting variations on the above patterns. More specifically, I argue that the V2 syntax of many (Germanic/Romance) languages facilitates the reanalysis of C-oriented clitics. Furthermore, it is shown that the reanalysis in question involves the creation of a dissociated Agr-morpheme on C which is inserted post-syntatically (see Chapter 3 above), at least in languages that already had an existing paradigm of verbal agreement markers prior to the reanalysis in question. In addition, I suggest that there are at least two further configurations where C-oriented clitics may turn into new agreement markers in languages lacking the V2 property. Note that the various changes are only briefly outlined in the present section. They are discussed in more detail in Chapter 5, which provides a number of more detailed case-studies of languages where C-oriented clitics developed into new markers of verbal agreement. First, we can observe that in V2 languages, new verbal agreement markers may arise out of clitic doubling structures where a full nominal is added to the clitic pronoun for reasons of emphasis (similar to (37) above). In the course of time, the enclitic, which undergoes cliticization movement to C (either syntactically or at MS), is reanalyzed as an agreement morpheme on C: (40) [CP XP [C’ C+V+pronounj [TP [DP double [D’ tj ]]i [T’ T [νP ti ... → [CP XP [C’ C+V+Agr [TP [TP subjecti [T’ T [νP ti ... Structural simplification: loss of big DP (+ loss of cliticization) Preservation of argument structure: θ-role previously assigned to big DP is now assigned to previous double Identification of feature content: dissociated Agrmorpheme licensed under structural adjacency with Agr-on-T Word building constraint: verb is located in C (via movement) With respect to the requirements of structural simplification and Preservation of argument structure, (40) is quite similar to (37).19 However, it differs in two important ways from the examples with Infl-oriented clitics discussed above. First, the reanalysis in (40) is dependent on the V2 property and can only take place in inversion contexts, since this is the only context where the reanalysis in question can satisfy the Word build-

JB[v.20020404] Prn:3/08/2005; 10:20

F: LA8104.tex / p.20 (1251-1316)

 The Rise of Agreement

ing constraint (at least in SOV languages, C-oriented clitics fail to be adjacent to the verb in embedded clauses). Second, the present reanalysis gives rise to an agreement morpheme on C, whereas it has been assumed that the reanalysis of Infl-oriented clitics leads to a new ‘canonical’ agreement morpheme on T. This typically gives rise to instances of multiple agreement where the subject’s φ-set is not only reflected by the existing verbal morphology (associated with Agr-on-T), but also shows up on the newly created agreement morpheme on C. In Chapter 3 above, it has been argued that this state of affairs results from the post-syntactic insertion of a dissociated agreement morpheme at MS. In the present context, this entails that the pronoun is reanalyzed as a dissociated Agr-morpheme on C that is licensed under structural adjacency with Agron-T, the content of which has been evaluated in the syntactic derivation via Agree. In Chapter 5, it is argued that this reanalysis is apparently underway in a number of (Swiss) Rhaeto-Romance dialects. Second, in inversion contexts, C-oriented clitic pronouns may be ‘directly’ reanalyzed as agreement markers on C, giving rise to pro-drop properties without involving a prior stage of clitic doubling (similar to (38) above): (41) [CP XP [C’ C+V+pronouni [TP ti ’ [T’ T [νP ti ... → [CP XP [C’ C+V+Agr [TP’ T [νP pro ... Structural simplification: overt movement replaced by an Agree-relation Preservation of argument structure: θ-role previously assigned to overt pronoun is now assigned to pro Identification of feature content: dissociated Agrmorpheme licensed under structural adjacency with Agr-on-T Word building constraint: verb is located in C (via movement) Again, this change is made possible by the special V2 syntax of the languages in question, which guarantees adjacency between the pronoun and the verb and thereby facilitates a reanalysis of the enclitic as a dissociated agreement morpheme on C. Similar to (38) above, the resulting structure involves the loss of overt movement, since the newly created empty subject pronoun stays behind in SpecνP and is accessed by an Agree operation initiated by the complex T+Agr (prior to movement of the complex V+T+Agr to C). Relevant examples for this structural change come from a variety of German and Rhaeto-Romance dialects, see Chapter 5 for details. In addition, C-oriented clitics can evolve into verbal agreement markers in languages that lack the V2 property. Here, I will mention only three relevant structural configurations, which presumably do not exhaust the range of possibilities.20 For example, ‘true’ second position clitics which occupy a position after the first prosodic

JB[v.20020404] Prn:3/08/2005; 10:20

F: LA8104.tex / p.21 (1316-1370)

Chapter 4. The transition from pronoun to inflectional marker 

phrase or word may turn into prefixal agreement markers in languages where they are adjacent to the finite verb (presumably located in T): (42) [CP XP [C’ C+pronouni [TP ti ’ [T’ V+T [νP ti ... → [CP XP [C’ C [TP’ Agr+V+T [νP pro ... Structural simplification: overt movement replaced by an Agree-relation Preservation of argument structure: θ-role previously assigned to overt pronoun is now assigned to pro Identification of feature content: via Agree with the subject (pro) in SpecνP Word building constraint: verb is located in T (via movement) In (42), the reanalysis of the pronoun as a prefixal agreement marker requires that the verb is adjacent to second position. Thus, the change in question is possible only in VO languages, in contexts where nothing intervenes between the C-oriented clitic and the finite verb. As in (38) above, the outcome of the reanalysis satisfies the requirement of structural simplification by the loss of overt movement (pro stays behind in SpecνP and is accessed by an Agree operation initiated by the complex T+Agr). Similar to V2 languages, there exists an alternative scenario which involves a doubling configuration. In Chapter 5, it is argued that this grammaticalization process can be observed in a number of Uto-Aztecan languages (Steele 1977), where second position clitics develop into prefixal agreement markers, attaching to the finite verb in T: (43) [CP XP [C’ C+pronouni [TP [T’ V+T [νP [DP double [D’ ti ]] ... → [CP XP [C’ C [TP’ Agr+V+T [νP subject ... Structural simplification: loss of big DP Preservation of argument structure: θ-role previously assigned to big DP is now assigned to the former double Identification of feature content: via Agree with the subject in SpecνP Word building constraint: verb is located in T (via movement) Similar to (40), this reanalysis leads to structural simplification via losing the big DP structure underlying clitic doubling in the target grammar. Preservation of argument structure is warranted by assigning the θ-role to the former double. The newly created agreement morpheme is valued by initiating an Agree operation that accesses the subject in SpecνP, which by assumption may stay in situ in the Uto-Aztecan languages under consideration (see Chapter 5 for details).

JB[v.20020404] Prn:3/08/2005; 10:20

F: LA8104.tex / p.22 (1370-1430)

 The Rise of Agreement

In Chapter 5, I will also discuss a set of Mongolian SOV languages (including Buryat, Dagur, Moghol, Kalmyk; Poppe 1960; Comrie 1980) which developed suffixal agreement markers in their recent recorded history. It is argued that these agreement formatives originated from a structure where enclitic pronouns attach to C while a constituent (presumably TP) containing the rest of the clause moves to a specifier in the C-domain (see Julien 2002 for a related proposal): (44) [CP [TP tj ’ [T’ [νP tj [VP object]] V+T]]i [C’ C+pronounj tTP ]]→ [CP [TP [T’ [νP pro [VP object]] V+T+Agr]]] Structural simplification: loss of TP movement; subject movement replaced by an Agreerelation Preservation of argument structure: θ-role previously assigned to overt pronoun is now assigned to pro Identification of feature content: via Agree with the subject (pro) in SpecνP Word building constraint: verb is located in T (via movement) or adjacent to T at MS (enabling Morphological Merger) In this way, (apparent) clause-final clitics may evolve into verbal agreement suffixes in an SOV language. The reanalysis depicted in (44) leads to a major structural reorganization in which TP movement and overt subject movement to SpecTP are lost, a rather clear-cut case of structural simplification. Note that the change in question represents another way (apart from inversion contexts) in which OV-languages can develop suffixal verbal inflection. Furthermore, in OV-languages there exists always the possibility that the verb stays in situ and combines with the resulting agreement morpheme (on T) via Morphological Merger at MS, due to the fact that in a strict OV grammar, the verb is always string-adjacent to the set of head-final inflectional heads (still, I opt for a structure with overt verb movement in (44), see Chapter 5 for discussion). To sum up, this section has introduced a set of environments where C-oriented clitics may develop into new markers of verbal agreement. First, it has been suggested that the V2 property provides a path for the grammaticalization of agreement markers in SOV languages, where clitics are otherwise not adjacent to the verb. The relevant change may involve either doubling structures where the double turns into the new subject or a ‘direct’ reanalysis of the clitic into an agreement morpheme, giving rise to new pro-drop properties. Importantly, it has been argued that the change in question proceeds via an initial stage where the pronoun is reanalyzed as a dissociated Agr-morpheme on C, resulting in multiple agreement configurations, similar to the phenomenon of complementizer agreement discussed in Chapter 3 above. In addition, I have outlined two further scenarios for the categorial reanalysis of C-oriented

JB[v.20020404] Prn:3/08/2005; 10:20

F: LA8104.tex / p.23 (1430-1482)

Chapter 4. The transition from pronoun to inflectional marker

clitics in languages that lack the V2 property. First, ‘true’ second position clitics may evolve into new prefixal agreement markers in languages where the finite verb is rightadjacent to the clitic. Second, I have argued that clitics which appear in (apparent) clause-final position in SOV languages (e.g., Mongolian) may be reanalyzed as agreement suffixes on the left-adjacent verb which has undergone movement to SpecCP as part of a fronted TP.

. Summary In this chapter, I have argued that the reanalysis of pronouns as agreement markers is not confined to a unique syntactic environment – in contrast to common beliefs in the literature. I suggested that this change can be triggered in a number of contexts where a set of syntactic restrictions on the reanalysis of pronominal elements is fulfilled. At least in part, the relevant syntactic restrictions were derived from the synchronic theory of agreement developed in Chapter 3. In addition, I specified a number of syntactic and morphological diagnostics for distinguishing between clitics and agreement markers which can also be conceived of as cues for the learner during language acquisition. The relevant syntactic criteria reviewed in 4.2.1 are repeated in (45). (45) Indications for an analysis of clitics as agreement markers a. The clitic is obligatory (doubling is not contextually restricted). b. The double in clitic doubling structures may be indefinite/non-specific. c. Anti-agreement between the clitic and the double. d. The clitic appears in subject gap environments. In addition, Section 4.2.2 presented a number of morphological criteria that can also be made use of for the present purpose, basically following Zwicky and Pullum (1983). The relevant criteria included the type of selectional restrictions imposed by the element under consideration (rather loose in the case of clitics, rather strict in the case of affixes), morphophonological effects on the host, and the distribution with respect to other clitics and affixes. In Section 4.3, I argued that the theory of agreement developed in Chapter 3 imposes a set of syntactic restrictions on the reanalysis of pronouns as agreement markers. More precisely, it was claimed that the change in question is only possible if it (i) preserves the predicate’s argument structure (either by assigning the relevant θ-role to the double in a clitic doubling configuration or to a newly introduced pro), (ii) provides a means to value the feature content of the resulting agreement morpheme (either in the syntax or at MS), and (iii) warrants that (bound) agreement morphemes can combine with the verb prior to Vocabulary Insertion. In addition, I followed Roberts and Roussou (2003) in assuming that the reanalysis must conform



JB[v.20020404] Prn:3/08/2005; 10:20



F: LA8104.tex / p.24 (1482-1549)

The Rise of Agreement

to a simplicity metric (formulated in terms of feature syncretism) which requires that the resulting structure is less complex than the target structure. Subsequently, I suggested that these conditions can be met in a variety of different syntactic environments. In Section 4.4, I focused on the reanalysis of Infl-oriented clitics in SVO and VSO grammars. It was shown that viable historical scenarios for the change in question are divided into two groups. In the first group, the θ-role previously assigned to the clitic is transmitted to a full DP that originally served to double the clitic in (i) clitic left dislocation structures, or (ii) doubling configurations where a full nominal is added to the clitic for reasons of emphasis. By assumption, both scenarios involve the reanalysis of a former big DP which contains the clitic and the double. In the second group, the clitic is ‘directly’ reanalyzed as an agreement marker, with the subject θ-role being assigned to an empty pronoun which is newly added to the structure. It was demonstrated that all these environments meet the requirement of structural simplification either via loss of phrase structure or loss of movement operations. In Section 4.5, I argued that languages with C-oriented clitics provide a number of further environments where pronominal elements may be reanalyzed as verbal agreement markers. First, I suggested that the reanalysis of C-oriented enclitics provides an alternative path to new verbal agreement morphology in languages that exhibit the V2 property. Again, the relevant reanalysis may involve either doubling structures where the double turns into the new subject or a ‘direct’ reanalysis of the clitic into an agreement morpheme, giving rise to new pro-drop properties. In line with the analysis of agreement presented in Chapter 3 above, I argued that the change in question may lead to the presence of a dissociated Agr-morpheme on C, giving rise to instances of multiple agreement. In addition, I claimed that C-oriented clitics may also give rise to new verbal agreement markers in languages lacking the V2 property. By assumption, new prefixal agreement markers may evolve from ‘true’ second position clitics in languages where the finite verb is right-adjacent to the clitic. In addition, it was proposed that clause-final clitics in SOV languages such as Mongolian may give rise to new verbal inflection if it is assumed that the clitics in question attach to right of C which hosts a fronted TP in its specifier. At this point let me address the question of whether it is possible to establish a connection between the suffixing preference (cf. Chapter 1) and the set of historical scenarios discussed in the present chapter. At first sight, it seems that this not the case. Of the nine syntactic environments presented above, five give rise to prefixal agreement, while only four result in agreement suffixes. Thus, it seems that there exists in fact a slight preference for the development of prefixal agreement. However, note that I have identified only two scenarios for the development of new agreement markers in SOV languages (inversion and TP-fronting), which both give rise to suffixal agreement. This captures the fact that SOV languages are predominantly suffixing (cf. Chapter 1). In addition, recall that SOV is the most common word order across the world’s languages, so the two scenarios in question are presumably much more frequent than other pathways which may lead to prefixal agreement. Moreover, the fact that postver-

JB[v.20020404] Prn:3/08/2005; 10:20

F: LA8104.tex / p.25 (1549-1613)

Chapter 4. The transition from pronoun to inflectional marker

bal material affixes more readily to the verb than preverbal material is presumably also influenced by morphophonological factors, to the effect that enclitics are more likely to fuse with the verb than proclitics (cf. e.g. Cutler, Hawkins, & Gilligan 1985; Hawkins & Gilligan 1988; Hall 1988; Bybee et al. 1990; Kirby 1999; Siewierska 2004 for discussion). Tentatively, I therefore suggest that the suffixing preference should receive a historical explanation, in the sense that it results at least partially from the way agreement markers come into existence historically. In other words, it seems that cross-linguistically, the grammaticalization process in question is more likely to be triggered in syntactic environments which give rise to suffixal agreement morphology.21 In the following chapter, I will discuss the historical scenarios involving the reanalysis of C-oriented clitics in some more detail, drawing on examples from German and Rhaeto-Romance dialects, Uto-Aztecan, and Mongolian languages.

Notes . Note that in contrast to Italian, Macushi exhibits an absolutive/ergative organization. The absolutive argument is cross-referenced on the verb by a prefix, while the ergative argument is marked by a suffix which follows the tense marker. . For the claim that subject clitic doubling constitutes the historical source of verbal agreement cf. e.g. Givón (1976), Haiman (1991), Haiman and Benincà (1992), Simpson and Wu (2002). For an analysis of the subject clitics of Northern Italian dialects as agreement heads, cf. Rizzi (1986), Brandi and Cordin (1989), Suñer (1992), Poletto (1995, 1999, 2000), among many others. For the claim that object clitic doubling in Spanish in fact constitutes an instance of object agreement cf. e.g. Givón (1976), Borer (1984), Suñer (1988). . Note that this relates to the observation that it is a characteristic of grammaticalization processes that the relevant element becomes more and more obligatory (see Chapter 2 above for discussion). . Similar facts can be observed in Non-Standard French (cf. Lambrecht 1981): (i)

a. b.

Il he I he

mange eats mange eats

et and et and

boit comme un drinks like a i boit comme he drinks like

cochon. pig un cochon. a pig

(Standard) (Non-Standard)

. Other criteria which have been proposed in the literature concern for example the placement of clitics relative to other elements such as negation (Haiman 1991; Haiman & Benincà 1992, and in particular Poletto 2000). See Gerlach (2002: Ch. 2) for further criteria and in-depth discussion and Ingram (1978) for a set of discourse related diagnostics. . According to Kayne (1984), the deficient subject pronouns of Standard French behave like NPs in the syntax, but attach to the verb as clitics at PF. . Here, the question arises of whether adjacency must hold without exceptions for a reanalysis to be possible or whether it is sufficient if adjacency is given in the majority of cases. Note that in many cases this creates no problems, since adjacency of the verb and the clitic is usually obligatory.



JB[v.20020404] Prn:3/08/2005; 10:20

F: LA8104.tex / p.26 (1613-1675)

 The Rise of Agreement . Note that (31) must be formulated in more general terms to account for cases where the agreement marker does not attach to the verb, but rather to other (free) inflectional morphemes such as tense. Relevant examples come from Australian languages such as Wambaya (showing a clitic Agrs/Agro/T marker, Nordlinger 1995), the Mayan language Jacaltec (Craig 1977), where Agrs attaches to a clause-initial completive marker, and Finnish, where Agrs may attach to the negation. (i)

Word building constraint (alternative) The reanalysis of a clitic as a (bound) agreement marker on a functional head X requires that X combines with the potential host of agreement prior to Vocabulary Insertion.

. To make sure that (34) fulfills the requirement of structural simplification, Roberts and Roussou (2003: 183) assume that the newly created AgrS-head receives the subject’s θ-role, in contrast to standard assumptions concerning the realization of argument structure. Otherwise, one would have to assume that the reanalysis of (34) requires the presence of a referential pro which receives the θ-role previously assigned to the pronoun. This, however, would annul the structural simplification shown in (34) if pro moves to SpecTP for Case/EPP reasons, similar to the overt pronoun. See below for some discussion and an alternative proposal. . Note that this interpretation of the workings of the simplicity metric differs somewhat from the proposals in Roberts and Roussou (2003), where feature syncretism and simplification is associated with the PF-realization of features. More specfically, Roberts and Roussou assume that a lexical item L that moves to a functional position F (and is spelled-out there) realizes both its own features and the formal features of F. Structural simplification is then understood in terms of the PF realization of these features, in the sense that a lexical item which PF-realizes two separate features X and Y is more complex than a lexical item which realizes only a single feature. . In this book, I will not deal with instances where new subject (absolutive) agreement markers develop from former object pronouns. This often involves the reanalysis of passives as actives, which in addition may give rise to an ergative alignment in the languages under investigation (cf. e.g. Anderson 1980). . Under a big DP analysis, Givón’s proposals might lead to suffixal agreement if it is assumed that the resumptive clitic may remain in situ (i.e., SpecνP), escaping the Case filter via cliticization onto the verb in T. In general, absence of subject movement to SpecTP may open up further possibilities with respect to syntactic environments where subject agreement can arise. . For more examples cf. Givón (1976) and the references therein. According to Givón, NonStandard French constitutes a further instance of this reanalysis. However, it seems equally possible that the change affecting French involves the structures in (37). . Note that only (36) is associated with certain person restrictions. In contrast, (37) may give rise to a full paradigm of agreement formatives. Thus, one might speculate that the cases commonly attributed to (36) can be subsumed under (37). I leave that point open for future research. . In cases of Spanish object clitic doubling such as (i) (Suñer 1988: 394), the double stays in a position to the right of the verb, while the clitic undergoes movement (either in the syntax or at PF/MS) to the left of the verb. This order may then be reanalyzed as involving a prefixal agreement marker and a postverbal object (presumably in situ):

JB[v.20020404] Prn:3/08/2005; 10:20

F: LA8104.tex / p.27 (1675-1741)

Chapter 4. The transition from pronoun to inflectional marker

(i)

La llamaron a ella. her call-3pl to her ‘They called her.’

. It is standardly assumed that the verb has to undergo movement in pro-drop languages to license the empty subject pronoun. Thus, the late combination of the verb and the Agr+T complex via Morphological Merger is presumably not available. . As already mentioned above in Chapter 2 (Note 28), the reanalysis in (38) may lead to the apparent multiple realization of agreement features in cases where the verb already carries an agreement ending, as in the Northern Italian dialects discussed by Poletto (1995, 2000) and Roberts and Roussou (2003). Note that the analysis of agreement proposed in Chapter 3 above makes available an alternative solution to the problem of multiple agreement. More precisely, I suggest that in cases where the verb already carries (suffixal) agreement morphology, the pronoun is reanalyzed as a dissociated agreement morpheme that adjoins post-syntactically to a functional head directly above the head that licenses verbal agreement (presumably Agr-on-T). Recall that the dissociated Agr-morpheme is then licensed under structural adjacency with Agron-T (to which the verb moves). Thus, I propose the following structures for the reanalysis in question (with the host of the dissociated Agr-morpheme simply labeled F): (i)

[TP DPi [T’ V+T+Agr ] ... [VP ti ... →[FP F+Agr [F’ [TP [T’ V+T+Agr ] ... [VP ...

. A similar reanalysis, albeit one step further up in the structure, may take place if VSO languages are analyzed as involving verb movement to C, with the subject pronoun in SpecTP. . In V2 languages, one might expect that there exists an analogue of (36) in which C-oriented resumptive enclitics turn into agreement markers in left dislocation contexts. However, in all Germanic languages, the relevant structures involve a resumptive demonstrative which precedes the verb, giving rise to V3 orders as for example in German (cf. Grewendorf 2002): (i)

Der Peter, der hat Maria geküsst. the Peter dem has Mary kissed. ‘As for Peter, he kissed Mary.’

. For example, as already mentioned above in Note 18, C-oriented clitics may turn into new agreement markers in verb-initial languages where the verb moves to C. . Furthermore, note that many present-day SVO languages were originally SOV languages. In contrast, the change from SVO to SOV seems to be quite rare cross-linguistically (cf. e.g. Kiparsky 1996). Moreover, it is a well-known fact that the suffixal agreement morphology of many present-day SVO languages is of considerable antiquity. Taken together, these facts suggest that in many present-day SVO languages, the suffixal agreement formatives actually developed when the language in question was SOV. In contrast, the converse development leading to agreement prefixes in a present-day SOV language which was previously SVO is much less likely. This presumably provides a diachronic explanation of the fact that there are much more SVO languages with agreement suffixes than SOV languages with agreement prefixes.



JB[v.20020404] Prn:2/08/2005; 15:48

F: LA8105.tex / p.1 (45-122)

Chapter 5

The reanalysis of C-oriented clitics

. Introduction This chapter deals with a particular subset of the structural reanalyses outlined in Chapter 4, focusing on instances where C-oriented subject clitics develop into new verbal agreement markers. Relevant examples of this historical scenario come from a variety of V2 languages such as Bavarian, German dialects spoken in Northern Italy (Cimbrian, Walser German), and a number of Rhaeto-Romance dialects spoken in Switzerland. In addition, it is shown that related changes can be observed in languages lacking the V2 property such as Uto-Aztecan and Mongolian. Based on the insights reached in Chapter 4, I argue that the different historical pathways leading to new verbal agreement markers share a well-defined set of properties, reflecting the universal nature of the abstract syntactic conditions which shape the grammaticalization process in question across languages. The chapter is organized as follows. Section 5.2 deals with the grammaticalization of new verbal agreement markers from C-oriented enclitics in the history of Germanic, focusing on Bavarian, where new agreement endings developed for 2nd person forms and 1st plural. Section 5.2.1 describes the course of the grammaticalization process while Section 5.2.2 presents a new account of the relevant changes based on the idea that enclitics were reanalyzed as exponents of a dissociated Agr-morpheme on C. In Section 5.2.3, I discuss further examples of this particular grammaticalization process which can be observed in the dialects of German speaking minorities in Northern Italy, drawing on data from Cimbrian and Walser German. Section 5.3 is concerned with historical developments in a set of Rhaeto-Romance dialects spoken in Switzerland. After providing a grammatical sketch of the languages under consideration in Section 5.3.1, I show in Section 5.3.2 that similar to Bavarian, the Swiss Rhaeto-Romance dialects were affected by a change in which C-oriented clitics were reanalyzed as enlargements of existing agreement endings in inversion contexts. Section 5.3.3 shows that some of these dialects exhibit another grammaticalization process in which Coriented clitics are reanalyzed as agreement markers in clitic doubling configurations. Section 5.3.4 provides an analysis of this change which is again based on the idea that clitics are reanalyzed as exponents of a dissociated morpheme on C. Section 5.4 deals with alternative paths to new agreement markers in languages lacking the V2 property. Section 5.4.1 focuses on second position clitics that turn into prefixal agreement in a number of Uto-Aztecan languages. Section 5.4.2 deals with the Mongolian SOV lan-

JB[v.20020404] Prn:2/08/2005; 15:48



F: LA8105.tex / p.2 (122-197)

The Rise of Agreement

guage Buryat, arguing that the reanalysis in question targets C-oriented clitics which show up in clause-final position due to TP-fronting.

. Bavarian In contrast to Standard German, Bavarian shows two series of pronominal elements: (i) a paradigm of full forms that may bear stress, and (ii) a set of reduced, atonic forms. The latter are C-oriented enclitics which attach to the right of elements located in the C-system. Table 1. Nominative pronominal forms of Bavarian (Bayer 1984: 230). Full form 1sg 2sg 3sg.masc 3sg.fem 3sg.neut 1pl 2pl 3pl

I du er, der (demonst.) sie, die (demonst.) es, des (demonst.) mir ihr, es sie, die (demonst.)

Enclitic /i/ /du/ /er/, /der/ /si/, /di/ /es/, /des/ /mir/ /ir/, /εs/ /si/, /di/

-a, -e -(s)t -a -s -s -ma -(t)s -s

/a/, /e/ /st/ /a/ /s/ /s/ /mer/ /ts/ /s/

The enclitic forms of the 2nd person nominative pronouns (2sg -st, 2pl -ts) exhibit special properties which set them apart from the other pronominal clitics. The particular status of the 2nd person forms has received quite some attention in the literature (cf. e.g. Pfalz 1918; Merkle 1975; Bayer 1984; Altmann 1984; Zehetner 1985; Werner 1988; Harnisch 1989; Wiesinger 1989; Nübling 1992; Rowley 1994; Abraham 1995; Weiß 1998, 2002). The special behavior of the 2nd person clitics is commonly taken to indicate that they are actually not pronominal elements. Instead, they are often analyzed as pieces of inflection which attach to other elements in the left periphery of the clause.1 The particular properties of the 2nd person ‘clitics’ are illustrated in the following. First, there are a number of syntactic (i.e., distributional) differences between 2nd person forms and the other pronominal clitics. In contrast to the latter, the 2nd person forms are obligatory in all contexts, that is, they cannot be simply replaced by the relevant full forms: (1) a.

ob-st noch Minga kumm-st whether-2sg to Munich come-2sg b. *ob du noch Minga kumm-st whether you.sg to Munich come-2sg ‘Whether you come to Munich’

JB[v.20020404] Prn:2/08/2005; 15:48

F: LA8105.tex / p.3 (197-278)

Chapter 5. The reanalysis of C-oriented clitics 

(2) a.

ob-ts noch Minga kumm-ts whether-2pl to Munich come-2pl b. *ob es/ihr noch Minga kumm-ts whether you-pl to Munich come-2pl ‘Whether you come to Munich’

In addition, a full 2nd person subject pronoun is only acceptable if it co-occurs with the relevant clitic form, giving rise to instances of subject clitic doubling: (3) a.

ob-st du noch Minga whether-2sg you.sg to Munich ‘Whether you come to Munich’ b. ob-ts es/ihr noch Minga whether-2pl you.pl to Munich ‘Whether you come to Munich’

kumm-st come-2sg kumm-ts come-2pl

This contrasts with the behavior of the other subject enclitics, which are not obligatory and which cannot be doubled by full pronouns:2 (4) a.

ob=e (*i) noch Minga kumm whether=clit.1sg I to Munich come-1sg b. ob i noch Minga kumm whether I to Munich come-1sg ‘Whether I come to Munich’

Furthermore, Altmann (1984) notes another difference between the subject enclitics for 2nd person and the other enclitic forms, having to do with the way the presence of a clitic affects the distinction between V1 and V2 structures. When a 1st or 3rd person subject enclitic follows a clause-initial finite verb, the structure is unambiguously interpreted as a V1 clause: (5) kh enn=6 enngg wõ. know=clit.1sg you.pl modprt ‘I know you guys.’ (Altmann 1984: 206) However, if a 2nd person form is attached to a clause-initial verb, a sentence such as (6a) is ambiguous between a V1 and a V2 structure. This ambiguity can be resolved if a full pronoun is added, as in (6b): hfbbd-s kh õ! geId nImm!. have-2pl no money not-anymore ‘You have no money anymore.’ b. hfbbd-s e:s kh õ! geId nImm!. have-2pl you.pl no money not-anymore ‘You have no money anymore.’ (Altmann 1984: 207)

(6) a.

JB[v.20020404] Prn:2/08/2005; 15:48

F: LA8105.tex / p.4 (278-345)

 The Rise of Agreement

Table 2. 2nd person full/reduced pronominal forms and verbal agreement.

2sg 2pl

Full pronoun

Enclitic pronoun

Verbal agreement

du es/ihr

-st -ts

-st -ts

Altmann concludes that the different behavior of 2nd persons forms follows directly from their inflectional character: since they are always obligatorily present, they cannot serve to distinguish between different word order or clause types. Apart from these syntactic differences, we can also observe some crucial morphophonological differences that underline the special status of 2nd person forms, cf. Table 2. First, in contrast to the other clitics, the 2nd person forms cannot be derived from the relevant full pronouns via phonological reduction. Second, the 2nd person clitics are identical with the relevant verbal agreement suffixes, which already suggests that they are in fact inflectional elements. Accordingly, Pfalz (1918), Bayer (1984), Altmann (1984), Zwart (1993b), and Weiß (1998, 2002, to appear) (among others) propose that the special properties of the 2nd person subject clitics can be explained on the assumption that these elements are in fact some form of inflection located in C. This view is also adopted in the present work. In line with the analysis proposed in Chapter 3, the 2nd person ‘enclitics’ are analyzed as exponents of a dissociated agreement morpheme that attaches post-syntactically to C (henceforth Agr-on-C). Note, however, that this assumption leaves an open question, namely why complementizer agreement is restricted to 2sg, 2pl forms. This question will be addressed in some more detail in Chapter 6. Another consequence of this analysis is that Bavarian is considered a ‘partial prodrop’ language (Bayer 1984; Weiß 1998): sentences which contain only a 2nd person clitic have to be analyzed as instances of pro-drop (according to Bayer 1984, the overt manifestation of agreement in C (with 2sg, 2pl) serves to license referential pro in present-day Bavarian): (7) a.

Kumm-st pro noch Minga, dann muas-st pro me bsuacha. come-2sg to Munich then must-2sg me visit ‘If you come to Munich you must visit me.’ b. Kumm-ts pro noch Minga, dann müass-ts pro me bsuacha. come-2pl to Munich then must-2pl me visit ‘If you come to Munich you must visit me.’

(8) a. *Kumm pro noch Minga ... come-1sg to Munich ‘If I come to Munich, ...’ b. *Kumm-t pro noch Minga? come-3sg to Munich ‘Will he/she/it come to Munich?’

JB[v.20020404] Prn:2/08/2005; 15:48

F: LA8105.tex / p.5 (345-408)

Chapter 5. The reanalysis of C-oriented clitics

Note that in examples such as (7), there is no double affixation of agreement morphology to verbs that move to C via T, that is, forms such as kummst-st ‘come-2sg-(clit)2sg’ are ungrammatical. On the assumption that -st and -ts are clitics, one has to resort to some sort of phonological filter to rule out the sequence st-st (and ts-ts) that should be possible after cliticization to an inflected verb has taken place. However, a more principled (and interesting) explanation becomes available if the 2nd person enclitics are treated as agreement markers, as suggested. Then, the absence of forms such as kummst-st can be analyzed in terms of a general morphological condition that restricts the number of overtly realized (identical) Agr-morphemes on the verb to one, see Chapter 3 above. This lends further support to an analysis of the 2nd person clitics as pieces of inflection.3 Again, the fact that pro-drop is restricted to 2nd person contexts is somewhat mysterious. For example, it cannot be attributed to some special morphological property of the 2nd person agreement suffixes, in the sense that 2nd person forms are morphologically ‘stronger’ or more distinctive than 1st or 3rd person forms, which enables only the former but not the latter to license pro: Table 3. Verbal agreement paradigm (pres. indic.) of Bavarian. Verbal agreement 1sg 2sg 3sg 1pl 2pl 3pl

-Ø -st -t -an -ts -an

Table 3 shows that the only non-distinctive (i.e., homophonous) agreement endings are 1pl and 3pl, whereas all other endings serve to unambiguously signal person and number of the subject. Thus, it appears that the restrictions on complementizer agreement and pro-drop observed in Bavarian escape a deeper synchronic account and can only be captured by a stipulation that simply rules out the non-existing cases (cf. Bayer 1984). In this work, I argue that the restrictions in question can be more fully understood if diachronic evidence is taken into account. More specifically, it is claimed that the restrictions on pro-drop and complementizer agreement follow from a set of syntactic and morphological factors that determined the reanalysis of subject clitics as verbal agreement markers in the history of Bavarian (see Chapter 6 for an explanation of the person/number restrictions in terms of morphological factors which shaped the development of Agr-on-C). In this section, it is shown that this reanalysis could only take place in contexts with V-to-C movement, forcing the learner to assume the existence of (i) pro-drop and (ii) an agreement morpheme on C, leading to complementizer agreement. This analysis of the diachronic facts is based on the analysis



JB[v.20020404] Prn:2/08/2005; 15:48

F: LA8105.tex / p.6 (408-494)

 The Rise of Agreement

of complementizer agreement developed in Chapter 3, where the presence of inflection in the C-domain was attributed to the post-syntactic insertion of a dissociated Agr-morpheme at Morphological Structure.

.. The diachronic development of Agr-on-C in Bavarian In all varieties of Bavarian, C-oriented enclitics were reanalyzed as enlargements of existing agreement markers, giving rise to new agreement suffixes for 2nd person (cf. Bayer 1984; Wiesinger 1989). Table 4 below lists the inherited verbal agreement endings and the forms resulting from the reanalysis of the pronominal clitics for 2sg, t(hu), and 2pl, (¯e)s (originally a dual that developed into a 2pl pronoun in Bavarian). Interestingly, it can be shown that this grammaticalization process was initially confined to certain syntactic environments. More specifically, the relevant data suggest that the historical development of the new 2nd person agreement suffixes affected first finite verbs in C and spread later to other verbal positions.

... 2sg /-st/ The development of 2sg /-st/ began already in early Old High German (OHG), presumably during the 9th century AD. The resulting new ending is found in all modern German varieties. It is commonly assumed (cf. Brinkmann 1931; Braune 1950: 252; Sommer 1994) that the agreement formative 2sg -s+t resulted from a reanalysis of the combination of verb and clitic pronoun t(hu) in inversion contexts, possibly on the analogy of the preterite-presents, which already showed /-st/ for the 2sg present indicative (kanst ‘can’, tarst ‘dare’, muost ‘must’, weist ‘know’ etc.), and the 2sg of ‘be’ bist, which resulted from an independent and earlier development (cf. Lühr 1984).4 Examples such as (9) from the early OHG Tatian seem to suggest that the new agreement ending was initially confined to verbs that have undergone V-to-C movement: (9) Ih forahta, [CP uuanta thu grim man bist, [C’ nimi-st [IP thaz thu I feared since you grim man are take-2sg that you ni sázto-s]]] inti [CP [C’ arno-st [IP thaz thu ni sáto-s.]]] neg plant-2sg and earn-2sg that you not sow-2sg ‘Since you are a grim man, I feared that you take what you haven’t planted and earn what you haven’t sowed.’ (Tatian ζ 151,7; Sievers 1961: 228) In (9), we can see that the new ending shows up on verbs that occupy the C position in V2 contexts, whereas finite verbs in clause-final position still exhibit the older ending /-s/.5 However, a second look at the relevant data reveals that the picture is acTable 4. Old and new agreement suffixes for 2nd person in Bavarian.

2sg 2pl

‘Old’ inherited ending

‘New’ enlarged ending

Lexical source

-s -t

-s+t -t+s

t(hu) (¯e)s

JB[v.20020404] Prn:2/08/2005; 15:48

F: LA8105.tex / p.7 (494-590)

Chapter 5. The reanalysis of C-oriented clitics 

tually more complex. In the following example, also taken from the OHG Tatian, the finite verb (ges ‘go-2sg’) does not show the enlargement /-t/ despite the fact that it has undergone inversion with the subject pronoun thu:6 (10) Ni ges thú thanan úz, ér thanne thu giltis then iungiston not go you from-there out before you return the newest scáz. treasure ‘You will never get out until you have paid the last penny.’ (Tatian, β, 27.3; Sievers 1961: 51) In general, it appears that the OHG Tatian reflects an early stage of the development of the 2sg formative /-st/. The following table shows that the old ending /-s/ by far outnumbers the enlarged form /-st/, even in inversion contexts (cf. Sommer 1994: 328): Table 5. 2sg endings in the OHG Tatian (without bist ‘be.2sg’). Total -s -s thu -st -st thu -st(t)u

166 15 26 4 9

As noted in Chapter 2 above, this kind of variation can be analyzed as an instance of grammar competition (Kroch 1989), where speakers have access to more than one internalized grammar. In the case at hand, the relevant grammars presumably differ with respect to the presence of an agreement morpheme in the C domain (giving rise to the ending /-st/). Of course, settling this matter requires a detailed quantitative analysis of the OHG texts in question (which is a topic for future research; see Chapter 6 for morphological aspects of the development of 2sg /-st/ in OHG/Bavarian). Nevertheless, it seems reasonable to suppose that the historical development of the new verbal agreement suffix /-st/ proceeded via the following stages: V + enclitic → V+Agr/inversion contexts (reanalysis of clitics as Agr-onC) b. extension to other elements located in the C-system such as complementizers, relative pronouns etc. (confined to Bavarian) c. extension to verbs in clause-final position

(11) a.

Note that at least with respect to the development of 2sg /-st/ there is not much empirical data available which can be used to evaluate the hypothesis in (11), due to the fact that the change in question predates the earliest Bavarian records. Still, even with no relevant empirical evidence available, the order of events assumed in (11) seems to be

JB[v.20020404] Prn:2/08/2005; 15:48

F: LA8105.tex / p.8 (590-616)

 The Rise of Agreement

well-motivated from a conceptual point of view. First, it is rather uncontroversial that sentences with inversion provided the only context where a pronoun could be reanalyzed as part of the verbal agreement morphology, since only in inversion contexts, the pronoun was adjacent to the finite verb in early Bavarian/OHG. Thus, it is fairly clear that the new ending could not develop first on clause-final verbs. In other words, stage (11a) must precede stage (11c). What about the chronological ordering of (11a) and (11b)? One might assume that enclitics were directly reanalyzed as agreement markers on complementizers, claiming that stage (11b) in fact might have preceded (11a) or developed simultaneously. However, I think that it is more natural to assume that agreement affixes spread from verbs (in inversion contexts) to complementizers rather than the other way around (cf. Altmann 1984: 200 for a similar view). In general we should expect new patterns of verbal agreement to develop first on verbs. It would be surprising if complementizers were leading the charge.7 This is also in line with the idea that during language acquisition, the relevant trigger experience comes from simple non-embedded contexts only (so-called Degree-0-Learnability, Lightfoot 1991). As a consequence, properties of embedded contexts must be deducible from properties of non-embedded contexts (however, see below for some qualifying remarks). Therefore, we should not expect the development of complementizer agreement to predate the development of new verbal agreement in inversion contexts. Next, it is shown that the set of hypotheses stated in (11) receives further support by later developments that affected the agreement markers for 2pl and 1pl in the history of Bavarian.

... 2pl /-ts/ As already noted above, in Bavarian, the original inherited ending for 2pl, /-t/, (which is still found today in most other German varieties) was replaced by /-ts/ via a reinterpretation of the enclitic form (-s) of the pronoun ¯es, originally a dual. It is usually assumed that this grammaticalization process began in the 13th century (cf. e.g. Wiesinger 1989). Again, it seems that the new ending appeared first on finite verbs in inversion contexts. Moreover, in the case of 2pl /-ts/, there are further indications supporting the sequence of historical stages assumed in (11) above. In other words, it can be shown that after the initial development of a new agreement marker which shows up on the finite verb in inversion contexts, the new ending first spread to other potential hosts in the C-domain such as complementizers, wh-phrases, or relative pronouns before it was generalized to verbs in clause-final position. Pfalz (1918: 232) observes that in some northeastern Bavarian dialects (spoken in the Lauterbach and Sangerberg area), the new ending for 2pl (/-ts/) still attaches only to conjunctions and verbs in C, but crucially not to verbs which occur in clause-final position: (12) wou-ts teham sat where-2pl at-home are ‘...where you are at home’

JB[v.20020404] Prn:2/08/2005; 15:48

F: LA8105.tex / p.9 (616-688)

Chapter 5. The reanalysis of C-oriented clitics 

(13) [CP wei-ts iw6 t’pruk khumt [C’ sea-ts [IP s’wi6tshaus]]] when-2pl over the-bridge come-2pl see-2pl the-tavern ‘When you cross the bridge, you see the tavern.’ These facts suggest that the new agreement marker was first grammaticalized as a realization of the feature content of C (attaching to verbs and other elements located in the C-system) and spread later to verbs in clause-final position.8 Further support for the claim that the change in question proceeded along these lines (i.e., via a transitional stage where the new ending was confined to the C-domain) comes from Lower Bavarian dialects which exhibit a related grammaticalization process affecting 1pl forms.

... 1pl /-ma/ in Lower Bavarian In a set of Lower Bavarian dialects, the 1st person plural subject enclitic -ma developed in a similar way as the 2nd person enclitics (cf. Pfalz 1918; Merkle 1975; Bayer 1984; Altmann 1984; Kollmer 1987; Wiesinger 1989; Rowley 1994; Abraham 1995; Weiß 1998, 2002, to appear). That is, the pronominal enclitic 1pl -ma shows a similar behavior as the 2nd person forms: it is obligatory in all contexts, it can be doubled by full forms for emphatic reasons, and 1pl contexts license pro-drop in these dialects (Bayer 1984: 252):9 (14) a.

wem-ma aaf Minga fon when-1pl to Munich drive b. wem-ma mia aaf Minga fon when-1pl we to Munich drive c. *wem mia aaf Minga fon when we to Munich drive ‘When we drive to Munich’ (Weiß 2002: 9)

(15) a.

Mia fom-ma hoam. we drive-1pl home ‘We go home.’ (Weiß 2002: 9) b. *Mia fon hoam. we drive home ‘We go home.’ (Helmut Weiß, p.c.)

(16) Fom-ma pro noch Minga? drive-1pl to Munich ‘Will we go to Munich?’ Therefore, it is plausible to assume that /-ma/ developed into an additional instance of Agr-on-C (cf. Bayer 1984; Weiß 1998, 2002, to appear). This claim is further supported by the fact that in V2 clauses, /-ma/ replaces the original agreement ending in bisyllabic verbs such as laffa ‘to run’, gengan ‘to go’, soucha(n) ‘to seek’ etc. (cf. Kollmer 1987; Weiß 2002):

JB[v.20020404] Prn:2/08/2005; 15:48

F: LA8105.tex / p.10 (688-753)

 The Rise of Agreement

(17) a.

Mia laff-ma/*laff-a hoam. we ran-1pl/ran-1pl home ‘We are running home.’ b. Mia gem-ma/*geng-an hoam. we go-1pl/go-1pl home ‘We are going home.’

Crucially, no such replacement is possible when the finite verb shows up in sentencefinal position: (18) wa-ma hoam laff-a/*laff-ma. because-1pl home go-1pl ‘Because we are going home’ In other words, the dialects in question show a complementary distribution of the new suffix /-ma/ and the old ending for 1pl, /-a(n)/ (cf. Kollmer 1987: I, 357): /-ma/ systematically appears on verbs in V2 clauses (main and embedded contexts), cf. (17), whereas verbs in sentence-final position maintain the old ending /-a(n)/, cf. (18). As already noted above in Note 3, a similar development is exhibited by a number of Carinthian dialects (Lessiak 1963; Wiesinger 1989; Abraham 1995). In the following examples, we can observe a multiple realization of 1pl subject marking: (i) via the ending -mr/-m6 on the verb, (ii) via a C-oriented clitic, and (iii) via an overt full pronoun (similar to West Flemish, cf. Chapter 3, Section 3.7): (19) a.

Kher-m6=mr wir a¯ aufn? belong-1pl=clit.1pl we up ‘Do we belong at the top?’ b. wi6-mr=mr wir wöln as-1pl=clit.1pl we want ‘The way we want (it)’ (Pernegg, Lessiak 1963: 204)

This clearly shows that the ending -mr/-m6 cannot be a clitic – it would be quite odd to have two clitics in these sentences – and accordingly must also be analyzed as an agreement marker, that is, another instance of Agr-on-C.10 Furthermore, it seems that at least in some dialects, the new agreement ending has spread to verbs in clause-final position, replacing the original agreement ending in all contexts: (20) ... wos m#r wöl-m#r what we want-1pl (Wiesinger 1989: 38) Interestingly, an analogous development has apparently taken place in a subset of the Lower Bavarian dialects as well. In some varieties spoken in the Bavarian Forest,11 /-ma/ appears on clause-final frequently used (light) verbs such as ‘have’ and ‘do’ (Kollmer 1987: I, 357; Wiesinger 1989: 38; Weiß 2002: 9). Note that /-ma/ must be an-

JB[v.20020404] Prn:2/08/2005; 15:48

F: LA8105.tex / p.11 (753-777)

Chapter 5. The reanalysis of C-oriented clitics 

alyzed as an agreement marker in the following examples, since enclitics cannot attach to clause-final verbs in Bavarian. (21) a.

dass-ma (mia) koã geid ned hã-ma [instead of 1pl hã-n] that-1pl we no money not have-1pl ‘That we have no money’ (Kollmer 1987: I, 362) b. we-ma (mia) des ned dou-ma ... [instead of 1pl dou-n] if-1pl we that not do-1pl ‘If we don’t do that...’ (Kollmer 1987: I, 358)

The above discussion has shown that in Bavarian, new verbal agreement morphology developed from enclitic subject pronouns in inversion contexts. With respect to 2nd person forms, the available evidence suggests that the new 2nd person agreement markers developed first on verbs located in C and spread to other verbal positions in a subsequent development. The latter claim is supported by data from Lower Bavarian dialects where the new agreement marker for 1pl, /-ma/, is obligatory on elements in C (i.e., on verbs and complementizers), but still excluded on most clause-final verbs.

.. Clitics, V2, and the rise of agreement The basic proposal for the rise of new verbal agreement markers in Bavarian is illustrated by the phrase markers in (22) and (23). That is, I assume that the clitic pronoun that attaches to C in the target grammar, cf. (22), is reanalyzed as the realization of an Agr-morpheme adjoined to C in the learner’s grammar, as shown in (23): CP

(22)

C’

Topic C + Vfin

TP DPi Dclit. ti

T’ íP

T í’

JB[v.20020404] Prn:2/08/2005; 15:48

F: LA8105.tex / p.12 (777-834)

 The Rise of Agreement CP

(23)

C’

Topic C C + Vfin

TP Agr Ø

T’ íP

pro

T í’

In the following, I will discuss the changes leading to the structure in (23) in some more detail, focusing on the set of syntactic conditions on this kind of structural reanalysis argued for in Chapter 4 (briefly repeated in (i)–(iv) below). (i)

Structural simplification – the resulting structure must be less complex than the target structure. (ii) Preservation of argument structure – the θ-role previously assigned to the pronominal clitic must be assigned to another element after the reanalysis. (iii) Identification of feature content – the content of the newly created Agr-morpheme must be valued. (iv) Word building constraint – a bound Agr-morpheme must be able to combine with the verb prior to Vocabulary Insertion. As shown in (23) above, I assume that the reanalysis in question satisfies the restriction imposed by θ-theory (Preservation of argument structure) by assigning the subject/agent θ-role previously assigned to the clitic pronoun to a referential pro that is inserted into SpecνP. This also serves to meet the requirement of structural simplification, since the empty subject pronoun stays behind in SpecνP throughout the derivation.12 The referential pro is Case-licensed by an Agree operation initiated by the X0 -complex T+Agr (prior to movement of V+T+Agr to C). This operation also serves to value the non-interpretable φ-set of the ‘canonical’ verbal agreement morpheme which is part of to T0 .13 As already noted above, the rise of verbal agreement from former pronominal elements may in this way give rise to new pro-drop properties formerly absent in the target grammar. In short, the reanalysis of the former subject clitic as an agreement marker forced the learner to assume the presence of a referential pro in the subject position, which is the historical source of the limited pro-drop properties of present-day Bavarian (cf. Bayer 1984 on pro-drop in Bavarian; Weiß 2002 for a related proposal). The rise of referential pro-drop was presumably facilitated by the fact that Bavarian is not a strict non-pro-drop language. For example, the subject position remains empty in impersonal passives, as illustrated in (24). In other words, Bavarian

JB[v.20020404] Prn:2/08/2005; 15:48

F: LA8105.tex / p.13 (834-890)

Chapter 5. The reanalysis of C-oriented clitics 

does not require the insertion of an expletive subject pronoun in these contexts – in contrast to English, but similar to Standard German (cf. Grewendorf 1989). (24) Heit wird geoabeid. today is-pass worked (Günther Grewendorf, p.c.) Note that in a V2 language, the reanalysis in question trivially fulfills the requirement that the resulting Agr-morpheme must combine with the verb prior to Vocabulary Insertion (the Word building constraint), since subject enclitics are always adjacent to the verb in inversion contexts. Moreover, it seems that the V2 property in fact provides the only pathway to new verbal agreement morphology in languages with C-oriented clitics and basic SOV order like Bavarian: in an SOV+V2 grammar, a C-oriented clitic cannot be reanalyzed as the realization of an Agr-morpheme on C that combines via Morphological Merger with a verb that occupies a lower head position, since these heads are usually not adjacent at MS (e.g., the subject regularly intervenes between C and T). Hence, overt movement of the finite verb to C is a necessary precondition for a reanalysis that leads to the existence of verbal agreement features on C (in this way, the presence of the finite verb in C can be said to signal that C is capable of hosting an Agrmorpheme). To conclude, the rise of new agreement formatives (in C) appears to be intimately linked to the V2 property. This corresponds to the observation that crosslinguistically, complementizer agreement/Agr-on-C is a marked phenomenon which goes hand in hand with another marked syntactic property, V2. The V2 property arguably plays also an important role in extending the domain of the new endings to other verbal positions. Recall that the change in question did not immediately lead to a wholesale replacement of the old ending (i.e., the phonological exponent of Agr-on-T) by the new one. Rather, it proceeded via a stage where the new agreement ending was confined to C (as the phonological realization of Agr-onC), presumably due to the fact that embedded clauses (without V-to-C) still provided enough evidence for the old ending. Thus, initially, the exponent of Agr-on-C replaced the old agreement ending only in inversion contexts.14 The new inflection then gradually spread to other verbal positions when the learner reanalyzed the exponent of Agr-on-C as the canonical verbal agreement ending, that is, as the exponent of Agron-T. The latter reanalysis depended on the V2 property as well: only instances where the original agreement ending was replaced by the exponent of Agr-on-C could feed a possible misinterpretation of the latter as the realization of Agr-on-T. Schematically, this later development is illustrated for the change from 2pl /-t/ to /-ts/ in (25): (25) [2pl/Agr-on-C] ↔ /-ts/ → [2pl/Agr-on-T] ↔ /-ts/ (in inversion contexts) Thus, the phonological exponent of Agr-on-C is eventually reanalyzed as the exponent of the canonical agreement morpheme, Agr-on-T. It is important to note that this later development is a counterexample to the claim that grammaticalization always leads to new exponents for higher functional heads (Roberts & Roussou 2003, see Chapter 2

JB[v.20020404] Prn:2/08/2005; 15:48

F: LA8105.tex / p.14 (890-961)

 The Rise of Agreement

above). Instead, it seems that in V2 languages, the exponent of a higher (dissociated) Agr-morpheme may be reanalyzed as a marker of canonical subject-verb agreement, that is, as an exponent of a lower agreement morpheme, Agr-on-T. So far, we have not addressed the question of how the content of the newly created Agr-morpheme on C is identified, that is, how the reanalysis in question can satisfy the condition that the feature content of the resulting Agr-morpheme must be licensed (Identification of feature content). Recall that in Chapter 3, I have provided a number of arguments against an analysis in which the agreement morpheme on C is valued by accessing the subject’s φ-set via an Agree operation. Instead, I presented an analysis where Agr-on-C is inserted post-syntactically as a dissociated Agr-morpheme that is licensed under structural adjacency with an agreement morpheme on T which has been valued in the syntactic derivation (via Agree). Accordingly, I assume that the subject clitic is reanalyzed as the realization of a dissociated Agr-morpheme that is post-syntactically adjoined to C and licensed under structural adjacency with Agron-T, thereby fulfilling the condition Identification of feature content. In this way, the post-syntactic insertion of dissociated morphemes provides an alternative way of reanalyzing a former movement dependency in terms of a ‘less costly’ syntactic structure, in addition to external Merge (Roberts & Roussou 2003). The various aspects of the change in question can be summarized as follows: (26) [CP XP [C’ C+V+pronouni [TP ti ’ [T’ T [νP ti ... → [CP XP [C’ C+V+Agr [TP’ T [νP pro ... Structural simplification: overt movement replaced by an Agree-relation Preservation of argument structure: θ-role previously assigned to overt pronoun is now assigned to pro Identification of feature content: dissociated Agr-morpheme licensed under structural adjacency with Agr-on-T Word building constraint: verb is located in C (via movement) This analysis of the grammaticalization of new agreement markers in the history of Bavarian raises the question of whether the reanalysis of clitics as dissociated Agrmorphemes constitutes a necessary intermediate stage of the development of new verbal agreement formatives (i.e., exponents of Agr-on-T) in languages that already exhibit a series of verbal agreement markers. The presence of more than one Agrmorpheme should then give rise to instances of multiple agreement where the subject’s φ-set is reflected in more than one position (see also Chapter 2, Note 28, and Chapter 4, Note 17, on the development of new agreement markers from Infl-oriented clitics in Northern Italian dialects). Initial support for this hypothesis comes from the diachronic facts observed in Bavarian, where the new agreement endings started on C as exponents of a disso-

JB[v.20020404] Prn:2/08/2005; 15:48

F: LA8105.tex / p.15 (961-1010)

Chapter 5. The reanalysis of C-oriented clitics



ciated morpheme and replaced the original exponents of Agr-on-T in a subsequent development. In addition, it appears that the difference between clitics and dissociated morphemes is rather subtle and therefore easily missed during language acquisition. Note that dissociated Agr-morphemes share basic properties with clitics such as postsyntactic positioning (for post-syntactic accounts of clitic placement cf. Bonet 1991; Halpern 1992; Schütze 1994, among others). In Bavarian, the clitic-like behavior of Agr-on-C is also evident in cases where the agreement morphology does not attach to an element in C, but rather to an XP located in SpecCP. The following examples illustrate that in the absence of a filled C-head, the inflection can attach to any element that occurs in the left periphery of the clause such as DPs (27a), adjectives (27b), or adverbs (27c): (27) a.

Du soll-st song [CP [ an wäichan Schuah]-st [IP du wui-st]]]. which shoe-2sg you want-2sg you should-2sg say ‘You should say which shoe you want.’ b. [CP [ Wia oit]-ts [IP ihr/es sei-ts]] is mir wurscht. how old-2pl you are-2pl is for-me not-important ‘How old you are makes no difference to me.’ c. [CP [ Wia schnäi]-ts [IP ihr/es fahr-ts ]]! how fast-2pl you.pl drive-2pl ‘How fast you drive!’ (Bayer 1984: 235)

Note that this behavior of Agr-on-C in Bavarian seems to contrast with the properties of well-behaved inflectional affixes, which usually select for a unique host/syntactic category they attach to.15 The systematic differences between dissociated Agr-morphemes and Agr-morphemes which are present in the syntactic derivation (adjacency effects, weak morphological selection, interaction with PF-processes such as deletion or clitic placement) can be used as a diagnostics for telling apart the two types of agreement. The similarities between clitics and dissociated morphemes presumably abet a reanalysis of the former as the latter, since they blur the boundary between these elements. Thus, it seems that at least in Bavarian, the development of new agreement endings proceeded via an initial reanalysis of subject clitics as exponents of a dissociated Agr-morpheme (on C). Whether this change represents a necessary stage on the diachronic pathway toward canonical argument-predicate agreement across languages is discussed in more detail in the sections to come.

.. Developments in other German varieties In a number of German dialects spoken in minority enclaves surrounded by Italianspeaking communities, we can observe grammaticalization processes that clearly resemble the historical developments in Bavarian. In the following, I will present data from Cimbrian, a variety of Bavarian spoken in a small number of villages in the Italian Alps (southeast of Trento) and a set of Alemannic dialects spoken in Northern Italy. Without giving further arguments, I will assume that the relevant developments

JB[v.20020404] Prn:2/08/2005; 15:48

F: LA8105.tex / p.16 (1010-1087)

 The Rise of Agreement

can be analyzed on a par with the Bavarian examples discussed in the previous section (with some qualifications added where it seems necessary). Thus, the following data are taken to represent further examples of enclitics turning into instances of Agr-on-C in inversion contexts.

... Cimbrian Cimbrian is a Germanic (Bavarian-Austrian) language spoken by about 2,000 speakers in Northeast Italy in Lusern, the Sette and Tredici Comuni (Sieben and Dreizehn Gemainde, southeast of Trento), Giazza (Glietzen, Ljetzen), and Roana (Rabam). Cimbrian is related to Bavarian and shows many archaic features which date back to the time when the area in question was settled (in the 11th and 12th century). Note, however, that there are many differences between present-day Bavarian and Cimbrian concerning grammar, lexicon and pronunciation which justify a classification of Cimbrian as a separate Germanic language in its own right. The following discussion is based on the unpublished Zimbrische Gesamtgrammatik by Bruno Schweizer (1952, Vol. 5: Syntax der zimbrischen Dialekte in Oberitalien), and Tyroller (2003). Some Cimbrian varieties such as Lusern are characterized by the V2 property, that is, the subject inverts with the finite verb if a non-subject is fronted, as shown in example (28). Lusern (28) disan libar hat ar gelest, dar pua. this book has he read, the boy ‘The boy has read this book.’ (Tyroller 2003: 226) In general, however, the Cimbrian V2 dialects allow more variation concerning the placement of the finite verb, including V1 and V3 order in contexts where other German varieties require V2 order (see Schweizer 1952; Tyroller 2003 for details). In other varieties of Cimbrian, the V2 property has been (at least partially) lost, probably due to Italian influence, cf. Heller (1975) on the Sette Comuni and Schweizer (1952: 25) on the Tredici Comuni. In contrast to German, all Cimbrian dialects exhibit basic VO word order. This is illustrated by the following example from the variety spoken in Lusern: Lusern (29) I hån s as khöt, ke dar mucht machan lautrar di I have it you.pl said that you.pl must make more-fluid the malt. mortar ‘I told you that you that the mortar must be more fluid.’ (Tyroller 2003: 234) All Cimbrian varieties exhibit a series of enclitic pronouns that attach to the right of the finite verb in inversion contexts and to the complementizer (or other elements of the C-domain) in embedded clauses (cf. Schweizer 1952: 26, 225). Schweizer (1952: 4,

JB[v.20020404] Prn:2/08/2005; 15:48

F: LA8105.tex / p.17 (1087-1150)

Chapter 5. The reanalysis of C-oriented clitics 

26) observes that the enclitics enter into a close relation with the finite verb, which sometimes gives the impression of amalgamated forms: Roana (30) a. morgen vrü ge=x übar van Roan. tomorrow morning go=clit.1sg over from Roana ‘Tomorrow morning, I will cross over from Roana.’ b. ba pis=to gabest hôüte? where are=clit.2sg been today ‘Where have you been today?’ (Schweizer 1952: 26) Interestingly, the clitic pronouns that attach to the inverted finite verb can be doubled by full pronominal forms. According to Schweizer (1952: 4, 28), this should be taken to indicate that the enclitics have developed into a new form of verbal inflection which he calls Pronominalkonjugation:16 Roana (31) a. ix gan ôux kim-ex nemmear. I to you.pl come-1sg never again ‘I will never come to you again.’ b. bazt üs-te du? What do-2sg you ‘What are you doing?’ c. denne miz-er gen ear. then must-3sg go he ‘Then he must leave.’ d. hat-er gahat tzo megelan ear ox. hat-3sg had to marry he too ‘He had to marry, too.’ e. bir tü-ber nixt. we do-1pl nothing ‘We do nothing.’ Giazza (32) a. pin-i nixt ploax haute i? am-1sg not pale today I ‘Am I not pale today?’ b. has-to du nixt gahoart lautan? have-2sg you not heard ring ‘Haven’t you heard the bells ring?’ Lusern (33) dopo möchten-ber essan biar. then wish-1pl eat we (Dal Negro 2004: 179)

JB[v.20020404] Prn:2/08/2005; 15:48

F: LA8105.tex / p.18 (1150-1217)

 The Rise of Agreement

Importantly, it seems that this form of clitic doubling is restricted to non-embedded clauses (cf. Schweizer 1952: 26ff.), which suggests that the various Cimbrian dialects exhibit an early stage of the grammaticalization of new verbal agreement markers where the newly created agreement morpheme on C is still confined to matrix clauses with subject-verb inversion, that is, the environments where the reanalysis of clitics originally took place. In other words, Cimbrian exhibits another instance of the structural reanalysis schematized in (26) above. In contrast to Bavarian, however, the relevant grammaticalization process is still confined to stage (34a): V + enclitic → V+Agr/inversion contexts (reanalysis of clitics as Agr-on-C) b. extension to other elements located in the C-system such as complementizers, relative pronouns etc. c. extension to verbs in clause-final positions

(34) a.

The hypothesis that Cimbrian exhibits an early stage of the change in question is further supported by the fact that the clitics are not yet obligatory. Thus, in contrast to Bavarian (see above), clitic pronouns can still be replaced by the relevant full forms, cf. the following examples from Roana and Lusern: Roana (35) a. ix pin bolaibet segen... I am stayed to-see ‘I stayed in order to see...’ (Schweizer 1952: 18) b. han ix gavunt an hasen have I found a rabbit ‘I found a rabbit.’ (Schweizer 1952: 24) Lusern (36) a. I pin gest bachant da gånz nacht. I am stayed awake the whole night ‘I stayed awake during the whole night.’ b. Bar håm gepitet viar urn. we have waited four hours ‘We waited for four hours.’ (Tyroller 2003: 215) The variation between new enlarged forms resulting from the reanalysis of pronouns and old, non-enlarged endings presumably represents an instance of grammar competition (cf. Kroch 1989). That is, the speakers have command over two internalized grammars which are characterized by the absence vs. presence of a dissociated Agrmorpheme on C, respectively. Over time, one of the grammatical options will win out over the other. If the change proceeds in a fashion parallel to the developments in Bavarian, this might eventually lead to a grammar in which the exponent of Agr-on-C is generalized to all verbal positions.

JB[v.20020404] Prn:2/08/2005; 15:48

F: LA8105.tex / p.19 (1217-1286)

Chapter 5. The reanalysis of C-oriented clitics

... Walser dialects Hotzenköcherle (1971), Nübling (1992: 257ff.), and Dal Negro (2004) report on a number of Walser German (Alemannic) varieties spoken in Northern Italy where enclitic pronouns apparently have developed into new markers of verbal agreement. For example, in the local dialects of Rima and Rimella, the grammaticalization process in question has led to new agreement endings for 1pl (-wer/-war) and 2pl (-(ed)er/ar), which are obligatory and clearly resemble the relevant pronominal forms, cf. the following examples from Rimella (taken from Nübling 1992: 257):17 Rimella (37) a. endsch andre ber-wer ... we others carry-1pl ‘We carry...’ b. ier andru ber-eder ... you.pl others carry-1pl ‘You (pl) carry...’ Dal Negro (2004: 161) notes that similar examples of ‘doubling’ can be observed in other person/number combinations as well. However, in contrast to 1pl and 2pl forms, the enclitics are not obligatory here. Still, they presumably represent early instances of a grammaticalization process in which the enclitic is reanalyzed as an agreement morpheme on C: Rimella (38) ix tjön-ex setsu. I do-1sg sit ‘I am sitting.’ Rima (39) schù gòòn-sch. they go-3pl ‘They go.’ Apparently, the new endings developed first on highly frequent, short (often monosyllabic) verbs such as ‘do’, ‘go’, ‘have’, ‘sit’ etc.18 A related development, albeit confined to 1pl forms, took place in another Walser dialect, Bosco Gurin, spoken in Switzerland. Again, the new agreement ending for 1pl (-w) is clearly derived from the corresponding pronoun (wiär): Bosco Gurin (40) wiär schtää-w dsch morgän-dsch üf, um fyfi. we stand-1pl the.gen morning-gen up at five ‘We get up at five in the morning.’ (Dal Negro 2004: 162) Recently, Dal Negro (2004) has demonstrated that a similar development can be observed in Pomattertitsch, a Walser dialect spoken in Formazza (Italy/Piedmont) which



JB[v.20020404] Prn:2/08/2005; 15:48

F: LA8105.tex / p.20 (1286-1340)

 The Rise of Agreement

is closely related to the Walser varieties discussed above. Similar to Bavarian, Cimbrian, and the other Walser dialects, the grammaticalization process in question is apparently triggered in inversion contexts, where enclitic pronouns are reanalyzed as agreement suffixes on the verb. As a result, the ‘clitics’ are obligatory, which suggests that the ‘clitics’ are presumably better analyzed as new suffixal agreement markers. Optionally, full pronouns can be added, giving rise to instances of apparent doubling.19 Similar to the other Walser dialects, the change seems to affect highly frequent, short (often monosyllabic) verbs (sometimes called Kurzverben ‘short verbs’) first:20 Pomattertitsch (41) a. dets bin-i ich. this am-1sg I ‘This is me.’ (Dal Negro 2004: 165) b. d möter set-sch ... the mother says-3sg ‘Her mother says ...’ (Dal Negro 2004: 171) Dal Negro (2004) considers declaratives with V1 order in which the clitic attaches to the right of the clause-initial verb as further instances where the clitic has turned into a verbal agreement marker (presumably in analogy to V1-sequences in Italian), cf. the following examples taken from Dal Negro (2004: 170f.): (42) a.

bin=i deheima gsi. am=I home been ‘I have been home.’ b. chan=der no nit heirate. can=he yet not marry cannot get married yet.’ c. si=wer äso. are=we so ‘We are like this.’ d. tön=tsch in titsch zellä. do=they in German speak ‘They speak German.’

However, it is not entirely clear to me whether these examples actually can be taken as evidence that the clitic has turned into an agreement marker. It is equally possible that V1-orders of this kind represent an inherited feature, which is reminiscent of the frequent occurrence of V1-declaratives in older stages of Germanic. As Dal Negro (2004: 39) herself notes, the German speaking enclaves in Northern Italy have preserved a number of archaic traits that “in some cases date back to Old High German.” So it is quite likely that these orders in fact do not constitute an innovation, in contrast to Dal Negro’s claims.21 According to Dal Negro, the grammaticalization process in question is intimately linked to the pro-drop property. Thus, even in contexts that do not show the new

JB[v.20020404] Prn:2/08/2005; 15:48

F: LA8105.tex / p.21 (1340-1404)

Chapter 5. The reanalysis of C-oriented clitics 

agreement endings, overt pronouns can be omitted (probably the result of Italian influence), which abets a reinterpretation of enclitics as verbal agreement suffixes (with the relevant θ-role assigned to pro):22 (43) a.

tösch leiga wi ä tiful. do-2sg lie like a devil ‘You are just lying.’ b. hä tischut ds wärch. have-1sg changed the work ‘I have a new job.’ (Dal Negro 2004: 167)

To sum up, it appears that a set of Walser dialects is affected by a change very similar to that observed in Bavarian, in which enclitic pronouns are reanalyzed as verbal agreement suffixes in inversion contexts. Note, however, that the grammaticalization process that affects the Walser dialects apparently differs in two important respects from the changes in Bavarian. First, it seems that in contrast to Bavarian, the categorial reanalysis of clitics does not give rise to new pro-drop properties in the Walser dialects. Instead, it appears that the dialects in question already exhibited extensive pro-drop features prior to the change in question. This supports the idea put forward in Chapter 4 that pro-drop may either be a result of the rise of agreement morphology (as in Bavarian) or constitute a factor which promotes the reanalysis of clitic pronouns, as in the Walser dialects. Second, it seems that in Pomattertitsch, the development of Agr-on-C has not led to the rise of complementizer agreement, in contrast to Bavarian. To account for this difference, I assume that the newly created agreement morpheme on C is still restricted to the contexts where the reanalysis of clitic pronouns has taken place (i.e., inversion environments where the finite verb occupies C) and has not been generalized to all instances of C.23

.. Section summary In this section, I have argued that there is a diachronic link between complementizer agreement, pro-drop and the development of new verbal agreement markers in Bavarian. More precisely, it has been shown that the development of new verbal agreement markers from clitic pronouns could only take place in inversion contexts. This change forced the learner to assume that the subject position is occupied by referential pro, in accordance with the requirement that the reanalysis preserves the verb’s argument structure by assigning the former pronoun’s θ-role to another element. In addition, we have seen that the change in question is also consistent with the other syntactic conditions on the grammaticalization of new agreement markers proposed in Chapter 4, leading to a simplified structure in which the content of the newly created agreement morpheme is identified under structural adjacency with Agr-on-T. Note that this diachronic analysis is based on the synchronic account of complementizer agreement developed in Chapter 3, making use of the post-syntactic insertion of a dissociated Agr-morpheme. On these assumptions, the diachronic development in

JB[v.20020404] Prn:2/08/2005; 15:48

F: LA8105.tex / p.22 (1404-1459)

 The Rise of Agreement

Bavarian has been shown to proceed via a stage where the learner assumes the existence of a dissociated Agr-morpheme which is initially confined to C (Agr-on-C). In a subsequent change, the phonological exponent of Agr-on-C is reanalyzed as the exponent of the canonical subject agreement morpheme, Agr-on-T, and spreads to other verbal positions. Furthermore, it has been demonstrated that similar grammaticalization processes can also be observed in a set of Germanic dialects spoken in Northern Italy. Both Cimbrian and Walser varieties have been shown to exhibit an early stage of the change in question, where enclitics are reanalyzed as instances of Agr-on-C in inversion contexts. In contrast to Bavarian, however, the newly created agreement markers are still confined to the environments where the reanalysis originally took place. Importantly, the developments observed in this section contradict the claim that grammaticalization always proceeds in an upwards fashion, leading to a new exponent of a higher functional head, as proposed by Roberts and Roussou (2003) (see Chapter 2 above). Rather, in Bavarian the exponent of a higher functional morpheme (Agron-C) is reanalyzed as the realization of a lower functional morpheme (Agr-on-T), which seems to be a typical grammaticalization path for the rise of agreement in V2 languages. Schematically, this can be depicted as in (44), where /-x/, the exponent of X (= Agr-on-C) is reanalyzed as the exponent of Y (= Agr-on-T). XP

(44) X

XP X

YP

/-x/ Y /-y/

...

YP

/-x/ Y

...

/-x/

Present-day Bavarian still shows a residue of these diachronic processes, namely a dissociated Agr-morpheme on C, which gives rise to the phenomenon of complementizer agreement. Thus, pro-drop and complementizer agreement are reflexes of the historical development of new verbal agreement morphology in the history of Bavarian, which had to proceed via C in inversion contexts.24 In addition, the above discussion has suggested that the grammaticalization of agreement markers from clitic pronouns generally proceeds via a stage where the clitic is reanalyzed as a dissociated Agr-morpheme before it turns into a syntactic Agr-morpheme in a subsequent development. This hypothesis is illustrated by the following grammaticalization path for verbal agreement markers, which is a more fine-grained description of the relevant historical developments than the traditional model depicted in (46). (45) free pronoun → weak pronoun → clitic pronoun → dissociated Agrmorpheme → syntactic Agr-morpheme → Ø (46) free pronoun → weak pronoun → clitic pronoun → (bound) agreement marker → Ø

JB[v.20020404] Prn:2/08/2005; 15:48

F: LA8105.tex / p.23 (1459-1508)

Chapter 5. The reanalysis of C-oriented clitics 

However, although we have managed to establish a diachronic connection between complementizer agreement and pro-drop in Bavarian, the investigation of syntactic aspects has not provided an answer to the question of why complementizer agreement and pro-drop are restricted to 2nd person contexts in present-day Bavarian. In Chapter 6, it is demonstrated that this restriction is a result of morphological conditions that shaped the development of new verbal agreement morphology in the history of Bavarian. In the next section, it is shown that C-oriented subject enclitics were repeatedly reanalyzed as agreement markers in a number of Swiss Rhaeto-Romance dialects, which are also characterized by the V2 property, but differ from Bavarian with respect to the particular syntactic environments where the reanalysis of subject clitics is triggered.

. Rhaeto-Romance The Rhaeto-Romance languages are particularly interesting with respect to the present research context, since it can be shown that some of them repeatedly underwent a change in which a former enclitic is reanalyzed as an agreement ending. This section focuses on a set of Swiss Rhaeto-Romance dialects spoken in Graubünden, drawing on data from Gartner (1883, 1910), Widmer (1959), Haiman (1971, 1974), Linder (1987), Oetzel (1992), Haiman and Benincà (1992), Ebneter (1994), and Liver (1999). Among the Swiss varieties of RR (sometimes referred to as Romansh) five major dialects are usually distinguished (each with a number of local varieties): Puter (spoken in the higher Engadin), Vallader (in the lower Engadin and Münstertal valleys), Surmeiran (in Oberhalbstein, Surmeir valley, and in Albulatal valley), Surselvan (in the higher Rhin or Vorderrheintal region), and Sutselvan (in the lower Rhin or Hinterrheintal, and in Nidwald). It is demonstrated that clitics repeatedly developed into new markers of verbal agreement in the recorded history of these dialects. Interestingly, it seems that the individual grammaticalization processes are triggered in different syntactic contexts. First, we can observe a development similar to Bavarian where C-oriented clitics are ‘directly’ reanalyzed as agreement markers on C, giving rise to a limited amount of pro-drop. In addition, it seems that in the present-day dialects, a different development is underway in which new agreement formatives are created from subject enclitics that occur in clitic doubling structures.

.. A grammatical sketch of the Swiss Rhaeto-Romance dialects Before turning to the diachronic processes that led to the development of a set of new agreement markers in the Swiss RR dialects, this section briefly introduces some basic syntactic characteristics of the RR dialects under investigation, focusing on basic word order, the syntax and morphology of pronominal forms, and pro-drop properties.

JB[v.20020404] Prn:2/08/2005; 15:48

F: LA8105.tex / p.24 (1508-1586)

 The Rise of Agreement

... Word order: V2 and SVO All Swiss varieties of RR are characterized by the V2 property (cf. e.g. Haiman 1974; Linder 1987; Haiman & Benincà 1992; Oetzel 1992). Usually the subject follows the finite verb in main clauses if the clause-initial position is occupied by another XP (an adverbial, a PP, an object, a wh-phrase etc.). This is illustrated by the following examples which exemplify inversion in main declaratives and wh-questions for the five major Swiss RR dialects: Puter (47) a. Las chavras charget el aint illa chaista... the goats put he into the box... ‘He put the goats into the box...’ (Linder 1987: 17) b. A chi ho l’homin sulvedi preschanto üna giuvna? to whom has the dwarf introduced a young-girl ‘To whom did the dwarf introduce a young girl?’ (Oetzel 1992: 7) Vallader (48) a. A quella sto el imprometter da nun ir mâ plü davent. to her must he promise to never go prt again away ‘He must promise to her that he will never go away again.’ (Oetzel 1992: 28) b. Da che discuorra el tuotta pezza? of what talks he all time ‘What does he talk about all the time?’ (Ebneter 1994: 798) Surmeiran (49) a. A quests pleds suonda ena parfetga quietezza. to these words followed a complete silence ‘A complete silence followed these words.’ (Oetzel 1992: 29) b. Tge fo pia ti onda? what makes then your aunt ‘What is your aunt doing?’ (Gartner 1910: 17) Surselvan (50) a. La brev ha la mumma scret. the letter has the mother written ‘The mother wrote the letter.’ (Liver 1999: 146) b. Cura eis ella morta? when is she died ‘When has she died?’ (Liver 1999: 145) Sutselvan (51) a. A que num porta igl cratsch ànc oz. and this name bears the baby-of-the-family still today ‘And the baby of the family still bears this name today.’ (Oetzel 1992: 29)

JB[v.20020404] Prn:2/08/2005; 15:48

F: LA8105.tex / p.25 (1586-1674)

Chapter 5. The reanalysis of C-oriented clitics

b. Cura va’la la Nona? when goes-3sg.fem the grandma ‘When will grandma leave?’ (Linder 1987: 155) However, note that the Swiss RR dialects are actually less strict with respect to the V2 constraint than the neighboring Germanic V2 languages.25 Oetzel (1992: 17ff.) observes a number of examples with non-inverted order in written records of Puter, Vallader and, in particular, Surmeiran, cf. the following examples from Vallader and Surmeiran:26 Vallader (52) In quel istess mumaint il trid uorsin as transmüdet in ün in this moment the ugly bear refl changed into a bel prinz. beautiful prince ‘In this moment, the ugly bear changed into a handsome prince.’ Surmeiran (53) La seira anturn las nov Tina sa prepara per sorteir. the evening against the nine Tina refl prepares for leaving ‘In the evening around nine, Tina gets ready for going out.’ For the time being, I assume that V2 orders result from the finite verb moving to C, with the clause-initial XP occupying the specifier of CP (see Section 5.3.4.1 for some modifications). Note that the V2 property appears to be restricted to root clauses in the Swiss RR dialects, in contrast to Icelandic or Yiddish, for example. In embedded clauses, the finite verb apparently moves out of the VP into a position in the inflectional domain. This is suggested by following examples where the finite verb precedes adverbials or the sentential negation:27 Puter (54) ch=i gnit dandettamaing ün ferm temporal... when=it came suddenly a great storm ‘When suddenly, a great storm came’ (Linder 1987: 136) Vallader (55) cha il tschêl vuless finalmantig as sclerir that the sky would finally refl clear ‘That the sky would finally clear up’ (Linder 1987: 35) Surmeiran (56) ci tö lesti óssa ün marus that you like-subjnc now a lover ‘That you would like to have a lover now’ (Linder 1987: 59) Sutselvan (57) perquei ca la furtga era betga daliensch since the gallows was not far-away ‘Since the gallows was not far away’ (Linder 1987: 162)



JB[v.20020404] Prn:2/08/2005; 15:48

F: LA8105.tex / p.26 (1674-1778)

 The Rise of Agreement

Table 6. Verbal agreement (present indicative) in five Swiss RR dialects.

1sg 2sg 3sg 1pl 2pl 3pl

Puter

Vallader

Surmeiran

Surselvan

Sutselvan

-Ø -ast -a -ains -ais -an

-Ø -ast -a -ain(a) -aivat -an

-Ø -as -a, -e -(g)n -es, -as -an

-6l -as -a -in, -ein -is, -eis -an

-Ø -(a)s -a -(g)n -(e)s, -(a)s -an

Moreover, under the assumption that there is a correlation between rich verbal inflection and overt verb movement (Kosmeijer 1986; Holmberg & Platzack 1988, 1991, 1995; Roberts 1993a; Bobaljik 1995; Rohrbacher 1999; Vikner 1994, 1995, 1997), the Swiss RR dialects are actually expected to exhibit verb movement into the inflectional domain. Table 6 illustrates the agreement endings (present indicative) found in the RR dialects under investigation.28 If we take a look at the paradigms listed in Table 6, it becomes clear that the verbal agreement morphology of the Swiss RR dialects qualifies as ‘rich’ under all common criteria for ‘richness’. Apart from Surmeiran and Sutselvan, where 2sg and 2pl are (at least partially) homophonous, the dialects show a fully distinctive agreement paradigm, which is expected to correlate with overt verb movement. For the sake of concreteness, I assume that the functional head targeted by verb movement is T in embedded clauses. Finally, the fact that the subject usually precedes the finite verb in embedded clause suggests that the subject occupies SpecTP in the overt syntax (which is standardly attributed to the presence of an EPP feature in T). In contrast to the surrounding German dialects, the Swiss RR varieties exhibit basic VO word order (which is a characteristic of RR in general). Thus, the non-finite verb is followed by its complements in the unmarked word order. This is illustrated for DP and PP complements with the following examples from Puter and Vallader taken from Oetzel (1992: 7ff.):29 Puter (58) L’homin sulvedi ho preschanto [üna giuvna] [al raig]. the dwarf has introduced a young-girl to-the king ‘The dwarf has introduced a young girl to the king.’ Vallader (59) Ed eu n’ha vis [davant mai] [ad ün pitschen homet]. and I have seen before me at a little man ‘And I saw a little man in front of me.’ Note that these examples exemplify merely the basic word order of these dialects. In addition, there is a variety of further ordering possibilities which can only be hinted at in the present discussion (for an overview and some discussion cf. Linder 1987 and in particular Oetzel 1992). As an illustration, consider the following embedded clauses.

JB[v.20020404] Prn:2/08/2005; 15:48

F: LA8105.tex / p.27 (1778-1858)

Chapter 5. The reanalysis of C-oriented clitics 

In (a) a PP has been scrambled to the left of the subject, while in (b), a quantified object occurs in preverbal position:30 (60) a.

chi [cun quel] els nun stetan plü... that with this they not stay any-longer ‘That they will not stay with this one any longer’ (Puter; Linder 1987: 31) b. Mò che ans güd’igl cur nus [quegl tutt] cretein? but what us help-it if we that all believe ‘But how does it help us if we believe all that?’ (Sutselvan; Linder 1987: 182)

The next section focuses on the inventory and distribution of personal pronouns, a topic which is particularly relevant for the diachronic analyses developed later (recall that pronouns commonly provide the lexical source for the grammaticalization of agreement markers).

... The distribution of pronominal forms and pro-drop The Swiss RR dialects exhibit three series of pronouns, one set of full pronouns that may bear stress and two sets of atonic, reduced pronouns: (i) proclitics that attach to the left of the finite verb in V2 clauses, and (ii) a set of enclitic forms which resemble the C-oriented clitics of Germanic (cf. Linder 1987). Similar to Bavarian, all Swiss RR dialects are characterized by the lack of enclitic 2nd person pronouns (cf. Haiman 1971; Linder 1987). In Puter and Surmeiran, we can observe a proclitic (Puter a-, Surmeiran i-) which can be used for several person/number combinations (1sg, 3sg.neut, and all plural forms).31 Table 5 illustrates the pronominal system found in Puter:32 Table 7. Pronominal forms (nominative) in Puter (Haiman 1971; Linder 1987; Liver 1999).

1sg 2sg 3sg.masc 3sg.fem 3sg.neut 1pl 2pl 3pl.masc 3pl.fem

Full pronoun

Proclitic

Enclitic

eau tü el ella ad nus vus els ellas

at(ü-) (e)l(el)laaaaaa-

-i, a Ø -l -la -a -a Ø -e -e

Recall that in Chapter 4 above, it has been claimed that the presence of clitic pronouns is a prerequisite for the grammaticalization of new agreement markers. In addition, it appears that the Swiss RR dialects exhibit (limited) pro-drop, another feature which by assumption plays an important role in the development of new verbal agreement

JB[v.20020404] Prn:2/08/2005; 15:48

F: LA8105.tex / p.28 (1858-1930)

 The Rise of Agreement

morphology, since it provides a means to preserve the verb’s argument structure if a pronominal argument is reanalyzed as an agreement marker (see Chapter 4 for discussion). Linder (1987) lists a number of examples where a subject pronoun is omitted. However, he notes that there is a significant difference between proclitic and enclitic pronouns with respect to pro-drop: the omission of proclitics (which appear in clause-initial position) is quite rare and is apparently the result of Italian influence or an instance of topic-drop similar to German. Linder (p. 21) further notes that apart from cases of topic-drop, the omission of proclitics is a phenomenon which is more or less confined to the written language of the 19th century and no longer accepted by present-day speakers. The following examples from Puter and Vallader illustrate omission of proclitics, leading to V1 order: (61) Puter Giavüschessans da preschenter a noss lectuors ils tips cracteristics wish-subjnc-1pl to present to our readers the typical characters da noss’istoria engiadinaisa. of our-story Engadinian ‘(We) would like to present our readers the typical characters of our Engadinian story.’ (Chalender Ladin 1911; Linder 1987: 24) (62) Vallader Il gigant es gni our e àn fat la partida... the giant is come out and have-3pl made the game ‘The giant came out and (they) made a game...’ (Linder 1987: 31) In contrast, enclitic pronouns can be more freely omitted (in all person/number combinations apart from 3sg, where no examples with pro-drop are attested, cf. Linder 1987: 34), even in the present-day language.33 2nd person pronouns play a special role in that they are normally dropped in inversion contexts in all Swiss RR dialects (cf. e.g. Linder 1987: 53). This is illustrated by the following examples from Vallader and Puter for 2sg, 2pl, respectively (recall that no enclitics are available for 2nd person forms). (63) Vallader Hoz est vaira nervus, Paul. today are really nervous Paul ‘Today, (you) are really nervous, Paul.’ (Linder 1987: 35) (64) Puter Cu fais que? how make-2pl that ‘How do (you) make that?’ (Linder 1987: 35) So far, I have illustrated the essential word order properties of the Swiss RR dialects, which are characterized by a V2 constraint that operates on a basic SVO syntax. In addition, I have shown that the Swiss RR dialects exhibit an elaborate series of clitic pronouns and thus satisfy a necessary condition for the development of new verbal

JB[v.20020404] Prn:2/08/2005; 15:48

F: LA8105.tex / p.29 (1930-1991)

Chapter 5. The reanalysis of C-oriented clitics 

agreement markers from former pronouns. In addition, all dialects apparently exhibit a limited amount of referential pro-drop, which still differs significantly from the kind of general pro-drop found in Italian, for example. Recall, however, that it has been claimed above that the availability of pro-drop (even if limited) may facilitate the grammaticalization of agreement markers by providing an alternative means to preserve the predicate’s argument structure if a pronoun is reanalyzed as an agreement morpheme. In the following, it is shown how the particular syntactic properties of the Swiss RR dialects presented so far shaped the development of new verbal agreement markers in the (recent) history of these languages.

.. Earlier grammaticalization processes affecting enclitic pronouns As already noted above, we can observe a set of distinct grammaticalization processes in the history of the Swiss RR dialects under investigation. This section focuses on developments that took place in previous historical stages. It can be shown that the agreement morphology found in the present-day Swiss RR dialects has been at least partially shaped by the reanalysis of former enclitic subject pronouns as enlargements of the existing verbal agreement endings, similar to the facts observed in Bavarian. In the following, this is illustrated for 1st and 2nd person forms.

... 1st person forms The Swiss RR dialects are characterized by the loss of the original Romance 1sg agreement suffix -o, similar to many Northern Italian dialects. In some Italian RhaetoRomance dialects, this loss was compensated by the development of new markers for 1sg: in present-day Ampezzan we again find -o, presumably under influence from the nearby Italian dialects (e.g., Venetian). In some varieties of Friulian, a new form for 1sg (-e ∼ -i) can be observed.34 A rather recent development took place in Surselvan, which exhibits 1sg -6l, while Old Surselvan showed consistently -Ø in this context. However, from 1700 on, -6l appeared as a new ending for 1sg, which is still used in the present-day language. The origin of the new ending is still unclear (cf. Haiman & Benincà 1992: 93f. for some discussion). In contrast, the origin of the present-day 1pl forms is more transparent. In Puter, a final /-s/ has been added to the older ending 1pl -(a)in.35 It is commonly assumed that the new agreement formative 1pl -ains developed via a reanalysis of an enclitic 1pl (no)s in inversion contexts (Gartner 1883; Widmer 1959; Linder 1987; Haiman & Benincà 1992). Thus, similar to the Bavarian examples discussed above, the V2 property seemingly provided an appropriate syntactic context for the development of new agreement endings from former clitic pronouns. In one of the earliest Puter records, a translation of the New Testament by Giachem Bifrun (dating from 1560), the new enlarged form is still confined to inversion contexts, while the older ending -(a)in is consistently found when the 1pl pronoun occupies a preverbal position (cf. Linder 1987: 70ff.).36 This is illustrated by the following example:

JB[v.20020404] Prn:2/08/2005; 15:48

F: LA8105.tex / p.30 (1991-2055)

 The Rise of Agreement

(65) Mu hauiand la spaisa, & a quellas chioses che nus ns’cuurin but having the food and some things that we us=put.on-1pl cun quaistes chioses dains esser cuntains. with these things shall-1pl be content ‘But having food and clothing, with these things we shall be content.’ (Bifrun, Timothy I. 6,8; Linder 1987: 70) In (65), the old ending -(a)in is still found on the verb cuurin which occurs in noninverted position inside a relative clause, while the enlarged form dains is used in a V2 clause with inversion, in which the clause-initial position is occupied by the PP cun quaistes chioses. Note that this distribution is identical to the facts observed in connection with the grammaticalization of new agreement markers in the history Bavarian: the new forms emerge first on verbs that have moved to C and spread to other verbal positions in a subsequent development. In the late 16th/early 17th century, final -s is already much more common. Linder (1987: 71) gives the following example: (66) Il Segner vulains luder, cha nus poassens a chiesa turner. the Lord want-1pl praise that we can-1pl to home turn ‘We want to praise the Lord, so that we can return to home.’ Interestingly, in some dialects the 1pl subject enclitic a has undergone a similar development, eventually giving rise to a second cycle of cliticization and reanalysis in inversion contexts (Linder 1987). In spoken Surselvan, Sutselvan, and Surmeiran, the clitic shows up obligatorily on oxytonic verbs (i.e., verbs that are stressed on the last syllable) in inversion contexts, cf. the following examples from Surmeiran: (67) a.

Davent da lò muntag-ns-a sen igls Corns greischs... from there climb-1pl-1pl onto the peaks gray ‘From there, we climb the Gray Peaks.’ b. Ea, alloura lai-ns-a parteir. yes then want-1pl-1pl leave ‘Yes, then we want to leave.’ (Linder 1987: 76)

However, -a cannot be added to verbs that receive stress on the penultimate syllable (paroxytonic verbs). Instead, the form with final -s is retained: (68) Finalmantg ischans anc ballos cugl auto segl Klausen... finally are-1pl still driven with-the car on-the Klausen ‘Finally, we drove with the car onto the Klausen...’ (Linder 1987: 77) This distribution is apparently governed by a stress rule which is characteristic of the Swiss RR dialects. This rule, called ‘penultimate stress target’ in Haiman (1971) and ‘three-syllable rule’ in Haiman and Benincà (1992), forbids antepenultimate stress on verbs. If the verb receives stress on the final syllable (as e.g. munÁtagns ‘climb-1pl’), final -a can be added to the existing verbal inflection, since the resulting form complies with the ‘penultimate stress target’, receiving stress on the penultimate syllable. In contrast,

JB[v.20020404] Prn:2/08/2005; 15:48

F: LA8105.tex / p.31 (2055-2115)

Chapter 5. The reanalysis of C-oriented clitics 

with verbs such as Áischans ‘are-1pl’, which already receive stress on the penultimate syllable, the integration of -a into the verbal inflection is impossible, since this would lead to a verb with a stressed antepenultimate (therefore *Áischans+a). Linder (1987: 76) shows that in some dialects such as Pontresina (a variety of Puter), the new ending -insa is used independent of word order: final -a is completely integrated into the verbal inflection and no longer confined to inversion contexts. That is, according to Linder the forms in (69) can also show up in embedded clauses: (69) a.

nus we b. nus we

curr-Áinsa run-1pl gia-Áinsa go-1pl

Thus, -(a)insa has been generalized as a new marker of 1pl verbal agreement. Again, these facts clearly show that the new agreement formatives arose first in inversion contexts and spread to other verb positions in a subsequent development. However, similar to the distribution of 1pl -a in Surmeiran (see above), the former clitic has only been added to verbs that do not bear stress on the penultimate syllable. Elsewhere, the old ending (-(a)ins) has been maintained: (70) a.

ad we b. a we

Áes-ans(-*a) are durm-Áivans(-*a) were-sleeping

To summarize, in the Swiss RR dialects under investigation, the possibility of reinterpreting a subject enclitic as a verbal agreement suffix is dependent on two factors: first, the reanalysis is only possible in inversion contexts. Hence, the V2 property plays an important role in the change in question, similar to the Bavarian data discussed above. In addition, the reanalysis depends on the syllable structure (and stress pattern) of a given verb and can only be carried out if the resulting agreement ending does not lead to a stress pattern where stress falls on the antepenultimate syllable. In the next section, it is shown that quite similar observations can be made with respect to 2nd person forms.

... 2nd person forms Similar to many German varieties, Puter, Vallader and Surmeiran have replaced the original verbal agreement ending 2sg /-s/ by the enlarged form /-st/.37 This change is commonly attributed to the reanalysis of the 2sg clitic -t(i) in inversion contexts (cf. Widmer 1959; Linder 1987; Haiman & Benincà 1992). Widmer (1959: 64f.) shows that the development of the new ending /-st/ proceeded via the following three historical stages. In stage 1, the weak pronoun follows the finite verb, which still carries the old ending /-s/ (inversion contexts):

JB[v.20020404] Prn:2/08/2005; 15:48

F: LA8105.tex / p.32 (2115-2175)

 The Rise of Agreement

(71) Or tutta quest na pôs ti provar il teu fagg. from all this not can you prove the your case ‘From all this, you cannot prove your case.’ (Nauli, 46; Widmer 1959: 64) In Stage 2, the progress of the relevant grammaticalization process is reflected by the fact that the inverted verb and the enclitic tü are written as one word: (72) Mo qui sarestü leid... but here will.be-you sad ‘But here, you will be sad...’ (Wietzel, Präf., 32; Widmer 1959: 64) Finally, in stage 3, the enclitic has been reanalyzed as part of the verbal agreement morphology and can be accompanied by a full pronoun. In addition, the new ending can appear on finite verbs in embedded clauses. This indicates that the clitic has developed into a marker of verbal agreement, cf. the following dialogue taken from Widmer (1959): (73) A: Che tscherchi-ast tü, juwen, in quaista strêda? What search-2sg you, young man in this street ‘What are you looking for?’ B: Meis frars veng eau à tscharchiêr, s tü m’sav-est my brothers come I to search if you me=know-2sg intraguidêr. to-guide ‘I come to search my brothers, if you know how to guide me (to them).’ (Travers, 4; Widmer 1959: 65) In one of the oldest Swiss RR text, the bible translation by Bifrun (dating from 1560), final /-t/ is still very frequently absent in subject-initial clauses, but consistently present in inverted word order. Moreover, even at this very early stage there are a few examples where the enlargement /-t/ co-occurs with an additional 2sg pronoun tü and thus cannot be analyzed as an enclitic (Linder 1987: 54, fn. 42). This resembles the present-day usage (which according to Haiman & Benincà 1992: 95 was already established around 1700), where the agreement ending /-st/ has been generalized to all contexts (examples taken from Haiman & Benincà 1992: 94f.): (74) a.

tü vaes you go (John 14: 5) b. tü nu pous you not can (John 13: 36)

(75) innua vaest tü? where goest thou (John 13: 36)

JB[v.20020404] Prn:2/08/2005; 15:48

F: LA8105.tex / p.33 (2175-2252)

Chapter 5. The reanalysis of C-oriented clitics 

Again, these observations emphasize the important role of the V2 property for the reanalysis of C-oriented clitics, suggesting that in the Swiss RR dialects, the historical development of 2sg /-st/ proceeded along similar lines as proposed for Bavarian (cf. (11) above). Interestingly, it can be observed that the RR dialects that developed 2sg /-st/ share another characteristic: as noted above, there are usually no overt 2sg subject enclitics present (full forms may be used when they receive stress). According to Linder (1987), the present-day Swiss RR dialects in fact lack 2nd person enclitics completely. The discussion of Bavarian in Section 5.2 above suggests a diachronic explanation of this fact (cf. Linder 1987: 53ff. for a related idea). More precisely, it seems reasonable to suppose that similar to present-day Bavarian, the historical development of the agreement formative /-st/ led to partial pro-drop for 2sg contexts in RR, due to the ‘conversion’ of former enclitics into verbal agreement suffixes.38 According to Meyer-Lübke (1894) and Linder (1987: 58), the RR dialects spoken in Graubünden show a related development for the agreement ending 2pl present indicative (e.g., chant-ais ‘sing-2pl’), suggesting that final /-s/ in the present-day suffix -(a)is developed via cliticization of the 2pl pronoun (vo)s in inversion contexts.39 Similar to 2sg, I assume that this development is the historical source of pro-drop in 2pl contexts.40 We can therefore conclude that the Swiss RR dialects exhibit similar historical developments as Bavarian. Again, the V2 property provided the syntactic context for the reanalysis of C-oriented enclitics as Agr-morphemes. Without giving further arguments, I assume that these developments proceeded more or less completely analogous to the changes that affected Bavarian (see (26) above). In other words, C-oriented enclitics were reanalyzed as dissociated Agr-morphemes the content of which is licensed under structural adjacency with a valued Agr-morpheme on T (which by assumption also hosts the finite verb in the overt syntax in RR, see above). This reanalysis depends on the V2 property, which brings about adjacency between the clitic and the verb and thereby licenses a reanalysis of the former as part of the verbal inflection, in line with the Word building constraint. The new agreement formatives were initially confined to C and spread later to other verbal positions as a result of a change in which the phonological exponent of Agr-on-C replaced the existing exponent of Agr-on-T. As a by-product of this change, the Swiss RR dialects developed limited pro-drop properties due to the fact that the reanalysis of enclitic pronouns forced the learner to posit the existence of a referential pro in the contexts where the reanalysis took place. On the assumption that the empty pronoun is accessed by an Agree operation and thus does not move to SpecTP, this change is also consistent with the requirement of structural simplification:41 (76) [CP XP [C’ C+V+pronouni [TP ti ’ [T’ T [νP ti ... → [CP XP [C’ C+V+Agr [TP’ T [νP pro ... In the next section, I discuss a more recent development in which the grammaticalization of new agreement formatives proceeds from clitic doubling structures.

JB[v.20020404] Prn:2/08/2005; 15:48

F: LA8105.tex / p.34 (2252-2304)

 The Rise of Agreement

.. Clitic doubling and the rise of agreement In a number of Rhaeto-Romance (RR) dialects, we can observe subject clitic doubling, the properties of which can be taken to indicate that the clitic elements in these contexts undergo a grammaticalization process leading to new (verbal) agreement markers.42 The change in question exclusively affects C-oriented subject enclitics in inverted position, similar to the data discussed so far. Again, it seems that the development of new verbal agreement morphology proceeds via a transitional stage where the new agreement morphemes are confined to the C-head, suggesting that the V2 property plays an important role in the reanalysis of second position clitics (cf. Haiman 1991 for a similar claim). Crucially, however, the present development differs from the phenomena discussed above in that it proceeds from a clitic doubling configuration where an additional DP is present. This DP turns into the overt subject of the clause when the clitic is reanalyzed as a marker of verbal agreement. While the Swiss varieties of RR show a relatively uniform behavior with respect to the basic syntactic properties illustrated above, the individual dialects exhibit some important differences concerning the availability of subject clitic doubling. In the remainder of this section, I will focus on these differences, arguing that they represent different stages of a new wave of enclitic pronouns developing into agreement formatives. In all RR dialects apart from Surselvan (which lacks subject enclitics, Linder 1987: 146), enclitic subject pronouns can be reinforced by full DPs or pronouns, which then usually receive stress (cf. Linder 1987: 146), giving rise to instances of subject clitic doubling. In principle, doubling is possible in all person/number combinations where a subject enclitic is available, that is, 1sg, 1pl and 3sg, 3pl. Examples with 2nd person subjects are not attested, since the dialects in question appear to lack 2nd person subject enclitics (possibly due to the fact that they have been reanalyzed as agreement markers in an earlier development, see above). Moreover, with 3rd person subjects, clitic doubling occurs much more frequently with full DP subjects than with full tonic pronouns.43 It appears that doubling is confined to inversion contexts where both the C-oriented enclitic and the full subject follow the finite verb occupying the C-head (see Section 5.3.4.1 for some discussion). The additional full nominal is often adjacent to the clitic (in particular with 1st person forms), but may also appear in a position further to the right of the clause (at least in Puter, Vallader and Surmeiran). The construction is optional in Puter and Vallader and occurs quite frequently in Surmeiran. The following data illustrate (optional) subject clitic doubling in Puter, Vallader and Surmeiran (examples taken from Linder 1987: 147ff.; the clitic and the corresponding full form are marked by boldface):44 Puter (77) a. Da’ls spisantêr he=ia eau grand dallett. them to-feed have=clit.1sg I great pleasure ‘I have great pleasure in entertaining them to a meal.’

JB[v.20020404] Prn:2/08/2005; 15:48

F: LA8105.tex / p.35 (2304-2370)

Chapter 5. The reanalysis of C-oriented clitics

b. Et per che giain=s êr nus in münchia hura à priuel? and why go=clit.1pl also we in jeopardy hour every ‘And why do we also stand (lit. ‘go’) in jeopardy every hour?’ c. Co haun=e ells crida tuotts duos. then have=clit.3pl they cried all both ‘Then, they both cried.’ Vallader45 (78) a. Alura à=l dit il figl: ... then has-clit.3sg.masc said the son ‘Then, the son said: ...’ b. Che effet ha=la gnü aint il pövel which effect has-clit.3sg.fem done in the people la nouva predgia? the new sermon ‘Which effect had the new sermon on the people?’ c. Cur cha’ls üns rivainan, surdaivn=a ils oters ad when that=the ones arrived handed-over=clit.3pl the others to els la scossa e partivan. them the herd and left ‘When they arrived, the others handed over the herd and left.’ Surmeiran (79) a. Schi te ist cuntaint, vign=a ia avant mezde... if you are content come=clit.1sg I before noon ‘If that’s fine with you, I will come before noon.’ b. Lagns=a nous dus betg eir dumang ainten en’otra want=clit.1pl we two go also tomorrow in another vischnanca a messa? village to mass ‘Should we go to mass in another village tomorrow?’ c. Chel’idea veve=l gia igl uestg Ziegler sez. this idea had=clit.3sg.masc had the bishop Ziegler himself ‘Bishop Ziegler himself had had this idea.’ d. Par tema e=lla la femna curoida ancheunter Zorten. for fear is=clit.3sg.fem the woman run to Zorten ‘Because she was frightened, the woman ran to Zorten.’ e. Dantant èn=igl rivos igls bernes e turitges. meanwhile are=clit.3pl arrived the Bernese and Zurich ‘Meanwhile, the Bernese and Zurich people have arrived.’ As briefly mentioned above, these instances of subject clitic doubling are restricted to inversion contexts where both the C-oriented enclitic and the full form follow the finite verb, which occupies C. In contrast, proclitics apparently do not allow doubling, that



JB[v.20020404] Prn:2/08/2005; 15:48

F: LA8105.tex / p.36 (2370-2441)

 The Rise of Agreement

is, a configuration such as *proclitic-finite verb-full DP is not attested in the data sample available to me. In addition, Haiman and Benincà (1992: 192) explicitly state that doubling is possible only in inverted order. Furthermore, full pronouns/DPs and weak forms/proclitics are in complementary distribution in clause-initial position. These restrictions can be summarized as follows: (80) Ordering restrictions on clitic doubling in the Swiss RR dialects a. *full DP-finite verb-enclitic... b. *proclitic-finite verb-full DP... c. *full DP-proclitic-finite verb... Furthermore, in Puter, Vallader, and Surmeiran, doubling exhibits a definiteness effect: the full DP must be definite (or specific), that is, sentences such as ‘yesterday arrivedhe a man’ are not possible (cf. the examples above). Together with the observation that the full nominal often receives stress, this can be taken to indicate that in Puter, Vallader, and Surmeiran, clitic doubling fulfills certain stylistic (or, rather, discourse) functions in which a full element is added to reinforce an enclitic pronoun for reasons of emphasis. Sutselvan shows a form of clitic doubling the properties of which differ from the kind of (optional) doubling found in the other Swiss RR dialects. In Sutselvan, clitic doubling appears to be a much more common, almost obligatory phenomenon that has lost its function as a stylistically marked structural option. For example, in contrast to the other dialects, the full nominal is not added for reasons of emphasis and does not receive stress (cf. Linder 1987: 150). Relevant example sentences from Sutselvan are shown in the following.46 (81) 1st person singular Egn da quels lev-i ear jou. one of those wanted-1sg also I ‘I also wanted one of those.’ (Sutselvan; Linder 1987: 148) (82) 1st person plural Ascheia vain-sa nus arviart igl mulegn... so have-1pl we unlocked the mill ‘So we have unlocked the mill...’ (Sutselvan, Linder 1987: 149) (83) 3rd person singular masculine Igl fetschi preaschas, â-l el getg. it is urgent has-3sg.masc he said ‘He said it’s urgent.’ (Sutselvan; Linder 1987: 153) (84) 3rd person singular feminine Cunquegl c’igl eara november, vev-la la scola antschiat. since it was November had-3sg.fem the school begun ‘Since it was November, the school had begun.’ (Sutselvan; Linder 1987: 155)

JB[v.20020404] Prn:2/08/2005; 15:48

F: LA8105.tex / p.37 (2441-2494)

Chapter 5. The reanalysis of C-oriented clitics 

(85) 3rd person plural Natiral vev-in las matàns radetg sei mailenders. of course had-3pl the girls brought up Milans ‘Of course, the girls had brought up some Milans [pastries].’ (Sutselvan; Linder 1987: 161) Note that the old agreement ending is replaced by the clitic in the examples (84) and (85). In other words, subject verb agreement is solely marked by the ‘clitic’ in these sentences. This already suggests that the former enclitics have developed into some form of agreement marking in Sutselvan. A further important difference between Sutselvan and the other Swiss RR dialects concerns the definiteness restriction, which requires the full nominal to be definite or specific in Puter, Vallader and Surmeiran (see above). In contrast, no such restriction can be observed in Sutselvan, where doubling has spread to non-definite nominals as well: (86) a.

Qua han-i schon blears via da tutta sorts... there have-3pl already many seen all kind of things ‘Many people have already seen all kind of things there...’ b. Mo igl lungatg da la dunnetta san-i nigns. but the language of the little-woman knows-3pl no-one-pl ‘But nobody knows the language of the little woman.’ (Sutselvan; Linder 1987: 162)

Under the assumption that the definiteness effect is a restriction imposed on clitic doubling by its discourse function (similar to the well-known definiteness/specificity restriction on topics, cf. Givón 1976; Grewendorf 2002), these facts again suggest that the doubling construction has lost a great deal of its stylistic force in Sutselvan. Furthermore, recall that the loss of definiteness/specificity effects is often understood as an indication that the clitic has evolved into an agreement marking element (cf. Uriagereka 1995). If taken together, the almost obligatory status of clitic doubling, the fact that it is not restricted to stressed and definite/specific full nominals, and the observation that the clitics can replace the original agreement morphology in V2 contexts show rather conclusively that the clitic has developed into a new marker of verbal agreement in Sutselvan (cf. Linder 1987: 162; Oetzel 1992: 49, 51; cf. Haiman & Benincà 1992: 197ff. for similar conclusions concerning RR varieties spoken in Northern Italy).

.. The reanalysis of emphatic doubling structures In this section, I present an analysis of the developments affecting Sutselvan that is based on the idea that new forms of agreement may result from a (formerly stylistic) strategy in which a full DP/tonic pronoun is added to reinforce a phonologically defective clitic, leading to clitic doubling. In the course of time, the construction loses its stylistic force and the originally reinforcing element is eventually reanalyzed as the

JB[v.20020404] Prn:2/08/2005; 15:48

F: LA8105.tex / p.38 (2494-2539)

 The Rise of Agreement

‘real’ argument, whereas the former clitic is reinterpreted as an (verbal) agreement marker.47 Before we can explore this scenario further, let me introduce a set of theoretical assumptions that the present analysis is based on, focusing on the structure of clitic doubling in the Swiss RR dialects under consideration. As noted above, the Swiss RR dialects are characterized by the V2 property. In addition, we have seen that they are characterized by a basic VO order and that the finite verb presumably occupies T in the overt syntax. With respect to the C-oriented enclitics found in these dialects, I suggest that they should be analyzed as D-heads. Following Kayne (1994: 61), Uriagereka (1995: 81), and Torrego (1995, 2002), I assume that the phrase structural representation of pronominal clitics is as in (87). (87)

DP D

NP

clitic

pro

In other words, a pronominal clitic enters the syntactic derivation as the determiner of a phonetically empty NP which occupies the clitic’s complement position (and presumably acts as the ‘real’ argument in this configuration). In the RR dialects that exhibit optional clitic doubling, full nominals may be added for reasons of emphasis, as a reinforcement of the enclitic pronoun which cannot bear stress. I assume that this state of affairs should be modeled by a structure where the clitic D-head selects for a reinforcing full nominal (henceforth called the ‘double’) in its specifier (Uriagereka 1995; Kayne 2002). The two elements are then merged together in a ‘big DP’:48 (88)

DP DP double

D’ D

NP

clitic

pro

This complex DP is base-generated in SpecνP, where it receives the θ-role for the external argument (which can be assumed to percolate to the elements contained in the big DP). Subsequently, the big DP moves to SpecTP for reasons of Case and EPP checking. From there, the clitic adjoins to C (either at MS/PF or in the overt syntax), similar to the C-oriented clitics found in the Germanic languages (e.g., German). Accordingly, optional clitic doubling in the present-day Swiss RR dialects (i.e., Puter, Vallader and Surmeiran) is analyzed as resulting from the following structure:

JB[v.20020404] Prn:2/08/2005; 15:48

F: LA8105.tex / p.39 (2539-2580)

Chapter 5. The reanalysis of C-oriented clitics  CP

(89) XP

C’

Vfin + C

TP T’

DPi DP double

D’

D clitic

íP

T NP pro

í’

ti í

VP

It seems reasonable to assume that (89) also represents the structure of clitic doubling in a historical stage of Sutselvan prior to the changes that affected this construction and the status of clitic pronouns in this variety of RR. The current status of clitic doubling in Sutselvan (doubling is obligatory, the double is not necessarily emphasized and does not obey the definiteness restriction observed in the other dialects) suggests that at some point in the (recent) history of Sutselvan, the doubling strategy lost its stylistic force (presumably due to an over-use). As a consequence, the doubling structure (which is associated with a certain stylistic effect) failed to be cued during language acquisition, giving way to a somewhat simpler structure:49 CP

(90) XP

C’ C

Vfin + C

TP Agr

DPi

T’ íP

T

í’

ti í

VP

In (90), the former clitic has been reanalyzed as an Agr-morpheme on C. The former double assumes the argument role previously assigned to the big DP. On the syntactic surface, this change gives the impression of obligatory ‘clitic doubling’. In the following, I will take a closer look at the factors that were involved in the categorial reanalysis of clitics, paying special attention to the set of syntactic conditions on the rise of new agreement formatives proposed above.50

JB[v.20020404] Prn:2/08/2005; 15:48

F: LA8105.tex / p.40 (2580-2650)

 The Rise of Agreement

Similar to the changes affecting Bavarian and earlier stages of Swiss RR, adjacency between the C-oriented clitic and the finite verb is obligatory in inversion contexts (where the reanalysis took place), so no problem arises with respect to the Word building constraint. The resulting Agr-morpheme on C attaches to the finite verb as a result of syntactic V-to-C movement, satisfying the requirement that the two elements combine prior to Vocabulary Insertion. In other words, the V2 property again provides a syntactic environment where new agreement formatives may develop. As noted above, the recent developments in Sutselvan differ from the changes that affected Bavarian and earlier stages of RR in that there is an overt element present in the structure which may assume the role of the subject. As a result of the reanalysis of the clitic, the agent/subject θ-role is no longer assigned to the big DP which contained both the clitic and its double. Preservation of argument structure is warranted by assigning the relevant thematic role to the former double, which turns into the ‘real’ subject of the structure that results from the reanalysis in question. Note that this change also leads to structural simplification due to the loss of the big DP structure and the relevant selectional features that required the addition of a full DP double in the specifier of the big DP. Finally, parallel to the analysis of the Bavarian data above, I suggest that the newly created agreement morpheme attached to C is inserted post-syntactically as a dissociated Agr-morpheme which is licensed (under structural adjacency) by an agreement head that has been valued in the syntactic derivation. Thus, the changes affecting clitic doubling structures in Sutselvan can be summarized as follows (cf. Chapter 4): (91) [CP XP [C’ C+V+pronounj [TP [DP double [D’ tj ...]]i [T’ T [νP ti ... → [CP XP [C’ C+V+Agr [TP [TP subjecti [T’ T [νP ti ... Structural simplification: loss of big DP Preservation of argument structure: θ-role previously assigned to big DP is now assigned to previous double Identification of feature content: dissociated Agr-morpheme licensed under structural adjacency with Agr-on-T Word building constraint: verb is located in C (via movement) This analysis makes an interesting prediction: similar to the Germanic dialects which exhibit complementizer agreement (i.e., West Flemish, Frisian, East Netherlandic, Bavarian etc.), the presence (or, rather, valuation) of an Agr-morpheme on C should require structural adjacency between C and T (hosting the syntactically valued Agrmorpheme). Thus, on the assumption that the subject occupies SpecTP in the RR dialects under investigation (see above), we should expect that no material (apart from modal particles and clitics) may intervene between the Agr-morpheme on C and the subject in SpecTP (see Chapter 3, Section 3.7). Note that this would then constitute a further deviation from the behavior of optional clitic doubling in the other dialects,

JB[v.20020404] Prn:2/08/2005; 15:48

F: LA8105.tex / p.41 (2650-2729)

Chapter 5. The reanalysis of C-oriented clitics 

where a number of elements may intervene between the clitic and the full subject DP, cf. the following examples from Puter, Vallader and Surmeiran:51 Puter (92) Co dalungia füttan=e à Seneca et a sa duonna avertas then immediately were=clit.3pl to Seneca and to his wife opened sü las avainas in la bratscha. up the veins in the arms ‘Immediately, then, they slashed the wrists of Seneca and his wife.’ (Linder 1987: 160) Vallader (93) Che effet ha=la gnü aint il pövel la nouva which effect has=clit.3sg.fem done in the people the new predgia? sermon ‘Which effect had the new sermon on the people?’ (Linder 1987: 154) Surmeiran (94) an quel mumaint è=la riveda er la mama. in this moment is=clit.3sg.fem arrived also the mother ‘In this moment, the mother arrived too.’ (Linder 1987: 154) Interestingly, this prediction seems to be borne out by the facts. Of the 42 examples of ‘clitic doubling’ in Sutselvan listed in Linder (1987: 148ff.), the subject is directly rightadjacent to the ‘clitic’ (i.e., the Agr-morpheme on C) in 34 cases (81%). Similar to the adjacency requirement observed in connection with complementizer agreement in Germanic, this can be attributed to the fact that the dissociated Agr-morpheme on C (resulting from the reanalysis of a former clitic) requires structural adjacency between C and T. However, note that there are five examples where a modal particle or a short adverb (marked by underlining) intervene between the clitic and the subject as for example in the following sentences: (95) a.

Egn da quels lev-i ear jou. one of those wanted-1sg also I ‘I also wanted one of those.’ (Linder 1987: 148) b. Anzucuras e-la lura la mort gnida... eventually is-3sg.fem then the death come ‘And eventually, the death then came...’ (Linder 1987: 155)

Among the intervening elements, I found only ear ‘also’ (one example), (al)lura ‘then’ (three examples) and schon ‘already’ (one example), that is, short, monomorphemic particles and adverbs. By assumption, these elements are to be analyzed on a par with the modal particles in languages such as Bavarian, which are base-generated as adjuncts on TP and therefore do not block structural adjacency between C and T (see

JB[v.20020404] Prn:2/08/2005; 15:48

F: LA8105.tex / p.42 (2729-2786)

 The Rise of Agreement

Chapter 3, Section 3.7). Finally, there are three examples where a participle intervenes between C and the subject: (96) a.

“Ah, nagn a tgea”, â-l getg igl fumegl... go-1pl to home has-3sg.masc said the servant “‘Well, let’s go home”, said the servant...’ (Arquint-Felix; Linder 1987: 155) b. Dalùnga en-i svanidas las pusadas. immediately are-3pl vanished the cutlery-pl ‘Immediately, the cutlery vanished.’ (Arquint-Felix; Linder 1987: 161) c. Igl mestar e-l sto igl Mia la Freana. the master is-3sg.masc been the Mia of Freana ‘The Mia of Freana was the master.’ (Linder 1987: 153)

Note that two of these examples originate from the same source (J.C. Arquint, Cuors da Rumantsch. Adataziun sutsilvana da A.-L. Felix. Cuera, 1958). Thus, it might be possible that (96a–b) reflect the grammar of a single speaker that has not yet completed the reanalysis of clitics as Agr-morphemes on C. Alternatively, one might attribute the placement of the participle in (96) to influence from Italian, where the participle always precedes the subject in inversion contexts. Of course, the limited amount of available data does not allow for firm conclusions concerning the status of clitic doubling in Sutselvan. This would require a survey over a greater data set (and in particular, speakers’ judgments), which I leave open as a research project for the future. Still, the evidence which is available suggests that Sutselvan has undergone a change in which clitic pronouns in doubling structures were reanalyzed as exponents of a dissociated agreement morpheme on C. Apparently, this change led to a further difference between Sutselvan and the other Swiss RR dialect which has gone unnoticed by Linder (1987), namely a requirement that C and the subject (or TP) be adjacent. Note that this (somewhat preliminary) conclusion lends further support to the hypothesis that the grammaticalization of agreement markers proceeds via an intermediate stage where pronouns are reanalyzed as postsyntactically inserted agreement morphemes (on C) which may eventually turn into syntactic agreement morphemes in a subsequent development. The analysis presented so far should lead us to expect that Sutselvan exhibits a similar behavior as other languages that acquired Agr-on-C. In other words, it should be possible that a full DP subject can co-occur with the agreement morpheme on C in all environments, for example in embedded clauses or when the subject occupies the clause initial positions, as in the following Bavarian sentences: (97) a.

ob-st du noch Minga kumm-st whether-2sg you.sg to Munich come-2sg ‘Whether you come to Munich’ b. Mia fom-ma hoam. we drive-1pl home ‘We go home.’

JB[v.20020404] Prn:2/08/2005; 15:48

F: LA8105.tex / p.43 (2786-2828)

Chapter 5. The reanalysis of C-oriented clitics 

Surprisingly, I have not managed to find sentences parallel to (97) in the sources available to me. This might be taken to indicate that these orderings are ungrammatical in Sutselvan. In fact, Haiman and Benincà (1992: 192) state that doubling (that is, the presence of Agr-on-C) is only possible in inverted order in the Swiss RR dialects. However, it is unclear whether this generalization refers only to main clauses or whether it is meant to exclude doubling in embedded contexts as well. Here, clearly more empirical work is necessary to settle this matter.52 In the following, I will focus on the apparently clear-cut cases, that is, the constraint against the presence of an agreement morpheme on C in main clauses that lack inversion.

... Syntactic restrictions on Agr-on-C In this section, it is argued that the peculiar distribution of Agr-on-C in Sutselvan can be traced back to restrictions on clitic doubling which hold in all Swiss RR dialects. In other words, I claim that certain synchronic properties of Sutselvan should ultimately receive a diachronic explanation, in the sense that the distribution of Agr-on-C still reflects restrictions on its historical source, clitic doubling. This suggests that the change in question is a rather recent development and that Agr-on-C has not yet been generalized to all instances of C. Above, I have noted that in all Swiss RR dialects, doubling is restricted to inversion contexts where both the enclitic and the full nominal follow the finite verb in C. In contrast, weak pronouns/proclitics do not license doubling and are in complementary distribution with full nominals: (98) Ordering restrictions on clitic doubling in the Swiss RR dialects a. *full DP-finite verb-enclitic... b. *proclitic-finite verb-full DP... c. *full DP-proclitic-finite verb... Here, I want to suggest that these restrictions on doubling can be explained if the RR dialects in question are analyzed as instances of ‘asymmetric’ V2. According to Travis (1984) and Zwart (1993a), superficially identical V2 orders are in fact the result of two different underlying structures in the Germanic V2 languages (hence the notion ‘asymmetric V2’):53 the finite verb moves to C only in inversion contexts, where SpecCP is occupied by a non-subject. In contrast, the verb stays behind in Infl/T in subject-initial main clauses, with the subject occupying SpecIP/TP. Note that this analysis seems to be much more natural for the Swiss RR dialects under investigation than for the Germanic OV languages for which this proposal was originally developed, due to the SVO character of RR and the fact that the dialects discussed here show general V-to-T in embedded clauses (see above). Combined with the analysis of clitic doubling proposed above, an ‘asymmetric’ analysis leads to the following structure with the big DP in SpecTP, that is, in clause-initial position:

JB[v.20020404] Prn:2/08/2005; 15:48

F: LA8105.tex / p.44 (2828-2886)

 The Rise of Agreement

(99)

*CP Ø

C’ C Ø

TP T’

DPi DP double D clitic

D’

íP

Vfin + T NP pro

í’

ti í

VP

The non-availability of doubling in subject-initial main clauses can then be explained in the following way. First, if we assume that the licensing of clitics is connected to certain functional categories (which must provide an appropriate prosodic host), that is, C in Germanic-type languages and Infl/T in many Romance languages, the order (98a) cannot be generated on the basis of a structure like (99), since there is no appropriate prosodic host available for the C-oriented subject enclitics of the Swiss RR dialects. More precisely, the enclitic has to attach either to a finite verb (in C) or to a complementizer. Therefore, the order (98a) (full DP–finite verb–enclitic) cannot be generated by (i) lowering the enclitic to the right of the finite verb in T or (ii) raising only the subject to SpecTP, while the enclitic stays behind in SpecνP. In both instances, the enclitic fails to be associated with C and thus does not meet its licensing requirements. What about (98b) and (98c), which both involve a proclitic instead of an enclitic pronoun? Tentatively, I assume that the capacity to license a specifier where the double is merged correlates with morphological properties of the clitic. Recall that in those dialects that exhibit a special series of proclitics (Puter, Surmeiran), a single proclitic (Puter a-, Surmeiran i-) is used for several person/number combinations (1sg, 3sg.neut, and all plural forms). Therefore, there is a significant difference between the proclitic forms, which are entirely homophonous and non-distinctive, and the enclitics which serve to signal several person/number distinctions (see Table 7 above). By assumption, there is a correlation between this morphological difference and the capability to project a specifier to host the double (cf. Uriagereka 1995 for related conclusions concerning the morphological strength of D and the capacity to license a specifier). As a consequence, doubling is only available with enclitic forms which are morphologically ‘strong’ enough to license a double in their specifier.54 Thus, it appears that at least some of the restrictions on clitic doubling observed in the present-day Swiss RR dialects can be better understood if these dialects are analyzed as instantiating a form of ‘asymmetric’ V2 in the sense of Travis (1984) and Zwart (1993a). Moreover, it seems reasonable to assume that similar restrictions on the availability of clitic doubling were at work in historical stages of Sutselvan. This makes

JB[v.20020404] Prn:2/08/2005; 15:48

F: LA8105.tex / p.45 (2886-2924)

Chapter 5. The reanalysis of C-oriented clitics 

available a diachronic explanation for the distribution of Agr-on-C in the present-day language. When instances of clitic doubling were reanalyzed as instances of Agr-on-C, the restrictions on clitic doubling made their way into the new grammar as restrictions on Agr-on-C. In other words, the restrictions on Agr-on-C observed in the present-day language reflect restrictions that formerly constrained clitic doubling: Agr-on-C could not arise from clitic doubling in subject-initial clauses, since clitic doubling was not licensed in these contexts in previous stages of Sutselvan. Furthermore, even after the general development of Agr-on-C, ‘doubling’ is not possible in subject-initial clauses, since there is no lexical host available for the Agr-suffix in C (due to the ‘asymmetric’ nature of V2 in Sutselvan): *CP

(100) Ø

C’ C C Ø

TP Agr

DPi

T’ íP

Vfin + T

í’

ti í

VP

In addition, the observation that ‘doubling’ (i.e., Agr-on-C) seems to be limited to non-embedded clauses in Sutselvan (see note 52) further supports the hypothesis that Sutselvan represents an early stage of the grammaticalization process in question, where Agr-on-C is still confined to the contexts where it originally arose, that is, inversion in non-embedded clauses.

.. Section summary This section has demonstrated that the Swiss RR dialects exhibit several waves of a grammaticalization process in which C-oriented enclitic pronouns are reanalyzed as agreement markers. I pointed out that we can identify two different syntactic contexts where this change took place. First, Section 5.3.2 showed that in a set of earlier developments affecting 1st and 2nd person forms, enclitic pronouns were ‘directly’ reanalyzed as agreement morphemes, giving rise to a limited amount of pro-drop, especially in 2nd person contexts. I argued that this diachronic process basically resembles the changes in Bavarian discussed in more detail in Section 5.2 above. In other words, enclitic pronouns were reanalyzed as realizations of a dissociated agreement morpheme that attaches to the C-head at Morphological Structure. In the Swiss RR dialects, this reanalysis has been further promoted by a stress rule (‘penultimate stress

JB[v.20020404] Prn:2/08/2005; 15:48

F: LA8105.tex / p.46 (2924-2985)

 The Rise of Agreement

target’, Haiman 1971) which may trigger elision of the existing agreement ending if a clitic attaches to the verb, ensuring that the antepenultimate syllable does not receive stress. In an independent later development, the new endings are generalized to other verbal positions, replacing the original agreement formatives, that is, the morphological exponent of Agr-on-C is reanalyzed as the exponent of Agr-on-T, the ‘canonical’ agreement morpheme. Similar to Bavarian, these changes are made available by the V2 property, which creates linear adjacency between the clitic and the verb. Moreover, I argued in Sections 5.3.3 and 5.3.4 that at least in Sutselvan, we can detect a rather recent process in which new agreement formatives develop from clitic doubling structures where an enclitic pronoun is reinforced by a full nominal for reasons of stress and emphasis. This doubling configuration was analyzed in terms of a structure where the clitic and the reinforcing DP are merged together as a ‘big DP’. After the reanalysis of the clitic as an exponent of (dissociated) Agr-on-C, the former double receives the θ-role previously assigned to the big DP, thereby preserving the predicate’s argument structure. Due to the loss of the big DP structure, this change is also in line with the requirement of structural simplification. Again, the reanalysis in question can only take place in inversion contexts, where adjacency of the clitic and the verb is guaranteed by the V2 property. The analysis presented in this section predicts that the newly created Agr-on-C must be structurally adjacent with T in order to be licensed. It was shown that this prediction appears to be borne out by the Sutselvan data, constituting a further deviation from the behavior of (optional) clitic doubling in the other Swiss RR dialects, which to the best of my knowledge has gone unnoticed in previous work. Finally, in Section 5.3.4.1, I claimed that a set of syntactic restrictions on Agr-on-C in Sutselvan should ultimately receive a diachronic explanation, arguing that the distribution of Agr-on-C reflects independent restrictions on clitic doubling, the historical source of the agreement morpheme on C (these restrictions can still be observed in the other Swiss RR dialects). This diachronic explanation of the distribution of Agron-C was based on an analysis of the Swiss RR dialects as asymmetric V2 languages, where the finite verb moves to C only in inversion contexts, but stays behind in T in subject-initial clauses. Thus, even in a set of very closely related dialects, we can observe different pathways toward new verbal agreement morphology. This observation further supports the hypothesis that the grammaticalization process in question may be triggered in a variety of different syntactic environments. In the next section, I will examine another set of historical scenarios in which C-oriented clitics may turn into new verbal agreement markers, drawing on data from languages lacking the V2 property.

. Reanalysis of C-oriented clitics in non-V2 languages This section discusses two instances of a grammaticalization process in which Coriented clitics turn into agreement formatives in a non-V2 environment, drawing

JB[v.20020404] Prn:2/08/2005; 15:48

F: LA8105.tex / p.47 (2985-3042)

Chapter 5. The reanalysis of C-oriented clitics 

on data from various Uto-Aztecan (Section 5.4.1) and Mongolian languages (Section 5.4.2). It is shown that the reanalysis of clitics takes place in different syntactic environments in these languages, either under adjacency with a verb moved to T or under adjacency with a finite verb that has undergone fronting as part of a TP moved into the left clausal periphery. Importantly, the languages under consideration differ from the cases discussed so far in that they lacked verbal agreement markers prior to the reanalysis of subject clitics. It is argued that this difference gives rise to a change where the clitic is directly reanalyzed as an instantiation of a syntactic Agr-morpheme on T. Thus, the grammaticalization process does not involve an intermediate stage where the clitic turns first into a dissociated Agr-morpheme.

.. Uto-Aztecan Steele (1977) shows that various Uto-Aztecan languages (spoken in Southern California and Mexico) exhibit different stages of a development in which independent pronouns first become second position clitics and eventually develop into prefixal verbal agreement morphology. Unfortunately, there exist only very few written records of previous historical stages of these languages. However, due to the fact that the relevant change seems to affect the individual languages in more or less parallel fashion, a comparative study of different Uto-Aztecan languages can be taken to provide important insights into the various historical stages of the grammaticalization process in question. Hence, the following discussion is based on the assumption that the individual Uto-Aztecan languages represent different stages of the development of prefixal agreement markers from former second position clitics (cf. Steele 1977). Many Uto-Aztecan languages exhibit a set of enclitic pronouns which historically derive from full tonic pronouns (cf. Steele 1977, 1978, 1995). These clitics occupy the second position of the clause, that is, they move into the left periphery and attach to the first word or constituent: Luiseño (101) hunwuti=pum se‘iwun. bear.object=clit.3pl are.shooting ‘They are shooting a bear.’ (Steele 1977: 539) Interestingly, in some languages such as Classical Aztec, Huichol, and Pochutla, the former clitics obligatorily attach to the left of the finite verb and are no longer confined to second position. This suggests that they have developed into person/number (i.e., agreement) prefixes on the verb (see below for further evidence in favor of an analysis in terms of verbal inflection): Classical Aztec (102) an-teeˇcLaso‘La. 2pl-love-us ‘You love us.’ (Steele 1977: 539)

JB[v.20020404] Prn:2/08/2005; 15:48

F: LA8105.tex / p.48 (3042-3133)

 The Rise of Agreement

In Classical Aztec, the (historical) relation between agreement markers and independent pronouns is still discernible. Table 8 shows that the onset of the individual agreement markers is identical to the onset of the respective pronouns (apart from 3rd person forms):55 Table 8. Pronouns and agreement prefixes of Classical Aztec (Steele 1978: 611f.).

1sg 2sg 3sg 1pl 2pl 3pl

Verbal agreement prefixes

Independent pronouns

nitiØ tianØ

ne‘waatl te‘waatl ye‘waatl te‘waan am‘waan ye‘waan

In the following, I will discuss the course of the grammaticalization of prefixal person/number markers in Uto-Aztecan, drawing on data from a set of languages that by assumption exhibit different developmental stages of the historical process in question. According to Steele (1977), the change in question proceeded via an initial stage where full pronouns attracted to second position turned into reduced clitic forms. In a later development, these second position clitics were eventually reanalyzed as person/number prefixes on the verb. The early beginnings of the grammaticalization of verbal inflection can still be observed in languages such as Mono (Western Numic) where full, non-reduced (subject) pronouns generally occur in second position: Mono (103) nopihweeh nGG miyawaih. to.home I will.go ‘I shall go home.’ (Steele 1977: 541) In Mono, the majority of the pronouns are still full forms which do not attach to the clause-initial phrase/word. However, 1st dual and plural inclusive forms have developed into enclitics which lean on to the clause-initial element:56 Mono (104) nopihweeh=taa miyawaih. to.home=clit-1pl will.go ‘We shall go home.’ (Steele 1977: 541) Tubatulabal and Comanche present a somewhat advanced stage and exhibit a full paradigm of second position clitics (in addition to full pronouns). The second position clitics are in complementary distribution with full pronouns (and other DPs), which clearly shows that the clitics still have argument status. Note that all three languages that show early stages of cliticization (i.e., Mono, Comanche, and Tubatulabal)

JB[v.20020404] Prn:2/08/2005; 15:48

F: LA8105.tex / p.49 (3133-3195)

Chapter 5. The reanalysis of C-oriented clitics 

exhibit no evidence that topic left dislocation is involved in this process. For example, clitics follow clause-initial elements such as infinitival clauses and time adverbials, which presumably do not constitute topics: Tubatulabal (105) aanayuwibG=gi iimi. fight.subordinate=1sg went ‘Without fighting, I went.’ (Steele 1977: 551) Tubatulabal (106) piš=bum pinahi tohiilin tuguwayin. then=2pl must.bring deer’s meat ‘Then you must bring the meat of the deer.’ (Steele 1977: 551) Furthermore, if second position clitics originated from reduced resumptive pronouns in topic left dislocation structures, we would expect the pronouns to co-occur with topicalized DPs in the early stages of this development. But this is not the case. Neither in Mono, nor in Comanche or Tubatulabal, clitic pronouns can co-occur with coreferential topicalized DPs (i.e., there is no clitic doubling in these languages). Thus, it seems that doubling phenomena are in fact a result, and not the original source of the grammaticalization process in question. Instead, it seems reasonable to suppose that clitic doubling came into existence when the former full pronouns underwent further phonological reduction, presumably due to their placement in clausal second position (see below for some discussion). For reasons of emphasis, the second position clitics are then reinforced by a full pronoun that bears stress, giving rise to instances of subject clitic doubling, presumably involving a big DP structure similar to the Swiss RR data discussed in Section 5.3 above. This can be observed in all Uto-Aztecan languages except for Mono, Comanche, and Tubatulabal: Luiseño (107) wunaalum=pum hunwuti se‘iwun. they=clit.3pl bear.object are.shooting ‘They are shooting the bear.’ (Steele 1977: 547) Papago (108) ‘áañi=‘áñ ñGok. I=clit.1sg am.speaking ‘I am/was speaking.’ (Steele 1977: 547) Serrano (109) ‘Gmi‘=ta=m‘ payGka‘ miib. you=modal=2sg away will.go ‘You will go away.’ (Steele 1977: 547)

JB[v.20020404] Prn:2/08/2005; 15:48

F: LA8105.tex / p.50 (3195-3271)

 The Rise of Agreement

Yaqui (110) ínepo ka=ni aman nóitik. I neg=clit.1sg there went ‘I didn’t go there.’ (Steele 1977: 547) Southern Paiute (111) qaˇcu=anga=ni nG‘ iminˇcuxwavaang‘wainiaanga. neg=object.clit=clit.1sg I shall.give.him.to.you ‘I shall not give him to you.’ (Steele 1977: 547) In (107) and (108) (from Luiseño and Papago), the clitic attaches directly to the right of the double (in both cases a full pronoun), which occurs in clause-initial position. In (109) and (110) (from Serrano and Yaqui), the clause-initial position is again occupied by the double. In contrast to the first two examples, however, the clitic does not attach directly to the full form, but rather to a free inflectional marker (a modal in (109) and a negation marker in (110)) which is placed in clausal second position.57 Finally, in (111) (Southern Paiute) the clitic attaches to a complex consisting of the negation marker and an object clitic, followed by the 1sg full pronoun nG‘. At first sight, the examples (107) and (108) seem to suggest that the whole big DP has moved into clause-initial position (presumably a specifier in the left clausal periphery). However, on the assumption that the relevant construction should receive a uniform analysis across Uto-Aztecan, the examples in (109) and (110) show that this cannot be correct: in both (109) and (110), an inflectional head intervenes between the double and the clitic. Therefore, it seems reasonable to suppose that the double moves independently into the left periphery (an instance of A’-movement), while the clitic attaches to an appropriate host in a separate operation, presumably some kind of prosodic repositioning which is part of MS or, more generally, the mapping to PF (cf. e.g. Bonet 1991; Halpern 1992; Schütze 1994; Embick & Noyer 2001).58 The clitic may attach to the double (in, say, SpecCP) if no other material is located in an intervening head position. For the time being, I therefore assume that the big DP moves to SpecTP prior to fronting of the double (but see below for some modifications). Then it seems likely that the intervening inflectional markers occupy a functional head in the Cdomain, which I will refer to simply as C. In some languages such as Cora, Cupeño, Tarahumara, Tepecano, and Yaqui, the second position clitics can co-occur with prefixal person/number markers on the verb, the shape of which clearly indicates that they developed from former pronominal elements (furthermore, these languages exhibit a series of full pronouns), cf. the following examples from Tarahumara and Yaqui:59 Tarahumara (112) ˇcú=mu šika ké mu-nakí muhé ko ba? wh=clit.2sg wh neg 2sg-want you emph emph ‘Why don’t you want it?’ (Steele 1977: 553)

JB[v.20020404] Prn:2/08/2005; 15:48

F: LA8105.tex / p.51 (3271-3355)

Chapter 5. The reanalysis of C-oriented clitics 

Yaqui (113) kwarénta péso dyáryota=ne ne-kóba íani ínine. forty peso daily=clit.1sg 1sg-earn now here ‘Now I make forty pesos a day here.’ (Steele 1977: 543) Since it is rather implausible that both elements constitute pronominal clitics, these data can be taken to indicate that the proclitic elements are better analyzed as prefixal agreement markers on the verb. This conclusion is further supported by the fact that in (112), the example from Tarahumara, there is a full pronoun present in addition to the second position clitic and the person/number marker on the verb.60 The claim that the prefixal elements in these languages are in fact agreement markers is further supported by the fact that in Cora (similar to Classical Aztec, see above), the prefixal element is not restricted to second position any longer. In the following example, the person/number prefix occurs on the left edge of a clause-initial complex verb which is followed by the second position clitic nú: Cora nú. (114) n-a’ana’iny íiche’e 1sg-was.going.to.build.a.fire clit.1sg ‘I was going to build a fire.’ (Steele 1977: 543) We can thus conclude that the various Uto-Aztecan languages exhibit different stages of the grammaticalization of new prefixal agreement markers from former second position clitics. The evidence available to us suggests that the change proceeded via an intermediate stage where the clitic is doubled by a full pronoun, presumably for reasons of emphasis. In a subsequent development, these doubling configurations gave rise to a reanalysis where the clitic is reanalyzed as a verbal agreement prefix.61 The empirical findings with respect to the different stages of the grammaticalization process in question are summarized in Table 9.62 In stage 1 (exemplified by Mono), full pronouns become attracted to second position.63 Presumably, the occurrence in an intonational gap after the clause-initial phrase accelerates the phonological reduction of the pronouns (stage 2). At some point, the Table 9. Different stages of the grammaticalization of person/number prefixes in UtoAztecan. Grammaticalization process

Languages

Stage 1:

Mono

Stage 2: Stage 3: Stage 4:

Full pronouns are attracted to second position Reduction to second position enclitics Second position enclitics are doubled by a full nominal Second position enclitics have turned into verbal prefixes

Tubatulabal, Comanche Luiseño, Papago, Serrano, Southern Paiute Classical Aztec, Cora, Huichol, Pochutla, Tarahumara, Tepecano, Yaqui

JB[v.20020404] Prn:2/08/2005; 15:48

F: LA8105.tex / p.52 (3355-3429)

 The Rise of Agreement

second position clitic is reinforced by a full DP (presumably for reasons of emphasis), leading to the big DP structure proposed above (stage 3). At some point, the resulting clitic doubling structures lose their stylistic force and the clitics are reanalyzed as agreement morphemes on T, leading to stage 4. The transition from stage 3 to stage 4 which leads to the presence of an agreement prefix on the verb is schematized in (115). (115) [CP XP [C’ C+pronouni [TP [T’ V+T [νP [DP double [D’ ti ]] ... → [CP XP [C’ C [TP’ Agr+V+T [νP subject ... Structural simplification: loss of big DP Preservation of argument structure: θ-role previously assigned to big DP is now assigned to the former double Identification of feature content: via Agree with the subject in SpecνP Word building constraint: verb is located in T (via movement) Similar to the change affecting Sutselvan (see Section 5.3.4), this reanalysis leads to structural simplification via losing the big DP structure underlying clitic doubling in the target grammar. Preservation of argument structure is warranted by assigning the θ-role to the former double. The newly created agreement morpheme is valued by initiating an Agree operation that accesses the subject in SpecνP, which by assumption may stay in situ (cf. Note 63). Finally, the question arises of how the Word building constraint can be satisfied. More generally speaking, it is fairly clear that the reanalysis in (115) is possible only if the former clitic is adjacent to the verb (or an auxiliary-like element base-generated in T, see above). Interestingly, Steele (1977: 556) notes that the change from second position clitics to verbal prefixes took place only in languages that fail to exhibit the basic SOV order characteristic of most Uto-Aztecan languages, but show instead SVO or VSO/VOS word order.64 This is exactly what is expected under the Word building constraint: the change described in (115) can only take place if the verb has moved high enough in the structure to warrant adjacency to the C-oriented (second position) clitic. By assumption, the target of verb movement is T in these contexts, which also carries the agreement morpheme resulting from the reanalysis.65 If the verb stays in a lower position it is not adjacent to the clitic, which renders a reanalysis of the clitic impossible in SOV orders. Of course, this section provided only a cursory discussion of the intriguing synchronic and diachronic facts which can be observed in the various Uto-Aztecan languages, leaving many details open for future research. Still, it seems that it is possible to subsume the individual grammaticalization processes under a single explanation which is based on the assumption that the agreement prefixes under consideration evolved from second position clitics in clitic doubling configurations. The individual steps taken by this process have been isolated by comparing a set of different languages that by assumption exhibit different stages of the grammaticalization of agreement

JB[v.20020404] Prn:2/08/2005; 15:48

F: LA8105.tex / p.53 (3429-3502)

Chapter 5. The reanalysis of C-oriented clitics 

markers from clitic pronouns. Interestingly, the grammaticalization process leading to new verbal prefixes apparently took place only in a subset of the Uto-Aztecan languages which is characterized by the lack of basic SOV order. This was attributed to the fact that only in this subset, the necessary adjacency between the clitic and the verb was guaranteed. Note that the change in question differs from the diachronic developments discussed in Sections 5.2 and 5.3 in that it did not proceed via an initial stage where the clitic is reanalyzed as a dissociated Agr-morpheme. Instead, the clitic is reanalyzed as an instantiation of Agr-on-T which is valued during the syntactic derivation. This derives from the assumption that dissociated agreement morphemes are parasitic on a syntactically licensed Agr-morpheme. Accordingly, the latter path is not available in languages that lack agreement altogether prior to the reanalysis. It follows that the reanalysis of clitics does not necessarily proceed via an initial stage where the clitic turns into a dissociated Agr-morpheme, in contrast to what has been hinted at in Section 5.2 above. As mentioned above, basic SOV order represents an obstacle for the development of verbal inflection from clitic pronouns which apparently can only be overcome by moving the verb to the left of the pronoun in V2 languages (as in the case of Bavarian, see Section 5.2 above). Still, I have noted in Chapter 1 and Chapter 3 that most SOV languages exhibit suffixal agreement morphology. Given the rareness of the V2 phenomenon across languages, we clearly cannot assume that all these languages developed agreement suffixes during a historical stage when they were all V2 languages. In the next section, I will therefore consider an alternative path which leads to the development of suffixal agreement markers in an SOV language.

.. Mongolian In a number of Mongolian SOV languages (Buryat, Dagur, Kalmyk, Moghol, and Oirat), we can observe a rather recent grammaticalization process that led to the development of suffixal agreement markers. This section focuses on Buryat, spoken in East Siberia and various parts of Mongolia (Poppe 1960; Sanžeev 1973; Comrie 1980). Even a cursory look at table 3 reveals the similarities between nominative pronouns and the subject agreement suffixes found on finite verbs. In addition, a comparison of the last two columns shows that the series of possessive markers found in that language apparently developed from the genitive pronouns in a similar fashion.66 Recall that SOV languages such as Buryat raise an important issue concerning the grammaticalization of agreement markers. Above, we have noted that basic SOV order often hinders the development of agreement markers from subject clitics, since the clitic is usually not adjacent to the finite verb. Even more problematic, it has been observed in Chapter 1 that the vast majority of SOV languages exhibits suffixal inflection. The assumption that the position of agreement morphemes on the verb reflects the relative order between verb and clitic at the stage when clitics were reanalyzed as agreement markers (Givón 1971, 1976) implies that the verb preceded the clitic when the latter turned into a verbal agreement marker. As noted in Section 5.2 above, V+clitic

JB[v.20020404] Prn:2/08/2005; 15:48

F: LA8105.tex / p.54 (3502-3549)

 The Rise of Agreement

Table 10. Pronouns and agreement in Buryat (Poppe 1960; Comrie 1980).

1sg 2sg 3sg 1pl 2pl 3pl

Nominative pronouns

Verbal Agr

Genitive pronouns

Possessive suffixes

bi ši – bide ta –

-b -š -Ø -bdi -t(a) -d

´ minii ´ šinii – manai tanai –

-m(ni), -ni -š(ni) -(i)n´ -(m)nai -tnai -(i)n´

orders may be the outcome of a V2 syntax. However, the fact that the V2 property is actually quite rare among the world’s languages suggests that it is not very likely that every SOV language with suffixal agreement went through a V2 stage at some point in its history. In this section, I will therefore give a sketch of an alternative scenario for the development of suffixal agreement in SOV languages. Before we turn to the details of the relevant proposal, let us review some more data which bear on the syntax of Mongolian and the development of suffixal agreement in particular. In present-day Buryat, the morpheme order found on the verb does not correspond to the position of free pronouns. Subject pronouns regularly precede the verb (and the object, if present), while the agreement markers – which still clearly resemble the pronominal forms – are realized as suffixes on the verb: (116) a.

bi jaba-na-b. I go-pres-1sg ‘I am going’ (Comrie 1980: 89) b. bide jerexe-bdi. we shall.come-1pl ‘We shall come.’ (Poppe 1960: 123)

To account for the mismatch between word and affix order, one might assume that the basic word order was VS at the historical stage when the pronouns were reanalyzed as agreement markers, followed by a subsequent change to SOV order. However, it can be shown that Buryat has not undergone any major word order change in its recorded history (cf. Comrie 1980). Moreover, according to Comrie (1980), the grammaticalization of agreement suffixes is in fact a rather recent development, since the relevant suffixes are absent in earlier stages of the language, that is, Classical Mongolian (attested in written records from the 13th century on, cf. Poppe 1954; Grønbech & Krueger 1955).67 Thus, an analysis of the diachronic developments in Buryat has to address the question of how a language with basic SOV word order can develop agreement suffixes on the verb. Given that subject pronouns regularly appear to the left of the verb, we should rather expect the development of agreement prefixes, contrary to the facts. Moreover, a further problem arises if we acknowledge that adjacency is a necessary precondition for the reanalysis of a (clitic) pronoun as a verbal affix. In an SOV

JB[v.20020404] Prn:2/08/2005; 15:48

F: LA8105.tex / p.55 (3549-3619)

Chapter 5. The reanalysis of C-oriented clitics

language like Buryat, this precondition is often not met, since the subject is regularly separated from the verb by intervening objects or adverbs. Interestingly, however, it can be shown that in written records of older stages of Mongolian, pronouns can also show up in other positions, often following the verb, cf. the following quote from Poppe (1954: 124): In the pre-classical language, personal and demonstrative pronouns are often placed after the finite verb. Sometimes a personal pronoun is placed, as usual, before the predicate, but is repeated after the latter.

This marked word order option is illustrated by the following examples from (pre-) Classical Mongolian (taken from Poppe 1954: 124f.; unfortunately, Poppe gives no examples with doubling): (117) a.

tere metü jalbarin ügülemüi bi. this like prayer say I ‘I am praying in that manner.’ b. inegeldüküi-yi yekin tayalamu ˇci? laughter why like you ‘Why do you like laughter?’

Even more interesting, Poppe (1954) assumes that this word order option provided the context for the grammaticalization of agreement suffixes on the verb:68 This repetition of the pronoun after a verb has produced personal endings on verbal forms in certain colloquial languages, e.g., in Buriat, Kalmuck, and others, e.g., Buriat bi abanab I take, ši abanaš you take, etc. (p. 125)

In other words, it seems fairly clear that the agreement suffixes found in present-day Buryat originated from a marked word order option in which a weak unstressed pronoun followed the finite verb, while an additional full form apparently could be added in preverbal position (presumably for reasons of emphasis). Thus, it is possible to assume that similar to other instances of clitic doubling, this marked stylistic option lost its communicative function over time, giving way to a reanalysis in terms of a neutral, non-marked construction, where the postverbal, unstressed/clitic pronoun is reanalyzed as a verbal agreement suffix, while the preverbal pronoun turns into the ‘true’ subject of the clause. Unfortunately, I have not been able to track down instances of doubling in the material available to me. As noted above, Poppe (1954) merely remarks that doubling is possible, but does not give a relevant example. In other words, at least for the time being it remains a speculation that instances of doubling were involved in the diachronic process under investigation. However, note that Classical Mongolian (Poppe 1954: 123; Grønbech & Krueger 1955: 39) as well as present-day Buryat (Poppe 1960: 119) seem to license pro-drop: similar to Chinese (and other Asian languages), pronouns need not be realized overtly if their reference can be inferred from the discourse context. This observation suggests an alternative scenario for the development of new agreement markers in Buryat. As shown above in this chapter (and Chapter 4), the pro-drop property may facilitate the reanalysis of pronouns as agreement markers,



JB[v.20020404] Prn:2/08/2005; 15:48

F: LA8105.tex / p.56 (3619-3685)

 The Rise of Agreement

since the relevant θ-role can be readily assigned to an empty pronoun. Accordingly, I assume that in the history of Buryat, the availability of pro-drop licensed a ‘direct’ reanalysis of postverbal pronouns as agreement markers. However, while this hypothesis captures the relevant historical developments, it still leaves the exact structural properties of the reanalysis in question in the dark. For example, it is still left unclear how weak subject pronouns can come to occupy a postverbal position in an SOV language such as Buriat. Comrie (1980: 92) suggests that the reanalysis in question affected pronouns that were postposed to a position to the right of the verb. He claims that in the Mongolian languages, this operation is necessary to avoid a sequence where an unstressed element precedes its selecting head (which seems to be ruled out in Mongolian in general).69 However, an analysis where the weak pronouns are assumed to undergo rightward movement seems to be problematic for the following two reasons. First, it is widely accepted that weak/clitic pronouns tend to move into the inflectional domain (or even higher) for syntactic licensing (cf. e.g. Cardinaletti & Starke 1999). Second, it is well-known that rightward movement (if it constitutes an option at all) is normally confined to ‘heavy’ elements, as in clausal extraposition or Heavy NP Shift. Hence, extraposition of pronouns (even full pronouns that may receive stress) usually leads to ungrammaticality (e.g., in the Germanic languages). Therefore, it is perhaps more promising to assume that the position of the weak pronouns of (Classical) Mongolian is the result of another syntactic process that guarantees that the pronoun ends up in a (well-established) defocused position, namely clausal second position. More specifically, I suggest that the weak pronouns of Mongolian adjoin to C, similar to the Wackernagel clitics in the other languages discussed so far. This cliticization movement is then followed by fronting of a larger constituent, presumably TP, into SpecCP, which creates the word order exhibited by examples such as (117) above (cf. Kayne 1994; Julien 2002 for similar analyses of SOV order; Julien 2002: 40 explicitly mentions an analysis of Buryat along these lines).70 CP

(118)

C’

TP (S) ti OV

C C

tTP clitici

Ø

On the assumption that in the history of Buryat, structures such as (118) provided the syntactic environment for the reanalysis of postverbal pronouns as agreement suffixes on the verb, the following scenario can be hypothesized for the grammaticalization process in question:

JB[v.20020404] Prn:2/08/2005; 15:48

F: LA8105.tex / p.57 (3685-3725)

Chapter 5. The reanalysis of C-oriented clitics 

(119) [CP [TP tj ’ [T’ [νP tj [VP object]] V+T]]i [C’ C+pronounj tTP ]] → [CP [TP [T’ [νP pro [VP object]] V+T+Agr]]] Structural simplification: loss of TP movement; subject movement replaced by an Agree-relation Preservation of argument structure: θ-role previously assigned to overt pronoun is now assigned to pro Identification of feature content: via Agree with the subject (pro) in SpecνP Word building constraint: verb is located in T (via movement) or adjacent to T at MS (enabling Morphological Merger) As already noted above, the reanalysis in question was presumably facilitated by the pro-drop properties which already existed in the language prior to the structural change in (119), providing a means to fulfill Preservation of argument structure by the insertion of an empty pronoun the vacant θ-role could be assigned to. The change described in (119) leads to a major structural reorganization in which TP movement and overt subject movement to SpecTP are lost. This seems to be the most likely outcome of the reanalysis in question if the requirement of structural simplification is understood as a pressure to posit the least complex structure compatible with the linguistic evidence. Still, other less dramatic changes are also conceivable. For example, we might assume that the reanalysis in question does not affect TP-movement, but merely leads to the loss of overt subject movement (via insertion of pro in SpecνP) which suffices to meet the requirement of structural simplification. In any case, the reanalysis in question differs from all other changes discussed so far in that the resulting agreement morpheme is realized on a functional head different from the head the clitic pronoun was attached to (thus, it seems that linear adjacency is sufficient for the reanalysis to take place). Note that the verb and the pronoun do not form a head complex in the target grammar. Instead, the verb is contained within TP (either in situ or moved to T), while the pronoun has undergone movement to C. In contrast, in the structure resulting from the reanalysis, the agreement morpheme is not attached to C, but part of the verb’s inflectional domain, presumably adjoined to T. Hence, similar to Uto-Aztecan, the reanalysis leads to a syntactic agreement morpheme that enters the syntactic derivation as an adjunct to a contentful functional head. Moreover, it must be ensured that the verb can combine with the agreement morpheme prior to Spell-out. By assumption, this is achieved by verb movement in (119). Note, however, that in principle, OV-languages always allow the verb to stay in situ and combine with the agreement morpheme on T via Morphological Merger at MS, due to the fact that in a strict OV grammar, the verb is always string-adjacent to the set of right functional heads. In fact, this alternative seems to be more economi-

JB[v.20020404] Prn:2/08/2005; 15:48

F: LA8105.tex / p.58 (3725-3783)

 The Rise of Agreement

cal than the derivation involving verb movement. Still, I prefer to stick to the analysis in (119), which is more in line with the well-established generalization that rich verbal morphology goes hand in hand with overt verb movement (whatever the ultimate reason for this correlation may be).

. Conclusion In this chapter, I have taken a closer look at the reanalysis of C-oriented clitics as markers of verbal agreement, showing that this change is shaped by a variety of different syntactic factors which (partially) derive from the way agreement is established in natural languages. It has been demonstrated that the grammaticalization process in question may take place in a number of different scenarios, depending on the specific syntactic properties of the languages under consideration. In the Germanic and Rhaeto-Romance varieties discussed in this chapter, the V2 property has been shown to play an important role in the reanalysis of C-oriented enclitics by creating the necessary adjacency between subject clitic and finite verb, which uniformly gives rise to suffixal agreement morphology. In non-V2 languages, the transition from C-oriented clitics to verbal agreement markers may be triggered in alternative environments, either leading to prefixal agreement in SVO languages where the verb moves to T (UtoAztecan) or suffixal agreement in SOV languages where by assumption the whole TP is fronted to a position to the left of the clitic (Mongolian). Another point of variation involves the realization of the θ-role assigned to the pronominal clitic in the target grammar. On the one hand, the relevant θ-role may be assigned to a phonologically empty pronoun. It has been argued that this change either exploits a pro-drop strategy already available prior to the reanalysis (as in Mongolian) or widens the range of pro-drop, leading to new pro-drop properties absent in the target grammar (as in Bavarian). In this way, the rise of agreement is intimately linked to the pro-drop phenomenon. On the other hand, doubling configurations provide an alternative path to new agreement formatives by assigning the vacant θ-role to the former double (as in the Swiss RR dialect Sutselvan). Additionally, languages may differ with respect to the nature of the agreement morpheme which results from the reanalysis of a clitic pronoun. In languages which develop agreement from scratch, the grammaticalization process in question directly results in a new syntactic agreement morpheme (e.g., Agr-on-T), which by assumption enters the syntactic derivation adjoined to a contentful functional head. In languages with an existing verbal agreement paradigm, however, the creation of new agreement formatives usually proceeds via an initial stage where a clitic is reanalyzed as the realization of a post-syntactically inserted dissociated Agr-morpheme (e.g., Agr-on-C), the licensing of which is parasitic on the presence of an Agr-morpheme that has been valued during the syntactic derivation. In a subsequent development, the phonological exponent of this newly created dissociated Agr-morpheme may then replace the original exponent of the (hierarchically lower) syntactic agreement morpheme (e.g.,

JB[v.20020404] Prn:2/08/2005; 15:48

F: LA8105.tex / p.59 (3783-3837)

Chapter 5. The reanalysis of C-oriented clitics 

Agr-on-T). As pointed out, this particular pathway also represents a counterexample to the claim that grammaticalization always leads to new exponents of higher functional heads, as proposed by Roberts and Roussou (2003). These findings motivate the following slightly modified grammaticalization cline which captures the distinction between different types of agreement morphemes (see (45) above, repeated here for convenience): (120) free pronoun → weak pronoun → clitic pronoun → (dissociated Agr- morpheme) → syntactic Agr-morpheme → Ø Thus, even if we concentrate on changes affecting C-oriented clitics, it is evident that the reanalysis of pronominal elements as agreement formatives can come about in a variety of different syntactic environments. Accordingly, attempts to reduce the grammaticalization process in question to a single syntactic scenario seem to be misguided. Still, I have demonstrated that the different historical pathways share a set of clearly identifiable common properties due to the universal nature of the restrictions the reanalysis of clitics is subject to. That is, all structural changes discussed in this chapter meet (and are shaped by) the requirements of Structural simplification (the outcome must be less complex than the target structure), Preservation of argument structure (the relevant θ-role must be assigned either to a former double or pro), Identification of feature content (the content of the newly created Agr-morpheme must be valued), and the Word building constraint (the Agr-morpheme must combine with the verb prior to Vocabulary Insertion). Inasmuch as these conditions derive from the general theory of agreement laid out in Chapter 3 above, the diachronic data examined in this chapter provides further support for the particular approach to syntactic agreement advocated in this book. In particular, the hypothesis that Agr-morphemes do not occupy a unique position in the structure of the clause, but rather are parasitic on contentful functional categories such as C, T, or ν, is corroborated by the observation that in V2 languages such as Bavarian, Cimbrian, Alemannic and Rhaeto-Romance, new agreement formatives are initially confined to the C head before they replace the original exponent of the ‘canonical’ subject agreement morpheme on T in a subsequent development. In addition, the fact that this historical development typically leads to instances of multiple agreement is successfully captured by an analysis according to which the change in question proceeds via an intermediate stage where clitics are reanalyzed as the realization of a dissociated morpheme which is inserted and valued post-syntactically. Finally, note that the particular analyses presented in this chapter can be taken to alleviate the problems raised by the observation that cross-linguistically, agreement markers tend to be realized as suffixes (the so-called suffixing preference, see Chapter 1): If we take a look at the scenarios discussed in this chapter, it appears that most of them give rise to suffixal agreement markers. Thus, a study of the transition from C-oriented clitics to agreement markers seems to support the conjecture (cf. Chapter 4) that the suffixing preference should receive a historical explanation, in the sense that it results at least partially from the way agreement markers come into existence

JB[v.20020404] Prn:2/08/2005; 15:48

F: LA8105.tex / p.60 (3837-3900)

 The Rise of Agreement

historically. More specifically, it seems that cross-linguistically, the grammaticalization process in question is more likely to be triggered in syntactic environments which give rise to suffixal agreement morphology. At this point, it should again be stressed that the syntactic factors discussed in this chapter are all taken to restrict the grammaticalization process in question. In other words, they are necessary, but crucially not sufficient conditions for the change to take place. This raises the question of whether it is possible to identify any causal factors which lead the learner to reanalyze pronouns as agreement markers. This question is dealt with in some more detail in the following chapter, where it is argued that the ultimate trigger for the reanalysis of pronominal elements is morphological in nature.

Notes . However, see Nübling (1992) for an analysis of the relevant elements as pronominal enclitics. . In a number of dialects, the 1pl clitic -ma has developed similar properties as the 2nd person forms, cf. Section 5.2.1.3. . Thus, it is assumed that the forms 2sg /-st/, 2pl /-ts/ are solely agreement markers; in other words, there are in fact no 2nd person clitics in most Bavarian dialects. Otherwise, the question would arise of why it is not possible to add a clitic onto an inflected verb or complementizer in C, giving again rise to forms such as *kummts-ts or *ob-ts-ts. The lack of 2nd person clitics is presumably the result of the reanalysis of these forms as part of the verbal agreement morphology, see below. Note, however, that there are some Carinthian dialects (spoken e.g. in Pernegg, cf. Lessiak 1963) where such forms seem to exist with 1pl mr (which apparently developed into another instance of Agr-on-C in these dialects, see below): (i)

Kher-m6=mr wir a¯ aufn? belong-1pl=clit.1pl we up ‘Do we belong at the top?’

(ii) wi6-mr=mr wir wöln the way-1pl=clit.1pl we want ‘The way we want (it)’ (Pernegg, Lessiak 1963: 204) The multiple realization of -mr can be accounted for if we assume that in this dialect (in contrast to other Bavarian varieties), the reanalysis of the clitic as an instance of Agr-on-C did not lead to the loss of the clitic itself, but rather to a situation where the source and the result of the grammaticalization process in question continue to exist side by side in the grammar (which is actually a common feature of grammaticalization processes; cf. Chapter 2). . The early OHG manuscripts written in the monastery of Fulda show this change in the process of its development, cf. the Hildebrandslied (preserved in an early 9th century copy of the original text dating from the late 8th century), the Basel Recipes (around 800), or the Tatian (translated around 830–840. This translation was then copied in the second half of the 9th century); see Brinkmann (1931), Moulton (1944), Sievers (1961), and Sommer (1994) for details.

JB[v.20020404] Prn:2/08/2005; 15:48

F: LA8105.tex / p.61 (3900-3972)

Chapter 5. The reanalysis of C-oriented clitics  . Note that verbs such as ‘fear’ may optionally take V2 complement clauses in German. Therefore, the verbs nimist ‘take-2sg’ and arnost ‘earn-2sg’ can be taken to occupy the C-head of an embedded V2 clause. . Here, one might speculate that the enlargement -t in (9) is actually to be analyzed as a 2sg enclitic. This would explain the absence of the enlargement in the presence of the full pronoun thu in (10), on the assumption that clitics and full pronouns are in complementary distribution in OHG. However, in the OHG Tatian, there is at least one example where a full pronoun and a clitic form do co-occur: (i)

Eno thu bistu mera unsaremo fater Iacobe... qprt you are-clit.2sg more our father Jacob ‘Are you more powerful than our father Jacob...?’ (Tatian, γ 87.3; Sievers 1961: 119)

. This was pointed out to me by Tom McFadden. . This predicts the existence of a historical stage of Bavarian that exhibits (apparent) doubling phenomena in main and embedded clauses as a result of Agr-on-C, but crucially lacks the new ending on verbs in clause-final position. . Rowley (1994: 494) reports that a similar development can be observed for 3pl in some North Bavarian and East Franconian dialects. In the following examples, an overt 3pl subject is accompanied by the relative marker wf which carries the (verbal) agreement ending 3pl -n: (i)

dez wfu-n fn6r! vf’dein moust-n that what-3pl others earn must-3pl ‘The things that others have to work for’ (dialect of Erkersreuth)

(ii) wf-n z! wøl-n what-3pl they want-3pl ‘What they want’ (dialect of Kastl) . Recall that in contrast to Bavarian, the reanalysis that led to the rise of Agr-on-C did not lead to the loss of the clitic, but gave rise to a situation where the source and the result of the grammaticalization process continue to exist side by side in the new grammar, resulting in the unusual form of triple subject marking in this dialect. . These dialects are spoken in an area the boundaries of which are (roughly) marked by Cham in the west, Lam in the east, Furth i. W. in the north, and Kötzting in the south, cf. Kollmer (1987, I). . Note that even if pro moved to SpecTP, the requirement of structural simplification would still be satisfied, due to the loss of clitic movement to C. However, recall that I have argued above (in Chapter 4) that empty pronouns cannot undergo overt movement since they fail to provide SpecTP with phonological content, which would be necessary to satisfy the EPP (cf. Holmberg 2000). . Thus, it appears that the change illustrated in (22)–(23) apparently led to the loss of an EPP feature in T in the contexts where the reanalysis took place. Given the often stated universality of the EPP (cf. e.g. Chomsky 1995, 2000; Collins 1997), this raises the question of whether such a change is possible at all. However, it is a well-known fact that subjects may remain in a lower (presumably νP-internal) position in German (cf. Grewendorf 1989; Haider 1993). This suggests that T may lack an EPP-feature in German. Accordingly, I am led to assume that the reanalysis in question was facilitated by this special trait of German.

JB[v.20020404] Prn:2/08/2005; 15:48

F: LA8105.tex / p.62 (3972-4052)

 The Rise of Agreement . Recall that in V2/V1 contexts, replacement of Agr-on-T by Agr-on-C is required by the assumption that in a complex head adjunction structure, only the highest Agr-morpheme (of a set of identical Agr-morphemes) is spelled-out (cf. Chapter 3, Note 62). . However, one might argue that in examples such as (27a–c), there is actually a complementizer dass ‘that’ present to which the inflectional ending attaches (cf. Harnisch 1989; Nübling 1992). On this assumption, one can maintain that the set of hosts which Agr-on-C selects for is rather limited (only C0 elements, i.e., complementizers and finite verbs). This analysis is supported by the fact that the complementizer can also be overtly present in the above examples, cf. (i)

Du soll-st song [CP [ an wäichan Schuah] (dass)-st [IP du wui-st ]]]. which shoe-2sg that-2sg you want-2sg you should-2sg say ‘You should say which shoe you want.’

. Interestingly, the same kind of doubling is found with objects, in contrast to Bavarian: (i)

a.

b.

(ii) a.

b.

vorsch=me net miar. ask=clit.1sg.acc not me ‘Don’t ask me.’ dar vorscht=de di. he asks=clit.2sg.acc you ‘He asks you.’ (Lusern; Tyroller 2003: 157) du pôxes=mar miar. you scold=clit.1sg.dat me.dat ‘You scold me.’ i bölta=ter dir contraren... I wished=clit.2sg.dat you.dat tell ‘Iwished to tell you...’ (Roana; Schweizer 1952: 28)

This might be taken to suggest that Cimbrian develops some form of object agreement previously absent in the grammar, similar to present-day Spanish (cf. e.g. Suñer 1988). . This development goes hand in hand with the rise of new full pronominal forms which are clearly influenced by the neighboring Italian dialects: wir/endsch andre ‘we others’, ir/ouw andre ‘you.pl others’. . This is reminiscent of the expansion of 1pl -ma to verbs in clause-final position in some Lower Bavarian dialects, see Section 5.2.1.3 above. . The distribution of ‘clitics’ and full pronouns parallels the distribution of pro and full pronouns in Standard Italian (and the 2nd person forms in Bavarian): “In Pomattertitsch stressed subject pronouns (as well as nouns) mainly occur in emphatic, focused or contrastive contexts, with an explicit referential function. They are usually placed pre-verbally but their position is in fact rather free. On the other hand, clitics always occur in post-verbal position in all pragmatically unmarked contexts.” (Dal Negro 2004: 164). . Dal Negro (2004: 178f.) speculates that the change in question is triggered by the fact that it serves to repair an otherwise defective agreement paradigm (see Chapter 6 for some discussion). . Interestingly, however, the verb-clitic sequences are very frequent in embedded contexts where they apparently replace the ‘German’ pattern in which clitics normally attach to the complementizer in embedded clauses, cf. the following examples in which the conjunction is followed by V1-orders (inversion):

JB[v.20020404] Prn:2/08/2005; 15:48

F: LA8105.tex / p.63 (4052-4117)

Chapter 5. The reanalysis of C-oriented clitics 

(i)

a.

b.

ds morgen-tsch wen si=wer fertig gsi im gadä the.gen morning-gen when are=we finished been in.the stall ‘In the morning, when we had finished working in the stall’ der manut wa si=wer döi gsi the month that are=we over.there been ‘During the month when we were over there’ (Dal Negro 2004: 168)

Indirect questions play a somewhat special role in that they exhibit this type of inversion with all kinds of subjects (clitics and full nominals) in the speech of all informants consulted by Dal Negro: (ii) gang löge wa ischt der nonno go look where is the grandfather ‘Go looking where the grandfather is.’ (Dal Negro 2004: 168) . Cf. the following statement taken from Dal Negro (2004: 166f.): “Also in Pomattertitsch ich and du are the most frequently omitted pronouns, although other pronouns can be omitted. [...] As a matter of fact, zero subjects in Walser dialects can be interpreted as one of the possible outcomes of unstressed subject pronouns. In the case of zero subject sentences both the phonic form and the referential strength of the pronoun are null. Besides, the grammaticality of zero subject sentences provides the necessary background that allows the interpretation of verb-subject clitic sequences as units of verb followed by its ending in which no explicit subject is present and the information on personal agreement is carried by the new agglutinative endings. [my emphasis, E.F.]” . On the assumption that agreement morphemes usually exhibit narrow selectional restrictions concerning their host, this outcome actually seems to be more likely than the development of complementizer agreement observed in Bavarian. . The fact that the development of 2sg /-st/ is not confined to Bavarian, but rather was already well underway in early Old High German, should lead us to expect that pro-drop and complementizer agreement (i.e., Agr-on-C) can be found in other German varieties as well in the context of 2sg. While standard German does not allow referential pro-drop, there are other non-standard varieties apart from Bavarian in which pro-drop is available with 2sg subjects, cf. the following examples from Swabian (Carola Trips, personal communication) and Zurich German (Weber 1987: 174): (i)

Hascht pro de Peter gesehe? have-2sg the Peter seen ‘Have you seen Peter?’

(ii) Häscht pro e gsee? have-2sg him seen? ‘Have you seen him?’

Swabian

Zurich German

In other words, it appears that referential pro-drop is actually quite widespread in (nonstandard) varieties of German in the context of 2sg. Concerning the presence of Agr-on-C, note that there is at least one example in the OHG Tatian in which a full pronoun co-occurs with an apparent clitic. Since clitic doubling is otherwise not attested in OHG, the ‘clitic’ is presumably to be analyzed as the realization of Agr-on-C:

JB[v.20020404] Prn:2/08/2005; 15:48

F: LA8105.tex / p.64 (4117-4200)

 The Rise of Agreement

(iii) Eno thu bistu mera unsaremo fater Iacobe... qprt you are-2sg more our father Jacob ‘Are you more powerful than our father Jacob...?’ (Tatian, γ 87.3; Sievers 1961: 119) Accordingly, one might speculate that complementizer agreement had originally been available in other varieties of German as well (albeit restricted to 2sg) and that it disappeared when the exponent of Agr-on-C was reanalyzed as the canonical realization of 2sg agreement (the loss of this dialectal trait was presumably furthered by sociolinguistic factors, that is, the development of a standard language). In contrast, further reanalyses of enclitic subjects leading to new formatives for 2pl (/-ts/) and 1pl (/-ma/) reinforced this grammatical option in Bavarian. . Linder (1987: 35) attributes this to the fact that some authors/speakers consider inversion to be an ‘un-RR’ Germanic trait that should be avoided in ‘pure’ RR. Interestingly, most examples cited by Oetzel (1992) involve a configuration where a scene-setting adverbial precedes the subject. This is reminiscent of the fact that in many older Germanic languages such as Old English and Old High German, we can observe similar violations of the V2 constraint in connection with scene-setting adverbs (cf. Fuß 2003 for some discussion), cf. the following examples from Old English: (i)

a.

b.

[Æfter þeossum wordum] [se Hælend] cwæþ to his after these words the Savior spoke to his (Blickling 135; Swan 1994: 241) [Her] [Oswald se eadiga arceb] forlet in-this-year Oswald the blessed archbishop forsook (ASC, Laud (992); Kroch & Taylor 1997: 304)

leornerum... disciples þis lif. this life

. Other differences between the Swiss RR dialects and the Germanic V2 languages concern the placement of negation and the behavior of clitic clusters. In Puter and Vallader, the sentential negation nu(n) precedes the finite verb in inversion contexts, cf. (i)

A Gustav nu pudaiv’la cuntradir... to Gustave not could-she contradict ‘She could not oppose Gustav...’ (Puter; Oetzel 1992: 28)

(ii) Ma il min brün nun han=a pudü verer. but the my brown not have=they can see ‘But they could not see my brown (cat).’ (Vallader; Oetzel 1992: 62) This resembles the behavior of negation in many other Romance languages. Note that similar facts can be observed in Old English and Old High German, where sentential negation is expressed by a clitic which accompanies verb movement to C, cf. the following example from the Old High German Isidor translation: (iii) endi mina milthissa ni nimu ih ab imu. and my compassion not take I from him ‘And I will not take my compassion from him.’ (Isidor, 627; Robinson 1997: 16) Another deviation from the kind of V2 observed in Germanic can be found in Puter, where the proclitic subject pronoun a can amalgamate with an object clitic, resulting in a clitic cluster in clause-initial position (Linder 1987: 104): (iv) A-t racumand mia Nanigna! clit.1sg.nom-clit.2sg.dat recommend my Nanigna ‘I recommend my Nanigna to you!’

JB[v.20020404] Prn:2/08/2005; 15:48

F: LA8105.tex / p.65 (4200-4268)

Chapter 5. The reanalysis of C-oriented clitics 

(v) A-l faregia sager qualchosa d’bun. clit.1sg.nom-clit.3sg.acc make-fut try something good ‘I will make him try something good.’ Note that (v) is particularly interesting, since it represents an example of clitic climbing where the subject of the embedded infinitival fuses with the subject of the causative. . Recall that in Puter and Vallader, the sentence negation nu(n) is a clitic that attaches to the left of the verb and accompanies verb movement to higher head positions (cf. note 26). Therefore, the relative position of the finite verb and negation cannot be used as a diagnostic of the position of the finite verb in these dialects. . Another series of agreement endings (the so-called ‘secondary desinences’, cf. Haiman and Benincà 1992 for details) is found in the imperfect indicative, the present subjunctive, the imperfect subjunctive, and the future(s). The secondary desinences are generally unstressed, which gave way to phonological reduction and a higher degree of syncretism. . At least in the Engadin dialects Puter and Vallader, the accusative precedes the dative in the unmarked order, in contrast to German. In the other Swiss RR varieties, the picture is less clear, possibly due to the influence of German. . Note that (60b) also exhibits a violation of the V2 constraint, since the object pronoun intervenes between the wh-phrase and the finite verb. . According to Linder (1987: 13f.), in Puter, the reduced form -a can be used as a C-oriented enclitic for 1sg and 3pl as well (attaching to the finite verb or a complementizer). In the latter usage, it competes with the (older) variants 1sg -i and 3pl -e. The enclitic -a is also found as 3sg.neut when its host ends in a consonant, and as 1pl if the resulting word complies with certain stress rules (‘penultimate stress target’, Haiman 1971; see below for details). -a is not found as a 2pl enclitic. In contrast, the corresponding vocalic clitic i of Surmeiran is apparently restricted to proclitic contexts (at least in the written language). . The distinction between 3pl.masc and 3pl.fem full forms is found in all five Swiss RR dialects under investigation. In addition, Puter and Vallader exhibit feminine variants of 1pl and 2pl (nussas and vussas, respectively). The individual dialects differ with respect to the shape and the availability of the clitic forms. . The dialects under investigation differ with respect to the frequency of enclitic pro-drop. For example, Linder (1987: 34) notes that enclitic pro-drop is much more frequent in Puter than in Vallader. . Haiman and Benincá (1992: 93) suggest that this development was caused by a pressure for paradigmatic coherence: in the first conjugation, 2sg and 3sg had two syllables, while 1sg had only one. This difference was then leveled by suffixing another vowel. . Haiman and Benincà (1992: 95) argue that the old ending for 1pl -(a)in is also an innovation, which derived historically from a former 3sg pronoun that was reinterpreted as an agreement suffix for 1pl (within the pronominal paradigm, a similar development is exhibited by French on): “First, the 1st plural was expressed by homo/unus + 3sg. [...] Second, this PRO form appeared postverbally in inverted order as a clitic. Finally, -VN was reinterpreted as a bound suffix on the verb stem, obligatory in both direct and inverted word order.” . Widmer (1959: 95) notes that in the translation by Bifrun there are still also examples which reflect the older usage. In the following example (Bifrun, Cor. I, 9.4), the verb exhibits the old, non-enlarged ending in an inversion context, with the 1pl pronoun following the verb:

JB[v.20020404] Prn:2/08/2005; 15:48

F: LA8105.tex / p.66 (4268-4341)

 The Rise of Agreement

(i)

Nun haivan nus forsa da mangiêr & da baiver? not have-1pl we power to eat and to drink ‘Have we not power to eat and to drink?’

. Other Swiss varieties of RR such as Surselvan and Sutselvan (and the majority of the RR dialects spoken in Northern Italian) still show the original agreement ending 2sg -s. . Note that this correlation holds only for Puter, Vallader and Surmeiran in an obvious form. Surselvan and Sutselvan show pro-drop in 2nd person contexts although they do not exhibit the innovation -s+t. Here we might speculate that the latter dialects underwent a similar historical process as Puter, Vallader, and Surmeiran (i.e., a reanalysis of enclitic -t(i)), but lost final -t in a later development (note that loss of final -t can also be observed in spoken present-day Puter and Vallader, cf. Linder 1987: 56). Alternatively, it is possible to assume that pro-drop with 2nd persons is the retention of an older system of pro-drop (similar to Old French, cf. Roberts 1993a; Vance 1997), as suggested by Linder (1987: 56). . Haiman and Benincà (1992: 97) note a similar development in the Gorizian dialect of Friulian, where the 2pl atonic pronoun subject -o has developed into an invariable suffix on the verb: (i)

o fevel-ez-o you.pl talk-2pl ‘You talk.’

. In the 2pl present indicative, Vallader exhibits the ending -aivat, which is unique among the Swiss RR dialects (e.g., chant-aivat ‘sing-2pl’). Traditionally, it is assumed that this form also involved the reanalysis of a clitic form and proceeded in two steps (cf. Gartner 1883; Widmer 1959; Linder 1987). First, a reduced form of the 2pl pronoun vos (clitic va without final -s; a similar form vo still exists in present-day Puter) was reanalyzed as an enlargement of the existing 2pl ending -ai, cf. the following example from early Vallader (Chiampel, Ps. 58) cited in Widmer (1959: 99): (i)

Pud-aiw wuo foars’ilg uaira dyr? can-2pl you perhaps=the truth say ‘Can you perhaps tell the truth?’

In a second step, the ending /-t/ was added in analogy to the other tenses where 2pl is signaled by the agreement suffix /-t/. A somewhat simpler account is proposed by Haiman and Benincà (1992), who argue that Vallader borrowed the unusual ending from the almost identical imperfect paradigm. . As noted above in Section 5.2.2, this analysis raises some questions concerning the status of the EPP. These are probably even more apparent in the present case since it seems that in the RR languages, the subject moves to SpecTP without exception, in contrast to German/Bavarian. However, note that even under the more traditional assumption that pro occupies SpecTP in the RR dialects under investigation, the requirement of structural simplification would still be satisfied, due to the loss of clitic movement to C. . As already mentioned, similar facts can be observed in Non-Standard French and a number of Northern Italian dialects (cf. Chapter 6 for more detailed discussion). . Note that different clitics exist for 3sg.masc and 3sg.fem forms. With 3rd person plural subjects, no gender distinctions are marked by the clitic. Doubling with 3rd person neuter forms is only found in Surmeiran, where it is apparently restricted to wh-questions:

JB[v.20020404] Prn:2/08/2005; 15:48

F: LA8105.tex / p.67 (4341-4410)

Chapter 5. The reanalysis of C-oriented clitics 

(i)

Ma tge è-gl chegl? but what is-clit.3sg.neut that ‘But what is that?’ (Surmeiran; Linder 1987: 156)

. Note that (77b) is a very early example of clitic doubling, taken from Bifrun’s bible translation (1560), where -s presumably is still a clitic form. This might be taken to suggest that doubling structures were also involved in the early development of 1pl final -s, in contrast to what has been claimed above. However, all other instances cited by Linder (1987) and Haiman and Benincà (1992) lack a postverbal full pronoun. In addition, the assumption that the new agreement ending -ains developed from doubling structures would leave us without an explanation for the fact that 2nd person contexts usually license pro-drop in RR. . Linder (1987) does not give any examples for clitic doubling with 1st person subjects in Vallader. . Note that there are no examples of doubling with 2nd person subjects, since there are no enclitic pronouns for 2nd person available. . A related idea is proposed in Simpson and Wu (2002), who suggest that in general, AgrPs develop out of FocPs that are originally selected by some higher functional head (cf. Chapter 1 for discussion). . See Grewendorf (2002) for an analysis of left dislocation in German which is based on the assumption that the dislocated phrase and the resumptive pronoun start out together as a big DP. . As pointed out to me by Ede Zimmermann, one might wonder whether the loss of stylistic force should not be analyzed as an earlier, separate structural change, which manifests itself in a different LF, for example. However, it seems to me that it is not necessary to posit such an intermediate stage. More specifically, I maintain that a direct reanalysis of a formerly marked construction is possible if the resulting structure is compatible with the original informationstructural function in the target grammar. For example, the reanalysis of a former topic as a subject does not necessarily lead to a change in information structure, since (definite/referential) subjects act as default topics in the unmarked case (cf. e.g. Givón 1976). Accordingly, a reanalysis as ‘neutral syntax’ is possible as soon as the corresponding structure is no longer unambiguously marked (e.g., by a special intonation) as a topic construction. . The reanalysis in question was presumably promoted by another trait of Sutselvan that further weakened the evidence for a pronominal status of the clitics. As noted above, in all Swiss RR dialects, the structure of finite verbs is governed by a stress rule which forbids that the main word stress falls on the antepenultimate syllable (the so-called ‘penultimate stress target’ Haiman 1971; cf. Linder 1987; Haiman & Benincà 1992 for discussion). In some contexts, attachment of the clitic forces elision of the original verbal agreement ending (3sg.fem. -a in the example at hand) to ensure that the antepenultimate syllable does not receive stress: (i)

ella Áchant-a → *Áchant-a=la → Áchant=la she sing-3sg sing-3sg=clit.3sg.fem sing=clit.3sg.fem

In other words, adherence to the ‘penultimate stress target’ leads to surface forms where a clitic apparently replaces the original verbal agreement morphology. Given that the necessary conditions for a reanalysis are fulfilled in Sutselvan (see below), it is reasonable to suggest that examples such as (i) tipped the scales in favor of a reinterpretation of the former clitics as agreement markers (see Chapter 6 for discussion).

JB[v.20020404] Prn:2/08/2005; 15:48

F: LA8105.tex / p.68 (4410-4494)

 The Rise of Agreement . I will not attempt to provide a detailed analysis of these word order facts here apart from suggesting that variable word order is associated with the optional projection of TopPs/FocPs within the middle field/IP domain (cf. Grewendorf 2004). . The following example from Sutselvan (Linder 1987: 240) seems to suggest that doubling/Agron-C is actually confined to main clauses (recall that doubling/Agr-on-C is otherwise almost obligatory in Sutselvan): (i)

Tscheart ear’ igl [c’igls pocs da nus earan egnea stos a Casti...] certain was it that=the very-few of us were once been to Casti ‘It was certain that very few of us have ever been to Casti.’

However, doubling is possible if the embedded clause exhibits inversion/V2. The following example from Puter exhibits doubling in a V2 clause introduced by the conjunction per che ‘since’: (ii) Sullet uain el dit, [per che da natüra e-l el sullet filg da Only is he mentioned since by nature is-clit.3sg.masc he only son of Dieu...] God ‘Only he is mentioned, since by nature, he is the only son of God...’ (Puter, Linder 1987: 151) . The term ‘asymmetric V2’ is sometimes also used to refer to the Germanic V2 languages that show a main/embedded asymmetry with respect to the position of the finite verb (in contrast to ‘symmetric’ V2 languages such as Icelandic and Yiddish). . Note that the order (98b) is presumably ruled out for independent reasons: after movement to SpecTP, the full nominal must occupy a preverbal position. Furthermore, if both the clitic and the full DP are merged together in a big DP headed by the clitic, movement of the clitic to SpecTP presumably cannot strand the subject in a lower position. This would require that the double is scrambled out of the big DP to a position below T (simply labeled FP here). In a subsequent operation, the big DP-remnant (containing the clitic) moves to SpecTP, thereby crossing the double: (i) *[CP [TP [DP ti [D’ clitic ...]]j [T’ V+T [FP doublei ... [νP tj ... Following Grewendorf (2003), such a derivation can be ruled out as an instance of Improper Remnant Movement, due to the fact that the movement that creates the remnant (in the case under consideration, scrambling of the subject) is of a higher type than the movement operation that affects the remnant subsequently (A-movement). . Note that Classical Aztec is another example of a language that lacks 3rd person agreement markers (cf. Chapter 1 and Chapter 6 for discussion). . Thus, similar to other grammaticalization processes discussed so far, the change in question affects certain pronominal elements before it affects others. Generally speaking, it seems that 1st and 2nd person elements are leading the charge (see the discussion and references in Chapter 1; see Chapter 6 for an explanation). . In many Uto-Aztecan languages, C-oriented clitics do not show up in second position if certain inflectional markers occupy this position. In these cases, the pronominal clitic usually attaches to the right of these markers, cf. the following quote from Steele (1977: 549) and the examples from Luiseño in (i): “In fact, elements which correspond roughly to what has been

JB[v.20020404] Prn:2/08/2005; 15:48

F: LA8105.tex / p.69 (4494-4563)

Chapter 5. The reanalysis of C-oriented clitics 

analyzed as the AUX in English – tense, aspect, and modality – commonly occur in sentential second position, as do conjunctions, negatives, and question-markers.” (i)

a.

b.

c.

nanatmalum xu=m=po henge‘malumi ‘ari. girls modal=clit.3pl=asp boys kick ‘Girls should kick boys.’ nawitmal qay hengeemali ‘ariq. is.kicking girl neg boy ‘The girl isn’t kicking the boy.’ nawitmal=su hengeemali ‘ariq? girl=qmarker boy is.kicking ‘Is the girl kicking the boy?’

. Further support for an analysis in terms of late repositioning comes from the fact that clitic placement does not respect (syntactic) phrasal boundaries. In the following example from Tarahumara, the clitic attaches to the first word of the fronted DP, separating the noun and the adjective: (i)

[semati=ne napaha] rarimea aré. nice=clit.1sg shirt gonna.buy probably ‘I am probably going to buy a nice shirt.’ (Steele 1977: 554)

. Further evidence for this assumption is provided by an 18th century grammar of Yaqui (Velasco 1737), where it is explicitly stated that subject pronouns must occur in second position. In other words, it can be concluded that the prefixal person/number markers which are similar in shape to the second position clitics have developed subsequently. . Still, there are indications that this grammaticalization process has not yet been completed in Tarahumara and Yaqui. Steele (1977: 553f.) observes that the prefixal person/number markers are not obligatory in these languages. In Tarahumara, the presence of the verbal prefix is bound to the presence of a second position clitic, that is, second position clitics can occur without the verbal prefix but not vice versa, cf. (i). In contrast, all three logical possibilities can be observed in Yaqui: second position clitics can co-occur with person/number prefixes (as in (113)), and both elements can occur alone, cf. (ii) and (iii) (all examples from Steele 1977: 554): (i)

semati=ne napaha rarimea aré. nice=clit.1sg shirt going.to.buy probably ‘I am probably going to buy a nice shirt.’

(Tarahumara)

(ii) tuká=ne antónyta biˇcak. yesterday=clit.1sg Antonio saw ‘Yesterday, I saw Antonio.’

(Yaqui)

(iii) ‘inepo ne-‘a-me‘ak. I 1sg-it-threw ‘I threw it.’

(Yaqui)

These differences can be analyzed as an instance of grammar competition (Kroch 1989 and Chapter 2 above), where speakers have command over more than one internalized grammar: an older grammar A, producing the sentences without the person/number prefix on the verb and a newer grammar B where former second position clitics have been reanalyzed as (obligatorily present) verbal agreement markers. The fact that in Tarahumara, the agreement marker cannot occur without a second position clitic suggests that the reanalysis in question has not given rise to a

JB[v.20020404] Prn:2/08/2005; 15:48

F: LA8105.tex / p.70 (4563-4636)

 The Rise of Agreement

pro-drop grammar: in grammar B, an overt pronominal element (e.g., a second position clitic) must be present in addition to the verbal agreement marker, similar to Standard French, for example. See Note 64 below for an alternative explanation that links the distribution of agreement markers to the syntactic environments (word order) where the change in question was licensed. . Steele (1977: 554ff.) argues that at least in Tepecano, the verbal prefixes developed via a ‘direct’ reanalysis of second position clitics, which possibly does not require an intermediate stage of clitic doubling. Evidence for this claim comes from Tepecano examples where the element in question can apparently act as both a prefix/proclitic on the verb and a second position clitic: (i)

ndedos n=an=ahohoinda. my.fingers introducer=1sg=will.shake.them ‘I will shake my fingers.’ (Tepecano; Steele 1977: 543)

Note, however, that the verbal person/number markers of Tepecano otherwise behave differently than second position clitics. For example, they can occur in clause-initial position, as in (ii), where the ‘clitic’ attaches to an auxiliary-like tense/aspect marker. (ii) an=ti nauw kGGs. 1sg=t/asp napal transplanted ‘I transplanted napal.’ (Tepecano; Steele 1977: 555) . As noted above, the grammaticalization process in question is not completed in languages such as Tarahumara and Yaqui which still show competition between old, non-prefixed verbs and new prefixed forms (presumably an instance of grammar competition). Steele (1977: 557) assumes that the change from second position clitics to verbal prefixes (i.e., stage 4 in Table 9) has been promoted by the fact that the verb attracts grammatical elements such as clitics (and modals, auxiliaries etc.): “It is the attraction of the verb for grammatical elements that encourages the development of clitic pronouns that are proclitic (or prefixed) to the verb from second position clitic pronouns, whether by copying them there or by reanalysis.” This intuition can be rendered into more formal terms by saying that the verb occupies a functional head such as T, which by assumption can host an agreement morpheme. This configuration facilitates the reanalysis of non-stressed clitics that are adjacent to the verb as markers of verbal agreement, especially in instances of doubling. . For the sake of concreteness, I suppose that (weak) subject pronouns still move to SpecTP at this stage. Following ideas by Roberts (1998), I assume that the set of [–interpretable] features located in T can trigger overt movement of pronominal elements since the latter consist of nothing but a bundle of formal features (cf. Fuß 2003 for a similar analysis of V3 orders with pronominal subjects in Old English). Further reduction of the pronoun presumably leads to a reanalysis as X0 -element which undergoes head movement into the inflectional domain (with the second position effects still resulting from repositioning at MS, see above). In contrast, full nominal subjects reside in their θ-position in the languages under consideration, presumably due to the absence of an EPP feature in T. Accordingly, the double in languages exhibiting stage 3 must remain in a lower position as well. Alternatively, it may undergo A’-movement into the left periphery, which seems more or less in line with the facts observed above. . In Classical Aztec and Pochutla, SVO or VOS orders are most common; Huichol exhibits regular OSV order (presumably “derivative from an older SVO stage”, Steele 1977: 556); Cora exhibits VSO order. Steele (1977) gives no information concerning the basic word order of Tarahumara, Tepecano and Yaqui. According to the Ethnologue database (http://www.ethnologue.com/) Tarahumara and Yaqui show basic SOV order, while Southeastern Tepehuan, the closest relative

JB[v.20020404] Prn:2/08/2005; 15:48

F: LA8105.tex / p.71 (4636-4704)

Chapter 5. The reanalysis of C-oriented clitics 

to Tepecano (Tepecano is by now extinct), shows both VSO and VOS orders. The SOV character of Tarahumara and Yaqui seems to pose a problem for the present analysis, since it is obvious that a second position clitic is not adjacent to the verb in this word order. Interestingly, however, it seems that these languages show a set of alternative word orders which apparently licensed the development of new verbal person/number prefixes, while the prefixes are still absent in SOV sentences: (i)

semati=ne napaha rarimea aré. nice=clit.1sg shirt going.to.buy probably ‘I am probably going to buy a nice shirt.’

(ii) tuká=ne antónyta biˇcak. yesterday=clit.1sg Antonio saw ‘Yesterday, I saw Antonio.’

(Tarahumara)

(Yaqui)

Thus, it seems that the distribution of verbal agreement prefixes still reflects the environments where their grammaticalization was licensed (absent in SOV orders, present in other orders). Note that this idea possibly can be used as a diachronic explanation of the fact that the verbal prefixes are sometimes absent in Tarahumara and Yaqui (cf. Note 60 above). . In VSO/VOS languages, adjacency between the verb and the C-oriented clitic is guaranteed if the verb moves to T to create V1-order (that is, the subject stays behind in a lower position, as assumed above). In SVO languages, the clitic moves to the left of the finite verb, while the double stays behind in a lower position or moves up into the left periphery (see Note 63 above). . As numerous other languages, Buryat lacks pronouns for 3rd person subjects. Instead, demonstratives are used to refer to 3rd person entities (Poppe 1960: 53). . It can be shown that the development of subject agreement suffixes predates the development of the possessive suffixes. Apart from Buryat, the ‘new’ verbal agreement markers are found in Oirat, Kalmyk, Dagur, and Moghol. In Dagur and Moghol, the relevant suffixes are very similar to the respective nominative pronouns, showing no signs of phonetic reduction (cf. Martin 1961; Comrie 1980). . See Comrie (1980: 90f.) for a similar conclusion. A similar word order option can still be found in colloquial Khalkha (Standard Mongolian): (i)

a.

b.

bi med-ne. I know-pres ‘I know.’ med-ne bi. ‘I know.’

However, in contrast to Classical Mongolian, colloquial Khalkha does not allow doubling in these contexts (Comrie 1980: 91), cf. (ii) *bi med-ne bi I know-pres I ‘I know.’ . Furthermore, Comrie assumes that the possibility of a reanalysis is sensitive to the major affixation pattern in a given language. Thus, in an overwhelmingly suffixing language such as Buryat, a clitic is more readily reanalyzed as an inflectional affix if it follows the verb, in accordance with the general direction of affixation in the language in question.

JB[v.20020404] Prn:2/08/2005; 15:48

F: LA8105.tex / p.72 (4704-4719)

 The Rise of Agreement . To be sure, more data and discussion are needed to evaluate the proposal in (118). For example, if all SOV orders of Classical Mongolian are taken to follow from this structure, we should expect that there is no wh-ex-situ, since SpecCP is already occupied by the fronted TP. Interestingly, this expectation seems to be borne out by the facts, lending further support to an analysis of SOV order in terms of TP-fronting (examples taken from Poppe 1954: 173): (i)

a.

b.

ˇci qamiγ-a-aˇca irebei? you where-from come-past ‘From where did you come?’ ˇci yaγun abumui? you what take ‘What will you take?’

JB[v.20020404] Prn:2/08/2005; 15:55

F: LA8106.tex / p.1 (45-143)

Chapter 6

Morphological blocking and the rise of agreement

. Introduction This chapter examines the way the grammaticalization of verbal agreement markers is shaped by properties of the morphological component of grammar. More specifically, it is claimed that the reanalysis of subject clitics is triggered by a morphological principle which scans the input for the most specific realization of a given agreement morpheme. This claim is based on the observation that the change in question does not replace existing agreement markers in a random fashion. Rather, it can be shown that the creation of new forms affects only those cells of the paradigm where the existing verbal agreement morphology is not distinctive. Thus, we can often observe that clitics turn into obligatorily present markers of verbal agreement only in some person/number combinations, whereas they continue to be merely optional in other contexts. The distribution of the obligatory clitics interacts in interesting ways with properties of the existing verbal agreement morphology. For example, in the Northern Italian dialect Piattino (Gerlach 2001, 2002) clitics are obligatory only in contexts where the existing suffixal agreement morphology fails to signal person and number of the subject in an unambiguous way (see Section 6.5.1 on the special status of 3pl, where the existing ending seems to be distinctive). This is illustrated in Table 1 (boldface signals that the clitic is obligatory). From Table 1 it becomes clear that the presence of subject clitics is merely optional for 1sg, 2sg and 2pl, but obligatory in all other contexts (i.e., 3sg, 1pl, and 3pl). Hence, it appears that the presence of the clitic is required only in those environments where the existing verb form is underspecified for agreement features. Thus, it seems Table 1. Subject agreement and clitics in Piattino.

1sg 2sg 3sg.masc 3sg.fem 1pl 2pl 3pl

Clitic + verb

Presence of the clitic

(a) guardi (te) guardesc al guarda la guarda an guarda (ve) guardé li guarden

optional optional obligatory obligatory obligatory optional obligatory

JB[v.20020404] Prn:2/08/2005; 15:55

F: LA8106.tex / p.2 (143-199)

 The Rise of Agreement

that the grammaticalization of new means of agreement marking serves to repair a previously defective agreement paradigm. Below, it is shown that similar phenomena can be observed in various languages, including German varieties, Non-Standard French, Northern Italian and Rhaeto-Romance dialects. Of course, this chapter cannot do justice to the vast amount of descriptive literature on the grammaticalization of inflectional morphology. Still, the discussion of selected examples suggests that (1) represents a valid generalization on the grammaticalization of agreement morphology across languages. (1) New verbal agreement morphology arises only for those slots of the agreement paradigm where the existing verbal inflection is non-distinctive. It is argued that (1) follows from innate principles which govern the acquisition of inflectional morphology. More specifically, I claim that the reanalysis of pronominal elements as functional agreement heads is triggered if the change leads to the elimination of syncretism in an agreement paradigm. This is modeled as the outcome of blocking effects which operate during language acquisition and block the acquisition of a less specified form if a more specified candidate is attested in the Primary Linguistic Data. The chapter is organized as follows. Section 6.2 introduces the main theoretical claim put forward in this chapter, namely that the grammaticalization of inflectional morphology is shaped by a principle (dubbed the Blocking Principle) which operates during language acquisition and requires new inflectional material to be more specified/distinctive than the relevant inflectional formatives already present in the language. This idea is motivated by taking a closer look at morphological aspects of the development of new verbal agreement markers in the history of Bavarian in Section 6.2.1. In the sections 6.2.2 and 6.2.3, I discuss a set of theoretical issues which arise in connection with the Blocking Principle (concerning the status of morphological doublets and competing changes which give rise to less specified forms), thereby sharpening and partially reshaping the account in terms of morphological blocking developed previously. In Section 6.3, it is shown that the Blocking Principle offers a new explanation for the observation that 1st and 2nd person play a pioneering role in the grammaticalization of agreement markers across languages (cf. Chapter 1). Sections 6.4–6.8 illustrate the workings of the Blocking Principle in the grammaticalization of agreement markers with examples from Non-Standard French (Section 6.4), Northern Italian dialects (Section 6.5), Rhaeto-Romance (Section 6.6), American Russian and Walser German (Section 6.7), and Skou, a Papuan language spoken in New Guinea (Section 6.8). Some of these languages have already been commented on in Chapter 5, which focused on syntactic aspects of the change in question. Note that Uto-Aztecan and Mongolian are not included in the discussion, since the Blocking Principle is trivially fulfilled in languages which completely lacked verbal agreement prior to the grammaticalization of relevant markers. Section 6.9 provides a summary of this chapter.

JB[v.20020404] Prn:2/08/2005; 15:55

F: LA8106.tex / p.3 (199-256)

Chapter 6. Morphological blocking and the rise of agreement

. The Blocking Principle It is a long-standing idea in morphological theory that inflectional paradigms do not tolerate doublets for a given paradigmatic slot. In recent approaches, this insight is usually modeled as a blocking effect which ensures that the availability of a more specified morphological form blocks the use/insertion of a less specified form, cf. the following quote from Sauerland (1996: 4) (for discussion of blocking effects cf. Kiparsky 1973, 1983; Aronoff 1976; Anderson 1986, 1992; Halle & Marantz 1993; Kroch 1994; Sauerland 1996; Halle 1997, Noyer 1997):1 If one form possesses all the properties of another form and some more specific properties, the more specified form must be used wherever possible.

In a realizational model of grammar such as Distributed Morphology, this idea can be modeled in terms of conditions on the process of Vocabulary Insertion (cf. Chapter 2 above), cf. the italicized passage of Halle’s (1997) Subset Principle in (2): (2) The Subset Principle (Halle 1997: 428) The phonological exponent of a Vocabulary item is inserted into a morpheme in the terminal string if the item matches all of a subset of the grammatical features specified in the terminal morpheme. Insertion does not take place if the Vocabulary item contains features not present in the morpheme. Where several Vocabulary items meet the conditions for insertion, the item matching the greatest number of features specified in the terminal morpheme must be chosen. In other words, if more than one Vocabulary item is compatible with the feature content of a certain terminal node, the most specified Vocabulary item must be inserted. This trait of the insertion procedure implies that phonological exponents may be underspecified with respect to the featural content of the morpheme they realize. That is, for insertion to take place, a phonological exponent need not be specified for all features present in a terminal node. This section serves to introduce the central theoretical claim of the present chapter, namely the idea that the rise of new verbal agreement morphology is guided by morphological blocking effects as well. More specifically, it is claimed that morphological blocking guides the acquisition of Vocabulary items to the effect that a new verbal agreement marker can only be coined if it is specified for a greater subset of agreement (i.e., φ-) features than a competing agreement formative that occupies the relevant paradigmatic slot in the target grammar. The relevant principle of grammar, the so-called Blocking Principle, is motivated by inspecting morphological details of the grammaticalization of new verbal agreement markers in the history of Bavarian (cf. Chapter 5 above). In addition, it is argued that these assumptions offer a new (diachronic) explanation of the person/number restrictions on complementizer agreement and pro-drop observed for Bavarian in Chapter 5.



JB[v.20020404] Prn:2/08/2005; 15:55

F: LA8106.tex / p.4 (256-357)

 The Rise of Agreement

.. The rise of new verbal agreement endings in Bavarian A closer look at morphological aspects of the rise of new agreement suffixes in Bavarian reveals that the individual diachronic developments share an interesting characteristic: the grammaticalization of 2pl /-ts/ and 1pl /-ma/ resolved existing homophony in the verbal agreement paradigm. The grammaticalization of 2pl /-ts/ (< clit. -(¯e)s) began in the 13th century (in Northern and Middle Bavarian, cf. Wiesinger 1989: 72f.). Prior to this development, the verbal agreement paradigm exhibited homophonous forms for 3sg and 2pl. Thus, the development of a new distinctive ending for 2pl can be said to have repaired this ‘defect’ of the paradigm. The relevant facts are shown in Table 2 (the 3sg, 2pl forms are marked by boldface): Table 2. Verbal agreement paradigms (pres. indic.), 13th century Bavarian.

1sg 2sg 3sg 1pl 2pl 3pl

Old paradigm

New paradigm

-Ø -st -t -an -t -ant

-Ø -st -t -an -ts -ant

Interestingly, it can be shown that a similar situation existed in those varieties that developed a new agreement ending for 1pl, /-ma/. Due to phonological reduction, final /-t/ was lost in the 3pl, leading to homophony of 3pl and 1pl forms in most Bavarian dialects in the 18th century. In some dialects, this form of syncretism was resolved by the development of 1pl /-ma/ (from the relevant subject clitic) as a new agreement ending (initially confined to C, see Section 5.2.1): Table 3. Verbal agreement paradigms (pres. indic.), late 18th century Bavarian.

1sg 2sg 3sg 1pl 2pl 3pl

Old paradigm

New paradigm

-Ø -st -t -an -ts -an

-Ø -st -t -ma -ts -an

We can thus conclude that both changes apparently repaired a previously defective agreement paradigm.2 Interestingly, it seems that this observation is a general characteristic of the grammaticalization of agreement markers across languages. Thus,

JB[v.20020404] Prn:2/08/2005; 15:55

F: LA8106.tex / p.5 (357-418)

Chapter 6. Morphological blocking and the rise of agreement 

new forms do not develop randomly, affecting any arbitrary slot of the paradigm. Instead, the development of new inflectional formatives is confined to those slots of the paradigm where the existing verbal inflection is not distinctive (e.g., due to phonological erosion). This has already been noted in the introduction to this chapter with respect to the development of new prefixal agreement markers in the Northern Italian dialect Piattino (Gerlach 2002). Similar observations can be made in various other Italian and Rhaeto-Romance dialects (cf. e.g. Spiess 1956; Kuen 1957; Renzi & Vanelli 1983; Linder 1987; Haiman & Benincà 1992; Poletto 1997), Non-Standard French (Wartburg 1970; Ashby 1977; Harris 1978; Auger 1993; Gerlach 2002), the Walser German dialects discussed in Chapter 5 above, and American Russian (Polinsky 2000). Further examples from somewhat more exotic areas include Khinalug (NorthEast Caucasian, Corbett 1991: 123), the Austronesian language Kisar (Blood 1992; Siewierska 1999, 2004), the Yuman language Maricopa (Siewierska 1999, 2004), and the Papuan language Skou (Donohue 2002, see Section 6.8).3 Let’s therefore assume that (3) represents a valid descriptive generalization on the grammaticalization of new verbal agreement morphology: (3) New verbal agreement morphology arises only for those slots of the agreement paradigm where the existing verbal inflection is non-distinctive. In other words, the reanalysis of clitics as agreement markers is licensed if the change leads to the elimination of syncretisms in an agreement paradigm. Of course, this requirement is trivially fulfilled if new agreement formatives develop in a language that previously showed no agreement morphology at all, as for example in the Uto-Aztecan and Mongolian languages discussed in Chapter 5 above. At the same time, (3) rules out the possibility that the grammaticalization of new agreement formatives leads to the replacement of inflectional markers which are fully distinctive. However, note that (3) is merely a description, but no explanation. Therefore, it would be desirable if (3) could be shown to follow from deeper principles of grammar. In this chapter, it is argued that the generalization in (3) is the outcome of blocking effects that operate during language acquisition and block the acquisition of a less specified form if a more specified form is attested in the Primary Linguistic Data. In a Late Insertion model such as Distributed Morphology, this general idea can be formalized as in (4):4 (4) Blocking Principle If several appropriate PF-realizations of a given morpheme are attested in the Primary Linguistic Data, the form matching the greatest subset of the morphosyntactic features included in the morpheme must be chosen for storage in the lexicon. The Blocking Principle is to be understood as a cognitive economy principle which applies during language acquisition and guarantees an optimal and non-redundant lexicon (and paradigm) structure. Note that this kind of morphological (or lexical) economy differs significantly from syntactic/structural economy principles that are part of the computational system and shape the course of syntactic derivations (cf.

JB[v.20020404] Prn:2/08/2005; 15:55

F: LA8106.tex / p.6 (418-457)

 The Rise of Agreement

e.g. Chomsky 1995; Collins 1997; Kitahara 1997). The latter exhibit usually a ‘least effort’ character, in the sense that they favor less complex, that is, less marked syntactic derivations/representations (cf. Clark & Roberts 1993; Roberts 1993b; Roberts & Roussou 2003 on the role of ‘least effort’ principles in language change). In contrast, the Blocking Principle prefers more specific, that is, more marked lexical entries over less marked ones. Still, this can be seen as a form of economy as well, as noted above.5 Similar to structural economy principles, the Blocking Principle is called into service only if the cues provided by the input data are for some reason ambiguous and not sufficient for the acquisition of certain properties of the grammar. Thus, in cases where the continued phonological erosion of a clitic leads to a situation where the trigger experience contains more than one potential exponent for a given agreement head/morpheme,6 the BP is invoked to decide which of the candidates is stored in the lexicon. The ultimate decision is then taken by comparing the feature specifications of the competing potential phonological exponents, favoring the candidate which realizes the greatest subset of the feature content of the underlying agreement head/morpheme. If prior to the reanalysis, the language lacked an agreement marker for the paradigmatic slot in question, the BP is trivially fulfilled, of course.7 In this way, the BP ensures that the development of new inflectional formatives can affect only weak/underspecified slots of the paradigm, replacing Vocabulary items that are not distinctive. Coming back to the historical developments observed in Bavarian, it can be shown that the new agreement suffixes 2pl /-ts/ and 1pl /-ma/ satisfy the Blocking Principle due to the fact that they are more specified than their respective predecessors. Turning first to the change that took place in 13th century Bavarian, a closer look at Table 2 shows that the formative /-t/ occurs in 3sg and 2pl contexts, that is, the relevant Vocabulary item must be underspecified for [person] as well as [number].8 In other words, /-t/ represents the elsewhere case and is inserted as the default agreement ending: (5) elsewhere ↔/-t/ Hence, the introduction of 2pl /-ts/ was licensed by the Blocking Principle, since the new form is specified for [person] and [number], resolving the existing homophony between 3sg and 2pl:9 (6) [2, pl] ↔/-ts/ Similar facts can be observed in 18th century Bavarian. There are only two different plural forms in the agreement paradigm: /-ts/ is inserted in the context [2, pl], whereas /-an/ realizes the feature combinations [1, pl] and [3, pl]. Thus, the lexical entry for /-an/ must be underspecified for the feature [person]. In other words, /-an/ is simply the elsewhere case among the plural forms, cf. the following insertion rules: (7) a. [2, pl] ↔/-ts/ b. [pl] ↔/-an/

JB[v.20020404] Prn:2/08/2005; 15:55

F: LA8106.tex / p.7 (457-536)

Chapter 6. Morphological blocking and the rise of agreement 

Again, the potential new realization of 1pl (/-ma/) is more specified than the existing Vocabulary item /-an/, since it is in addition specified for [person]. This state of affairs facilitates the grammaticalization process in question, leading to a fully distinctive set of plural agreement markers:10 (8) a. [1, pl] ↔/-ma/ b. [2, pl] ↔/-ts/ c. [pl] ↔/-an/ It now becomes clear that the person/number restrictions on complementizer agreement (and pro-drop) which we observed in Chapter 5 above reflect the workings of the Blocking Principle in the grammaticalization of new verbal agreement morphology in the history of Bavarian. As noted above, the Blocking Principle licensed the development of new agreement formatives only for those slots of the agreement paradigm where the existing phonological exponents were less specified than their newly emerging competitors, that is, in 2nd person and 1pl contexts. Accordingly, the reflexes of this historical process (which leads to the presence of a dissociated Agr-morpheme on C, see Chapter 5 above) – complementizer agreement and pro-drop – are found only in these particular slots of the paradigm in the present-day language. So far, I have shown that an account in terms of the Blocking Principle provides a new explanation for the question of why new agreement suffixes developed only for 2pl and 1pl forms in the history of Bavarian (still reflected by the restrictions on complementizer agreement and pro-drop in the present-day language). In the next section it is argued that this approach can also handle the earlier grammaticalization process which created the 2sg suffix /-st/ during the Old High German period.

... The diachrony of 2sg /-st/ As already noted above in Chapter 5, it is commonly assumed that the form /-st/ resulted from a reanalysis of -s+thu sequences, in which the onset of the pronoun (or the enclitic -t(u)) was mistakenly analyzed by the learner as part of the verbal agreement ending (in analogy with the 2sg of the preterite-presents). As noted in Chapter 5, this process could only take place in inversion contexts, where the finite verb preceded the subject clitic. However, it seems that the development of 2sg /-st/ presents a problem for an account in terms of the BP. Consider the forms listed in Table 4:11 Table 4. Agreement paradigms (pres. indic.) for nëmen ‘take’, early OHG.

1sg 2sg 3sg 1pl 2pl 3pl

Old paradigm

New paradigm

nim-u nim-is nim-it nëm-emês (-êm, -ên) nëm-êt nëm-ant

nim-u nim-ist nim-it nëm-emês (-êm, -ên) nëm-êt nëm-ant

JB[v.20020404] Prn:2/08/2005; 15:55

F: LA8106.tex / p.8 (536-642)

 The Rise of Agreement

A look at Table 4 suggests that the change from 2sg /-s/ to /-st/ apparently did not involve the creation of a Vocabulary item that is more specific than its predecessor. Both items seem to realize the same set of morphosyntactic features, cf. (9) a. [2, sg, pres.] ↔/-s/ b. [2, sg, pres.] ↔/-st/ Thus, it appears that the creation of the new ending /-st/ conflicts with the BP, since it apparently does not lead to a more specified form. In what follows, however, I will argue that the change in question proceeded in accordance with the BP, since it helped to create an inflectional formative for 2sg that was additionally specified for verbal mood, in contrast to its predecessor. Thus, the feature specifications of the relevant Vocabulary items are actually those in (10). (10) a. [2, sg, pres.] ↔/-s/ b. [2, sg, pres., indic.] ↔/-st/ In early OHG, the 2sg endings of many verbs were identical in the present indicative and the present subjunctive, while verbal mood was clearly distinguished in other person/number combinations (apart from 2pl). This is illustrated in Table 5 and Table 6 for the weak verbs salbôn ‘anoint’ (conjugation class 2) and the very frequent habên ‘have’ (conjugation class 3), which exhibit the characteristic inflections of their respective verb classes.12 Table 5. Conjugation of salbôn ‘anoint’ (class 2, present tense), early OHG.

1sg 2sg 3sg 1pl 2pl 3pl

Present indicative

Present subjunctive

salbôm salbôs salbôt salbômês salbôt salbônt

salbo salbôs salbo salbôm salbôt salbôn

Table 6. Conjugation of habên ‘have’ (class 3, present tense), early OHG.

1sg 2sg 3sg 1pl 2pl 3pl

Present indicative

Present subjunctive

habêm habês habêt habêmês habêt habênt

habe habês habe habêm habêt habên

JB[v.20020404] Prn:2/08/2005; 15:55

F: LA8106.tex / p.9 (642-686)

Chapter 6. Morphological blocking and the rise of agreement 

A closer look at the Tables 5 and 6 reveals that identical verb forms show up in the slots for 2sg present indicative and 2sg present subjunctive. We can therefore conclude that the forms for 2sg are underspecified with respect to verbal mood. Interestingly, the development of the new formative /-st/ began in the present indicative (cf. Brinkmann 1931). This suggests that the development in question was licensed by the fact that the new ending was unambiguously specified for verbal mood (i.e., indicative), in contrast to the earlier formative /-s/. Thus, despite appearances, the original change leading to 2sg /-st/ does not represent a counterexample to the BP. Rather, it proceeded in accordance with the requirement that new inflectional formatives realize a greater subset of morphosyntactic features than their predecessors. In a later development, the new ending spread via analogical extension to all verb classes, tenses and verbal moods including the present subjunctive. This subsequent development blurred the original motivation for the change in question and eliminated the mood distinction in the 2sg. It is important to note, however, that the original grammar change leading to the creation of 2sg /-st/ can be shown to comply with the Blocking Principle if a larger portion of the inflectional paradigm (including the distinction of verbal mood) is taken into account, thereby providing further support for the central claim of this chapter. To sum up, this section has introduced the central theoretical claim of this chapter, namely that the grammaticalization of agreement markers is guided by blocking effects which shape the acquisition of inflectional morphology. The Blocking Principle has been motivated by looking at the history of Bavarian, arguing that an approach in terms of morphological blocking offers an explanation for the observation that the development of new agreement formatives affected only underspecified slots of the agreement paradigm. In addition, it has been suggested that the peculiar person/number restrictions on complementizer agreement and pro-drop that can be observed in present-day Bavarian should receive a diachronic explanation, in the sense that they reflect the workings of the Blocking Principle in the grammaticalization process under consideration. So far, I have focused on the basic mechanisms of the Blocking Principle, glossing over several theoretical and empirical issues which arise in connection with this approach to language acquisition and change. Some of these are discussed in some more detail in the following two sections.

.. Morphological blocking versus analogical leveling In this section, I discuss the interaction of grammaticalization processes governed by the Blocking Principle with other forces in language change which have opposite effects and lead to the loss of distinctions and the creation of less marked, more regular forms. It is well-known that in many languages, the historical development of verbal agreement paradigms (and inflectional morphology in general) is characterized by the loss of distinctive morphology (the Germanic languages, in particular English and Mainland Scandinavian are often-studied examples).13 It is immediately clear that this observation appears to constitute a serious problem for the assumption that the acqui-

JB[v.20020404] Prn:2/08/2005; 15:55

F: LA8106.tex / p.10 (686-755)

 The Rise of Agreement

sition of inflectional morphology is guided by something like the Blocking Principle, repeated here for convenience as (11). (11) Blocking Principle If several appropriate PF-realizations of a given morpheme are attested in the Primary Linguistic Data, the form matching the greatest subset of the morphosyntactic features included in the morpheme must be chosen for storage in the lexicon. If a principle such as (11) is assumed to be operative during the course of language acquisition, we should expect that the loss of inflectional distinctions is considerably hindered and that the learner should always go for the most specified form attested in the input. Still, we can observe that in most languages, the effects of phonological erosion and analogical leveling win out over the processes that create new inflectional morphology. While phonological erosion presumably does not constitute a real issue in connection to the Blocking Principle (when phonological processes lead to the erosion of inflectional morphology, the relevant forms simply disappear from the input data and fail to be acquired), the phenomenon of analogical leveling appears to be more problematic, since it leads to the replacement of more specified, irregular forms by less specified and more regular forms that are absent in the target grammar (cf. e.g. Bybee 1985; Hock 1991; Albright 2002 for discussion). In other words, the learner does not acquire a non-distinctive form due to phonological erosion. Rather, he/she chooses to expand the domain of a more regular inflection, substituting less distinctive forms for more distinctive forms. A typical example of analogical change with respect to agreement marking is the loss of stem vowel alternations in the development of the English past tense forms, as shown in Table 7 for the verb ‘to drink’. In Old English, agreement distinctions were not only marked by different agreement suffixes, but also by alternations of the verb’s stem vowel (in the past tense). However, in the historical development of the relevant forms, the vowel changes were lost and drank surfaced as the only past tense form of the verb drink with the stem vowel /æ/ generalized to all person/number combinations. This represents a typical example of analogical change, leading to a set of forms that are more regular and show less distinctions than the inventory found in the target grammar (note that this Table 7. Analogical leveling of agreement distinctions in the historical development of to drink.

1sg 2sg 3sg 1pl 2pl 3pl

Old English past tense forms

Expected Modern English past tense forms

Actual Modern English past tense forms

dranc drunce dranc druncon druncon druncon

drank drunk drank drunk drunk drunk

drank drank drank drank drank drank

JB[v.20020404] Prn:2/08/2005; 15:55

F: LA8106.tex / p.11 (755-838)

Chapter 6. Morphological blocking and the rise of agreement 

Table 8. Analogical leveling of agreement distinctions in Yiddish.

1sg 2sg 3sg 1pl 2pl 3pl

Present tense of grfbn ‘to dig’

Present tense of visn ‘to know’

grfb grebst → grfbst grebt → grfbt grfbn grfbt grfbn

veys veyst veys(t) visn → veysn vist → veyst visn → veysn

change cannot be attributed to phonological erosion, since the latter affected only non-stressed final syllables, but crucially not the verb stem in the history of English). Similar changes affected the verb’s stem vowel in the history of Yiddish present tense forms, where the 2nd and 3rd singular forms of ‘to dig’ and the plural forms of ‘to know’ have been changed to match the relevant verb stem found in the 1sg, cf. Table 8 (see Albright 2002 for detailed discussion).14 From Table 8 it becomes clear that the process of analogical leveling led to the loss of agreement distinctions in the paradigm. In the case of grfbn ‘to dig’, the change conflated the forms for 3sg and 2pl, which were formerly distinct due to the existence of stem vowel alternation. Similarly, analogical leveling led to homophonous forms for 2sg and 2pl of visn ‘to know’. These changes raise a problem for an approach in terms of the Blocking Principle: the elimination of alternations resulting from analogical leveling is not expected if it is assumed that the acquisition procedure prefers more specified over less specified forms. In order to address this problem, some more has to be said on the nature and possible analysis of analogical leveling. Traditionally, it is often assumed that the drift towards more uniformity within a paradigm is a natural development towards a situation where, ideally, a single form corresponds to a single meaning/function (in the case at hand, a single non-alternating form of the verb stem). In other words, it is assumed that learners/speakers tend to accept a change eliminating stem alternations in a given paradigm.15 Still, this kind of explanation runs counter the idea that the acquisition of inflectional paradigms is governed by a principle that singles out the most specified candidate (robustly) attested in the input. In an interesting new approach, Albright (2002) analyzes analogical leveling as an effect of the way learners acquire the base form of the verb in a given paradigm. Albright argues that learners scan the input for a base form of the verb which allows to generate unknown forms as accurately as possible: The way that they do this [...] is by seeking a base form within the paradigm that is “maximally informative” – that is, that suffers the least serious phonological and morphological neutralizations – and then deriving the remaining forms in the paradigm from the base form. (p. 7)

Albright uses the term neutralization to refer to the disappearance of (morpho-) phonemic contrasts in a certain stem form due to the affixation of inflectional ma-

JB[v.20020404] Prn:2/08/2005; 15:55

F: LA8106.tex / p.12 (838-916)

 The Rise of Agreement

Table 9. Neutralization due to voicing assimilation in Yiddish (present tense).

1sg 2sg 3sg 1pl 2pl 3pl

lib6n ‘to love’

zip6n ‘to sift’

lib lipst lipt lib6n lipt lib6n

zip zipst zipt zip6n zipt zip6n

terial. He illustrates the effects of neutralization with an example from Yiddish, where – similar to German or English – the affixation of a suffix consisting of voiceless obstruents requires that a root-final obstruent become voiceless as well, cf. Table 9. This phonological process leads to the neutralization of morphophonemic contrasts in the 2sg, 3sg and 2pl, where it is no longer clear whether the base form ends in a voiced (as in lib6n) or voiceless (as in zip6n) obstruent. Albright then goes on to show that of all base forms of the present tense paradigm, the 1sg has a unique status in Yiddish, since it is much less affected by phonological neutralization than all other base forms. For this reason, the 1sg form is more informative than any other base form as far as phonemic distinctions are concerned and can therefore be considered the maximally informative base form. For example, in the case of lib6n ‘to love’, only the 1sg base form would serve to unambiguously differentiate the verb lib6n from another possible (but non-existing) verb *lip6n ending in a voiceless obstruent. How does this relate to the phenomenon of analogical leveling? Albright claims that the learner has to select a single surface form as the base form of the verb when he acquires a given inflectional paradigm (the so-called single surface base hypothesis). By assumption, this form is always the “maximally informative” candidate attested in the input (in the case at hand the 1sg form). In other words, children acquiring an inflectional paradigm seek out the base form which manifests the distinctive phonological properties of the verb in question (number of phonemes, voicing etc.) in the most unambiguous and informative way. After the learner has detected the most suitable candidate, he/she goes on to derive the remainder of the paradigm from the base form. Of course, the learner may be later compelled to store separate forms with stem alternations if they are robustly attested in the relevant input data. However, if for some reason he/she fails to pay attention to the relevant data, he/she will go on to use the over-regularized forms, which will eventually replace the old irregular forms if the change gains a wider distribution in the speaker community.16 In Yiddish, this kind of analogical leveling affected the present tense paradigm, in which the 1sg developed into the only base form for all slots in the paradigm (see Table 8 above). Interestingly, under these assumptions, it appears that even analogical leveling can be attributed to the workings of morphological blocking, in the sense that the learner selects the stem alternant which is most informative/specified with respect to phone-

JB[v.20020404] Prn:2/08/2005; 15:55

F: LA8106.tex / p.13 (916-985)

Chapter 6. Morphological blocking and the rise of agreement 

mic distinctions as the base form of a given lexical root. This intuition is phrased in somewhat more formal terms in (12). (12) Blocking Principle – phonemic distinctions If several realizations of a given lexical root are attested in the Primary Linguistic Data, the form manifesting unambiguously the greatest number of phonemic distinctions is chosen for storage in the lexicon. The fact that the effects of analogical leveling seem to win out over the effects of the Blocking Principle (as defined above in (4) for the acquisition of inflectional distinctions) can then be explained if it is assumed that the acquisition of phonemic distinctions (and base forms) is more fundamental than the acquisition of inflectional morphology and therefore has priority over the latter, thereby bleeding the effects of the ‘inflectional’ Blocking Principle in the historical development of languages. A related but somewhat different scenario arises in cases where paradigm leveling does not affect stem allomorphy, but rather inflectional affixes, in the sense that a certain affix gains a wider distribution in a paradigm. As a rule, this kind of change involves the expansion of underspecified, ‘elsewhere’ formatives to other slots of the paradigm. This can be illustrated with a change that affected the plural forms in the verbal agreement paradigm of many Alemannic dialects. The majority of these dialects spoken in Switzerland and Southwest Germany exhibit only a single plural ending for all persons, -ed (so-called Einheitsplural). Interestingly, it can be shown that this ending was originally confined to 3pl (-end prior to elision of /n/) and spread later to the other persons, affecting first 2pl forms (cf. e.g. Paul 1952: 194; Weber 1987: 173; Paul 1998: 240). This change is illustrated in Table 10. In other words, the plural agreement morpheme originally confined to 3rd person evolves into the general plural morpheme, replacing more specified morphemes that formerly signaled person distinctions. Following Noyer (1997: lxxx–lxxxi), I suggest that this kind of change is to be analyzed as the extension (or development) of Impoverishment rules which delete a subset of the morphosyntactic features of a given inflectional head in a certain context (cf. Chapter 2 above).17 Recall that Impoverishment generally leads to the insertion of a less marked exponent, due to the fact that the impoverished inflectional head is no longer compatible with the feature matrix of a more specified Vocabulary item, leading to an “expansion of the domain of the unmarked exponent” (Halle 1997: 431). In the change at hand, the relevant Impoverishment rule would lead to the deletion of person features in plural contexts, thereby

Table 10. The development of Einheitsplural in Alemannic (pres. indic.)

1pl 2pl 3pl

Original paradigm

Intermediate Stage

New paradigm

-en -ed -end

-en -end -end

-ed -ed -ed

JB[v.20020404] Prn:2/08/2005; 15:55

F: LA8106.tex / p.14 (985-1047)

 The Rise of Agreement

forcing the insertion of the default plural ending -ed (which is specified only for the feature [+pl]):18 (13) [person] → Ø /[+pl] Here, it is important to note that Impoverishment and morphological blocking operate in different parts of the grammar and give rise to different kinds of historical effects: while Impoverishment rules are part of MS and affect the output of the syntactic derivation, that is, the feature content of terminal nodes, the Blocking Principle governs the acquisition of Vocabulary items realizing inflectional heads (and, if a principle such as (12) is on the right track, the acquisition of lexical roots as well), favoring more specified over less specified Vocabulary items. In this way, the effects of the (inflectional) Blocking Principle can be thwarted by the expansion of Impoverishment rules: after deletion of morphosyntactic features in the relevant underlying inflectional heads, forms which realize a greater set of features simply cannot be inserted (or created) and give way to less specified, unmarked inflectional formatives. It is therefore perhaps not accidental that in most languages, the effects of paradigm leveling and phonological erosion win out over the processes that create new inflectional morphology.19 In the next section, I will deal with another problem in relation to morphological blocking, namely the observation that the grammaticalization processes discussed so far do not lead to an instant replacement of the old forms by the new forms. Rather, the change in question introduces a more specified inflectional formative which replaces the older, less specified form first in a certain context before it gradually gains a wider distribution, apparently leading to the existence of morphological doublets.

.. On the status of morphological doublets In Chapter 5 above, I have noted that in the history of Bavarian the development of new agreement suffixes proceeded via an intermediate stage where the new form appears on verbs that occupy C (in inversion contexts), while the original (less specified) agreement suffix is still found on verbs in clause-final position (i.e., in embedded clauses). This is illustrated for the rise of 1pl /-ma/ (replacing the less specified form /-a(n)/) by the following pair of examples from Lower Bavarian:20 (14) Mia gem-ma/*geng-an hoam. we go-1pl/go-1pl home ‘We are going home.’ (15) wa-ma hoam laff-a/*laff-ma. because-1pl home go-1pl ‘Because we are going home’ At first sight, one would perhaps expect the Blocking Principle (and the Subset Principle) to rule out this kind of variation. First, if the insertion of phonological exponents

JB[v.20020404] Prn:2/08/2005; 15:55

F: LA8106.tex / p.15 (1047-1110)

Chapter 6. Morphological blocking and the rise of agreement 

is governed by the Subset Principle, the availability of a more specified form (1pl /-ma/) should always block the use of a less specified competing form (1pl /-a(n)/). Second, if the Blocking Principle is assumed to operate during language acquisition, one would perhaps expect that the presence of a more specified form in the Primary Linguistic Data prevents the acquisition of the less specified form in the first place. However, the fact that most modern theories of morphology assume some version of morphological blocking does not mean that languages never show morphological doublets. On the contrary, competing forms of a certain word or grammatical formative can be shown to be quite common in the world’s languages. Well-known examples include co-existing (new) regular and (old) irregular past tense and participle forms in German (taken from Bittner 2004), cf. (16) a. backte (regular) vs. buk (irreg.) ‘I/he/she/it baked’ b. erbleichte (regular) vs. erblich (irreg.) ‘I/he/she/it paled’ c. glimmte (regular) vs. glomm (irreg.) ‘I/he/she/it glowed’ (17) a. gegärt (regular) vs. gegoren (irreg.) ‘fermented (participle)’ b. geschnaubt (regular) vs. geschnoben (irreg.) ‘snorted (part.)’ c. gebleicht (regular) vs. geblichen (irreg.) ‘bleached (part.)’ Note that in contrast to the competition between the forms in (14) and (15), which arose via the grammaticalization of a new, more specific agreement ending for 1pl, the variation illustrated in (16) and (17) is again the result of analogical leveling which creates more regular (i.e., in a certain sense less specific) alternatives to the existing inventory of forms. After some period of coexistence, the regular forms usually drive the old forms out of the grammar. The latter type of change leading to morphological doublets is discussed at length in Kroch (1994), who argues that the development of doublets is the result of sociolinguistic factors. Kroch’s argument is based on work by Taylor (1994), who discusses the occurrence of doublets in the history of English past tense morphology. Taylor argues that the rise of morphological doublets in the Middle English (ME) period resulted from language contact of northern ME dialects with Scandinavian and subsequent dialect mixture with southern dialects.21 Interestingly, the life span of such doublets is apparently rather limited: there are no cases where a doublet that was already present in ME continues to exist in the present-day language (apart from cases where the two forms have developed different meanings). This can be taken to indicate that doublets can come into existence due to sociolinguistic factors such as language contact. However, in the course of time, blocking effects ensure that one form wins out over the other: Doublets arise through dialect and language contact and compete in usage until one or the other form wins out. Due to their sociolinguistic origins, the two forms often appear in different registers, styles, or social dialects: but they can only coexist stably in the speech community if they differentiate in meaning, thereby ceasing to be doublets. Speakers learn either one or the other form in the course of basic language acquisition, but not both. Later in life, on exposure to a wider range of

JB[v.20020404] Prn:2/08/2005; 15:55

F: LA8106.tex / p.16 (1110-1161)

 The Rise of Agreement

language, they may hear and come to recognize the competing form, which for them has the status of a foreign element. They may borrow this foreign form into their own speech and writing for its sociolinguistic value or even just because it is frequent in their language environment. Over time, however, as dialects and registers level out through prolonged contact, the doublets tend to disappear. (Kroch 1994: 6)

Does this explanation carry over to the kind of morphological doublets created by the rise of new agreement formatives (cf. e.g. the Bavarian data in (14) and (15) above)? For the following two reasons, this is not very likely. First, it is fairly clear that language/dialect contact usually does not play an important role in the reanalysis of clitics as agreement markers (however, see Siewierska 2004: 274ff. for a set of examples where person markers are apparently borrowed from other languages). Rather, the change in question is most probably a language-internal process which leads to the rise of new, more specified inflectional affixes. In addition, we have seen that at least in Bavarian, the old and new forms do not vary freely. Instead, they are distributed complementarily, with the new forms initially restricted to inversion contexts, while the old forms are confined to (embedded) OV clauses.22 Thus, we have to look for an alternative explanation of the existence and development of morphological doublets which come into existence as the result of grammaticalization processes similar to the Bavarian case at hand. Note that this explanation must satisfy two apparently conflicting demands. On the one hand, in order to not violate the Blocking Principle or the Subset Principle, the connection between the competing doublets must be loose enough to allow the coexistence of different agreement formatives for a given slot of the paradigm. On the other hand, the connection must be close enough in order to warrant that the grammaticalization process affects only underspecified slots of the paradigm, in line with the BP. The basic idea I want to pursue in the following makes use of the assumption that the BP operates in a local fashion, that is, it ensures replacement of a less specified form only in a certain insertion context (or, with respect to a certain morpheme), but not in a wholesale fashion. In (18), the relevant passage is marked by italics. (18) Blocking Principle If several appropriate PF-realizations of a given morpheme are attested in the Primary Linguistic Data, the form matching the greatest subset of the morphosyntactic features included in the morpheme must be chosen for storage in the lexicon. Coming back to the Bavarian data under consideration, it is important to note that according to the analysis proposed in Chapter 5 above, the new and the old agreement ending are different phonological realizations of two different instances of one and the same underlying agreement head/morpheme. While the old ending (2sg /-s/, 2pl /-t/, 1pl /-a(n)/) is confined to the syntactic Agr-on-T head, the new ending (2sg /-st/, 2pl /-ts/, 1pl /-ma/) is inserted as the realization of a dissociated agreement morpheme which attaches to the C-head as a copy of the syntactically licensed agreement morpheme adjoined to T:

JB[v.20020404] Prn:2/08/2005; 15:55

F: LA8106.tex / p.17 (1161-1211)

Chapter 6. Morphological blocking and the rise of agreement 

(19)

CP C’ C

TP

C Agr (subj.) {/-st/, /-ts/, /-ma/}

T’ íP

T T Agr {/-s/, /-t/, /-a(n)/}

Recall that due to the SOV+V2 syntax of Bavarian, new agreement suffixes could only develop in V2 contexts via an intermediate stage where the new agreement marker was introduced as the realization of a dissociated Agr-morpheme attached to C. Thus, initially the new ending replaced the old form only in a certain context (inversion), gradually gaining a wider distribution.23 However, the reanalysis did not only introduce a new phonological exponent of verbal agreement. In addition, it created another agreement head/morpheme which is post-syntactically adjoined to C.24 When the reanalysis in question became possible (e.g., due to phonological erosion of the pronoun), it is only in this insertion context (i.e., Agr-on-C) that the BP ensured that a former pronoun (or, rather, a suffix consisting of the old ending and the onset of the clitic) was stored in the lexicon as the exponent of an Agr-morpheme, leading to the obsolescence of the old, less specified verbal agreement endings in inversion contexts. In other words, initially the BP forced the replacement of the less specified form only in a certain insertion context (Agr-on-C), while verbs in clause-final position continued to exhibit the old agreement ending, triggering the acquisition of two different exponents for a certain slot of the verbal agreement paradigm.25 Hence, the BP (and similarly the Subset Principle) does not prohibit the existence of different agreement formatives for a given slot of the paradigm as long as these formatives are realizations of different instances of an Agr-morpheme which occur in different syntactic contexts. What the BP does rule out, however, is the possibility that the learner acquires competing phonological realizations for one and the same Agr-morpheme. Interestingly, this prediction seems to be borne out by the empirical facts. To the best of my knowledge, there are no instances where complementizer agreement (i.e., Agr-on-C) is realized by the old agreement formatives, that is, forms such as *ob-s ‘whether-2sg’ *ob-t ‘whether2pl’ or *ob-an ‘whether-1pl’ are not attested in any stage or variety of Bavarian. This can be taken as further support for the analysis proposed above according to which the BP required the use of the new, more specified exponents when this insertion context was created via the reanalysis of former clitic pronouns.26 In other words, the BP (and the Subset Principle) allows for the existence of morphological doublets if the relevant agreement formatives are realizations of different agreement morphemes with identi-

JB[v.20020404] Prn:2/08/2005; 15:55

F: LA8106.tex / p.18 (1211-1266)

 The Rise of Agreement

cal feature content (or, rather, different instances of a single agreement morpheme), in the case at hand Agr-on-C (inserted post-syntactically) and Agr-on-T (present in the syntax). Still, this account is restrictive enough to guarantee that the new exponents develop only for previously underspecified slots in the verbal agreement paradigm. This is achieved by the assumption that the BP forces the acquisition of the new, more specified exponents when the latter compete with the old, less specified formatives for the realization of the newly created agreement morpheme on C.

.. Section summary In this section, I have established the claim that the grammaticalization of agreement markers is shaped by morphological blocking effects. The central theoretical device, the so-called Blocking Principle is understood as an economy principle which shapes the acquisition of inflectional morphology, warranting an optimal and nonredundant Lexicon (and paradigm) structure. It was demonstrated that this approach offers an explanation for the observation that the grammaticalization of new agreement markers in the history of Bavarian affected only underspecified slots of the agreement paradigm. In addition, I proposed a diachronic explanation of the peculiar person/number restrictions on complementizer agreement and pro-drop exhibited by present-day Bavarian (cf. Chapter 5 above), arguing that they reflect the workings of the Blocking Principle, in the sense that only in these contexts (2nd person, 1pl) the Blocking Principle licensed the reanalysis of clitic pronouns as agreement morphemes attached to C (Agr-on-C). Subsequently, I discussed the status of morphological doublets, which are often a by-product of this kind of change, and the phenomenon of paradigm leveling, which expands the domain of less marked, more regular forms. It was argued that analogical leveling affecting stem allomorphy can also be attributed to morphological blocking if it is assumed that the learner prefers to acquire the stem alternant which manifests the characteristic phonemic distinctions of a given lexical root in the most unambiguous way. In addition, I considered cases where an underspecified agreement exponent gains a wider distribution in a paradigm, arguing that this kind of paradigm leveling is to be attributed to the expansion of the domain of Impoverishment rules which bleed the effects of the Blocking Principle. Furthermore, I argued that the Blocking Principle allows for the existence of morphological doublets only if the relevant agreement formatives are realizations of different agreement morphemes with identical feature content (in the case at hand Agr-on-C and Agr-on-T). In other words, the Blocking Principle operates in a local fashion, restricted to a certain insertion context. In this way, it is possible to maintain the hypothesis that the Blocking Principle forces the acquisition of more specified exponents: in the insertion context where the new forms came to compete with the less specified old forms (Agr-on-C), only the former, more specified exponents were acquired, leading first to a complementary distribution of old and new agreement suffixes and eventually to the loss of the old forms.

JB[v.20020404] Prn:2/08/2005; 15:55

F: LA8106.tex / p.19 (1266-1329)

Chapter 6. Morphological blocking and the rise of agreement 

In the next section, it is shown that the theoretical hypotheses developed in the present section receive further support from a typological tendency already noted in the introductory Chapter 1, namely the observation that 1st and 2nd person play a pioneering role in the development of new agreement formatives across languages.

. The pioneering role of 1st and 2nd person In this section I want to add another piece of evidence from language typology which can be taken to indicate that something like the Blocking Principle is in fact on the right track. More specifically, it is argued that the pioneering role of 1st and 2nd person forms in the rise of agreement paradigms across the world’s languages (cf. Chapter 1 above) can directly be attributed to the workings of blocking effects during language acquisition if we adopt certain assumptions about the feature inventory that underlies person distinctions. As noted in Chapter 1 above, we can observe that across the world’s languages, verbal agreement markers are much more common for 1st and 2nd person subjects than for 3rd person subjects (cf. Bybee 1985; Ariel 2000; Cysouw 2003a). For example, Bybee (1985) shows that 54% of the languages (in her sample) which manifest agreement do not mark 3rd person on the verb (cf. Chapter 1 for examples). Similar person restrictions can also be observed in languages that are currently developing new agreement markers from clitic pronouns such as Non-Standard French, where new agreement formatives are found for 1st/2nd person, but not in 3rd person contexts (see Section 6.4 below).27 The special role of 1st and 2nd person in grammaticalization processes has inspired numerous functionalist explanations which mostly make use of the fact that speaker and hearer are the most salient participants in a speech event (cf. Mithun 1991; Ariel 2000), that is, 1st and 2nd person forms exhibit a high degree of ‘givenness’, ‘discourse accessibility’ or ‘discourse prominence’ etc. Since the reference of pronominal forms for 1st and 2nd persons can be easily deduced from the immediate discourse situation, so the argument goes, they are more readily reduced to unstressable clitics, affixes, and eventually zero than 3rd person forms, which denote a ‘more remote’ agent. In this section it is argued that these facts can be explained in purely formal terms if we combine the Blocking Principle with certain assumptions concerning the featural basis of person distinctions in natural languages. These assumptions are laid out in the following, which requires a brief digression on the structure of person features before we can turn to explaining the typological tendency in question in terms of the workings of the Blocking Principle. The basic assumption I want to adopt is that the feature system behind person distinctions is not organized around a tripartite division ([1st], [2nd], [3rd]) as traditionally assumed. Instead, I take up the idea that [3rd person] actually does not constitute a separate person feature at all. Instead, ‘3rd person’ is analyzed as the result of the absence of (positive values for) the features 1st and 2nd person (cf. Benveniste

JB[v.20020404] Prn:2/08/2005; 15:55

F: LA8106.tex / p.20 (1329-1419)

 The Rise of Agreement

Table 11. Person features according to Halle (1997). Feature combination

Interpretation

[+Auth, +PSE] [+Auth, –PSE] [–Auth, +PSE] [–Auth, –PSE]

‘1st person inclusive’ ‘1st person exclusive’ ‘2nd person’ ‘3rd person’

1972; Bayer 1984; Halle 1997; Grimshaw 1997; Noyer 1997; Poletto 1999; Ariel 2000; Harley & Ritter 2002; Baker & Bobaljik 2002; Cysouw 2003a; Frampton 2002, among others). In a model of person specification that does not include a separate feature for 3rd person, the traditional three-way system is replaced by a binary feature system which only refers to the immediate participants in a speech event, for example [±speaker] and [±hearer]. Within this approach, the interpretation ‘3rd person’ results from the absence of positive values for the features [speaker] and [hearer]. The set of person distinctions listed in Table 11 makes use of the feature inventory introduced in Halle (1997), namely Participant in Speech Event (PSE) and Author in Speech Event (Auth). This system gives rise to a four-way person distinction for non-singular forms. In addition to the traditional notions ‘1st person’, ‘2nd person’ and ‘3rd person’, it also serves to distinguish between ‘1st person inclusive’ (non-singular 1st person forms that include the addressee) and ‘1st person exlclusive’ (non-singular 1st person forms that exclude the addressee).28 In languages which lack the distinction between inclusive and exclusive ‘we’ (as many Indo-European languages such as English, German, Italian etc.), the feature [±PSE] (or, [±hearer]) is not relevant for the phonological realization of 1st person plural pronouns. Likewise, the feature [±Auth] does not matter for the realization of 2pl. As a consequence, the system of plural pronouns is most economically described by assuming Vocabulary items which are underspecified for the relevant person features, cf. the paradigm of plural nominative pronouns in English:29 (20) a. [+Auth, +pl] ↔ we b. [+PSE] ↔ you c. [+pl] ↔ they With respect to 3pl, the logic of underspecification requires that 3pl be analyzed as the elsewhere case in the plural paradigm: due to the absence of positive values for person features (i.e., [–Auth], [–PSE]), ‘3rd person’ is most economically captured by the absence of any person specification in the relevant Vocabulary item. The same goes for 3sg forms, which lack person specifications, but are specified for gender in English (if singular is treated as the unmarked default value, they lack number features as well). In other words, the 3rd person forms represent the maximally underspecified (elsewhere) cases in the paradigm. This of course carries over to 3rd person forms across languages, including agreement formatives.30

JB[v.20020404] Prn:2/08/2005; 15:55

F: LA8106.tex / p.21 (1419-1458)

Chapter 6. Morphological blocking and the rise of agreement 

Let us now turn to the question of how this binary system of person specifications helps us to explain the pioneering role of 1st and 2nd person in the historical development of verbal agreement marking. Here, the important assumption is that 3rd person forms are generally underspecified for person features, that is, the relevant Vocabulary items lack specifications for [±Auth] and [±PSE]. The fact that cross-linguistically, 3rd person agreement formatives arise later (if at all) than markers for 1st and 2nd person can then be attributed to the workings of the Blocking Principle in the following way: if UG requires that newly grammaticalized markers realize a greater subset of agreement features than existing markers, the development of 3rd person forms is considerably hindered, due to the inherent underspecification of 3rd person forms with respect to the set of person features. As a consequence, even in a grammar that previously lacked agreement markers completely, a true 3rd person marker can only develop if it is specified for some other inflectional feature like [gender], [number] etc. We can therefore conclude that the workings of the Blocking Principle ensure that the grammaticalization of new 3rd person forms is less likely to be triggered than the development of forms specified for a separate [person] feature, that is, 1st and 2nd person markers. This explains the typological tendency in question, lending further support to the proposal that the acquisition of inflectional morphology is guided by blocking effects that favor more specified over less specified forms.31 In the remainder of this chapter, the workings of the Blocking Principle are illustrated with examples from a variety of different languages, beginning with NonStandard French, where the grammaticalization of new agreement formatives is also characterized by the dichotomy between 1st/2nd person forms and 3rd person forms discussed in the present section.

. French In the history of French, we can observe a cluster of changes involving pronouns, verbal agreement, and the pro-drop property which is cyclic in nature (Guiraud 1968; Wartburg 1970; Ashby 1977; Harris 1978; Roberts 1993a; Vance 1997): (21) Distinctive verbal Agr/pro-drop >>> loss of Agr/loss of pro-drop (overt subject pronouns become obligatory) >>> subject pronouns become clitics >>> clitics are reanalyzed as verbal Agr/rise of pro-drop Thus, subject pronouns were merely optional in Old French. Overt pronouns were used either for reasons of emphasis (similar to the present-day Romance pro-drop languages like Italian) or inserted to avoid V1 orders, presumably due to the V2 character of Old French (cf. e.g. Kuen 1957; Haiman 1991; Siewierska 1999).32 It is generally assumed that in later stages, the subject pronouns gradually lost their emphatic value. In the 14th century then, the phonological erosion of the inherited verbal agreement suffixes reached a point where most endings had lost their distinctive character. In turn, overt subject pronouns became more and more obligatory, in line with the stan-

JB[v.20020404] Prn:2/08/2005; 15:55

F: LA8106.tex / p.22 (1458-1518)

 The Rise of Agreement

dard hypothesis that rich verbal agreement is a precondition for pro-drop. The fact that the original personal pronouns could no longer be used to convey emphasis led to the use of object pronouns in emphatic contexts (which are doubled by the weak subject pronouns), as in the following topic left dislocation example from Modern Standard French (cf. Guiraud 1968; Wartburg 1970; Ashby 1977 on the connection between the rise of this construction and the weakening/grammaticalization of the subject pronouns): (22) Moi, je sors. me I leave ‘As for me, I leave.’ In other words, even in the traditional descriptive literature it is often assumed that the subject pronouns of present-day French undergo a development from referential elements to some form of agreement marker: It is, then, possible to see in the history of the minimum finite verb group that a semantic restructuring has taken place, whereby the once referentially important subject pronoun has become a morpheme whose only importance is now structural. In this sense, the subject pronoun has become a mere satellite of the verb. (Ashby 1977: 55)

This development is even further advanced in present-day Non-Standard French, where subject clitics are obligatory (see below for details) and exhibit characteristics typical of prefixal inflectional morphology (Ashby 1977; Harris 1978; Lambrecht 1981; Auger 1993, 1994; Gerlach 2002). Accordingly, Ashby (1977: 59) predicts that Modern Non-Standard French is currently developing into a pro-drop language (with prefixal person/number markers), completing the cycle depicted in (21). Several arguments have been put forward in the literature on Non-Standard French that examples such as (22) should not be analyzed as instances of (clitic) left dislocation, in contrast to present-day Standard French, where je is usually treated as a resumptive clitic. In what follows, I will review only a selection of the arguments suggesting that the subject clitics in Non-Standard French constitute some form of prefixal agreement marking on the verb (cf. Lambrecht 1981; Auger 1993, 1994; Haiman 1991). First, the relevant subject clitics are obligatory and cannot be replaced by the full form (historically an oblique form, see above): Non-Standard French (23) a. (Moi) je porte la table. me clit.1sg carry the table ‘I carry the table.’ b. Moi *(je) porte la table. me clit.1sg carry the table ‘I carry the table.’ (Gerlach 2002: 224)

JB[v.20020404] Prn:2/08/2005; 15:55

F: LA8106.tex / p.23 (1518-1612)

Chapter 6. Morphological blocking and the rise of agreement

The obligatory presence of these clitic elements is also reflected by their peculiar behavior in conjoined clauses. In most languages, subject pronouns are usually elided in second conjuncts if they are coreferent with the subject of the first conjunct. However, no such ellipsis takes place in Non-Standard French (cf. Lambrecht 1981: 24; Haiman 1991): Standard French (24) Il mange et boit comme un cochon. he eats and drinks like a pig Non-Standard French (25) I mange et i boit comme un cochon. he eats and he drinks like a pig Again, this behavior points toward an analysis of the preverbal pronouns as bound agreement markers which are obligatory regardless of the syntactic context. Second, sentences with apparent clitic doubling generally favor a basic, nondislocated interpretation. Thus, sentences such as (23a) receive an interpretation like (26a) instead of (26b). This indicates that they do not constitute instances of topic left dislocation, again suggesting an analysis of the apparent clitics as prefixal agreement markers. (26) a. ‘I carry the table.’ b. *‘As for me, I carry the table.’ Third, in contrast to the pronominal clitics of Standard French, cf. (27), the preverbal ‘clitics’ of colloquial French fail to undergo subject-verb inversion in matrix interrogatives, as can be seen in the example (28) (taken from Lambrecht 1981: 6): Standard French (27) Où vas=tu? where go=you ‘Where are you going?’ Non-Standard French (28) Où tu-vas? where you-go ‘Where are you going?’ In other words, the obligatory preverbal clitics of Non-Standard French seem to accompany verb movement to C, in contrast to the subject clitics of the standard language. Again, this can be taken to indicate that they should be analyzed in terms of prefixal agreement markers. To sum up, it seems that Non-Standard French exemplifies another instance of a grammaticalization process in which clitic pronouns develop into some form of verbal inflection. Interestingly, the distribution of the newly created agreement markers of NonStandard French exhibits some parallels to the distribution of complementizer agree-



JB[v.20020404] Prn:2/08/2005; 15:55

F: LA8106.tex / p.24 (1612-1663)

 The Rise of Agreement

Table 12. Distribution of subject clitics in Non-Standard French. Presence of the clitic 1sg 2sg 3sg 1pl 2pl 3pl

obligatory obligatory optional obligat. (on)/option. (nous) optional optional

ment in the Bavarian varieties discussed above. Table 12 shows that the clitic is obligatory in some person/number combinations while it is merely optional in others. The contrast between 1st person and 2sg clitics, which are always obligatorily present, and 3rd person and 2pl forms, which can be replaced by full DPs, is illustrated by the following examples (taken from Gerlach 2002: 224): (29) a.

(Moi) je porte la table. me 1sg carry the table ‘I carry the table.’ b. Moi *(je) porte la table. me 1sg carry the table ‘I carry the table.’

(30) a.

(Jules) il porte la table. Jules clit.3sg carry the table ‘Jules/he carries the table.’ b. Jules (il) porte la table. Jules clit.3sg carry the table ‘Jules carries the table.’

(31) a.

(Alexandra et moi) on porte la Alexandra and I 1pl carry the ‘Alexandra and I carry the table.’ b. Alexandra et moi *(on) porte la Alexandra and I 1pl carry the ‘Alexandra and I carry the table.’

table. table table. table

(32) a.

(Alexandra et moi) nous portons la table. Alexandra and I clit.1pl carry the table ‘We/Alexandra and I carry the table.’ b. Alexandra et moi (nous) portons la table. Alexandra and I clit.1pl carry the table ‘Alexandra and I carry the table.’

Thus, it seems that the grammaticalization of new agreement markers has affected only a subset of the relevant clitic forms. A further indication that 3sg forms have

JB[v.20020404] Prn:2/08/2005; 15:55

F: LA8106.tex / p.25 (1663-1746)

Chapter 6. Morphological blocking and the rise of agreement 

Table 13. Subject agreement in written and spoken French.

1sg 2sg 3sg 1pl 2pl 3pl

Written language

Phonetic form

porte portes porte (on) porte (nous) portons portez portent

[pfrt] [pfrt] [pfrt] [pfrt] [pfrÁt«f] [pfrÁte:] [pfrt]

retained their pronominal character comes from the fact that clitic doubling with 3rd person forms still exhibits properties of topic left dislocation such as incompatibility with (negative) quantifiers (Friedemann 1995; Laenzlinger 1998: 339): (33) *Personne il rigole jamais. nobody he laughs never ‘Nobody never laughs.’ So the question arises of why the reanalysis in question has affected only a subset of the subject clitics. Interestingly, Gerlach (2002: 225f.) observes that the distribution of the obligatory subject ‘clitics’ correlates with properties of the verbal agreement paradigm. More specifically, she argues that clitics are obligatory only when the verbal inflection is underspecified for subject agreement features. According to the written standard language, French apparently distinguishes a number of different agreement endings on the verb. However, the spoken language actually makes less distinctions than the written forms may suggest. This is shown in Table 13, where the phonetic forms of the present indicative of the verb porter ‘to carry’ are compared with the relevant forms of the written language. In Table 13, we can see that the written language distinguishes five different agreement endings; only the forms for 1sg and 3sg (and the agreement endings for the colloquial 1pl form on) are homophonous. In contrast, the spoken language distinguishes merely three forms, -Ø (1sg, 2sg, 3sg, 1pl (on), 3pl), -«f (1pl, nous) and -e: (2pl). One of these forms, the 1pl -«f, shows up only in connection with the pronoun nous, which is considered rather formal and is rarely used in the spoken language. Thus, we actually deal with a highly syncretic agreement paradigm, in which only the 2pl ending signals unambiguously person and number of the subject. In all other contexts, we find the ending -Ø, which is underspecified for person and number and accordingly represents the elsewhere case: (34) a. [–Auth, +PSE, +pl] ↔ /-e:/ b. elsewhere ↔ -Ø If we combine our observations concerning the distribution of the obligatory ‘clitics’ with the findings concerning the existing suffixal agreement paradigm in the spoken language, we arrive at the following generalization:

JB[v.20020404] Prn:2/08/2005; 15:55

F: LA8106.tex / p.26 (1746-1828)

 The Rise of Agreement

(35) Correlation between the distribution of subject clitics and properties of suffixal agreement In Non-Standard French, subject clitics are obligatory only a. in non-3rd person contexts, and b. if they serve to express φ-features not marked by the existing suffixal agreement morphology. This is illustrated in Table 14 where agreement-like elements which reflect the φfeatures of the subject are in boldface (obligatory ‘clitics’ and agreement suffixes) while optional clitics (which are presumably still pronominal elements) are set in parentheses. In the following, it is argued that the distribution of obligatory subject clitics in Non-Standard French can be attributed to the workings of the Blocking Principle if it is assumed that 3rd person forms are underspecified for person features (see Section 6.3 above). Similar to Bavarian (cf. Section 6.2.1 and Note 31), new agreement formatives developed only in 1st and 2nd person contexts, where the new agreement formatives are specified for a greater subset of the relevant φ-features than the existing verbal agreement morphology (the elsewhere form -Ø): (36) a. [+Auth, –pl] ↔ /Š6/ b. [+PSE, –pl] ↔ /ty / c. [+Auth, +pl] ↔ /«f/

(1sg) (2sg) (1pl)

Hence, it is quite obvious that the grammaticalization of the new prefixal forms is licensed by the Blocking Principle. In contrast, the potential new 3rd person agreement prefixes il/elle and ils/elles carry no person specification. That is, at least with respect to person features they fail to be stronger specified than the existing forms. However, it is fairly clear that the pronouns carry gender and (probably) number specifications which at first sight should qualify them as more specific candidates if compared with the relevant existing agreement suffixes. Still, the 3rd person clitics are merely optional in Non-Standard French, which suggests that they have not yet developed into agreement prefixes. This fact is at first sight not expected, given the way the Blocking Principle is formulated above.

Table 14. Suffixal agreement and clitics in Non-Standard French. Clitic + verb 1sg 2sg 3sg 1pl 2pl 3pl

je porte tu portes (il/elle) porte on porte (vous) portez (ils/elles) portent

JB[v.20020404] Prn:2/08/2005; 15:55

F: LA8106.tex / p.27 (1828-1887)

Chapter 6. Morphological blocking and the rise of agreement 

First, I will address the question of whether gender specification alone is sufficient to trigger the grammaticalization of new verbal agreement markers. Note that Greenberg’s (1966a) Universal 32 states that verbal agreement for gender entails agreement in number,33 that is, it seems that verbal agreement in gender becomes available only if a language has developed a full paradigm of number distinctions. Moreover, it appears that at least on verbs, gender agreement is a highly marked grammatical trait that is found only in a couple of languages. This generalization on the distribution of morphosyntactic features can be explained if it is assumed that φ-features are not unordered sets, but rather are organized hierarchally in a feature geometry where number features dominate gender features (cf. e.g. Noyer 1997; Harley & Ritter 2002), as illustrated by the following tree structure:34 (37)

Agr-morpheme/pronoun  Auth

 PSE



  Animate

Inanim./Neuter

It is generally assumed that in a feature structure such as (37), the individual nodes (and the features/nodes they dominate) are present only if the grammar in question exhibits the relevant morphological distinctions. On these assumptions, the observation that the possibility of gender distinctions seems to depend on the existence of number distinctions (Greenberg’s Universal 32) can be accounted for in the following way: since the Gender node is a dependent of the Number node, gender agreement is only possible if the Number node is present as well, but not vice versa. Furthermore, if the feature geometry in (37) is taken to represent a universal property of grammar we should expect that it has some bearing on the historical development of inflectional morphology as well. More specifically, the grammaticalization of morphosyntactic feature distinctions is predicted to proceed in a top-down fashion, in the sense that a language can develop verbal agreement in gender only if it has previously grammaticalized a set of number distinctions. Moreover, I assume that this restriction operates over paradigms: languages do not develop gender agreement in a certain person/number combination (e.g., 3sg.fem) when there are still forms in the paradigm that fail to show number distinctions. In other words, the development of verbal gender agreement is bound to the existence of a full paradigm of number specifications for all persons.35 These assumptions suggest the following explanation for the fact that the gender specification of the 3rd person pronouns is not sufficient for triggering a reanalysis as agreement markers in Non-Standard French: it is fairly clear that the existing agreement paradigm does not distinguish number for all persons. Accordingly, gender distinctions are presumably not visible to the workings of the Blocking Principle and can therefore not motivate the grammaticalization of new agreement markers.

JB[v.20020404] Prn:2/08/2005; 15:55

F: LA8106.tex / p.28 (1887-1937)

 The Rise of Agreement

Still, it appears that il/elle and ils/elles are apparently specified for number (via plural -s), which should nonetheless suffice to mark them as more specified than the existing ‘elsewhere’ agreement suffix -Ø. As is well known, however, the number marking of the plural forms is only perceivable if the verb following the pronoun begins with a vowel. In all other contexts, final -s is not pronounced and the plural forms are homophonous with the relevant 3sg forms. Hence, the number marking of the 3rd person forms is actually less salient in the PLD than it appears at first sight. Tentatively, I suggest that the restricted distribution of unambiguous number marking (plural is perceivable only in front of verbs beginning with a vowel; in all other contexts the 3sg/pl forms are identical) fails to be robust enough to activate the Blocking Principle (cf. Note 7 above on the influence of frequency on the workings of the Blocking Principle). Note, however, that the existence of clearly different 3sg and 3pl clitics is expected to trigger the grammaticalization of new agreement formatives if all other necessary conditions are met. A relevant example comes from the Northern Italian dialect Piattino (Lombardy) which is discussed in the next section.

. Northern Italian dialects Many Italian dialects exhibit similar developments as Non-Standard French. In particular, it has often been observed that Northern Italian dialects exhibit obligatory subject clitics which may be doubled by an additional full subject. The peculiar properties of these clitics (including e.g. obligatory presence in conjoined clauses and relative clauses, obligatory adjacency to the verb, compatibility with a quantified full subject etc.) have led many scholars to the conclusion that these elements developed into some form of verbal inflection in the history of these dialects. Even more interesting in the present context is the observation that the change in question correlates with properties of the existing verbal agreement paradigm (cf. e.g. Meyer-Lübke 1894; Rohlfs 1949; Spiess 1956; Kuen 1957; Renzi & Vanelli 1983; Poletto 1995, 1997, 1999, 2000). It appears that the subject clitics are obligatory only in those instances where the existing verbal agreement morphology fails to unambiguously reflect the subject’s φ-features. If it is assumed that the obligatory clitics have in fact developed into (prefixal) verbal agreement markers, their distribution can be explained by assuming that the newly grammaticalized agreement formatives served to compensate for the loss of distinctive agreement endings, in line with the Blocking Principle. Of course, this section cannot serve as a comprehensive overview of the phenomenon in question in the multitude of Northern Italian dialects. Instead, I chose to focus on a pair of dialects which represent two of the major dialectal areas of Northern Italy: Piattino, a variant of Lombardo spoken in and around the village of Piatta in the Italian Alps and the dialect of Vicenza (Vicentino) which belongs to the dialect group usually referred to as Veneto. The Piattino data are taken from Gerlach (2001, 2002) while the section on Vicentino reports on my own research on this dialect.

JB[v.20020404] Prn:2/08/2005; 15:55

F: LA8106.tex / p.29 (1937-2009)

Chapter 6. Morphological blocking and the rise of agreement 

.. Piattino First, let us take a look at Piattino, a Lombardian dialect discussed in Gerlach (2001, 2002). According to Gerlach (2002: 225f.), the Piattino subject clitics are obligatory “whenever the verbal forms lack subject person agreement features”. From this, it appears that Piattino behaves similar to Non-Standard French, exhibiting clitics which take up the function of agreement markers in contexts where the existing verbal morphology is not distinctive anymore. However, at a closer look it becomes clear that the Piattino facts differ in important ways from the situation observed in Non-Standard French. Consider Table 15, where obligatory clitics and distinctive verbal agreement morphology are marked by boldface (relevant example sentences are shown in (38)– (40)).36 (38) a.

Mi (a) guardi. I clit.1sg watch-1sg ‘I am watching.’ b. (Mi) a guardi. I clit.1sg watch-1sg ‘I am watching.’ (Gerlach 2002: 224)

(39) a.

(Nigun/un omen/l’omen/Alessio/lu) al guarda. nobody/a man/the man/Alessio/he clit.3sg.masc watch.3sg ‘Nobody/a man/the man/Alessio/he is watching.’ b. Nigun/un omen/l’omen/Alessio/lu *(al) guarda. nobody/a man/the man/Alessio/he clit.3sg.masc watch.3sg ‘Nobody/a man/the man/Alessio/he is watching.’ (Gerlach 2002: 223)

(40) a.

(Noaltri) an guarda. we clit.1pl watch.1pl ‘We are watching.’ b. Noaltri *(an) guarda. we clit.1pl watch.1pl ‘We are watching.’ (Gerlach 2002: 223)

Table 15. Subject agreement and clitics in Piattino, present indicative.

1sg 2sg 3sg.masc 3sg.fem 1pl 2pl 3pl.masc 3pl.fem

Clitic + verb

Presence of the clitic

(a) guardi (te) guardesc al guarda la guarda an guarda (ve) guardé i guarden li guarden

optional optional obligatory obligatory obligatory optional obligatory obligatory

JB[v.20020404] Prn:2/08/2005; 15:55

F: LA8106.tex / p.30 (2009-2072)

 The Rise of Agreement

Table 15 and the examples in (38)–(40) illustrate that the presence of subject clitics is merely optional in 1sg and 2nd person contexts, but obligatory in all other contexts (i.e., 3sg, 1pl and 3pl), giving rise to clitic doubling if a full pronoun/subject is added for reasons of emphasis. The obligatory presence of the 3rd person and 1pl clitics and the fact that the 3rd person clitics can be doubled by quantified expressions like nigun ‘nobody’ suggest that the relevant clitics have changed into agreement markers. As already mentioned above, the obligatory presence of the subject clitic seems to correlate with properties of the suffixal agreement morphology. That is, the clitic is obligatory in the context of 1pl and 3sg, where the existing agreement endings are homophonous.37 This is also the conclusion reached by Gerlach, who attributes the obligatory presence of the clitics to the lack of (distinctive) person agreement features on the verb.38 In other words, subject clitics became obligatory in contexts where the finite verb is underspecified for agreement features, presumably to recomplete a defective agreement paradigm. In contrast, the clitic is merely optional in 1sg and 2nd person contexts, where the verbal agreement morphology is still fully distinctive, reflecting unambiguously the subject’s set of φ-features. This sensitivity to properties of the suffixal agreement morphology can be taken as a further indication that the clitics are in fact better analyzed as some form of agreement marking. Thus, it seems that the behavior of subject clitics in Piattino resembles the situation observed in Non-Standard French. However, note that there are also important differences between Piattino and NonStandard French. First, there is a context (3pl) where a clitic is obligatory although the suffixal agreement morphology is not homophonous with any other agreement ending (and thus seems to be distinctive). Second, in contrast to Non-Standard French, Piattino exhibits obligatory clitics for all verb forms that are homophonous (in NonStandard French, all singular forms and 1pl, 3pl are identical, but the clitic is obligatory only in non-3rd person contexts), cf. the following quote from Gerlach (2002: 225): A particular feature of the Piattino subject paradigm is that the 3rd person may not remain unspecified. Consequently, the clitic is obligatory for 3rd person and 1st person plural.

That is, while the optional presence of clitics in 1sg and 2sg contexts is in line with the Blocking Principle (the reanalysis of clitics as agreement markers is not called for as long as the existing agreement morphology is distinctive), the behavior of the 3rd person forms is unexpected and demands some further consideration, in particular if it assumed that 3rd person forms are inherently underspecified with respect to person features (cf. Section 6.3 above). Let us therefore take a closer look at the agreement distinctions resulting from the development of obligatory clitics in Piattino. The change affecting the 1pl is rather straightforward from the viewpoint of the Blocking Principle: while the existing agreement ending consists merely of the theme vowel and represents the elsewhere case in the present indicative paradigm, the new agreement formative is clearly specified for both person and number (i.e., [+Auth, +pl]).

JB[v.20020404] Prn:2/08/2005; 15:55

F: LA8106.tex / p.31 (2072-2140)

Chapter 6. Morphological blocking and the rise of agreement 

Table 16. Development of gender agreement in Piattino.

3sg.masc 3sg.fem 3pl.masc 3pl.fem

New agreement formatives

Old agreement formatives

allaili-

-a -a -en -en

What about the development of obligatory 3rd person clitics for both singular and plural? Table 16 compares the new 3rd person agreement formatives with the existing suffixal agreement morphology. From this, it becomes apparent that the grammaticalization process led to the development of gender agreement in the context of 3rd person. Due to their specification for gender, the new 3rd person agreement formatives count as stronger agreement exponents, even if they do not carry any person specification. Of course, this raises the question of why gender information is accessible in Piattino, while it has been argued above that gender distinctions are invisible to the grammaticalization of agreement formatives in Non-Standard French. Tentatively, I suggest that the access to gender specifications is made available by the existence of a – compared to French – much more elaborate inventory of number distinctions in the existing agreement paradigm of Piattino. Recall that in Section 6.4 above, I have proposed a feature geometry for agreement distinctions where gender features are dominated by the node encoding number information. By assumption, the robust morphological representation of number in Piattino paved the way for the development of gender distinctions on the verb by making the gender node accessible for grammaticalization processes. Summing up, it appears that after a closer look, it is possible to maintain the idea that the grammaticalization of new verbal agreement morphology is governed by morphological blocking effects which require new agreement formatives to be more distinctive than existing forms. More specifically, it has been argued that the obligatory presence of subject clitics in Piattino can be attributed to the fact that the emerging prefixal agreement formatives realize distinctions such as person/number (1sg) or gender agreement (3sg/3pl) which are not marked by the respective suffixal agreement morphology. In the next section, it is shown that similar correlations can be detected in Vicentino, another Northern Italian dialect which shows an even richer inventory of existing suffixal agreement morphology.

.. Vicentino This section reports on my research on Vicentino, a variant of Veneto spoken in and around Vicenza (northern Italy).39 Similar to Piattino, the obligatory presence of subject clitics seems to be linked to properties of the existing verbal agreement paradigm in Vicentino. However, as we will see shortly, Vicentino differs in interesting ways from

JB[v.20020404] Prn:2/08/2005; 15:55

F: LA8106.tex / p.32 (2140-2217)

 The Rise of Agreement

Table 17. Subject agreement and clitics in Vicentino, present indicative of mangiare ‘to eat’.

1sg 2sg 3sg 1pl 2pl 3pl

Clitic + verb

Presence of the clitic

(a) magn-o te magn-i (enclitic: magn-i=to) el/ła magn-a (enclitic: magn-e=ło/ła) (a) magn-emo (a) magn-è (enclitic: magn-è=o) i/łe magn-a (enclitic: magn-e=łi/łe)

optional obligatory obligatory optional proclitic optional enclitic obligatory obligatory

Piattino. Table 17 illustrates the agreement paradigm and the distribution of subject clitics in Vicentino. Distinctive person/number marking (either via suffixes or clitics) is marked by boldface while optional clitics are set in parentheses.40 Even a first glance at Table 17 reveals that the correlation between defective agreement endings and the obligatory presence of subject clitics is less straightforward than in Piattino. The only homophonous endings are 3sg and 3pl /-a/ (which is traditionally analyzed as a theme vowel). In all other contexts, the agreement morphology seems to be distinctive. Still, subject clitics are obligatory for all person/number combinations apart from 1sg/1pl (2pl forms show a peculiar behavior: while the enclitic o is obligatory, the corresponding proclitic form a is merely optional; see below for discussion). The distribution of the various clitics is illustrated in the following, beginning with the observation that clitics are merely optional in 1st person contexts: (41) (A) vegno da Vicensa. clit.1sg come-1sg from Vicenza ‘I come from Vicenza.’ (42) (A) vegnemo da Vicensa. clit.1pl come-1pl from Vicenza ‘We come from Vicenza.’ Thus, in examples such as (41) and (42), the verbal morphology apparently suffices to identify subject pro. As a result, the overt pronoun a is merely optionally present. In all other contexts (apart from 2pl, see below), the absence of (distinctive) subject clitics leads to ungrammaticality (note that the shape of the proclitics differs from that of the enclitics used in inversion contexts): (43) a.

Da ‘ndó vien=*(to) ti? where-from come=clit.2sg you ‘Where do you come from?’

JB[v.20020404] Prn:2/08/2005; 15:55

F: LA8106.tex / p.33 (2217-2281)

Chapter 6. Morphological blocking and the rise of agreement 

b. Ti te vien da Montecio. you clit.2sg come from Montecio ‘You come from Montecio.’ c. *Ti vien da Montecio. As already noted above, 2pl forms exhibit an asymmetry between inversion and noninversion forms: while the enclitic o used in inversion contexts is obligatory, the corresponding proclitic form a (used in all other contexts) can be omitted:41 (44) a.

Da ‘ndó vegni=*(o) (voaltri)? where-from come=clit.2pl you.pl ‘Where do you come from?’ b. (A) vegní da Durlo. clit.2pl come-2pl from Durlo ‘You come from Durlo.’

Similar to Piattino, all 3rd person clitics (in both inversion and non-inversion contexts) are obligatory. Note that the obligatory 3pl clitics serve to signal gender distinctions in addition to person and number (łe unambiguously marks 3pl.fem, while (ł)i may be used to refer to either 3pl.masc or general 3pl without gender specification). The (a) examples illustrate the use of subject enclitics while the (b) examples depict the series of 3rd person proclitics: (45) 3sg.masc a. Da ‘ndó vien=*(ło) sto omo? where-from come=clit.3sg.masc this man ‘Where does this man come from?’ b. Elo/łu *(el) vien da Durlo. he clit.3sg.masc come from Durlo ‘He comes from Durlo.’ (46) 3sg.fem a. Da ‘ndó vien=*(ła) sta dona? where-from come=clit.3sg.fem this woman ‘Where does this woman come from?’ b. Ela *(ła) vien da Arzegnan. she clit.3sg.fem come from Arzignano ‘She comes from Arzignano.’ (47) 3pl.masc a. Da ‘ndó vien=*(łi) lori? where-from come=clit.3pl.masc they.masc ‘Where do they come from?’ b. Lori *(i) vien da Malo. they.masc clit.3pl.masc come from Malo ‘They come from Malo.’

JB[v.20020404] Prn:2/08/2005; 15:55

F: LA8106.tex / p.34 (2281-2336)

 The Rise of Agreement

(48) 3pl.fem a. Da ‘ndó vien=*(łe) lore? where-from come=clit.3pl.fem they.fem ‘Where do they come from?’ b. Lore *(łe) vien da Ciampo. they.fem clit.3pl.fem come from Ciampo ‘They come from Ciampo.’ Before turning to the issues raised by these data for the Blocking Principle, let me take a closer look at the inventory of subject clitics available in Vicentino. In general, Vicentino exhibits a remarkably rich inventory of different types of subject clitics. There are three different series of clitic pronouns: a set of enclitics which attach to the right of the verb in inversion contexts and two sets of proclitics. The latter subdivide into a proclitic a, which can be used for various person/number combinations (1sg, 1pl, 2sg, and 2pl), and a set of distinctive forms, which are unambiguously specified for person and number. Interestingly, the two types of proclitics also differ in their syntactic properties. First, only the distinctive subject clitics are obligatory in Vicentino while the non-distinctive pronominal clitic a, which is underspecified for both person and number features, is merely optional (see below for examples). Second, the distinctive clitics follow the negation no, while the non-distinctive forms precede the negation (similar phenomena can be observed in various other Northern Italian and Rhaeto-Romance dialects, cf. Meyer-Lübke 1894: 101f.; Rohlfs 1949: 168ff.; Renzi & Vanelli 1983; Linder 1987; Poletto 1997, 1999, 2000; Vanelli 1997: 109):42 (49) a.

(A) no vegno da Vicensa. clit.1sg not come-1sg from Vicenza ‘I do not come from Vicensa.’ b. (A) no vegnemo da Vicensa. clit.1pl not come-1pl from Vicenza ‘We do not come from Vicenza.’

(50) a.

No te vien da Vicensa. not clit.2sg come from Vicenza ‘You do not come from Vicenza.’ b. No ła vien da Vicensa.43 not clit.3sg.fem come from Vicenza ‘She does not come from Vicenza.’

The different distribution of the two kinds of clitics becomes particularly clear in 2sg contexts, where both types of clitics are available. Thus, in examples such as (51), up to three different types of pronouns may co-occur: a full pronoun, which usually bears stress (ti), the non-distinctive clitic a (unspecified for person and number), which precedes the negation, and the obligatory clitic te, which follows the negation and is always adjacent to the finite verb.

JB[v.20020404] Prn:2/08/2005; 15:55

F: LA8106.tex / p.35 (2336-2439)

Chapter 6. Morphological blocking and the rise of agreement 

(51) Ti a no te vien da Vicensa. you clit.2sg not clit.2sg come from Vicenza ‘You do not come from Vicenza.’ The different subject clitics available in Vicentino and their diverging properties are summarized in Table 18 and Table 19. Similar to Non-Standard French and Piattino, the special properties of the enclitics and distinctive proclitics suggests that these elements are presumably better analyzed as agreement marking elements. In contrast, the fact that the non-distinctive clitics are merely optional can be taken to indicate that these are still pronominal elements. The observation that the non-distinctive clitics are not affected by the grammaticalization process in question can be directly attributed to the workings of the Blocking Principle: due to the underspecification of the clitic a for both person and number features, it fails to meet the requirement that new agreement exponents must realize a greater subset of agreement features than existing agreement formatives. As a consequence, the clitic a cannot be reanalyzed as an agreement marker and maintains its status as a pronominal element. The 2sg forms provide a nice minimal pair: while the non-distinctive clitic a is merely optional and occurs to the left of negation, the distinctive form te is obligatorily present and always adjacent to the verb. In this way, the distributional asymmetries between distinctive and non-distinctive forms reflect the workings of morphological blocking in the grammaticalization of new agreement formatives in Vicentino: while the distinctive clitics were reanalyzed as agreement markers and became obligatory, the morphosyntactic defectiveness of the non-distinctive forms prevented a categorial reanalysis, preserving the pronominal properties of the elements in question.45 Table 18. Three series of subject clitics in Vicentino. Enclitics distinctive 1sg 2sg 3sg.masc 3sg.fem 1pl 2pl 3pl.masc 3pl.fem

– to ło ła – o łi łe

– te (e)l ła – – i łe

Proclitics non-distinctive a a – – a a – –

Table 19. Diverging properties of subject clitics in Vicentino.

Enclitics Distinctive proclitics Non-distinctive proclitics

φ-feature specification

Presence

Placement relative to verb and negation

person/number person/number Ø

obligatory obligatory optional

neg-V-clitic44 neg-clitic-V clitic-neg-V

JB[v.20020404] Prn:2/08/2005; 15:55

F: LA8106.tex / p.36 (2439-2593)

 The Rise of Agreement

However, while the behavior of the non-distinctive forms is expected from the viewpoint of the Blocking Principle, it is the change affecting the distinctive subject clitics which raises a more serious issue, in particular if a strong version of the Blocking Principle is adopted. On this assumption, the agreement formatives developing from former subject clitics are required to realize a greater subset of agreement features than existing inflectional markers. Above, however, we have seen that Vicentino exhibits a rather rich paradigm of suffixal agreement morphology, where all endings apart from 3rd person (3sg/3pl /-a/) are apparently fully distinctive. So, again the question arises of how the grammaticalization process in question can satisfy the constraints imposed by the Blocking Principle. In the following it is argued that the changes affecting Vicentino met these requirements via a conspiracy of a set of independent factors. First, if we take a look at the resulting paradigm of prefixal agreement markers, it becomes apparent that similar to Piattino, the change in question introduces verbal agreement in gender in 3rd person contexts, a trait formerly absent in the grammar of Vicentino, cf. the comparison of 3rd person forms in Table 20. Thus, even if it is assumed that the absence of person features hinders the grammaticalization of new agreement formatives in 3rd person contexts (see Section 6.3 above), it is fairly clear that the new formatives qualify as stronger specified elements by unambiguously marking gender (and number) of the subject.46 While this answers the question of why the 3rd person clitics were eligible for being grammaticalized as agreement markers, this account cannot be extended to the change affecting 2nd person forms, which does not show gender distinctions. Concerning the historical development of new 2nd person agreement formatives in Vicentino, consider the paradigm of the frequent irregular verb vegnere ‘to come’ in Table 21. Table 20. Development of gender agreement in Vicentino.

3sg.masc 3sg.fem 3pl.masc 3pl.fem

New agreement formatives

Old agreement formatives

el- (inversion: -ło) ła (inversion: -ła) i (inversion: -łi) łe (inversion: -łe)

-a -a -a -a

Table 21. Present indicative of vegnere ‘to come’ (Vicentino). vegnere ‘to come’ 1sg 2sg 3sg 1pl 2pl 3pl

vegno vien vien vegnemo vegní vien

JB[v.20020404] Prn:2/08/2005; 15:55

F: LA8106.tex / p.37 (2593-2594)

Chapter 6. Morphological blocking and the rise of agreement 

Table 22. 2sg, 3sg and 3pl of irregular verbs (present indicative) in Vicentino.

2sg 3sg 3pl

dare ‘to give’

fare ‘to do’

nare ‘to go’

savere ‘to know’

de da da

fe fa fa

ve va va

se sa sa

In Table 21, we can observe that the 2sg form vien is homophonous with the respective 3rd person forms in the present indicative (tegnere ‘to hold’ behaves similar). In other words, it seems that at least in the context of irregular verbs such as vegnere and tegnere, the reanalysis of the 2sg clitics proceeded in accordance with the Blocking Principle, creating a new agreement formative which is unambiguously specified for 2sg (i.e., [+PSE, –pl]). Subsequently, the newly created agreement formative spread to other verb classes by analogical extension, becoming obligatory in all contexts. However, we might ask why the presence of distinctive endings in the other verb classes did not prevent the reanalysis of 2sg clitics. For this reason, it would be desirable if we could detect another factor which helped to promote the reanalysis of 2sg clitics. In the following, it is claimed that the change in question was promoted by the fact that a number of highly frequent irregular verbs signal subject agreement only via stem vowel alternations (and not via affixal agreement morphology). Consider Table 22 which lists the 2sg, 3sg and 3pl present indicative forms of four very frequent irregular verbs. Note that these forms are characterized by the absence of suffixal agreement morphology and the fact that they consist merely of the (monosyllabic) verb root, lacking even the theme vowel found with other short verbs. Interestingly, this verb class (sometimes referred to as Kurzverben ‘short verbs’, cf. Nübling 1995; Dal Negro 2004) apparently plays a pioneering role in the grammaticalization of new verbal agreement morphology (at least in Germanic, cf. the development of 1pl -ma in Lower Bavarian and the grammaticalization of new agreement formatives in Walser German discussed in Chapter 5 above). Why? Tentatively, I suggest that this verb class is affected by the grammaticalization process in question more readily because the learner scans the input for the most salient realization of a given inflectional marker. As pointed out by Nübling (1995: 148), agreement distinctions via stem vowel alternation are much more prone to accidental homophony than other means of agreement marking (i.e., suffixes). In other words, in a situation where the learner has to decide whether the subject clitic or the stem vowel alternation is the primary exponent of 2sg verbal agreement, it is quite likely that he/she will go for the clitic if the latter meets all other necessary requirements for being regarded as a verbal inflection. This hypothesis is also supported by research on the acquisition of inflectional morphology, where it has been observed that children generally prefer to use affixes over other morphological means (such as stem vowel alternations) for marking additional (inflectional) distinctions (cf. Clark 1998: 384). Now, if this results in a scenario where the reanalysis of 2sg clitics is not only triggered in the context of vegnere and tegnere (where 2sg is defective), but also with the set of highly frequent verbs such as ‘to go’, ‘to do’ ‘to give’

JB[v.20020404] Prn:2/08/2005; 15:55

F: LA8106.tex / p.38 (2594-2699)

 The Rise of Agreement

and ‘to know’, it is quite probable that the change in question catches on and develops into a model of agreement marking which spreads to other verb classes via analogy. Finally, let me address the development of a new agreement formative for 2pl. At first sight, it appears that this change cannot be accounted for in ways similar to the changes affecting 3rd person forms and 2sg. Apparently, the reanalysis of the clitic neither leads to the development of an exponent which realizes a greater subset of φ-features such as person, number or gender (even the 2pl of short verbs such as vegnere is fully distinctive with respect to person and number, see Table 21), nor are there suffixless ‘short verbs’ where 2pl is marked solely by stem vowel alternations. However, at a second look it appears that the rise of 2pl is susceptible to an explanation analogous to the account offered for the development of 2sg /-st/ in the history of Old High German/Bavarian in Section 6.2.1, where it has been argued that the grammaticalization of the new agreement ending created a form which was additionally specified for verbal mood, distinguishing between present indicative and present subjunctive.47 Consider the agreement paradigms for the present indicative and present subjunctive of a selection of four highly frequent irregular verbs in Table 23 and Table 24. If we compare the 2pl forms (marked by boldface) in Table 23 and Table 24, it becomes apparent that the 2pl present indicative and 2pl present subjunctive forms of these irregular verbs are identical in Vicentino. Thus, we face a situation which is reminiscent of the explanation offered above for the grammaticalization of 2sg /-st/ in Old High German, which by assumption led to the distinction of verbal mood in the relevant contexts. More or less the same scenario can be conceived for the reanalysis of the 2pl enclitic -o in Vicentino.48 Thus, I suggest that the reanalysis in question Table 23. Present indicative of four frequent irregular verbs in Vicentino.

1sg 2sg 3sg 1pl 2pl 3pl

dare ‘to give’

fare ‘to do’

poder ‘can’

vegner ‘to come’

do de da demo dazi da

fo fe fa femo fé/fazi fa

posso poi pole podemo podí pole

vegno vien vien vegnemo vegní vien

Table 24. Present subjunctive of four frequent irregular verbs in Vicentino.

1sg 2sg 3sg 1pl 2pl 3pl

dare ‘to give’

fare ‘to do’

poder ‘can’

vegner ‘to come’

daga dai/daghi daga demo dazi daga

fassa fassi fassa fasemo fazi fassa

possa possi possa podemo podí possa

vegna vegni vegna vegnemo vegní vegna

JB[v.20020404] Prn:2/08/2005; 15:55

F: LA8106.tex / p.39 (2699-2778)

Chapter 6. Morphological blocking and the rise of agreement 

first affected the present indicative of verbs where the existing 2pl present indicative and present subjunctive forms were identical. The development of the new agreement ending 2pl -e+o was licensed due to the fact that the resulting form unambiguously realized verbal mood (i.e., [+indicative]), in contrast to its predecessor. By analogy, the new agreement formative later extended to other verbs and moods, which blurred the morphological motivation behind the original change (similar to Old High German, where 2sg /-st/ spread to the subjunctive as well). Note that in present-day Vicentino the formative -o is still confined to the environment where the reanalysis became structurally possible, that is, inversion contexts.49 Finally, I want to address a general theoretical issue arising in connection with the kind of analysis presented in this section. If the obligatory subject clitics of dialects like Piattino or Vicentino are analyzed as agreement morphemes, this seems to give rise to an instance of multiple exponence, since agreement is realized by both the ‘clitic’ and the existing suffixal agreement morphology. However, as is well-known, most current theories of morphology (with the notable exception of Anderson’s 1992 A-Morphous Morphology) rule out the possibility that a given inflectional feature/morpheme is realized by more than one primary exponent (cf. e.g. Spencer 1991 or Noyer 1997 for some discussion). If this view is adopted, it follows that the grammaticalization of new agreement exponents cannot lead to an outcome where a given underlying agreement morpheme is realized by more than one primary exponent.50 However, recall that I have argued above in Chapters 4 and 5 that the grammaticalization of new agreement exponents often proceeds via a stage where the clitic is reanalyzed as the realization of a dissociated Agr-morpheme which is parasitic on another Agr-morpheme valued in the syntax. Thus, I suggest that in dialects such as Vicentino, the subject clitic is reanalyzed as a dissociated Agr-morpheme which adjoins post-syntactically to a functional head directly above the head that gives rise to suffixal agreement on the verb. The newly created dissociated Agr-morpheme is then licensed under structural adjacency with Agr-on-T (which hosts the verb). The reanalysis in question is illustrated by the bracketed structures in (52) (with the host of the dissociated Agr-morpheme simply labeled F). (52) [TP DPi [T’ V+T+Agr ] ... [νP ti ... → [FP F+Agr [F’ [TP [T’ V+T+Agr ] ... [νP ... Note that the outcome of this reanalysis represents merely a superficial instance of multiple exponence. In fact, the two agreement exponents correspond to two different underlying agreement morphemes (with identical feature content), so no real problem arises here. In other words, the development of multiple agreement is not blocked if each agreement exponent is associated with a separate agreement morpheme, in the case at hand Agr-on-T (spelled-out by the existing suffixal agreement morphology) and a dissociated Agr-morpheme attached to a higher functional head.51

JB[v.20020404] Prn:2/08/2005; 15:55

F: LA8106.tex / p.40 (2778-2812)

 The Rise of Agreement

.. Section summary In this section I have discussed the development of new agreement markers in Piattino and Vicentino, two Northern Italian dialects where the existing verbal agreement morphology is still quite rich. At first sight, this kind of inflectional richness seems to pose a problem for the idea that the grammaticalization of inflectional morphology serves to repair a defective agreement paradigm. However, it has been argued that the individual historical developments in these dialects can be shown to have in fact led to the creation of stronger specified forms if more inflectional distinctions are taken into consideration. More specifically, I demonstrated that the rise of new 3rd person agreement markers in the singular as well as the plural led to the grammaticalization of gender agreement, which was previously absent in Piattino and Vicentino. Furthermore, it was claimed that in Vicentino, the grammaticalization process in question was also sensitive to the realization of verbal mood, arguing that the reanalysis of the 2pl clitic was licensed by giving rise to an inflectional formative which was unambiguously specified for [+indicative], in contrast to its predecessor. Based on the observation that the agreement paradigms of a number of very frequent irregular verbs exhibit certain defects which do not occur with regular verbs, a more complex scenario was developed for the change affecting 2sg clitics in Vicentino. First, it was shown that in the present indicative, the 2sg and 3sg/pl forms of verbs like vegnere ‘to come’ and tegnere ‘to hold’ are homophonous. Second, I demonstrated that in another set of frequent irregular verbs (including ‘to do’, ‘to give’, ‘to go’, ‘to know’), agreement marking for 2sg is signaled only via stem vowel alternations. It was then argued that the grammaticalization of new agreement markers was triggered first only in these two contexts, leading to more distinctive (and salient) inflectional formatives. In a later development, the relevant changes gradually expanded to other verbs via analogical extension, giving rise to the system exhibited by present-day Vicentino. Thus, even in languages where the existing paradigm of agreement marking is quite rich, it is possible to maintain the idea that the grammaticalization of new agreement markers must conform to the Blocking Principle. Crucially, as already hinted at in Section 6.2.1.1 above, to prove that a new form is in fact more specified than its predecessor often requires that a larger portion of the existing inflectional morphology is taken into consideration. Here it is important to note that the additional specification of a given new form does not necessarily involve agreement features such as person, number or gender. Instead, the relevant additional specification may involve any kind of verb-related inflectional features like verbal mood, tense etc. In other words, the Blocking Principle requires (i) that the set of inflectional features realized by the new form includes the set of features realized by the old form and (ii) that the cardinality of the set of features realized by the new form exceeds that realized by the old form at least by one. The next section discusses the development of new agreement formatives in Sutselvan, a Swiss Rhaeto-Romance dialect where the existing agreement morphology appears to be even richer than in the Italian dialects discussed in the present section.

JB[v.20020404] Prn:2/08/2005; 15:55

F: LA8106.tex / p.41 (2812-2888)

Chapter 6. Morphological blocking and the rise of agreement 

. Rhaeto-Romance This section deals with morphological aspects of the changes affecting the Swiss Rhaeto-Romance (RR) dialects examined in Chapter 5 above. Note that the following discussion focuses on the rather recent development of obligatory clitic doubling in Sutselvan (and, partially, Surmeiran, cf. the previous chapter, Section 5.3.3), leaving aside previous developments that led to a reanalysis of C-oriented clitics in the Swiss RR dialects (cf. Section 5.3.2). Most Rhaeto-Romance dialects are quite conservative with respect to their inflectional morphology, that is, they preserve the system of conjugation classes and agreement endings inherited from earlier stages of Romance quite faithfully (cf. Haiman and Benincà 1992 for an overview). This is also true of the Swiss RR dialects. Apart from Surmeiran and Sutselvan, where 2sg and 2pl are (at least partially) homophonous, the dialects show a fully distinctive agreement paradigm in the present indicative, as illustrated in Table 25. At first sight, the partial homophony of 2sg and 2pl should lead us to expect that new agreement formatives develop preferably in this context. However, as noted in Section 5.3.3, the phenomenon of obligatory clitic doubling is restricted to 1st and 3rd person contexts, due to the fact that Sutselvan lacks subject enclitics for 2nd person (similar to the other Swiss RR dialects). As a consequence, there is simply no lexical material available which may feed the grammaticalization of new 2nd person agreement formatives. In the following, I will therefore focus on the question of whether the reanalysis of 1st and 3rd person enclitics has taken place in accordance with the Blocking Principle. First, note that in many instances, enclitics replace the existing agreement morphology in inversion contexts due to a stress rule sometimes referred to as penultimate stress target (Haiman 1971; Linder 1987; Haiman & Benincà 1992). Recall that in all Swiss RR dialects, the syllable structure of finite verbs is governed by a phonological rule which forbids that the main word stress fall on the antepenultimate syllable. In instances where the inflected verb consists of two syllables and receives stress on the penultimate syllable, attachment of the clitic would lead to a stress pattern which is not well-formed. This conflict is often resolved by eliding the original verbal agreement ending (which in the example at hand consists only of the theme vowel -a), ensuring that the antepenultimate syllable does not receive stress (cf. Haiman 1971: 806):52 Table 25. Verbal agreement (present indicative) in five Swiss RR dialects.

1sg 2sg 3sg 1pl 2pl 3pl

Puter

Vallader

Surmeiran

Surselvan

Sutselvan

-Ø -ast -a -ains -ais -an

-Ø -ast -a -ain(a) -aivat -an

-Ø -as -a, -e -(g)n -es, -as -an

-6l -as -a -in, -ein -is, -eis -an

-Ø -(a)s -a -(g)n -(e)s, -(a)s -an

JB[v.20020404] Prn:2/08/2005; 15:55

F: LA8106.tex / p.42 (2888-2984)

 The Rise of Agreement

(53) ella Áchant-a →*Áchant-a=la → Áchant=la she sing-3sg sing-3sg=clit.3sg.fem sing=clit.3sg.fem As a consequence, clitics regularly replace the original verbal agreement morphology in inversion contexts in the Swiss RR dialects, cf. the following example from Sutselvan: (54) A tut puder curr-la sper l’aua giou. at all can run-3sg.fem alongside the-water down ‘She runs down alongside the water as fast as she can.’ (Sutselvan, Linder 1987: 63) Similarly, the clitic appears as the only formative marking the subject’s φ-features in a number of frequent irregular verbs: (55) Egn da quels lev-i ear jou. one of those wanted-1sg also I ‘I also wanted one of those.’ (Sutselvan, Linder 1987: 148) (56) Cunquegl c’igl eara november, vev-la la scola antschiat. since it was November had-3sg.fem the school begun ‘Since it was November, the school had begun.’ (Sutselvan; Linder 1987: 155) (57) Natiral vev-in las matàns radetg sei mailenders. of course had-3pl the girls brought up Milans ‘Of course, the girls had brought up some Milans [pastries].’ (Sutselvan; Linder 1987: 161) Similar to other dialects, 3pl forms are somewhat special in this respect (cf. Note 52). First, there is a 3pl clitic -i which attaches either to the existing ending, in violation of the penultimate stress rule, or may cause elision of the vowel of the agreement ending: (58) A tgea fagev-an=i quegl betga savens ascheia. at home made-3pl=clit.3pl that not often like-this ‘At home, they didn’t do that often in this way.’ (Sutselvan, Linder 1987: 86) (59) Mo gid-n=i ear a tgei, scha tei âs malavetta? but help-3pl=clit.3pl also to you when you have sorrows ‘But do they also help you when you are troubled?’ (Sutselvan, Linder 1987: 87) Only in Sutselvan, however, there is still another ‘clitic’ 3pl -in which completely replaces the original verbal agreement ending in inversion contexts, cf. (60) and (57) above: (60) Qua bavev-in vinars e sa divertetan cun gois. then drank-3pl brandy and refl entertain-3pl with games ‘Then they drank brandy and entertained themselves with games.’ (Sutselvan, Linder 1987: 88)

JB[v.20020404] Prn:2/08/2005; 15:55

F: LA8106.tex / p.43 (2984-3037)

Chapter 6. Morphological blocking and the rise of agreement 

Thus, from the perspective of the learner, the clitics are often the only realization of verbal agreement in inversion contexts. This may give rise to a special verbal inflection confined to inversion contexts, which is sometimes referred to as Inversionsformen ‘inversion forms’ in the traditional literature on the Swiss Rhaeto-Romance dialects (see Linder 1987: Ch. 8 and Section 5.3 above).53 Thus, if the learner realizes that the relevant verb form results from eliding the original agreement ending due to a phonological rule, he will acquire the clitic as a pronominal element. In contrast, if for some reason, he/she misses this fact, he/she will possibly acquire a grammar which differs from the target grammar in that the clitic is reanalyzed as an agreement formative.54 Still, this raises the question of whether this reanalysis is in fact licensed by the Blocking Principle – after all, the present indicative agreement endings are still distinctive in non-inversion contexts. Furthermore, recall that in Bavarian, the development of new agreement formatives in inversion contexts (via Agr-on-C) was apparently sensitive to properties of the existing agreement endings (showing up on clause-final verbs) and was licensed only in contexts where the existing agreement morphology was not distinctive. This seems to suggest that the failure to detect any other potential agreement formatives in inversion contexts apart from the enclitics is probably not sufficient to trigger the change in question if the original (distinctive) agreement endings are robustly attested in other contexts. In other words, we have to look for other factors which helped to promote the reanalysis in question (in line with the Blocking Principle). In the following, it is argued that there are at least two settings where the potential new agreement markers realize more distinctions than the existing forms. First, at least in the case of 3sg, the new forms are also specified for gender, in contrast to the existing agreement formatives (which consist merely of the theme vowel). This is shown in Table 26. Accordingly, the new 3sg agreement formatives clearly satisfy the requirements imposed by the Blocking Principle. However, it is quite obvious that no additional gender specification is involved in the other contexts where obligatory clitic doubling is observed in Sutselvan, that is, 1sg/1pl and 3pl. Concerning the latter, I suggest that the relevant grammaticalization process (leading to Agr-on-C) affected first verb forms which exhibit an agreement paradigm which is less rich than that found in the present indicative. One characteristic of Rhaeto-Romance languages is the presence of two series of agreement endings, referred to by Haiman and Benincà (1992) as primary and secondary desinences. The primary agreement endings are immediately adjacent to the verb root. They signal a rich array of person/number distinctions and at least partially preserve the four conjugation classes inherited from early Romance, leading to Table 26. Development of gender agreement in Sutselvan.

3sg.masc 3sg.fem

Old agreement formative

New agreement formative

-a -a

-a+l -la

JB[v.20020404] Prn:2/08/2005; 15:55

F: LA8106.tex / p.44 (3037-3137)

 The Rise of Agreement

conjugational allomorphy. Furthermore, at least the 1st and 2nd person endings are stressed and trigger a stress shift from the verb root to the agreement suffix. In contrast, the secondary endings do not attach directly to the verb root, but rather to other inflectional suffixes conveying tense or mood information. The secondary endings are generally unstressed, lack conjugational allomorphy and show a greater number of syncretisms in the agreement paradigm (i.e., they signal less person/number distinctions than the primary endings).55 The differences between primary and secondary endings are summarized in Table 27 (cf. Haiman & Benincà 1992: 77). In the dialects under consideration, the primary endings are usually found in the present indicative and the imperative while the secondary endings show up in the imperfect indicative, the present subjunctive, the imperfect subjunctive, and the future. The following discussion focuses on the imperfect indicative and present subjunctive. Consider Table 28, which lists the different agreement endings found in the present indicative (primary) and imperfect indicative (secondary) in Sutselvan.56 In contrast to the present indicative paradigm, where only the 2sg and 2pl forms are similar, the secondary endings found in the present subjunctive and imperfect indicative are much less distinctive. As can be seen from Table 28, 1sg is identical with 3sg, 2sg is identical with 2pl, and 1pl is identical with 3pl. This state of affairs is most economically represented by assuming that there are in fact only three separate Vocabulary items, where /-a/ is simply the elsewhere case.57 (61) a. [+PSE, –Auth] ↔ /-as/ b. [+pl] ↔ /-an/ c. elsewhere ↔ /-a/

(2sg, 2pl) (1pl, 3pl) (1sg, 3sg)

Recall that due to the penultimate stress rule, the clitics often replace the old agreement formatives in inversion contexts. Only in 3sg.masc contexts, the resulting form is Table 27. Primary and secondary agreement endings of Rhaeto-Romance.

Adjacency to verb root Conjugational allomorphy Moveable stress Loss of agreement distinctions

Primary agreement endings

Secondary agreement endings

yes yes yes no

no no no yes

Table 28. Agreement formatives in Sutselvan, present and imperfect (indic.).

1sg 2sg 3sg 1pl 2pl 3pl

Present indicative

Imperfect indicative & present subjunctive

-Ø -as -a -(g)n -es, -as -an

-a -as -a -an -as -an

JB[v.20020404] Prn:2/08/2005; 15:55

F: LA8106.tex / p.45 (3137-3203)

Chapter 6. Morphological blocking and the rise of agreement 

a combination of the old (secondary) agreement ending -a and the clitic -l. When the clitics are reanalyzed as the realization of an agreement morpheme on C, this gives rise to a special verbal inflection in inversion contexts. What I suggest is that this change affected first the impoverished secondary endings, which show up in tenses/moods other than present indicative and imperative. Subsequently, the agreement endings created by this grammaticalization process may spread to other tenses/moods via analogical extension (eventually replacing/enlarging the existing exponents of Agr-on-T in all contexts). If we compare the existing set of secondary agreement endings with the set of agreement formatives resulting from the reanalysis of enclitic pronouns, it is immediately clear that the new agreement paradigm includes more agreement distinctions than the old one (marked by boldface), cf. Table 29. In particular, it becomes apparent that apart from 2nd person forms, which retain the ending -as (due to the fact that there are no clitics available), there is a distinct agreement ending for each slot of the paradigm. Let me now take a closer look at the set of Vocabulary items resulting from the change in question and check whether each of the resulting Vocabulary items actually realizes a greater subset of agreement features than its predecessor. (62) a. b. c. d. e. f.

[+Auth, –pl] [+Auth, +pl] [–pl, +masc] [–pl, +fem] [+PSE] elsewhere

↔ ↔ ↔ ↔ ↔ ↔

/-i/ /-nsa/ /-al/ /-la/ /-as/ /-in/

(1sg) (1pl) (3sg.masc) (3sg.fem) (2sg, 2pl) (3pl)

Note that if the set of Vocabulary items in (62) is posited (which presumably represents the most economical analysis of the paradigm in question),58 it appears that the change in question actually led to some forms which are less specified than their predecessors: 2sg, 2pl /-as/ was previously specified for [+PSE, –Auth], while it is now analyzed as the realization of [+PSE]. In addition, 3pl /-in/ is now the elsewhere case, while its predecessor /-an/ was specified for [+pl]. In other words, while a superficial inspection of the resulting paradigm shows that the change led to distinct forms for each slot of the paradigm (apart from 2nd person contexts), a more fine-grained morTable 29. Development new agreement formatives in Sutselvan, secondary endings.

1sg 2sg 3sg.masc 3sg.fem 1pl 2pl 3pl

Old agreement formative

New agreement formative

-a -as -a -a -an -as -an

-i -as (no change) -al -la -nsa -as (no change) -in

JB[v.20020404] Prn:2/08/2005; 15:55

F: LA8106.tex / p.46 (3203-3275)

 The Rise of Agreement

phological analysis seems to suggest that some changes violated the Blocking Principle. This curious discrepancy calls for some further discussion. First, note that the 2nd person forms in fact do not constitute a problem for the claim that the reanalysis of clitics is constrained by morphological blocking effects, since there is no such change involved. Phonologically, the 2sg/2pl exponent /-as/ remained the same. However, due to changes affecting other parts of the paradigm (namely the creation of stronger specified 1st person forms which block the insertion of /-as/ in these contexts, see Note 57 above), the specification [–Auth] is not any longer necessary to warrant that /-as/ is inserted only in 2nd person contexts. Thus, if it is assumed that the child acquires the most economical paradigm structure involving the least number of feature specifications compatible with the input (Halle 1997; see Note 18 above), the redundant feature specification [–Auth] should not be part of the Vocabulary items realizing 2nd person in (62). In this way, changes affecting one part of an inflectional paradigm (namely, the creation of new 1st person agreement formatives in inversion contexts) may have an effect on other members of the paradigm (in the case at hand, the feature specification of 2nd person forms). What about the new agreement ending 3pl /-in/? In (62) above, /-in/ has been analyzed as the elsewhere form, which is completely underspecified for person and number. Accordingly, /-in/ is apparently less specified than its predecessor /-an/, which realizes the feature /+pl/. This is somewhat surprising if the Blocking Principle is adopted. However, note that the new formative /-in/ is defective only in the context of the newly established agreement paradigm. The apparent problem disappears if we become clear about the fact that the Blocking Principle is not concerned with complete paradigms (i.e., the outcome of the acquisition process), but rather with selecting one out of a set of individual forms which compete for a particular slot in the paradigm. It does not look at the paradigm as a whole.59 If the input contains more than one potential candidate for realizing a specific feature set (say, [+past], [+indicative], [+pl]), the learner must select one of these forms for storage in the Lexicon. By assumption, he/she will go for the most explicit realization of the relevant set of features. Thus, in the case at hand, the learner first compares the immediate competitors (i.e., 3pl /-in/ vs. /-an/) with each other before he/she considers further information like the feature specification of other inflectional formatives already acquired up to this point. With respect to the potential realizations of 3pl, this procedure delivers a clear-cut result: while the formative /-an/ shows up in 3pl as well as 1pl contexts (present subjunctive/imperfect indicative), the immediate competitor /-in/ is confined to 3pl and therefore represents the most explicit realization of this feature combination. Consequently, /-in/ is stored as the realization of 3pl present subjunctive/imperfect indicative in inversion contexts. The eventual complete underspecification of 3pl /-in/ presumably results from a later operation which compares the feature matrices of all members of the paradigm to determine the most economical paradigm/feature structure (cf. Note 18). To summarize, the present section has focused on changes affecting the categorial status of enclitics in the Swiss RR dialect Sutselvan, giving rise to a special form

JB[v.20020404] Prn:2/08/2005; 15:55

F: LA8106.tex / p.47 (3275-3315)

Chapter 6. Morphological blocking and the rise of agreement 

of verbal inflection in inversion contexts (which has not yet spread to verbs in other positions). Similar to the Northern Italian dialects discussed in the previous section, the existing agreement paradigm of Sutselvan seems to be quite rich. However, it has been claimed that the grammaticalization of new agreement formatives did not start in the present indicative, but rather in those tenses/moods which exhibit a defective paradigm of secondary agreement endings. Subsequently, the newly created agreement morpheme on C (and its exponents) spread to other contexts via analogical extension, leading to a special inflection in inversion contexts for all tenses/moods. In addition, I have argued that the change in question was promoted by the fact that the attachment of enclitics often affects the existing suffixal agreement morphology in the Swiss RR dialects, due to a stress rule (penultimate stress target) which requires word stress to not fall on the antepenultimate syllable of the verb. In many instances, this gives rise to the impression that the enclitic replaces the existing agreement morphology, which abets a reanalysis of the enclitic as the realization of a special form of inflection associated with inversion contexts. In the next section, it is argued that language obsolescence may constitute another setting where the loss of existing agreement morphology (due to incomplete acquisition of the target grammar) accelerates the development of new, alternative means of agreement marking.

. Language loss and the grammaticalization of agreement markers In this section, it is argued that minority languages which are in danger to disappear may instantiate an alternative scenario where the grammaticalization of new agreement markers correlates with the existence of a defective agreement paradigm. More specifically, it appears that pronouns may be employed as new markers of verbal agreement if a generation of learners fails to fully acquire the original inflectional paradigm in a sociolinguistic setting where the language in question competes with another language which is dominant and associated with social or cultural prestige. Relevant examples come from the Walser German dialects discussed in Section 5.2.3.2 and American Russian, that is, the form of Russian spoken by the second (or third) generation of Russian immigrants in the United States (Polinsky 2000). Polinsky (2000) describes a set of structural properties of American Russian, focusing on speakers whose linguistic competence exhibits a higher degree of attrition. The speakers under consideration are usually born in the US and acquired Russian as their first language. Later, but still as a child, they switched to English as their primary language, failing to acquire many aspects of the grammar of Russian properly. In particular, the speakers have problems with the complex inflectional morphology of Russian, producing numerous verbal and nominal forms that are not acceptable in standard Russian. Polinsky (2000: 87) summarizes the relevant structural changes as follows:60

JB[v.20020404] Prn:2/08/2005; 15:55

F: LA8106.tex / p.48 (3315-3385)

 The Rise of Agreement

[...] loss of case distinctions; loss of verbal agreement; elimination of the conditional; loss or simplification of reflexivization rules; development of resumptive pronouns; loss of null copying under clause linkage, increased redundancy in discourse.

In this section, I will focus on the use of resumptive pronouns in American Russian and its correlation with the loss of agreement distinctions. Speakers of American Russian tend to use only the masculine singular form of the verb in the past tense while they use the 3rd person singular (or the infinitive) in all other contexts, giving arise to apparent agreement violations (examples taken from Polinsky 2000: 58): (63) a.

American Russian moi roditeli oni kupil drugoj dom. my parents they bought-masc another house ‘My parents bought another house.’ b. Standard Russian moi roditeli kupili ešˇce odin dom. my parents bought-pl more one house ‘My parents bought another house.’

American Russian (64) v universitet knig-i budet dorogo. in university.nom book-nom.pl will-be.3sg expensively ‘Books will be costly when you go to the university.’ Interestingly, this trait of American Russian correlates with another characteristic of this language, namely the extensive use of resumptive pronouns in preverbal position, as shown for example in (63a) above, where the subject ‘my parents’ is resumed by the pronoun oni ‘they’, giving rise to a doubling configuration. The resumptive pronoun is obligatory if the subject is not adjacent to the verb or if the subject consists of two conjoined DPs (the latter condition applies only to 1st and 2nd person contexts): (65) ty vˇcera ty pozvonila moja mat’ dlja manikjur? you yesterday you called.perf my.nom mother.nom for manicure ‘Did you call my mother for a manicure appointment yesterday?’ (Polinsky 2000: 59) (66) [Dima i ja] my byli vmeste v škole. Dima and I we were together at school ‘Dima and I went to school together.’ (Polinsky 2000: 59) There are some indications that the resumptive pronouns in these examples do not resume a left dislocated topic but should rather be analyzed as agreement marking elements (Polinsky 2000: 60ff.). First, whereas in Standard Russian, this kind of topic left dislocation is only used to signal a change of topic in discourse, no such restriction can be observed in American Russian. Second, in Standard Russian the resumptive pronoun is separated from the preceding phrase by an intonational break which may be

JB[v.20020404] Prn:2/08/2005; 15:55

F: LA8106.tex / p.49 (3385-3453)

Chapter 6. Morphological blocking and the rise of agreement 

absent in American Russian.61 Finally, in conjoined clauses, the resumptive pronouns of American Russian exhibit a similar behavior as the subject clitics of Non-Standard French, that is, they are obligatorily present in the second conjunct clause: (67) ona togda ona uvidela moju mamu, i ona govorila s moej she then she saw my mom and she spoke with my mamoj. mom ‘Then she met with my mom and finally spoke to her.’ (Polinsky 2000: 63) Thus, American Russian differs significantly from languages such as English, where subject pronouns are usually elided in second conjuncts if they are coreferent with the subject of the first conjunct. Again, this suggests that the resumptive pronouns of Standard Russian have evolved into agreement markers which are obligatorily present independent of the syntactic context. Polinsky (p. 58f.) proposes a functionalist explanation of these facts which links the development of obligatory resumptive pronouns to the incomplete acquisition of the verbal inflection in American Russian: As verbal agreement deteriorates, there arises a need for some other grammatical mechanism marking the relation between the subject and the predicate. This explains, if only partially, another striking feature of American Russian, namely, the widespread occurrence of the subject resumptive pronoun before the verb.

In the context of the present chapter, the basic idea of Polinsky’s can be captured in more formal terms if it is assumed that the reanalysis of resumptive pronouns (presumably in topic left dislocation contexts) was guided by the workings of the Blocking Principle: in environments where the resumptive pronouns met the necessary preconditions on a reanalysis as agreement markers, the change in question was promoted by the fact that the learner failed to develop full command over the existing agreement paradigm. Instead, he/she acquired the (former) resumptive pronouns, which carry unambiguous person/number/gender distinctions, as an alternative means of agreement marking, in line with the Blocking Principle. It seems that this kind of explanation carries over to the case of Pomattertitsch (Walser German) discussed in Chapter 5 above, Section 5.2.3.2. Recall that in Pomattertisch, enclitic pronouns have evolved into suffixal agreement markers in inversion contexts: Pomattertitsch (68) a. dets bin-i ich. this am-1sg I ‘This is me.’ (Dal Negro 2004: 165) b. d möter set-sch. the mother says-3sg ‘Her mother says.’ (Dal Negro 2004: 171)

JB[v.20020404] Prn:2/08/2005; 15:55

F: LA8106.tex / p.50 (3453-3515)

 The Rise of Agreement

Dal Negro (2004) argues that this change is linked to the special socio-linguistic situation of Pomattertitsch, which is less prestigious than Standard Italian and the surrounding Northern Italian dialects, facing extinction. As a consequence, younger speakers often fail to fully acquire their dialect and in particular the inflectional morphology of the language. Similar to American Russian, this appears to lead to a grammaticalization process which serves to recomplete an otherwise defective agreement paradigm:62 In fact, original verb endings may have started to seem to be too opaque, at least in the case of less fluent dialect speakers, making it difficult to distinguish single forms from one another, while the trend to omit subject pronouns was also spreading. Hence, old, worn out endings have been replaced or rather reinforced by new, more salient and agglutinative endings. (Dal Negro 2004: 178)

Again, the development of new agreement formatives appears to be linked to the incomplete acquisition of the existing agreement paradigm which is a characteristic of language loss. Similar to the case of American Russian, this observation can be attributed to the workings of the Blocking Principle, which favors the acquisition of the most specified agreement formative attested in the linguistic input. In the case at hand, the BP stimulates the reanalysis of former subject enclitics in inversion contexts, replacing the original agreement endings that are not acquired properly.63 Let me now summarize the conclusions reached in the present section. It has been shown that in languages which are threatened by obsolescence, the younger generation often develops only a restricted competence of the target grammar, in particular failing to fully acquire the complex inflectional morphology of the original system. More specifically, it has been demonstrated that in American Russian and the Walser German dialect Pomattertitsch, the incomplete acquisition of the verbal agreement morphology goes hand in hand with the development of obligatory subject clitics, giving rise to instances of clitic doubling if a full subject is present. It has been argued that this phenomenon is also to be attributed to the workings of morphological blocking effects during language acquisition which scan the input for the most specified candidate realizing a given inflectional morpheme. Thus, the limited linguistic competence characteristic of language loss/death may be accompanied by a set of rapid and catastrophic changes which consist not only in the loss of structures of the target language, but interestingly at times may also create new grammatical structures absent in the target language, thereby reflecting properties of UG such as the Blocking Principle (similar observations can be made in pidgin and creole languages, cf. e.g. Bickerton 1984, 1998; Lightfoot 1991, 1999 for discussion). Note, however, that the present discussion covered only a minute fraction of the ever growing number of languages facing extinction. In particular, the languages under consideration most probably do not represent typical examples. Usually, the disappearance of inflectional morphology is not compensated by innovative developments in languages facing extinction. Rather, it appears that the effects of phonological and morphological reduction more often than not win out over processes creating new morphology, similar to ‘healthy’ languages which are not in danger of disappearing (cf. Section 6.2.2 above).

JB[v.20020404] Prn:2/08/2005; 15:55

F: LA8106.tex / p.51 (3515-3587)

Chapter 6. Morphological blocking and the rise of agreement 

. Grammaticalization and multiple agreement in Skou So far, this chapter focused primarily on examples from Indo-European languages. Of course, we expect that similar phenomena can be observed in other language families as well. In this section, I present one such example coming from the Papuan family, drawing on data collected by Mark Donohue. Donohue (2002) discusses the phenomenon of multiple agreement in Skou, an SOV language spoken in some areas along the north coast of central New Guinea. Skou exhibits a set of different, but sometimes co-occurring strategies to mark agreement with a subject: all verbs obligatorily carry pronominal clitics for this purpose. In addition, some verbs undergo vowel and consonant alternations, as well as combinations of these. Donohue shows that the kind of multiple agreement found in Skou most likely resulted from the repeated reanalysis of pronominal clitics as means of verbal agreement marking in the history of this language. Accordingly, the consonant alternations found on the verb root are attributed to an earlier process in which subject pronouns fused with the verb. The subsequent loss of this kind of distinctive agreement morphology (due to phonological reduction) is in turn compensated for by a second wave of cliticization that led to the peculiar kind of multiple agreement marking that can be observed in the present-day language. Before we turn to the historical developments affecting clitic pronouns in some more detail, the following section briefly reviews the different strategies of verbal agreement marking which can be observed in Skou today.

.. Strategies of agreement marking in present-day Skou With all verbs, subject-verb agreement is obligatorily marked by pronominal clitics that attach to the left of the verb (or to the left of a so-called ‘adjunct nominal’ which is part of the verbal complex). The proclitics, which are listed in Table 30, are more or less similar in shape to the free pronouns. For a certain number of verbs (according to Donohue, this class consists of 20 verbs) agreement is solely marked by the use of proclitics. The following examples show that the proclitic must be present to form a grammatical sentence: (69) a. *Peangku fue a wung. girl that die ‘That girl died.’ b. Peangku fue a pe=wung. girl that 3sg.fem=die ‘That girl died.’ Table 30. Proclitic agreement markers in Skou.

1. 2. 3. masc/neut 3. fem

Singular

Plural

nì mè ke pe

ne e te

JB[v.20020404] Prn:2/08/2005; 15:55

F: LA8106.tex / p.52 (3587-3684)

 The Rise of Agreement

Table 31. Agreement via stem vowel alternation: lú weng ‘to sleep’.

1. 2. 3. masc/neut 3. fem

Singular

Plural

lú weng lú weng lú weng ló weng

lú weng lú weng lé weng

Table 32. Agreement via stem vowel alternation: fue ‘to see’.

1. 2. 3. masc/neut 3. fem

Singular

Plural

fue fue fue fu

fi fi fi

Furthermore, the proclitic apparently has no pronominal function: an independent overt subject (a pronoun or a full DP) is always required, as illustrated in (70). According to Donohue, the pronominal clitics are therefore better analyzed as simple agreement markers.64 (70) a. *Pe=wung 3sg.fem=die b. Pe pe=wung. she 3sg.fem=die ‘She died.’ A second strategy to mark agreement with the subject is realized by stem vowel alternations affecting the verb root. In contrast to agreement marking via a proclitic, these vowel changes affect only a couple of verbs and mark only a restricted set of distinctions.65 More specifically, stem vowel alternation is used to mark only the features [feminine] and (less often) [plural] (mostly 3pl), where rounding and backing signals gender and fronting signals plurality. This is illustrated in Table 31 and Table 32 for the verbs lú weng ‘to sleep’ and fue ‘to see’, respectively. With the complex verb lú weng ‘to sleep’, vowel changes are found in the forms for 3sg.fem and 3pl. Note that in the case of lú weng, the vowel alternations affect the socalled ‘adjunct nominal’, a nominal element which cannot be separated from the verb root, while the verb root itself remains unchanged.66 In the case of fue ‘to see’, the vowel changes affect the verb root. Moreover, in contrast to lú weng ‘to sleep’, all plural forms exhibit the relevant stem vowel alternation. As noted above, the vowel changes alone are not sufficient for marking agreement with the subject – an additional proclitic is always required: (71) a.

Ke móe ke=fue. he fish 3sg.masc=see ‘He saw a fish.’

(*Ke móe fue.)

JB[v.20020404] Prn:2/08/2005; 15:55

F: LA8106.tex / p.53 (3684-3782)

Chapter 6. Morphological blocking and the rise of agreement 

(*Pe móe fu.)

b. Pe móe pe=fu. she fish 3sg.fem=see.fem ‘She saw a fish.’

The majority of verbs (68% according to Donohue 2002: 13) exhibits an additional way of agreement marking via consonant alternations which affect the onset of the verb root (or the adjunct nominal). This kind of agreement serves to signal more distinctions than the vowel changes discussed previously. Some verbs (such as há ‘to close’) exhibit up to six different onsets dependent on the φ-features of the subject.67 Still, all verbs that signal agreement via consonant changes must appear with an additional proclitic and an overt subject:68 (72) a.

Pe mè pe=wé. she you 3sg.fem=catch.3sg.fem ‘She caught you.’ b. *Mè wé. c. *Mè pe=wé. (Donohue 2002: 13)

Although there is considerable idiosyncratic variation between different verbs, Donohue identifies a set of underlying consonantal prefixes, which are listed in the last column of Table 33 (Donohue 2002: 12). Another interesting property of the consonant changes comes to light if we compare the shape of the consonantal prefixes which replace the onset of the verb root with the proclitics which are used on all verbs to mark agreement, cf. Table 34. Table 33. Agreement via onset alternations: selected verbal paradigms.

1sg 2sg 3sg.masc/neut 3sg.fem 1pl 2pl 3pl

ké ‘catch’

há ‘close’

wé ‘get’

láng ‘chop’

ang ‘eat’

o ‘seawards’

Prefix

ké bè ké wé ké ké ki

há má ká wá ná há yà

wé pé wé wé wé wé wé

làng pàng làng wàng tàng làng yàng

kang mang kang pang nang ang tang

o mo ko po no o to

Ø-, k-, nmkpnØt-, y-

Table 34. Proclitics and consonantal agreement prefixes in Skou.

1sg 2sg 3sg.masc/neut 3sg.fem 1pl 2pl 3pl

Proclitics

Prefixes

nì mè ke pe ne e te

Ø-, k-, nmkpnØt-, y-

JB[v.20020404] Prn:2/08/2005; 15:55

F: LA8106.tex / p.54 (3782-3844)

 The Rise of Agreement

Even a superficial inspection of the paradigms listed in Table 34 reveals the systematic similarities between the proclitics and the corresponding consonantal prefixes. These similarities clearly indicate that there is a diachronic relation between the proclitics and the consonantal prefixes which is briefly discussed in the next section.

.. The historical origin of multiple agreement marking in Skou As already noted above, it is fairly clear that the consonant prefixes found in Skou developed from former clitic pronouns that underwent phonological reduction und “fused onto the verb” (Donohue 2002: 28) in the course of time. Furthermore, it has been noted that these prefixes occur only with a highly restricted set of onsets in Skou. A comparison of the agreement prefixes found in Skou with those of closely related languages reveals a set of interesting differences. In contrast to Skou, languages such as Dumo, Dusur, and Leitre exhibit a more articulated system of consonantal agreement prefixes. In these languages, the consonantal prefixes attach to a wider range of onsets and the resulting verbal agreement paradigms are generally more complete, that is, there are more distinctions available and there is less amount of syncretism. Donohue (2002: 29) summarizes the differences between Skou and the neighbouring eastern languages as follows: (73) Agreement marking via consonantal prefixes in Skou (i) phonological mergers have collapsed a number of contrasts; (ii) there are less syllable onset types available for inflecting verb roots (only four in Skou, as opposed to six or seven in the east); (iii) phonotactic constraints rule out clusters on onsets; (iv) there is a large number of uninflecting verb roots. Donohue shows that the reduction of consonant clusters led to a collapse of contrasts in many agreement paradigms of Skou (due to the simple (C)V syllable structure of Skou, cf. Donohue 2002: 4, Fn. 4). In some cases, cluster reduction resulted in the complete loss of consonantal agreement marking, giving rise to the class of ‘non-inflected’ verbs, where agreement is solely marked by the use of a proclitic in the present-day language.69 Crucially, all languages that retained a rather rich system of consonant prefixes have not developed a further way of agreement marking via proclitics. Donohue therefore concludes – rightly in my view – that there is a connection between the loss of agreement distinctions and the second wave of pronominal clitics that became reanalyzed as agreement markers. However, in contrast to Donohue, who proposes a functionalist explanation of this development, I suggest that the loss of agreement distinctions paved the way for the reanalysis of subject clitics as new means of verbal agreement marking in accordance with the Blocking Principle. More specifically, it seems likely that the grammaticalization of new verbal agreement markers began in those contexts where cluster reduction led to the complete loss of consonantal

JB[v.20020404] Prn:2/08/2005; 15:55

F: LA8106.tex / p.55 (3844-3914)

Chapter 6. Morphological blocking and the rise of agreement 

agreement distinctions. In a subsequent change, the new agreement markers spread to other verb classes via analogical extension, similar to the examples from Vicentino and Sutselvan discussed above. Presumably, the latter change involved the insertion of a dissociated agreement morpheme to some higher functional head, giving rise to the form of multiple agreement exhibited by Skou (cf. Chapter 3 above).

. Conclusion In this chapter, I have focused on the observation that across languages, the creation of new inflectional morphology generally serves to rebuild previously defective paradigms, exchanging weak old forms for stronger new forms. It has been claimed that this observation can be explained if it is assumed that the grammaticalization of new agreement markers from former clitic pronouns is triggered if this change leads to new agreement markers which are more distinctive than the relevant agreement formatives which already exist in the language. This has been modeled as the outcome of blocking effects which operate during language acquisition and block the acquisition of a less specified form if a more specified form is attested in the Primary Linguistic Data. Thus, if the learner encounters more than one potential realization of a certain feature combination in the input, he/she will select the candidate which realizes the greatest subset of the relevant agreement features for storage in the Lexicon. The relevant principle of grammar, dubbed the Blocking Principle, is understood as an economy principle which shapes the acquisition of inflectional morphology, warranting an optimal and non-redundant Lexicon (and paradigm) structure. Similar to other (structural) economy principles (cf. Clark & Roberts 1993), the Blocking Principle is called into service only if the cues provided by the input data are for some reason ambiguous and not sufficient for the acquisition of certain properties of grammar. The Blocking Principle was introduced in Section 6.2 by looking at the grammaticalization of new agreement formatives in the history of Bavarian, arguing that an approach in terms of morphological blocking explains that this change affected only underspecified slots of the agreement paradigm. It was shown that this approach accounts not only for the development of new agreement exponents in Bavarian, which unambiguously mark 2pl (/-ts/) and 1pl (/-ma/), but also for the much earlier development of 2sg /-st/, which took place in the Old High German period and led to a 2sg present form that was unambiguously specified for verbal mood (indicative), in contrast to its predecessor, /-s/. Thus, assessing of whether a certain change complies with the Blocking Principle sometimes requires that inflectional distinctions other than person and number are taken into consideration. Moreover, I argued that the diachronic analysis presented in this section makes available a new explanation of the peculiar person/number restrictions on complementizer agreement and pro-drop in present-day Bavarian which have been observed in Chapter 5. More specifically, it was claimed that the limited distribution of these properties reflects the morphological contexts where the reanalysis in question was licensed by the Blocking Principle.

JB[v.20020404] Prn:2/08/2005; 15:55

F: LA8106.tex / p.56 (3914-3962)

 The Rise of Agreement

Subsequently, I discussed a set of further theoretical issues in connection with this approach to the rise of inflectional morphology, focusing on the status of morphological doublets (Section 6.2.3) and the fact that most changes seem to expand the domain of less marked, more regular forms (Section 6.2.2), which seems to be at odds with the claim that stronger inflectional formatives are preferred. In Section 6.2.2, it was argued that analogical changes which lead to the loss of stem allomorphy can also be attributed to the workings of morphological blocking if it is assumed that the learner prefers to acquire the stem alternant which manifests the characteristic phonemic distinctions of a given lexical root in the most unambiguous way. In addition, I considered cases where an underspecified exponent gains a wider distribution in a paradigm, arguing that this kind of paradigm leveling is to be attributed to the expansion of the domain of Impoverishment rules which bleed the effects of the Blocking Principle. Concerning the existence of morphological doublets, that is, apparent multiple competing realizations of a given inflectional head, I argued in Section 6.2.3 that the Blocking Principle allows the coexistence of morphological doublets only if the relevant agreement formatives are realizations of different agreement morphemes with identical feature content (e.g., Agr-on-C and Agr-on-T). In Section 6.3, it was demonstrated that the pioneering role of 1st and 2nd person in the historical development of verbal agreement marking (cf. Chapter 1) can be attributed to the workings of the Blocking Principle if it is assumed that 3rd person forms are underspecified for person features. As a result, the grammaticalization of 3rd person forms is less likely to be triggered than the development of 1st and 2nd person markers, which are generally specified for a set of [person] features. The Sections 6.4–6.8 presented a set of case studies which further corroborated the hypothesis that the reanalysis of pronominal clitics is guided by morphological blocking effects. Section 6.4 dealt with the development of new agreement formatives in Non-Standard French. It was shown that the grammaticalization process in question proceeded in line with the Blocking Principle, affecting only those slots of the paradigm where the existing agreement morphology is non-distinctive. Concerning the question of why the number and gender specification of the 3rd person clitics il(s) and elle(s) is apparently not sufficient to trigger the grammaticalization process in question, I claimed that verbal agreement in gender cannot evolve unless the language has not acquired a full paradigm of number distinctions for all persons (assuming that φ-features are organized in a feature geometry where number features dominate gender features). Furthermore, I conjectured that the limited distribution of plural -s (which is only perceivable if the following word begins with a vowel) reduces its visibility for the workings of the Blocking Principle. In Section 6.5, I discussed the development of new agreement markers in Piattino and Vicentino, two Northern Italian dialects where the existing verbal agreement morphology is quite rich. It was argued that the changes in these dialects can be shown to have led to the creation of stronger specified forms if more inflectional distinctions are taken into consideration. More specifically, I demonstrated that in both dialects, the grammaticalization process in question led to 3rd person agreement markers which

JB[v.20020404] Prn:2/08/2005; 15:55

F: LA8106.tex / p.57 (3962-4013)

Chapter 6. Morphological blocking and the rise of agreement 

are additionally specified for gender. Furthermore, it was claimed that in Vicentino, the grammaticalization process in question was also sensitive to the realization of verbal mood, giving rise to a 2pl form which was unambiguously specified for [+indicative], in contrast to its predecessor. Finally, a more complex scenario was devised for the development of new 2sg agreement markers in Vicentino. It was argued that 2sg clitics were first reanalyzed in the context of a set of very frequent irregular verbs which are characterized by a defective agreement paradigm. By assumption, the new agreement formatives expanded to other verbs via analogical extension in a later gradual development. A similar scenario, where a change begins in a certain verb class before it gains a wider distribution, was proposed for the Swiss Rhaeto-Romance dialect Sutselvan in Section 6.6. I argued that the reanalysis of enclitics was initially confined to contexts where Sutselvan exhibits a set of impoverished (secondary) agreement endings which appear in all tenses/moods apart from the present indicative and imperative. Subsequently, the newly created agreement exponents spread to other contexts via analogical extension, leading to a special inflection in inversion contexts for all tenses/moods. In addition, I suggested that the change in question was promoted by the fact that the existing agreement morphology is often truncated when enclitics attach to the verb in inversion contexts, due to a stress rule (penultimate stress target) which requires word stress to not fall on the antepenultimate syllable of the verb. In Section 6.7, I showed that the Blocking Principle can also be invoked to account for the observation that new means of agreement marking may evolve in sociolinguistic contexts where speakers develop only a restricted competence of the target grammar. It was argued that the development of obligatory clitic doubling in American Russian and certain Walser German dialects which are threatened by extinction is a consequence of the incomplete acquisition of the existing verbal agreement morphology: if the learner fails to acquire the existing agreement distinctions properly, he/she might scan the input for a more salient (and more specified) candidate realizing a certain inflectional morpheme, which might lead to rapid and catastrophic changes before a language finally vanishes. In Section 6.8, I discussed data from the Papuan language Skou, in which subject clitics were repeatedly reanalyzed as agreement markers. It was argued that the grammaticalization of these new proclitic agreement markers balanced the loss of distinctive verbal agreement markers, which resulted from the phonologically triggered reduction of consonant clusters. Although the brief discussion could not really do justice to the complex agreement patterns exhibited by Skou, the evidence presented was nevertheless suggestive, indicating that the reanalysis of subject clitics as new means of verbal agreement marking proceeded in line with the Blocking Principle.

JB[v.20020404] Prn:2/08/2005; 15:55

F: LA8106.tex / p.58 (4013-4076)

 The Rise of Agreement

Notes . Often the intuition behind the blocking effect is that the availability of an irregular form blocks the use of the regular form. Note that a formulation of the blocking effect that refers to the irregular/regular distinction covers a wider array of cases and is not confined to inflectional morphology (e.g., it includes cases such as the famous thief vs. *stealer contrast). . Note that this use of the term ‘defective’ actually deviates from the traditional understanding of ‘defectiveness’ in morphological theory. As pointed out by Albert Ortmann to me, the notion ‘defective paradigm’ is commonly used to refer to a situation where a given lexical item fails to exhibit certain inflected forms (such as some of the English modals which cannot show up as infinitives: *Peter wants to can play the guitar.). However, for expository reasons, I will continue to use the notion ‘defective paradigm’ to refer to an agreement paradigm which exhibits a number of syncretisms. . The correlation between a defective agreement paradigm and the obligatory presence of pronominal clitics has already been noted in traditional work on the history of the Romance languages (including French and various Northern Italian dialects), see Kuen (1957) or Wartburg (1970), for example. Note that the idea (going back to Gabelentz 1891) that grammaticalization processes are motivated by the need to compensate for the loss of distinctions due to phonological reduction is widely held in the literature on grammaticalization, cf. Lüdtke (1980), Hopper and Traugott (1993), Haspelmath (1995), Siewierska (1999, 2004), Ariel (2000), and Lehmann (2002), among others. . Possibly, the Blocking Principle can be reduced to general principles governing the procedure of Vocabulary Insertion, that is, the Subset Principle (cf. Chapter 2). Recall that the Subset Principle requires that Vocabulary items realizing functional morphemes (f -morphemes) enter into a competition which warrants that only the most specified candidate is inserted in a given insertion context. In contrast, exponents of lexical morphemes (l-morphemes) are freely inserted, that is, there is no competition. If it is further assumed that grammaticalization is to be analyzed as a change in which an exponent of an l-morpheme changes into an exponent of an f -morpheme, one might posit that this transition is only possible if the resulting inflectional formative fulfills the Subset Principle, that is, qualifies as the most specified candidate with respect to the insertion context at hand. Note that the enhanced paradigmaticity of grammaticalized elements (cf. Chapter 2 above) is presumably also to be attributed to this change concerning the insertion procedure, in the sense that the Vocabulary items which compete for realizing a given f -morpheme share a certain feature specification which qualifies them for entering into this competition and gives rise to the impression that these Vocabulary items constitute a certain paradigm. . Another piece of evidence which helps us to gain some insights into the workings of blocking effects during language acquisition comes from the chronological sequence of the acquisition of inflectional morphology (cf. Clark 1998). It is a well-known fact that children typically begin to use a certain inflection sporadically until they have mastered the relevant morphological rule. From this point on, they use the given inflection consistently and produce over-generalized forms such as foot-s and man-s (in the case of the English plural inflection). In a subsequent development, they come to recognize the limits of the rule in question and replace the overregular forms with the correct irregular ones (e.g., feet and men). If we apply these observations to the acquisition of (irregular) inflectional formatives in general, this suggests the following two things: (i) presumably, the Blocking Principle becomes relevant only after the acquisition

JB[v.20020404] Prn:2/08/2005; 15:55

F: LA8106.tex / p.59 (4076-4157)

Chapter 6. Morphological blocking and the rise of agreement 

of basic morphological rules (perhaps due to maturation processes); (ii) at this stage of language acquisition, the storage of inflectional formatives for a given paradigm is not yet fixed, so irregular forms can replace over-regular forms (irregular forms are presumably learned in a one-by-one fashion, cf. Bybee & Slobin 1982; Kiparsky 1983; Prasada & Pinker 1993). . As noted in Chapter 4 above, the pronoun must fulfill a whole set of independent necessary conditions before it can be reanalyzed as the exponent of an agreement morpheme (e.g., it must be a phonologically reduced clitic that cannot receive stress, the reanalysis has to satisfy a set of syntactic conditions such as adjacency to the host, etc.). . Presumably, the BP ensures acquisition of the more specified form only if the relevant candidate form is already robustly attested in the PLD. In other words, frequency is another factor that determines the acquisition of morphology – if a less specified form is much more frequent than the more specified form, then the learner will probably acquire the less specified form, despite the workings of the BP. Furthermore, regularization is a factor that might work against the effects of the BP: In the absence of robust evidence for a irregular/more specified form, the learner may acquire a regular/underspecified form for a given verb as a default. Evidence from language acquisition (Prasada & Pinker 1993; Clark 1998) and language change (Taylor 1994) shows that this process affects primarily verbs that are less frequent in the PLD, a fact which is in line with the properties of the BP suggested above (i.e., the BP selects between forms that are robustly attested in the input). . As Sauerland (1996: 24) puts it: “If two feature complexes in the same position and in the same syntactic environment get spelled out by the same phonological string, the lexical entry of this phonological string has to be unspecified with respect to all contrasting features of the two feature complexes.” . For 13th century Bavarian, the sets of Vocabulary items that competed for insertion into the Agr-morpheme (present tense indicative) are thus as follows. (i) lists the set of items prior to the development of 2pl /-ts/, (ii) shows the situation after the change in question has taken place. [1, +pl] [1, –pl] [2, –pl] [pl] elsewhere

↔ ↔ ↔ ↔ ↔

/-an/ Ø /-st/ /-ant/ /-t/

(ii) [1, +pl] [2, +pl] [1, –pl] [2, –pl] [pl] elsewhere

↔ ↔ ↔ ↔ ↔ ↔

/-an/ /-ts/ Ø /-st/ /-ant/ /-t/

(i)

. The relevant sets of Vocabulary items for 18th century Bavarian are listed below. (i) lists the items prior to the rise of 1pl /-ma/, while (ii) shows the resulting paradigm. (i)

[2, +pl] [1, –pl] [2, –pl] [pl] elsewhere

↔ ↔ ↔ ↔ ↔

/-ts/ Ø /-st/ /-an/ /-t/

JB[v.20020404] Prn:2/08/2005; 15:55

F: LA8106.tex / p.60 (4157-4223)

 The Rise of Agreement

(ii) [1, +pl] [2, +pl] [1, –pl] [2, –pl] [+pl] elsewhere

↔ ↔ ↔ ↔ ↔ ↔

/-ma/ /-ts/ Ø /-st/ /-an/ /-t/

. Note that the initial vowel in formatives such as -emês is actually not part of the agreement suffix, but rather a so-called ‘theme vowel’ that originally served to derive verb stems from roots. . Strong verbs and the weak verbs of conjugation class 1 exhibit -is and -ês for 2sg present indicative and 2sg present subjunctive, respectively. Here, the difference in vowel quality was perhaps not salient enough to differentiate the forms. Furthermore, the difference was presumably further weakened by phonological erosion that affected non-stressed final syllables. Alternatively, one might assume that the change first affected the weak verbs of the conjugation classes 2 and 3 and spread later to other verb classes by analogy. . In fact, the syntactic consequences of morphological erosion represent one of the best studied phenomena in diachronic generative syntax (cf. e.g. van Kemenade 1987; Lightfoot 1991, 1999; Roberts 1993a, 1997; Falk 1993; Allen 1995; Haeberli 1999; Pintzuk 1999; Lightfoot 2002). . The stem vowel alternations found in older Yiddish have different historical origins. With verbs such as grfbn ‘to dig’, the vowel alternations were introduced by a phonological process (Umlaut) operative in Middle High German (note that Yiddish is assumed to have evolved from this historical stage of German) which changed an a in the verb stem to an e in the 2sg and 3sg, due to the presence of an /i/ in the suffixes of these forms (strong classes VI and VII). In contrast, vowel alternations between singular and plural forms (in the present tense) of verbs such as visn ‘to know’ are of a more ancient origin. Verbs such as visn are so-called preteritepresents, the present tense of which derives from Proto-Indo-European perfect forms where stem vowel alternations distinguished between singular and plural (Ablaut). . Recently, this intuition has been formalized in Optimality Theoretic terms, positing constraints such as UNIFORM EXPONENCE or PARADIGM UNIFORMITY (cf. e.g. Kenstowicz 1997; Steriade 2000), which prefer inflectional paradigms to be uniform. Analogical leveling is then described in terms of constraint re-ranking. . To be sure, further questions arise concerning the diffusion of analogical leveling in a speaker community. Concerning the changes affecting Yiddish, Albright (2002: 29) stresses that his approach is not primarily concerned with the latter problem, but rather deals with the question which forms may be affected by analogical leveling (my emphasis, E.F.): “There are clearly many factors at play in determining how willing a community is to adopt new forms; I conjecture that the thoroughness of the change in Yiddish may have been facilitated by the lack of a standard language or widespread literacy, and perhaps even by a conscious desire to differentiate Yiddish from German. The model that I am presenting here is simply an attempt to predict which forms would have been available as potential regularizations for Yiddish speakers, and which would not.” . Noyer (1997) discusses several examples of analogical change which can be observed in the historical development of a number of Afroasiatic languages and analyzes them in terms of the gain or loss of Impoverishment rules. Another example which is quite similar to the change depicted in Table 10 involves the loss of gender distinctions in the plural part of the verbal agreement paradigm of Egyptian Arabic (Noyer 1997: 58ff.).

JB[v.20020404] Prn:2/08/2005; 15:55

F: LA8106.tex / p.61 (4223-4293)

Chapter 6. Morphological blocking and the rise of agreement  . Halle (1997: 430) speculates that the expansion of Impoverishment rules is linked to the economy principle in (i), which is supposed to minimize the load on memory for the learner (see Noyer 1997: lxxxi for similar considerations): (i)

The number of features mentioned in the Vocabulary must be minimized.

Due to the fact that Impoverishment expands the domain of underspecified forms, (i) may promote the development of Impoverishment rules which delete certain morphosyntactic features of syntactic heads in certain contexts. Thus, at first sight, an economy principle such as (i) seems to contradict the workings of the Blocking Principle, since the latter requires ‘new’ Vocabulary items to realize more feature distinctions than the existing items. However, note that (i) is in fact not intended to mean that the learner selects the least marked/specified of several potential candidates and stores it in the lexicon. Rather, it is supposed to ensure that the child acquires the most economical lexical inventory compatible with the input he/she is exposed to. For example, if a feature specification [+speaker] is sufficient to guarantee that a 1sg exponent is inserted in the contexts where it appears in the input, the child will not acquire a redundant feature specification [+speaker, –hearer] for this exponent. Crucially, however, this function of (i) does not interfere with the claim that the learner scans the input for the most marked (and therefore most salient) realization of a given inflectional head. In other words, the Blocking Principle ensures that the most specified candidate is selected, while (i) warrants that this candidate is assigned a non-redundant feature specification. Thus, the two principles actually work hand in hand during acquisition, thereby warranting that the acquisition process leads to an optimal paradigm and Lexicon structure. . It is quite obvious that languages with agreement syncretisms may exist happily over centuries. This seems to support the view that language change is a contingent process (cf. Lightfoot 1999). On this view, properties of UG draw up the boundaries of possible changes, but crucially, they do not trigger changes. Note that this apparently conflicts with the workings of the BP as portrayed above, where it has been argued that the BP actually may trigger a change in the grammar if more than one phonological exponent is available for a given inflectional head. Here, I will simply note this tension without trying to resolve it. . As noted above in Chapter 5, the development of 2sg -s+t and 2pl -t+s presumably proceeded in a similar fashion, with the new formative initially confined to inversion contexts. . In most cases, the observed variation can be traced back to northern innovations that produced ‘new’ regular (weak) forms for historically irregular (strong) verbs such as welk vs. walked, awoke vs. awaked etc. Note that this kind of morphological regularization is a characteristic trait of imperfect second language acquisition by adults. Taylor therefore concludes that the weak past tense variants were the outcome of an intense language contact that obtained between Scandinavian immigrants and speakers of northern English dialects from the late Old English period on. Subsequently, the new forms spread to other (southern) ME dialects via dialect mixture. A remarkable number of past tense doublets can be found in the contemporary language of the London area, which was a melting pot of different dialects in the ME period. For example, Taylor shows that in the work of Chaucer, 24 verbs can be found that exhibit varying forms. In contrast, the Paston letters, which were written in a local Norfolk dialect approximately at the same time, contain only three pairs of morphological doublets for past tense verbs. . However, the scenario envisaged by Kroch can probably account for the gradual replacement of the old ending in clause-final contexts, which apparently affects one verb after the other, starting with frequently used short verbs such as ‘have’ and ‘do’ (see Chapter 5).

JB[v.20020404] Prn:2/08/2005; 15:55

F: LA8106.tex / p.62 (4293-4362)

 The Rise of Agreement . In addition, we have to acknowledge that there is a difference between the grammar change as such and its diffusion in the speaker community. Thus, even if a given change has taken place (in the grammar of some speakers), it does not necessarily prevail, due to socio-linguistic factors. In the case of 1pl /-ma/, for example, the change is restricted to some Lower Bavarian varieties which became marginalized due to the influence of urban (i.e., Munich) Bavarian and are nearly extinct today (cf. Altmann 1984; Kollmer 1987). . Recall that in Chapter 3, Note 62, it has been assumed that an independent principle of Morphological Structure requires that in a given head complex, only the hierarchically highest of two identical agreement morphemes is subject to Vocabulary Insertion, ruling out forms of the type V+Agr-on-T+Agr-on-C. Presumably, this principle forces the learner to assume the existence of a separate, higher agreement head when the relevant exponent resulting from the reanalysis of a clitic differs from the (existing) agreement ending found in other contexts: if the new exponent (found in inversion contexts) differs from the old agreement ending (found in clause-final position), then the learner will assume that the new form is the exponent of an additional, hierarchically higher agreement morpheme the spell-out of which ‘overwrites’ the old exponent in a certain syntactic context. Note that the relevant input did not contain forms which could be reanalyzed as instances of double agreement (i.e., V+Agr-on-T+Agr-on-C): in the case of the enlargements 2sg -s+t and 2pl -t+s, the combination of old agreement marker and clitic is reinterpreted as a single new agreement ending while in the case of 1pl /-ma/, the reanalysis first affected bisyllabic verbs, where the old agreement ending has been deleted in the presence of a clitic, quite similar to the effects of the penultimate stress rule in the Swiss Rhaeto-Romance dialects (cf. Section 5.3.4). . The condition that the newly created agreement exponent is confined to the C-system can be expressed by specifying the syntactic insertion context in the relevant lexical entry, for example: (i)

[1, +pl/C] ↔ /-ma/ [+pl] ↔ /-an/

Note that initially, the old and the new forms do not compete for the realization of Agr-on-T. The gradual expansion of the new form to Agr-on-T can then be modeled as the loss of the contextual restriction in the lexical entry of the more specified form, which eventually drives the old form out of the grammar, presumably along the lines envisaged in Kroch (1994). . Note that when the reanalysis in question took place, the input presumably still included examples with full (instead of clitic) pronouns where the old, underspecified agreement ending showed up on verbs in inversion contexts. Still, the old endings were not acquired as realizations of Agr-on-C. . The special role of 1st and 2nd person is also reflected by the historical developments in Bavarian discussed above. In cases where 1st/2nd person forms and 3rd person forms were homophonous, the grammaticalization processes always created new, more specified 1st/2nd person forms, while in the 3rd person, the old forms were preserved. . Further support for a model that makes use of only two instead of three person features comes from an interesting cross-linguistic restriction on possible pronoun systems (Bobaljik & Baker 2002). On the basis of the traditional model that includes separate features for 1st, 2nd, and 3rd person, we should expect to find languages that distinguishes up to seven different morphologically simple pronouns in the plural:

JB[v.20020404] Prn:2/08/2005; 15:55

F: LA8106.tex / p.63 (4362-4425)

Chapter 6. Morphological blocking and the rise of agreement 

(i)

a. 1+2 inclusive ‘we’ b. 1+3 exclusive ‘we’ c. 1+2+3 complete ‘we’ (refers not only to speaker and hearer, but also to other persons) d. 1 groupspeak ‘we’ (refers to a plurality of speakers, e.g., a crowd shouting we want more! at the end of a concert) e. 2 exclusive ‘you’ f. 2+3 inclusive ‘you’ g. 3 ‘they’

However, the attested maximum number of distinct morphologically simple plural pronouns is apparently four (as exemplified e.g. by the Australian Aboriginal language Kitja, Dixon 2002: 247). In descriptive terms, the fact that only four out of seven possible interpretations are realized by separate pronominal forms is due to the following pattern of syncretisms, which seems to hold universally across languages (cf. Bobaljik & Baker 2002): (ii) a. b. c.

inclusive ‘we’ = complete ‘we’ exclusive ‘we’ = groupspeak ‘we’ exclusive ‘you’ = inclusive ‘you’

(iia) asserts that in all languages that show the maximum number of four different plural pronouns the form for inclusive ‘we’ is also used to express the reading of complete ‘we’. Similarly, according to (iib), there are no morphologically simple pronouns that refer solely to groupspeak ‘we’, whereas (iic) states that there can be only a single form that is used for reference to both exclusive ‘you’ and inclusive ‘you’. Interestingly, exactly this pattern is predicted by a binary person feature system, in which it is not possible to distinguish between inclusive ‘we’ and complete ‘we’. Since there is only a single feature combination that corresponds to both inclusive ‘we’ and complete ‘we’, one and the same pronoun must be used to refer to these different readings of 1st person plural. The other universal syncretisms receive a similar explanantion, cf. (iii) a. b. c.

[+Auth, +PSE] = inclusive ‘we’ & complete ‘we’ (cf. iia) [+Auth, –PSE] = exclusive ‘we’ & groupspeak ‘we’ (cf. iib) [–Auth, +PSE] = inclusive ‘you’ & exclusive ‘you’ (cf. iic)

. Note that you is also the form used in the context of 2sg. Therefore, you is underspecified for number features. Since you and they realize the same number of features, the ordering between (20b) and (20c) is necessary to prevent they from being inserted as the realization of 2pl. . Note that this generalization is apparently contradicted by English, where 3.sg.pres.indic. /-z/ appears to be the only verbal agreement ending. This seems to suggest that the relevant Vocabulary item carries a person specification which is absent otherwise (in all other contexts we find the elsewhere case -Ø, which is completely underspecified for person and number). However, see e.g. Halle (1997) for an analysis where 3.sg.pres.indic. /-z/ is analyzed as an inflectional marker that in fact does not realize person distinctions, but rather carries the feature specification [–pl, +pres, +finite]. Ian Roberts pointed out to me an alternative way of analyzing 3.sg.pres.indic. /-z/, in which /-z/ is treated as the non-default ending in a system that makes use only of a two-way default/non-default distinction in the present tense indicative (see Haeberli 2004 for related considerations). . This reasoning also provides an answer to the question of why new agreement suffixes developed for 2pl and 1pl, but not for 3rd person forms in the history of Bavarian. Note that the latter

JB[v.20020404] Prn:2/08/2005; 15:55

F: LA8106.tex / p.64 (4425-4492)

 The Rise of Agreement

development would have been equally sufficient to repair a defective paradigm where 2nd (and 1st) person forms are homophonous with 3rd person forms. However, if 3rd person forms are inherently underspecified for person features, they are less prone to be replaced by new, equally underspecified exponents, which explains the fact that the new suffixes developed only in 1st and 2nd person contexts. . Guiraud (1968: 98) notes that in Old French, SVO order requires an overt subject pronoun in 97% of all cases, whereas the ratio of overt subject pronouns is only 15% in OVS orders where the clause-initial object is a full DP (i.e., where the V2 constraint is satisfied for independent reasons). . Cf. Greenberg (1966a: 94): “Universal 32: Whenever the verb agrees with a nominal subject or nominal object in gender, it also agrees in number.” . The structure in (37) represents a modified version of proposals by Harley and Ritter (2002), who assume for example an elaborate structure of number features to account for intricate number systems which show distinctions such as dual, trial or paucal. For expository reasons, however, I chose to adopt a somewhat simplified model, which is sufficient for the present purposes. . In addition, we might speculate that, somewhat paradoxically from the viewpoint of the Blocking Principle, it is exactly the gender distinctions which lead the learner to assume that il/elle and ils/elles are still pronouns and not agreement markers: recall that gender features are a well-established aspect of the pronoun system of French, but completely absent in the verbal agreement paradigm (at least in the case of finite verbs). It is therefore conceivable that it is this difference which inhibits a categorial reanalysis of il/elle and ils/elles. . Gerlach (2001, 2002) proposes an Optimality Theoretic account of the distribution of clitics in Piattino. Note furthermore that according to Gerlach (2002: 225), the 3sg and 1pl forms actually do not have an agreement ending at all, but end merely in the theme vowel -a. For expository reasons, I nevertheless labeled the respective forms 3sg and 1pl in the glosses. . Historically, the 1pl an developed from impersonal structures of the kind homo (‘man’) + verb-3sg, quite similar to the use of on + verb-3sg in colloquial French (cf. e.g. Savoia 1997: 78). . As noted above in Note 36, -a is usually analyzed as a theme vowel and does not constitute an agreement suffix. Furthermore, Gerlach (2002: Fn. 24, p. 235) argues that 3pl -en is only the realization of [number] and not of [person]. . I am grateful to Adriana Castagna and Ermenegildo Bidese for much discussion and native speaker judgements on Vicentino. . Note that in the 3sg and 3pl, the attachment of the enclitic leads to a change in the quality of the theme vowel ([a] → [e]). . Of course, the full 2pl pronoun voaltri can be added for reasons of emphasis: (i)

Voaltri a vegní da Durlo. you.pl clit.2pl come-2pl from Durlo ‘You come from Durlo.’

. Poletto (1999, 2000) links the morphological properties of the different kinds of clitics with their different syntactic distribution. More precisely, she identifies four different types of clitics in the Northern Italian dialects and claims that these clitics realize different kinds of agreement heads (some above and some below NegP) each of which is associated with a different set of mor-

JB[v.20020404] Prn:2/08/2005; 15:55

F: LA8106.tex / p.65 (4492-4565)

Chapter 6. Morphological blocking and the rise of agreement 

phosyntactic features. Accordingly, AgrS is split into a series of functional heads with different content, corresponding to the features [±deictic], [person], [gender], and [number]. . Interestingly, in the case of 3sg.masc, the proclitic (e)l does not attach to the verb, but rather to the preceding negation no, presumably for phonological reasons (no+l leads to a wellformed syllable structure, while the combination of l+vien would create a consonant cluster in the onset). (i)

Łu no=l vien a casa stasera. he not=clit.3sg.masc come to home tonight ‘He is not coming home tonight.’

. Similar to other Romance languages, the negation no accompanies verb movement in inversion contexts: (i)

No vien=łe mia a casa stasera? not come=clit.3pl.fem prt to home tonight ‘Are they really not coming home tonight?’

. This leaves open why distinctive and non-distinctive ‘clitics’ occupy different positions relative to the negation no (see e.g. (51) above). However, there are some indications that this difference is in fact another consequence of the grammaticalization process in question. In a study of the grammaticalization of agreement markers from former subject clitics in the history of Veneto, Poletto (1995) shows that the linear order of clitic, verb and negation changed as a result of the reanalysis of clitics as realizations of an AgrS head. Thus, while in Renaissance Veneto the clitics preceded the negation no, they follow no in the present-day language (according to Poletto, this change took place in the 17th century): (i)

La no vaga a mio conto. she not goes on my count (Renaissance Veneto, Calmo, 79; Poletto 1995: 301)

(ii) No la vien. not she come ‘She does not come.’ (Veneto; Poletto 1995: 297) It is thus likely that a similar change in word order accompanied the grammaticalization of new agreement markers in the history of Vicentino. More precisely, the reanalysis in question was presumably only possible in contexts where the clitic was adjacent to the verb (in line with the Word building constraint, cf. Chapters 4 and 5 above). After the reanalysis, the newly developed agreement markers could only show up in a position adjacent to the verb, that is, to the right of negation. It seems likely that the change in question proceeded via an intermediate stage which exhibited variation between pre-neg and post-neg placement of clitics (the latter being in fact agreement markers). When the change was completed and the former clitics had been fully established as agreement formatives, only the post-neg placement survived. . Similar to Piattino (but in contrast to Non-Standard French), the grammaticalization of gender agreement was made available by the existence of a much more elaborate inventory of number distinctions in the existing agreement paradigm of Vicentino, which by assumption renders the gender node visible for grammaticalization processes (see Section 6.4 above).

JB[v.20020404] Prn:2/08/2005; 15:55

F: LA8106.tex / p.66 (4565-4636)

 The Rise of Agreement . It is a well-known fact that many Northern Italian dialects exhibit a highly syncretic present subjunctive paradigm where most forms are identical to the relevant present indicative forms (cf. e.g. Meyer-Lübke 1890: 225; Meyer-Lübke 1894: 184ff.; Rohlfs 1949: 346f.; Savoia 1997: 84). . Recall that the proclitic a is not distinctive and therefore not eligible for the grammaticalization process in question. . This observation suggests a possible alternative explanation of the change in question according to which the reanalysis of the enclitics satisfied the Blocking Principle by creating a verb form which is specified for clause type, that is, [+interrogative]. Note that this idea is also already present in historical comparative grammars of the Romance languages such as MeyerLübke (1894: 367f.) or Rohlfs (1949: 179f.). See also Haiman and Benincà (1992) and Savoia (1997: 85f.). . However, note that it is possible that grammaticalization processes lead to instances of secondary exponence, where the shape of a given exponent which is the primary exponent of another inflectional feature such as Tense is influenced by the feature content of an agreement morpheme, giving rise to conditioned allomorphy. . A related question arises in connection with the claim (see Chapter 3, Note 62) that only the highest Agr-morpheme is spelled out if a certain head complex contains two Agr-morphemes with identical feature content (Kinyalolo 1991; Carstens 2003): (i)

Morphological Economy In an adjoined structure, Agr on a lower head is inert iff its features are predictable from Agr on a higher head.

Accordingly, a reanalysis such as (52) giving rise to a verb form which exhibits a prefixal agreement marker in addition to the existing suffixal agreement morphology should be possible only if the new higher Agr-morpheme is located in a separate X0 -complex which does not contain the old Agr-morpheme. Thus, I assume that prefixal agreement morphology may also result from a syntactic configuration where the relevant agreement head does not form an X0 -complex with the verb, but is merely linearly adjacent to it (cf. e.g. Embick & Noyer 2001; Julien 2002). . The Swiss RR dialects actually make use of a variety of different morphological means to satisfy the penultimate stress target. For example, apart from causing deletion of the existing agreement suffix/theme vowel of the verb, the attachment of a clitic may also affect the clitic itself, leading to elision of the clitic’s vowel (e.g., 3sg.masc chanta + el → chanta’l). In other contexts (often 3sg.neut or 1pl) a clitic simply cannot be used if this would result in a verb form which violates the penultimate stress target. In Vallader, a semantically empty, stressed suffix -ésch is sometimes added to the verb in addition to the clitic. As a consequence, the word stress shifts from the stem to -ésch, leading to acceptable penultimate stress if a clitic is added (cf. Haiman 1971: 805). In addition, for reasons which are unclear, some dialects such as Puter apparently tolerate violations of the penultimate stress rule in 3pl contexts (cf. Haiman 1971: 807; Linder 1987: 82ff.). The individual dialects differ with respect to the use of these mechanisms. Moreover, the distribution of these mechanisms is influenced by the stress pattern of the verb (which is partially dependent on tense and mood), verb class, shape of clitics and a set of other factors such as interference with other dialects. As a consequence, more often than not, different strategies appear to exist side by side for a single form in a single dialect (see Haiman 1971 and in particular Linder 1987: Ch. 8 for details). . Here, we might speculate that these forms also serve to realize morphosyntactic features associated with the C-head (thereby fulfilling the Blocking Principle). Recall that the new

JB[v.20020404] Prn:2/08/2005; 15:55

F: LA8106.tex / p.67 (4636-4700)

Chapter 6. Morphological blocking and the rise of agreement 

agreement formatives are confined to inversion contexts and cannot show up in embedded clauses. This might be taken to indicate that these inflections also serve to mark the distinction [±subordinate], which is associated with C (the fact that the new endings do not show up in subject-initial clauses follows from the assumption that the verb does not move to C in these contexts, cf. Chapter 5, Section 5.3.4.1). . In the resulting grammar, deletion of the original agreement ending (realizing Agr-on-T) is then the result of a morphological rule which guarantees that of two agreement morphemes (with identical feature content) in a given head complex, only the higher one (here: Agr-on-C) is subject to Vocabulary Insertion (see Chapter 3, Note 62). . The loss of conjugational allomorphy and agreement distinctions, which characterizes the secondary endings, is to be attributed to the fact that these endings do not receive stress and are therefore more likely to undergo phonological reduction than the primary endings. . Some Sutselvan dialects (e.g., the dialect of Flem, cf. Decurtins 1958) show 1sg, 3sg -i instead of -a. Still, what is important is that 1sg and 3sg forms are identical in these dialects as well. . The 2nd person form must carry the specification [–Auth] to prevent that /-as/ is inserted as the realization of the feature matrix [+Auth, +PSE] (‘we-inclusive’). Furthermore, note that the insertion of /-an/ as the realization of 2pl is blocked by the existence of a stronger specified form /-as/, which realizes two inflectional features. . In the case of the 3sg forms, the specification [–pl] is necessary to prevent that the exponents /-al/, /-la/ are inserted in the context of 3pl, where the relevant agreement morpheme also carries gender specification (recall that by assumption the complete set of the subject’s φ-features are copied into the agreement morpheme, cf. Halle 1997). . Note that the (complete) underspecification of the formative /-in/ results from inspecting other parts of the paradigm (e.g., the existence of a form which signals unambiguously 1pl). However, it is a well-known fact that children do not acquire entire paradigms at once. Rather, the individual formatives which constitute a paradigm are acquired one after the other (the exact sequence is subject to cross-linguistic variation, cf. e.g. Clark 1998). Accordingly, determining the exact feature (under-) specification of a given element is a task which can be accomplished only after all forms that constitute a given paradigm have been acquired. In contrast, the Blocking Principle is concerned with a more basic choice which must be decided on earlier during language acquisition. . These changes are reminiscent of the characteristic properties of creoles, a fact also noted by Polinsky (p. 86): “The syntactic redundancy demonstrated for American Russian is paralleled by similar tendencies in extended pidgins and early creoles, which points to a more general correspondence between language loss and pidginization.” . Note, however, that Polinsky (2000: 60) expresses some reservations with respect to the latter criterion, since “American Russian is characterized by aberrant pauses”. . The special status of short, monosyllabic verbs where agreement distinctions are marked solely by stem vowel alternations is presumably another factor which played a role in this development (cf. Section 6.5.2 above): “In this perspective it is not by chance that Kurzverben have been the first context in which subject clitics grammaticalized into verb endings. In fact, this verbal class presents a very reduced word final morphology, as most grammatical information is carried by stem vowel alternation rules (cf. Nübling 1995). Therefore, Walser dialects provided Kurzverben with longer endings, transferring the functional load from the stem to the end of the word.” (Dal Negro 2004: 178f.)

JB[v.20020404] Prn:2/08/2005; 15:55

F: LA8106.tex / p.68 (4700-4781)

 The Rise of Agreement . Presumably, these considerations carry over to the development of new agreement formatives in Cimbrian discussed in Chapter 5, Section 5.2.3.1. . However, Donohue (2002: 7) notes that the requirement for subjects to be overtly realized is dependent on person: “With first or second person clitics this restriction is relaxed somewhat, and sometimes the clitic pronoun alone is sufficient for a grammatical reading (especially with the singular nì and mè).” . According to Donohue, agreement via vowel changes is found only in seven verbs (in combination with a proclitic and sometimes consonant changes): fu ‘to fear’, fue ‘to see’, lu weng ‘to sleep’, lóe ‘to shave’, lóeng ‘to tell’, re ‘to go’, and i ri ‘to fall’. . The factors that condition inflection of the adjunct nominal (instead of the verb root) remain somewhat unclear. Donohue (2002: 11) observes that the adjunct nominal remains unchanged in cases where the proclitic intervenes between the adjunct nominal and the verb stem (in contrast to lú weng ‘to sleep’, where the proclitic attaches to the left of the whole complex verb), as in pìng lú ‘to shoot at someone’: (i)

Nì pe pìng nì=lú. I 3sg.fem arrow 1sg=release ‘I shot her.’

(ii) Pe nì pìng pe=rú. she me arrow 3sg.fem=release.3sg.fem ‘She shot me.’

(*Nì pe nì=pìng lú.)

(*Pe nì pe=pìng lú, *Pe nì pe=pìng rú)

. The availability of consonant alternation as a means of agreement marking is apparently restricted by the onset of the verb root: it is found only with verb roots that begin with the consonants k, h, w, l or with the vowels a, o, oe. . It has already been noted that agreement marking via changes affecting the onset or stem vowel is always accompanied by agreement marking via a proclitic. Some verbs display even more redundant agreement patterns, where subject-verb agreement is marked by up to three different morphological means. For example, there are verbs such as lóe ‘shave’ and lóeng ‘tell, order, promise, persuade’, which display a combination of vowel and consonant alternations (plus the obligatory proclitic, cf. Donohue 2002 for details). This gives rise to sentences such as (i), where subject-verb agreement is realized in three different ways on the verb (marked by boldface): (i)

Pe yu-pe-pè=pe she brother-dat.3sg.fem-poss.3sg.fem=dat.3sg.fem ta-ké=ke pe=r-úe. hair-poss.3sg.masc=dat.3sg.masc 3sg.fem=3sg.fem-shave.3sg.fem ‘She shaved her brother’s hair.’ (Donohue 2002: 14)

Donohue shows that the different agreement patterns (i.e., proclitic, proclitic+consonant prefix etc.) found in Skou cannot be correlated with different verb classes (such as e.g. transitive vs intransitive verbs) or phonological factors. Apparently, then, there are no systematic synchronic factors/criteria that determine the way a given verb inflects for agreement (apart from the fact that the presence of consonant prefixes is limited to a certain set of onsets, see Note 67 above). . Cf. Donohue (2002: 30): “[...] verbs [in Skou] can begin with a larger range of consonants than k h w l a o or oe [...] – additionally, p j m f h y are found, but these verbs do not inflect by prefix. Historically it is likely that these forms showed inflection, just as in the Eastern languages.

JB[v.20020404] Prn:2/08/2005; 15:55

F: LA8106.tex / p.69 (4781-4787)

Chapter 6. Morphological blocking and the rise of agreement 

Extreme simplification in Skou resulted in the complete loss of the inflectional system with these verbs; there is no longer any evidence in the paradigm for any inflection. Taking these verbs into account, assuming seven distinct paradigm sets as in the majority of the Eastern languages, we arrive at a figure of 79% for loss of contrast in verbal paradigms.”

JB[v.20020404] Prn:2/08/2005; 15:57

F: LA8107.tex / p.1 (44-121)

Chapter 7

Concluding summary

This study has investigated the historical paths leading from pronominal clitics to markers of verbal agreement, focusing on the rise of subject-verb agreement. I have argued that the grammaticalization process in question cannot be reduced to a single scenario – such as the reanalysis of resumptive clitics in NP-detachment constructions – which proceeds uniformly across languages (in contrast to claims widely held in the literature). Instead, I have tried to show that the reanalysis of clitic pronouns as agreement markers is shaped by a variety of different but interwoven syntactic and morphological factors which derive from the way syntactic agreement is established in natural languages and the way language change is anchored in the workings of language acquisition. Accordingly, the change in question can be triggered in a multitude of historical scenarios which still share a set of clearly identifiable common properties (such as structural simplification, adjacency between the finite verb and the pronoun, and the presence of an alternative carrier of the θ-role previously assigned to the pronoun). These underlying similarities have been shown to result from the universal nature of the restrictions that the reanalysis of clitics is subject to. A subset of the relevant syntactic contexts has been explored in more detail, focusing on historical developments affecting C-oriented clitics which either attach to the finite verb in V2 languages or occupy clausal second position in languages such as Uto-Aztecan or Mongolian. A central claim put forward in this book is that the grammaticalization of agreement markers from pronominal elements is shaped by a certain division of labor between syntax and morphology in which the workings of the syntactic component define a set of necessary conditions which a reanalysis must satisfy, while the ultimate trigger for the change in question is morphological in nature. More specifically, I have argued that the reanalysis of pronominal clitics (and the acquisition of inflectional morphology in general) is governed by morphological blocking effects which trigger the grammaticalization of new inflectional markers in contexts where the existing morphology is defective, that is, signals fewer distinctions than the potential new formatives. In the following, the content of the individual chapters is briefly recapitulated, highlighting the thread of the individual arguments put forward in this study. In the introductory Chapter 1, I introduced the empirical phenomenon under consideration and gave an overview of general characteristics of the transition from pronominal elements to markers of verbal agreement. Based on a critical discussion of previous analyses, I highlighted a set of descriptive and explanatory issues which an adequate theoretical account of the grammaticalization of agreement markers should

JB[v.20020404] Prn:2/08/2005; 15:57

F: LA8107.tex / p.2 (121-182)

 The Rise of Agreement

address, including the pioneering role of 1st and 2nd person in the development of agreement markers and the fact that cross-linguistically, verbal agreement is preferably realized by suffixes. In Chapter 2, I introduced the basic theoretical assumptions advocated in this book, including an outline of minimalist syntax (i.e., the probe-goal mechanism) and key properties of Distributed Morphology. In addition, I provided an overview of generative approaches to language change. With respect to the latter, I focused on the phenomenon of grammaticalization, claiming that this type of change is to be analyzed as a historical process in which a phonological exponent formerly realizing a lexical head (or l-morpheme) turns into the realization of a functional head (or f morpheme). It was demonstrated that this model provides a unified explanation for a number of surface characteristics of grammaticalization phenomena, attributing the latter to universal properties of functional categories. In this context, however, I already hinted at the special status of agreement, calling into doubt whether agreement should be analyzed on a par with categories such as T or D, which are commonly assumed to belong to a universally present inventory of functional heads. The peculiar properties of agreement were explored in some more detail in Chapter 3, where I developed a synchronic analysis of predicate-argument agreement. The central claims of this approach can be summarized as follows. I argued that agreement morphemes do not head their own projection in the syntax. Rather, they are parasitic on other contentful functional heads such as C, T, or ν, with which they can combine in two different ways. First, the agreement morpheme may attach to its functional host prior to the insertion of that host into the syntactic derivation (‘canonical’ predicate-argument agreement). Second, the agreement morpheme may be added post-syntactically as a dissociated Agr-morpheme at the level of Morphological Structure. I claimed that Agr-morphemes present in the syntactic derivation initiate an Agree-operation, accessing the closest interpretable φ-set to value their feature content and that this operation may in principle be independent from Case licensing, contra Chomsky (2000). In contrast, dissociated Agr-morphemes are parasitic on an agreement relation established in the syntax. That is, they are licensed under structural adjacency with an Agr-morpheme that has been licensed in the syntax. As a consequence, the insertion of dissociated morphemes typically gives rise to instances of multiple agreement where the argument’s φ-features are reflected in more than one location across the clause. It has been shown that this approach makes available a new analysis of complementizer agreement in Germanic which avoids the shortcomings of previous analyses and successfully captures the ‘PF-flavor’ of this phenomenon (adjacency effects, sensitivity to elision of the finite verb). In addition, I have taken up the question of how agreement morphemes are attached to the verb stem, defending the traditional idea that affix order is determined by lexical properties of individual Vocabulary items (i.e., the prefix/suffix distinction), in addition to structural restrictions imposed by the hierarchical relations created in the syntax. Here, I argued that the effects of the Mirror Principle can be derived if it is assumed that head adjunction structures are processed in a bottom-up fashion during Vocabulary Insertion, creating

JB[v.20020404] Prn:2/08/2005; 15:57

F: LA8107.tex / p.3 (182-250)

Chapter 7. Concluding summary 

the linear ordering of morphemes. Furthermore, I pointed out that the workings of the mapping procedure can also account for the fact that agreement markers exhibit more ordering possibilities than other inflectional affixes. This explanation crucially relies on the special phrase-structural status of agreement morphemes as adjuncts to other inflectional heads, introducing a point of variation in the linearization of head complexes. As a consequence, an Agr-morpheme adjoined to T may give rise to either V+T+Agr or V+Agr+T in individual languages. The diachronic implications of this approach to predicate-argument agreement were explored in greater detail in Chapters 4–6. In Chapter 4, I demonstrated that the specific analysis advocated in the previous chapter makes certain predictions with respect to the syntactic environments where a reanalysis of clitic pronouns as exponents of agreement morphemes is possible. In addition, I introduced a set of criteria for distinguishing between pronominal elements and markers of verbal agreement, arguing that these criteria can also be conceived of as cues which determine the acquisition of agreement markers during language acquisition. On these assumptions, I developed a set of necessary requirements that the grammaticalization of agreement markers must satisfy. More specifically, I argued that the change in question is only possible if (i) it preserves the predicate’s argument structure, (ii) provides a means to value the feature content of the resulting agreement morpheme, (iii) warrants that (bound) agreement morphemes combine with the verb prior to Vocabulary Insertion, and (iv) satisfies economy considerations which require that the resulting structure is (derivationally) less complex than the target structure. Based on these assumptions, I presented a (presumably non-exhaustive) set of syntactic scenarios in which the reanalysis of clitic pronouns as agreement markers is licensed. In addition, I claimed that the suffixing preference should receive a diachronic explanation, arguing that the grammaticalization process in question is more likely to be triggered in syntactic environments which give rise to suffixal agreement morphology. In Chapter 5, I took a closer look at a subset of the scenarios outlined in Chapter 4, focusing on syntactic aspects of the reanalysis of C-oriented clitics in languages with (Bavarian, Cimbrian, Walser German, Rhaeto-Romance) and without the V2 property (Uto-Aztecan and Mongolian). I demonstrated that in the former, the V2 property played a key role in this change by creating adjacency between subject clitics and the finite verb. In addition, I showed that we still must distinguish between two different historical scenarios where new agreement markers may arise in V2 languages. First, in Bavarian, Walser German, Cimbrian and older stages of Rhaeto-Romance, the conversion of subject clitics into agreement markers led to structures where the subject’s θ-role is assigned to a referential pro, either exploiting pro-drop properties already available in the target grammar or creating new instances of pro-drop. Second, I demonstrated that in some Swiss Rhaeto-Romance dialects (in particular Sutselvan), the reanalysis targets subject clitics in clitic doubling structures, resulting in a configuration where the relevant θ-role is assigned to the former double (which was originally added for reasons of emphasis). I further argued that in both scenarios, the creation of new agreement markers proceeded via an initial stage where a C-oriented clitic is

JB[v.20020404] Prn:2/08/2005; 15:57

F: LA8107.tex / p.4 (250-290)

 The Rise of Agreement

reanalyzed as the realization of a dissociated agreement morpheme on C, giving rise to instances of multiple agreement (still reflected by complementizer agreement in present-day Bavarian). In a subsequent development, the exponents of Agr-on-C may then replace the older exponents of ‘canonical’ subject-verb agreement (associated with Agr-on-T). The theoretical insights gained from the discussion of V2 languages can be summarized as follows. First, there is a systematic link between the pro-drop property and the rise of new agreement markers, which can be taken to suggest that historically, the former is a by-product of the latter. Second, the observation that exponents of a higher functional morpheme (Agr-on-C) may turn into realizations of a lower functional morpheme (Agr-on-T) contradicts the claim that grammaticalization always leads to new exponents of higher functional heads (cf. Roberts & Roussou 2003). Third, at least in languages which already exhibited verbal agreement morphology prior to the reanalysis in question, the grammaticalization of new agreement formatives proceeds via an initial stage where a clitic is reanalyzed as a dissociated Agr-morpheme which is parasitic on another Agr-morpheme that is valued in the syntactic derivation. Accordingly, the grammaticalization path in question can be depicted as in (1). (1) free pronoun → weak pronoun → clitic pronoun → (dissociated Agr-morpheme) → syntactic Agr-morpheme → Ø In the remainder of Chapter 5, I discussed two examples in which C-oriented clitics were ‘directly’ reanalyzed as realizations of a syntactic Agr-morpheme (i.e., Agr-onT) in languages which previously lacked agreement markers. First, I showed that a set of Uto-Aztecan languages exhibits various stages of a historical process where second position clitics (in doubling configurations) turn into prefixal agreement markers in a set of SVO languages. Second, I discussed an alternative historical path which may lead to suffixal agreement morphology in ‘true’ SOV languages, arguing that in a set of Mongolian languages, C-oriented clitics were reanalyzed as suffixal agreement markers in a structural configuration where the TP is fronted into the specifier of CP, creating the necessary adjacency between the clitic (adjoined to C) and the finite verb which occurs at the right end of the fronted TP. In Chapter 6, I focused on morphological factors which shape the grammaticalization of agreement markers, starting out from the observation that across languages, the grammaticalization of new agreement formatives does not proceed in a random fashion, but rather affects only contexts where the existing verbal morphology is nondistinctive. I claimed that this observation can be explained if it is assumed that the acquisition of inflectional morphology is guided by blocking effects which prefer new verbal agreement formatives to be more specific than the existing morphology. Framed in a realizational model of grammar (i.e., Distributed Morphology), this idea is expressed by the so-called Blocking Principle: (2) Blocking Principle If several appropriate PF-realizations of a given morpheme are attested in the

JB[v.20020404] Prn:2/08/2005; 15:57

F: LA8107.tex / p.5 (290-358)

Chapter 7. Concluding summary 

Primary Linguistic Data, the form matching the greatest subset of the morphosyntactic features included in the morpheme must be chosen for storage in the lexicon. By assumption, the Blocking Principle operates during language acquisition as an economy principle which warrants an optimal and non-redundant Lexicon (and paradigm) structure. Similar to structural economy principles which guide the acquisition of syntactic properties (cf. Clark & Roberts 1993), the Blocking Principle is invoked only if the trigger experience is not sufficient for determining which Vocabulary item realizes a certain inflectional head, as in cases where the learner is confronted both with an existing, underspecified exponent of a certain Agr-morpheme and a pronominal clitic which is stronger specified and fulfills all necessary conditions for being reanalyzed as an agreement formative. In turn, I demonstrated that the workings of the Blocking Principle can be detected in the development of new agreement formatives in languages such as Bavarian, Non-Standard French, Northern Italian dialects, Rhaeto-Romance and the Papuan language Skou. Furthermore, in the case of Bavarian, I argued that the peculiar person/number restrictions on pro-drop and complementizer agreement observed in Chapter 5 should receive a diachronic explanation, in the sense that these restrictions reflect the contexts where the grammaticalization of new forms was licensed by the Blocking Principle. In many instances, we observed that the reanalysis initially affects clitics only in a restricted context in which the relevant existing verb forms are defective (e.g., a certain verb class, tense or mood) before the change gains a wider distribution. Here, it was argued that only the initial reanalysis which gives rise to a new agreement formative proceeds in accordance with the Blocking Principle, whereas the gradual expansion to other contexts was attributed to the workings of analogical change. In addition, I argued that the analysis presented in this chapter is further supported by the fact that the Blocking Principle provides a new explanation for the pioneering role of 1st and 2nd person in the development of predicate-argument agreement across languages (assuming that 3rd person forms, including potential new agreement formatives, lack person features and are therefore inherently underspecified). As noted repeatedly throughout this work, the present study focused only on a small portion of the vast array of diachronic phenomena which involve the creation of new agreement markers. At least in part, the decision to concentrate on two major issues (syntactic aspects of the reanalysis of C-oriented subject clitics and the workings of morphological blocking in the rise of agreement) was governed simply by practical considerations. An in-depth treatment of the complete set of grammaticalization processes leading to new agreement formatives is a task simply too ambitious to be accomplished within a single monograph. So I decided to remain silent about other major topics such as the grammaticalization of object or wh-agreement (cf. e.g. Chung 1998 on the latter), the rise of DP-internal (head-modifier) agreement (cf. e.g. Lehmann 2002), the development of ergative/absolutive agreement (cf. Anderson 1977, 1980; Givón 1984), alternative paths to agreement which do not involve

JB[v.20020404] Prn:2/08/2005; 15:57

F: LA8107.tex / p.6 (358-381)

 The Rise of Agreement

pronominal elements (such as the reinterpretation of other pieces of inflection, cf. e.g. Chafe 1977 on the 3rd person subject prefixes of Iroquoian, see Siewierska 2004 for an overview), or the connection between the rise of agreement and other syntactic changes, each of which justifies a monograph in itself.1 Still, I reckon that the analysis developed in the previous chapters is not limited to the data set discussed in this book and can be extended to at least some of the above mentioned phenomena. The factors and principles proposed in this work such as the Blocking Principle, Preservation of argument structure or Structural simplification are not bound to a single surface phenomenon (such as the reanalysis of subject clitics). Rather, they derive from (and model) deeper properties of grammar and are therefore expected to carry over to other changes creating new inflectional markers across languages. In this way, I hope that the insights gained in this work add to our understanding of the processes which underlie the rise of agreement and the phenomenon of grammaticalization in general.

Note . The latter includes such intriguing topics as the link between the rise of agreement and the development of free word order (cf. e.g. Mithun 1991), the development of verb movement (Alexiadou & Fanselow 2001, 2002) or changes affecting the syntactic alignment (nominative/accusative vs. absolutive/ergative) of a language. For example, it appears that in a number of ergative/absolutive languages, the development of person agreement on the verb affects the overall organization of grammar, giving rise to nominative/accusative patterns (cf. e.g. Harris 1994; Schulze 1998 on the Caucasian family, Cysouw 2003b on a set of Austronesian languages spoken in Sulawesi). Thus, the new person markers do not follow the ergative/absolutive pattern of the existing agreement markers (such as noun class agreement marking in the Caucasian languages). Instead, they always signal agreement with the subject DP of both intransitives and transitives.

JB[v.20020404] Prn:8/09/2005; 13:19

F: LA81RE.tex / p.1 (44-168)

References

Abney, Steven (1987). The English Noun Phrase in its Sentential Aspect. Doctoral dissertation, MIT. Abraham, Werner (1995). Deutsche Syntax im Sprachenvergleich. Tübingen: Gunter Narr. Abraham, Werner (2003). Autonomous and non-autonomous components of ‘grammaticalization’: Economy criteria in the emergence of German negation. Sprachtypologie und Universalienforschung, 56 (4), 325–365. Abraham, Werner (2004). The grammaticalization of the infinitival preposition – toward a theory of ‘grammaticalizing reanalysis’. Journal of Comparative Germanic Linguistics, 7, 111–170. Ackema, Peter, & Ad Neeleman (2003). Context-sensitive Spell-out. Natural Language and Linguistic Theory, 21, 681–735. Ackema, Peter, & Ad Neeleman (2004). Beyond Morphology. Interface Conditions on Word Formation. Oxford: Oxford University Press. Albright, Adam C. (2002). The Identification of Bases in Morphological Paradigms. Doctoral dissertation, UCLA. Alexiadou, Artemis, & Elena Anagnostopoulou (1998). Parametrizing AGR: Word order, Vmovement and EPP-checking. Natural Language and Linguistic Theory, 16, 491–539. Alexiadou, Artemis, & Gisbert Fanselow (2001). Laws of diachrony as a source for syntactic generalisations: The case of V to I. GLOW Newsletter, 46, 57–58. Alexiadou, Artemis, & Gisbert Fanselow (2002). On the correlation between morphology and syntax: The case of V-to-I. In J.-W. Zwart & W. Abraham (Eds.), Studies in Comparative Germanic Syntax (pp. 219–242). Amsterdam: John Benjamins. Allen, Cynthia (1995). Case Marking and Reanalysis: Grammatical Relations from Old to Early Modern English. Oxford: Clarendon. Allen, Cynthia (1997). The origins of the ‘group genitive’ in English. Transactions of the Philological Society, 95 (1), 111–131. Altmann, Hans (1984). Das System der enklitischen Personalpronomina in einer mittelbairischen Mundart. Zeitschrift für Dialektologie und Linguistik, 51 (2), 191–211. Andersen, Torben (1988). Ergativity in Pari, a Nilotic OVS language. Lingua, 75, 289–324. Anderson, Stephen R. (1977). On mechanisms by which languages become ergative. In C. N. Li (Ed.), Mechanisms of Syntactic Change (pp. 317–363). Austin: University of Texas Press. Anderson, Stephen R. (1980). On the development of morphology from syntax. In J. Fisiak (Ed.), Historical Morphology (pp. 51–69). Berlin: Mouton de Gruyter. Anderson, Stephen R. (1986). Disjunctive ordering in inflectional morphology. Natural Language and Linguistic Theory, 4, 1–31. Anderson, Stephen R. (1992). A-morphous Morphology. Cambridge: Cambridge University Press.

JB[v.20020404] Prn:8/09/2005; 13:19

F: LA81RE.tex / p.2 (168-291)

 The Rise of Agreement

Ariel, Mira (1990). Accessing NP Antecedents. London: Routledge [Croom Helm Linguistics Series]. Ariel, Mira (1998). Three grammaticalization paths for the development of person verbal agreement in Hebrew. In J.-P. Koenig (Ed.), Discourse and Cognition: Bridging the Gap (pp. 93–111). Stanford, CSLI. Ariel, Mira (2000). The development of person agreement markers: From pronoun to higher accessibility markers. In S. Kemmer & M. Barlow (Eds.), Usage-based Models of Language (pp. 197–261). Stanford, CSLI. Aronoff, Mark (1976). Word Formation in Generative Grammar. Cambridge, MA: The MIT Press. Arregi, Karlos (1999). Person and number inflection in Basque. In K. Arregi, B. Bruening, C. Krause, & V. Lin (Eds.), MIT Working Papers in Linguistics 33: Papers on Morphology and Syntax, Cycle One (pp. 229–264). Cambridge, MA: Department of Linguistics and Philosophy, MIT. Ashby, William (1977). Clitic Inflection in French. An Historical Perspective. Amsterdam: Rodopi. Auger, Julie (1993). More evidence for verbal agreement marking in Colloquial French. In W. J. Ashby, M. Mithun, G. Perissinotto, & E. Raposo (Eds.), Linguistic Perspectives on the Romance Languages (pp. 177–98). Amsterdam: John Benjamins. Auger, Julie (1994). Pronominal Clitics in Quebec Colloquial French: A Morphological Analysis. Doctoral dissertation, University of Pennsylvania. Baker, Mark C. (1985). The Mirror Principle and morphosyntactic explanation. Linguistic Inquiry, 16 (3), 373–415. Baker, Mark C. (1988). Incorporation. A Theory of Grammatical Function Changing. Chicago: Chicago University Press. Baker, Mark C. (1996). The Polysynthesis Parameter. Oxford: Oxford University Press. Baker, Mark C. (2002). Building and merging, not checking: The non-existence of (Aux)-S-V-O languages. Linguistic Inquiry, 33 (2), 321–328. Baker, Mark. C., & Jonathan Bobaljik (2002). Introduction to Morphology. Ms., Rutgers and McGill. Barlow, Michael (1992). A Situated Theory of Agreement. New York, NY: Garland. Battye, Adrian, & Roberts, Ian (Eds.). (1995). Clause Structure and Language Change. Oxford: Oxford University Press. Bayer, Josef (1984). COMP in Bavarian syntax. The Linguistic Review, 3, 209–274. Beard, Robert (1988). The separation of derivation and affixation: Toward a Lexeme-Morpheme Base Morphology. Quarderni di semantica, 5, 277–287. Beard, Robert (1995). Lexeme-Morpheme Base Morphology. A General Theory of Inflection and Word Formation. Albany NY: SUNY Press. Bech, Gunnar (1955). Studien zum deutschen Verbum infinitum. Tübingen: Niemeyer. Beckman, Jill N. (Ed.). (1995). Proceedings of NELS 25. Vol. 2: Papers from the Workshops on Language Acquisition and Language Change. University of Pennsylvania. Bejar, Susana (2003). Phi-syntax: A Theory of Agreement. Doctoral dissertation, University of Toronto. Belletti, Adriana (1990). Generalized Verb Movement: Aspects of Verb Syntax. Turin: Rosenberg and Sellier. Belletti, Adriana (2001). Agreement projections. In M. Baltin & C. Collins (Eds.), The Handbook of Contemporary Syntactic Theory (pp. 483–510). Oxford: Blackwell.

JB[v.20020404] Prn:8/09/2005; 13:19

F: LA81RE.tex / p.3 (291-404)

References 

Bennis, Hans, & Liliane Haegeman (1984). On the status of agreement and relative clauses in West Flemish. In W. de Geest & Y. Putseys (Eds.), Sentential complementation (pp. 33–55). Dordecht: Foris. Benveniste, Emile (1968). Mutations of Linguistic Categories. In W. Lehmann & Y. Malkiel (Eds.), Directions for Historical Linguistics: A Symposium (pp. 83–94). Austin: University of Texas Press. Benveniste, Emile (1972). Problèmes de linguistique générale. Paris: Editions Gallimard. Berwick, Robert C. (1985). The Acquisition of Syntactic Knowledge. Cambridge, MA: The MIT Press. Besten, Hans den (1982). On the interaction of root transformations and lexical deletive rules. In W. Abraham (Ed.), On the Formal Syntax of the Westgermania (pp. 47–132). Amsterdam: John Benjamins. Bickerton, Derek (1984). The language bioprogram hypothesis. Behavioral and Brain Sciences, 7, 173–221. Bickerton, Derek (1998). How to acquire language without positive evidence: What acquisitionists can learn from creoles. In M. DeGraff (Ed.), Language Creation and Language Change: Creolization, Diachrony and Development (pp. 49–74). Cambridge, MA: The MIT Press. Bittner, Andreas (2004). Wenn Spracherwerb Sprachwandel wär. Paper presented at the University of Frankfurt, January 14th, 2004. Blood, Cindy (1992). Subject-verb agreement in Kisar. NUSA. Linguistics Studies of Indonesia and Other Languages in Indonesia, Vol. 34 (pp. 1–21). Jakarta: Badan Penyelenggara NUSA. Bobaljik, Jonathan (1994). What does adjacency do? In H. Harley & C. Phillips (Eds.), MIT Working Papers in Linguistics 22: The Morphology-Syntax Connection (pp. 1–32). Cambridge, MA: Department of Linguistics and Philosophy, MIT. Bobaljik, Jonathan (1995). Morphosyntax: The Syntax of Verbal Inflection. Doctoral dissertation, MIT. Bobaljik, Jonathan (2002). A-chains at the PF interface. Natural Language and Linguistic Theory, 20, 197–267. Bobaljik, Jonathan (2003). Realizing Germanic inflection: Why morphology does not drive syntax. Journal of Comparative Germanic Linguistics, 6 (2), 129–167. Bobaljik, Jonathan (2005). Where’s φ? Agreement as a postsyntactic operation. Ms., University of Connecticut, Storrs. Bobaljik, Jonathan, & Samuel Brown (1997). Interarboreal operations: Head movement and the Extension Requirement. Linguistic Inquiry, 28 (2), 345–56. Bobaljik, Jonathan, & Diane Jonas (1996). Subject positions and the role of TP. Linguistic Inquiry, 27 (2), 195–236. Bobaljik, Jonathan, & H. Thráinsson (1998). Two heads aren’t always better than one. Syntax, 1, 37–71. Bobaljik, Jonathan, & Susi Wurmbrand (2004). The domain of agreement. To appear in Natural Language and Linguistic Theory. Bonet, Eulalia (1991). Morphology after Syntax. Pronominal Clitics in Romance. Doctoral dissertation, MIT. Bopp, Franz (1816). Über das Conjugationssystem der Sanskritsprache in Vergleichung mit jenem der griechischen, lateinischen, persischen und germanischen Sprache. Frankfurt am Main: Andreäische Buchhandlung. Borer, Hagit (1984). Parametric Syntax. Dordrecht: Foris.

JB[v.20020404] Prn:8/09/2005; 13:19

F: LA81RE.tex / p.4 (404-534)

 The Rise of Agreement

Borer, Hagit (1998). Morphology and Syntax. In A. Spencer & A. Zwicky (Eds.), The Handbook of Morphology (pp. 151–189). Oxford: Blackwell. Brandi, Luciana, & Patrizia Cordin (1989). Two Italian dialects and the null subject parameter. In O. Jaeggli & K. Safir (Eds.), The Null Subject Parameter (pp. 111–142). Dordrecht, Kluwer. Branigan, Phil, & Marguerite MacKenzie (2002). Altruism, A-movement and object agreement in Innu-aimûn. Linguistic Inquiry, 33, 385–407. Braune, Wilhelm (1950). Althochdeutsche Grammatik (7th edition). Halle: Niemeyer. Bresnan, Joan (1973). The syntax of the comparative clause construction in English. Linguistic Inquiry, 4, 275–343. Bresnan Joan, & Sam A. Mchombo (1987). Topic, pronoun, and agreement in Chichewa. Language, 63, 741–782. Brinkmann, Hennig (1931). Sprachwandel und Sprachbewegung in althochdeutscher Zeit. Jena. Brugmann, Karl (1916). Vergleichende Laut-, Stammbildungs- und Flexionslehre der indogermanischen Sprachen (2nd edition). Vol. 2/3: Lehre von den Wortformen und ihrem Gebrauch: Verbum finitum. Strassburg: Trübner. Bruening, Benjamin (2001). Syntax at the Edge: Cross-clausal Phenomena and the Syntax of Passamaquoddy. Doctoral dissertation, MIT. Burenhult, Niclas (2002). A Grammar of Jahai. Lund: Department of Linguistics and Phonetics, Lund University. Bybee, Joan L. (1985). Morphology. Amsterdam: John Benjamins. Bybee, Joan L., William Pagliuca, & Revere Perkins (1990). On the asymmetries in the affixation of grammatical material. In W. Croft, K. Denning & S. Kemmer (Eds.), Studies in Typology and Diachrony (pp. 1–42). Amsterdam: John Benjamins. Bybee, Joan L., & Dan I. Slobin (1982). Rules and schemas in the development and use of the English past tense. Language, 58, 265–289. Campbell, Lyle (1991). Some grammaticalization changes in Estonian. In E. C. Traugott & B. Heine (Eds.), Approaches to Grammaticalization, Vol. I: Theoretical and Methodological Issues (pp. 285–99). Amsterdam: John Benjamins. Cardinaletti, Anna, & Michal Starke (1999). The typology of structural deficiency: A case study of the three classes of pronouns. In H. van Riemsdijk (Ed.), Clitics in the Languages of Europe (pp. 145–232). Berlin: Mouton de Gruyter. Carstens, Vicki (2003). Rethinking complementizer agreement: Agree with a Case-checked goal. Linguistic Inquiry, 34 (3), 393–412. Cecchetto, Carlo (2000). Doubling structures and reconstruction. Probus, 12, 93–126. Chafe, Wallace L. (1977). The evolution of third person verb agreement in the Iroquoian Languages. In C. N. Li (Ed.), Mechanisms of Syntactic Change (pp. 493–524). Austin: University of Texas Press. Chomsky, Noam (1957). Syntactic Structures. The Hague: Mouton. Chomsky, Noam (1965). Aspects of the Theory of Syntax. Cambridge, MA: The MIT Press. Chomsky, Noam (1981). Lectures on Government and Binding. Dordrecht: Foris. Chomsky, Noam (1982). Some Concepts and Consequences of the Theory of Government and Binding. Cambridge, MA: The MIT Press. Chomsky, Noam (1986a). Knowledge of Language: Its Nature, Origin and Use. New York, NY: Praeger. Chomsky, Noam (1986b). Barriers. Cambridge, MA: The MIT Press. Chomsky, Noam (1991). Some notes on economy of derivations and representations. In R. Freidin (Ed.), Principles and Parameters in Comparative Grammar (pp. 417–454). Cambridge, MA: The MIT Press.

JB[v.20020404] Prn:8/09/2005; 13:19

F: LA81RE.tex / p.5 (534-672)

References 

Chomsky, Noam (1993). A minimalist program for linguistic theory. In K. Hale & S. J. Keyser (Eds.), The View from Building 20: Essays in Linguistics in Honor of Sylvain Bromberger (pp. 1–52). Cambridge, MA: The MIT Press. Chomsky, Noam (1995). The Minimalist Program. Cambridge, MA: The MIT Press. Chomsky, Noam (2000). Minimalist inquiries: The framework. In R. Martin, D. Michaels, & J. Uriagereka (Eds.), Step by Step (pp. 89–155). Cambridge, MA: The MIT Press. Chomsky, Noam (2001a). Derivation by phase. In M. Kenstowicz (Ed.), Ken Hale. A Life in Language (pp. 1–52). Cambridge, MA: The MIT Press. Chomsky, Noam (2001b). Beyond explanatory adequacy. MIT Occasional Papers in Linguistics, 20, 1–28. Cambridge, MA: Department of Linguistics and Philosophy, MIT. Chomsky, Noam (2002). On Nature and Language. Cambridge: Cambridge University Press. Chomsky, Noam (2005). On phases. Ms., MIT. Chomsky, Noam, & Howard Lasnik (1993). Principles and Parameters theory. In J. Jacobs, A. von Stechow, W. Sternefeld, & T. Vennemann (Eds.), Syntax: An International Handbook of Contemporary Research (pp. 496–569). Berlin: Walter de Gruyter. Chung, Sandra (1998). The Design of Agreement. Evidence from Chamorro. Chicago: Chicago University Press. Cinque, Guglielmo (1999). Adverbs and Functional Heads: A Cross-linguistic Perspective. Oxford: Oxford University Press. Clark, Eve V. (1998). The acquisition of morphology. In Andrew J. Spencer & Arnold M. Zwicky (Eds.), The Handbook of Morphology (pp. 374–389). Oxford: Blackwell. Clark, Robin, & Ian Roberts (1993). A computational model of language learnability and language change. Linguistic Inquiry, 24, 299–345. Collins, Chris (1997). Local Economy. Cambridge, MA: The MIT Press. Comrie, Bernhard (1978). Ergativity. In W. P. Lehmann (Ed.), Syntactic Typology: Studies in the Phenomenology of Language (pp. 329–394). Austin, TX: University of Texas Press. Comrie, Bernard (1980). Morphology and word order reconstruction: Problems and prospects. In J. Fisiak (Ed.), Historical Morphology (pp. 83–96). The Hague: Mouton. Comrie, Bernard (2003). When agreement gets trigger-happy. Transactions of the Philological Society, 101 (2), 313–337. Corbett, Greville (1991). Gender. Cambridge: Cambridge University Press. Corbett, Greville (1995). Agreement. In J. Jacobs, A. von Stechow, W. Sternefeld, & T. Vennemann (Eds.), Syntax: An International Handbook of Contemporary Research Vol. 2 (pp. 1235–1244). Berlin: Walter de Gruyter. Corbett, Greville (1998). Morphology and agreement. In A. Zwicky & A. Spencer (Eds.), The Handbook of Morphology (pp. 191–205). Oxford: Blackwell. Corbett, Greville (2000). Number. Cambridge: Cambridge University Press. Corbett, Greville (2003). Agreement: The range of the phenomenon and the principles of the Surrey Database of Agreement. Transactions of the Philological Society, 101 (2), 155–202. Craig, Colette Grinevald (1977). The Structure of Jacaltec. Austin, TX: University of Texas Press. Cutler, Anne, John A. Hawkins, & Gary Gilligan (1985). The suffixing preference: A processing explanation. Linguistics, 23, 723–758. Cysouw, Michael (2003a). The Paradigmatic Structure of Person Marking. Oxford: Oxford University Press. Cysouw, Michael (2003b). From pronouns to agreement (and back again). Paper presented at the symposium Grammatikalisierung: Wege von der Performanz zur Kompetenz. FU Berlin, 28/29.11.2003.

JB[v.20020404] Prn:8/09/2005; 13:19

F: LA81RE.tex / p.6 (672-790)

 The Rise of Agreement

Dal Negro, Silvia (2004). The Decay of a Language. The Case of a German Dialect in the Italian Alps. Frankfurt: Peter Lang. Decurtins, Alexi (1958). Zur Morphologie der unregelmäßigen Verben im Bündnerromanischen. Historisch-deskriptive Studie mit besonderer Berücksichtigung des Sur- und Sutselvischen [Romanica Helvetica 62]. Bern: A. Francke. Dimmendaal, Gerrit Jan (1983). The Turkana Language. Dordrecht: Foris. Dixon, R. M. W. (1980). The Languages of Australia. Cambridge: Cambridge University Press. Dixon, R. M. W. (2002). Australian Languages. Their Nature and Development. Cambridge: Cambridge University Press. Donohue, Mark (2002). A most agreeable language. Ms., University of Sydney. Dresher, Elan (1999). Charting the learning path: Cues to parameter setting. Linguistic Inquiry, 30 (1), 27–67. Dresher, Elan, & Jonathan Kaye (1990). A computational learning model for metrical phonology. Cognition, 34, 137–195. Ebneter, Theodor (1994). Syntax des gesprochenen Rätoromanischen. Tübingen: Niemeyer. Ellegård, Alvar (1953). The Auxiliary Do: The Establishment and Regulation of its Use in English. Stockholm: Almquist and Wiksell. Embick, David (1997). Voice and the Interfaces of Syntax. Doctoral dissertation, University of Pennsylvania. Embick, David, & Rolf Noyer (2001). Movement operations after syntax. Linguistic Inquiry, 32(4), 555–595. Emonds, Joseph (1978). The verbal complex V’–V in French. Linguistic Inquiry, 21, 49–77. Falk, Cecilia (1993). Non-referential Subjects in the History of Swedish. Doctoral dissertation, University of Lund. Fintel, Kai von (1995). The formal semantics of grammaticalization. In J. Beckman (Ed.), Proceedings of NELS 25: Papers from the Workshops on Language Acquisition & Language Change (pp. 175–189). Amherst, MA: GLSA. Fodor, Janet (1998). Unambiguous triggers. Linguistic Inquiry, 29 (1), 1–36. Frampton, John (2002). Syncretism, impoverishment and the structure of person features. In M. Andronis, M. Debenport, A. Pycha & K. Yoshimura (Eds.), Proceedings of CLS 38 (pp. 207–222). Chicago: Chicago Linguistic Society. Friedemann, Marc-Ariel (1995). Sujets Syntaxiques: Positions, Inversions et pro. Doctoral dissertation, Université de Genève. Friedmann, Na’ama, & Yosef Grodzinsky (2000). Split inflection in neurolinguistics. In M.A. Friedemann & L. Rizzi (Eds.), The Acquisition of Syntax: Studies in Comparative Developmental Linguistics (pp. 84–104). Geneva: Longman. Fukui, Naoki (1986). A Theory of Category Projection and Its Application. Doctoral dissertation, MIT. Fuß, Eric (2003). On the historical core of V2 in Germanic. Nordic Journal of Linguistics, 26 (2), 195–231. Fuß, Eric, & Carola Trips (2002). Variation and change in Old and Middle English. On the validity of the Double Base Hypothesis. Journal of Comparative Germanic Linguistics, 4, 171–224. Gabelentz, Georg von der (1891). Die Sprachwissenschaft. Ihre Aufgaben, Methoden, und bisherigen Ergebnisse. Leipzig: Weigel. Gartner, Theodor (1883). Rätoromanische Grammatik. Heilbronn: Henninger. Gartner, Theodor (1910). Handbuch der rätoromanischen Sprache und Literatur. Halle: Niemeyer.

JB[v.20020404] Prn:8/09/2005; 13:19

F: LA81RE.tex / p.7 (790-924)

References

Gelderen, Elly van (1993). The Rise of Functional Categories. Amsterdam: John Benjamins. Gelderen, Elly van (2001). Modal particles in Germanic. Folia Linguistica Historica, 22 (1–2), 310–33. Gelderen, Elly van (2004a). Economy, innovation, and prescriptivism: From spec to head and head to head. Journal of Comparative Germanic Linguistics, 7, 59–98. Gelderen, Elly van (2004b). Grammaticalization as Economy. Amsterdam: John Benjamins. Gerlach, Birgit (2001). Optimality aspects of clitic doubling in Romance. In P. de Carvalho, L. Labrune, F. Lambert, CL. Muller, & K. Ploog (Eds.), Clitiques et cliticisation (pp. 343– 363). Paris: Champion. Gerlach, Birgit (2002). Clitics Between Syntax and Lexicon. Amsterdam: John Benjamins. Gibson, Edward, & Kenneth Wexler (1994). Triggers. Linguistic Inquiry, 25 (3), 407–454. Givón, Talmy (1971). Historical syntax and synchronic morphology. An archeologist’s field trip. In Papers from the Seventh Regional Meeting Chicago Linguistic Society (pp. 394–415). Chicago, IL: Chicago Linguistic Society. Givón, Talmy (1976). Topic, Pronoun, and Grammatical Agreement. In C. N. Li (Ed.), Subject and Topic (pp. 149–188). New York, NY: Academic Press. Givón, Talmy (1984). Syntax: A Functional-Typological Introduction, Vol. I. Amsterdam: John Benjamins. Greenberg, Joseph (1963). Essays in Linguistics. Chicago: Chicago University Press. Greenberg, Joseph. (Ed.). (1966a). Universals of Language (2nd edition). Cambridge, MA: The MIT Press. Greenberg, Joseph (1966b). Language Universals, with Special Reference to Feature Hierarchies. The Hague: Mouton. Greenberg, Joseph (1978). How does a language aquire gender markers? In J. Greenberg, C. A. Ferguson, & E. Moravcsik (Eds.), Universals of Human Language Vol. II (pp. 47–82). Stanford CA: Stanford University Press. Grewendorf, Günther (1987). Kohärenz und Restrukturierung. Zu verbalen Komplexen im Deutschen. In B. Asbach-Schnitker & J. Roggenhofer (Eds.), Neuere Forschungen zur Wortbildung und Histographie. Festgabe für Herbert Brekle zum 50. Geburtstag (pp. 123– 144). Tübingen: Gunter Narr. Grewendorf, Günther (1989). Ergativity in German. Dordrecht: Foris. Grewendorf, Günther (2002). Minimalistische Syntax. Tübingen: A. Francke. Grewendorf, Günther (2003). Improper remnant movement. Gengo Kenkyo: The Journal of the Linguistic Society of Japan, 123, 47–94. Grewendorf, Günther (2004). The discourse configurationality of scrambling. To appear in J. Sabel & M. Saito (Eds.). The Free Word Order Phenomenon: Its Syntactic Sources and Diversity. Berlin: Mouton de Gruyter. Grimshaw, Jane (1997). The best clitic: Constraint conflict in morphosyntax. In L. Haegeman (Ed.), Elements of grammar (pp. 169–196). Dordrecht: Kluwer. Groat, Erich, & John O’Neil (1996). Spell-Out at the LF-Interface. In W. Abraham, S. D. Epstein, H. Thráinsson, & J.-W. Zwart (Eds.), Minimal Ideas (pp. 113–139). Amsterdam: John Benjamins. Grønbech, Kaare, & John R. Krueger (1955). An Introduction to Classical Mongolian. Wiesbaden: Otto Harrassowitz. Guiraud, Pierre (1968). L’ancien français (3rd edition). Paris: Presses Universitaires de France. de Haan, Germen, & Fred Weerman (1986). Finiteness and verb fronting in Frisian. In H. Haider & M. Prinzhorn (Eds.), Verb Second Phenomena in Germanic Languages (pp. 77–110). Dordrecht: Foris.



JB[v.20020404] Prn:8/09/2005; 13:19



F: LA81RE.tex / p.8 (924-1049)

The Rise of Agreement

Haas, Mary (1977). From auxiliary verb phrase to inflectional suffix. In C. N. Li (Ed.), Mechanisms of Syntactic Change (pp. 525–537). Austin, TX: University of Texas Press. Haeberli, Eric (1999). Features, Categories and the Syntax of A-positions. Synchronic and Diachronic Variation in the Germanic Languages. Doctoral dissertation, University of Geneva. Haeberli, Eric (2002). Features, Categories and the Syntax of A-Positions. Cross-Linguistic Variation in the Germanic Languages. Dordrecht, Kluwer. Haeberli, Eric (2004). Syntactic effects of inflectional morphology and competing grammars. In E. Fuß & C. Trips (Eds.), Diachronic Clues to Synchronic Grammar (pp. 101–130). Amsterdam: John Benjamins. Haegeman, Liliane (1990). Subject pronouns and subject clitics in West Flemish. The Linguistic Review, 7, 333–363. Haegeman, Liliane (1992). Theory and Description in Generative Syntax: A Case Study in WestFlemish. Cambridge: Cambridge University Press. Haider, Hubert (1986). Fehlende Argumente: Vom Passiv zu kohärenten Infinitiven. Linguistische Berichte, 101, 3–33. Haider, Hubert (1993). Deutsche Syntax – generativ. Tübingen: Gunter Narr. Haiman, John (1971). Targets and paradigmatic borrowing in Romantsch. Language, 47 (4), 797–809. Haiman, John (1974). Targets and Syntactic Change. The Hague: Mouton. Haiman, John (1991). From V/2 to subject clitics: Evidence from Northern Italian. In E. Traugott & B. Heine (Eds.), Approaches to Grammaticalization, Vol. 2 (pp. 135–157). Amsterdam: John Benjamins. Haiman, John, & Paola Benincà (1992). The Rhaeto-Romance Languages. London: Routledge. Hale, Kenneth (1973). Person marking in Warlbiri. In S. Anderson & P. Kiparsky (Eds.), A Festschrift for Morris Halle (pp. 308–344). New York, NY: Holt, Rinehart and Winston. Hale, Mark (1996). Theory and method in historical linguistics. Ms., Concordia University, Montreal. Hale, Mark (1998). Diachronic syntax. Syntax, 1 (1), 1–18. Hall, Christopher J. (1988). Integrating diachronic and processing principles in explaining the suffixing preference. In J. A. Hawkins (Ed.), Explaining Language Universals (pp. 321–349). Oxford: Blackwell. Halle, Morris (1990). An approach to morphology. In J. Carter, R.-M. Déchaine, B. Philip, & T. Sherer (Eds.), Proceedings of the North East Linguistics Society 20, Vol. 1 (pp. 150–184). Amherst, MA: GLSA, University of Amherst, Massachusetts. Halle, Morris, & Alec Marantz (1993). Distributed Morphology and the pieces of inflection. In S. J. Keyser & K. Hale (Eds.), The View from Building 20 (pp. 111–176). Cambridge, MA: The MIT Press. Halle, Morris, & Alec Marantz (1994). Some key features of Distributed Morphology. In A. Carnie, H. Harley, & T. Bures (Eds.), MIT Working Papers in Linguistics 21: Papers on Phonology and Morphology (pp. 275–288). Cambridge, MA: Department of Linguistics and Philosophy, MIT. Halle, Morris (1997). Distributed Morphology: Impoverishment and Fission. In B. Bruening, Y. Kang, & M. McGinnis (Eds.), MIT Working Papers in Linguistics 30: PF: Papers At the Interface (pp. 425–450). Cambridge, MA: Department of Linguistics and Philosophy, MIT. Halpern, Aaron (1992). Topics in the Placement and Morphology of Clitics. Doctoral dissertation, Stanford University.

JB[v.20020404] Prn:8/09/2005; 13:19

F: LA81RE.tex / p.9 (1049-1168)

References

Harley, Heidi (1994). Hug a tree: Deriving the morphosyntactic feature hierarchy. In A. Carnie & H. Harley (Eds.), MIT Working Papers in Linguistics 21: Papers on phonology and morphology (pp. 275–288). Cambridge, MA: Department of Linguistics and Philosophy, MIT. Harley, Heidi, & Rolf Noyer (1999). State-of-the-article: Distributed Morphology. Glot International, 4 (4), 3–9. Harley, Heidi, & Elizabeth Ritter (2002). Structuring the bundle: A universal morphosyntactic feature geometry. In H. J. Simon & H. Wiese (Eds.), Pronouns – Grammar and Representation (pp. 23–39). Amsterdam: John Benjamins. Harnisch, Rüdiger (1989). Die sogenannte ‘sogenannte Flexion der Konjunktionen’. Ein Paradigma aus der Bavaria thuringica. In E. Koller, W. Wegstein, & N. R. Wolf (Eds.), Bayerisch-österreichische Dialektforschung (pp. 283–290). Würzburg: Königshausen & Neumann. Harris, Alice (1981). Georgian Syntax: A Study in Relational Grammar. Cambridge: Cambridge University Press. Harris, Alice (1994). Ergative-to-accusative shift in agreement: Tabassaran. In Howard I. Aronson (Ed.), Linguistic Studies in the Non-Slavic Languages of the Commonwealth of Independent States and the Baltic Republics (NSL 7) (pp. 113–131). Chicago: Chicago Linguistic Society. Harris, Alice, & Lyle Campbell (1995). Historical Syntax in Cross-linguistic Perspective. Cambridge: Cambridge University Press. Harris, Alice, & Ricks Smeets. (Eds.). (2003). The Languages of the Caucasus. London: Routledge Curzon. Harris, Martin (1978). The Evolution of French Syntax. A Comparative Approach. London: Longman. Haspelmath, Martin (1995). The growth of affixes. In G. Booij & J. van Marle (Eds.), Yearbook of Morphology 1994 (pp. 1–29). Dordrecht: Kluwer. Hawkins, John A., & Gary Gilligan (1988). Prefixing and suffixing universals in relation to basic word order. Lingua, 74, 219–259. Heine, Bernd, Ulrike Claudi, & Friederike Hünnemeyer (1991). Grammaticalizazion: A Conceptual Framework. Chicago: University of Chicago Press. Heine, Bernd, & Tania Kuteva (2002). World Lexicon of Grammaticalization. Cambridge: Cambridge University Press. Heine, Bernd, & Mechthild Reh (1984). Grammaticalization and Reanalysis in African Languages. Hamburg: Helmut Buske. Heller, Karin (1975). Syntaktische Einflüsse des Italienischen im Zimbrischen. Incontri Linguistici, 2, 165–176. Hetzron, Robert (1976). The Agaw languages. Afroasiatic Linguistics, 3 (3), 1–45. Hewitt, George (1996). Georgian: A Structural Reference Grammar. Amsterdam: John Benjamins. Hock, Hans H. (1991). Principles of Historical Linguistics (2nd ed.). Berlin: Mouton de Gruyter. Hoekstra, Jarich, & Laszlo Marácz (1989). On the position of inflection in West Germanic. Working Papers in Scandinavian Syntax, 44, 75–88. Hoekstra, Eric, & Caroline Smits (1999). Everything you always wanted to know about complementizer agreement. In E. van Gelderen & V. Samiian (Eds.), Proceedings of WECOL 1998. Fresno, CA: California State University Press. Höhle, Tilman N. (1978). Lexikalische Syntax: Die Aktiv-Passiv-Relation und andere Infinitkonstruktionen im Deutschen. Tübingen, Niemeyer. Holmberg, Anders, & Christer Platzack (1988). On the role of inflection in Scandinavian syntax. Working Papers in Scandinavian Syntax, 42, 25–42.



JB[v.20020404] Prn:8/09/2005; 13:19

F: LA81RE.tex / p.10 (1168-1303)

 The Rise of Agreement

Holmberg, Anders, & Christer Platzack (1991). On the role of inflection in Scandinavian syntax. In W. Abraham, W. Kosmeijer, & E. Reuland (Eds.), Issues in Germanic Syntax (pp. 93–118). Berlin: Mouton de Gruyter. Holmberg, Anders, & Christer Platzack (1995). The Role of Inflection in Scandinavian Syntax. Oxford: Oxford University Press. Holmberg, Anders (2000). Scandinavian Stylistic Fronting: How any category can become an expletive. Linguistic Inquiry, 31 (3), 445–484. Hopper, Paul, & Elizabeth Traugott (1993). Grammaticalization. Cambridge: Cambridge University Press. Hotzenköcherle, Rudolf (1971). Die südwalserisch-ernetbirgischen Mundarten im Spiegel ihrer Verbalformen. In M. Bindschedel, R. Hotzenköcherle, & W. Kohlschmidt (Eds.), Festschrift für Paul Zinsli (pp. 79–98). Bern: Francke. Humboldt, Wilhelm von (1822). Über das Entstehen der grammatischen Formen und ihren Einfluß auf die Ideenentwicklung. Abhandlungen der Akademie der Wissenschaften zu Berlin. Reprint: Humboldt 1972, 31–63. Humboldt, Wilhelm von (1836). Über die Verschiedenheit des menschlichen Sprachbaues und ihren Einfluß auf die geistige Entwicklung des Menschengeschlechtes. Reprint: Humboldt 1972, 368–756. Humboldt, Wilhelm von (1972). Schriften zur Sprachphilosophie. (= Werke in fünf Bänden, ed. A. Flitner & K. Giel, Bd. III). Darmstadt: Wissenschaftliche Buchgesellschaft. Hyams, Nina (1986). Language Acquisition and the Theory of Parameters. Dordrecht: Reidel. Iatridou, Sabine (1990). About Agr(P). Linguistic Inquiry, 21, 551–577. Ingram, David (1978). Typology and universals of personal pronouns. In J. H. Greenberg (Ed.), Universals of Human Language. Vol. 3 (pp. 213–247). Stanford: Stanford University Press. Jespersen, Otto (1922). Language: Its Nature, Development and Origin. London: Allen and Unwin. Julien, Marit (2002). Syntactic Heads and Word Formation. Oxford: Oxford University Press. Kachru, Yamuna (1990). Hindi-Urdu. In B. Comrie (Ed.), The Major Languages of Southeast Asia, the Middle East and Africa (pp. 53–72). London: Routledge. Kayne, Richard (1975). French Syntax. The Transformational Cycle. Cambridge, MA: The MIT Press. Kayne, Richard (1984). Connectedness and Binary Branching. Dordrecht: Foris. Kayne, Richard (1989). Facets of past participle agreement in Romance. In P. Benincà (Ed.), Dialect Variation and the Theory of Grammar (pp. 85–103). Dordrecht: Foris. Kayne, Richard (1994). The Antisymmetry of Syntax. Cambridge, MA: The MIT Press. Kayne, Richard (2002). Pronouns and their antecedents. In S. Epstein & T. D. Seely (Eds.), Derivation and Explanation in the Minimalist Program (pp. 133–166). Oxford: Blackwell. Keegan, John M. (1997). A Reference Grammar of Mbay. Munich: Lincom Europa. Keenan, Edward (2002). Explaining the creation of reflexive pronouns in English. In D. Minkova & R. Stockwell (Eds.), Studies in the History of English: A Millennial Perspective (pp. 325– 355). Berlin: Mouton de Gruyter. Kemenade, Ans van (1987). Syntactic Case and Morphological Case in the History of English. Dordrecht: Foris. Kemenade, Ans van, & Nigel Vincent (Eds.). (1997). Parameters of Morphosyntactic Change. Cambridge: Cambridge University Press. Kenstowicz, Michael (1994). Phonology in Generative Grammar. Oxford: Blackwell.

JB[v.20020404] Prn:8/09/2005; 13:19

F: LA81RE.tex / p.11 (1303-1438)

References

Kenstowicz, Michael (1997). Base-identity and uniform exponence: Alternatives to cyclicity. In J. Durand & B. Laks (Eds.), Current Trends in Phonology: Models and Methods (pp. 363– 394). CNRS and University of Salford Publications. King, Tracy Holloway (1994). SpecAgrP and case: Evidence from Georgian. In H. Harley & C. Phillips (Eds.), MIT Working Papers in Linguistics 22: The Morphology-Syntax Connection (pp. 91–110). Cambridge, MA: Department of Linguistics and Philosophy, MIT. Kinyalolo, Kasangati (1991). Syntactic Dependencies and the Spec-head Agreement Hypothesis in Kilega. Doctoral dissertation, UCLA. Kiparsky, Paul (1973). ‘Elsewhere’ in phonology. In S. Anderson & P. Kiparsky (Eds.), A Festschrift for Morris Halle (pp. 93–106). New York, NY: Holt, Rinehart and Winston. Kiparsky, Paul (1983). Word-formation and the lexicon. In F. Ingemann (Ed.), Proceedings of the 1982 Mid-America Linguistics Conference (pp. 3–29). Lawrence, University of Kansas. Kiparsky, Paul (1996). The shift to head-initial VP in Germanic. In H. Thráinsson, S. D. Epstein, & S. Peter (Eds.), Studies in Comparative Germanic Syntax II (pp. 140–179). Dordrecht: Kluwer. Kiparsky, Paul (1997). The rise of positional licensing. In A. van Kemenade & N. Vincent (Eds.), Parameters of Morphosyntactic Change (pp. 460–493). Cambridge: Cambridge University Press. Kirby, Simon (1999). Function, Selection, and Innateness: The Emergence of Language Universals. Oxford: Oxford University Press. Kitahara, Hisatsugu (1997). Elementary Operations and Optimal Derivations. Cambridge, MA: The MIT Press. Kollmer, Michael (1987). Die schöne Waldlersprach. Bd. I–III. Prackenbach: Kollmer (Eigenverlag). Koopman, Hilda (1992). On the absence of Case chains in Bambara. Natural Language and Linguistic Theory, 10, 555–594. Koopman, Hilda (1996). The spec head configuration. Ms., UCLA. Koopman, Hilda (2003). Inside the ‘Noun’ in Maasai. In A. Mahajan (Ed.), Syntax at Sunset 3: Head Movement and Syntactic Theory (UCLA Working Papers in Linguistics 10) (pp. 77– 115). Los Angeles, CA: UCLA Linguistics Department. Koopman, Hilda, & Anna Szabolcsi (2000). Verbal Complexes. Cambridge, MA: The MIT Press. Kornfilt, Jaklin (1990). Turkish and the Turkic languages. In B. Comrie (Ed.), The Major Languages of Eastern Europe (pp. 227–252). London: Routledge. Kosmeijer, Wim (1986). The status of the finite inflection in Icelandic and Swedish. Working Papers in Scandinavian Syntax, 26, 1–41. Krifka, Manfred (1995). Swahili. In J. Jacobs, A. von Stechow, W. Sternefeld, & T. Vennemann (Eds.), Syntax: An International Handbook of Contemporary Research (pp. 1397–1418). Berlin: Walter de Gruyter. Kroch, Anthony (1989). Reflexes of grammar in patterns of language change. Language Variation and Change, 1, 199–244. Kroch, Anthony (1994). Morphosyntactic variation. In K. Beals, J. Denton, B. Knippen, L. Melnar, H. Suzuki, & E. Zeinfeld (Eds.), Proceedings of the Thirthieth Annual Meeting of the Chicago Linguistics Society. Vol. II (pp. 180–201). Chicago, IL: Chicago Linguistics Society. Kroch, Anthony (2001). Syntactic change. In M. Baltin & C. Collins (Eds.), The Handbook of Contemporary Syntactic Theory (pp. 699–730). Oxford: Blackwell. Kroch, Anthony, & Ann Taylor (1997). Verb movement in Old and Middle English: Dialect variation and language contact. In A. van Kemenade & N. Vincent (Eds.), Parameters of Morphosyntactic Change (pp. 297–325). Cambridge: Cambridge University Press.



JB[v.20020404] Prn:8/09/2005; 13:19

F: LA81RE.tex / p.12 (1438-1568)

 The Rise of Agreement

Kuen, Heinrich (1957). Die Gewohnheit der mehrfachen Bezeichnung des Subjekts in der Romania und die Gründe ihres Aufkommens. In G. Reichenkron (Ed.), Syntactica und Stilistica. Festschrift für Ernst Gamillscheg zum 70. Geburtstag (pp. 293–326). Tübingen: Niemeyer. Kuryłowicz, Jerzy (1964). The Inflectional Categories of Indo-European. Heidelberg: Carl Winter. Laenzlinger, Christopher (1998). Comparative Studies in Word Order Variation. Amsterdam: John Benjamins. Laka, Itziar (1990). Negation in Syntax: On the Nature of Functional Categories and Projections. Doctoral dissertation, MIT. Lambrecht, Knud (1981). Topic, Antitopic and Verb-agreement in Non-Standard French [Pragmatics and beyond Vol. II: 8]. Amsterdam: John Benjamins. LaPolla, Randy (1992). On the dating and nature of verb agreement in Tibeto Burman. Bulletin of the School of Oriental and African Studies, 55 (2), 298–315. LaPolla, Randy (1994). Parallel grammaticalization in Tibeto-Burman languages: Evidence of Sapir’s drift. Linguistics of the Tibeto-Burman Area, 17 (1), 61–79. Lechner, Winfried (2001). Reduced and phrasal comparatives. Natural Language and Linguistic Theory, 19, 683–735. Legate, Julie (2003). Some interface properties of the phase. Linguistic Inquiry, 34 (3), 506–516. Lehmann, Christian (1988). On the function of agreement. In M. Barlow & Charles A. Ferguson (Eds.), Agreement in Natural Language (pp. 55–65). Stanford, CA: CSLI. Lehmann, Christian (1993). Kongruenz. In J. Jacobs, A. von Stechow, W. Sternefeld, & T. Vennemann (Eds.), Syntax: An International Handbook of Contemporary Research Vol. 1 (pp. 722–729). Berlin: Walter de Gruyter. Lehmann, Christian (2002). Thoughts on Grammaticalization (2nd edition). Arbeitspapiere des Seminars für Sprachwissenschaft der Universität Erfurt. Lehmann, Winfred (1993). Theoretical Bases of Indo-European Linguistics. London: Routledge. Lessiak, Primus (1963). Die Mundart von Pernegg in Kärnten. Marburg: N. G. Elwert. Lightfoot, David (1979). Principles of Diachronic Syntax. Cambridge: Cambridge University Press. Lightfoot, David (1991). How to Set Parameters: Arguments from Language Change. Cambridge, MA: The MIT Press. Lightfoot, David (Ed.). (2002). Syntactic Effects of Morphological Change. Oxford: Oxford University Press. Linder, Karl Peter (1987). Grammatische Untersuchungen zur Charakteristik des Rätoromanischen in Graubünden. Tübingen: Gunter Narr. Liver, Ricarda (1999). Rätoromanisch: Eine Einführung in das Bündnerromanische. Tübingen: Gunter Narr. Longobardi, Giuseppe (2001). Formal syntax, diachronic minimalism, and etymology: The history of French chez. Linguistic Inquiry, 32 (2), 275–302. Lüdtke, Helmut (1980). Auf dem Wege zu einer Theorie des Sprachwandels. In H. Lüdtke (Ed.), Kommunikationstheoretische Grundlagen des Sprachwandels (pp. 182–252). Berlin: de Gruyter. Lühr, Rosemarie (1984). Reste der athematischen Konjugation in den germanischen Sprachen. In J. Untermann & B. Brogyanyi (Eds.), Das Germanische und die Rekonstruktion der Indogermanischen Grundsprache (pp. 25–90). Amsterdam: John Benjamins. MacKenzie, D. N. (1990). Pashto. In B. Comrie (Ed.), The Major Languages of Southeast Asia, the Middle East and Africa (pp. 132–150). London: Routledge.

JB[v.20020404] Prn:8/09/2005; 13:19

F: LA81RE.tex / p.13 (1568-1691)

References

Manzini, Rita, & Leonardo Savoia (2002). Parameters of subject inflection in Italian dialects. In P. Svenonius (Ed.), Subjects, Expletives, and the EPP (pp. 159–200). Oxford: Oxford University Press. Marantz, Alec (1984). On the Nature of Grammatical Relations. Cambridge, MA: The MIT Press. Marantz, Alec (1988). Clitics, morphological merger, and the mapping to phonological structure. In M. Hammond & M. Noonan (Eds.), Theoretical Morphology: Approaches in Modern Linguistics (pp. 253–270). San Diego, CA: Academic Press. Marantz, Alec (1992). Case and licensing. In G. Westphal, B. Ao, & H.-R. Chae (Eds.), Proceedings of ESCOL 1991 (pp. 234–253). Columbus: Department of Linguistics, Ohio State University. Marantz, Alec (1997). No escape from syntax: Don’t try morphological analysis in the privacy of your own Lexicon. In A. Dimitriadis, L. Siegel, C. Surek-Clark, & A. Williams (Eds.), Proceedings of the 21st Annual Penn Linguistics Colloquium. Penn Working Papers in Linguistics 4.2 (pp. 201–225). Philadelphia, PA: University of Pennsylvania, Martin, Samuel. E. (1961). Dagur Mongolian Grammar, Texts, and Lexicon. Based on the Speech of Peter Onon. Bloomington, IN: Indiana University Publications. Matthews, Peter H. (1991). Morphology (2nd edition). Cambridge: Cambridge University Press. Meillet, Antoine (1912). L’evolution des formes grammaticales. Scientia (Rivista di Scienza 12, No. 26.6). Reprinted in A. Meillet 1958, Linguistique historique et linguistique générale (pp. 130–148). Paris: Champion. Meinhof, Carl (1936). Die Entstehung flektierender Sprachen. Berlin: Dietrich Reimer. Merchant, Jason (2001). The Syntax of Silence: Sluicing, Islands, and the Theory of Ellipsis. Oxford: Oxford University Press. Merkle, Ludwig (1975). Bairische Grammatik. München: Heimeran. McCloskey, James (1996). On the scope of verb movement in Irish. Natural Language and Linguistic Theory, 14, 47–104. McCloskey, James (1997). Subjecthood and subject positions. In L. Haegeman (Ed.), Elements of Grammar (pp. 197–235). Dordrecht: Kluwer. McCloskey, James, & Kenneth Hale (1984). On the syntax of person-number inflection in Modern Irish. Natural Language and Linguistic Theory, 1, 487–533. Means, Nathalie, Gordon P. Means, & Paul B. Means (1986). Sengoi-English, English-Sengoi Dictionary. Toronto: Joint Centre on Modern East Asia, University of Toronto/York University. Meyer-Lübke, Wilhelm (1890). Italienische Grammatik. Leipzig: O. R. Reisland. Meyer-Lübke, Wilhelm (1894). Grammatik der romanischen Sprachen. Bd. II: Romanische Formenlehre. Leipzig: O. R. Reisland. Mitchell, Erica (1994). When Agro is fused to Agrs: What morphology can tell us about the functional categories. In H. Harley & C. Phillips (Eds.), MIT Working Papers in Linguistics 22: The Morphology-Syntax Connection (pp. 111–130). Cambridge, MA: Department of Linguistics and Philosophy, MIT. Mithun, Marianne (1991). The development of bound pronominal paradigms. In W. P. Lehmann & H.-J. Jakusz Hewitt (Eds.), Language Typology 1988: Typological Models in Reconstruction (pp. 85–104). Amsterdam: John Benjamins. Moscati, Sabatino, Anton Spitaler, Edward Ullendorff, & Wolfram von Soden (1969). An Introduction to the Comparative Grammar of the Semitic Languages. Wiesbaden: Otto Harassowitz. Moulton, William (1944). Scribe γ of the Old High German Tatian translation. PMLA, 59 (2), 307–334.



JB[v.20020404] Prn:8/09/2005; 13:19

F: LA81RE.tex / p.14 (1691-1811)

 The Rise of Agreement

Müller, Friedrich Max (1875). Vorlesungen über die Wissenschaft der Sprache, Bd. 1. Leipzig: Julius Klinkhardt. Müller, Gereon (2004). Verb-second as νP-first. The Journal of Comparative Germanic Linguistics, 7 (3), 179–234. Newmeyer, Frederick (1998). Language Form and Language Function. Cambridge, MA: The MIT Press. Newmeyer, Frederick (2001). Deconstructing grammaticalization. Language Sciences, 23, 187– 229. Niyogi, Partha, & Robert C. Berwick (1998). The logical problem of language change: A case study of European Portuguese. Syntax, 1 (2), 192–205. Noonan, Michael (1992). A Grammar of Lango. Berlin: Mouton de Gruyter. Nordlinger, Rachel (1995). Split tense and mood inflection in Wambaya. Berkeley Linguistics Society, 21, 226–236. Nordlinger, Rachel (1998). A Grammar of Wambaya, Northern Territory (Australia). (Pacific Linguistics C-140). Canberra: Pacific Linguistics. Noyer, Rolf (1997). Features, Positions, and Affixes in Autonomous Morphological Structure. New York: Garland. Nübling, Damaris (1992). Klitika im Deutschen. Schriftsprache, Umgangssprache, alemannische Dialekte. Tübingen: Gunter Narr. Nübling, Damaris (1995). Kurzverben in germanischen Sprachen. Unterschiedliche Wege – gleiche Ziele. Zeitschrift für Dialektologie und Linguistik, 62 (2), 127–154. Oetzel, Annette (1992). Markierte Wortstellung im Bündnerromanischen. Frankfurt am Main: Peter Lang. O’Grady, William (1997). Syntactic Development. Chicago, IL: Chicago University Press. Ouhalla, Jamal (1990). Sentential negation, Relativised Minimality and the aspectual status of auxiliaries. The Linguistic Review, 7, 183–231. Ouhalla, Jamal (1991). Functional Projections and Parametric Variation. London: Routledge. Paul, Hermann (1880). Prinzipien der Sprachgeschichte. (5th edition, published 1920, Konzepte der Sprach- und Literaturwissenschaft, 6). Tübingen: Niemeyer. Paul, Hermann (1952). Deutsche Grammatik II.3: Flexionslehre. Halle: Max Niemeyer. Paul, Hermann (1998). Mittelhochdeutsche Grammatik (24th edition). Tübingen: Niemeyer. Pesetsky, David, & Esther Torrego (2001). T-to-C movement: Causes and consequences. In M. Kenstowicz (Ed.), Ken Hale: A Life in Language (pp. 355–426). Cambridge, MA: The MIT Press. Pesetsky, David, & Esther Torrego (2004). The syntax of evaluation and the interpretability of features. Ms., MIT and University of Massachusetts/Boston. Pfalz, Anton (1918). Suffigierung der Personalpronomina im Donaubairischen. Republished 1983 in P. Wiesinger (Ed.), Die Wiener dialektologische Schule. Grundsätzliche Studien aus 70 Jahren Forschung [Wiener Arbeiten zur germanischen Altertumskunde und Philologie 23] (pp. 217–235). Wien: Peter Lang. Pintzuk, Susan (1999). Phrase Structures in Competition: Variation and Change in Old English Word Order. New York, NY: Garland. Pintzuk, Susan, Geroge Tsoulas, & Anthony Warner. (Eds.). (2000). Diachronic Syntax. Models and Mechanisms. Oxford: Oxford University Press. Platzack, Christer (1988). The emergence of a word order difference in Scandinavian subordinate clauses. McGill Working Papers in Linguistics: Special Issue on Comparative Germanic Syntax, 215–238.

JB[v.20020404] Prn:8/09/2005; 13:19

F: LA81RE.tex / p.15 (1811-1937)

References 

Poletto, Cecilia (1995). The diachronic development of subject clitics in North-Eastern Italian dialects. In A. Battye & I. Roberts (Eds.), Clause Structure and Language Change (pp. 295– 324). Oxford: Oxford University Press. Poletto, Cecilia (1997). Pronominal syntax. In M. Maiden & M. Parry (Eds.), The Dialects of Italy (pp. 137–145). Cambridge: Cambridge University Press. Poletto, Cecilia (1999). The internal structure of Agrs and subject clitics in the Northern Italian dialects. In H. van Riemsdijk (Ed.), Clitics in the Languages of Europe (pp. 581–620). Berlin: Mouton de Gruyter. Poletto, Cecilia (2000). The Higher Functional Field. Evidence from Northern Italian Dialects. Oxford: Oxford University Press. Polinsky, Maria (2000). American Russian: An endangered language? Ms., University of California, San Diego. Polinsky, Maria (2003). Non-canonical agreement is canonical. Transactions of the Philological Society, 101 (2), 279–312. Polinsky, Maria, & Bernard Comrie (1999). Agreement in Tsez. Folia Linguistica, 33, 109–130. Polinsky, Maria, & Eric Potsdam (2001). Long distance agreement and topic in Tsez. Natural Language and Linguistic Theory, 19 (3), 583–646. Pollock, Jean-Yves (1989). Verb movement, UG and the structure of IP. Linguistic Inquiry, 20 (3), 365–424. Poppe, Nicholas (1954). Grammar of Written Mongolian. Wiesbaden: Otto Harrassowitz. Poppe, Nicholas (1960). Buriat Grammar. Bloomington, IN: Indiana University Publications. Prasada, Sandeep, & Steven Pinker (1993). Generalization of regular and irregular morphological patterns. Language and Cognitive Processes, 8, 1–56. Ramat, Anna Giacalone, & Paul Hopper (Eds.). (1998). The Limits of Grammaticalization. Amsterdam: John Benjamins. Rezac, Milan (2004). Elements of Cyclic Syntax: Agree and Merge. Doctoral dissertation, University of Toronto. Renzi, Lorenzo, & Laura Vanelli (1983). I pronomi soggeto in alcune varietà romanze (Subject pronouns in some Romance varieties). In P. Benincà et al. (Eds.), Scritti linguistici in onore die Giovan Battista Pellegrini (Linguistic Papers in Honor of G. B. Pellegrini) (pp. 121–145). Pisa: Pancini. Richards, Norvin (2001). Movement in Language: Interactions and Architectures. Oxford: Oxford University Press. Rizzi, Luigi (1986). On the status of subject clitics in Romance. In O. Jaeggli & C. Silva-Corvalan (Eds.), Studies in Romance Linguistics (pp. 391–419). Dordrecht: Foris. Rizzi, Luigi (1997). The fine structure of the left periphery. In L. Haegeman (Ed.), Elements of Grammar: Handbook in Generative Syntax (pp. 281–337). Dordrecht: Kluwer. Rizzi, Luigi, & Ian Roberts (1989). Complex inversion in French. Probus, 1, 1–30. Republished 1996 in A. Belletti & L. Rizzi (Eds.), Parameters and Functional Heads (pp. 91–116). Oxford: Oxford University Press. Roberts, Ian (1993a). Verbs and Diachronic Syntax. A Comparative History of English and French. Dordrecht: Kluwer. Roberts, Ian (1993b). A formal account of grammaticalization in the history of Romance futures. Folia Linguistica Historica, 13, 219–258. Roberts, Ian (1994). Second position effects and agreement in Comp. Ms., University of Bangor. Roberts, Ian (1995). Object movement and verb movement in Early Modern English. In H. Haider, S. Olsen, & S. Vikner (Eds.), Studies in Comparative Germanic Syntax (pp. 269–284). Dordrecht, Kluwer.

JB[v.20020404] Prn:8/09/2005; 13:19

F: LA81RE.tex / p.16 (1937-2063)

 The Rise of Agreement

Roberts, Ian (1997). Directionality and word order change in the history of English. In A. van Kemenade & N. Vincent (Eds.), Parameters of Morphosyntactic Change (pp. 397–427). Cambridge: Cambridge University Press. Roberts, Ian (1998). Have/Be raising, Move F, and Procrastinate. Linguistic Inquiry, 29 (1), 113– 125. Roberts, Ian (1999a). Verb movement and markedness. In M. DeGraff (Ed.), Language Creation and Language Change: Creolization, Diachrony and Development (pp. 287–329). Cambridge, MA: The MIT Press. Roberts, Ian (1999b). Agreement marking in Welsh and Romance. In H. van Riemsdijk (Ed.), Clitics in the Languages of Europe (pp. 621–638). Berlin: Mouton de Gruyter. Roberts, Ian, & Anna Roussou (1999). A formal approach to ‘grammaticalization’. Linguistics, 37, 1011–1041. Roberts, Ian, & Anna Roussou (2003). Syntactic Change. A Minimalist Approach to Grammaticalization. Cambridge: Cambridge University Press. Robinson, Orrin W. (1997). Clause Subordination and Verb Placement in the Old High German Isidor Translation. Heidelberg: C. Winter. Rohlfs, Gerhard (1949). Historische Grammatik der italienischen Sprache und ihrer Mundarten. Bd. II: Formenlehre und Syntax. Bern: A. Francke. Rohrbacher, Bernhard (1994). The Germanic VO languages and the Full Paradigm: A Theory of V-to-I Raising. Doctoral dissertation, University of Massachusetts, Amherst. Rohrbacher, Bernhard (1999). Morphology-driven Syntax: A Theory of V-to-I Raising and Prodrop. Amsterdam: John Benjamins. Ross, John R. (1969). Guess who? In R. Binnick, A. Davison, G. Green, & J. Morgan (Eds.), Papers from the 5th regional meeting of the Chicago Linguistic Society (pp. 252–286). Chicago, IL: Chicago Linguistic Society. Rowley, Anthony R. (1994). Morphologie aus Syntax – natürlich. Zur Flexion der Nebensatzeinleiter in nordostbayerischen Dialekten. In W. Viereck (Ed.), Verhandlungen des Internationalen Dialektologenkongresses Bamberg, 29.7. – 4.8.1990). Band 3: Regionalsprachliche Variation, Umgangs- und Standardsprachen (Zeitschrift für Dialektologie und Linguistik, Beiheft 76) (pp. 488–497). Stuttgart: Franz Steiner Verlag. Sabel, Joachim (1996). Restrukturierung und Lokalität. Berlin: Akademie Verlag. Sapir, Edward (1921). Language. New York, NY: Harcourt, Brace and World. Sanžeev, Garma Dancaranovic (1973). The Modern Mongolian Language. Moscow: Nauka. Sauerland, Uli (1996). The late insertion of Germanic inflection. Ms., MIT. Savoia, Leonardo (1997). Inflectional morphology of the verb. In M. Maiden & M. Parry (Eds.), The Dialects of Italy (pp. 75–86). London: Routledge. Schlegel, August Wilhelm von (1818). Observations sur la langue et la littérature provençales. Paris: Librairie grecque-latine-allemande. Schulze, Wolfgang (1998). Person, Klasse, Kongruenz. Fragmente einer Kategorialtypologie des einfachen Satzes in den ostkaukasischen Sprachen. Band II: Die Person. Ms., University of Munich. Schütze, Carson (1994). Serbo-Croatian second position clitic placement and the phonologysyntax interface. In A. Carnie, H. Harley, & T. Bures (Eds.), MIT Working Papers in Linguistics 21: Papers on Phonology and Morphology (pp. 373–473). Cambridge, MA: Department of Linguistics and Philosophy, MIT. Schweizer, Bruno (1952). Zimbrische Gesamtgrammatik. Fünfter Teil: Syntax der zimbrischen Dialekte in Oberitalien. Ms. Shlonsky, Ur. (1994). Agreement in Comp. The Linguistic Review, 11, 351–375.

JB[v.20020404] Prn:8/09/2005; 13:19

F: LA81RE.tex / p.17 (2063-2204)

References

Sievers, Eduard (Ed.). (1961). Tatian. Darmstadt: Wissenschaftliche Buchgesellschaft. Siewierska, Anna, & Dik Bakker (1996). The distribution of subject and object agreement and word order type. Studies in Language, 20 (1), 115–161. Siewierska, Anna (1999). From anaphoric pronoun to grammatical marker: Why objects don’t make it. Folia Linguistica, 33 (2), 225–251. Siewierska, Anna (2004). Person. Cambridge: Cambridge University Press. Sigurðsson, Halldór A. (1996). Icelandic finite verb agreement. Working Papers in Scandinavian Syntax, 57 1–46. Sigurðsson, Halldór A. (2000). The locus of case and agreement. Working Papers in Scandinavian Syntax, 65, 65–108. Sigurðsson, Halldór A. (2001). Inflectional features and clausal structure. In J. Niemi & J. Heikkinen (Eds.), Nordic and Baltic Morphology: Papers from A NorFA Course, Tartu, June 2000 [Studies in Languages 36] (pp. 99–111). University of Joenssu. Sigurðsson, Halldór A. (2002). Agree and agreement: evidence from Germanic. Working Papers in Scandinavian Syntax, 70, 101–156. Simpson, Andrew, & Tanmoy Bhattacharya (2003). Obligatory overt wh-movement in a wh-insitu language. Linguistic Inquiry, 34 (1), 127–142. Simpson, Andrew, & Zoe Wu (2002). Agreement, shells and focus. Language, 78 (2), 287–313. Sommer, Thomas (1994). Flexionsmorphologie des Verbs im althochdeutschen Tatian. München: Tuduv. Speas, Margaret (1991). Functional heads and inflectional morphology. The Linguistic Review, 8, 389–417. Spencer, Andrew (1991). Morphological Theory. Oxford: Blackwell. Spencer, Andrew (1997). Inflectional morphology and functional heads. In W. Dressler, M. Prinzhorn, & J. Rennison (Eds.), Advances in Morphology (pp. 31–49). Berlin: Mouton de Gruyter. Spiess, Federico (1956). Die Verwendung des Subjektpersonalpronomens in den lombardischen Mundarten [Romanica Helvetica 59]. Bern: A. Francke. Sportiche, Dominique (1998). Partitions and Atoms of Clause Structure. London: Routledge. Steele, Susan (1977). Clisis and diachrony. In C. N. Li (Ed.), Mechanisms of Syntactic Change (pp. 539–579). Austin, TX: University of Texas Press. Steele, Susan (1978). Word order variation: A typological study. In J. H. Greenberg, C. A. Ferguson, & E. A. Moravcsik (Eds.), Universals of Human Language IV: Syntax (pp. 585– 623). Stanford, CA: Stanford University Press. Steele, Susan (1995). Clisis. In J. Jacobs, A. von Stechow, W. Sternefeld, & T. Vennemann (Eds.), Syntax: An International Handbook of Contemporary Research Vol. 2 (pp. 1226–1235). Berlin: Walter de Gruyter. Steriade, Donca (2000). Paradigm uniformity and the phonetics/phonology boundary. In M. Broe & J. Pierrehumbert (Eds.), Papers in Laboratory Phonology V: Acquisition and the Lexicon (pp. 313–334). Cambridge: Cambridge University Press. Stowell, Tim (1996). The phrase structure of tense. In J. Rooryk & L. Zaring (Eds.), Phrase Structure and the Lexicon (pp. 277–292). Dordrecht: Kluwer. Stump, Gregory T. (1998). Inflection. In A. Spencer & A. Zwicky (Eds.), The Handbook of Morphology (pp. 13–43). Oxford, MA: Blackwell. Suñer, Margarita (1988). The role of agreement in clitic-doubled constructions. Natural Language and Linguistic Theory, 6, 391–434. Suñer, Margarita (1992). Subject clitics in the Northern Italian vernaculars and the matching hypothesis. Natural Language and Linguistic Theory, 10, 641–672.



JB[v.20020404] Prn:8/09/2005; 13:19

F: LA81RE.tex / p.18 (2204-2331)

 The Rise of Agreement

Swan, Toril (1994). Old English and Old Norse initial adverbials and word order. In T. Swan, E. Mørck, & O. J. Westvik (Eds.), Language Change and Language Structure: Older Germanic Languages in Comparative Perspective (pp. 233–270). Berlin: Mouton de Gruyter. Szemerényi, Oswald (1989). Einführung in die vergleichende Sprachwissenschaft (3rd edition). Darmstadt: Wissenschaftliche Buchgesellschaft. Taylor, Ann (1994). Variation in past tense formation in the history of English. In R. Izvorski, M. Meyerhoff, B. Reynolds, & V. Tredinnick. (Eds.), University of Pennsylvania Working Papers in Linguistics 1 (pp. 143–159). University of Pennsylvania, Philadelphia. Thráinsson, Höskuldur (1996). On the (non-)universality of functional categories. In W. Abraham, S. D. Epstein, H. Thráinsson & J.-W. Zwart (Eds.), Minimal Ideas. Syntactic Studies in the Minimalist Framework (pp. 253–281). Amsterdam: John Benjamins. Traugott, Elizabeth, & Bernd Heine (Eds.). (1991). Approaches to Grammaticalization. Amsterdam: John Benjamins. Travis, Lisa deMena (1984). Parameters and Effects of Word Order Variation. Doctoral dissertation, MIT. Trips, Carola (2002). From OV to VO in Early Middle English. Amsterdam: John Benjamins. Torrego, Esther (1995). On the nature of clitic doubling. In H. Campos & P. Kempchinsky (Eds.), Evolution and Revolution in Linguistic Theory: Essays in Honor of Carlos Otero (pp. 399–418). Dordrecht: Kluwer. Torrego, Esther (2002). Arguments for a derivational approach to syntactic relations based on clitics. In S. Epstein & T. D. Seely (Eds.), Derivation and Explanation in the Minimalist Program (pp. 249–268). Oxford: Blackwell. Tortora, Christina (2003). The Syntax of Italian Dialects. Oxford: Oxford University Press. Tyroller, Hans (2003). Grammatische Beschreibung des Zimbrischen von Lusern. Stuttgart: Steiner. Ura, Hiroyuki (2000). Checking Theory and Grammatical Functions in Universal Grammar. Oxford: Oxford University Press. Ura, Hiroyuki (2001). Case. In M. Baltin & C. Collins (Eds.), The Handbook of Contemporary Syntactic Theory (pp. 334–373). Oxford: Blackwell. Uriagereka, Juan (1995). Aspects of the syntax of clitic placement in western Romance. Linguistic Inquiry, 26, 79–123. Vance, Barbara (1997). Syntactic Change in Medieval French: Verb Second and Null Subjects. Dordrecht: Kluwer. Vanelli, Laura (1997) (with Lorenzo Renzi). Personal pronouns and demonstratives. In M. Maiden & M. Parry (Eds.), The Dialects of Italy (pp. 106–115). London: Routledge. Velasco, J. B. de (1737). Arte de la Lengua Cahita. (Edited and reprinted by Eustaquio Buelna, Mexico, 1890). Vikner, Sten (1994). Finite verb movement in Scandinavian embedded clauses. In N. Hornstein & D. Lightfoot (Eds.), Verb Movement (pp. 117–147). Cambridge: Cambridge University Press. Vikner, Sten (1995). Verb Movement and Expletive Subjects in the Germanic Languages. Oxford: Oxford University Press. Vikner, Sten (1997). V-to-I movement and inflection for person in all tenses. In L. Haegeman (Ed.), The New Comparative Syntax (pp. 189–213). London: Longman. Vogel, Irene (1999). Subminimal constituents in prosodic phonology. In S. J. Hannahs & M. Davenport (Eds.), Issues in Phonological Structure (pp. 251–267). Amsterdam: John Benjamins.

JB[v.20020404] Prn:8/09/2005; 13:19

F: LA81RE.tex / p.19 (2331-2451)

References 

Vogelaer, Gunther de, Annemie Neuckermans, & Guido vanden Wyngaerd. (2002). Complementizer agreement in the Flemish dialects. In S. Barbiers, L. Cornips, & S. van der Kleij (Eds.), Meertens Institute Electronic Publications in Linguistics (MIEPiL) II: Syntactic Microvariation (pp. 97–115). Amsterdam. Wartburg, Walther von (1970). Einführung in die Problematik und Methodik der Sprachwissenschaft (3rd edition). Tübingen: Niemeyer. Watanabe, Akira (1996). Case Absorption and WH-agreement. Dordrecht: Kluwer. Watanabe, Akira (2000). Feature copying and binding: Evidence from complementizer agreement and switch reference. Syntax, 3 (3), 159–181. Watkins, Calvert (1969). Indogermanische Grammatik III.1: Geschichte der indogermanischen Verbalflexion. Heidelberg: Carl Winter. Weber, Albert (1987). Zürichdeutsche Grammatik. Ein Wegweiser zur guten Mundart. Zürich: Hans Rohr. Weiß, Helmut (1998). Die Syntax des Bairischen. Studien zur Grammatik einer natürlichen Sprache. Tübingen, Niemeyer. Weiß, Helmut (2002). Agr-in Comp and partial pro-drop. Ms., Universität Regensburg. Weiß, Helmut (To appear). Inflected complementizers in Continental West Germanic dialects. Zeitschrift für Dialektologie und Linguistik. Werner, Otmar (1988). Mundartliche Enklisen bei Schmeller und heute. In L. M. Eichinger & B. Naumann (Eds.), Johann Andreas Schmeller und der Beginn der Grammatik (pp. 127– 147). München: Oldenburg. Wexler, Kenneth, & Peter Culicover (1980). Formal Principles of Language Acquisition. Cambridge, MA: The MIT Press. Wexler, Kenneth, & Rita M. Manzini (1987). Parameters, Binding theory, and learnability. Linguistic Inquiry, 18, 413–444. Widmer, P. Ambros (1959). Das Personalpronomen im Bündnerromanischen in phonetischer und morphologischer Schau. (Romanica Helvetica 67). Bern: A. Francke. Wiesinger, Peter (1989). Die Flexionsmorphologie des Verbums im Bairischen. Wien: Verlag der Österreichischen Akademie der Wissenschaften. Wischer, Ilse, & Gabriele Diewald (Eds.). (2002). New Reflections on Grammaticalization. Amsterdam: John Benjamins. Wright, Joseph (1925). Old English Grammar. (3rd edition). Oxford: Oxford University Press. Wu, Zoe (2004). Restructuring and the development of functional categories. In E. Fuß & C. Trips (Eds.), Diachronic Clues to Synchronic Grammar (pp. 191–217). Amsterdam: John Benjamins. Wurmbrand, Susi (2001). Infinitives: Restructuring and Clause Structure. Berlin: Mouton de Gruyter. Zanuttini, Raffaella (1997). Negation and Clausal Structure. Oxford: Oxford University Press. Zehetner, Ludwig (1985) (unter Mitarbeit von Ludwig M. Eichinger, Reinhard Rascher, Anthony Rowley, & Christopher J. Wickham). Das bairische Dialektbuch. München: Beck. Zeller, Jochen (1994). Die Syntax des Tempus. Opladen: Westdeutscher Verlag. Zeller, Jochen (2001). Particle Verbs and Local Domains. Amsterdam: John Benjamins. Zwart, Jan-Wouter (1993a). Dutch Syntax: A Minimalist Approach. Doctoral dissertation, University of Groningen. Zwart, Jan-Wouter (1993b). Clues from dialect syntax: Complementizer agreement. In W. Abraham & J. Bayer (Eds.), Dialektsyntax (Linguistische Berichte Sonderheft 5) (pp. 246– 270). Opladen: Westdeutscher Verlag.

JB[v.20020404] Prn:8/09/2005; 13:19

F: LA81RE.tex / p.20 (2451-2468)

 The Rise of Agreement

Zwart, Jan-Wouter (1997a). Morphosyntax of Verb Movement. A Minimalist Approach to the Syntax of Dutch. Dordrecht: Kluwer. Zwart, Jan-Wouter (1997b). Words and features. On lexicalization in Generative Grammar. Ms., University of Groningen. Zwart, Jan-Wouter (2001). Syntactic and phonological verb movement. Syntax, 4 (1), 34–62. Zwicky, Arnold, & Geoffrey Pullum (1983). Cliticization vs. inflection: English N’T. Language, 59 (3), 502–513.

JB[v.20020404] Prn:8/09/2005; 14:07

F: LA81IND.tex / p.1 (45-137)

Index

A A-movement , , , , –,  feeding agreement  A-position , , , ,  A/A’-distinction  A’-agreement  A’-movement , , , , , –, ,  Ablaut  Abney, Steven  Abraham, Werner , , , , ,  absolutive, see Case Accessibility marker ,  theory , ,  accusative, see Case Ackema, Peter , –, , ,  activating Case feature, see Case adjacency between clitic and finite verb, see clitic adjacency effect, see complementizer agreement adjective , , , , ,  inflection  adjunction , , , , , , , , , , –, , , , , ,  head adjunction , , , –, , , ,  adverb , , , , , –, , , ,  VP-adverb ,  affix-hopping  Afro-Asiatic ,  agglutinative ideal  Agree , –, , , , , , , , , , –, –, , , , , –, –, , , , , , , –,

–, , , , , ,  loss of an Agree relation  multiple Agree  agreement (Agr) 3rd person agreement, lack of , , , ,  as a relational category ,  canonical (via Agree) , , , , , , , –, , , , ,  checking , , , ,  context-sensitive  controller , , , , , , , –, , , , , , , ,  default , , , , , , ,  defective paradigm , , , , , , ,  domain , , ,  ergative/absolutive agreement  features , , , , , , , , , , , , , , , , , , , ,  features, interpretability of  fused marking , , , , ,  grammaticalization path , , , , ,  head , , , , , , , , , , , , ,  in gender , , , , ,  in noun class  in number , ,  in person , –, , , , , , , , , ,

, , –, , , , , ,  in tense features  licensing of , , , , , ,  loss of , , –, , , ,  morphological realization of – multiple agreement , , , , , , , , , , , , , , , , , , ,  object agreement , , , , , , , , , ,  participle agreement , –, , ,  phrase-structural representation of – prefixal , , , , , , , , , , , , , , , , , , , , , , , , , , , ,  suffixal , , , , , , , , –, , , , , , , –, –, , , , , , , , , , , , , , , , , , , , , , , , ,  and suffixing preference –, , , , , ,  via stem vowel alternation -, , , , ,  via consonant alternations (in Skou) ,  weakening (in Dutch) ,  agreement morpheme

JB[v.20020404] Prn:8/09/2005; 14:07

F: LA81IND.tex / p.2 (137-236)

 Index Agr-on-C , , , –, , , , , , , –, , , , , , , , –, , , ,  Agr-on-C, distribution of (in Sutselvan) , ,  Agr-on-T , , , , , , , , , , , ,  linearization of –, – parasitic on other functional heads , , , , , , , ,  agrammatism  AgrC , – adjunction to AgrCP  AgrC-to-C movement  SpecAgrCP , ,  agreement weakening (in Dutch), see agreement AgrO , , , , ,  SpecAgrOP  Agr-on-C, see agreement morpheme Agr-on-T, see agreement morpheme AgrP , , , , , ,  AgrS , , , –, , , , , , , ,  AgrS-to-C movement – SpecAgrSP , , ,  Albright, Adam –,  Alemannic , , ,  Alexiadou, Artemis  Algonquian ,  Allen, Cynthia ,  Altmann, Hans , , , , –, , ,  American Russian, see Russian Ampezzan ,  Anagnostopoulou, Elena  analogical change , , , , , , , , , , ,  analogical extension , ,  analogical leveling –, , ,  Anderson, Stephen , , , , , ,  anti-agreement , , ,  aorist , , 

argument licensing  argument order  argument structure , , , , , , , , , , , ,  Ariel, Mira , –, , , ,  Aronoff, Mark , ,  Arregi, Karlos , ,  Ashby, William , , , , ,  Aspect (Asp) , , , , –, , , , , –, –, , ,  asymmetric V2, see V2 Athabaskan  Auger, Julie , , , ,  Austronesian , , ,  Author in Speech Event (Auth) , , –, , –, ,  auxiliaries , , , –, , –, , ,  B Badiot  Baker, Mark , , , , , , , , ,  Bantu , ,  Bare Phrase Structure ,  Barlow, Michael  Basque , , ,  Bavarian , –, , , , , , –, –, , , , , , , , , , –, –, , , , , , , , ,  Lower Bavarian , , , –, , ,  Northern Bavarian ,  Bayer, Josef , , , , , , –, , ,  Beard, Robert  Bejar, Susana , , ,  Belletti, Adriana , ,  Benincà, Paola , –, , , , , , –, , , , –, , , , ,  Bennis, Hans ,  Benveniste, Emile ,  Berber 

Berwick, Robert  Besten, Hans den  Bickerton, Derek  big DP, see D bilingualism  Blackfoot  blocking effects, see morphological blocking effects Blocking Principle , , –, , , –, , –, , , , , , , –, , , , ,  and elimination of syncretisms ,  and phonemic distinctions  operates in a local fashion ,  Bobaljik, Jonathan , , , , , , , , , , , , –, , ,  Bonet, Eulalia , , ,  Bopp, Franz ,  Borer, Hagit , ,  Bosco Gurin  Brabants  Brandi, Luciana , , ,  Branigan, Phil  Bresnan, Joan , , , , ,  Brinkmann, Hennig , , ,  Brown, Samuel ,  Bruening, Benjamin ,  Brugmann, Karl ,  Buryat , –,  Bybee, Joan –, , , , , , , , ,  C C , , , , , , , , , , , , , , , , , , , –, , , , , –, , –, , , , , , , , , , , –, , , , , ,  CP , , , , , , , , , , , , , , , , , , , -, , 

JB[v.20020404] Prn:8/09/2005; 14:07

F: LA81IND.tex / p.3 (236-313)

Index  SpecCP , , , , ,  Campbell, Lyle ,  Cardinaletti, Anna  Carinthian ,  Carstens, Vicki , , , , , , , ,  Case –, , , , , –, , , , –, , –, ,  absolutive , –, , ,  accusative , , , , , , , ,  checking , , ,  dative , , ,  ergative , , , , , , , ,  feature , , , , ,  feature, activating , ,  feature, marked for deletion , , , , , ,  genitive , , ,  licensing of , , , , , , , ,  nominative , , , , –, , , , , , , , , ,  Caucasian , ,  Cecchetto, Carlo  Celtic , , ,  Chafe, Wallace ,  Chamorro  Chichewa , ,  Chinese  Chomsky, Noam , –, –, –, , , , , –, , , , , , –, –, , , , , , , , , , , , , , , ,  Chung, Sandra , , , ,  Cimbrian , –, , , ,  Cinque, Guglielmo , , , , , ,  Clark, Eve , , ,  Clark, Robin , , , , ,  Classical Aztec , , , ,  Classical Mongolian , , 

clause structure , , , , ,  clause type , , ,  clause union  clitic adjacency between clitic and finite verb , , , , , , , , , , , ,  complementary distribution of clitic and full forms , , , , ,  C-oriented , , , –, , , , , , , , -, , , , , , , , – C-oriented clitics, reanalysis of , , , , , , , , , , , , , , ,  Infl-oriented clitics, reanalysis of –, ,  movement of , , ,  non-distinctive (in Vicentino) ,  obligatorily present , , , –, , , , ,  reanalysis of , –, , , , , , , , , , , , , , , , , , , , , , , , , , , , , ,  second position clitics , , , , , , , , –, –,  Wackernagel clitics , , ,  clitic climbing  clitic cluster  clitic doubling –, , , , , , –, , , , –, –, , –, –, , , , , , , , ,  definiteness/specificity restriction on , , – double , , , , , , , , , –, , , , , , ,

, , , , , ,  emphatic function of , , , , , , , , , , , , , , , , , ,  loss of stylistic force , , , , ,  object clitic doubling , ,  restrictions on (in Swiss RR dialects) , – clitic left dislocation , , ,  clitic object pronouns  cliticization , , , , , , , ,  loss of  closest c-command , , , , , , , , ,  coalescence  Collins, Chris ,  Comanche , ,  comparatives , , , , , ,  complementizer agreement , , , , , , –, –, , , –, , –, , , , , , , , , , , ,  adjacency effect , , –, , , ,  parasitic on verbal agreement ,  replaces verbal agreement , , , , , , , , , , , , , , , , , ,  restrictions on , , , , , , , , ,  strict adjacency between inflected C and the subject (West Flemish) ,  Comrie, Bernhard , , , , , , , ,  concord ,  condensation  conditioned allomorphy  conjugational allomorphy ,  context-sensitive agreement, see agreement

JB[v.20020404] Prn:8/09/2005; 14:07

F: LA81IND.tex / p.4 (313-421)

 Index Cora ,  Corbett, Greville , –, , , ,  Cordin, Patrizia , , , ,  core functional categories , , , , ,  covert movement , , , ,  covert syntax ,  Cree ,  creole languages , ,  cue (in language acquisition) , , , , , , , , , ,  Culicover, Peter  cumulative exponence ,  Cupeño  Cutler, Anne ,  cyclicity ,  Cysouw, Michael , , , ,  D D , , , , –, , , , , , , , , , ,  DP , , , , , , , , , –, –, , , , , , –, , , , , , , ,  big DP , , , , –, , , ,  big DP, loss of , , , , , ,  Dagur , ,  Dal Negro, Silvia , , , , ,  Danish  dative, see Case de Haan, Germen ,  declarative , ,  Decurtins, Alexi  default agreement, see agreement defective intervention effect  Degree-0-Learnability  degree phrase  demonstrative , , ,  diffusion , , , ,  dissociated agreement morpheme , , , , , , –, , , ,

, , , , , , , , , , , , –, , , , , , , , , ,  insertion procedure , ,  parasitic on syntactic Agr-morpheme , , , , , ,  similarities with clitics  dissociated morpheme , , , , , , , , , , ,  Distributed Morphology (DM) , , , , , , , ,  Dixon, R. M. W. ,  do-support ,  Donohue, Mark , , , ,  double, see clitic doubling doublets, see morphological doublets Dresher, Elan  Dutch , , ,  E E-language  East Franconian  East Netherlandic ,  economy principles , , , , , , , , ,  Least Effort ,  Einheitsplural  elsewhere , , , , , , , , , –, , ,  Embick, David , , , , , , , ,  Emonds, Joseph ,  English , , , , , , , , , , , , , , , , , , , , , , ,  EPP (feature) , , , , , , , , , , , , ,  loss of an EPP feature , , , ,  equidistance  ergative, see Case Estonian  expletive , , 

Extension Condition  external argument ,  F f-morpheme, see functional morpheme Falk, Cecilia ,  Fanselow, Gisbert  Faroese  Fassa ,  feature checking , , , , ,  formal , , , , , ,  geometry , , ,  matching , –, , , , , , , ,  syncretism , ,  Fin , , ,  Finnish ,  Fintel, Kai von –, , , ,  Fiorentino  Fission –,  fixation  focus (Foc) , –, , , , , ,  FocP , , – Fodor, Janet  formal features, see feature Fox  French , , , ,  Friedemann, Marc-Ariel  Friedmann, Na’ama  Frisian –, , ,  Friulian , , , ,  Fukui, Naoki  full pronoun, see personal pronoun functional categories , –, –, , , , , , , , –, , , , , , ,  phonological deficiency of , , ,  universal hierarchy of ,  universal properties of ,  functional morpheme (f-morpheme) , , , , , , , , , , , 

JB[v.20020404] Prn:8/09/2005; 14:07

F: LA81IND.tex / p.5 (421-519)

Index  functionalist explanations , , ,  Furlan  Fusion , , ,  future, synthetic , , 

G Gabelentz, Georg von der ,  Gartner, Theodor ,  Gelderen, Elly van , , , ,  gender , , , , , , , , , , , , , , , ,  features , , , ,  gender agreement, see agreement genitive, see Case Georgian , , , , ,  Gerlach, Birgit , , , , , , , , , –,  German , , , , , , , , , –, , , , , , , , , , , , , , , , , , , , , , , , , , , , , ,  Germanic , , , , , , , , , , , , , , , , –, , , , , , , , , , , , , , , , , ,  Giazza ,  Gibson, Edward  Gilligan, Gary , ,  Givón, Talmy , , –, , , , , , , , ,  goal , , , , , ,  active , , ,  grammar change –, , ,  grammar competition , , , , ,  grammaticalization cline/path , , , , , ,  formal analysis of – of 1st and 2nd person agreement , , ,  of 3rd person agreement , , , , 

of agreement, complementary distribution of old and new marker ,  of agreement, pioneering role of 1st and 2nd person , , , , , , , , , , , ,  of agreement, syntactic conditions , , , , , , , , ,  of agreement prefixes , , , , , , , , , , , , , , , , , , , , , , ,  of agreement suffixes , , , –, , –, –, , ,  of inflectional morphology , ,  of object agreement ,  as transition from lexical to functional elements , , , , ,  unidirectional process  Greek  Greenberg, Joseph , , ,  Grewendorf, Günther , , , , , , , , , ,  Groat, Erich  Grodzinsky, Yosef  Guiraud, Pierre , 

H Haas, Mary  Haeberli, Eric , , , , , , ,  Haegeman, Liliane , , , ,  Haider, Hubert ,  Haiman, John , , , , , , , , –, , , , , , –, , –, , , ,  Hale, Kenneth , , , , , , ,  Hale, Mark , , , , , , ,  Hall, Christopher 

Halle, Morris , , –, , , , , , , , , , , , , , ,  Halpern, Aaron , ,  Harley, Heidi , , , , , , ,  Harnisch, Rüdiger ,  Harris, Alice , , , ,  Harris, Martin , , ,  Haspelmath, Martin  Hawkins, John , ,  head adjunction, see adjunction head complex , , , , , ,  head domain , ,  head movement , , , , , –, , , , , , ,  interarboreal  Head Movement Constraint (HMC) ,  Hebrew ,  Heine, Bernd  Hellendoorn , , –, ,  Hindi ,  Hoekstra, Eric , ,  Hoekstra, Jarich ,  Holmberg, Anders , ,  Hopper, Paul , , , , , ,  Huichol  Humboldt, Wilhelm von , 

I Iatridou, Sabine ,  Icelandic , , ,  Identification of feature content , –, , , , , ,  Identity Generalization  I-language  imperfect , , , , ,  Impoverishment rules , , , , –, , , ,  expansion of ,  loss of  incorporation ,  Indic 

JB[v.20020404] Prn:8/09/2005; 14:07

F: LA81IND.tex / p.6 (519-621)

 Index indicative , , , , , , , , –, –, , , ,  Indo-European , , ,  Infl , , , , , ,  Infl-to-C movement  IP , , , ,  Infl-oriented clitics, see clitics inflectional head , , , , , , ,  inflectional markers , , , , , , , , , , , , , , ,  linear ordering of –, –, –, , , , ,  non-attested sequences of ,  non-bound , ,  prefix/suffix distinction , , , , ,  selectional properties of , , , ,  inflectional morpheme, see functional morpheme inflectional morphology , , , , , , , , , , , , ,  acquisition of , , , , , ,  loss of ,  rich verbal inflection  inflectional paradigm , , ,  acquisition of ,  information structure ,  Innu-aimûn  interarboreal head movement, see head movement interface level , ,  interrogative , ,  intervention effect , , , ,  inversion , , , , , , , , , , , , , , , –, , , , –, , –, –, , , , , , –, , , , , , –, , –, , , , , , ,  Irish ,  Italian , , –, , , , , 

J Jacaltec  Jahai  Japanese  Jespersen’s cycle ,  Julien, Marit , , , , –, –, , , , , , , , , , , ,  K Kalmyk , , ,  Kaye, Jonathan  Kayne, Richard , , , –, , , , ,  Keenan, Edward  Kemenade, Ans van , , ,  Khalkha  Khinalug  Kim, Shin-Sook ,  King, Tracy Holloway ,  Kinyalolo, Kasangati  Kinyarwanda  Kiparsky, Paul , , , ,  Kipsigis  Kirby, Simon  Kisar ,  Kitahara, Hisatsugu  Kitja  Kollmer, Michael , , , ,  Koopman, Hilda , –, , ,  Korean  Kornfilt, Jaklin  Krifka, Manfred  Kroch, Anthony , , , , , , , , , ,  Kuen, Heinrich , , , , ,  Kurzverben , , , ,  L l-morpheme, see lexical morpheme Lühr, Rosemarie , ,  Laenzlinger, Christopher  Laka, Itziar  Lambrecht, Knud , , , ,  Lango 

language acquisition , –, , , , , , –, , , , , , –, , – logical problem of  language change , , –, , , , ,  abrupt , , , ,  gradual , , , , ,  logical problem of  language contact , ,  language loss ,  LaPolla, Randy  Lasnik, Howard  Late Insertion , , , , , , , ,  Late Linearization Hypothesis  Latin , ,  Least Effort, see economy principles left periphery , , , , , , , , , , , , ,  Legate, Julie  Lehmann, Christian , , , , –, , , ,  Lehmann, Winfred , – lexical category , , –, , –,  lexical morpheme (l-morpheme) , , , ,  Lexical Parametrization Hypothesis  Lexicon , , , , , , , , , ,  Encyclopedia  LF , , , ,  Lightfoot, David , –, , , , , , , ,  Linder, Karl Peter , , , –, , , , , , , –, , , –,  Linear Correspondence Axiom (LCA)  linearization of X0 -complexes , , , , –, , ,  of words  linear ordering of inflectional markers, see inflectional markers

JB[v.20020404] Prn:8/09/2005; 14:07

F: LA81IND.tex / p.7 (621-730)

Index

linguistic variation  Livonian  Local Dislocation  locality of syntactic operations –, , , , , , , , ,  of morphological operations ,  logical meaning , ,  Lombardo  long passive, see passive long-distance agreement (LDA) , , , , –, , , , , ,  Longobardi, Giuseppe ,  Lower Bavarian, see Bavarian Luiseño , , ,  Lusern – M Macushi , ,  Mainland Scandinavian, see Scandinavian Malagasy  Mandarin  Manzini, Rita , , ,  Marácz, Laszlo ,  Marantz, Alec , , , –, , , , , , , , , , , , , ,  Maricopa  Matthews, Peter ,  Mbay  McCloskey, James , ,  McFadden, Tom ,  Mchombo, Sam , ,  Meillet, Antoine ,  Merge , , , , , , , , , ,  Merge over Move  Meyer-Lübke, Wilhelm , , , ,  Middle English (ME) ,  minimal domain  Mirror Principle , , , , , , ,  Mitchell, Erica , , , , ,  Mithun, Marianne , , , ,  modal particle , , , , 

modal verb , ,  Moena  Moghol  Mon-Khmer  Mongolian , , , , , , , , , –, , , ,  Mono ,  mood , , , , , ,  morphological blocking effects , , , –, , –, , , , , , , , , , , , ,  morphological component , , , , , , ,  morphological doublets , , , –, ,  Morphological Economy , , ,  Morphological Merger , , , , , , , , , ,  morphological operations , , , , , , ,  Morphological Structure (MS) , , , , , –, , , , , , , , , ,   morphosyntactic features , , , , , , , , , , , , , , ,  Move –, , , , , ,  Move > Merge reanalysis  movement, loss of , , , , ,  Müller, Friedrich Max ,  Müller, Gereon  multiple Agree, see Agree multiple agreement, see agreement multiple exponence ,  multiple specifiers  N Nübling, Damaris , , , , , ,  Nakh-Dagestanian , ,  Native American languages , ,  Neeleman, Ad , –, , , 

Negation (Neg) –, , , , , , , , , , , , , , ,  negative concord  neutralization (of phonemic distinctions) ,  Newmeyer, Frederick  Nilotic ,  Niyogi, Partha  nominative, see Case Non-Standard French , , , , , –,  Northern Bavarian, see Bavarian Northern Italian dialects , , , , , , , , , , , , , , , , , , –, , , , , , ,  noun class , ,  Noyer, Rolf , , , , , , , , , , , , , , , , , , ,  NP-detachment –, , , ,  NP-movement ,  number , –, , , , , , , , , –, , , –, , , , , –,  features , , , ,  NumP  numeration ,  O object agreement, see agreement object movement , , –, ,  obligatorification  Oetzel, Annette –, ,  Oirat ,  Old English (OE) , , , , , ,  Old French  Old High German (OHG) , , , , , , , , ,  O’Neil, John  Optimality Theory ,  Ortmann, Albert ,  OSV  Ouhalla, Jamal , , , 



JB[v.20020404] Prn:8/09/2005; 14:07

F: LA81IND.tex / p.8 (730-812)

 Index OV , , , , , ,  OVS  P Papago  Papuan , , , , , ,  paradigmatization  paradigm leveling , , ,  parameter , , ,  change –, , ,  default setting ,  doublets  parametrization , , ,  Pari  Participant in Speech Event (PSE) , , –, , , , ,  participle , , , , –, , , , ,  participle agreement, see agreement Pasamaquoddy ,  Pashto ,  passive , , ,  impersonal passive  long passive (in German) – past tense , , , , ,  Paul, Hermann , , , , , ,  penultimate stress rule (in Swiss RR dialects) , , , , , , , , , , ,  percolation , ,  permutation invariance  person , , , , , , , , , , –, –, , ,  1st , , –, , , , , , , , , , , , , , , , , , –, , , , , , , , , –, ,  2nd , , –, , , , , , , , , , , , –, , , , –, , , , , , , –, , ,

, , , , , , –, , , , , , ,  3rd –, , , , , , , , , , , , , , , , , , –, –, , , , –, , , , , –, ,  binary feature system –,  features , , , , –, , , , , ,  PersP  personal pronoun (free form) , , , , , , , , , , –, , – extraposition of  weak , , , , , ,  Pesetsky, David , , ,  PF , , , , , , –, , ,  mapping to ,  PF feature checking , ,  PF movement  PF rules  Pfalz, Anton , , , ,  phase , , , , –, , , ,  Phase Impenetrability Condition (PIC) , ,  phi (φ)-features –, –, , , , , , , , –, , , –, , , , , , , , , , , , , , , ,  checking of  interpretable , , , , , , , , , , ,  marked for deletion ,  non-interpretable , , –, , , , , , , , ,  phonological erosion , , , , , , , , , , , , , , , , , , , , , 

Piattino , , –, , , ,  pied-piping , ,  Pinker, Steven  Pintzuk, Susan , ,  Platzack, Christer ,  Pochutla , ,  Poletto, Cecilia , , , , , , , , , , , , ,  Polinsky, Maria –, , , , –,  Pollock, Jean-Yves , , –, , ,  Pomattertitsch , , ,  Pontresina  Poppe, Nicholas , , , ,  portmanteau , , ,  possessive , , ,  Potsdam, Eric –, , ,  PP , , , , ,  prefix/suffix distinction, see inflectional markers prefixation ,  present tense , , , , ,  Preservation of argument structure , –, , , , , , ,  preterite-presents , ,  Primary Linguistic Data (PLD) , , , , , , , , ,  Principles and Parameters theory , , , , , , ,  pro –, , , , , , , , , , –, , , , , –, ,  referential , , , , ,  pro-drop , , , , , , , , , , , , , , , , , –, –, , –, , –, , , , , , , , , – pro-drop parameter  probe –, , ,  Prosodic Inversion  prosodic phrase , , ,  Proto-Indo-European , , 

JB[v.20020404] Prn:8/09/2005; 14:07

F: LA81IND.tex / p.9 (812-918)

Index  Pullum, Geoffrey , –,  Puter –, , , –, , ,  Q Quechua  quirky subject  R raising , –, ,  Ramat, Anna Giacalone  realizational model of grammar , , ,  reanalysis of clitics as agreement markers, see clitics Reh, Mechthild  relative pronoun , ,  remnant movement ,  Renzi, Lorenzo , , ,  restructuring , ,  resumptive pronoun , , , ,, ,  reanalysis of , ,  Rezac, Milan  Rhaeto-Romance , , , , , , , ,  Northern Italian dialects  Swiss dialects , , , , –, –, –, , , –, , , , , , , , , , ,  Richards, Norvin  Rima  Rimella  Ritter, Elizabeth , , ,  Rizzi, Luigi , , , , , , , , ,  Roana – Roberts, Ian , , , , , , –, –, , , , , , , , , , , , , , , , , , , , , , , , , , , , ,  Rohlfs, Gerhard , , ,  Rohrbacher, Bernhard ,  roll-up movement ,  Romance , , , , , , , , , , , , , , 

root , , , , , , , , , –, , ,  root-out insertion  Roussou, Anna , , , , , , , , –, , , , , , , , , , , , , , , , ,  Rowley, Anthony , ,  Russian – American Russian , , , –, ,  S Sabel, Joachim , ,  Sanskrit ,  Sapir, Edward , ,  Sauerland, Uli , ,  Savoia, Leonardo , ,  Scandinavian , , ,  Mainland Scandinavian ,  Schütze, Carson , , ,  Schulze, Wolfgang , , ,  Schweizer, Bruno –,  scrambling , ,  disrupting structural adjacency , – semantic content/features  loss of – Semitic ,  Sengoi  Separation Hypothesis  Serrano ,  Shlonsky, Ur , , , –, ,  short verbs, see Kurzverben Siewierska, Anna , , , , , , , , , , , , , , , ,  Sigurðsson, Halldór , ,  simplicity metric, see structural simplification Simpson, Andrew , , , –, , , , ,  Siouan  Skou , , , , ,  Slavic  sluicing  Smits, Caroline , ,  Sommer, Thomas , ,  Southern Paiute ,  SOV –, , , , , , , , , –, , , ,

–, , , –, ,  Spanish , , , , ,  Speas, Margaret , , , , , , ,  specifier-head configuration , , , , , , –, , , , ,  Spell-out , , , , , , , , , ,  at LF  Spencer, Andrew , ,  Spiess, Federico , ,  split ergativity  Sportiche, Dominique  Standard Arabic  Starke, Michal  Steele, Susan , , , , , – stem vowel alternation -, , , , ,  loss of  Stowell, Tim  Stray Affix Filter  structural adjacency , –, , , , , , , , , , , , , , , ,  structural simplification , , , –, , , , , , , , , , , , , , ,  simplicity metric , ,  Stump, Gregory ,  Suñer, Margarita , , ,  subject-initial clauses , , , –,  subjunctive , , , ,  Subset Principle in language acquisition , ,  governing Vocabulary Insertion , , , –,  suffixation , , ,  suffixing preference –, , , , , , , , , ,  suppletion  Surmeiran –, , , , , , , , ,  Surselvan , , , , , , 

JB[v.20020404] Prn:8/09/2005; 14:07

F: LA81IND.tex / p.10 (918-1016)

 Index Sutselvan –, , , –, , , , –, , , , ,  SVO –, , , , , , , , , , , , ,  Swabian  Swahili , , ,  Swedish  Swiss Rhaeto-Romance dialects, see Rhaeto-Romance syncretism , , ,  syntactic change , –, ,  Szemerényi, Oswald , 

T Tagalog  Tarahumara , – target grammar , , , , , , ,  Taylor, Ann , ,  Tense (T) , , , , , –, –, , –, , , , , , , , , , –, , –, , , , , , –, , ,  defective T  T+Agr , , , , , , ,  TP , , , , , , , , , , ,  TP fronting , , ,  SpecTP , , , , , , , , , , , , , , , , , , , , ,  Tepecano ,  Tepehuan  theme vowel , , ,  Theta (θ) position ,  role , , , , –, , , , , , , , , –, ,  theory , ,  Thráinsson, Höskuldur ,  Tibeto-Burman ,  topic , , , , , , , , , , , , , , , , 

definiteness/specificity restriction  TopP – topic-drop  topic left dislocation , , , , , , , ,  of 1st and 2nd person forms ,  loss of topic movement  Torrego, Esther , , , ,  Traugott, Elizabeth , , , , , ,  Travis, Lisa deMena , , , ,  trigger experience , , , , ,  Trips, Carola , ,  Tsakhur  Tsez , , , , , , ,  Tubatulabal ,  Turkana ,  Turkish ,  Tyroller, Hans ,  U Umlaut  unaccusative  underspecification , , , , , , , , , , –, ,  for agreement features , ,  for person , , , , , , , , , ,  for number , , , ,  Universal Grammar (UG) , , , , , , ,  Ura, Hiroyuki  Urdu  Uriagereka, Juan , , , ,  Uto-Aztecan , , , , , , , –, , , , ,  V V1 , , , , ,  V2 , , , , , , , –, , , , , , , , , ,

–, –, , , , , , –, , , , , , , , , , , , ,  asymmetric analysis of ,  embedded V2 ,  V3 , ,  Vallader –, , , , , , , , ,  Vanelli, Laura , , ,  Veneto , ,  verb movement , , , , , , , , , , , , , , , , , , , ,  loss of verb movement , ,  V-to-Infl movement  V-to-Infl movement, loss of  V-to-C movement , , , , ,  V-to-T movement , ,  verb stem , –, , , , –, , , , , ,  verb-initial languages , ,  Vicentino , –,  Vikner, Sten  VO , , , , , ,  Vocabulary Insertion , , , , , , , , , , ,  proceeds in a bottom-up fashion , ,  Vocabulary items –, , , , , , , , , –, , , , , , , , , , ,  Voice , , , , ,  VOS , ,  VP , , , , , , ,  ν , –, , , , –, , , , , , , , , , , , ,  νP , , , , , , , , , , , ,  SpecνP , , , , , , , , –, –, , , , , ,  VSO , , , , , 

JB[v.20020404] Prn:8/09/2005; 14:07

F: LA81IND.tex / p.11 (1016-1070)

Index 

W Walser German , –, , , , , , , , ,  Wambaya ,  Wartburg, Walther von , , , , ,  Watanabe, Akira ,  Weerman, Fred ,  Weiß, Helmut , , , , , , ,  Welsh , ,  West Flemish –, , , , ,  Wexler, Kenneth , , ,  wh-agreement  wh-movement  wh-phrase , ,  wh-questions ,  Widmer, P. Ambros , , , , , 

Wiesinger, Peter , , , –,  Word building constraint , –, , –, , , , , ,  word formation –, , , , ,  Word Formation Rules  word order , , , , , –, , , , , , , , –, , ,  basic , ,  free  word order change  from OV to VO ,  from SVO to SOV  Wu, Zoe , , , , , , ,  Wurmbrand, Susi , , , –

X X-bar theory 

Y Yaqui , , – Yiddish , , ,  Yuman , 

Z Zanuttini, Raffaella  Zeller, Jochen , ,  Zeneyze  Zimmermann, Ede , ,  Zurich German  Zwart, Jan-Wouter , –, , , , , , ,  Zwicky, Arnold , –, 

The Rise of Agreement: A Formal Approach to the Syntax And Grammaticalization of Verbal Inflection (Linguistik Aktuell Linguistics Today)

The Emergence of Order in Syntax (Linguistik Aktuell Linguistics Today)

Grammaticalization As Economy (Linguistik Aktuell Linguistics Today, LA 71)

Balkan Syntax and Semantics (Linguistik Aktuell Linguistics Today)

Agreement Systems (Linguistik Aktuell Linguistics Today, Volume 92)

Advances in Comparative Germanic Syntax (Linguistik Aktuell Linguistics Today)

Experimental Pragmatics Semantics (Linguistik Aktuell Linguistics Today)

Multiple Wh-Fronting (Linguistik Aktuell Linguistics Today)

The Bantu-Romance Connection: A comparative investigation of verbal agreement, DPs, and information structure (Linguistik Aktuell Linguistics Today)

Parentheticals (Linguistik Aktuell Linguistics Today, 106)

Argument Structure (Linguistik Aktuell Linguistics Today)

The Derivation of Anaphoric Relations (Linguistik Aktuell Linguistics Today)

The Copy Theory of Movement (Linguistik Aktuell Linguistics Today)

The Structure of Stative Verbs (Linguistik Aktuell Linguistics Today)

The Grammar of Focus (Linguistik Aktuell Linguistics Today)

The Grammar of Focus (Linguistik Aktuell Linguistics Today)

Phase Theory (Linguistik Aktuell Linguistics Today)

Causatives in Minimalism (Linguistik Aktuell Linguistics Today)

Cyclical Change (Linguistik Aktuell Linguistics Today)

The Syntax of Jamaican Creole: A Cartographic Perspective (Linguistik Aktuell Linguistics Today)

Syntatic Aspects of Topic and Comment (Linguistik Aktuell Linguistics Today)

Adjectives: Formal Analyses in Syntax and Semantics (Linguistik Aktuell Linguistics Today)

The Linguistics Enterprise: From knowledge of language to knowledge in linguistics (Linguistik Aktuell Linguistics Today)

The Minimalist Syntax of Defective Domains: Gerunds and Infinitives (Linguistik Aktuell Linguistics Today)

Verbal Complement Clauses: A Minimalist Study of Direct Perception Constructions (Linguistik Aktuell Linguistics Today

Adverbials and the Phase Model (Linguistik Aktuell Linguistics Today, 177)

Scrambling and the Survive Principle (Linguistik Aktuell Linguistics Today)

The Grammar of Repetition: Nupe grammar at the syntax-phonology interface (Linguistik Aktuell Linguistics Today)

Movement Theory of Control (Linguistik Aktuell Linguistics Today)

Beyond Coherence: The syntax of opacity in German (Linguistik Aktuell Linguistics Today)

Current Issues in Generative Hebrew Linguistics (Linguistik Aktuell Linguistics Today)

The Rise of Agreement: A Formal Approach to the Syntax And Grammaticalization of Verbal Inflection (Linguistik Aktuell Linguistics Today)

The Emergence of Order in Syntax (Linguistik Aktuell Linguistics Today)

Grammaticalization As Economy (Linguistik Aktuell Linguistics Today, LA 71)

Balkan Syntax and Semantics (Linguistik Aktuell Linguistics Today)

Agreement Systems (Linguistik Aktuell Linguistics Today, Volume 92)

Advances in Comparative Germanic Syntax (Linguistik Aktuell Linguistics Today)

Experimental Pragmatics Semantics (Linguistik Aktuell Linguistics Today)

Multiple Wh-Fronting (Linguistik Aktuell Linguistics Today)

The Bantu-Romance Connection: A comparative investigation of verbal agreement, DPs, and information structure (Linguistik Aktuell Linguistics Today)

Parentheticals (Linguistik Aktuell Linguistics Today, 106)

Argument Structure (Linguistik Aktuell Linguistics Today)

The Derivation of Anaphoric Relations (Linguistik Aktuell Linguistics Today)

The Copy Theory of Movement (Linguistik Aktuell Linguistics Today)

The Structure of Stative Verbs (Linguistik Aktuell Linguistics Today)

The Grammar of Focus (Linguistik Aktuell Linguistics Today)

The Grammar of Focus (Linguistik Aktuell Linguistics Today)

Phase Theory (Linguistik Aktuell Linguistics Today)

Causatives in Minimalism (Linguistik Aktuell Linguistics Today)

Cyclical Change (Linguistik Aktuell Linguistics Today)

The Syntax of Jamaican Creole: A Cartographic Perspective (Linguistik Aktuell Linguistics Today)

Syntatic Aspects of Topic and Comment (Linguistik Aktuell Linguistics Today)

Adjectives: Formal Analyses in Syntax and Semantics (Linguistik Aktuell Linguistics Today)

The Linguistics Enterprise: From knowledge of language to knowledge in linguistics (Linguistik Aktuell Linguistics Today)

The Minimalist Syntax of Defective Domains: Gerunds and Infinitives (Linguistik Aktuell Linguistics Today)

Verbal Complement Clauses: A Minimalist Study of Direct Perception Constructions (Linguistik Aktuell Linguistics Today

Adverbials and the Phase Model (Linguistik Aktuell Linguistics Today, 177)

Scrambling and the Survive Principle (Linguistik Aktuell Linguistics Today)

The Grammar of Repetition: Nupe grammar at the syntax-phonology interface (Linguistik Aktuell Linguistics Today)

Movement Theory of Control (Linguistik Aktuell Linguistics Today)

Beyond Coherence: The syntax of opacity in German (Linguistik Aktuell Linguistics Today)

Current Issues in Generative Hebrew Linguistics (Linguistik Aktuell Linguistics Today)

Recommend Documents