Clause Structure and Language Change

Clause Structure and Language Change

Oxford Studies in Comparative Syntax Richard Kayne, General Editor

Principles and Parameters of Syntactic Saturation Gert Webelhuth Verb Movement and Expletive Subjects in the Germanic Languages Sten Vikner Parameters and Functional Heads: Essays in Comparative Syntax Edited by Adriana Belletti and Luigi Rizzi Discourse Configurational Languages Edited by Katalin E. Kiss Clause Structure and Language Change Edited by Adrian Battye and Ian Roberts

Clause Structure and Language Change

New York Oxford OXFORD UNIVERSITY PRESS 1995

Oxford University Press Oxford New York Toronto Delhi Bombay Calcutta Madras Karachi Kuala Lumpur Singapore Hong Kong Tokyo Nairobi Dar es Salaam Cape Town Melbourne Auckland Madrid and associated companies in Berlin Ibadan

Copyright (c) 1995 by Oxford University Press, Inc. Published by Oxford University Press, Inc. 200 Madison Avenue, New York, New York 10016 Oxford is a registered trademark of Oxford University Press, Inc. All rights reserved. No part of this publication may be reproduced, stored in a retrieval system, or transmitted, in any form or by any means, electronic, mechanical, photocopying, recording or otherwise, without the prior permission of Oxford University Press. Library of Congress Cataloging-in-Publication Data Clause structure and language change / edited by Adrian Battye, Ian Roberts. p. cm. — (Oxford studies in comparative syntax) Collection of papers based on material presented at the 1 st Generative Diachronic Syntax Conference which was held at the University of York, Apr. 1990. Includes bibliographical references. Contents: Why UG needs a learning theory : triggering verb movement / David Lightfoot — Two types of verb second in the history of Yiddish / Beatrice Santorini — The locus of verb movement in non-asymmetric verb-second languages : the case of Middle French / Monique Lemieux and Fernande Dupuis — Evidence for a verb-second phase in Old Portuguese / Ilza Ribeiro — Indo-European origins of Germanic syntax / Paul Kiparsky — On the decline of verb movement to comp in Old and Middle French / Barbara Vance — The loss of verb second in English and French / Christer Platzack — Verb second, prodrop, functional projections, and language change / Aafke Hulk and Ans van Kemenade — Null subjects in verb-first embedded clauses in Philippe de Vigneulles' Cent nouvelles nouvelles / Paul Hirschbuhler — The diachronic development of subject clitics in northeastern Italian dialects / Cecilia Poletto — Complement clitics in medieval Romance: the Tobler-Mussafia law / Paola Beninca — Cases of verb third in Old High German / Alessandra Tomaselli. ISBN 0-19-508632-5. — ISBN 0-19-508633-3 (pbk.) 1. Grammar, Comparative and general—Clauses—Congresses. 2. Grammar, Comparative and general—Syntax—Congresses. 3. Linguistic change—Congresses. 4. Principles and parameters (Linguistics)—Congresses. 5. Grammar, Comparative and general—Verb—Congresses. 6. Grammar, Comparative and general—Clitics— Congresses. I. Battye, Adrian. II. Roberts, Ian G. III. Generative Diachronic Syntax Conference (1st: 1990 : University of York) IV. series. P297.C55 1995 415—dc20 94-34278

246897531 Printed in the United States of America on acid-free paper

Preface

This collection of papers developed from material presented at the First Generative Diachronic Syntax Conference, held at the University of York in April 1990. The idea for that meeting was Adrian Battye's. Since then, the Conference has developed into a series, with a second meeting at the University of Pennsylvania in November 1992 and a third one currently planned for the Free University of Amsterdam in March 1994. Adrian Battye passed away in March 1993. Until a short time before his death, we collaborated in editing this collection. With the exception of one final revision of the Introduction, and the choice of the running order of the papers, the book is our joint work. I would like to dedicate this book to Adrian's memory. Ian Roberts

Bangor, August 1993

This page intentionally left blank

Contents

1. Adrian Battye and Ian Roberts: Introduction, 3 PART ONE: The Diachrony of Verb Second, 29 2. David Lightfoot: "Why UG Needs a Learning Theory: Triggering Verb Movement", 31 3. Beatrice Santorini: "Two Types of Verb Second in the History of Yiddish", 53 4. Monique Lemieux and Fernande Dupuis: "The Locus of Verb Movement in Non-Asymmetric Verb-Second Languages: The Case of Middle French", 80 5. Ilza Ribeiro: "Evidence for a Verb-Second Phase in Old Portuguese", 110 6. Paul Kiparsky: "Indo-European Origins of Germanic Syntax", 140 PART TWO: Verb Second and the Null-Subject Parameter, 171 7. Barbara Vance: "On the Decline of Verb Movement to Comp in Old and Middle French", 173

vui

CONTENTS

8. Christer Platzack: "The Loss of Verb Second in English and French", 200 9. Aafke Hulk and Ans van Kemenade: "Verb Second, Pro-drop, Functional Projections and Language Change", 227 10. Paul Hirschbiihler: "Null Subjects in Verb-First Embedded Clauses in Philippe de Vigneulles' Cent Nouvelles Nouvelles", 257 PART THREE: Clitics and Verb Second, 293 11. Cecilia Poletto: "The Diachronic Development of Subject Clitics in North Eastern Italian Dialects", 295 12. Paola Beninca: "Complement Clitics in Medieval Romance: the Tobler-Mussafia Law", 325 13. Alessandra Tomaselli: "Cases of Verb Third in Old High German", 345

Contributors

Professor Paola Beninca, State University of Milan Professor Fernande Dupuis, University of Quebec, Montreal Professor Paul Hirschbuhler, University of Ottawa Professor Aafke Hulk, Free University of Amsterdam Professor Ans van Kemenade, Free University of Amsterdam Professor Paul Kiparsky, Stanford University Professor Monique Lemieux, University of Quebec, Montreal Professor David Lightfoot, University of Maryland at College Park Professor Christer Platzack, University of Lund Doctor Cecilia Poletto, Universities of Venice and Padua Ilza Ribeiro, State University of Sao Paulo, Campinas Professor Beatrice Santorini, Northwestern University Doctor Alessandra Tomaselli, University of Pavia Professor Barbara Vance, Indiana University

This page intentionally left blank

Clause Structure and Language Change

This page intentionally left blank

1 Introduction Adrian Battye and Ian Roberts

Since the nineteenth century beginnings of linguistics as a scientific discipline, it has been recognised that comparative and historical linguistics are mutually reinforcing (in fact, synchronic and diachronic linguistics were not rigourously distinguished prior to de Saussure (1916)). The development of the "principles and parameters" approach to syntax in the wake of Chomsky (1981) and related work (see in particular Jaeggli (1982), Kayne (1984) and Rizzi (1982) for groundbreaking work on the Romance languages) has provided a solid foundation for the development of a rich and insightful approach to comparative syntax, with the result that during the mid-to-late 1980s there was a veritable explosion of work on comparative syntax, covering many of the world's major language families. In the context of these developments, it is natural that interest in diachronic syntax should be sparked off anew. The papers in this volume are the first fruit of this renewed interest in diachronic syntax, all proposing analyses of diachronic phenomena that are framed in terms of the principles and parameters approach. In this introduction, we will try to provide a backdrop for the papers. First, we discuss the general issue of how questions about historical and diachronic syntax can be framed in terms of the principles and parameters model and we try to indicate what the particular theoretical interest of such questions might be. Second, we provide some of the specific technical background regarding current theories of verb placement, since this is the central unifying theme for the papers here. Third, we briefly summarise the papers in the volume, attempting to bring out the conceptual and empirical connections among them. 3

4

INTRODUCTION

1. Diachronic Syntax and the "Principles and Parameters" Approach Before discussing the motivations for studying diachronic syntax in terms of the principles and parameters approach, let us very briefly describe what that approach involves. In the present context, we cannot give a full introduction; for a full discussion and illustration of the concept of a parameter of Universal Grammar, see the introduction to Jaeggli and Safir (1989), and for a general introduction to current grammatical theory, see Haegeman (1991). The notion of Universal Grammar in the context of modern linguistics was originated by Chomsky, and the reader is referred to Chomsky's own writings (in particular Chomsky (1975, 1980, 1986a)) for a fuller discussion and defence of this idea. Universal Grammar (UG) is taken to consist of an invariant core of constitutive principles, common to all the world's languages, and to all possible human languages. Because it is difficult to see how such principles can be arrived at on the basis of primary linguistic experience, Chomsky maintains that these principles are innately given, i.e. that they form part of the genetic endowment of every human being. In order to account for the attested variation among the world's languages, Chomsky proposes that these principles may be associated with parameters of variation which make it possible for a principle to be realised in different ways in different languages. Since the principles of UG are fairly abstract in nature, a minimal difference in the value of an associated parameter in two different languages may—and frequently does—give rise to dramatic surface differences in the well-formed sentences of the languages. In this way, the principles and parameters approach is able to account for differences among languages while maintaining the idea that all languages are cut from the same cloth, whose nature is determined by the innately-given principles of UG. Moreover, the differences among languages that the parameters encode may be structured, to the extent that the principles and parameters given by UG are structured. This gives rise to the possibility of parametrically-driven typologies; these typologies may overlap empirically with the more superficial kind of typological work originated by Greenberg (1963). Taking our cue from this last point, we can illustrate the interaction of principles and parameters from a simple example of word-order variation. The standard assumption regarding the base position of direct objects since Chomsky (1965) is that direct objects must be sisters of V. Suppose that this is a principle of UG (or a direct consequence of some principle of UG). Now, it is well known that languages vary according to the basic order of direct object and verb: there are languages in which the basic order is object-verb (OV languages like Turkish, Japanese, Latin, German, etc.), and languages where the basic order is verb-object (VO languages like English, the Romance languages, Bantu languages, many Papuan languages, etc.). The order of direct object and verb is then a locus of parametric variation. The attested crosslinguistic variation and the requirement on the structural relation between direct object and verb can be reconciled in the principles and parameters model, since

INTRODUCTION

5

the hierarchic structural relation between verb and object is a principle of UG, while the variation in linear order is stated as a parameter. Schematically, we can summarise the situation as in (1) (this mode of presentation is borrowed from Rizzi (1988), although Rizzi uses it in a slightly different way): (1) a. VP -> {V, NP,...} (principle of UG)

b. V precedes/follows NP

(parameter of UG)

Every natural language conforms to principle (la). We have written the material on the right of the arrow in set notation so as to indicate that the relative order of V and NP is not determined by this principle, only the hierarchical structure of (part of) VP. The parametric choice is given in (Ib): UG allows, in fact requires, individual languages to choose either "precedes" or "follows" here. Each choice is a different value for the parameter. It is clear that this choice, although extremely simple and easy to state in itself, has far-reaching consequences for the surface orders each language allows. Of course, the "principle" in (la) is not a real principle of UG. The statement given there is an instance of a more general schema for the internal structure of syntactic constituents given by X-bar theory. One of the most important aspects of X-bar theory is the fact that all syntactic constituents are assumed to have the same internal structure. So, a closer approximation to the relevant UG principle is given by (2), where X is a variable taking syntactic categories as values: (2)

a. XP -> { X', YP } b. X' -> { X°, ZP }

Here YP is the specifier of X' and ZP is the complement of X°. Whether these categories are present in a given representation depends on other considerations that we cannot go into here. Still supposing the parameter to be (Ib), this restatement of (la) leads us to expect that head-complement relations involving heads of different categories are "harmonic" in the sense that heads consistently precede or consistently follow their complements. Exactly this observation formed the basis for the typological work initiated by Greenberg (see Hawkins (1983) for a refinement). Here we see how the principles and parameters approach can overlap with, and in fact supersede, the earlier, more superficial typological studies. This outline of the basic ideas of the principles and parameters approach to cross-linguistic variation and typology, although extremely sketchy and simplified, is enough for our purposes here. We now have a notion of how current theory accounts for syntactic differences between languages. The principles and parameters approach was designed to account for synchronic variation, and, as we mentioned at the outset, it has been very successful in this. It should be immediately clear how the approach can extend from the synchronic to the diachronic dimension: just as we try to account for syntactic differences among, say, the contemporary Romance languages in terms of differing parametric values, so we want to account for differences between Latin and the contemporary Romance languages in terms of differing para-

6

INTRODUCTION

metric values. This naturally implies parametric change over time. The agenda for the study of diachronic syntax is thus set: we can analyse the historical development of a given language (or language family) in terms of differing parametric values at different times. Syntactic change is thus seen as parameter change. This approach has the immediate consequence that studying historically attested (but no longer spoken) etats de langue enlarges the data base available for comparison. Just as synchronic work on languages not previously studied from the principles and parameters perspective can bring to light new arrays of data which require the postulation either of new parameters or of new combinations of settings of existing parameters, so work on earlier etats de langue may yield the same result. A case in point is the important work on Old French in Adams (1987a,b). Adams showed that certain observations about Old French word order that were made in earlier philological studies (notably in Thurneysen (1892) and in Foulet (1919)) could be elegantly captured by saying that this language is both a verb-second language and a null-subject language, with the added restriction that null subjects are only possible in verb-second clauses (i.e. matrix clauses and a restricted class of embedded clauses). This kind of analysis of Old French word order is discussed, and in some cases questioned, in several of the articles included in this volume—see SECTION 3 for introductory comments, and the references given there. It seems that this particular combination of parameter settings (which we could schematize rather simplistically as +V2, +null-subject) is rare among contemporary languages (but cf. the remarks on Romansch at the end of Vance's contribution—Chapter 7). If one takes the terms "synchronic" and "diachronic" in their strict, Saussurean sense, we have up to now been discussing synchronic historical syntax rather than genuine diachronic syntax. "True" diachronic syntax is the study of how one etat de langue evolves into another, while the above remarks have been concerned largely with past etats de langue viewed synchronically. A properly diachronic analysis of Old French, then, would take into account the development of this +V2, +null-subject system into Middle and perhaps Modern French, or how Old French developed from Latin. The study of how grammatical systems change over time raises a range of questions which do not arise in purely synchronic work: Can we construct a typology of changes? Is there a theory of language change independently of linguistic theory generally? Is there a tendency for languages to "drift" towards particular kinds of systems (cf. Sapir (1921))? What are the relations between language change and other factors (extralinguistic events, socio- and psycholinguistic phenomena, language contact, etc.)? What are the relations between language change in different components of the grammar: phonology, syntax, etc.? These are all traditional questions of diachronic linguistics, and they illustrate the extent to which the study of language change can interact with other areas of linguistics, and possibly with other disciplines. Any answers we may devise to the questions given above will depend upon our answers to this question: What are the mechanisms of parametric change?

INTRODUCTION

7

The papers in this volume are primarily concerned with parametric changes, and this is clearly the central question for diachronic syntax in terms of the principles and parameters approach. It is probable, in fact, that more is at stake in fully understanding the nature of parametric change; following Lightfoot (1979, 1991) and Clark and Roberts (1993), we believe that the study of diachronic syntax, because it gives us an insight into the mechanisms of parameter change, can tell us something about parameter setting, i.e. about language acquisition. This is the point that we wish to pursue in the remainder of this section. It is a long-standing idea, probably due originally to Hermann Paul (see Paul (1920)), that grammatical systems can be changed through the process of language acquisition. To put the idea rather crudely and simplistically, a new generation of acquirers may, as it were, "misacquire" the parental system in the sense that they may restructure the adult system in various subtle and presumably imperceptible ways. In the course of time, the "misacquired" system becomes the parental system for a new generation of acquirers who in their turn may indulge in further "misacquisition." In this way, the system alters over time. Framing the question of parametric change in terms of aspects of language acquisition in this way puts diachronic linguistics in the centre of generative theoretical endeavour, given the fundamental importance of language acquisition to generative theory. In fact, the mechanisms of parameter changing can be identified with the mechanisms of parameter setting, which are held to play a fundamental role in the acquisition of syntax (see Hyams (1986); Pierce (1989); Pierce and Duprez (1990); and the papers in Roeper and Williams (1987)). This point emerges more clearly if we adopt the distinction between I(nternal)-language and E(xternal)-language made in Chomsky (1986a: 19-24). An E-language is a collection of actual or potential linguistic objects associated with some population. In these terms, Bloomfield's (1933) notion of language as a collection of speech events corresponds to an E-language, as do the variations on the notion of "set of well-formed formulae", "set of soundmeaning pairs" etc. that are inspired by work in analytic philosophy (see, for example, Quine (1960); Montague (1974)). Also, most traditional work in historical linguistics and philology is work on E-language. The important point that such otherwise quite different approaches to the study of language have in common, in Chomsky's view, is that the analysis of a language as an Elanguage is quite independent in principle of any properties that may be ascribed to the mind/brain of native speakers of that language. An I-language, on the other hand, is "some element of the mind of the person who knows the language, acquired by the learner and used by the speaker/ hearer" (Chomsky (1986a:22)). Clearly, then, generative grammar is primarily concerned with I-language. Chomsky goes on to make the following point: The I-language L may be the one used by a speaker but not the I-language L', even if the two generate the same class of expressions ...; L' may not even be a possible human I-language, one attainable by the language faculty (ibid:23).

8

INTRODUCTION

Thus different I-languages may in principle generate identical E-languages. Chomsky makes this point for philosophical reasons which need not concern us directly here, but we will see below that a slightly weaker version of this point is relevant for the connection between parameter setting and parameter change. In terms of this distinction, language acquisition, i.e. parameter setting, proceeds on the basis of two things. On the one hand, there is Universal Grammar "construed as the theory of human I-languages, a system of conditions deriving from the human biological endowment that identifies the I-languages that are humanly accessible under normal conditions" (ibid). On the other hand, there is a finite corpus of utterances in the child's environment: the primary linguistic data of acquisition. These are E-language tokens. Most importantly, the child has no direct access to the I-language underlying the Elanguage tokens that make up the primary linguistic data (Lightfoot (1979) also insists on the importance of this point). We can illustrate the situation by adapting the following schema from Andersen (1973) (although Andersen does not use the terms E-language and I-language): (3) Parents' I-language Child's I-langua ge

Parents' E-language Child's E-language

UG mediates the diagonal link between the parental E-language and the child's I-language (for a specific proposal regarding how this happens, see Clark (1990), and for an application of Clark's approach to parametric change, see Clark and Roberts (1993)). As we said above, there can be no arrow linking the parents' I-language with the child's I-language in (3) (for the obvious reason that no human being can have direct access to the contents of another human being's mind/ brain). Moreover, given Chomsky's assertion that different I-languages may in principle generate identical E-languages, it is possible that the child's Ilanguage representation for some E-language utterance U may differ from the parents' I-language representation for U. Of course, the different representations must both fall within the confines defined by UG (unlike the case imagined by Chomsky in the above quotation; as we already mentioned, Chomsky brings up this possibility for purely philosophical purposes). Although UG must for general reasons admit only a small class of I-languages, we can certainly allow that different I-languages in this class may generate identical strings. Thus there may be partial, and indeed close to total, overlap between E-languages produced by different I-languages. Under the conditions of language acquisition, this partial overlap may in effect be imperceptibly different from total overlap. Hence it is possible for the child to acquire a different I-language from that which underlies the parents' E-language in the schema in (3). The result of this will be that the child's E-language is subtly, almost imperceptibly, different from the parents'. When the child's E-language comes to serve as the basis for acquistion by a further generation, these differences may be amplified.

INTRODUCTION

9

I-languages are collections of parameter settings. Hence the above discussion can be seen as an abstract account of how parameters may change over time, and of how the process of parameter change is, from the point of view of a single generation, the same as the process of parameter setting. Of course, the above discussion avoids the crucial question of why acquirers may opt for some parameter setting which differs from the parents' I-language setting. In principle, this can arise in one of two ways: either through some inherent "opacity" in the primary linguistic data which favours a novel analysis and parameter setting on the part of the acquirer, or through some inherent preference on the part of the acquirer for one parameter setting over another. These two ways of choosing new parameter settings are by no means exclusive; indeed, it is likely that they very frequently coincide. The idea that "opacity" of the trigger experience can lead to a parameter change has its roots in traditional ideas about language change, and has been discussed in the generative framework above all in Lightfoot (1979). In Roberts (1992), it is suggested that this notion of "opacity" should be understood in terms of a notion of "least effort": acquirers assign the smallest structure possible to the strings with which they are presented, where the constraints on "smallness" are given in part by UG and in part by the trigger experience. In that case, many strings will be assigned an excessively "small" structure, and, under the appropriate conditions, some of these may survive into the steady-state grammar that is the end-point of acquisition (what the "appropriate conditions" are is to some extent an open question, which it is difficult to discuss in abstraction from specific cases). This approach can explain many cases of what has been referred in the typological literature in diachronic syntax as "grammaticalization"; for a variant of this idea see the account of the development of English auxiliaries in Roberts (1985). The idea that opacity is related to some notion of "least effort" derives from the assumption that acquirers are conservative, an assumption we take to be minimal. It should be stressed that the notion of "least effort" involved is a strategy of acquisition rather than a principle or guideline of UG; it is therefore conceptually distinct from the economy principles of Chomsky (1991). The notion of "opacity" involves an assumption of conservatism on the part of learners at the level of the structural analysis of given strings. At the level of parameters, the assumption of conservatism underlies the well-known Subset Principle, originally proposed by Berwick (1985). Loosely, the Subset Principle states that acquirers will always opt for the parameter setting which gives the "smallest" grammar, i.e. the smallest set of grammatical sentences. For example, null-subject languages can be thought of (possibly wrongly, but the example suffices to make the point) as allowing more grammatical sentences than non-null-subject languages since they frequently allow a choice between an overt and a null pronoun. Hence, null-subject languages are a superset of non-null-subject languages. Hence, acquirers "prefer" non-null-subject languages and need positive evidence to set the null-subject parameter positively. Something like the Subset Principle is clearly necessary, since negative evidence is not available to acquirers. If no such principle is assumed, then we

10

INTRODUCTION

have no way of explaining why acquirers would retreat from superset grammars, since they would have no positive evidence that their parametric choice did not correspond to the adult grammar. In the context of language change, acquirers may "wrongly" choose a subset grammar, and never receive sufficiently robust positive evidence that the parental grammar in fact has the superset value. Clark (1990) has shown that the Subset Principle is relevant for systems of parameters, even where two parameters may give rise to an intersection (rather than an inclusion) relation among the associated grammars. Clark refers to this situation as a "shifting" relationship among parameters. Abstractly, we can illustrate it as follows: suppose we have two parameters P 1 and P2 each of which individually gives rise to an intersecting grammar. However, where both parameters have the positive value, a grammar arises which is a superset of either of those which emerge when only one parameter has the positive value. The point can be illustrated by the following diagrams, where the shaded area indicates the set of grammatical sentences: (4)

a. P 1 positive, P2 negative:

b. P1 negative, P2 positive:

c. Both parameters positive: In a shifted system, the Subset Principle may lead to a change in the value of one of the intersecting parameters—see Clark and Roberts (1993) for a case study of exactly this kind. The conclusion is that, other things being equal, shifted systems are disfavoured, or "marked," ultimately because of the properties of the acquirer and the process of acquisition. Of course, given the enormous complexity of grammatical systems, it is frequently the case that other things are not precisely equal. Nevertheless, the result is interesting, and indicates how diachronic work can be important in developing the theory of parameters and of parameter setting. Our purpose in the foregoing remarks has been to sketch the conceptual background to the papers in this volume, and to indicate what we see as the implications and the potential importance of work of the type reported here. It should be clear that contemporary linguistic theory is in a position to provide genuinely new insights into a range of both novel and traditional questions. To illustrate this last assertion, consider the kinds of answers we may give to the traditional questions about language change that we raised earlier. First, can we construct a typology of changes? The answer to this question is that all parametric changes must convert one possible collection of parameter val-

INTRODUCTION

11

ues into another; in this very general sense, the answer to the question is positive. However, it should also be clear that the typology of changes will be indistinct from the typology of possible languages in general. Is there a theory of language change independently of linguistic theory more generally? If the core notion of the theory of language change is that of parameter change, and that notion is not distinct from the notion of parameter setting, then we see that the answer to this question is negative. Linguistic theory as constructed to account for the attested synchronic variation and the facts of language acquisition gives us all the conceptual tools we need (this point has also been argued by Lightfoot (1991)). Is there a tendency for languages to "drift" in Sapir's sense? Parameter theory again suggests that the answer to this question is negative. Language change is essentially a random "walk" through the space of possible combinations of parameter settings. However, at the level of individual constructions there may be more to say. The kinds of reanalyses of "opaque," or excessively effortful, input that may trigger parametric changes by altering the nature of the trigger experience will, if Roberts (1992) is right, always simplify limited classes of structures. We suggested above that this kind of change underlies the descriptive notion of "grammaticalization," which is certainly a very frequently attested kind of change. This kind of reanalysis is, however, distinct from parameter change; reanalysis simplifies given structures while parameter change alters whole systems. Note that we cannot view parameter change as simplifying or as complicating grammatical systems; this would be as misleading as claiming that synchronically attested languages differ in overall complexity. Finally, on the question of the relations between language change and other aspects of language (whether intra- or extralinguistic), we have nothing to add to whatever general view may be taken on how parameter setting is influenced by these factors. Here again, questions of language change reduce to more general questions about language acquisition. Here, as above, we see how formulating the agenda for diachronic syntax in terms of parameter theory holds out the prospect of understanding traditional questions of diachronic linguistics in terms of general aspects of linguistic theory.

2. Verb Syntax 2.1 The X-bar Structure of the Clause Most of the work contained in this volume takes its fundamental impetus from the recent upsurge of research within Government-Binding theory on verb syntax and, in particular, verb movement. The antecedents of the present work on verb movement can be traced back to Chomsky (1957), Klima (1964) and Emonds (1978). The current interest in verb movement in relation to a system of non-lexical categories carrying essentially "morphological" information—the functional categories—was stimulated initially by Travis (1984), and

12

INTRODUCTION

this approach was further developed by Chomsky (1986b) and Baker (1985b, 1988). In SECTION 1 we introduced the X-bar schema for the internal structure of syntactic categories. We repeat it here for convenience: (5)

a. XP -» { X', YP} b. X' -» { X , Z P }

Originally this schema was taken to apply only to the lexical categories N, V, A and P. It was assumed that the "clausal" categories S and S' were generated by the following PS-rules: (6)

a. S' -> COMPS b. S -> NP Aux VP

This was clearly an undesirable state of affairs given the general programme, initiated by Chomsky (1981) and Stowell (1981), for the elimination of independent PS-rules from the theory. Accordingly, Chomsky (1986b) proposed that S and S' are projections of the non-lexical categories I(nfl) and C(omp), respectively. I corresponds roughly to the Aux node of earlier work, and contains in particular, features specifying tense and agreement; this position had originally been argued to be the head of S in Hale, Jeanne and Platero (1977). In the system of Chomsky (1986b), the specifier of I is the subject position, and the complement of I is VP. C is the earlier Comp position; its specifier is the landing site for wft-movement and (in verb-second languages-see below) fronted topics, and its complement is IP. So we have the following clause structure:

This structure is the basis for nearly all the papers in this volume, although some of them elaborate it in various ways.

INTRODUCTION

13

2.2 V-to-I Movement Subsequent to Chomsky (1986b), the most important development for the theory of verb movement came from Pollock (1989). Pollock's starting point was the earlier work by Emonds on English and French. Emonds observed the following contrasts in the positioning of tensed main verbs in French and in English: (8) a. French verbs precede VP-adverbs, English ones do not: Jean embrasse souvent Marie. *John kisses often Mary. b. French verbs precede floated quantifiers, English ones do not: Les enfants aitnent tous le chocolat. *Children like all chocolate, (where all quantifies children) c. French verbs precede clausal negation, English ones do not: Jean n'aime pas Marie. *John likes not Mary. Taking as an initial assumption the idea that VP-adverbs, floated quantifiers and clausal negation (in French, this is taken to be pas, not ne) occupy the same position in the two languages, the facts in (8) are evidence for a rule of verb movement in French which does not exist in English. More precisely, if we suppose, still as a first approximation, that VP-adverbs, floated quantifiers and clausal negation are all left-adjoined to VP, we conclude that the French verb-movement rule moves the verb out of VP. Following the Structure Preservation Condition of Chomsky (1986b), the verb must move to a head-position. The obvious conclusion is that this rule moves the verb to I. So, as Pollock shows in detail, the relevant aspect of Emonds' system is to say that French has a rule which places V in I in tensed clauses. Despite initial appearances, English also has this V-to-I movement rule, but its scope is much more restricted than its French counterpart. In English, only auxiliaries, i.e. essentially aspectual have and be, can move to I; this is the have/beraising rule of Emonds (1976). The evidence for this is as follows: (9) a. John has often kissed Mary. b. The children have all eaten the chocolate. c. John hasn't seen Mary. (10) a. John is often kissing Mary. b. The children are all eating the chocolate. c. John isn't talking to Mary. So we see that the difference between English and French is not that French has a rule which English lacks, but rather that both languages have the same rule, whose scope of operation differs. Pollock suggests an account of this difference based on two ideas: (i) that the fundamental difference between auxiliaries and main verbs is that auxiliaries do not assign its 6-roles while main verbs do; (ii) that I is a position from which 0-roles can be assigned in

14

INTRODUCTION

French (a "transparent" position) but not in English (where I is an "opaque" position). Thus, English sentences like those in (8) are violations of the 9criterion, since the moved verb is unable to assign its 9-roles. Of particular relevance to this volume is the well-known fact that sentences of this type were grammatical until roughly the 17th century; Pollock's interpretation of this is that English I was formerly transparent. He suggests, following observations in Roberts (1985), that "transparency" is connected to the presence of a relatively "rich" agreement system. A number of the papers in this volume discuss V-to-I movement in English, French and Germanic languages. Lightfoot's contribution in particular focusses on the loss of this operation in the history of English, addressing in this connection the kinds of learnability questions that were alluded to in the previous section.

2.3 The "Split Infl" Hypothesis Another important result of Pollock (1989) concerns the structure of the clause and the behaviour of French infinitives. First, Pollock observes that French infinitives show the same split as English finite verbs in that only the auxiliaries etre and avoir can move over the negative pas: (11) a. N'etre pas content est une condition pour ecrire. 'To be not happy is a condition for writing' b. *Ne sembler pas content... To seem not happy...' In Pollock's terms, this is evidence that only finite I is a transparent position in French; non-finite I is opaque. However, the situation regarding the placement of main-verb infinitives in relation to adverbs is more complex. Infinitives can precede some adverbs, for example: (12) a. A peine parler 1'italien apres cinq ans d'etude...

Hardly to-speak Italian after five years of study... b. Parler a peine 1'italien apres cinq ans d'etude... This leads Pollock to propose a "short" movement of main-verb infinitives. What is the landing site of this movement? Here Pollock capitalizes on the fact that I node of Chomsky (1986b) was a rather uncomfortable combination of the features of Tense (T) and Agreement (Agr), and proposes that these two kinds of features should each project their own X-bar structure. This gives the two separate functional projections TP and AgrP. In these terms, the "short" movement of main-verb infinitives in French should be taken as a movement to the lower of these two heads, while the longer movement of tensed main verbs in French is to the higher of these heads. Pollock assumed that TP dominates AgrP, so this gives the following clause structure:

INTRODUCTION

15

Pollock's "split-Infl" hypothesis has given rise to a vast amount of work on basic clause structure. At the time of writing, it is difficult to see what kind of consensus will emerge. However, two elaborations of Pollock's system are worthy of particular note, and should be discussed here since they are relevant to some of the papers in this volume. Belletti (1990) argues that AgrP should be taken to dominate TP. There are two arguments for this. The first is that this proposal places the subject in SpecAgr', which is most natural since the subject is the element that agrees with Agr. The second is based on the Mirror Principle of Baker (1985a). The Mirror Principle claims that the order of affixes in a morphologically complex word reflects the order of syntactic operations triggered by that word. One way of understanding this is to say that the linear order of affixes is a direct consequence of syntactic head movement. Now, the morphologically complex verb forms in Italian and other Romance languages are built up according the schema root + tense + agreement. This cannot always be seen, owing to the intricacy of the morphological system, but emerges clearly in the future, conditional and imperfect tenses. Here a form like Italian leggevano ("they were reading") consists of the three morphemes legg- (the root), -eva(the imperfect marker) and -no (the 3pl marker). A derivation of the following type naturally gives rise to this order of morphemes, following the Mirror Principle:

16

INTRODUCTION

Here the verbal root first incorporates into the tense affix to give a complex verb form which, in turn, incorporates with the agreement affix to give the full verb form. Assuming this clause structure, i.e. the Mirror Principle and the Head Movement Constraint (which requires heads to move cyclically through locally governing head-positions, with no "skipping" of intermediate positions), we derive the form of the complex verb automatically. Chomsky (1991) elaborates Belletti's proposal further by suggesting that, in a sense, both Belletti and Pollock are right: there is an Agr-projection above T—this is the position of affixes specifying agreement with the subject, or AgrS—and there is also an Agr-projection below TP, the position for agreement with the object, Agr-O. In any case, whichever variant of the "split-Infl" proposal is correct, it is now fairly clear that the basic clausal structure of many languages provides evidence for at least two functional projections (in addition to CP). Belletti (1990) shows that a range of what appear to be minor word-order differences between French and Italian can be profitably analysed as the direct consequence of the different scopes of verb-movement rules, rather than as rules affecting particular lexical items. Consider for example the different distribution of negative polarity items with respect to the infinitive in French and Italian. While in Italian the negative polarity item invariably follows the infinitive, in French its most usual position (and, for some items, the only position) is that preceding the infinitive: (15) Gianni ha deciso di non tornare mai/piu. Gianni has decided to not return never/more. (16) *Gianni ha deciso di non mai/piu tornare.

INTRODUCTION

17

(17) Jean a décidéde ne pas rentrer. Jean has decided to not return. What these contrasts and many others show is that infinitives (and also other non-finite verb forms) always move out of VP in Italian. In other words, the higher functional projections are transparent to non-finite verbs in Italian. Pollock introduced a further functional projection in his analysis of negation in English and French. This was NegP, the projection of a negative morpheme. Belletti adopts this idea, and proposes the following D-structure for the complement clauses in (15-17):

In the derivation to S-structure, the Neg-elements ne/non, which are clitics, adjoin to Agr thereby changing their linear order with respect to the negativepolarity items. In Italian, the verb also obligatorily moves to Agr to give the order seen in (15). In French, on the other hand, the infinitive optionally raises but only as far as T (when it does not raise, Affix-Hopping lowers the infinitive ending onto the verb-root). Other polarity items in French, for example, plus 'no more' and point 'not at all' may follow infinitives; for these cases Belletti proposes a different D-structure, which we will not go into here. As we said earlier, it is not clear at the time of writing what the correct basic structure for clauses may be. However, the different proposals by Pollock, Belletti and Chomsky have led to an enormous amount of fruitful work, and a number of the contributions in this collection assume some variant of the "split-Infl" hypothesis.

18

INTRODUCTION

2.4 Verb Second The above paragraphs have been concerned mainly with verb movement within. IP, and consequently have not addressed the analysis of inversion constructions, i.e. when the verb moves over the subject (taking the usual position of the subject to be Spec!'). The most prominent type of construction where the verb appears to regularly move out of IP is the verb-second phenomenon, found in German and other Germanic languages (but not contemporary English). The verb-second constraint requires that the finite verb be preceded by exactly one constituent in matrix declarative clauses: formally, it says *Y XP V in such clauses, where Y is non-null. The precise nature of XP is immaterial; it may be the subject, a complement, or an adverbial element. The following German sentences (from Tomaselli (1989)) illustrate: (19) a. Ich las schon letztes Jahr diesen Roman. I read already last year this book. b. Ich habe schon letztes Jahr diesen Roman gelesen. I have already last year this book read. (20) a. Diesen Roman las ich schon letztes Jahr. This book read I already last year. b. Diesen Roman habe ich schon letztes Jahr gelesen. This book have I already last year read. (21) a. Schon letztes Jahr las ich diesen Roman. Already last year read I this book. b. Schon letztes Jahr habe ich diesen Roman gelesen. Already last year have I this book read. A number of the papers in this volume discuss this phenomenon from a diachronic perspective. The topic is of considerable interest, as both English and French have lost this constraint in their recorded history; this matter is discussed by Platzack and by Hulk and van Kemenade, while Lemieux and Dupuis suggest that verb second in Old French was not quite the same as what we find in most contemporary Germanic languages. Santorini discusses the development of embedded verb second in the history of Yiddish, while Ribeiro gives evidence that Medieval Portuguese was verb second. Kiparsky proposes an account of the development of the Germanic complementizer system, and thus of the conditions that led to the development of Germanic verb second. Finally, Tomaselli discusses some apparent exceptions to this constraint in Old High German and Old English. The most widely accepted analysis of verb second (although certainly not the only one) was first put forward in den Besten (1983). Den Besten proposed that the inflected verb moves to the C position in matrix declaratives in verb-second languages. This operation is associated with the fronting of some XP to SpecC. The derivation of (20a) would thus be as in (22):

INTRODUCTION

19

In this way, the root nature of the phenomenon is explained: embedded complementizers are frequently filled (and at a more abstract level of analysis, they perhaps always are) and so cannot serve as the landing site for the fronted verb. Moreover, this view implies that the subject is in the same position in verb-second languages as in related non-verb-second languages (although there is the further question of the treatment of SVO sentences like (19) in verb-second languages—is the subject in SpecAgr' or in SpecC? For differing points of view, see Zwart (1990) and Vikner and Schwartz (forthcoming)). An analysis of verb second of the den Besten type immediately raises two questions: (i) what forces the verb to move to C? (ii) what forces topicalization of XP to SpecC'? Den Besten's original proposal regarding the first of these questions was that C contained the Tense feature, and hence the finite verb was attracted there as a subcase of the general attraction of finite verbs to Tense morphemes. This proposal has been followed up in various ways. Koopmann (1984) proposed that C is responsible for Nominative-Case assignment in verb-second languages, and, as such, must be lexicalized by the finite verb. Platzack (1987) proposed that in verb-second languages C and I were united as Confl. In more recent work with Anders Holmberg (Holmberg and Platzack (1988); Platzack and Holmberg (1989)), Platzack has proposed that verb-second languages feature an abstract finiteness operator in C. In these

20

INTRODUCTION

terms, the loss of verb second is seen as the loss of this feature on C; in Platzack's contribution to this volume he proposes that this operator has shifted its position from C to I in the recorded history of both English and French, thereby explaining the loss of verb second in these languages. Hulk and van Kemenade propose a similar account based on a notion of "licenser." On the other hand, Lemieux and Dupuis question the existence of a root/embedded distinction in Old French verb second. Tomaselli (1989) takes a slightly different tack by suggesting that verb-second C contains an abstract agreement specification; this account of verb second is taken up by Roberts (1992) in his account of the loss of verb second in French. Accounts vary as to the nature of the obligatory topicalization. In Roberts (1992:71) this is taken to be forced by the following condition: (23) A head containing Agr must have a filled specifier. Since a verb-second C always contains Agr (unlike other Cs), this condition requires the presence of some constituent in the specifier of those Cs. However, no further condition specifies the nature of that constituent. Roberts suggests that the condition in (23) is related to the Extended Protection Principle of Chomsky (1982). Another approach is to say that the topicalization process triggers the V-fronting process. This idea was first developed by Taraldsen (1986), and is argued for here by Lightfoot (Chapter 2). Despite a large and growing research literature on verb second, however, the questions concerning how this operation is triggered have not been fully answered; for a detailed survey of all the main proposals that have been made (including some we have not mentioned here), see Vikner (1990, ch. 2). In this section we have given only the barest outline of the theory of verb movement, and of the various proposals that have been made to account for the different kinds of verb movement that have been observed. The analyses to be found in the articles contained in this volume, which we will review in the next section, will refine this outline somewhat.

3. Overview of the Collection As we have already mentioned, the papers in this collection deal with a fairly coherent, overlapping set of themes: verb second, V-to-I movement, null subjects and clitics. We have grouped the papers into sections reflecting the main themes that are treated. However, owing to the overlapping nature of the themes, it should not be assumed that assigning a paper to one section implies that it has nothing to say on issues that are at the forefront of discussion in another section. We now introduce the papers in order, and try to bring out the connections between them. Part One of the book contains the papers that focus mainly or exclusively on verb second. Lightfoot looks at the implications for the theory of learnability of the loss of verb second in English. Santorini shows how verb second has changed its character in the recorded history of Yiddish. Lemieux

INTRODUCTION

21

and Dupuis argue for a particular analysis of Middle French verb second, and Ribeiro shows that Old Portuguese had verb-second properties. Finally, Kiparsky presents an account of the development of verb second in Germanic. Lightfoot's paper deals with a question central to all theoretically driven work on diachronic syntax: the precise nature of the interaction of language acquisition and parameter setting. As we saw in SECTION 1, Universal Grammar has, in recent generative research, been considered to be made up of a set of principles and a set of parameters. Language acquisition may be thus conceived of as the child setting the various parameters of the particular language she is exposed to. A language will therefore change over time as acquirers of that language come to set parameters differently (see the discussion around (3) in SECTION 1). Lightfoot sets out to show, however, that the loss of the verb-second phenomenon in the diachronic development of English, when examined in some detail, shows that the whole question of parameter setting is no simple matter. The relationship between grammatical data and the triggering experience which sets the parameters for these data may be a very indirect one, Lightfoot maintains. He goes on to conclude "that a modern historical linguist cannot say that some changing phenomenon is due to the new parameter setting which accounts for it.... Rather, it manifests and provides evidence for that parameter setting, it is due to changes in the triggering experience which in turn entailed the new parameter setting." This is an important conclusion for our overall conception of language change as outlined in SECTION 1. Beatrice Santorini's paper "The Generalisation of the Verb Second-Phenomenon in the History of Yiddish," also deals with the diachronic development of verb second. Santorini shows that early Yiddish exhibits the same asymmetry between root and embedded contexts as is to be found today in German, while no such distinction is observable in Modern Yiddish. A quantitative investigation of over 2,200 subordinate clauses from over 40 Yiddish texts allowed the author to examine the generalisation of verb second from root to subordinate clauses. This diachronic change with respect to verb second is attributed to two other syntactic phenomena in Yiddish: a change in phrase structure from I-final to I-medial configuration and a change in NominativeCase assignment which allows for SpecIP to be occupied by non-subjects. The modification in Nominative-Case assignment is accounted for by postulating a shift of the finiteness operator from the C-position to the I-position (see Platzack's paper in this volume for more on the finiteness operator). Monique Lemieux and Fernande Dupuis' contribution focusses on a "symmetric" verb-second system, i.e. one in which verb second is not restricted to root clauses. Their paper, "The Locus of Verb Movement in Non-Asymmetric Verb-Second Languages; the Case of Middle French" looks at the verbsecond data from the history of French and gives an analysis of them which is at variance with the other papers in this volume dealing with verb second in French. The fundamental difference between this approach and the others is their claim that verb second in French can be best analysed as an IP-internal syntactic phenomenon. This analysis implies that, as in Yiddish (cf.

22

INTRODUCTION

Santorini's contribution) and Icelandic, Old and Middle French should not display asymmetries between root and embedded environments. It is argued that, although verb second is far less frequent in embedded contexts, it is nevertheless possible and that extraction from verb-second embedded clauses suggests that the exceptional "CP-recursion" analysis usually offered for them in the context of a CP-analysis for verb second is not applicable to the French data. The rarity of verb-second embedded clauses is considered to be amenable to an analysis which has recourse to illocutionary elements in the sentence. Data from Portuguese are examined in Ilza Ribeiro's paper "Evidence for a Verb-Second Phase in Old Portuguese." The analysis of a corpus of examples drawn from 14th-century texts suggests that verb-second phenomena can be identified in Old Portuguese. These data are shown to be compatible with the sort of CP-analysis for verb-second order which has often been proposed for Old and Middle French. Ribeiro offers an explanation of how the verb-second configuration was lost from this language (with the exception of a certain residue). She bases her account on the proposals made in Roberts (1992), which in turn derives from Koopmann and Sportiche (1991), that various parametric choices can be made with respect to the assignment of Nominative Case by either Agr or by T. The verb-second nature of Old Portuguese is also briefly discussed in Beninca's contribution (Chapter 12), where Portuguese is integrated into a typology of verb second in Medieval Romance. Paul Kiparsky's paper "Indo-European Origins of Germanic Syntax" attempts to deal with the origins of verb second in the Germanic languages. Kiparsky identifies the close relationship between the phrase structure of Old English, Old High German and Old Icelandic together with their parent language, Proto-Germanic, and other branches of Indo-European represented by Vedic Sanskrit, Hittite, and Homeric Greek. Evidence based on data concerning verb movement, clitic movement, coordination and relative-clause formation is given which argues for a Germanic clause structure which distinguishes an adjoined topic position, a Specifier position and a head-position within CP (it is interesting to compare this proposal with Beninca's structure for Medieval Romance, discussed below). Kiparsky argues that this structure descends directly from Indo-European, except that Indo-European had no C-position (in the sense of having a slot where indeclinable complementizers are inserted in embedded clauses). This in turn is connected to the fact that finite embedded clauses are not found in older Indo-European languages. The development of complementizers, and consequently of a C-position capable of hosting the finite verb, is tied to the development of finite subordination. Part Two groups together papers which relate aspects of verb second to the null-subject parameter. Most of the papers focus on French, although both Platzack's paper and Hulk and van Kemenade's paper compare the development of French with the development of English in these respects. Barbara Vance's paper concentrates on French. As Vance points out in her introduction, her focus is on the decline, rather than the loss, of verb second.

INTRODUCTION

23

What Vance argues is that many cases of verb-subject order in Middle French which are superficially cases of verb second, are not to be taken as instances of "true" verb second (which Vance takes to involve movement to C), but are in fact cases of Italian-style "free inversion," in the sense of Rizzi (1982). The free-inversion construction is characteristic of null-subject languages and consequently available in Old and Middle French. Vance's argument is that the availability of free inversion led to the weakening and eventual loss of productive verb second at the end of the Middle French period. She supports her arguments by showing that there is a difference in frequency in Middle French between verb-subject orders where the subject is a pronoun and such orders where the subject is a non-pronominal NP: the former decline through this period while the latter remain constant. Since "free inversion" with a pronominal subject was not possible in Old French or Middle French, inversion of a pronominal subject was an unambiguous indication of a "true" verb-second structure with the verb in C, while inversion with a non-pronominal subject was often amenable to either a verb-second analysis (with the verb in C) or a "free-inversion" analysis (with the verb in I). Accordingly, Vance interprets the statistical data as evidence that "true" verb second was declining throughout Middle French. This idea is further supported by the development of "verb-third" clauses, which is much more frequent after the 13th century for clauses with pronominal subjects than for those with non-pronominal subjects; again, Vance interprets this as evidence that non-pronominal subjects were able to appear in a different kind of inversion construction, and that this very fact was responsible for the decline of verb second. An attempt to reach a unified account of the loss of verb second in French and English is to be found in Christer Platzack's article ("The Loss of Verb Second in French and English"). The author's starting point is the observation that scholars agree that the loss of verb second takes place around the 14th and 15th century in these languages, but different reasons, usually language-specific ones, are put forward for this loss. The leading idea in this paper is that the loss of verb second is to be correlated with an abstract finiteness operator which indicates the existence of a predication at the time of the utterance. A mutual dependency is postulated between the position of this feature and the assignment of Nominative Case. In a verb-second language, the feature is associated with the C position and in a non-verb-second language with I. The change of position of the feature from C to I takes place at the same time in French and English, hence the simultaneous loss of verb second. However the other consequences of this change are not the same in the two languages. The main consequence of the shift in the position of the finiteness operator for English has been the growth of do-support and Infl-lowering while for French (at least for Middle French) it is to be witnessed in a change in the distribution of null subjects. The loss of verb-second configurations in the diachronic development of Modern English and French is also the central concern of Aafke Hulk and Ans van Kemenade's article "Verb Second, Pro-drop, Functional Projections and

24

INTRODUCTION

Language Change." The authors consider that two different kinds of abstract structures can give rise to verb second: structures involving movement of the verb to I and subsequent movement of V+I to C, and structures which involve the movement of V to I. Depending on whether the verb-second language makes use of the first or the second strategy, it is classified as a "Complementizer Verb-Second" (CV2) or an "Inflectional Verb-Second" (IV2) language. With this background Hulk and van Kemenade examine in some depth the interrelated phenomena of verb second and the distribution of referential and expletive null subjects. While Old French and Middle French (at least initially) are considered to have an I which is morphologically rich enough to license referential null subjects, Old English does not possess such morphological richness in I and, therefore, only expletive null subjects can be identified. Both Old French and Old English are, in terms of this analysis, to be classified as CV2 languages. The major change which takes place around the beginning of the 15th century is the move from CV2 to IV2 in both languages. A consequence of this change is that SpecIP becomes the only position to which Nominative Case can be assigned in English and, therefore, English becomes a non-verb-second language. On the other hand, French, with a morphologically rich I, even after the shift from CV2 to IV2, can still license and Case-mark referential null subjects in either VP-internal position or in SpecIP; a consequence of this situation is that for a short period (until this morphological richness is lost) French passes through an IV2 phase before becoming a non-verb-second language. Paul Hirschbiihler's contribution, "Null Subjects in Verb-First Embedded Clauses in Philippe de Vigneulles' Cent Nouvelles Nouvelles" focusses on the possibilities for null subjects in particular contexts in this early 16th-century French text. The paper aims to delimit as precisely as possible the syntactic contexts in which embedded verb-first configurations are found in this text. The results are that it is necessary to draw up a three-way division between the different kinds of subordinate-clause introducers according to whether they contribute to the formal licensing of SpecIP or not. The three groupings are shown to be those subordinators that do not license SpecIP, those that do (whphrases) and those that appear with hybrid properties (allowing preverbal subjects but not pro-drop): the last of these categories is instantiated by relative and comparative que. A similar tripartite division has to be made with respect to the person of the verb. Pro-drop is possible with second person plural, but first person and second person singular never allow it. Third person singular, expletive third person and third person plural do allow pro-drop in embedded contexts introduced by a w/i-item. Careful examination of the various kinds of w/i-clause shows that pro-drop in verb-first constructions is not equally represented in all of them. The paper is presented as strong justification of the need for more precise and accurate empirical studies of this kind. The last three papers all deal in various ways with the relation between clitics and verb second. Cecilia Poletto's object of study is the status and dis-

INTRODUCTION

25

tribution of subject clitics in Northern Italian dialects. The analysis presented connects the role of subject clitics in these dialects to at least three components of the grammar: the pro-drop parameter, conditions on Case assignment and the visibility of empty agreement heads. Detailed examination of the subject clitics in French and Renaissance and 17th-century Veneto varieties reveals that there are at least two types of subject clitics: those generated in SpecVP (argumental subject clitics) and those base-adjoined to Agr (expletive subject clitics). It is shown that some dialects can have both types of subject clitic. Their different function is examined in some depth drawing primarily on data from Veneto varieties, as well as Trentino and Friulian. Interestingly, Poletto shows that Renaissance Veneto had a similar null-subject system to that identified in Vigneulles' French by Hirchbiihler in his paper in this volume, and suggests that this kind of system is characteristic of a null-subject system which formerly depended on verb-second configurations for licensing null subjects, and from which verb second has recently been lost. The structure of the clause above the CP-level in the syntax of Medieval Romance is discussed in Paola Beninca's "Complement Clitics in Medieval Romance: The Tobler-Mussafia Law." This study seeks specifically to recast the 19th-century "law" of Medieval Romance syntax formulated by Adolf Tobler and Adolfo Mussafia—which states that clitic pronouns cannot appear in sentence-initial position—within the context of a model of generative grammar which makes use of functional projections and movement to C. The analysis proposes that Medieval Romance languages can be divided into two types: those which have free recursion of an unanalysed Top position above CP and those which do not, the latter group of languages being fairly rigidly verbsecond. Data drawn from Old French, Medieval Northern Italian dialects, Medieval Southern Italian dialects, and Old and Modern (European) Portuguese allow the conclusion to be drawn that enclisis of the clitic pronouns to the verb always occurs in Medieval Romance when the Spec,CP position is empty and the verb has moved to C. Old High German (OHG) was, like Modern German, characterised by the verb-second constraint. However OHG allowed for a number of patterns which deviate from the general verb-second configuration and which are not to be found in the modern language. These phenomena are examined by Alessandra Tomaselli in her paper "Cases of Verb Third in Old High German." She focusses particularly on the verb-third order which involves the surface configuration XP - subject pronoun -finite verb. The same verb-third order is also shown to be found in Old English (OE), following van Kemenade (1987). In both OHG and OE, it is argued that the data can best be accounted for if the functional projections CP and IP are considered to be independent and, furthermore, that the I-node is medial within the IP projection. This medial position of the I node will allow for cliticisation of the subject pronoun to I prior to verb movement from I to C, thus giving rise to an apparent verb-third order. The loss of this verb-third configuration in the diachronic development of the two languages is attributed to different factors. In German, it is proposed, the parameter setting of the head-complement relation within IP changed

26

INTRODUCTION

so that I became final, following its structural complement, VP. The loss of the medial I brought in its wake the loss of the possibility for the subject clitic to attach to I. In the development of English, however, I has remained medial but the loss of the possibility of cliticisation to this node is to be attributed to the effects of deflexion (both Hulk and van Kemenade and Platzack discuss this point further). This collection of papers testifies to the liveliness of current work in diachronic syntax. We see that, on the one hand, traditional questions (the Tobler-Mussafia Law, the nature of Germanic innovations) of historical linguistics are being seen in a new light, and can now be investigated using the powerful tools of contemporary linguistic theory. On the other hand, these papers show how diachronic data can be brought to bear in developing those theoretical tools. In these ways, diachronic syntax is becoming fully integrated in contemporary syntactic theory.

References Adams, M. (1987a) "From Old French to the Theory of Pro Drop." Natural Language and Linguistic Theory 5:1-32. Adams, M. (1987b) Old French, Null Subjects and Verb Second Phenomena. PhD Dissertation, UCLA. Andersen, H. (1973) "Abductive and Deductive Change." Language 49:567-595. Baker, M. (1985a) "The Mirror Principle and Morpho-Syntactic Derivations." Linguistic Inquiry 16:373-416. Baker, M. (1985b) Incorporation: A Theory of Grammatical Function Changing. PhD Dissertation, MIT. Baker, M. (1988) Incorporation: A Theory of Grammatical Function Changing. Chicago: University of Chicago Press. Belletti, A. (1990) Generalized Verb Movement: Aspects of Verb Syntax. Turin: Rosenberg and Sellier. Berwick, R. (1985) The Acquisition of Syntactic Knowledge. Cambridge, Mass: MIT Press, den Besten, H. (1983) "On the Interaction of Root Transformations and Lexical Deletive Rules." In W. Abraham, ed., On the Formal Syntax of the Westgermania, 47-131. Amsterdam: John Benjamins. Bloomfield, L. (1933) Language. New York: Holt, Reinhart and Winston. Chomsky, N. (1957) Syntactic Structures. The Hague: Mouton. Chomsky, N. (1965) Aspects of the Theory of Syntax. Cambridge, Mass.: MIT Press. Chomsky, N. (1975) Reflections on Language. New York: Pantheon. Chomsky, N. (1980) Rules and Representations. New York: Columbia. Chomsky, N. (1981) Lectures on Government and Binding. Dordrecht: Foris. Chomsky, N. (1982) Some Concepts and Consequences of the Theory of Government and Binding. Cambridge, Mass.: MIT Press. Chomsky, N. (1986a) Knowledge of Language: Its Nature, Origin and Use. New York: Praeger.

INTRODUCTION 27

Chomsky, N. (1986b) Barriers. Cambridge, Mass.: MIT Press. Chomsky, N. (1991) "Some Notes on Economy of Derivations and Representations." In R. Friedin, ed., Principles and Parameters in Comparative Grammar, 417-454. Cambridge, Mass.: MIT Press. Clark, R. (1990) Papers on Learnability and Natural Selection. Technical Reports on Formal and Computational Linguistics vol. 1. Geneve: Universite de Geneve. Clark, R. and I. Roberts (1993) "A Computational Model of Language Learnability and Language Change." Linguistic Inquiry 24:299-345. Emonds, J. (1976) A Transformational Approach to English Syntax: Root, Structure-Preserving and Local Transformations. New York: Academic Press. Emonds, J. (1978) "The Verbal Complex V - V in French." Linguistic Inquiry 9:151-175. Foulet, L. (1919) Petite syntaxe de 1'ancienfrancais. Paris: Editions Champion [3rd ed. rev. 1982]. Greenberg, J. (1963) "Some universals of grammar with particular reference to the order of meaningful elements." In J. Greenberg, ed., Universals of Language. Cambridge, Mass.: MIT Press [2nd edition 1966]. Haegeman, L. (1991) Introduction to Government and Binding Theory. Oxford: Blackwell. Hale, K., L. Jeanne and P. Platero (1977) "Three Cases of Overgeneration." In P. Culicover, T. Wasow and A. Akmajian, eds., Formal Syntax. New York: Academic Press. Hawkins, J. (1983) Word Order Universals. New York: Academic Press. Holmberg, A. and C. Platzack (1988) "On the Role of Inflection in Scandinavian Syntax." Working Papers in Scandinavian Syntax 42:25-42. Hyams, N. (1986) Language Acquisition and the Theory of Parameters. Dordrecht: Kluwer. Jaeggli, O. (1982) Topics in Romance Syntax. Dordrecht: Foris. Jaeggli, O. and K. Safir, eds. (1989) The Null Subject Parameter. Dordrecht: Kluwer. Kayne, R. (1984) Connectedness and Binary Branching. Dordrecht: Foris. van Kemenade, A. (1987) Syntactic Case and Morphological Case in the History of English. Dordrecht: Foris. Klima, E. (1969) "Relatedness between Grammatical Systems." In D.A. Reibel and S.A. Schane, eds., Modern Studies in English: Readings in Transformational Grammar. New Jersey: Prentice Hall. Koopmann, H. (1984) The Syntax of Verbs: From Verb Movement Rules in the Kru Languages to Universal Grammar. Dordrecht: Foris. Koopmann, H. and D. Sportiche (1991) "The Position of Subjects." Lingua 85:211-258. Lightfoot, D. (1979) Principles of Diachronic Syntax. Cambridge: Cambridge University Press. Lightfoot, D. (1991) How to Set Parameters: Arguments from Language Change. Cambridge, Mass.: MIT Press. Montague, R. (1974) Formal Philosophy: Selected Papers of Richard

28

INTRODUCTION

Montague. Edited and with an introduction by R.H. Thomason. New Haven: Yale University Press. Paul, H. (1960) Prinzipien der Sprachgeschichte. 5th ed. Tubingen: Niemeyer. Pierce, A. (1989) The Emergence of Syntax: A Cross-linguistic Study. PhD Dissertation, MIT. Pierce, A. and V. Deprez (1990) "A Crosslinguistic Study of Negation in Early Syntactic Development." Ms. University of Pennsylvania/Rutgers University. Platzack, C. (1987) "The Scandinavian Languages and the Null Subject Parameter." Natural Language and Linguistic Theory 5:377-401. Platzack, C. and A. Holmberg (1989) "The Role of AGR and Finiteness in Germanic VO Languages." Working Papers in Scandinavian Syntax 43:51-76. Pollock, J.-Y. (1989) "Verb Movement, UG and the Structure of IP." Linguistic Inquiry 20:365-424. Quine, W.V.O. (1960) Word and Object. Cambridge, Mass.: MIT Press. Rizzi, L. (1982) Issues in Italian Syntax. Dordrecht: Foris. Rizzi, L. (1988) "The Structural Uniformity of Syntactic Categories." In Proceedings of the Conference on the Basque Language. Vitoria/Gasteiz: Eusko Jaurlaritzaren Argitalpen-Zerbitzu Nagusia/Servicio Central de Publicaciones del Gobierno Vasco. Roberts, I. (1985) "Agreement Parameters and the Development of English Modal Auxiliaries." Natural Language and Linguistic Theory 3:21-58. Roberts, I. (1992) Verbs and Diachronic Syntax. Dordrecht: Kluwer. Roeper, T. and E. Williams (1987) Parameter Setting. Dordrecht: Kluwer. Sapir, E. (1949) Language. New York: Harcourt. Saussure, F. de (1916) Cours de Linguistique Generale. Paris: Payot [4th ed. 1969]. Stowell, T. (1981) Origins of Phrase Structure. PhD Dissertation, MIT. Taraldsen, K.T. (1986) "On Verb Second and the Functional Content of Grammatical Categories." In H. Haider and M. Prinzhorn, eds., Verb Second Phenomena in Germanic Languages, 7-26. Dordrecht: Foris. Thurneysen, R. (1892) "Die Stellung des Verbums im Altfranzosischen." Zeitschrift fur Romanische Philologie 16:289-371. Tomaselli, A. (1989) La sintassi del verbo finito nelle lingue germaniche. Dottorato di Ricerca in Linguistica, Universita di Pavia. Travis, L. (1984) Parameters and Effects of Word Order Variation. PhD Dissertation, MIT. Vikner, S. (1990) Verb Movement and the Licensing of NP-Positions in the Germanic Languages. These de Doctoral, Universite de Geneve. Vikner, S. and B. Schwartz (1991) "The Verb Always Leaves IP in V2 Clauses." In A. Belletti and L. Rizzi, eds., Parameters and Functional Heads. Oxford: Oxford University Press. Zwart, J.-W. (1991) "Clitics in Dutch: Evidence for the Position of INFL." Groninger Arbeiten zur Germanistischen Linguistik 33:71-92.

Part One The Diachrony of Verb Second

This page intentionally left blank

2 Why UG Needs a Learning Theory: Triggering Verb Movement* David Lightfoot University of Maryland at College Park

For more than a decade generativists have viewed the linguistic genotype or "Universal Grammar" (UG) as consisting of principles and a set of option-points or parameters. Correspondingly, language acquisition proceeds as children set those parameters, sometimes characterized as switches with ON and OFF positions. And languages change over time as parameters come to be set differently. The notion of a parameter has stolen the limelight, but particular parameters are surprisingly evanescent. That is, for only very few parameters can one point to a solid basis of evidence, a clear understanding of how a particular parameter-setting interacts with other elements of various grammars, and a plausible screen-play for how the parameter might come to be set in the appropriate fashion. Our analytical performances do not match the script very closely. That script is offering too simplistic a view of language acquisition, and the discrepancy between script and performance could bring us some bad reviews. Like other authors in this volume, I shall discuss one of the most tantalizing topics in the current theatre of research, the verb-second phenomenon. I hope to show that the loss of this phenomenon in the history of English suggests that parameter setting is not a simple matter and that substantial ideas are needed about how parameters come to be set. If UG consists of principles and parameters, then one needs a distinct type of theorizing to connect more directly with acquisition, dealing with what it takes to set parameters; this does not follow from the nature of the parameters themselves and therefore reflects a distinct theory of acquisition or, properly construed, a learning theory.

31

32

THE DIACHRONY OF VERB SECOND

1. The Verb-Second Phenomenon, Some Analyses and Degree-0 Learnability The verb-second phenomenon in its most familiar form has a finite verb occurring in second position, following a category of arbitrary grammatical/ thematic/semantic function. The phenomenon is often (although not always) restricted to matrix clauses. Paardekooper (1971) linked the non-occurrence of verb second in Dutch embedded clauses to the presence there of an overt complementizer, usually dot. Den Besten (1983) embedded this insight into generative analyses by positing a transformation moving the finite verb to a position in Comp; this movement would be blocked if the Comp position were already filled by an overt complementizer. This analysis, in turn, has been incorporated into standard wisdom, along with the more recent notion (due to Travis (1984)) that it can be seen as movement of a head if Comp is the head of a clause. In that case, the verb moves first to I(nflection). If, following Stowell (1981) and Chomsky (1986), I and C(omp) are each heads which project to a maximal phrasal category, the structure of a verb-second clause would be as in (1), where the verb moves first to I and then V+I moves to C. (1) (1)CP[Spec c-tlVj+H IP[NP I'[VP[... ei ej] ] ] ] Spec of CP may be filled by any phrasal category through what is often referred to misleadingly as a "topicalization" operation; this yields to the familiar verb-second phenomenon of Dutch, German, the Scandinavian languages, etc. Principles of UG force heads to move locally and therefore preclude movement of an uninflected (infinitive) verb directly to C. The inflected verb (V+I) cannot move to C if C is already filled, for example, by a complementizer in an embedded clause. This yields straightforwardly what have been taken to be core features of the verb-second phenomenon, and this analysis will provide a vocabulary with which we can discuss some interesting issues. In (2) I provide some simple examples of verb-second structures in Dutch. (2) a.a.CP[ den Haag; [bezoekj+t]k IP[hij VP[ei ej] ek] ] (2) The Hague visits 'he is visiting The Hague'

he

b.b. CP[in den Haagi [bezoekj+t]k IP[hij VP[ei het museum ej] ek] ]

'in The Hague he is visiting the museum' c. CP[Peteri [bezoekj+t]k IP[ei VP[den Haag ej] ek] ] c. 'Peter is visiting The Hague'. If one considers some non-verb-second languages, one sees some of the parameterization. For example, French allows finite verbs to move to C only if the initial element is +wh and the subject NP is pronominal (3). (3)

a.

[alli+ez]j IP [vous ej vp[ei a Paris] ] ] why go you to Paris 'why are you going to Paris?'

CP [pourquoi

WHY UG NEEDS A LEARNING THEORY

33

b. *CP[ a Parisk [alli+ez]j IP[vous ej vp[ei ek]]] English, on the other hand, allows only a narrow class of finite verbs to move to C, namely "auxiliaries," i.e. elements base-generated under I and the aspectual markers have and be; and they move to C only if the initial element is +wh or a negative (4). (4)

a. CP[whati havej IP[you ej VP[seen ei ] ] b. CP[never havej IP[you ej VP[seen such a mess] ] ] c. *CP[whati [discussj+ed]k IP[you ek VP[ej ei] ] ] d. *CP[never [discussj+ed]k IP[we ek VP[ejsuch a thing] ] ]

This suggests that, unlike French, English verbs do not move to I, as argued in a slightly different framework by Emonds (1978). This predicts correctly, for example, that English verbs must occur adjacent to their complements (5), whereas French verbs may be separated from their c mplements through movement to I (6). (5)

a. she always vp[reads the newspapers] b. *she reads always the newspapers

(6)

elle liti toujours VP[ei les journaux]

Other parameters may be involved, of course, but these are a few clear cases: verbs may or may not move to I; and any element may occur in Spec of CP and require I-to-C movement (Dutch, German, Swedish) or only categories bearing certain features may occur there (English, French). On the basis of the latter parameter, Rizzi (1990a) distinguishes "full" and "residual" verbsecond languages; we shall return to this distinction later. A more basic issue is: What exactly does the verb-second parameter consist of? That is, under what conditions do finite verbs move to C in Dutch, Swedish, etc.? Under what conditions do arbitrary phrasal categories occur in Spec of CP? What is the relation between these two properties? And how do children acquire the relevant parameter settings? Although we have a plausible descriptive framework, these questions continue to tantalize us and there are no entirely satisfactory answers. What requires explanation in the verb-second languages, whether underlyingly verb-object (like Swedish) or object-verb (like German and Dutch), is the obligatoriness of I-to-C movement in all main clauses and the obligatoriness of what is often called "topicalization," i.e. the movement of a phrasal category to Spec of CP, in declarative main clauses. Some element of UG will be needed to explain the obligatoriness of the processes. Since children generally do not have access to negative data, the Dutch child cannot "learn" that, for instance, I-to-C movement is obligatory directly on the basis of the ungrammaticality of *in den Haag hij het museum bezoekt 'in The Hague he visits the museum'. If that datum is not available to the child, there can be no consequence or inference or learning based on it. One must leave open the possibility that this learning might take place if children have indirect access to such data: that is, failure to hear such a simple expression might

34

THE DIACHRONY OF VERB SECOND

lead the child to deduce its ungrammaticality and thus to acquire some specific device which will block its derivation. This cannot be ruled out in principle but there are reasons to be sceptical of appeals to indirect negative data (Lightfoot (1991, ch.l)). If one is eventually forced to appeal to negative data, one would expect to find that this kind of deductive learning would emerge in children somewhat differently than the usual parameter-setting, perhaps later or after a stage of misgeneralization. Certainly an argument will be required that children do have indirect access to negative data under, one hopes, narrowly prescribed circumstances. Nonetheless, one finds several proposals in the extensive verb-second literature postulating language-specific devices, and in such a way that they implicitly (but never explicitly, as far as I know) presuppose indirect access to negative data. For example, the proposals of den Besten (1983), de Haan and Weerman (1986), Haider (1986), Koopman (1984) and others each postulate some element in C which must, given some principle of UG, "attract" I (or the finite verb) to that position, but the evidence for this element is the obligatoriness of the movement, i.e. the ungrammaticality of structures where movement has not taken place. In the absence of an appropriate learning theory whereby children have access to the appropriate negative data, the proposal simply re-states the problem. Platzack's (1986) proposal suffers from the same defect when he postulates different projections in verb-second and non-verbsecond languages. His idea is that S is a projection of I in English and other languages which do not have verb-second properties, and a projection of C in verb-second languages. He also allows grammars to differ in terms of whether the subject NP is base-generated as part of the projection of I or as part of the projection of C. Then a principle of UG, his "Case assignment rule," forces a verb to move to C in grammars where the projection of I does not include the subject NP. This postulates major differences in projection-types and the evidence for the particular projection-scheme for a verb-second language is the obligatoriness of I-to-C movement and the ungrammaticality of structures where I fails to move to C, and again there is no discussion of how the child would have access to such data. It is unclear how one might treat grammars (like those of earlier forms of English) allowing verbs to move to C as an option, and the proposal also fails to account for the necessity of topicalization in declarative main clauses, a point to be taken up below.1 The same sort of attainability problem arises with the quite different proposals of Safir (1981) and Evers (1982). Safir has UG require that verbs (and other heads) be uniquely governed, and he makes the structure of German such that verbs must move to C in main clauses in order for this requirement to be satisfied. Evers keys movement of the finite verb to the assumption that the tense element has a scope-bearing property and must c-command the clause; in embedded clauses the verb does not need to move because the matrix verb determines the scope of tense. In each case the child has to learn something specific about verb-second languages: Safir's child must learn the structural properties which entail that a verb in a base-generated main clause I position (assuming that verbs move to I in all clause-types) fails to be uniquely gov-

WHY UG NEEDS A LEARNING THEORY

35

erned in the appropriate sense (unlike, for instance, equivalent verbs in French). Evers' child must learn that C is the only position which c-commands the clause, unlike in English, French, Italian, etc. In each case the evidence is not explicitly discussed but it would appear to be the ungrammaticality of structures where movement to C has not taken place. Also, neither Safir nor Evers discuss the relationship between movement to C and topicalization. One solution to the problem of an appropriate triggering experience for the proposed analysis would be to make the verb-second phenomenon unmarked. So, if, for example, one postulates that I moves to C by attraction to some element there, one might argue that generating that element in C reflects the unmarked situation; English and French children, on the other hand, are exposed to positive data which show that the attractive element is absent, namely the occurrence of the finite verb in some other position. This might lead one to expect English and French children to go through an early verb-second stage, just as Hyams' (1986) English children go through an early pro-drop stage reflecting the alleged unmarked status of the null-subject option in her analysis. I know no evidence along these lines. A striking feature of the basic properties of verb-second languages is the correlation between the topicalization process and the obligatory movement of I to C. A fundamental shortcoming of many analyses is that while they offer some account for why the finite verb must be in some C-like position, they have nothing to say about why some other phrasal category must occur in initial position. Since the I-to-C movement is a common option in many languages which are not verb-second, it is unlikely that the obligatoriness of the movement would entail obligatory topicalization. Indeed, some verb-subject-object languages seem to have obligatory verb fronting without any obligatory topicalization; so, for example, Welsh seems to have underlying I-subject-verb-object order and surface verb-subject-object order in all clause-types by virtue of the verb moving obligatorily to I (Harlow (1981); Sproat (1985)), but there is no requirement that another phrasal category be moved forward. It is possible, of course, that the verb must not be just at the front of the clause but actually in C to entail topicalization, but it is hard to see why the obligatoriness of the movement should require topicalization, i.e. why topicalization should not be required in English or French when a verb happens to be in C. However, the reverse relationship might be more plausible: topicalization might entail verb movement, perhaps by a kind of predication requirement. So Taraldsen (1986), developing work by Cinque (1982) and Kayne (1982), argued interestingly that a topicalized argument phrase must be locally licensed and that a verb in C effectively turns the position filled by the topic into an argument position by yielding a predicate structure; so verb movement has the effect of providing a local licensing environment for a displaced argument phrase, which in turn permits its trace to be construed as a variable. He offered some intriguing evidence based on the absence of verb-second effects with bare w/z-words introducing root interrogatives in certain northern dialects of Norwegian. Also, it is a theory of the right type in that UG forces move-

36

THE DIACHRONY OF VERB SECOND

ment to C in order to license initial argument phrases; in that case, the child has to learn that Dutch and German sentences begin with some argument phrase of arbitrary function. Hellan and Koch Christensen (1986, introduction) noted that the verb-second phenomenon occurs in those languages which allow long distance anaphors, which are sometimes argued to be sensitive to predication; they went on to speculate that the common denominator of these languages may be some sort of predication sensitivity. However, there are two fundamental problems. First, it seems unnatural to construe expletives like Dutch er and German es as elements requiring predication, just as they cannot be taken as topics. Second, topicalization does not entail verb movement in languages that are not verb-second (English, French, etc.). If one keys this to a structural difference in the position of the topicalized element, then the requirement of a certain kind of predication relationship becomes unclear. Rizzi (1990a), nonetheless, builds on this idea of verb-second languages fulfilling a predication requirement. He distinguishes "residual" verb-second structures which occur in non-verb-second languages, like English, subject-auxiliary inversion and French subject-clitic inversion, and he compares them with the "full verb-second" phenomenon which determines the order of constituents in all main (and some embedded) clauses in Dutch, German and the Scandinavian languages. The distinction lies in the features of the head of CP, [+C.-IJ in the case of a residual verb-second construction and a kind of hybrid [+C,+I] in the case of full verb-second languages (7). Rizzi interprets the feature +C as "prepositional" and +I as "predicational." Thus a [+C,-I] category designates a proposition, projecting to the familiar CP of non-verb-second languages and of non-verb-second clauses in verb-second languages. [-C.+I] designates a predication, projecting to IP, and [+C.+I] is the hybrid category characteristic of verb-second languages, being both prepositional and predicational. For Rizzi this is the category of verb-second clauses in full verb-second languages. The inflected element [-C.+I] moves to the head of the CP, necessarily in the case of a full verb-second language because of a universal principle that the tense specification must c-command all other +1 categories in a given clause. (7)

a. Residual Verb Second

WHY UG NEEDS A LEARNING THEORY

37

b. Full Verb Second

The moved element properly governs a subject trace in (7b) under the "minimality" assumptions of Rizzi (1990b), because the trace is within the immediate projection of [+I]°, i.e. [+!]'; in (7a) "the moved head and the host are disjoint feature bundles, hence in no sense can the moved head be said to govern the subject trace within its (immediate) projection" (p. 385). This accounts for the symmetry of subject-object extractability in full verb-second languages, while subjects are relatively immobile in non-verb-second languages where a subject trace would generally fail to be properly governed: this difference between verb-second and non-verb-second languages was the central puzzle addressed by Rizzi's paper. Furthermore, different kinds of functional heads license different kinds of specifiers: so a [+C] head licenses an operator or trace in an A-bar chain and a [+I] head licenses a subject in the specifier position. So the hybrid case [+C,+I] allows its specifier to be a w/i-phrase or trace by virtue of being +C (8) or the subject of predication by virtue of being +1 (9). (8)

a. wer ist [e gekommen Infl] 'who has come?' b. wer hat Johann gesagt [e ist [e gekommen Infl]] 'who did Johann say has come?'

(9)

Maria ist [e gekommen Infl] 'Maria has come'.

A comparable approach, which does not involve notions of predicate formation, was adopted in Lightfoot (1991, ch.3) and can be stated in terms of constraints on phrase structure. Suppose that Dutch children learn that sentences begin with an arbitrary phrasal category, NP or PP etc., which has no fixed functional or thematic role. Since the category is utterance-initial and is not necessarily the complement of an adjacent verb, it cannot be treated as a complement of some head and must therefore be a specifier, since these are the only two positions in which a phrasal category may occur in D-structure; this assumes the familiar strong form of X-bar theory which claims that a head

38

THE DIACHRONY OF VERB SECOND

X projects to a X' which may also contain a complement, and the X' projects to a XP which may also contain a Specifier. Lexical specifiers always project to another phrasal category which must have a head, presumably C in this case, projecting to CP; UG dictates that a lexical specifier requires a lexical head. Since the initial phrasal category has no fixed thematic or functional role, the corresponding head cannot be I or any other element associated with a particular thematic, functional or case-assigning role, and therefore will be an empty position at D-structure, which is subsequently filled as another head moves to that position. The only element which is local enough to move to the empty head position and head-govern its trace, is I (with its associated verb). Under this approach, principles of UG dictate that an initial XP be locally licensed in a fashion which in turn requires obligatory movement of a verb to C. Universal 11 of Greenberg (1966:83) says: "Inversion of statement order so that verb precedes subject occurs only in languages where the question word or phrase is normally initial. This same inversion occurs in yes-no questions only if it also occurs in interrogative word questions." If languages generally do not move verbs to the front in questions unless they also front interrogative phrases, then it is plausible to interpret the initial phrase as licensed by the moved verb. If children acquiring a verb-second language learn that utterances begin with an element of arbitrary (functional or thematic) role, then it would follow from UG that the inflected verb must occur in second position, specifically, let us assume, in C, in order to license the initial phrasal category. This analysis provides a different answer to Rizzi's puzzle. (lOa) is well formed in a full verb-second language, because the subject trace is properly governed by the inflected verb in C (details omitted); but (lOb) is ill-formed in English because did is not sufficiently lexical to act as a proper governor for the subject trace. (10) a.CpLwiei bezoekt ip[eiden Haag] ] b.cp[whoi did IP[ei visit The Hague] ] An alternative account, adopted by Aoun, Hornstein, Lightfoot and Weinberg (1987), would construe (lOb) as a violation of the doubly-filled Comp filter, which requires Comp to contain no lexical material other than the head. This filter is "unlearned" in Dutch when children are exposed to structures like (10a).2 Although current theoretical mechanisms explain certain aspects of the verbsecond phenomenon, some things remain mysterious. Nonetheless children attain these verb-second languages readily, more readily in fact than children acquire the "residual" verb-second constructions in English. I have shown in Lightfoot (1991, ch.3) that simple data from matrix Domains suffice to trigger underlying object-verb order in V for Dutch and underlying verb-object for Swedish, and children do not need rich or complex data in order to learn that verbs move and from where they move. A crucial element in this account was that the X-bar schemata of UG require that verbs occur alongside their complements at D-structure; therefore Dutch children have ample evidence that verbs move syntactically, because they encounter plenty of utter-

WHY UG NEEDS A LEARNING THEORY

39

ances where verbs and complements are not adjacent (e.g. (2b) above). Furthermore, there is ample evidence from unembedded Domains that Dutch verbs follow their complements underlyingly; therefore Dutch children do not need access to embedded Domains to establish the underlying position of the verb. So the underlying position of the verb is readily attainable by a "degree-0 learner," i.e. a child who sets her parameters only on the basis of simple, unembedded structures. Not only is this a possible and theoretically pleasing account, but acquisitional data strongly suggest that something along these lines is correct, that there are simple, unembedded indicators which enable the child to adopt an object-verb setting and to posit the relevant verb movement operation. Clahsen and Smolka (1986) identify four stages in the acquisition of German verb movement properties (11). (11) a. stage 1 (25-29 months): no fixed order between sentence constituents; all verbal elements (including verbal complexes) occur in first/second and final position with a preference for final position. b. stage 2 (31-33 months): verbal elements with particles occur regularly in final position; other finite verbs occur in both first/ second and final position. c. stage 3 (36-39 months): all and only finite verbs occur in first/ second position; verbal complexes with finite and non-finite parts appear in discontinuous positions. d. stage 4 (41-42 months): as soon as embedded sentences are produced, their finite verbs are in final position. Strikingly, from the earliest relevant stage, children identify sentence-final position as one of the possible positions for verbs, including finite verbs despite the fact that they are almost never heard in this position in main clauses. At stage 3 there is a dramatic increase in the frequency of verb-second structures: in stages 1 and 2 they are used in only 20-40% of the utterances but at stage 3 they are used in 90%; Clahsen and Smolka (p. 149) report that this increase takes place explosively, within a month for all the children studied. At this stage children seem to have the object-verb D-structure order and an operation moving a finite verb obligatorily to a C-like position. In simple clauses there are two positions for verbs: verbal elements with the suffix -t and modals occur in second position while infinitives and verbs with other inflections occur sentence-finally. To this extent the adult system is in place (Clahsen (1990)). Importantly, when they begin to use embedded structures (stage 4), the finite verbs are invariably in final position and there seems to be no "experimentation" or learning based on embedded clause data. Clahsen and Smolka go further and make stronger claims: they take "move V" to operate from the earliest stage, initially affecting verbs and verbal complexes of all types, affecting only simple verbs at stage 2, and affecting finite simple verbs at stage 3. This is exactly what one would expect if children are degree-0 learners, and not at all what one would expect if children were sensi-

40

THE DIACHRONY OF VERB SECOND

tive to embedded Domains as they set grammatical parameters. If children are degree-0 learners, as these and other facts suggest, then learning theory is implicated; degree-0 learnability could not follow from the principles and parameters that make up UG and could only follow from some conditions on the way that parameters are set, i.e. from some sort of learning theory. Thus a condition of degree-0 learnability would exist alongside Berwick's (1985) Subset Principle, which must also be construed as part of a learning theory (the Subset Principle says that children first adopt parameter settings which yield smaller sets of sentences). For some arguments for degree-0 learnability, see Lightfoot (1991, chs. 3 and 4). In short, it seems reasonable to suppose that in so-called verb-second languages, where verbs move to I and then to C, the underlying position of the verb is degree-0 learnable whether it precedes its complement, as in the Scandinavian languages and in Yiddish (den Besten and Moed-van Walraven (1986)), or follows them, as in Dutch and German. Furthermore, the relevant parameters are set early, and with no apparent difficulty. From the earliest stages children manifest both object-verb and verb-object order, reflecting two verbal positions, whereas English-speaking children are more consistently verb-object at the two-word stage. In addition, Dutch and German children seem not to make the kinds of errors that English-speaking children make in acquiring inverted forms; at least, such errors are not part of the common lore of language acquisition and they reportedly do not occur. So, failure to invert (12a) or copying instead of substitution (12b) are not typical childhood forms, whereas the equivalent forms are standard for English-speaking children (13). Failure to invert with an initial why, for example, often persists in seven-year-olds (13c), but apparently not in other languages; see Davis (1987) for extensive discussion of the acquisition of English auxiliaries. (12) a. *wat Jan moet schoenmaken? what Jan must clean 'what must Jan clean?' b. * kan hij kan de vloer schoenmaken? can he can the floor clean (13) a. what John must clean? b. can he can clean the floor? c. 'why John must clean the floor? There is a puzzle here. The acquisition of English auxiliary verbs, the only elements to occur in verb-second contexts, is clearly data-driven and involves a significant amount of learning. For example, Newport, Gleitman and Gleitman (1977) showed that a child's ability to use auxiliaries results from exposure to noncontracted, stressed forms in initial position in yes-no questions: the greater the exposure to these subject-auxiliary inversion forms, the earlier the use of auxiliaries in medial position. Richards (1990) demonstrates a good deal of individual variation in the acquisition of English auxiliaries. Nothing in any of the accounts mentioned so far would make English verb-

WHY UG NEEDS A LEARNING THEORY

41

second constructions (i.e. subject-auxiliary inversion) in any way hard to attain. This makes Weinberg's (1990) account interesting because she argues that subject-auxiliary inversion involves marked and highly data-driven operations, thereby explaining why it is attained late in English. However, the equivalent verb-second constructions in Dutch are at least as marked from her perspective (probably more so, because the "doubly-filled Comp filter" has to be relaxed beyond what is required in English grammars), but the work of Clahsen (1990) and others suggests strongly that these constructions are acquired early and without the kinds of errors that English-speaking children produce. So the puzzle remains.

2. The Triggering Experience and Change The approaches of Rizzi (1990a) and Lightfoot (1991) to verb-second phenomena have the virtue of making the presence of the finite verb in C conditional on a certain kind of initial element; in this regard they seem to be theories of the right type. Presumably both analyses would invoke the same triggering experience; that is, hearing utterances which begin with phrasal categories with arbitrary grammatical functions and thematic roles followed by a finite verb. Rizzi's child determines that the verbal head is both prepositional and predicational (his "hybrid" case), and my child determines that there is an "extra" projection beyond IP. Given the scant ideas we have about possible triggers for particular parameters, we cannot know whether the trigger that both analyses need would more plausibly set a parameter relating to predication relations or a parameter relating to structural projections. Since both analyses make the same claims, I assume, about the triggering experience, they would also make the same predictions about the loss of full verbsecond properties historically. In languages which have lost full verb-second properties, like English and French, Rizzi and I would expect a period when the arbitrariness of the initial phrasal category gradually declines and comes to be predominantly the subject of the clause. Such a statistical shift would not reflect a change in people's grammars but just a change in the way that grammars were used. However, one would suppose that the statistical changes eventually would reach a point where the primary linguistic data would trigger a different grammar, and, whatever the nature of the parameter, one would expect the loss of verb-second to take place rapidly and catastrophically, reflecting the new parameter setting. Such are the expectations; what are the facts? The second prediction seems to be correct. It seems to be the case that verb-second constructions dropped out of English and French rather suddenly. Schmidt (1980) shows that verbsecond forms are standard in Chaucer but disappear rapidly after 1400; this view is echoed by van Kemenade (1987). From Old English through Chaucer, finite verbs occurred quite uniformly in C when preceded by an initial pa or ponne (literally 'then', but used effectively as sentence connectives); in a study of Chaucer's prose in the Treatise on the Astrolabe (written for his young son

42

THE DIACHRONY OF VERB SECOND

and therefore in a conversational style) and the Equatorie of Planets, Schmidt (p. 191) noted only one case of tho/thanne which was not followed by inversion (although Mandeville 's Travels, translated from a French source, was less consistent in this regard, and the "very formal," non-narrative prose in Chaucer's translation of Boethius shows still less inversion). However, Schmidt's study of early 15th-century prose (Book of Margery Kempe and Malory's Tale of King Arthur) showed a very different pattern, and after than, the environment historically most conducive to inversion, she found only 18 cases of inversion in 88 clauses (20%) (p. 250). She offers more data supporting the rapid, catastrophic disappearance of verb-second forms after 1400. The lack of a good, continuous prose tradition prior to the mid-14th century suggests some caution, but Vance (1990) has also argued that verb-second constructions disappeared rapidly in French in the 14th century (cf. also Vance's paper in this volume): when movement of finite verbs to C ceased to be obligatory, instances of verbs preceding subjects dropped out quickly. On the first prediction, that the loss of verb second should have been preceded by a steady increase of utterances beginning with subject+verb and a decline of non-subject+verb (while remaining verb second), I can offer some supporting evidence from English but it is far from conclusive. One is bedevilled here again by the lack of a continuous prose tradition. However, Bean's (1983) study of the Anglo-Saxon Chronicle shows 28% of matrix clauses showing subject-verb order in the first three sections, which were probably written in 891, and 41% in the last three sections, which are presumed to have been written contemporaneously with the events they describe (respectively 1048-1066, 1122-1124, and 1132-1140). This increase in subject+verb order is significant but not enormous. Gerritsen (1984:110) reports that modern Dutch, German and Norwegian, all verb-second languages, show about 60% subject+verb order in conversational modes. Similarly Jorgensen (1976) reports spoken Swedish showing 62% of initial subjects in interview contexts and 73% in more formal radio broadcasts of news and commentary (thanks to Kjartan Otto5sson for this reference). These figures suggest that the Chronicle may be showing an artificially high number of non-subject+verb sequences, perhaps due to what many commentators have called its "vivid style." If this is correct and if as few as 40% or even fewer non-subject+verb sequences suffice to set the relevant parameter in the verbsecond mode, then even the later sections of the Chronicle remain well above that threshhold. Recall, however, that Old English showed verb second obligatorily only when clauses were introduced by interrogative or negative phrases; otherwise verb second was just one option, albeit a prevalent one (see Stockwell (1984) for discussion), unlike in Dutch, German and Norwegian, where verb second is generally obligatory. This may reflect an important difference in the grammars and thus in what experience is required to set the relevant parameters. So there is some uncertainty about what it takes to set parameters in the verb-second mode and about how those requirements ceased to be met in English by the beginning of the 15th century.3 However, we do know that

WHY UG NEEDS A LEARNING THEORY

43

there were two major steps in the loss of the old verb-second system and the evolution of the "residual" system of modern English. We know that English ceased to be a full verb-second language in Rizzi's sense by the early 15th century. From this time verbs ceased to occur generally in C, occurring there only when preceded by certain kinds of elements, particularly a +wh feature. When the full verb-second properties were first lost, initial negative phrases also ceased to trigger verb second on any regular basis (compare never did I see such a mess, etc.). Schmidt (1980:209) claims that verb second declines in frequency with initial negative phrases, being re-established in the 16th century. If correct, this suggests that wh- and negative features do not have to be treated in parallel as triggers for verb-second properties. In any case, the first step is the restriction of verb-second forms to certain environments, namely sentences introduced by interrogative (or negative) phrases. In those environments verbs continued to move to C obligatorily, as in earlier English. The restriction of verb second to a narrow class of environments seems to have been complete by the early 15th century. Modern English, however, shows a further restriction: only certain verbs occur in verb-second forms, namely the modals, have and be. Since Emonds (1978), this lexical restriction has been viewed as following from the fact that modern English lacks the V-to-I operation, raising a verb into a position in which it acquires various inflectional features (for tense, person, and number). Modal verbs are base-generated in I and are therefore free to move to C, but a verb cannot move from its D-structure position directly to C without violating the Empty Category Principle (its trace would not be properly governed). The major evidence offered by Emonds and others for this parametric distinction is that English-type languages without the V-to-I operation have their verbs strictly adjacent to their complements (14), whereas languages with the V-to-I operation (like French) allow adverbs, negatives, and floating quantifiers to intervene (15). (14) a. *he watches always/never television on Wednesdays b. *they watch not television c. *they watch all television on Wednesdays (15) a. il regarde toujours/seulement la t616vision le mercredi b. ils regardent pas la television c. ils regardent tous la television le mercredi By this criterion, English used to have the V-to-I operation but lost it. One finds sentences parallel to French (15) (examples from Kroch (1989)): (16) a. I wende wel thys nyght to have deyed. Caxton, The Ryall Book, lines 20-25. 'I managed almost tonight to die'. b. ...if thay do noghte all. Rolle, "The Bee and the Stork," lines 23-24. '...if they don't do everything'.

44

THE DIACHRONY OF VERB SECOND

c. ...that is to seyn, whyl that they liven bothe. Chaucer The Parson's Tale, line 916. '...while they both live'. One reflection of the loss of V-to-I is the transitional occurrence in the 16th and 17th centuries of forms like he not spoke those words, discussed in Visser (1969, sect.1440): "Before 1500 this type is only sporadically met with, but after 1500 its currency increases and it becomes pretty common in Shakespeare's time." Visser cites 57 examples but none involves an auxiliary verb. This suggests that not continues to occur between I and VP, as in Middle and Modern English; the novel forms result from failure of the V-to-I operation (at this stage "Jo support" has yet to become categorical and "affix hopping" may apply across the intervening negative; the forms died out as do became a tense carrier). The loss of V-to-I is dated as mid-16th century by Kroch (1989:222) and as early 17th century by Lightfoot (1991), significantly later than the restriction of verb second to interrogative and negative environments and therefore a separate and distinct change. Before we ask why this second change took place, let us ask what triggers the presence/absence of the V-to-I operation for French/English children. It could, of course, be precisely the data that motivated Emonds to postulate the operation. This, however, is unlikely: adverbs are notoriously flexible in their distribution, being subject to a "transportability convention." Adverbs such as always and rarely may occur in any of the positions indicated in (17). (17)

John Wednesdays

must .

have

watched television

on

It is hard to imagine that failure to hear adverbs precisely between the verb and its complement would trigger the absence of V-to-I for an English-speaking child; similarly, it is unlikely (although not impossible) that French children must hear (and properly analyze) adverbs between a verb and its complement in order to acquire the V-to-I operation. From a diachronic perspective, if the V-to-I operation were triggered in this way, then one would expect the loss of the operation to reflect changes in the use of grammars prior to an actual change in grammars. That is, one would expect the historical record to show a steadily declining number of postverbal adverbs; as the decline reached a certain critical point, there would be a parametric change reflecting the categorical loss of the operation. An alternative scenario would posit a more indirect trigger, for example V-in-C forms. If verbs occur in C, then, since traces must be properly governed, the verbs could only move there via I and therefore there must be a V-to-I operation. Now the relevant difference for French (and Dutch, German, Swedish etc.), on the one hand, and English children on the other is that the former hear V-in-C forms (regardei-vous la television!, (2), (3) etc.), while English children do not. There can be no doubt that these forms are sufficiently robust and salient to be plausible triggers. If the latter scenario is along the right lines, then an explanation for the loss of V-to-I reduces to an explanation for the loss of V-in-C. This might

WHY UG NEEDS A LEARNING THEORY

45

be partially correct, but it cannot be the whole story. We have seen that V-in-C forms came to be greatly curtailed by the early 15th century but this had no apparent effect on the V-to-I operation. After the curtailment of V-in-C to interrogative contexts, V-to-I continued to operate: one still finds forms like (16) robustly attested, negatives like John spoke not these words, and interrogatives like how great and greuous tribulations suffered the Holy Appostyls ...?, all of which involve V-to-I (and I-to-C for the interrogatives). Similarly the loss of general verb second in French had no effect on the attainability of V-to-I, which persists in the present day language even though V-in-C forms are much rarer than in English (occurring only in interrogatives with pronominal subject NPs). This suggests that one needs to look for other changes in the primary linguistic data taking place between 1400 and 1600, which might have the effect of making V-to-I.harder to attain. Two likely candidates are the demise of inflectiorial endings on the verb and the rise of periphrastic do. The inflectional changes were effectively complete by 1400 and helped to distinguish shall, may, must etc. as a distinct subclass of lexical items, instances of I rather than of V (Lightfoot (1991, ch.6)). Therefore, the morphological changes took place too early to affect the primary linguistic data in the relevant period, and one cannot correlate V-to-I entirely with morphological properties (cf. Platzack and Holmberg (1989), who make such a correlation at least for the verb-object Germanic languages). V-to-I persisted for some time after verb morphology had become impoverished and English had lost its rich system of subject-verb agreement.4 Furthermore the mainland Scandinavian languages have lost their verbal morphology but retain full verb-second properties and, therefore, a Vto-I operation.5 Periphrastic do, on the other hand, occurred first at the beginning of the 15th century and steadily increased in frequency until it stabilized into its modern usage by the mid-17th century. This change has been analyzed extensively and Ellegard (1953) shows that the sharpest increase came in the period 1475-1550; for discussion and analysis, see Kroch (1989) and Lightfoot (1991), both of whom reproduce Ellegard's important graph showing the rise of do in different construction-types. Each insertion of a periphrastic do to carry inflectional markers represents a case where the V-to-I operation has not applied, so a steady increase in the distribution of do entails fewer and fewer instances of V-to-I; the two operations are mutually incompatible. The historical facts, then, suggest that lack of strong subject-verb agreement cannot be a sufficient condition for absence of V-to-I despite suggestions along these lines by some authors, but it may be a necessary condition. That is, if a language has strong verbal inflection, it will have V-to-I. Furthermore, if certain individual elements of grammar show strong inflection, then they undergo V-to-I: so English be is generally inflected richly and undergoes V-to-I wherever possible (i.e. whenever I is not otherwise lexically occupied).6 However, there are some forms of English which do not inflect be, and in those cases be does not raise to I. Thus so-called Black English uses George be president now, but the usual negative is George don't be presi-

46

THE DIACHRONY OF VERB SECOND

dent now rather than George ben't president and the uninflected be does not invert with the subject: *be George president now! and *what be George! (Myhill (1988)). Similarly children often use uninflected forms of be, but they do not invert them or use them to the left of not: instead they use do-support forms: did it be funny!, do clowns be a boy or girl!, I don't be angry. Under this view, the possibility of V-to-I not being triggered first arose in the history of English with the loss of rich verbal inflection; similarly in Swedish. That possibility never arose in Dutch, French, German, etc., where verbal inflection remained relatively robust. Despite this possibility, V-to-I continued to be triggered and it occurred in grammars well after verbal inflection had been reduced to its present day level. Although V-to-I continued to be triggered, it did not apply obligatorily as it had in Old and Middle English. There were two alternatives: the use of do as a "dummy" tense carrier and a morphological operation lowering the tense marker on to the verb ("affix hopping"). As already noted, the first option was exercised with steadily increasing frequency from the 15th century onwards. It is impossible to know when the second option of "affix hopping" came into use, because any effects of affix hopping could also be produced by V-to-I. However, it is worth noting that affix hopping or tense lowering could not be a syntactic operation (it would leave an unbound trace) and therefore must be a morphological operation:7 this entails automatically that it applies only to contiguous elements (modulo Visser's (1969) transitional John not spoke those words examples cited above) and that it does not feed other syntactic operations (compare V-to-I, which applies across intervening adverbs and feeds I-to-C operations). In that case it is reasonable to suppose that this morphological operation is generally available, even in grammars which also have the V-to-I operation, and that V-to-I (in grammars which have the operation) applies only where necessary, i.e. where there would otherwise be a stranded affix. We know that "do support" was exercised increasingly during the relevant period, and it seems that it became sufficently frequent that there was no "need" for V-to-I. That is, with the rise of periphrastic do there was no longer anything very robust in the primary linguistic data which required V-to-I, given that the morphological operation was always available. In particular, post-verbal adverbs and quantifiers (16) were not triggers for V-to-I and they simply disappeared quietly. Under this analysis, the absence of V-to-I in modern English grammars is a result of an historical convergence: at the time that verbal inflections were simplified, verbs underwent a mitosis whereby a subset (must, shall, etc.) came to be generated under I and there was in addition an element do which could be analyzed as a "dummy" tense carrier generated in I and which came to be used more and more frequently. As a result, V-to-I ceased to be triggered. Now that we have some idea of how the I-to-C and V-to-I parameters might be set, we can ask why the "residual" I-to-C operation (or "subject-auxiliary inversion") in English should be so hard for children to attain. Weinberg's. (1990) account whereby children must learn to relax the demands of the doubly-filled Comp filter predicts that Dutch and German children should have

WHY UG NEEDS A LEARNING THEORY

47

even more difficulty, because the filter must be relaxed still further than in English. This seems to be contrary to the acquisitional facts discussed earlier, which suggest that Dutch and German children acquire the I-to-C operation relatively early and without the kinds of errors made by English-speaking children. The difference might arise through Rizzi's full vs residual distinction: I-to-C is harder to learn because it is restricted to a narrow class of syntactic environments in English (i.e. to sentences introduced by +wh or a negative phrase). However, the acquisition of French militates against this view. Clark (1985) and others have pointed out that young French children show a preference for verb-subject and even verb-object-subject order over subject-verb in their early utterances. It is unclear whether such forms should be analyzed as cases of subject-verb inversion or as right dislocation of the subject. Whatever the analysis, this preference is surprising because other studies show that sentences with inverted subjects are quite rare in speech addressed to children, unlike in English; Lightbown (1977) noted almost total absence of subject-verb inversion in yes-no questions addressed to children. Questions with inverted subjects (18a) occur much more in "text-book" French than in colloquial forms, where people tend to indicate an interrogative by intonation (18b) or with a wh-in-situ construction (18c) or by an est-ce que form without inversion (18d). (18) a. b. c. d.

que manges-tu? tu manges la poire? tu manges quoi? quand est-ce qu'il vient?

A common error, noted by Clark (1985) is failure to invert with subject pronouns (19), but children produce forms like (20) from an early age. (19) ou ils sont? que ce c'est? (20) ou est cheval? Again, it is unclear whether (20) should be analyzed as a case of V-in-C (cf. ou est le chevall) or as right dislocation of the subject with a missing pronoun (cf. ou il est, le chevalT) (Amy Pierce, p.c.). Clark notes various types of word order errors made by French children but she does not note errors with inversion comparable to those of English-speaking children. She confirms in a personal communication that she has not noted errors of copying (2la) or of failure to invert with a fronted wh-NP (21b). (21) a. *vient il/Jean vient a Toulon? b. *que tu manges? Although French subject-verb inversion is restricted to +wh contexts as in English and, even more narrowly, to contexts where the subject NP is pronominal, and although the construction is attested much less robustly in what children hear, nonetheless French children seem to acquire the I-to-C opera-

48

THE DIACHRONY OF VERB SECOND

tion readily and without the kinds of systematic errors noted in Englishspeaking children. If I-to-C is easy for French children to learn despite being restricted to +wh contexts with pronominal subjects, then the source of the difficulty experienced by English children presumably lies elsewhere, probably in the fact that only a small set of "auxiliary" verbs may occur in C: do, be, have and the modals. It has been argued that these "functional" or "grammatical" items are opaque to children and are in some sense not perceived at early ages. This may be true but it would not explain the difficulty in moving them to C at a time when they occur in medial position, i.e. in I. It seems that it is relatively difficult to attain an operation which is manifested by a small class of lexical items, even though that operation is attested widely and robustly in the triggering experience. This would explain why English-speaking children seem to acquire the I-to-C operation later than Dutch, French and German children, and usually via systematic errors.

3. Conclusion We have considered the loss of the verb-second phenomenon in English and examined the conditions under which the V-to-I operation was lost and the Ito-C operation was restricted first to +wh contexts and then to a small class of lexical items. In doing so, we have learned something about what might trigger these operations in grammars which have them. What emerges is that the triggering experience may be only distantly related to the data that the operation immediately accounts for. That is, what triggers an operation in a child is by no means equivalent to the (positive) data that force a linguist to postulate that operation. This should not be a surprise: if a grammatical operation were triggered by precisely the positive data that the operation accounted for, one would expect languages to change only by the kind of arbitrary fluctuations of population genetics; there would be nothing very systematic and the historical foundations of the discipline laid in the 19th century would be shown to be weak. It means that a modern historical linguist cannot say that some changing phenomenon "is due to" the new parameter setting which accounts for it, although this is often said. Rather, it manifests and provides evidence for that parameter setting; it is due to changes in the triggering experience which in turn entailed the new parameter setting. It may be true that UG consists of principles and parameters, but an account of language acquisition needs to show how those parameters are set. This raises substantive issues and a substantive "learning theory" is required; that is, we need some theorizing about what it takes to set a parameter one way or another. I hope that my discussion of the triggering of the V-to-I operation has provided some relevant suggestions. In general, I believe that one can learn much about this from figuring out the conditions under which a parameter comes to be set differently at some historical stage in a language's development.

WHY UG NEEDS A LEARNING THEORY

49

Notes * Thanks to Norbert Hornstein and Peter Coopmans for comments on a preliminary version of this paper. This paper appeared in slightly different form in C. Jones, ed. Historical Linguistics: Problems and Perspectives, London: Longman. 1. Van Kemenade (1987) adopts a form of Platzack's analysis in her discussion of the history of English and claims that the projection-types changed as the verb-second construction was lost. But she offers no explanation for the change and does not notice the problems with the analysis, as I have outlined them. 2. English children are never exposed to structures equivalent to (lOa). Sentences like who has seen The Hague! have has still in its I position and thus only one element in Comp: CP[who IP[e has VP[seen The Hague] ] ]. Sentences like what have you seen?are immune to the doubly-filled Comp filter because the Comp has no head, i.e. no index percolates to Comp (or to CP in the framework adopted here). See Aoun, Hornstein, Lightfoot and Weinberg (1987) for details. 3. Alternative explanations have been offered for the loss of verb-second constructions, seeking to make it a consequence of another parametric shift. Some have sought to relate it to the earlier object-verb to verb-object word order change, and van Kemenade (1987:221) relates it to the demise of the clitic status of subject pronouns, but the explanations are deficient. First, Swedish shows that a language can acquire verb-object order and maintain full verbsecond status. Second, while van Kemenade claims that the non-clitic status of subject pronouns blocks a verb-second analysis of structures like XF'-pronoun-verb..., she does not show why such pronominal structures did not become obsolete. There is no obvious reason why such structures should have driven verb-second constructions to their death, rather than vice versa. After all, the pronoun forms were a small subset of all the former verb-second constructions and considerations of robustness would lead one to expect verbsecond forms to win out over XP-pronoun-verb constructions. 4. The second person -st ending survived longer than the other endings, but it is hard to imagine that this particular ending was the key to the V-to-I operation. 5. Swedish is sometimes analyzed as lacking the V-to-I operation because it has negatives preceding finite verbs in embedded clauses: ... om Jan inte kopte boken '...if John didn't buy the book' (Platzack and Holmberg (1989)). But this indicates that inte 'not' and other such adverbs occur to the left of I, and does not provide evidence against the application of V-to-I. Occurrence of verbs in C is strong evidence of movement through I, given almost any version of the proper government condition on traces. 6. Have is ambiguous in this regard. It shows no more inflection than regular verbs, but it nonetheless raises to I under certain circumstances, although some of those circumstances are subject to dialectal variation: Kim hasn 't a car vs Kim doesn't have a car, etc.

50

THE DIACHRONY OF VERB SECOND

7. Tense lowering might leave an unbound (and thus illicit) trace which is subsequently erased, for example, by a LF operation raising verbs to I (Chomsky (1986)). Lightfoot (1991, ch.6) criticizes this approach, which fails to explain why tense lowering affects only contiguous elements and why unstressed forms like John did write books (which permit the most economical derivations) entered the language along with the other periphrastic forms and then disappeared rapidly. Emonds (1987) also treats tense lowering as morphological.

References Aoun, J., N. Hornstein, D. Lightfoot and A. Weinberg (1987) "Two Types of Locality." Linguistic Inquiry 18:537-578. Bean, M. (1983) The Development of Word Order Patterns in Old English. London: Croom Helm. Berwick, R. (1985) The Acquisition of Syntactic Knowledge. Cambridge, Mass.: MIT Press. Besten, H. den (1983) "On the Interaction of Root Transformations and Lexical Deletive Rules." In W. Abraham, ed. On the Formal Syntax of the Westgermania, 47-131. Amsterdam: John Benjamins. Besten, H. den and C. Moed-van Walraven (1986) "The Syntax of Verbs in Yiddish." In Haider and Prinzhorn, eds., 111-135. Chomsky, N. (1986) Barriers. Cambridge, Mass.: MIT Press. Cinque, G. (1982) "Constructions with Left Peripheral Phrases, Subjacency and ECP." Ms. University of Venice. Clahsen, H. (1990) "Constraints on Parameter Setting: A Grammatical Analysis of Some Acquisition Stages in German Child Language." Paper presented at the Boston University Conference on Language Development. Clahsen, H. and K.-D. Smolka (1986) "Psycholinguistic Evidence and the Description of V2 Phenomena in German." In Haider and Prinzhorn, eds. Clark, E. (1985) "The Acquisition of Romance, with Special Reference to French." In D. Slobin, ed. The Cross-Linguistic Study of Language Acquisition. Hillsdale, NJ: Lawrence Erlbaum Associates. Davis, H. (1987) The Acquisition of the English Auxiliary System and Its Relation to Linguistic Theory. PhD Dissertation, University of British Columbia. Ellegard, A. (1953) The Auxiliary Do: The Establishment and Regulation of Its Use in English. Stockholm: Almqvist and Wiksell. Emonds, J. (1978) "The Verbal Complex V'-V in French." Linguistic Inquiry 9:151-175. Emonds, J. (1987) "Invisible Category Principle." Linguistic Inquiry 18:613632. Evers, A. (1982) "Twee functionele principes voor de regel 'Verschuif het werkwoord'." GLOT 5:11-30.

WHY UG NEEDS A LEARNING THEORY

51

Fisiak, J., ed. (1984) Historical Syntax. Trends in Linguistics: Studies and Monographs, vol. 23. Amsterdam: Mouton. Gerritsen, M. (1984) "Divergent Word Order Developments in Germanic Languages: A Description and a Tentative Explanation." In Fisiak, ed., 107135. Greenberg, J. (1966) "Some Universals of Grammar with Particular Reference to the Order of Meaningful Elements." In J. Greenberg, ed., Universals of Language. Cambridge, Mass.: MIT Press. Haan, G. de and F. Weerman (1986) "Finiteness and Verb Fronting in Frisian." In Haider and Prinzhorn, eds., 77-110. Haider, H. (1986) "V-Second in German." In Haider and Prinzhorn, eds., 4975. Haider, H. and M. Prinzhorn, eds. (1986) Verb Second Phenomena in Germanic Languages. Dordrecht: Foris. Harlow, S. (1981) "Government and Relativization in Celtic." In F. Heny, ed., Binding and Filtering, 213-255. Cambridge, Mass.: MIT Press. Hellan, L. and K. Koch Christensen, eds. (1986) Topics in Scandinavian Syntax. Dordrecht: Reidel. Hyams, N. (1986) Language Acquistion and the Theory of Parameters. Dordrecht: Kluwer. Jacobsson, B. (1951) Inversion in English with Special Reference to the Early Modern English Period. Uppsala: Almqvist and Wiksell. Jorgensen, N. (1976) Meningsbyggnaden i talad svenska. Lundastudier i nordisk sprakvetenschap c7. Lund: Studentlitteratur. Kayne, R. (1982) "Predicates and Arguments, Verbs and Nouns." GLOW Newsletter 8,24. Kemenade, A. van (1987) Syntactic Case and Morphological Case in the History of English. Dordrecht: Foris. Koopman, H. (1984) The Syntax of Verbs: From Verb Movement Rules in the Kru Languages to Universal Grammar. Dordrecht: Foris. Kroch, A. (1989) "Reflexes of Grammar in Patterns of Language Use." Language Variation and Change 1.3:199-244. Lightbown, P.M. (1977) Consistency and Variation in the Acquisition of French; A Study of First and Second Language Development. PhD Dissertation, Columbia University, NY. Lightfoot, D.W. (1991) How to Set Parameters: Arguments from Language Change. Cambridge, Mass.: MIT Press. Myhill, J. (1988) "The Rise of be as an Aspect Marker in BE Vernacular." American Speech 63:304-326. Newport, E., H. Gleitman and L. Gleitman (1977) "Mother, I'd Rather Do It Myself: Some Effects and Non-effects of Maternal Speech Style." In C. Snow and C. Ferguson, eds., Talking to Children: Language Input and Acquisition. Cambridge: Cambridge University Press. Paardekooper, P.C. (1971) Beknopte ABN-syntaxis. Den Bosch: L.C.G. Malmberg.

52

THE DIACHRONY OF VERB SECOND

Platzack, C. (1986) "COMP, INFL and Germanic Word Order." In Hellan and Koch Christensen, eds., 185-234. Platzack, C. and A. Holmberg (1989) "The Role of AGR and Finiteness in Germanic VO Languages." Working Papers in Scandinavian Syntax 43:51-76. Richards, B.J. (1990) Language Development and Individual Differences: A Study of Auxiliary Verb Learning. Cambridge: Cambridge University Press. Rizzi, L. (1990a) "Speculations on Verb Second." In J. Mascard and M. Nespor, eds., Grammar in Progress; GLOW Essays for Henk van Riemsdijk, 375-385. Dordrecht: Foris. Rizzi, L. (1990b) Relativized Minimality. Cambridge, Mass.: MIT Press. Safir, K. (1981) "Inflection-Government and Inversion." The Linguistic Review 1:417-467. Schmidt, D. (1980) A History of Inversion in English. PhD Dissertation, The Ohio State University. Sproat, R. (1985) "Welsh Syntax and VSO Structure." Natural Language and Linguistic Theory 3.2:173-216. Stockwell, R. (1984) "On the History of the Verb-Second Rule in English." In Fisiak, ed., 575-592. Stowell, T. (1981) Origins of Phrase Structure. PhD Dissertation, MIT. Taraldsen, K.T. (1986) "On Verb Second and the Functional Content of Syntactic Categories." In Haider and Prinzhorn, eds., 7-26. Travis, L. (1984) Parameters and Effects of Word Order Variation. PhD Dissertation, MIT. Vance, B. (1990) "Inversion and Pro-drop in Middle French." Paper presented at the First Generative Diachronic Syntax Conference. Visser, F. Th. (1969) An Historical Syntax of the English Language. Vol. IIIa. Leiden: E.J. Brill. Weinberg, A. (1990) "Markedness Versus Maturation: The Case of SubjectAuxiliary Inversion." Language Acquisition 1:165-194.

3 Two Types of Verb Second in the History of Yiddish Beatrice Santorini Northwestern University

As is well known, all the Germanic languages except modern English exhibit the verb-second phenomenon, as do several (mostly medieval) varieties of Romance. According to this word order constraint, the inflected verb appears as the second overt constituent of a clause—regardless of whether the first constituent is the subject. Thus, in a verb-second language, clauses in which the first constituent is not the subject, exhibit obligatory subject-verb inversion. The earliest and best-known generative analyses of the verb-second phenomenon (Thiersch (1978); den Besten (1983)) were formulated to describe languages like German and Dutch, where root clauses exhibit verb second, but formally subordinate clauses (that is, embedded clauses introduced by an overt complementizer or wft-phrase) do not. Since the phrase structure of these languages is verb-final, the root/subordinate asymmetry they display with regard to the position of inflected verbs is striking; it is illustrated for German in (1) and (2) ("Vf" refers to the inflected verb). (1)

a. Root clause, subject-initial—Vf in second position: Der Junge wird auf dem Weg eine Katze sehen. the boy will on the way a cat see 'The boy will see a cat on the way'. b. Root clause, non-subject-initial—Vf in second position: Auf dem Weg wird der Junge eine Katze sehen on the way will the boy a cat see 'On the way, the boy will see a cat'.

(2) Formally subordinate clause—Vf in final position: ob der Junge auf dem Weg eine Katze sehen wird whether the boy on the way a cat see will 'whether the boy will see a cat on the way'. 53

54

THE DIACHRONY OF VERB SECOND

In the mainland Scandinavian languages, verb second is a root phenomenon as well. Given the head-initial phrase structure of these languages, the root character of verb second is more difficult to establish than in the verb-final West Germanic languages, but clear evidence for it is available based on the position of inflected verbs in relation to sentence adverbs and negation (Holmberg (1986)). More recent studies have revealed that not all verb-second languages exhibit a root/subordinate asymmetry (Adams (1988); Diesing (1988, 1990); Dupuis (1989); Hirschbtihler and Junker (1988); Kosmeijer (1991); Rognvaldsson and Thrainsson (1990); Santorini (1989, 1992); Thrainsson (1986)). One of the languages that exhibits verb second in root clauses and formally subordinate clauses alike is Yiddish, as shown in (3) and (4). (3)

a. Dos yingl vet oyfn veg zen a kats. the boy will on-the way see a cat 'The boy will see a cat on the way'. b. Oyfn veg vet dos yingl zen a kats. on-the way will the boy see a cat 'On the way, the boy will see a cat'. c. * Oyfn veg dos yingl vet zen a kats. on-the way the boy will see a cat Intended meaning: 'On the way, the boy will see a cat'.

(4)

a. oyb dos yingl vet oyfn veg zen a kats whether the boy will on-the way see a cat 'whether the boy will see a cat on the way' b. oyb oyfn veg vet dos yingl zen a kats whether on-the way will the boy see a cat 'whether on the way, the boy will see a cat' c. * oyb oyfn veg dos yingl vet zen a kats whether on-the way the boy will see a cat Intended meaning: 'whether on the way, the boy will see a cat'.

Yiddish is an interesting verb-second language to study diachronically because it has exhibited verb second throughout its recorded history, but has not always allowed verb second in subordinate clauses as it does at present. Rather, we find the same root/subordinate asymmetry in early Yiddish1 texts that is familiar to us from German—a state of affairs that is hardly surprising in view of the origins of Yiddish in the vernacular German spoken by Jews who settled in German-speaking territory in the Middle Ages. In contrast to English and most of the Romance languages, then, which exhibited verb second during the medieval period but lost it (except for residual traces) in the course of their history, Yiddish developed in the opposite direction, extending the domain of verb second to include subordinate clauses. This paper investigates how Yiddish changed from an "asymmetric" to a "symmetrical" verb-second language (that is, from a verb-second language with a root/sub-

TWO TYPES OF VERB SECOND IN THE HISTORY OF YIDDISH

55

ordinate asymmetry to one without such an asymmetry). The analysis presented below is based on an investigation of over 2,200 formally subordinate clauses from about forty texts dating from the early 1400s to the mid-1900s (for further information about the corpus, see Appendices A and B of Santorini (1992)). The paper is organized as follows. SECTION 1 lays out my theoretical assumptions concerning the landing site of the inflected verb in verb-second clauses and nominative-case assignment. Adopting the VP-internal subject hypothesis, I assume that inflected verbs in a symmetrical verb-second language move into a clause-medial Infl and that SpecIP is available for nonsubjects to move into. Further, I extend the analysis of verb movement in asymmetrical verb-second languages put forward by Rizzi (1990b), according to which the inflected verb moves to a "hybrid" head with the feature composition [+C, 4-1]; specifically, I propose that the character of a verb-second language as asymmetric or symmetrical depends (among other things) on whether the highest [+I category in a clause is Comp or Infl. These assumptions form the basis for describing the transition of Yiddish from an asymmetric to a symmetrical verb-second language as the result of two changes. First, since early Yiddish, like the German from which it is descended, was Infl-final—an important factor in the transition was a change in phrase structure from Infl-final to Infl-medial.2 This phrase structure change is discussed in SECTION 2. After briefly reviewing evidence that early Yiddish exhibited both Infl-final and Infl-medial phrase structure, I document the rise of Inflmedial phrase structure in the two main dialects of Yiddish: West and East Yiddish.3 Second, I argue in SECTION 3 that the highest [+I] category in Yiddish went from being Comp to being Infl. Evidence for this analysis comes from the loss of empty categories and the concomitant rise of lexical expletives and non-subjects in SpecIP. I conclude the paper by addressing the relation between the two parametric changes described in the body of the paper, arguing that the generalization of verb second is not a necessary consequence of the emergence of Infl-medial phrase structure in Yiddish.

1. Theoretical Assumptions 1.1 The Landing Site of Verb Movement in Verb-Second Clauses Given their focus on asymmetric verb-second languages, early generative studies proposed to derive the position of the inflected verb in verb-second clauses by verb movement to Comp. Since Comp in formally subordinate clauses is filled by a complementizer, it is not available for substitution by the inflected verb—providing an elegant analysis of the root/subordinate asymmetry illustrated in (1) and (2). The clause-initial constituent in a verb-second clause occupies a non-thematic position immediately preceding the inflected verb in Comp. Adopting current assumptions regarding X' theory (Chomsky (1986:6)), the phrase structures for the verb-second clauses in (1) are then as shown in (5).

56

THE DIACHRONY OF VERB SECOND

(5)

a. Derived structure of (la): [ [Der Junge], [Comp wird,. ] [IP ... ti ... tj ]] CP b. Derived structure of (Ib): [CP [Auf dem Weg]i [Comp wirdj ] [IP... ti.... tj ]]

Given this standard analysis of verb second in terms of verb movement to Comp, word order facts such as those in (4b) are unexpected. In order to resolve the dilemma raised by the acceptability of such clauses, it has been proposed to treat them as instances of CP-recursion (Vikner (1991)) and to give embedded verb-second clauses like (4a) and (4b) the structures in (6). (6)

a. Derived structure of (4a)—assuming CP-recursion: [CPI [comp1 oyb ] [CP2[dos yingl]i [Comp2 vetj [IP... ti. ... tj... ]]] b. Derived structure of (4b)—assuming CP-recursion: [[CPI [comp1 oyb ] [cp2 [oyfn veg]i- [Comp2 vetj ] [IP... ti.... tj ... ]]]

But there are reasons to believe that this attempt to bring embedded verb-second clauses in line with the standard analysis of verb second is not on the right track.4 First, in languages where CP-recursion is independently motivated by word order facts, as in Frisian and Swedish, embedded verb second is restricted to embedded clauses that are asserted (de Haan and Weerman (1986); Holmberg (1986); Platzack (1986))—a restriction not observed in Modern Yiddish (den Besten and Moed-van Walraven (1986); Diesing (1988, 1990); Santorini (1989); latridou and Kroch (1992)). Second, whereas recursive CP's in Frisian and Swedish are islands for extraction, Diesing (1990) has shown that extraction out of non-subject-initial subordinate clauses is possible in Modern Yiddish. For these two reasons, then, CP-recursion is not the proper analysis of embedded verb second in Yiddish. Following much recent work (Adams (1988) for Old French; Diesing (1988, 1990) for Yiddish; Thrainsson (1986), Rognvaldsson and Thrainsson (1990) and Kosmeijer (1991) for Icelandic; Pintzuk (1991) for Old English; cf. also Koopman (1984) for Vata; Chung and McCloskey (1987) for Irish; Sproat (1985) for Welsh), I will assume instead that the inflected verb in symmetrical verb-second languages moves to Infl rather than to Comp and that the constituent preceding the inflected verb in such languages occupies SpecIP. Under this analysis, the derived structures of (4a) and (4b) are as shown in (7). (7)

a. Derived structure of (4a)—assuming verb movement to Infl: [ oyb [Ip [dos ying1]i [Inf1 vetj ] [VP ti tj zen a kats oyfn veg]]] cp b. Derived structure of (4b)—assuming verb movement to Infl: [CP oyb [IP [oyfn veg]i[Inf1 vetj] [VP dos ying1 tj zen a kats ti]]]

Since the constituent in SpecIP may be a non-subject, as in (7b), this analysis relies on (and in turn supports) the VP-internal subject hypothesis, according to which subjects originate within the verb phrase, rather than outside of it in SpecIP.5

TWO TYPES OF VERB SECOND IN THE HISTORY OF YIDDISH

57

1.2 Parameters of Nominative-Case Assignment Adopting the VP-internal subject hypothesis raises the important question of why SpecIP is available for non-subjects in certain languages (like Yiddish), but not others (like English). Drawing upon recent work on the feature content of functional categories, I would like to propose an answer to this question that is based on two parameters: one concerning the position of the highest [+I] category in a clause (Rizzi (1990b)), and the other concerning the feature content of Agr (Platzack and Holmberg (1990)). 1.2.1 The Feature Content ofComp and Infl Rizzi (1990b:382ff.) suggests that just as lexical heads are defined in terms of the features [N] and [V], so functional heads should be defined in terms of the features [C] and [I]. He proposes further that verb-second clauses are headed by a "hybrid" category. In contrast to a "pure" Comp node ([+C, -I]) or a "pure" Infl node ([-C, +1]), the feature composition of this hybrid head is [+C, +1]. Extending work by Laka (1989), Rizzi then motivates verb movement in verb-second clauses by requiring the tense specification in a clause to c-command all the other [+1] categories in the same clause. Rizzi's discussion is restricted to asymmetric verb-second languages, but his analysis can be extended to symmetrical verb-second languages very straightforwardly. With Rizzi, I assume that the position of the highest [+1] category in a verbsecond language is open to parametric variation. In asymmetrical verb-second languages like German, the highest [+1] category is Comp, whereas in non-verb-second languages like English or in symmetrical verb-second languages like Yiddish, the highest [+1] category is Infl.6 Thus, in German, a hybrid [+C, +1] Comp c-commands a pure [-C, +1] Infl, whereas in Yiddish and English, a pure [+C, -I] Comp c-commands a pure [-C, +1] Infl. Further extending Rizzi's analysis, I will assume that in addition to its role in triggering verb movement, the highest [+1] category of a (finite) clause assigns nominative case under government. I define government as in (8), adopting the strict definition of c-command in (9).7 (8)

Definition of government: X governs Y iff (i) X is a lexical head or is [+1], (ii) X c-commands Y, (iii) no barrier intervenes between X and Y, and (iv) minimality is respected.

(9)

Definition of c-command: X c-commands Y iff the node immediately dominating X dominates Y.

58

THE DIACHRONY OF VERB SECOND

7.2.2 The Feature Content ofAgr The distribution of subjects depends not only on the position of the highest [+1] category, but on a second parameter as well: the feature composition of Agr (which I take in turn to be a feature of Infl). Platzack and Holmberg (1990) propose distinguishing between two types of Agr: a type that is [+N] and inherently nominative, and a type that is categorically neutral.8 I adopt their distinction, but instead of attempting to relate it to the overt realization of certain features of agreement morphology, as they do (Platzack and Holmberg (1990:22-28)), I take the distinction between nominative and neutral Agr to be related to the availability of overt nominative case-marking on full noun phrases (the same approach is explored by Cardinaletti and Roberts (1991), Haider (1989), Holmberg (1986), Holmberg and Platzack (1988) and Trosterud (1989)).9 Specifically, I take Agr to be inherently nominative in languages that exhibit overt nominative case-marking on full noun phrases, but neutral in ones that do not. The nominative/neutral Agr parameter then allows us to state the distribution of subjects in terms of the licensing condition in (10). (10) Licensing condition on Agr: Agr must be identified as nominative at S-structure. In languages with inherently nominative Agr, the licensing condition in (10) is satisfied trivially; in languages with neutral Agr, on the other hand, Agr must be identified under antecedent-government (that is, by being c-commanded by a local antecedent bearing nominative case). 1.2.3 The Interaction Between the Two Parameters The two parameters just discussed interact to determine whether SpecIP can be occupied by non-subjects in the following way. Since each of the two parameters has two values, there are four cases to consider in all, depending (i) on whether Agr is nominative or neutral and (ii) on whether the highest [+1] category is Comp or Infl. First, in languages like English, French or Italian, where Agr is neutral and the highest [+1] category is Infl, subjects are assigned nominative case by [+1] in their underlying position according to (8), but unless they move to SpecIP, Agr is not identified as nominative. As a result, SpecIP is restricted to subjects in these languages. Second, in Dutch or the modern mainland Scandinavian languages, where Agr is neutral and the highest [+1] category is Comp, subjects must move to SpecIP for the same reason as in English, French and Italian: for neutral Agr to be identified. Furthermore, if subjects in such languages were to remain in their underlying position, considerations of minimality would prevent them from being assigned nominative case, since Infl, potentially the highest [+1] category, is a closer potential governor than Comp. As a result, subjects remaining in their underlying position would violate the case filter—a violation that can be avoided only if the subject moves to SpecIP. Third, suppose that the highest [+1] category is Comp and Agr is inherently nominative, as in German. Even though

TWO TYPES OF VERB SECOND IN THE HISTORY OF YIDDISH

59

subjects in such a language need not move from their underlying position in order to identify Agr, they are forced to move by the considerations of minimality just described for Dutch and mainland Scandinavian. Finally, consider the case where Agr is inherently nominative and the highest [+1] category is Infl, as is the case in modern Yiddish (and at least for some speakers, in Icelandic). This is the only case in which subjects are able to remain in their underlying position, since they are assigned nominative case there and do not need to raise to SpecIP in order to identify Agr. In such languages, then, non-subjects are free to occupy SpecIP. Independent support for the analysis just presented comes from the correlation between the loss of embedded verb second and the loss of overt case-marking on noun phrases in Old French (Cardinaletti and Roberts (1991:55f.)).10 Table 3.1 summarizes the above discussion concerning nominative-case assignment.

2. The Phrase Structure Position of Infl Infl-final verb-second languages like German are asymmetric verb-second languages par excellence; given the clause-final position of Infl, they cannot possibly be symmetrical verb-second languages. Since the earliest Yiddish texts overwhelmingly exhibited Infl-final phrase structure, a necessary condition in the transformation of Yiddish from an asymmetric to a symmetrical language was a change in its phrase structure from Infl-final to Infl-medial. I have argued in detail elsewhere (Santorini (1992)) that early Yiddish exhibited synchronic variation between Infl-final and Infl-medial phrase structure.11 In this paper, I will therefore only briefly recapitulate the evidence that Inflfinal and Infl-medial phrase structure are both attested in early Yiddish, before documenting the rise of Infl-medial phrase structure in West and East Yiddish. Table 3.1 Distribution of Non-subjects in SpecIP in Various Languages

Type of Agr

Highest [+I] Category

Languages

Non-subjects in SpecIP?

Dutch, Modern Mainland Scandinavian

No

Infl

English, French, Italian

No

Nominative

Comp

German, Early Yiddish

No

Nominative

Infl

Modern Yiddish, Icelandic

Yes

Neutral

Comp

Neutral

60

THE DIACHRONY OF VERB SECOND

2.1 Infl-Final Phrase Structure Many early Yiddish subordinate clauses are not consistent with verb second because more than one constituent precedes the inflected verb. In the simplest case, the inflected verb is in absolute clause-final position, as in (II). 12 (11) a. ds zi droyf givarnt vern that they there-on warned were 'that they might be warned about it' (Bovo 39.6, 1507) b. ven der vatr nurt doyts leyan kan if the father only German read can 'provided only that the father can read German' (Anshel 11, ca. 1534) c. vas er zeyn tag fun zeynm r. gilernt hat what he his day from his rabbi learned has 'what he learned from his rabbi in his day'. (Preface to Shir ha-shirim 2, 1579) In other cases, various well-known rightward movement processes such as PP extraposition, heavy NP shift and verb (projection) raising13 result in superficial word orders in which the inflected verb is after second but before absolute clause-final position. The treatment of such word orders as reflecting Infl-final phrase structure is consistent not only with current generative assumptions, but also with the traditional practice in German philology of taking so-called Spaterstellung of the inflected verb (positions intermediate between second and final) as a variant of Endstellung (absolute clause-final position) rather than of Zweitstellung (second position). 2.2 Infl-Medial Subordinate Clauses Using the distribution of certain diagnostic elements (particles, sentence negation and unstressed pronouns), which must precede uninflected main verbs but are stranded after inflected main verbs, Travis (1984:114) and den Besten and Moed-van Walraven (1986:116-128) provide compelling evidence that modern Yiddish is Infl-medial. Pursuing a classic line of argumentation (Koster (1975); Emonds (1978); cf. also Pollock (1989)), they propose that diagnostic elements are stranded in modern Yiddish because inflected main verbs move across them into a clause-medial Infl node. I have shown in Santorini (1992) that their argument extends straightforwardly to early Yiddish, since the same generalizations obtain concerning the distribution of the diagnostic elements as in the modern language. Thus, in addition to Infl-final subordinate clauses, early Yiddish also exhibited subordinate clauses that are structurally parallel to those discussed for the modern language by Travis and den Besten and Moed-van Walraven. Some examples are given in (12).

TWO TYPES OF VERB SECOND IN THE HISTORY OF YIDDISH

61

(12) a. dz der mensh git erst oyf in di hikh that the human goes first up in the height 'that people first grow in height' (Preface to Shir ha-shirim 6, 1579) b. dz ez iz nit az andri shlekhti bikhr that it is not like other bad books 'that it isn't like other bad books' (Preface to Megilat Ester 2, 1589) c. vi es izt mir zu kit how it is me so cold 'how I feel so cold' (Purim-shpil 424, 1697)

I argue further in Santorini (1992) that in certain cases, subordinate clauses which can apparently be derived either from an Infl-medial base or from an Infl-final base by verb projection raising must in fact be analyzed as Inflmedial, since a verb projection analysis of them is inconsistent with languageinternal evidence from early Yiddish as well as with comparative evidence from other varieties of West Germanic.

2.3 The Rise of Infl-Medial Phrase Structure Having briefly reviewed the evidence that early Yiddish allowed both Inflfinal and Infl-medial phrase structure, I turn now to documenting the rise of the latter over time. I focus separately on West and East Yiddish, distinguishing two types of texts: vernacular and literary. The vernacular texts in my corpus consist of private letters and verbatim transcriptions of court testimony, and they are characterized by pro-drop, variable number agreement with postposed and w/z-moved subjects (Prince (1988:407-409)), and verb-first narrative root clauses—all characteristic features of modern colloquial Yiddish as well. All other texts are classified as literary. Among the literary texts, I have not specially distinguished prose from verse texts since a multivariate analysis fails to reveal a statistically significant correlation of this distinction with the phrase structure position of Infl (Santorini (1989:149)). In order to arrive at conclusive results, I consider only structurally unambiguous clauses in tracking the rise of Infl-medial phrase structure. I have treated all subordinate clauses like those in (11), in which more than one constituent precedes the inflected verb, as Infl-final.14 On the other hand, I have treated subordinate clauses in which the inflected verb occupies second position (henceforth, "Vf-second" clauses) as underlyingly Infl-medial only if they contain a stranded diagnostic element or if they are one of the apparent instances of verb projection raising mentioned at the end of SECTION 2.2. Thus, I have excluded from consideration (i) subordinate clauses containing only a subject and an inflected verb, (ii) Vf-second instances of wh-movement or subject postposing (since the position of the trace of movement cannot be determined), and (iii) Vf-second clauses like the one in (13), where the position of the inflected verb

62

THE DIACHRONY OF VERB SECOND

might reflect underlying Infl-medial phrase structure, but might just as well reflect the combination of Infl-final phrase structure and rightward movement processes that are independently motivated in early Yiddish (namely, verb raising and PP extraposition). (13) dz zi verdn bshirrnt fun irh bitrh peyn that they become protected from their bitter pain 'that they might be protected from their bitter pain' (Purim-shpil 876, 1697) For each time period, dialect and style, I determined / (the number of unambiguously Infl-final subordinate clauses) and m (the number of unambiguously Infl-medial subordinate clauses) and calculated the relative frequency of unambiguously Infl-medial subordinate clauses (defined as m/ (f + m)).15 Figure 3.1 shows the rise of Infl-medial phrase structure over time in West Yiddish. It is evident that Infl-medial phrase structure is a marginal option in West Yiddish vernacular texts. For those time periods for which we have both vernacular and literary texts—up to the late 1600s—all but two literary texts closely reflect the vernacular ones, with the percentage of Infl-medial subordinate clauses never exceeding 5%.16 After about 1700, Infl-medial subordinate clauses in West Yiddish literary texts rise in relative frequency, though they never come to make up the majority of subordinate clauses. Given the attested close relationship in the earlier periods of West Yiddish between literary and vernacular usage, I take this development to reflect a rise in Inflmedial phrase structure in the vernacular.

Figure 3.1 Relative Frequency of Infl-medial Subordinate Clauses in West Yiddish

TWO TYPES OF VERB SECOND IN THE HISTORY OF YIDDISH

63

Figure 3.2 Relative Frequency of Infl-medial Subordinate Clauses in East Yiddish Figure 3.2 gives the corresponding results for East Yiddish. Two aspects of these results are noteworthy. First, in contrast to West Yiddish literary texts, East Yiddish literary texts do not faithfully reflect vernacular usage for the period before 1700 for which both literary and vernacular texts are available. Infl-medial subordinate clauses are already well attested in East Yiddish vernacular texts from before 1700, but they are still marginal in the literary texts from the same period, where they occur with a frequency comparable to that found for West Yiddish. Second, after about 1700, the great majority of structurally unambiguous subordinate clauses in the literary texts have suddenly become Infl-medial. The first fully modem texts (that is, ones without any Infl-final subordinate clauses at all) date back to the early 1700s, though the change from Infl-final to Infl-medial phrase structure takes roughly another century to go to completion, with Infl-final phrase structure surviving into the first half of the 1800s as an increasingly marginal option in literary East Yiddish.17 The quantitative differences between West and East Yiddish evident in Figures 3.1 and 3.2 are consistent with and in turn confirm conclusions reached by traditional philologists like Max Weinreich who have studied the history of Yiddish phonology and morphology within its wider social context. According to Weinreich (1980), the Jews of western and eastern Europe belonged to a single cultural community until the 1700s. The linguistic correlate of this unity was the existence of a fairly uniform supraregional literary standard based on West Yiddish, which Weinreich calls "Written Language A." While Written Language A remained essentially impervious to Slavic influence even in eastern Europe, vernacular Yiddish gradually developed into the two main dialects of West and East Yiddish, the earliest reports of differences between

64

THE DIACHRONY OF VERB SECOND

the two dialects date back to the beginning of the 1600s, and by the middle of the 1700s, they had diverged in speech to the point of causing difficulties in mutual comprehension (Weinreich (1980:284)). This growing rift between West and East Yiddish led to the emergence in the course of the 1700s of a literary language based on East Yiddish vernacular usage—what Weinreich calls "Written Language B." It is this replacement of Written Language A by Written Language B that we see reflected in the quasi-instantaneous rise of Infl-medial subordinate clauses in Figure 3.2.

3. The Character of SpecIP As we have seen in SECTION 2.1, early Yiddish allowed Infl-final subordinate clauses that are inconsistent with verb second because too many constituents precede the inflected verb. In addition, early Yiddish also allowed subordinate clauses that are inconsistent with verb second; because too few (overt) constituents precede the inflected verb—namely, none at all. Thus, we find subordinate clauses as in (14). (14) a. dz iz mir ydue dz... that is me-DAT known that 'that it is known to me that...' (Court testimony 197, ca. 1643) b. dz zoyln zikh dran kern manin un' veybr that shall REFL thereon turn men and women 'that men and women shall take heed of this' (Duties n.p., 1704) The subordinate clauses in (14) are unambiguously Infl-medial: (14a) contains the stranded pronoun mir, and (14b) is one of the apparent instances of verb projection raising mentioned in SECTION 2.2. Such clauses therefore exhibit one of the properties of a symmetrical verb-second language—that it have Inflmedial phrase structure (so that the inflected verb can occur in second position), but not the other—that SpecIP be occupied by a single overt constituent. The modern Yiddish counterparts of subordinate clauses like those in (14) (henceforth, "gap-Vf' clauses) are completely unacceptable.18 In this section, I analyze the loss of gap-Vf clauses in the history of Yiddish as the result of a parametric change in the position of the highest [+I] category of a clause— a change that has consequences for the distribution of empty categories as well as for the availability of SpecIP for lexical expletives and non-subjects. 3.1 Empty Categories in SpecIP Let us assume that impersonal and subject postposing constructions as in (14) have the structure in (15), where SpecIP is filled by an empty expletive. (15) [ C P ... [ IP e [ I N F L Vf i ] [ V P ...t i . ...]]]

TWO TYPES OF VERB SECOND IN THE HISTORY OF YIDDISH

65

In the spirit of recent work (Platzack and Holmberg (1990:21); Rizzi (1986:524); Safir (1985:206)), I assume further that empty expletives are subject to the licensing condition in (16).19 (16) Licensing condition on empty expletives: Empty expletives must be governed by a case-assigner. Given (16) and the minimality clause in the definition of government in (8) that it relies upon, the occurrence of gap-Vf subordinate clauses like (14) forces us to conclude that the highest [+1] category in early Yiddish is Comp, just as it is in German. On the other hand, since Comp is not a [+1] category in modern Yiddish, gap-Vf subordinate clauses are correctly ruled out in the modern language. It is worth noting that the unacceptability of gap-Vf clauses is not due to the unavailability of empty expletives in Yiddish. As the contrast in (17) shows, empty expletives are available; however, they are restricted to the underlying subject position, precisely as expected, given the c-command clause in (8). (17) a. oyb in shtub iz e varem whether in room is warm 'whether it is warm in the room' b. *oyb e iz varem in shtub whether is warm in room Intended meaning: 'whether it is warm in the room'. A second class of gap-Vf subordinate clauses that has been lost in the history of Yiddish consists of indirect questions and free relative clauses as in (18). (18) a. vil zehn ... velkhr ihudi vil mir nitn vr zi want see which Jew will me force for them tsu msptn to [exert-influence?] '[I] want to see which Jew will force me to [exert influence?] on their behalf (Court testimony 74, 1555) b. velkhr which

vert gifindll eyn hurg fun di hrugi gzirh hn"l will find a body from the massacre said da velin zi eyn talr gebin there want they a taler give 'they would give a taler to whoever found the body of a person killed in said massacre' (Court testimony 200, 1638)

c. ver veyz di simnim ... da zal zagin who knows the signs there shall say 'whoever knows the signs...he shall say' (Court testimony 171, 1640)

66

THE DIACHRONY OF VERB SECOND

d. ver vert leyann di kinh vert zikh gvis bzinh who will read the lament will REFL surely decide tshubh tfilh tsdkh tsu tan penance prayers charity to do 'whoever reads the lament will surely decide to repent, pray and give to charity'. (Preface to Kine, 1648) Prince (1988) argues that subject extraction in indirect questions and free relative clauses in modern Yiddish proceeds from a postverbal position rather than from SpecIP. If her analysis is correct and extends to early Yiddish, then SpecIP in the subordinate clauses in (18) is occupied by an empty expletive, and this class of examples can be assimilated to the subject-postposing cases illustrated by (14b). On the other hand, if subject extraction in (18) proceeds via SpecIP, leaving a trace, then the acceptability of these clauses in early Yiddish and the complete unacceptability of their counterparts in modern Yiddish can be derived by adopting a conjunctive formulation of the ECP and imposing a head-government requirement on all traces. In early Yiddish, traces in SpecIP can satisfy the head-government requirement, since Comp is a headgovernor by virtue of bearing the feature [+I], but they can no longer do so in modern Yiddish. Thus, under either analysis, the loss of this class of gapVf clauses supports an analysis according to which the highest [+I] category went from being Comp to being Infl. 3.2 Expletive es and Non-Subjects Whereas SpecIP becomes unavailable for empty categories in the transition from Early to Modern Yiddish because it is no longer governed, the lexical expletive es emerges in the same position. In the modern language, the lexical expletive is in complementary distribution with empty expletives: as shown by the contrast between (19) and (20), it can appear in SpecIP but not VPinternally.20 (19) a. ... volt er gepaskent vi es iz gut far got would he decided how it is good for God '...he would decide in God's favor' b. az es iz gekumen in zayn shtot a groyser magid that it is come in his town a great teacher 'that a great teacher came into his town' c. Ikh veys nit ver es hot nekhtn telefonirt. I know not who it has yesterday telephoned 'I don't know who called yesterday.' d. Ikh veys nit, vemen es hobn nekhtn farbetn zayne I know not who-ACC it have yesterday invited his tate-mame. parents 'I don't know who his parents invited yesterday'.

TWO TYPES OF VERB SECOND IN THE HISTORY OF YIDDISH

67

(20) a. *...volt er gepaskent vi far got iz es gut would he decided how for God is it good Intended meaning: '...he would decide in God's favor' b. *az in zayn shtot iz es gekumen a groyser magid that in his town is it come a great teacher Intended meaning: 'that a great teacher came into his town' c. *Ikh veys nit, ver nekhtn hot es telefonirt. I know not who yesterday has it telephoned Intended meaning: 'I don't know who called yesterday.' d. *Ikh veys nit, vemen nekhtn hobn es farbetn zayne I know not who-ACC yesterday have it invited his tate-mame. parents Intended meaning: 'I don't know who his parents invited yesterday'. The unacceptability of the clauses in (20) shows that the lexical expletive in Yiddish, as in other languages, is restricted to ungoverned positions.21 We would therefore expect expletive es not to occur in SpecIP in early Yiddish subordinate clauses, given that it would be governed by [+1] Comp, but to become available in that position as soon as Infl becomes the highest [+I] category. This expectation is borne out by the data in my corpus: es in SpecIP is not attested in subordinate clauses in the earliest Yiddish sources and first appears in the mid-1600s. Note that the same change that turns SpecIP into an ungoverned position— namely, the loss of the hybrid character of Comp—also allows nominativecase assignment to the underlying subject position. We would therefore expect SpecIP to become available for non-subjects at the same time as it becomes available for expletive es. Again, this expectation is borne out by the data: non-subjects in SpecIP are first attested in Yiddish subordinate clauses in the second quarter of the 1600s.22

4. Conclusion The diachronic data discussed in this paper raise an intriguing and difficult question—namely, whether the loss of hybrid Comp can be considered a necessary consequence of the emergence of Infl-medial phrase structure. In order to address this question, let us ask ourselves what evidence a child acquiring an Infl-medial verb-second language has for positing hybrid Comp. The first source of evidence comes from root/subordinate asymmetries. In Inflmedial verb-second languages without verb movement to Infl.like the modern mainland Scandinavian languages, the failure of the verb to move to Infl in subordinate clauses containing sentence negation or sentence adverbials results in verb-third word orders that are inconsistent with verb second. The

68

THE DIACHRONY OF VERB SECOND

word order asymmetry between such verb-third subordinate clauses and verb second root clauses therefore provides evidence that verb second in these languages involves verb movement to a hybrid Comp. By contrast, in an Inflmedial language with verb movement to ML, like Yiddish, the root/subordinate asymmetry resulting from verb-third subordinate clauses does not arise. It might therefore be argued that although the first Infl-medial clauses in Yiddish may have been produced by a grammar with hybrid Comp, they would provide children with no positive evidence for it and that at least some children would therefore analyze Infl in Infl-medial clauses as the highest [+I] category. Indeed, it might be argued further that in the absence of evidence for verb movement to Comp in Infl-medial clauses, economy of derivation (Chomsky (1991)) would force all children to acquire a grammar without hybrid Comp. The appearance of lexical expletives and non-subjects in SpecIP in embedded clauses would then follow directly, given the inherently nominative character of Agr in Yiddish. The attraction of this argument is that it relates the generalization of verb second to embedded contexts in Yiddish to the emergence of Infl-medial phrase structure (cf. Vikner (1994)). However, it neglects to take into account the second source of positive evidence for hybrid Comp in Infl-medial clauses—namely, the existence of gap-Vf clauses. Given this evidence, we would expect to find languages just like modern Yiddish—that is, with Infl-medial phrase structure, verb raising to Infl and overt case morphology on full NP's—except that they do not allow lexical expletives or non-subjects in SpecIP. West Yiddish is a candidate for such a language, though the evidence is murkier than one would like. As we saw in SECTION 2.3, West Yiddish exhibited some Infl-medial phrase structure, but did not in general exhibit embedded verb second. One exceptional text—the Purim-shpil of 1697—does contain Vf-second subordinate clauses with lexical expletives in SpecIP. However, it is not clear to what the occurrence of these clauses is attributed. On the one hand, the scribe who copied the manuscript of the Purlm-shptl from an earlier West Yiddish version, now lost, was from Cracow, a transition zone between West and East Yiddish (Weinryb (1936:417)), and the occurrence of embedded verb second in the manuscript might therefore reflect interference from the scribe's native dialect. On the other hand, of all the West Yiddish texts in the corpus, the Purim-shpil contains the largest absolute number of subordinate clauses that are unambiguously Infl-medial as well as the largest number that are consistent with Infl-medial phrase structure. Since embedded verb second is not common even in languages whose grammars allow it, the presence of embedded verb second in the Purim-shpil and its absence in

the other West Yiddish texts might simply reflect the statistically expected values for each text. Clearly, further evidence from other late West Yiddish texts would be needed to decide the issue. Unfortunately, however, since written West Yiddish was being replaced by German during precisely this period, such texts are rare, and I do not know of any besides the ones that I have investigated. While the case of West Yiddish brings us up against the limits of the historical record, there is synchronic evidence from insular Scan-

TWO TYPES OF VERB SECOND IN THE HISTORY OF YIDDISH

69

dinavian that embedded verb second is not a necessary concomitant of Inflmedial phrase structure in languages with verb raising to Infl and overt case morphology. Some speakers of Icelandic accept gap-Vf subordinate clauses (Maling (1990:84-85); SigurSsson (1990:51-56); Kosmeijer (1991:192, fn. 1)); in addition, not all speakers accept embedded verb second in non-CP-recursion contexts. Further, Rohrbacher (1993) notes that some speakers of Faroese allow verb raising to Infl in non-CP-recursion contexts, while topicalization in these environments is virtually ruled out (Barnes (1987:24)). This is precisely the pattern that we would expect if Comp were hybrid for some speakers of Icelandic and Faroese.23 While there is evidence, then, for the possibility of a parametric constellation of Infl-medial phrase structure, verb raising to Infl, overt case marking and hybrid Comp, the historical development in (East) Yiddish and the synchronic variation in the acceptability of embedded verb second reported for Icelandic do suggest that this constellation is not a stable one. A possible explanation is that the crucial evidence—namely, gap-Vf clauses—occurs so rarely that at least some speakers either fail to register it or treat it on a par with performance errors. Of the 131 unambiguously Infl-medial subordinate clauses in the Early Yiddish portion of the corpus, four (3.0%) are gap-Vf clauses; of these, two (1.5%) are indirect questions and two are that-clauses with an empty expletive in first position. For Icelandic, SigurSsson (1990:51) gives the following figures: 1) two out of 542 subordinate clauses (0.4%) in Modern Icelandic are gap-Vf clauses (both are that-clauses with an empty expletive in first position). 2) In a sample of Old Icelandic, 23 out of 2988 subordinate clauses are gap-Vf clauses (0.8%); of these, seven (0.2%) are thatclauses with an empty expletive in first position. At least for the Icelandic case, these frequencies are comparable to the frequencies with which performance errors occur in Modern English (Anthony Kroch, p.c.) and Early Yiddish (Santorini (1992:609, fn. 17 and 19, 612, fn. 20, 615, fn. 22)). Thus, while not required by Universal Grammar, it is likely that Infl-medial phrase structure correlates with the availability of non-hybrid Comp in Yiddish and Icelandic as a result of the quantitative character of the input to acquisition.

Appendix Table 3.2 provides statistics on the subordinate clauses in my corpus. For each text (or group of texts), the columns list the number of subordinate clauses with the following properties: 1. unambiguously Infl-final (more than one constituent precedes the inflected verb) 2. unambiguously Infl-medial (instance of stranding or apparent verb projection raising) 3. overt subject or trace of local subject extraction in headed relative clause in first position, Vf in second position

70

THE DIACHRONY OF VERB SECOND

4. empty expletive or trace of other subject extraction in first position, Vf in second position (the number of gap-Vf clauses with empty expletives in first position is given in parentheses) 5. non-subject in first position, Vf in second position 6. lexical expletive in first position, Vf in second position The total number of structurally unambiguous subordinate clauses is the sum of (1) and (2). The percentage of (2) over the sum of (1) and (2), upon which Figures 1 and 2 are based, is given after column (2) (figures rounded to the nearest percentage point). The total number of clauses consistent with the verbsecond constraint is the sum of (3), (5) and (6), and the total number of subordinate clauses considered for each text (or group of texts) is the sum of (1), (3), (4), (5) and (6). Table 3.2 West Yiddish Vernacular Texts

(4)

(5)

(6)

1 (1%) 6 1 (1%) 6 0 (0%) 10 0 (0%) 10 0 (0%) 3 0 (0%) 3 0 (0%) 6 4 (6%) 39 4 (6%) 39 West Yiddish Literary Texts

KO) 1 (0) 0(0) 0(0) 0(0) 1(0) 1(0) 1(0) 1(0)

0 0 0 0 0 0 0 0 0

0 0 0 0 0 0 0 0 0

(3) 3 3 21 4 6 31 33 54 7 12 33 139 14 19 4 37

(4) 0(0) 0(0) 4(0) 0(0) 0(0) 4(0) 3(0) 1(0) 0(0) 0(0) 1(0) 5(0) 6(1) 0(0) 0(0) 6(1)

(5) 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0

(6) 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0

(1) Court testimony, 1440-1489 Total: 1440-1489 Court testimony, 1540-1589 Total: 1540-1589 Court testimony, 1590-1639 Letters, 1619 Total: 1590-1639 Court testimony, 1640-1689 Total: 1640-1689

(2)

(3)

80 80 20 20 1 19 20 67 67

(1) 3 Mints, third quarter of 1400s 3 Total: 1440-1489 73 Bovo, 1507 15 Goetz, 1518 32 Anshel, ca. 1534 120 Total: 1490-1539 Preface, Shir ha-shirim, 1579 35 77 Shir ha-shirim, 1579 10 Officials, 1588 9 Preface, Megilat Ester, 1589 68 Megilat Ester, 1589 199 Total: 1540-1589 22 Kine, 1648 32 Messiah, 1666 4 Witzenhausen, 1677 58 Total: 1640-1689

(2) 1 (25%) 1 (25%) 3 (4%) 0 (0%) 1 (3%) 4 (3%) 7 (17%) 4 (5%) 0 (0%) 3 (25%) 1 (1%) 15 (7%) 1 (4%) 1 (3%) 0 (0%) 2 (3%)

71

TWO TYPES OF VERB SECOND IN THE HISTORY OF YIDDISH

(1) 66

Purim-shpil, 1697 Total: 1690-1739 Zeeb, 1740 Moses, ca. 1750 Total: 1740-1789

66 4 22 26

(2) 13 (16%) 13 (16%) 2 (33%) 9 (29%) 11 (30%)

(3) 70 70 6 44 50

(4) 4(0) 4(0) 1(0) 0(0) 1(0)

(5) 0 0 0 0 0

(6) 3 3 0 0 0

East Yiddish Vernacular Texts Court testimony, 1540-1589 Total, 1540-1589 Court testimony, 1590-1639 Total, 1590-1639 Court testimony, 1640-1689 Total, 1640-1689

(1) 6 6 25 25 10 10

(2)

(3)

(4)

(5)

(6)

2 (25%) 2 (25%) 11 (31%) 11 (31%) 8 (44%) 8 (44%)

8 8 64 64 46 46

1(0) 1(0) 3d) 3(1) 1 (1) 1(1)

0 0 0 0

1

0 0 0 0 2 2

(4)

(5)

(6)

0 0 0 0 0

0 0 0 0 0 0 0 0 0 0 0 5 0 1 0 0 6 3 5 9 9 2 28 22 8 16 46

1

East Yiddish Literary Texts

(1) 1 Meineket Rivkah, ca. 1550 Total: 1540-1589 1 Sam Hayyim, 1590 35 Preface, Lev Tov, 1620 32 Lev Tov, 1620 74 Preface, Ha-magid, 1623-1627 25 Hagen Abraham, 1624 32 Total: 1590-1639 198 Vaad, 1671 10 Ashkenaz un Polak, ca. 1675 38 Total: 1640-1689 48 Vilna, 1692 3 Sarah, first half of 1700s 0 Ellush, 1704 1 Duties, 1716 3 Poznan, 1717 1 Total: 1690-1739 8 Nakhman, ca. 1800 0 Naphthali, 1803 3 Geography, 1818 7 Ukraine, 1834 1 El Male Rakhamim, 1834 0 Total: 1790-1839 11 Judah, 1848 0 Grine Felder, ca. 1910 0 Royte Pomerantsen, 1947 0 Total: 1840-present 0

(2) 1 (50%) 1 (50%) 1 (3%) 3 (9%) 3 (4%) 3 (11%) 1 (3%) 11 (5%) 1 (9%) 5 (12%) 6 (11%) 16 (84%) 5(100%) 8 (89%) 8 (73%) 4 (80%) 41 (84%) 31(100%) 32 (91%) 20 (74%) 34 (97%) 11(100%) 128 (92%) 27(100%) 42(100%) 52(100%) 121(100%)

(3)

1 0(0) 1 0(0) 15 0(0) 14 3(0) 42 4(0) 11 3(0) 13 3(1) 95 13(1) 9 0(0) 18 2(1) 27 2(1) 31 0(0) 16 0(0) 15 0(0) 13 1(1) 9 0(0) 84 1(1) 50 0(0) 61 2(0) 52 0(0) 93 2(0) 31 1(0) 287 5(0) 64 0(0) 91 0(0) 116 0(0) 271 0(0)

1

2 3 0 2 2 0 0 0 0 0 0 5 0 4 1 2 12 0 4 4 8

72

THE DIACHRONY OF VERB SECOND

Notes It is a pleasure to thank the following people for their help in writing this paper: Caroline Heycock, Anthony Kroch, Janet Pierrehumbert, Bernhard Rohrbacher and Sten Vikner for helpful discussion, Ian Roberts for making me work faster than I ever thought possible, and Sue Johnson, Beth Levin and Adair Waldenburg for their refreshing senses of humor and perspective. Needless to say, I am responsible for all errors and shortcomings that remain. 1. For the purposes of this paper, two periods need to be distinguished in the history of Yiddish: "Early Yiddish" (from the earliest texts dating back to the 1400s until about 1800) and "Modern Yiddish" (from about 1800 on). For a more detailed periodization, see Weinreich (1980). 1. Yiddish verb phrases also appear to have changed from head-final to headinitial. Further research is needed to resolve this issue, since modern Yiddish still allows OV word order fairly productively. For discussion, see den Besten and Moed-van Walraven (1986), GeilfuB (1991), Hall (1979), and Santorini (1993b). 3. West Yiddish was spoken on German-speaking territory and in transition areas in contact with Slavic, and East Yiddish was originally spoken on Slavicspeaking territory only. The earliest continuous West Yiddish texts date from the late 1300s, and the language has been essentially extinct since 1800, at least in written form, having given way to German. The West Yiddish sources on which this paper is based cover a period from the early 1400s to the mid1700s. East Yiddish texts are more recent than West Yiddish ones, and the oldest ones that I have examined date from the early 1500s. East Yiddish continues to be spoken and written today both in eastern Europe and by the descendants of eastern European Jews who migrated abroad, particularly to North and South America. Since West Yiddish is for all practical purposes extinct, I will generally refer to Modern East Yiddish simply as "Modern Yiddish." 4. Arguing against Vikner (1991), Cardinaletti and Roberts (1991:6) raise the objection that analyzing embedded verb-second clauses in symmetrical verbsecond languages as instances of CP-recursion provides "no way to avoid unlimited recursion of C°." However, since unlimited recursion of Comp must be avoided in any event, even in asymmetric verb-second languages like German or residual verb-second languages like English, it is difficult to see what would preclude extending whatever mechanism is invoked to rule it out in those cases—say, a distinction between two types of Comp, as suggested by Cardinaletti and Roberts themselves—to the case of symmetrical verb-second languages. 5. The VP-internal subject hypothesis was originally proposed by Fillmore (1968) and McCawley (1970), and much recent work in phrase structure theory has been devoted to arguing in favor of it (Fukui (1986); Fukui and Speas (1986); Kitagawa (1986); Koopman and Sportiche (1991); Kuroda (1987); Manzini (1988); Sportiche (1988); Zagona (1988)). Several different variants of the

TWO TYPES OF VERB SECOND IN THE HISTORY OF YIDDISH

73

hypothesis have been proposed and discussed in the literature. For the purposes of this paper, the precise position in which subjects originate is irrelevant as long as it is dominated by a maximal projection of the verb, and I will simply refer to the position in question as the subject's "underlying" position. 6. Kosmeijer (1991) puts forward a similar analysis, taking the finiteness operator [+F] to be a feature of Comp in German, but of Infl in Icelandic. 7. The analysis of nominative-case assignment proposed here is inspired by that which is put forward by Platzack and Holmberg (1990), according to whom nominative case in the verb-second languages is assigned under government by the finiteness operator [+F]. However, there is an important difference between their and my analysis with regard to the notion of government: according to Platzack and Holmberg (1990:7(9)), governors m-command their governees, whereas according to the definition in (8), they must satisfy the stricter condition of c-command; cf. Deprez (1989:366ff.), Koopman and Sportiche (1991:229f.) and Rizzi (1990a:30-32) for relevant discussion. As we will see in SECTION 3.1, the distribution of empty expletives in modern Yiddish provides empirical justification for a c-command over an m-command clause in the definition of government. 8. Platzack and Holmberg's distinction between nominative and neutral Agr has roughly the same empirical consequences as the distinction made by Koopman and Sportiche (1991) between nominative-case assignment under Spec-head agreement and under government. 9. Platzack and Holmberg's idea that embedded verb second is related to rich subject-verb agreement, incorrectly leads one to expect embedded verb second in a language like Italian. The unacceptability of embedded verb second in Italian is a serious difficulty for their approach and supports the case-marking approach instead. 10. Old Spanish is problematic for my analysis, since it allows embedded verb second but fails to exhibit overt case-marking on full NP's (Josep Fontana, p.c.). 11. Synchronic variation between Infl-final and Infl-medial phrase structure is also attested in Old English texts (Pintzuk (1991) contra van Kemenade (1987)). 12. The first and second numbers following each reference indicate the page, verse or line number and the year of the text, respectively. Where no page number is available, I indicate this by "n.p."; where the exact date is unknown, I give a range of dates or an estimate from the secondary literature. In contrast to modern Yiddish, vowels are not consistently represented in early Yiddish, and I have not attempted to insert them where they are absent in the original texts; the transliteration conventions I follow are set out in Santorini (1989:15-17). 13. Verb projection raising is a variant of verb raising, a common and muchstudied phenomenon in West Germanic. Verb raising permutes the order of auxiliary verbs and the heads of their infinitival complements. In verb projection raising, the post-auxiliary sequence includes arguments and adjuncts of

74

THE DIACHRONY OF VERB SECOND

the infinitive head in addition to the infinitive itself. See Evers (1975), Lotscher (1978), Zaenen (1979), den Besten and Edmondson (1983), Haegeman and van Riemsdijk (1986), and Kroch and Santorini (1991) for discussion. 14. See Santorini (1992:614-616) for more detailed discussion. 15. The raw numbers on which Figures 3.1 and 3.2 are based are given in the Appendix. 16. The relative frequency of Infl-medial subordinate clauses in the two exceptional texts is comparable to East Yiddish vernacular usage of the same period. Their author was from Cracow, a transition zone between West and East Yiddish, and they are both prefaces to other works (one of these texts is by the same author but contains hardly any Infl-medial subordinate clauses). A plausible explanation for the unusually high incidence of Infl-medial phrase structure is that the usage of the author in prefaces tends to reflect his vernacular usage. 17. Modern Yiddish still preserves some relics of Infl-final phrase structure—for instance, the negative polarity idiom vos a/der hor vert iz 'in the slightest bit' (literally, 'what a/the hair is worth'). Their status is comparable to that of such relics of verb-final phrase structure in modern English as Indictments do not a conviction make. 18. The only gap-Vf clauses that have remained acceptable in modern Yiddish are headed relative clauses with local extraction of the subject. In such clauses, SpecIP may exceptionally be filled by an empty category—the trace of the subject. I will not attempt to give an analysis here of what licenses the empty initial position in these clauses. It is worth noting, however, that the exceptional character of local subject extraction is not peculiar to Yiddish, as is evidenced by the violations of the that-trace filter in precisely this context in English; cf. Deprez (1989, ch.4) for discussion. 19. In addition to the formal licensing condition in (16), empty expletives must also satisfy an identification requirement regarding their feature content (cf. Rizzi 1986:520(41)). I do not wish to deny that the availability of empty expletives in Yiddish might be related to properties like the "richness" of subject-verb agreement; however, the details of the identification requirement are not at issue here. 20. Without the postverbal es, the sentences in (20) are all acceptable. 21. Kosmeijer (1991) discusses parallel facts in Icelandic and German. For an analysis of the complementary distribution of empty and lexical expletives, see Cardinaletti (1990). 22. As in SCCTION 2.3,1 have excluded from consideration all Vf-second instances of w/z-movement or subject postposing; the clauses under consideration are therefore all bonafide instances of movement to SpecIP and cannot be ana-

lyzed as instances of stylistic fronting. Stylistic fronting is a process attested in Icelandic, Faroese and Medieval Scandinavian in which a non-subject constituent occupies clause-initial position, resulting in word orders that are superficially consistent with verb second. Since stylistic fronting is restricted to subject-gap clauses (Maling (1990:77-81)) and affects heads rather than maximal projections, there is reason to distinguish it from movement of non-

TWO TYPES OF VERB SECOND IN THE HISTORY OF YIDDISH

75

subjects to SpecIP (contra Cardinaletti and Roberts (1991)). The analysis of stylistic fronting is a matter of some debate in the literature (J6nsson (1991); Platzack and Holmberg (1990); Rognvaldsson and Thramsson (1990); Cardinaletti and Roberts (1991); Santorini (1994); Fontana (1993)). 23. Those speakers of Icelandic and Faroese who accept both gap-Vf clauses and embedded verb second in non-CP-recursion contexts must be assumed to have access to two grammars: one in which Comp is hybrid, and one in which it is not (Santorini (1992, 1994)).

References Adams, M.P. (1.988) "Embedded pro." In J. Blevins and J. Carter, eds. Proceedings ofNELS 18, 1-21. Amherst, Mass.: GSLA. Barnes, M. P. (1987) "Some Remarks on Subordinate-Clause Word-Order in Faroese." Scripta Islandica 38:3-35. Besten, H. den (1983) "On the Interaction of Root Transformations and Lexical Deletive Rules." In W. Abraham, ed. On the Formal Syntax of the Westgermania, 47-131. Amsterdam: John Benjamins. Draft version circulated in 1977. Besten, H. den and J.A. Edmondson (1983) "The Verbal Complex in Continental West Germanic." In W. Abraham, ed. On the Formal Syntax of the Westgermania, 155-216. Amsterdam: John Benjamins. Besten, H. den and C. Moed-van Walraven (1986) "The Syntax of Verbs in Yiddish." In H. Haider and M. Prinzhorn, eds. Verb Second Phenomena in Germanic Languages. Publications in Language Sciences 21:111135. Dordrecht: Foris. Cardinaletti, A. (1990) Pronomi nulli e pleonastici nelle lingue germaniche e romanze: Saggio di sintassi comparata. PhD Dissertation, Universiti di Padova e Venezia. Cardinaletti, A. and I. Roberts (1991) "Clause Structure and X-second." To appear in W. Chao and G. Horrocks, eds. Levels, Principles and Processes: the Structure of Grammatical Representations. Berlin: Foris/de Gruyter. Chomsky, N. (1986) Barriers. Cambridge, Mass.: MIT Press. Chomsky, N. (1991) "Economy of Derivation and Representations." In R. Freidin, ed. Principles and Parameters in Comparative Grammar, 417454. Cambridge, Mass.: MIT Press. Chung, S. and J. McCloskey (1987) "Government, Barriers and Small Clauses in Modern Irish." Linguistic Inquiry 18:173-237. Deprez, V. (1989) On the Typology of Syntactic Positions and the Nature of Chains: Move a to the Specifier of Functional Projections. PhD Dissertation, MIT. Diesing, M. (1988) "Word Order and the Subject Position in Yiddish." In J. Blevins and J. Carter, eds. Proceedings of N ELS 18, 124-140. Amherst, Mass.: GSLA.

76

THE DIACHRONY OF VERB SECOND

Diesing, M. (1990) "Verb Second in Yiddish and the Nature of the Subject Position." Natural Language and Linguistic Theory 8:41-79. Dupuis, F. (1989) L'expression du sujet dans les subordonnees en ancien francais. PhD Dissertation, Universite de Montreal. Emonds, J.E. (1978) "The Verbal Complex V'-V in French." Linguistic Inquiry 9:151-175. Evers, A. (1975) The Transformational Cycle in Dutch and German. PhD Dissertation, University of Utrecht. Distributed by the Indiana University Linguistics Club. Fillmore, CJ. (1968) "The Case for Case." In E. Bach and R.T. Harms, eds. Universals in Linguistic Theory, 1-88. New York: Holt, Rinehart and Winston. Fontana, J.M. (1993) Phrase Structure and the Syntax of Clitics in the History of Spanish. PhD Dissertation, University of Pennsylvania. Fukui, N. (1986) A Theory of Category Projection and its Applications. PhD Dissertation, MIT. Fukui, N. and M. Speas (1986) "Specifiers and Projection." In N. Fukui, T. Rapoport and E. Sagey, eds. MIT Working Papers in Linguistics 8:128172. GeilfuB, J. (1991) "Jiddisch als SOV-Sprache." In Verb- und Verbphrasensyntax, 3-17. Arbeitspapiere des Sonderforschungsbereichs 340: "Sprachtheoretische Grundlagen fur die Computerlinguistik," Bericht Nr. 11. Also in Zeitschrift fur Sprachwissenschaft 9 (1990). Haan, G. de and F. Weerman (1986) "Finiteness and Verb Fronting in Frisian." In H. Haider and M. Prinzhorn, eds. Verb Second Phenomena in Germanic Languages, Publications in Language Sciences 21:77-110. Dordrecht: Foris. Haegeman, L. and H. van Riemsdijk (1986) "Verb Projection Raising, Scope and the Typology of Rules Affecting Verbs." Linguistic Inquiry 17:417— 466. Haider, H. (1989) "0-Tracking Systems—Evidence from German." In L. Maracz and P. Muysken, eds. Configurationality. The Typology of Asymmetries. Studies in Generative Grammar 34:185-206. Dordrecht: Foris. Hall, B. L. (1979) "Accounting for Yiddish Word Order, or What's a Nice NP Like You Doing in a Place Like This?" In J.M. Meisel and M.D. Pam, eds. Linear Order and Generative Theory. Amsterdam Studies in the Theory and History of Linguistic Science IV. Current Issues in Linguistic Theory 7:263-287. Amsterdam: John Benjamins. Hirschbiihler, P. and M.-O. Junker (1988) "Remarques sur les sujets nuls en subordonnees en ancien et en moyen francais." Revue quebecoise de linguistique theorique et appliquee 7:63-84. Holmberg, A. (1986) Word Order and Syntactic Features in the Scandinavian Languages and English. PhD Dissertation, University of Stockholm. Holmberg, A. and C. Platzack (1988) "On the Role of Inflection in Scandinavian Syntax." Working Papers in Scandinavian Syntax 42:25-42.

TWO TYPES OF VERB SECOND IN THE HISTORY OF YIDDISH

77

latridou, S. and A.S. Kroch (1992) "The Licensing of CP-Recursion and its Relevance to the Germanic Verb-Second Phenomenon." Working Papers in Scandinavian Syntax 50:1-24. Jonsson, J.G. (1991) "Stylistic Fronting in Icelandic." Working Papers in Scandinavian Syntax 48:1-43. Kemenade, A. van (1987) Syntactic Case and Morphological Case in the History of English. Dordrecht: Foris. Kitagawa, Y. (1986) Subject in Japanese and English. PhD Dissertation, University of Massachusetts, Amherst. Koopman, H. (1984) The Syntax of Verbs. From Verb Movement Rules in the Kru Languages to Universal Grammar. Dordrecht: Foris. Koopman, H. and D. Sportiche (1991) "The Position of Subjects." Lingua 85:211-258. Kosmeijer, W. (1991) "The Distribution of bad: an Argument Against a Generalized COMP Analysis." In H.A. Sigur5sson, ed. Papers from the 12th Scandinavian Conference of Linguistics, 182-193. Reykjavik: Linguistic Institute, University of Iceland. Koster, J. (1975) "Dutch as an SOV Language." Linguistic Analysis 1:111-136. Kroch, A.S. and B. Santorini (1991) "The Derived Constituent Structure of the West Germanic Verb Raising Construction." In R. Freidin, ed. Proceedings of the Princeton Workshop on Comparative Grammar, Current Studies in Linguistics 20:269-338. Cambridge, Mass.: MIT Press. Kuroda, S.-Y. (1988) "Whether We Agree or Not: A Comparative Syntax of English and Japanese." In WJ. Poser, ed. Papers from the Second International Workshop on Japanese Syntax, 103-143. Stanford: CSLI. Laka, I. (1989) "Constraints on Sentence Negation: The Case of Basque." MIT Working Papers in Linguistics 10:199-216. Lotscher, Andreas (1978) "Zur Verbstellung im Zurichdeutschen und in anderen Varianten des Deutschen." Zeitschrift fur Dialektologie und Linguistik 45:1-29. Maling, J. (1980) "Inversion in Embedded Clauses in Modern Icelandic." Islenskt mal og almenn mdlfrceSi 2:175-193. Reprinted in J. Maling and

A. Zaenen, eds. Modern Icelandic Syntax. Syntax and Semantics 24:7191. San Diego: Academic Press. Manzini, M.R. (1988) "Constituent Structure and Locality." In A. Cardinaletti, G. Cinque, and G. Giusti, eds. Constituent Structure, 157-201. Dordrecht: Foris. McCawley, J.D. (1970) "English as a VSO Language." Language 46:286-299. Pintzuk, S. (1991) Phrase Structures in Competition: Variation and Change in Old English Word Order. PhD Dissertation, University of Pennsylvania. Platzack, C. (1986) "The Position of the Finite Verb in Swedish." In H. Haider and M. Prinzhorn, eds. Verb Second Phenomena in Germanic Languages. Publications in Language Sciences 21:27-47. Dordrecht: Foris.

78

THE DIACHRONY OF VERB SECOND

Platzack, C. and A. Holmberg (1990) "The Role of AGR and Finiteness in Some European VO Languages." Paper presented at the 1989 GLOW Colloquium, Utrecht. Ms., University of Lund. Pollock, J.-Y. (1989) "Verb Movement, UG, and the Structure of IP." Linguistic Inquiry 20:365-424. Prince, E.F. (1988) "Yiddish Wh-clauses, Subject-postposing, and Topicalization." Proceedings of ESCOL 6:403-415. Rizzi, L. (1986) "Null Objects in Italian and the Theory of Pro," Linguistic Inquiry 17:501-557. Rizzi, L. (1990a) Relativized Minimality. Cambridge, Mass.: MIT Press. Rizzi, L. (1990b) "Speculations on Verb-Second." In J. Mascaro and M. Nespor, eds. Grammar in Progress: GLOW Essays for Henk van Riemsdijk, 375-385. Dordrecht: Foris. Rognvaldsson, E. and H. Thralnsson (1990) "On Icelandic Word Order Once More." In J. Maling and A. Zaenen, eds. Modern Icelandic Syntax. Syntax and Semantics 24:3-40. San Diego: Academic Press. Rohrbacher, B. (1993) "V-AGR-Raising in Faroese." Proceedings of the Third Meeting of the Formal Linguistic Society of Midamerica, 281-296. Bloomington, Indiana: Indiana University Linguistics Club. Safir, K. (1985) "Missing Subjects in German." In J. Toman, ed. Studies in German Grammar. Studies in Generative Grammar 21:193-229. Dordrecht: Foris. Santorini, B. (1989) The Generalization of the Verb-Second Constraint in the History of Yiddish. PhD Dissertation, University of Pennsylvania. Santorini, B. (1992) "Variation and Change in Yiddish Subordinate Clause Word Order." Natural Language and Linguistic Theory 10:595-640. Santorini, B. (1993) "Das Jiddische als OV/VO-Sprache." Linguistische Berichte 123:230-245. Santorini, B. (1994) "Some Differences Between Icelandic and Yiddish." In N. Hornstein and D. Lightfoot, eds. Verb Movement, 87-106. Cambridge: Cambridge University Press. Sigur5sson, H.A. (1990) "Verb-Initial Declaratives and Verb Raising in Icelandic." In J. Maling and A. Zaenen, eds. Modern Icelandic Syntax, Syntax and Semantics 24:41-69. San Diego: Academic Press. Sportiche, D. (1988) "A Theory of Floating Quantifiers and Its Corollaries for Constituent Structure." Linguistic Inquiry 19:425-449. Sproat, R. (1985) "Welsh Syntax and VSO Structure." Natural Language and Linguistic Theory 3:173-216. Thiersch, C. (1978) Topics in German Syntax. PhD Dissertation, MIT. Thrdinsson, H. (1986) "VI, V2, V3 in Icelandic." In H. Haider and M. Prinzhorn, eds. Verb Second Phenomena in Germanic Languages, Publications in Language Sciences 21:169-194. Dordrecht: Foris. Travis, L. deMena (1984) Parameters and Effects of Word Order Variation. PhD Dissertation, MIT. Trosterud, T. (1989) "The Null Subject Parameter and the New Mainland Scandinavian Word Order: A Possible Counterexample from a Norwe-

TWO TYPES OF VERB SECOND IN THE HISTORY OF YIDDISH

79

gian Dialect." In J. Niemi, ed. Papers from the Eleventh Scandinavian Conference of Linguistics, 1:87-100. Joensuu, Finland: University of Joensuu. Vikner, S. (1991) Verb Movement and the Licensing of NP-Positions in the Germanic Languages. PhD Dissertation, Universite de Geneve. Revised version. Vikner, S. (1994) "Finite Verb Movement in Scandinavian Embedded Clauses." In N. Hornstein and D. Lightfoot, eds. Verb Movement, 117-147. Cambridge: Cambridge University Press. Weinreich, M. (1980) History of the Yiddish Language. Translated by S. Noble and J. Fishman. Chicago: University of Chicago Press. Weinryb, B. (1936) "Zur Geschichte des alteren judischen Theaters (uber die Leipziger Hs. des Ahasveros-Esther-Spiels.)." Monatsschrift fitr die Geschichte und Wissenschaft des Judentums 80:415-424. Zaenen, A. (1979) Infinitival Complements in Dutch." In Papers from the 15th Regional Meeting of the Chicago Linguistic Society, 378-389. Zagona, K. (1988) Verb Phrase Syntax: a Parametric Study of English and Spanish. Dordrecht: Kluwer.

4 The Locus of Verb Movement in Non-Asymmetric Verb-Second Languages: the Case of Middle French Monique Lemieux and Fernande Dupuis Universite du Quebec a Montreal

1. Introduction In this article, we discuss the derivation of the verb-second phenomena in Old French and Middle French (hereafter OF and MidF) in a model of grammar where the subject is generated internal to the VP and where the Spec of IP (or AgrP) is either an A or an A-bar position. We will provide arguments for a V-to-I derivation of this phenomena for both main and embedded clauses arguing that CP will be projected only when necessary (cf. Diesing (1990)). In the spirit of Chomsky (1989), a V-to-I derivation of the verb-second constraint corresponds to the minimal derivation, and thus enables us to minimize the function of transformations. We propose an analysis where verb second has not been lost as a rule but is a consequence of the interaction of very general principles of grammar. One important feature of this article will be to give an explicit characterization of a small class of discursive elements triggering verb second in root clauses in order to provide some new insights into the putative asymmetry between main and embedded clauses. Our analysis will be based on the following premises: (i) It will be shown that MidF, like OF, exhibits the main characteristics of a verb-second language, and that the theory that the subject is base-generated inside VP makes interesting predictions for both stages of the language; (ii) contrary to what has been proposed so far (Adams (1987a,b, 1988a,b); Vance (1988); Roberts (1992)) we will suggest that the licensing of the subject in verb-second clauses does not involve double verb raising to CP, that is V-to-I and V-to-C. More importantly, we will show that the facts addressed in this paper entail theoretical assumptions concerning the function of some illocutionary morphemes involved in verb second. The discussion will proceed as follows: SECTION 2 presents the data illustrating the verb-second characteristics of OF and MidF. SECTION 3 focuses on 80

THE LOCUS OF VERB MOVEMENT

81

the similarities and dissimilarities among verb-second languages such as Yiddish and Icelandic, showing that there are sufficient similarities between OF, MidF and other non-asymmetric verb-second languages to justify a unitary analysis of this sub-group without positing verb movement to C. We present our analysis in SECTION 4, where we elaborate on the relation between Nominative-Case assignment and agreement; finally, in SECTION 5, we examine a few predictions of our analysis, focusing on two main points. First we give an answer to the question "Why is verb-second less productive in embedded clauses than in root clauses?" Second, we show how the proposed analysis explains the violations of the verb-second constraint characterizing what we take as a true violation.

2. Middle French as a Verb-Second Language The verb-second status of Old French has been well known since Thurneysen (1892). While the situation is fairly uncontroversial for the 12th and 13th centuries, facts are less obvious for MidF (14th and 15th centuries).1 Despite some differences between these two stages of the language, we will demonstrate in this section that MidF shares one of the most important characteristics of OF in that it allows null and postverbal subjects when the preverbal position is filled by any maximal category.

2.1 XP V Constructions: Null and Postverbal Subject This can be observed in (1), (2) and (3) below, which represent typical constructions with null and postverbal subjects (in italics). Null subject (1) Atant regarda pro contreval la mer, (...) Then looked (he) down at the sea 'Then he looked down at the sea'

Berinus, I, p. 240

Postverbal full NP (2)

Longtemps

fu

ly roys Ellnas

For a long time was King Elinas

en la montaigne (...) Melusine, p. 14. on the mountain

Postverbal pronoun (3) Et puis qu'il m' assailli, mains n'en pouoyey'e faire que de And because he me attacked but only could I do that moy deffendre Berinus, I, p. 285 myself defend 'And because he attacked me, what could I do but defend myself.' It should be pointed out that such structures are not a marginal phenomenon in MidF, since they represent nearly half of the root declarative sentences.

82

THE DIACHRONY OF VERB SECOND

It is commonly held that the verb-second constraint of root clauses is derived through CP by two verb movements, V-to-I and V-to-C while another constituent fills the Spec position (cf. Adams (1987a,b, 1988b,c); Vance (1988, 1989); Hirschbiihler and Junker (1988); Roberts (1992); Hulk and van Kemenade (this volume)). Our main goal in this article will be to challenge this view and to motivate, following Dupuis (1989), a V-to-I derivation for root as well as for embedded verb-second sentences in both OF and MidF. In the analysis that will be presented below, the characteristics of the examples (l)-(3) are accounted for by moving (or base-generating) an XP to the Spec position immediately dominated by IP.

2.2 Preverbal Subject When comparing the different subject positions, that is preverbal, postverbal and null, as illustrated in Table 4.1 below, there can be little doubt that MidF shows a good deal of variation. Of course, one could raise the legitimate question whether the subject in preverbal position is still the reflection of the verb-second requirement rather than the manifestation of a new development into the basic SVO order. This is a complex issue. As we will demonstrate in SECTION 4 evidence required to capture the right generalization underlying verb-second violations rests on the assumption that MidF is still verb-second. For the moment, let us adopt the view that, when the verb-second constraint is still active in a language, a structure where the [Spec.IP] is filled by a preverbal subject can be analyzed as a verb-second structure. This is the position adopted for MidF in Roberts (1992). This will also be our position for the data drawn from a corpus of six texts covering the 14th and 15th centuries. Table 4.1 summarizes the distribution of the verb-second constructions in the period under study. The first column of the table gives the number of constructions with an XP in preverbal position, in other words the number of null and postverbal subjects. Table 4.1 Verb Second in Middle French %

Total

336

91

371

126

419

93

452

47

71

266

83

319

113

57

175

345

85

404

CNNA

152

64

130

346

87

397

Memoires

214

90

73

377

91

416

1027

380

682

2089

89

2359

Nom. Subj. V Pron. Subj. V Subtotal

Texts

XPV

Berinus

189

40

107

Melusine

211

82

Policie

148

QJM

TOTAL

THE LOCUS OF VERB MOVEMENT

83

The second and the third column indicate the number of preverbal subjects, nominal or pronominal, immediately preceding the tensed verb. The residual cases represent violations of the verb-second property. We will discuss the importance of such data in SECTION 5. Leaving aside for the moment the details which we will take up in SECTION 3, we will posit, along the lines of Diesing (1990) (see also Rognvaldsson and Thr&nsson (1990)) that, in the model of grammar where the subject is generated internal to the VP and where the Spec of IP is either an A or an A-bar position, examples (l)-(3) are derived via a single verb movement to 1° and via the movement of some XP into the Spec of IP as shown in (4):

3. Similarities and Dissimilarities of Verb-Second Languages At a purely descriptive level, it is easier to find differences among the verbsecond languages than common points among these languages. The term verb second is itself no more than a useful descriptive characterization of a large number of languages where the verb of a root declarative clause is usually in second position as is the case, for instance, in most of the Germanic languages. Such languages also share the property that any constituent can appear before the verb, including the subject. Since it has been argued in generative grammar that a CP analysis provides an interesting explanation of the differences between the verb-second languages and any other language that does not share this property, the generalization has come naturally that all verb-second languages are best analyzed as derived by movement of V-to-C. In this section, we show that there are at least two groups of verb-second languages, and that even among the Germanic languages, it is not evident that a uniform derivation should be postulated. If crucial arguments against a derivation to CP are

84

THE DIACHRONY OF VERB SECOND

to be postulated for some verb-second languages which share superficial characteristics, this will motivate an alternative solution, especially if these arguments weaken the presumed advantages of the CP analysis. Two of these arguments are based on facts related to extraction from embedded clauses, and embedded clauses in the subjunctive mood. In this section we show that, regarding the analysis of these facts, OF and MidF differ from German but pattern with Yiddish and Icelandic.

3.1 Typical Verb-Second Constructions: the Germanic Languages Considering the typological observations that any analysis must explain, there are some striking differences among the verb-second languages: A.

B.

Some of the verb-second languages obey the so-called root/embedded asymmetry while others do not. There is asymmetry when the position of the verb is triggered by the verb-second constraint in root clauses while the embedded clauses represent the basic word order. Two languages that do not exhibit this asymmetry have been widely discussed in the literature: Yiddish (Diesing (1990)) and Icelandic (Rognvaldsson and Thrainsson (1990); Sigur5sson (1990)). Some verb-second languages are strictly constrained as to the possibility of dropping the subject: this is the case with German, a semipro-drop language where an expletive is the only subject that can be omitted; omission is less constrained in some other dialects, such as Bavarian (cf. Bayer (1984)), where the second person can be omitted and Zurich dialect (Cooper and Engdahl (1989)), freely omitting the first and the second persons singular.

The first characteristic is the most important one for our discussion since the main argument in favor of a verb-raising to CP analysis has always been the asymmetry between root and embedded clauses.2 As for the question of B above, OF and MidF seem to be the most permissive of the verb-second languages and since the pro-drop contexts at least in main clauses are strictly verb-second, we will use this characteristic in our demonstration, but it is important to notice that pro-drop and verb-second are not necessarily related.

3.2 The Asymmetry in Germanic Languages 3.2.1 The Case of Yiddish Diesing (1990) proposes a V-to-I analysis for the verb-second phenomena in Yiddish. This author argues that Yiddish is a language that does not show the well-known asymmetry of German or Dutch between root and subordinate clauses. Indeed, even a quick glance at the data in (5) (examples from Diesing (1990:43-44)) enables one to check that the verb second is possible in both types of clauses. The alternation in (5) illustrates the verb-second effect in a root structure.

THE LOCUS OF VERB MOVEMENT

85

Yiddish (5)

a. Ikh wel avekshikn dos bukh. I will away-send the book

(Diesing (1990), example (Id))

b. Dos bukh shil ikh avek. the book sent I away 'I will send away the book'.

(Diesing (1990), example (lc))

By comparing the sentences (6) and (7), one can observe that Yiddish is an SVO language and not SOV like German. German (6)

Sigrid glaubt daB Waltraud wahrscheinlich das Buch gekauft hat Sigrid believes that Waltraud probably the book bought has (Diesing (1990), example (2b)) 'Sigrid believes that Waltraud has probably bought the book.'

Yiddish (7)

Avrom gloybt az Max shikt avek dos bukh 'Avrom believes that Max sent away the book' (Diesing (1990), example (3b))

In Yiddish, a topicalized XP can appear before the finite verb in any embedded clause as is the case in the following sentences where vayn 'wine', shabes bay nakht 'Saturday night' and hayntike tsaytn nowadays' trigger the verbsecond effect. (8)

a. Ir zolt visn You (pi) should know ken men makhn fun can one make from

zayn, mayne libe kinderlekh, az vayn be my dear children that wine troybn oykn. grapes also (Diesing (1990), example (5a)) 'You should know, my dear children, that one can make wine from grapes also'.

b. Der yid vos shabes bay nakht vet Khayim zen the man that Saturday at night will Khayim see 'The man that Khayim will see Saturday night'. (Diesing (1990), example (35b)) c. Es iz a shod vos hayntike tsaytn kenen azoy fil mentshn it is a shame that today's time can PRT many people nit leyenen even not read 'It is a shame that nowadays so many people can't even read'. (Diesing (1990), example (5b))

86

THE DIACHRONY OF VERB SECOND

According to Diesing, the fact that Yiddish does not show the same distribution as German is to be explained as a parametric choice for the landing site of the verb: V-to-I rather than V-to-C. A second parameter is also responsible for the dual nature of [Spec.IP]: this position is able to function either as an A-bar position as or an A position. 3.2.2 The Case of Icelandic Icelandic word order offers additional evidence for splitting verb-second languages into at least two different subgroups. Based upon similar observations as those given for Yiddish, Rognvaldsson and Thrainsson (1990) (hereafter R&T) argue that the V-to-C rule does not apply in Icelandic, although V-to-I does. Notice first the strict verb-second property of the typical root alternation where a topicalized XP is fronted as in (9): Icelandic (9)

a. Eg hef aldrei hitt Mariu I have never met Mary (ACC)

(R&T, example (2a))

b. Mariu hef eg aldrei hitt Mary (ACC) have I never met 'I have never met Mary'.

(R&T, example (2b))

As in Yiddish, verb-second structure can be observed in Icelandic embedded clauses: Icelandic (10) J6n efast um ao a morgun fari Maria snemma afajtur John doubts that tomorrow get Mary early up 'John doubts that Mary will get up early tomorrow.' (R&T, example (32a)) As similar facts in Old and MidF will be discussed extensively in the following sections, let us adopt R&T and Diesing's analysis without further elaboration. In both cases, interesting predictions are drawn from the base-generation of the subject inside VP and the assumption of a single verb movement to I.

3.3 The Case of Old and Middle French and the So-called Asymmetry While numerous analyses of the verb-second phenomenon have been provided since Thiersch (1978), most linguists in the generative framework give an account of the root/embedded asymmetry by a double movement of the tensed verb: to the head of IP, then to the head of CP. Adams (1987a,b) has shown that this analysis makes interesting predictions for OF and MidF. The

THE LOCUS OF VERB MOVEMENT

87

analysis she proposed capitalizes on the Germanic asymmetry as it involves raising of the verb to the head of CP in main clauses and of the subject to the position of [SpecJP] in embedded clauses, the verb remaining in 1°. To explain the fact that pro is encountered in some subordinate clauses with a class of verbs subcategorized for complementizer que and some degree clauses, Adams argues for a defective Comp [C, CP], an analysis already proposed for German by Haider (1986) and for Frisian by de Haan and Weerman (1986). However, according to Hirschbiihler and Junker (1988) and Dupuis (1988), verb-second effects are not restricted to the que-clauses. Rather the phenomenon can also occur in clauses headed by a w/i-element. In Adams' (1988a,b) analysis, this is to be explained by the fact that the Spec of IP, being an A or an A-bar position in OF and MidF, is available as a host for any XP in embedded w/i-clauses, including an expletive pro. 3.3.1 Some Problems: Solution in a CP Analysis At first sight, the approach just mentioned seems to deal with the major facts attested in the languages under study. In the following, we will give arguments showing that the defective CP analysis is not well motivated for the embedded verb-second. The first argument concerns the possibility of extracting from verb-second clauses, the second one is related to the clearly subordinate character of the subjunctive que-clause. To account for the verb-second effect in Frisian subordinate clauses, de Haan and Weerman argue for two different structures. To justify their analysis, they show that different conditions are linked to the possibility of extracting an element from an embedded clause. Extraction is possible in sentence (l1b), adapted from de Haan and Weerman, since a pre-Comp position is available. Frisian (11) a. Hy sei [CP dat dizze oersetting net maklik lest] He said that this translation not easy reads (de Haan and Weerman (1986:87)) b. Hokker oersettingj sei hy [cp dat ti net maklik lest] Which translation said he that not easy reads Extraction cannot occur when the upper Comp (C') is filled as in (12b). An adapted structure for this example is given in (13): (12) a. hy sei [c- dat [CP dizze oersetting; lestj [Iptitj net maklik]]] he said that this translation reads not easy b. * hokker oersettingi sei hy [c dat [CP ti lestj [IP ti tj net maklik]]]

which translation said he

that

reads

not easy

88

THE DIACHRONY OF VERB SECOND

Generally, extraction is limited to adjacent domains, it cannot cross more than one cyclic node. The ungrammaticality of (12b) can thus be explained if the defective upper C' blocks extraction, the resulting structure being excluded as an ECP violation. Long movements are seldom found in older stages of French but as can be seen in (14a,b), non-w/i-objects (anemis 'enemies' icestui convenant 'this convenant') may be extracted out of the subordinate clause to serve as the topic of the higher main clause. (14) a. car

anemis^ pense

il bienfcptique

because enemies believes he well that 'because he believes that there are enemies'

[IP ce soit t; ]] Queste, 112,1 there is

b. Icestui convenanti volunsnos [ CP t i que [IP vos asseurez ti alsi ]] Vill., 188 This convenant want we that you assure as well 'We want you to enter into convenant as well'. In (15) the modifier plus 'no longer' is pulled out from a Spec position in the lower IP to the Spec of CP suggesting that a doubly articulated CP is required. Notice also that the embedded clause is in the subjunctive mood, whose relevance will be discussed below.

THE LOCUS OF VERB MOVEMENT

89

(15) Pour tant se souvent ne vous voy, Nevertheless if often (NEG) you see (Is SUBJ) Pensez vous [CP plusi que [IPvostre soye ti ?]]

Orleans, p. 30 Think you no longer that yours be (Is SUBJ) 'Nevertheless if I do not see you often / Do you think that I am no longer yours'.

Moreover, if we say that embedded verb-second clauses in OF have a structure parallel to (13) we should expect extraction to be blocked. As (16a) shows, a direct object, fa 'this' is moved out of the subordinate clause where an infinitival PP, a faire 'to do' has been fronted thus creating a verb-second effect. (16) a. fa( ne sai

je [ CP t i qu' [ IP a fairej ayez ti tj ]] Erec, 211 (Skarup, p. 183) this not believe I that to do will have 'I don't believe that you will have to do this'

b. *cai

ne sai

je [c qu' [CP a fairej ayez [Ip ti tj ]]

Since movement is cyclic, in a derivation such as (16b) there is no way out for the object in a defective CP analysis because the CP embedded under the defective C' would be necessarily filled by the topic XP and the raised verb. This entails a two-CP projection. We take this to be a strong argument for a V-to-I analysis of the verb-second effect. In addition, as mentioned earlier, verb-second word order can also take place in subordinate clauses headed by a wft-element. Notice that a proposal which restricts movement to IP is quite compatible with this possibility. If this is so, the simplest analysis is one that uses the Spec of IP as a landing site for the preverbal element in subordinate clauses mirroring the verb-second effect. Another point should be added to clarify the status of subordinate queclause. The French subjunctive mood expressed by the verb in embedded clauses is strictly dependent upon the higher predicate. Obviously, there exists a strong dependence relation between both clauses such that the embedded one cannot be interpreted as a "root embedded clause." In this respect, Medieval French is no different from Modern French. We might therefore expect subjunctive clauses to behave similarly to the other types of embedded clauses discussed so far and to show instances of the verb-second effect. Consider (17) below: (17) a. Et luy dirent que bien a haste a sa mere And to him said (3p SUBJ) that quickly to his mother venist CNNA, LXXVII, p. 459 should come (3s SUBJ) 'And they said to him that he should come quickly to his mother' b. que vous tant de bien n'i avez ou vostre chief mettre ne puissiez Palatinus, p. 46 'that you are so uncomfortable that you cannot rest your head'.

90

THE DIACHRONY OF VERB SECOND

In each sentence, a complement has been moved to an XP position in front of the verb. To express the general character of the verb-second phenomena in subordinate clause, we make two assumptions: first, embedded clauses are not to be derived through a defective CP node, i.e. they are always "true CPs," i.e. not root embedded clauses; second, the Verb-second constraint is the result of a single verb movement to Intl. Let us summarize the discussion so far. Extraction processes and dependency relations in subjunctive subordinates provide empirical and theoretical motivations for a standard CP structure for embedded verb-second clauses. Our analysis pushes the similarity between embedded clauses a step further by assuming that the Spec of IP is the landing site for the XP that yields the verb-second word order. We will now explore the consequences of this proposal. 3.3.2 Predictions of the V-to-I Analysis Comparative considerations become relevant at this point. More evidence for the single movement of the verb comes from the fact that adverbs of negation like pas in OF and MidF regularly appear between the finite verb and the postverbal subject. Keeping the discussion at an informal level for the moment, if we compare MidF with similar constructions in verb-second languages like Dutch and Danish as below, we note that in the latter, the subject is always adjacent to the fronted verb while the negation element follows the subject. MidF (18) Mais a present n'est pas tele regie gardee But now is not such rule respected 'But now such a rule is not respected'

Policie, p. 17

Dutch (19) Het boek heeft Jan (waarschijnlijk) niet gelesen. The book has John probably not read (Weerman (1988:23)). Danish (20) Bogen har Jens ikke laest The book has John not read

(Weerman (1988:23)).

Notice that in the double verb movement analysis required for Dutch and Danish, these facts are not surprising even if we assume that the negation is higher than the base-subject in VP (adapting Pollock's (1989:383) proposal). Movement of the deep subject to [Spec,IP] will account for the surface position of the following negative element. But the same derivation would obviously yield the wrong result for MidF. The order of the negation in (18) is explained straightforwardly if we assume that the productive verb-second effect of MidF results from a single verb movement to Infl, and that the subject remains in its base ("NP*") position inside VP.

THE LOCUS OF VERB MOVEMENT

91

(21) [IP XP [r Vi [pas [VPn NP* NP* tti Vpp ]]]]

A V-to-I derivation as in (21) allows one to explain the contrast between the two grammatical systems illustrated in the preceding paradigm. Assuming that in (18) the subject has remained under VP", two potential problems for our analysis are raised: what triggers agreement, and how is the subject licensed? In the following section we will show how these different word orders can be accommodated maintaining our hypothesis.

4. The Structure of Inflexion In line with recent developments concerning the structure of Infl, we will adopt Chomsky's (1989) (see also Belletti (1989)) revision of the more articulated structure of inflectional projection proposed in Pollock (1989). According to this proposal, Agreement and Tense each head their own functional projection: AgrP and TP. We will adopt Pollock's original idea extended by Pesetsky (1989) suggesting that French pas (English not), as in (18) above, is not the head of an autonomous projection NegP but a modifier in the Spec of such a projection. At S-Structure, the head of NegP is empty, triggering movement of the verb through it as prescribed by the Head Movement Constraint.3 Thus, the representation of (18) would be as in (22):

92

THE DIACHRONY OF VERB SECOND

One important consequence of this analysis is that the strict locality requirement necessary for subject-verb agreement to take place is not fulfilled. As the verb raises up to amalgamate with the inflectional features, it goes further away from the base subject. Hence the government relation for agreement cannot be met in any way since the agreement features are outside the VP projection. Anticipating a little at this point, let us suggest that the agreement relation is mediated through a pro in the Spec position of TP. Consider now the examples in (23): (23) a. Et

encore

n'estoie il mie delivre

sa maleurte Berinus, I, p. 390 And still CL-NEO was he not delivered from his unhappiness 'And he still was not delivered from his unhappiness'

b. Atant regarda pro

de

contreval la

mer, (...) Berinus, I, p. 240 Then looked (3s SUBJ) down at the sea 'Then he looked down at the sea'.

Here we have to account for the fact that pronominal subjects manifest a quite different pattern from nominal subjects since there is here a parallelism between the structure of MidF and the Dutch and Danish examples of (19) and (20). Notice that the pronominal subject in (23) appears before the negation. On the basis of general theoretical considerations having to do with Spec-Head agreement and Case assignment, we will assume that in the above examples the subject has been moved to the Spec of TP. Let us start by reviewing certain issues involved in the theory of agreement. Although there is actually no clear consensus on how the agreement relation is to be implemented, recent proposals have provided strong arguments showing that it consists of a structural relation between a Spec position and a head in a local environment. Moreover, as noted by Rizzi (1990) (see also Koopman and Sportiche (1991)), the theory of Agreement and the theory of Case can be conceived of along the same lines since they both involve the same structural relation. Bearing this in mind, let us explain how the subject in TP, pro or an overt pronominal, can trigger agreement while being assigned Nominative Case. Our analysis is based on ideas developed in Rizzi (1990) following Tomaselli (1990).4 We assume that Agr is a governor when it has features, hence, when it is in a tensed clause. When there is no subject in its Spec position to agree with, Agr can transmit its features to TP. This means that when its governing capacities are restricted to the right, Agr will govern the pro or pronominal subject in [Spec.TP] which, in turn, being in a Spec-Head relation with T bearing tense features, will then trigger agreement since it is c-commanded by Agr. Let us state that this particular type of agreement is available only because of the fully pro-drop characteristics of OF and MidF, therefore being unavailable for partly pro-drop languages like Dutch and Danish. So, we make the assumption that, in languages where pro can be

THE LOCUS OF VERB MOVEMENT

93

canonically governed by AgrP, the agreement relation can be achieved through [Spec.TP]. We now address the question of the Case assignment to the VP-internal subject telle regie in (18) above. We propose that Nominative-Case assignment to the VP-internal subject follows from the Principle of Full Interpretation and expletive replacement at LF. We will assume that the Nominative NP is subject to some Case-checking principle at LF (cf. Chomsky (1986b)).5 An alternative to this proposal would be to admit with Diesing (1990) (see also Pesetsky (1990)) that Case can be assigned under government by a raised verb along the line of the Government Transparency Corollary (Baker (1988)). If this mechanism is available for Case assignment to the base subject, it introduces a dichotomy between Case and agreement and does not account for the strict locality requirement necessary if one maintains that "rich" agreement is a matter of Spec-Head relation. However, as argued in Koopman and Sportiche (1991), there are two ways to assign Nominative Case: Spec-Head agreement, as for instance in French and English, and structural Case assignment, i.e. Case assignment under government as in Standard Arabic. According to this proposal, quite different predictions are drawn from each procedure with respect to the possibilities of agreement. In Standard Arabic, when Case is assigned structurally to the NP in the VP internal position, agreement is a default mechanism and the verb exhibits only third person singular agreement, whereas while the NP is moved to [Spec.IP], Spec-Head agreement triggers full agreement. Since OF and MidF are full agreement languages, we will argue that Case assignment under Spec-Head agreement prevails thus triggering movement of the subject out of the VP to [Spec.TP]. Let us review the characteristics of OF and MidF which we have examined so far. It was shown that subordinate clauses tend to pattern with parallel constructions in Yiddish and Icelandic with respect to the verb-second effect they manifest. We have claimed that verb-second que-clauses have a true embedded status and so, cannot be headed by a defective C node. Evidence for this assumption has been provided by the extraction facts that can be found in OF and MidF. In such structures the antecedent-trace relation would be blocked by the intervening C'. One additional argument to support our claim is brought by the tense requirement that the main verb prescribes to its clausal complements. Since some verbs require their clausal complement verbs to be in the subjunctive mood, the verb-second effects observed in these cases cannot be plausibly attributed to the "root" character of the embedded clause. On the basis of comparative evidence between OF, MidF, and other Germanic languages such as Danish and Dutch, we have argued that the relative order of the postverbal subject and the negation element is best accounted for by an analysis in which the pronominal subject is moved to the Spec of TP while the verb raises to the head of [AgrP] but not to the head of CP.

94

THE DIACHRONY OF VERB SECOND

5. Predictions from an AgrP Analysis In the preceding sections, we have shown that the verb-second characteristics of embedded clauses in OF and MidF are best analyzed as properties of AgrP, and have developed a similar analysis for verb-second in root clauses. In this section, we show how some crucial differences between root and embedded clauses, which we call illocutionary factors, are predicted by our analysis. In SECTION 5.1, we demonstrate that some facts demand an explanation at a level higher than AgrP, and explore a suggestion by Pollock (1989) concerning an Assertion node. This suggestion has recently been expanded by Laka (1991), who labels the relevant projection IP. In SECTION 5.2, the violations of the verb-second constraint in root clauses will be taken into account; this is also a point where an AgrP analysis makes better predictions than a CP analysis. In SECTION 5.3 we will argue that the AgrP analysis is consistent with current theories of language change in the GB model (Chomsky (1989)). 5.1 The Asymmetry Revisited Given our analysis thus far, we must conclude that if there is an asymmetry between root and embedded clauses, it has to be explained by some means other than verb movement. Recall that verb-second is a possible word order for embedded clauses. So, since verb movement operates through AgrP in both types of clauses, it is now necessary to account for the unexpected figures we find in Table 4.2. This table shows that there is much less verb-second in embedded clauses than in root clauses. The solution we propose for the surprising distribution displayed below, rests on the idea that although verb second can surface freely in root and embedded clauses, it does not mean that the constraint is obeyed in the same way in both contexts. For instance, it is not difficult to show that some adverbial elements inducing the verb-second effect in OF and MidF never occur in embedded clauses. Table 4.2 Verb-Second Distribution in Root and Embedded Clauses in Six Texts of Middle French Postverbal Subject

Subtotal

Total

#

#

#

#

Root Clauses

645

382

1027

2385

Embedded Clauses

106

26

132

2819

TOTAL

751

408

1159

5204

Texts

Null Subject

THE LOCUS OF VERB MOVEMENT

95

5.1.1 Assertive Elements In the data we have studied, the verb-second requirement is often satisfied by the presence of different adverbial elements preceding the inflected verb. We will provide evidence that, while some of these elements are moved from [Spec.VP], the others are base-generated in a Spec position higher than [Spec,IP]. So we find cases like (24) where the VP adverb bien is generated in [Spec.VP] and moved to [Spec,AgrP]: (24) Car bien 1' ai desservy Because truly it have (Is SUBJ) deserved 'Because I have truly deserved it'.

B6rinus, 1,4

Sentential adverbials like si in (25) must be generated in some Spec position higher than VP: (25) si

lui cria

mercy devotement (...) Berinus I, p. 60 then him begged (3s SUBJ) for mercy devotedly 'then he devotedly begged him for mercy'.

We also find the elements or and et, as in (26a,b) respectively, which function like adverbials as far as verb-second is concerned, but whose categorial status is not clearly established: (26) a. Or

de Presine. M61usine, p. 14 Now have you heard of the King Elinas and of Presine 'Now, you have heard of the King Elinas and of Presine'

b. Et

avez vous ouy

n'

du

estoit nul qui

roy

Elinas et

conforter le

sceust; CNNA, p. 460 And NEG was none who to comfort him knew 'And there was no one who knew how to comfort him'.

As we will now demonstrate, the discourse elements that participate in verbsecond constructions do not necessarily constitute a homogeneous class; in the following discussion, we will distinguish the si-class, which is clearly an adverbial from the ef-class, whose status has long been the subject of debate.6 Since we find examples such as (27) where et and si co-occur, we must determine whether they occupy two distinct positions. (27) et

si est

bien

a noter comment (...) Policie, p. 149 and so is (3s SUBJ) important to notice how 'and it is important to notice how'

Let us first consider the behavior of si. The status of this sentential adverbial has been established in the work of Marchello-Nizia (1979). Below, we summarize the main characteristics of si, according to Marchello-Nizia:

96

THE DIACHRONY OF VERB SECOND

si is a sentence adverbial;7 the normal position for si is at the beginning of the sentence (never at the beginning of a text); si is never used in subordinate clauses and never in untensed clauses; si is never used with negative clauses; si is strictly contiguous to the verb form. In our analysis, si is base-generated in the Spec of AgrP, an A-bar position. According to our model, the sentential adverbial si has the status of a clitic element on a tensed verb. Like the clitic subjects in Modern French, si can only be separated from the verb by clitic objects. No adverbial element can be inserted between si and the verb; si is never found in untensed clauses, i.e. in infinitival or participial clauses. Our claim that si occupies the [Spec,AgrP] position, subject to some conditions on identification, explains its distinct behavior. It is important to notice that Spec of AgrP is not the only possible position where the so-called sentential adverbial could arguably be generated. For example, R&T adopt a V-to-I analysis for Icelandic, but they propose that some root narrative facts can be explained by generating in CP an E node, a suggestion inspired by Banfield (1982).8 We shall pursue this idea that there is something distinct about root narrative clauses, but without recourse to V-toC movement. In his discussion of emphatic contexts with imperative do in English, Pollock (1989) suggests, as an alternative to NegP, that in such cases, do occupies the head of a projection he calls "Assertion" (see also Pesetsky (1989)). Adopting Pollock's suggestion, Laka (1991) develops a similar analysis for Spanish in which negative polarity elements such as nadie, nada, ningun, nunca, etc., involve movement to the Specifier of a functional projection (IP) which is generated between CP and IP, a position also involved in emphatic fronting. The structure Laka assigns to the sentence in (28) is depicted in (29): (28) a nadie yo pedire perdon to nobody I will ask forgiveness

Laka (1991)

THE LOCUS OF VERB MOVEMENT

97

Spec.ZP is also available in emphatic fronting contexts, i.e. in contexts containing a fronted topicalized constituent, such as (30). (30) vestidos compraria yo con ese dinero clothes would buy (I) with this money 'I would buy clothes with this money'. Pursuing this idea, we argue that the head of IP is filled by a discourse operator like Assertion which identifies si in [Spec.AgrP] as a legitimate assertive element.9 As a discourse operator, this assertive element is not expected in embedded clauses. Since si is generally used to relate the current sentence to the preceding one, or to indicate how the locutor situates himself through the discourse, we have a natural explanation for the fact that si is not found in embedded clauses. The main characteristics of the sentence adverbial si follow from the fact that it is base-generated in the category Assertion and this category will be projected in root clauses, but never in embedded ones. Our analysis also correctly explains that si XP V is ungrammatical in both OF and MidF. This prediction would be difficult to make in a V-to-C analysis involving a VP-internal subject and some base-generated element si in [Spec.CP] as proposed by Adams (1987a,b). To explain the MidF period, Adams assumes that verb movement to C became optional. Notice that under this analysis, nothing prevents Spec of AgrP from being filled by any XP, yielding the ungrammatical si XP V. However, on a V-to-Agr analysis, this is correctly predicted to be illicit. While the evidence clearly indicates that si must be generated in Spec,AgrP, the case of et is less clear. Et had an ambiguous status in MidF: as in Modern French, et could be used as a coordinating particle. Notice also that et could immediately precede the verb in root clauses, inducing a verb-second effect, as illustrated in (3la), or it could precede the subject (or some XP) in preverbal position, as illustrated in (31b). (31) a. Et se taist 1'ystoire a parler de lui Melusine, p. 5 And became silent the story to talk about him b. Et les autres deux respondirent And the other two to answer

Melusine, p. 6

In view of its participation in the verb-second structures, we analyze this et as a discursive category whose unique function in a narrative context is to relate what is being said to what precedes it.10 As has just been pointed out, when et precedes another element that occupies the preverbal position, as in (31b), it is a discursive element. We take this as evidence that et must be generated in a position higher than Spec.AgrP. Before developing our proposal, let us briefly restate the problem. Should sentences such as (3la) be analyzed as verb-second constructions? Vance (1989) has argued that clauses with et and a null or postverbal subject are not verb-second constructions. However, one can argue in favor of their verbsecond status, if one takes the verb-second constraint as a constraint which operates only at S-Structure. On this view, all root declarative sentences which

98

THE DIACHRONY OF VERB SECOND

do not obey the constraint constitute violations of the verb-second constraint; on the other hand, if the verb-second constraint can be checked at different levels, nothing forbids apparent verb-second violations at S-Structure, as long as verb second is obeyed at another level, such as LF. In OF and MidF, the verb-first constructions in root clauses occur in similar contexts as in German and Icelandic. These include yes/no questions, conditionals, etc., as illustrated below: (32) a. As tu bien regarde tous les cons de ces femmes qui Have you well looked at all the cunts of these women who etaient aux estuves CNNA, p. 413 were at the baths 'Did you look at all the cunts of the women who were bathing' b. Plaise vous en paix le laissier 'Would you please leave him in peace'.

Podsies, p. 3

With Roberts (1992), we suggest that the so-called verb-first root sentences are those which contain a phonetically null element in the highest Spec position.11 The preverbal position is filled by an operator in the Spec which is licensed by discourse conditions. We argue that in such superficially verbfirst root clauses, the discourse operator in the head of Z identifies an empty position in [Spec, AgrP], spreading its features to this position, giving the sentence the verb-second interpretation. In our corpus, these and the imperative clauses are the only cases where the verb is in first position at S-Structure. So we do not see why et V should be an exception. Since we have already motivated a position higher than AgrP for et si, we now suggest that et may be generated in a position outside AgrP. In this case, we assume that the [Spec.AgrP] position is empty as in the verb-first contexts just mentioned, and that et itself may fill the head of IP.12 Our preliminary investigations indicate that or and mats pattern with et rather than with si. However, further research is required to verify this result since illocutionary factors are a complex issue which requires detailed study. Our analysis predicts that et, mais and or can be followed by a subject or by any other XP while playing a discursive role, and this is in fact what we find in the data.13 (33) a. or il est bien verites So it is really (the) truth b. mais avant vous

M61usine, p. 29

diray

comment le roy Elinas (...) Melusine, p. 14 but before to you tell (1s SUBJ) how the king Elinas 'But before I tell you how the king Elinas... '

To illustrate that the discursive elements are the most important among the categories involved in the verb-second construction in root clauses, we have separated these elements from other XPs. Table 4.3 shows the distribution of discursive elements in root clauses.

99

THE LOCUS OF VERB MOVEMENT

Table 4.3 Proportion of Discursive Elements in Prose with Respect to XP Elements "et"

"si"

3

56

36

95

50%

189

M61usine

61

28

49

138

65%

211

Policie

19

26

8

53

36%

148

QJM

40

6

21

67

59%

113

CNNA

25

35

11

71

47%

152

Memoires

108

4

12

124

58%

214

Subtotal

256

155

137

548

53%

1027

Texts Berinus

Other Discourse Total Discourse Total XPVs XPs XPs

If we compare root and embedded clauses, omitting from the root clauses the total number of discourse elements such as si, et, or, mais, etc., that are never found in subordinate ones, the asymmetry between the two types of clauses radically decreases. As is shown in Table 4.4, the three texts of poetry in our corpus (Palatinus, Miracles and Poesies) use far fewer discursive elements than the texts in prose. This is not a coincidence and this peculiar behavior is predicted by the proposed analysis. Since poetry does not involve a narrative style, we do not expect a large number of discursive elements. Let us summarize the results of the discussion. This section was intended to show that there remains some kind of asymmetry between root and embedded clauses since verb-second structures are used more frequently in root clauses than in embedded ones. We have claimed that this asymmetry can be partly explained by discursive factors typically found in root contexts. Table 4.4 Proportion of Discursive Elements in Verses with Respect to XP Elements

"et"

"si"

Palatinus

3

31

36

70

35%

201

Miracles

1

14

12

27

11%

240

Poesies

14

6

14

34

15%

226

Subtotal

18

51

62

131

20%

667

Texts

Other Discourse Total Discourse Total XPVs XPs XPs

100

THE DIACHRONY OF VERB SECOND

We have proposed an analysis which accounts for the discursive particles most frequently encountered in our data: si, et and or. The assertive level argued for offers a new way to derive some discourse phenomena related to the verbsecond constraint within a government-binding model, without reference to some extra mechanisms. Finally, this category gives support to our claim that V-to-C is an undesirable analysis of the derivation of verb-second word order in MidF.

5.2 Violations and Apparent Violations Despite the importance of the verb-second phenomena during the period under study, one cannot deny that violations exist in OF and in MidF root clauses14 and that they must be accounted for. (The discussion in this section will deal exclusively with root clauses.) This is a crucial point since it has been proposed that the increased number of verb-second violations is the reflex of a parametric change. Specifically, it has been suggested that the loss of V-to-C movement in the grammar eventually led to the emergence of SVO order. Since there is never a V-to-C movement in our analysis, the change in the grammar of French cannot be attributed to the loss of this rule. In this section we develop an alternative account for the emergence of SVO. We begin by considering the violations of verb second found in MidF. The sentence in (34) illustrates a marked construction but one which is common to Old and Middle French: (34) et la dame si fu appellee Guigamo. B6rinus I, p. 120 'and the woman was thus named Guigamo'. This sentence involves left dislocation of the subject. This is in fact the interpretation of Marchello-Nizia (1979) who analyzes si in such constructions as a sentence adverbial. In a V-to-C derivation, si is in the [Spec.CP] and the subject la dame in a position outside CP. An alternative would be to move the subject in [Spec.CP] and to adjoin si to AgrP. The sentence in (34) is clearly a case of focus on the subject. In our V-to-Agr analysis, si is in [Spec.AgrP] and the subject is in [Spec.IP]. While both analyses take sentences like (34) as violations of the verb-second constraint, a V-to-Agr approach has nothing special to say about these examples. In fact, a large proportion of the cases involving discursive elements discussed in SECTION 4.1 constitute verb-second violations for any theory. Nevertheless, these violations are better accounted for in a V-to-Agr analysis coupled with an independently motivated EP projection than in a V-to-C analysis. It should be pointed out that violations of the verb-second constraint in root declarative clauses are documented in every stage of the history of French. We call these false violations. We suggest that this fact undermines Adams' (1987a,b) proposal that the cliticization of the subject was the trigger for the loss of the verb-second constraint on word order because the evidence will show that her proposal does not accurately reflect the development of the Ian-

THE LOCUS OF VERB MOVEMENT

101

guage. We then develop an alternative proposal which accounts for the gradual nature of the change, as a consequence of the reanalysis of individual lexical items. Finally, we demonstrate that a V-to-Agr analysis has nothing special to say about most of the violations of the constraint. 5.2.7 False Violations Some discursive elements which are never met with pro or with postverbal subjects, the typical contexts of the verb-second constructions, are not taken as verb-second violations, neither for the proponents of a V-to-C analysis, nor for us. Nevertheless, they must be placed somewhere in the structure. A Vto-C analysis has to postulate a special mechanism for these elements: either a double CP, or adjunction of some element to CP. In a V-to-Agr analysis, such elements are likely to be found inside the Assertion projection. For example, certes 'certainly', and premierement 'firstly' are realized in the [Spec.ZP] position. A V-to-C analysis would have to assume that they are adjoined to CP, the landing site of the raised verb in verb-second constructions. Second, we show that the AgrP analysis permits a superior account over the CP analysis for the cases where a pronominal subject is preceded by an adverbial XP or a discursive et. According to the analysis presented so far, if the pronominal subject vous in a sentence, as in (35), is to be taken as an autonomous element in a Spec position, such a case is a true verb-second violation. (35) et en telle maniere vous arez en pou d'eure la compaignie and thus you will in little time the company de moy oubliee,... B6rinus I, p. 15 of me forget 'and thus you will shortly forget my company, ...' This is not so for Adams (1987a,b) who assumes that the cliticization of the subject pronoun is the first step in a series of changes leading to a reanalysis of the position of the verb from C to I. Adams proposes that the pronominal subject is cliticized in syntax and forms a verbal complex before moving to C°. Hence there is a step where the derivation is still to CP: the XP is available because the pronoun is on the verb. In her analysis, such apparent verbsecond violations would result in the following derivation: (36) [ C P XP[ c .Scl+V[, P tt]]] Unfortunately, this elegant proposal does not accurately reflect the data in our corpus because it implies that the first violations of verb second arise with a clitic subject. However, as argued in Dufresne (1989), the first violations were in fact observed more frequently with full NP subjects than with pronominal ones. The figures in Table 4.5 support this conclusion. These facts suggest that clitic status of the subject during this period is not involved in the change.

102

THE DIACHRONY OF VERB SECOND

Table 4.5 Verb Second (V2) and V>2 in Middle French Texts

V2

Nominals V>2 %

Total

V2

Pronominals V>2 %

Total

Berinus

40

15

27

55

107

20

16

127

Melusine

82

16

16

98

126

17

12

143

Policie

47

32

41

79

71

21

23

92

CNNA

64

20

24

84

130

31

19

161

Memoires

90

18

17

108

73

21

22

94

380

122

24

502

682

148

18

830

Total

Figure 4.5 following, shows that violations with both NP subjects and with pronominal ones represent a stable phenomenon that is attested over two centuries. Thus, the violations are surprising for two reasons: First, they occur in contexts where they should not occur on the subject clitic hypothesis, and second, they reflect a relatively stable property of MidF.

Figure 4.5 Independent Percentages of Verb Second Violations with Pronominal and Nominals

THE LOCUS OF VERB MOVEMENT

103

On the other hand, when comparing the elements responsible for the violations, there does not seem to be a clear cut difference between those that show up with pronouns, as in (37), or with nouns, as in (38). Indeed it is easy to find both in the same text. (37) a. Et auxi elles s'en scevent bien

mocquer entre elles QJM, p. 23 perfectly laugh between

and also they know themselves 'and so they perfecly know to laugh about it between themselves' b. briefment il court par la maison quickly he runs through the house (38) a. Et auxi toutes les aultres sont en la buee 'And also all the other ones are in the laundry'

QJM, p. 52 QJM, p. 53

b. briefment, le pouvre corps de lui n'

avrajames repoux (...) QJM, p. 93 in brief the poor body of him NEG will never rest 'In brief his poor body will never rest in peace'.

In our analysis if two elements precede the pronominal subject, like et and auxi in sentences (37) and (38), the first, et, must be in 1°, and the second, the adverb, must be adjoined to AgrP. 5.2.2 Adjunctions to AgrP? In a V-to-C analysis of a strict verb-second language, sentences containing more than one constituent before the tensed verb in root declarative clauses, must be analyzed as cases of adjunction to CP. To avoid this solution for MidF, Roberts (1992), following Vance (1989), takes into account the increasing number of XP Subj V constructions during this stage to propose an alternative solution.15 According to Roberts, the Adv XP V order is rare in Germanic languages. This is why the adverb never occurs before a preverbal subject in a true verb-second clause. Crucially, in cases of true violations in MidF, Roberts proposes that CP has been reanalyzed as AgrP. Adjunction would be to this projection, rather than to CP. In our V-to-Agr analysis, we arrive at the same result, without having to postulate a reanalysis. 5.3 Loss of the Verb-Second Constraint We have argued that MidF is a non-asymmetric verb-second language, like Yiddish and Icelandic. These languages share the characteristic that verbsecond is possible both in root and in embedded clauses. Our proposal entails that no verb movement has been lost in the change from MidF to Modern French. If this hypothesis is correct, then we must seek another explanation for the loss of verb second in Modern French. In fact there are two important properties which distinguishes MidF from Modern French: pro-drop and verb second. As has been argued by Adams,

104

THE DIACHRONY OF VERB SECOND

there are two reasons for the loss of these two characteristics of MidF. They are the loss of directional government and a change in the accentual pattern of the language. We agree with Adams about the importance of these two factors, despite the differences between the analysis presented here and the one she proposes. We have also shown that the loss of the superficial manifestations of the verb-second constraint in Modern French must take into account some changes in the narrative structure of the language. A small class of elements whose original syntactic function was to satisfy the verb-second requirement and at the same time establish a predicative relation between the verb and its subject is no longer able to fulfill these two functions. Any analysis which aims to account for syntactic change in the history of French must take into account all three of these factors.

6. Conclusion In conclusion, there are at least two classes of verb-second languages: Those that show a strict asymmetry between root and embedded clauses and those that show little or no asymmetry. In the spirit of recent work, the former involve long movement of V through Agr to C, while the latter involve short movement of V to Agr. The evidence indicates that OF and MidF are nonasymmetric verb-second languages, like Yiddish and Icelandic. Thus, we pre dict that apparent root-embedded asymmetries in these other non-asymmetric languages should also be amenable to an analysis which has recourse to illocutionary elements in the sentence.

Notes *

Research for this article was made possible by grants from SSHRC No. 41089-1409 and 410-89-0785. We thank all the members of our research project for their comments and support. Special thanks to Anne Rochette, Betsy Ritter and Isabelle Hai'k for helpful comments and suggestions.

1. The 16th-century data not considered in this paper are part of work in progress. 2. One can maintain that there is no asymmetry between root and embedded clauses, while proposing a movement to Comp: this is indeed what is pro posed by Haan and Weerman (1986). 3. It has been noticed that LF movement is not possible across negation, while S-Structure raising is possible. Pesetsky's suggestion rules out the movement at LF, arguing'that the position is filled at this level, the null version of pas blocking movement. For a different view of the problem with far reaching consequences, see Rizzi (1990). 4. Tomaselli (1990) has argued that the agreement in COMP is a characteristic of all the Germanic languages, even those which do not manifest agreement at a superficial level. This question will not be discussed in this paper.

THE LOCUS OF VERB MOVEMENT

105

5. Deprez (1988, cf. fn. 3) proposed that Stylistic Inversion licenses an expletive pro in [Spec,IP]. This expletive must be replaced at LF along the lines of Chomsky. The subject must consequently move to SpecIP at LF where Nominative Case is checked. 6. We borrow from Vance (1989:88-89) the following summary of these debates: "Many scholars have puzzled over the role of et in introducing VS clauses in Old and Middle French. Some of them (for example Foulet (1963); Nissen (1943); Franzen (1939); Lewinsky (1949); Crabb (1955); Skarup (1975) and Adams (1987a,b)) maintain that et may in some cases be used as an adverb, acting as the intial constituent in what is really an CVS structure. Others (for example Baulier (1956)) find in their data no basis for this claim. My data argue for the latter position." For more references and a discussion of this question, we send the reader to Marchello-Nizia (1979:281-287). For us et has a role in the verification of the verb-second constraint, even though it does not have the same categorial status as si. 7. The discussion will be limited to si used as a sentence adverbial. For other distributions of si in different contexts, see Marchello-Nizia. 8. The node E in Banfield is used as an abbreviation for "Expression." The theory developed by this author includes in a generative framework narrative facts that are related to the writer. Banfield elaborates these notions under the term of "Self." Most of the assertive considerations about si, drawn from Marchello-Nizia, could be translated into Banfield's.theory. 9. If we argue that the Spec,IP position is a possible A-bar position, si may be generated there. This can be bound to the pecularities of the nominal subject: an NP has to be identified by its thematic role and by its Case; both are necessary to express the relation of predication between the verb with its external argument. If this relation can be expressed elsewhere than in Spec,IP, as it is the case in OF and MidF, this position is available for another type of predication, the one called "a second level predication" by Marchello-Nizia: as it has been demonstrated by this author, si expresses just such a relation between the locutor and the sentence. In the case of si, this means that the locutor takes the meaning of the sentence as true. This is why si itself cannot be negated. 10. For more considerations on this element see Marchello-Nizia (1979, ch. 20). 11. Roberts (1992) assumes that the relevant Spec position is [Spec.CP]. However, we propose that the position in question is [Spec,AgrP]. 12. As an alternative, one could argue that et occupies the head of CP. This solution would have the advantage of giving a unified analysis of et in coordination and et in discursive contexts, but in this view one does not explain why these discursive elements like et, or, or mais are involved in verb-second structures with null or postverbal subjects. 13. It can also be suggested that imperative sentences would be analyzed through the category Assertion since one of their most striking characteristics at this period of French is the presence of discursive elements like et, si, or, etc. 14. In his study, Herman (1954) discusses a few cases ofXPS V violations in OF and says that most of them are only apparent violations. With this observa-

106

THE DIACHRONY OF VERB SECOND

tion in mind and considering the kind of violations given by Foulet (1963), it is possible to argue that MidF did not show more violations of the verb-second constraint in root clauses than OF. Thus expressions premierement, certes, nonpour tant and the like as in (15) are not violations either in OF or in MidF. (i)

Certes, elle disoit verite, 'Surely, she was telling the truth'

(ii) Premierement je vueil respondre... 'First I want to answer...'

Berinus, I, p. 15 Berinus, I, p. 90

(iii) Non pourtant, je ne le dy pas... QJM, p. 53 'Nevertheless, I do not say so'. In OF, violations occurred in two contexts each one with a subject NP and with a pronominal: 1. an XP could separate the subject and the verb as in (iv): (iv) Li

dist QLDR, li tiers livres, p. 223 The king promptly to him says 'The king promptly says to him'. 2. an XP precedes the subject as illustrated in (v): (v)

reis erranment li

De ces

nuveles tuit furent esfrees QLDR, li tiers livres, p. 227 By these news all were afraid 'This news frightened everyone'. The main difference in the corpus of MidF that we have studied is that the pronominal subject is almost never separated from the verb. On the contrary, NP XP V word order is still sometimes met with a nominal subject: (vi) Li chevalier tous communement tindrent ce conseil... Berinus I, p. 120 'The knights all collectively followed this advice...' (vii) et

la dame

si fu

appellee Guigamo. Berinus I, p. 120 'and the woman was thus named Guigamo'. 15. Vance has systematically compared the violations in word order in La Queste del Graal, a 12th-century prose text to those in Jehan de Saintre also in prose, written around 1456 by Antoine de la Sale. The author attributes the increase of violations in MidF to a greater liberty in adjunction to IP which has multiple consequences on word order. For Vance, this new possibility explains the transitional period between verb second to CP and SVO.

References Adams, M. (1987a) "From Old French to the Theory of Prodrop." Natural Language and Linguistic Theory 5:1-32.

THE LOCUS OF VERB MOVEMENT

107

Adams, M. (1987b) Old French, Null Subjects and Verb Second Phenomena. PhD Dissertation, UCLA. Adams, M. (1988a) "Embedded Pro." In J. Blevins and J. Carter, eds. Proceedings ofNELS 18, 1-21. Amherst, Mass.: GLSA. Adams, M. (1988b) "Les effets V2 en ancien et en moyen francais." Revue quebecoise de linguistique theorique et appliquee 7:13-40. Banfield, A. (1982) Unspeakable Sentences, Narration and Representation in the Language of Fiction. Boston, Mass.: Routledge and Kegan Paul. Baker, M.C. (1988) Incorporation: a Theory of Grammatical-Function Changing. Chicago: University of Chicago Press. Baulier, F. (1956) "Contribution a 1'etude de 1'inversion du sujet apres la conjouction 'et'." Le Francois Moderne 24:249-257. Bayer, J. (1984) "COMP in Bavarian Syntax." The Linguistic Review 3:209274. Belletti, A. (1989) Agreement and Case in Past Participial Clauses in Italian. Ms. Universite de Geneve. Bennis, H. and L. Haegeman (1983) "On the Status of Agreement and Relative Clauses in West-Flemish." In W. de Geest and Y. Putseys, eds. Sentential Complementation, 33-55. Dordrecht: Foris. Besten, H. den (1983) "On the Interaction of Root Transformations and Lexical Deletive Rules." In W. Abraham, ed. On the Formal Syntax of Westgermania, 47-131. Amsterdam: John Benjamins. Chomsky, N. (1986a) Barriers. Cambridge, Mass.: MIT Press. Chomsky, N. (1986b) Knowledge of Language. New York: Praeger Publishers. Chomsky, N. (1991) "Some Notes on the Economy of Derivations and Representations." In R. Friedin, ed. Principles and Parameters in Comparative Grammar, 417-454. Cambridge, Mass.: MIT Press. Chung, S. and J. McCloskey (1987) "Government, Barriers and Small Clauses in Modern Irish." Linguistic Inquiry 18:173-237. Cooper R. and E. Engdahl (1989) "Null Subjects in Zurich German." In L.O. Delsing, C. Falk and C. Platzack, eds. Comparative Germanic Syntax. Working Papers in Scandanavian Syntax 44:31-44. Lund, Sweden. D6prez, V. (1989) On the Typology of Syntactic Positions and the Nature of Chains: Move a to the Specifier of Functional Projections. PhD Dissertation, MIT. Diesing, M. (1990) "Verb Second in Yiddish and the Nature of the Subject Position." Natural Language and Linguistic Theory 8:41—79. Dufresne, M. (1989) "Les pronoms personnels sujets en moyen francais et le contrainte V2." Talk given at the Colloque de syntaxe historique, 57th Conges de 1'ACFAS, Universitd du Quebec a Montreal. Dupuis, F. (1988) "Pro-drop dans les subordonnees en ancien francais." Revue quebecoise de linguistique theorique et appliquee 7:41-62. Dupuis, F. (1989) L'expression du sujet dans les subordonnees en ancien francais. PhD Dissertation, Universit6 de Montreal. Foulet, L. (1963) Petite syntaxe de I 'ancien francais. Paris: Champion.

108

THE DIACHRONY OF VERB SECOND

Franzen, T. (1939) La syntaxe des pronoms personnels sujets en ancien francais. Uppsala: Almqvist. Fukui, N. and M. Speas (1986) "Specifiers and Projections." In N. Fukui, T. Rapaport and E. Sagey, eds. MIT Working Papers 8:128-172. Haan, G.J. de and F. Weerman (1986) "Finiteness and Verb Fronting in Frisian." In H. Haider and M. Prinzhorn, eds. Verb-Second Phenomena in Germanic Languages, 77-110. Dordrecht: Foris. Haider, H. (1986) "V-Second in German." In H. Haider and M. Prinzhorn, eds. Verb-Second Phenomena in Germanic Languages, 49-75. Dordrecht: Foris. Herman, J. (1954) "Recherches sur 1'ordre des mots dans les plus anciens textes francais en prose." Acta Linguistica Academiae Scientiarum Hungaricae 4:69-94, 351-379. Hirschbiihler, P. and M. O. Junker (1988) "Remarques sur les sujets nuls et pronominaux dans 1'histoire du francais." Revue quebecoise de linguistique theorique et appliquee 7:63-84. Hulk, A. and A. van Kemenade (this volume) "Verb Second, Pro-drop, Functional Projections and Language Change." Jaeggli, O. and K. Safir, eds. (1989) The Null Subject Parameter. Dordrecht: Kluwer. Kitagawa, Y. (1986) Subject in Japanese and English, PhD Dissertation. University of Massachusetts, Amherst. Koopman, H. and D. Sportiche (1986) "A Note on Long Extraction in Vata and the ECP." Natural Language and Linguistic Theory 4:357-374. Koopman, H. and D. Sportiche (1991) "The Position of Subjects." Lingua 85:211-258. Kuroda, S.Y. (1988) "Whether we Agree or Not: A Comparative Syntax of English and Japanese." In W. Poser, ed. Papers from the Second International Workshop on Japanese Syntax, 103-143. Stanford: CSLI. Laka, I. (1991) "Negative Fronting in Romance: Movement to E." Paper presented at LSRL, Santa Barbara, Feb. 1991. Lewinsky, B. (1949) L'ordre des mots dans Berinus, roman en prose du XlVe siecle. Goteberg: Pendkuists Peletryskeri. Marchello-Nizia, C. (1979) Histoire de la langue francaise aux XlVe et XVe siecles. Paris: Bordas. Marchello-Nizia, C. (1985) 'Question de methode." Romania 106:481-492. Martin, R., ed. (1978) Etudes de syntaxe du moyen francais. Paris: Kliencksieck. Moignet, G. (1965) Le pronom personnel francais. Essai de psychosystematique historique. Paris: Klincksieck. Nissen, H. (1943) L'ordre des mots dans La Chronique de Jean d'Oubremente. Boktrychen: Almquist and Wilsells. Oilier, M.-L. (1989) "La sequence OR/SI en ancien francais: une strategic de persuasion." To appear in Romania 1992. Pesetsky, D. (1989) "Language-Particular Processes and the Earliness Principle." Ms. MIT.

THE LOCUS OF VERB MOVEMENT

109

Pollock, J.-Y. (1989) "Verb Movement, UG and the Structure of IP." Linguistic Inquiry 20:365-424. Rizzi, L. (1990) Relativized Minimality. Cambridge, Mass.: MIT Press. Rizzi, L. and I. Roberts (1989) "Complex Inversion in French." Probus 1:130. Roberts, I. (1992) Verbs and Diachronic Syntax. Dordrecht: Kluwer. Rognvaldsson, E. and H. Thrainsson (1990) "On Icelandic Word Order Once More." In J. Maling and A. Zaenen, eds. Modern Icelandic Syntax. Syntax and Semantics 24:3-40. San Diego: Academic Press. Santorini, B. (1989) The Generalization of the Verb-Second Constraint in the History of Yiddish. PhD Dissertation, University of Pennsylvania. SigurSsson, H.A. (1990) "VI Declaratives in Icelandic: Their Syntax and Semantics." In J. Maling and A. Zaenen, eds. Modern Icelandic Syntax. Syntax and Semantics 24:41-69. San Diego: Academic Press. Skarup, P. (1975) Les premieres zones de la proposition en ancienfrancais. Copenhagen: Akademisk Forlag. Thiersch, C. (1978) Topics in German Syntax. PhD Dissertation, MIT. Thurneysen, R. (1892) "Die Stellung des Verbums im Altfranzosisehen." Zeitschrift fur Romanische Philologie 16:289-371. Tomaselli, A. (1990) "COMP° as a Licensing Head - An Argument Based on Cliticization." In J. Mascaro and M. Nespor, eds. Grammar in Progress: GLOW Essays for Henk van Riemskijk, 433-445. Dordrecht: Foris. Vance, B.S. (1988) "L'evolution de pro-drop en franc.ais medieval." Revue quebecoise de linguistique theorique et appliquee 7:85-109. Vance, B.S. (1989) Null Subjects and Syntactic Change in Medieval French. PhD Dissertation, Cornell University. Weerman, F. (1989) The V2 Conspiracy, A Synchronic and a Diachronic Analysis of Verbal Positions in Germanic Languages. Dordrecht: Foris. Zanuttini, R. (1990) "On the Relevance of Tense for Sentential Negation." Ms. University of Pennsylvania.

5 Evidence for a Verb-Second Phase in Old Portuguese* Ilza Ribeiro State University of Sao Paulo, Campinas

The basic facts related to the verb-second phenomenon have been studied in generative grammar based on data from Modern Germanic languages (except Modern English) and both Old Germanic and Romance languages. There is a consensus among researchers who have developed studies on this phenomenon that the verb-second effects are derived from two rules of movement: a) a rule which moves the finite verb to the second position of the sentence, the verb fronting being obligatory for every root clause: b) a rule which moves any XP constituent (the NP-subject or a VP element or any XP of the sentence [e.g., an adverbial]) to the first position of the sentence. This movement is obligatory for every root declarative clause. In the string Wh V S in matrix Wh-questions, the Wh-word counts as the first element of the sentence with respect to verb-second. In other words, displacing the verb to a position adjacent to the Wh-word creates a verbsecond structure. Rizzi (1991) calls "residual verb-second" the Wh-constructions which show subject-auxiliary inversion in English, and verb-subject clitic inversion in French. In those languages there is no generalization of the verbsecond order to main declarative clauses. On the other hand, "full verb-second" identifies the phenomenon that determines the order of the constituents in all matrix clauses and in some embedded ones in languages such as German, Dutch and Scandinavian. The distinction between the two types of phenomenon lies in the fact that residues of verb second are restricted to certain limited environments. The examples in (la) and (Ib) are instances of the French type of verbsecond residue while (Ic) illustrates the English type:1 (1)

a. A qui as-tu parle? To whom have you spoken? 110

EVIDENCE FOR A VERB-SECOND PHASE IN OLD PORTUGUESE

b.

Que manges-tu? What do you eat?

c.

Whom did Mary see?

111

Current linguistic analyses assume that these constructions instantiate Agr° to C° movement by virtue of the fact that Spec/C' is filled in by a Wh-element (cf. Rizzi (1990b, 1991), among others). Wh-questions with subject-verb inversion can be found in Brazilian Portuguese (henceforth BP), as shown by the following sentences:2 (2)

a. Para quando querem os diretores esses relat6rios? For when want the directors these reports 'When do the directors want these reports?' b. Com quern tinha Maria pretendido sair ontem a noite? With whom had Maria intended to go out last night? With whom did Maria intend to go out last night?'

The respective sequences Wh - V / Aux [+fin] - Subject - Object / V [-fin] can be analysed as evidence that the verb is in C° and the subject in Spec/ Agr', and the Wh-phrases (para quandolcom quern) in Spec/C'. The constructions above, however, do not exhibit a homogeneous pattern in those languages, as can be observed from the following cases: (3)

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

A qui tu as parle? *Que tu manges? *Who Mary saw t? Who t saw Mary? *Who did t see Mary? Para quando os diretores querem esses relat6rios? Com quern Maria tinha pretendido sair ontem a noite?

In French, verb-subject inversion is obligatory with the Wh-element que (cf. (lb/3b)), but optional with other interrogative elements in front of the sentence (cf. (la/3a)). In BP, verb-subject inversion is optional (cf. (2a/3f) and (2b/3g)), the order with no inversion (subject-verb order) being the most common in contemporary spoken BP.3 English, on the other hand, presents a peculiar behavior when compared to those two languages: subject-auxiliary inversion is obligatory whenever a non-subject Wh-element is moved into C° (cf. (lc/3c)), and prohibited if the moved element is a Wh-subject (cf. (3d/ 3e)). Assuming that the English and the French constructions are residues of verb second and that the observed differences in their syntax derive from other properties of these languages (cf. Rizzi (1991)), the first question to be posed is: Would the interrogative constructions of BP presented above be residues of verb second? Studies within the variationist model have shown two things regarding the subject-verb order in BP: on the one hand, the occurrence of the VS order is low in interrogative structures in modern spoken BP, but, on the other hand,

112

THE DIACHRONY OF VERB SECOND

it had a very high frequency of occurrence in past periods. Duarte (1989) studies the evolution of the VS order to SV in Wh-interrogatives. She observes that in 1734—the starting point of the time period considered—all direct Whinterrogatives showed VS order. The same goes for the year 1845, since only 4% of the cases displaying SV order corresponded to occurrences of only one single sentence of a total of 28. From 1845 on, a slight increase in the occurrence of SV order (especially in 1882 and 1918) is observed, but a clear preference for it in the period from 1937 to 1989 is shown, and this may characterize a change. Thus, Duarte's data, especially the fact that in 1734 100% of the Wh-root constructions occurred with VS order, indicate the existence of a movement rule from Agr° to C° in 18th-century Portuguese. We may then conclude that the high frequency of Wh+VS constructions in the 18th century must be analysed as resulting from syntactical mechanisms related to the verb-second phenomenon. Therefore, the second question to be posed is: If it is the case that Old Portuguese (henceforth OP) was a verb-second language, how and why has it lost this syntactic property? In this work we shall limit ourselves to offering some evidence which seems to favor a characterization of OP as a veb-second language, and to presenting some considerations on the loss of this property.4 In SECTION 2, we shall compare some analyses of the French verb-second phenomenon with data from 14th-century Portuguese. In 2.1 we shall discuss the verb-second constructions in root and embedded clauses; in 2.2 we shall focus on verb-first clauses and in 2.3 on verb-third ones. In 2.4 we shall deal with the clitic complement placement and the feature [+Agr] in C°. In SECTION 3, we shall sketch a proposal for analysing the loss of verb-second in a diachronic perspective, following Roberts' proposal of parametric possiblities for Nominative assignment by Agr° (Roberts (1992)).

2. The Verb-Second Phenomenon in Old Portuguese We established the 14th century as an initial period and built a corpus from the OP document "Dialogos de Sao Gregorio" (Saint Gregory Dialogues; henceforth DSG), presented in Mattos e Silva (1989).5 We collected a total of 113 constructions from pages 781-790 and 815-821. Table 5.1 below shows the results: Table 5.1 The Order of Constituents in Old Portuguese6 Order

SV(C)

XPV(S)

VSV

(C)

Total

Root

15

31

3

8

57

Embedded

19

12

7

18

56

Total

34

43

10

26

113

EVIDENCE FOR A VERB-SECOND PHASE IN OLD PORTUGUESE

113

2.1 The Verb-Second Clauses 2.1.1 The Root Clauses Verb-second effects are typically found in root clauses. In a language showing verb-final basic word order (SOV) such as German, for instance, the inflected verb must appear in second position in all root clauses, being preceded by any phrasal constituent: an XP-adverbial, an XP part of the VP, or even an NP-subject. Therefore, in a SOV language, the verb-second structure may show up either as XPVS or as SVXP. Taking into account the basic order SOV and the principles that govern the syntactical representations, current studies have assumed (cf. Adams (1988); Roberts (1992); Haider (1986); among others) that the orders XPVS/SVXP may be derived by movement of the V to the second position in the sentence (C° position), and by displacement of the constituents XP or S to the sentence-initial position (Spec/C position).7 However, since OP is a language whose basic word order is SVO, the constructions in which the subject NP takes the first position in the sentence and the verb takes the second obliterate the effects of the movement of the subject to Spec/C' and of the verb to the head C°; that is to say, no evidence of structure may be derived from constructions such as in (4), in which the NP subject satisfies the verb-second configuration, because the sequence in surface presents the same linear word order as its corresponding D-structure. (4)

a. O honrado padre Sao Beento deu todalas cousas (2.28.2) the honorable priest S. B. gave all-the things "The honorable priest S. B. gave all the things' b. Aquestas tres moravam en hua casa (4.13.8) 'Those three lived in a house' c. Tu es homen boo (3.37.53) you are man good 'You are a good man' d. Roma secara en si meesma (2.15.13) Rome wither-will in itself 'Rome will wither in itself.'

We consider it a relevant fact that, even in a small corpus out of 57 root clauses, 31 were occurrences of XPV(S) constructions . The constructions with realized subjects in these cases have the forms shown in (5): (5) a. Ca assi temian todalas bestas a agua (1.2.38) so then feared all-the beasts the water 'So all the beasts feared the water' b. Daqueste miragre diz San Gregorio que... (3.12.12) of-this miracle says San Gregorio that... 'San Gregorio says of this miracle that...'

114

THE DIACHRONY OF VERB SECOND

c. E todo o contrairo faz a Escritura (3.34.27) and all the contrary does the Scripture 'And the Scripture does all the contrary' d. E tanto creceu a agua derredor da eigreja (3.20.7) and so much grew the water around the church 'And the water around the church grew so much' e. E desto se nembrou el (2.16.7) and of-this REFL remembered he 'And he remembered this' f.

Com tanta paceenga sofria ela esta enfermidade (4.13.13) with so much patience suffered she this disease 'She suffered this disease so patiently'

g. Ca no meu mosteiro foi um frade (4.24.15) so in my monastery went a friar 'So a friar went into my monastery'. Here the XP is represented by constituents generated under the VP or by adverbial phrases. In the construction types illustrated in (5), the NP subjects (r-expressions or pronominal elements) occupy a postverbal position, and the verb comes in second position. Thus, we conclude that the strings in (5) are manifestations of verb-second structures. The examples in (5a-e,g) instantiate the use of the connective elements e 'and' and ca 'so/thus' in sentence-initial position. According to Mattos e Silva (1989), e and ca are largely used in initial position of root sentences in the DSG. In her work on the verb-second phenomenon in OF, Adams (1987) identifies the frequent use of the adverbs such as ainsi 'thus', si 'thus, so, and', lors 'so', or 'now' and et 'and' with a syntactical recourse to comply with the verbsecond restriction without the need for the raising of any other constituent to Spec/C: (6)

a. Ensi fut Joseph perdue une grant piece Thus was Joseph lost for a long time (R.Gr.27) (Adams, p. 104) b. Et vous dit — que entre Brinde en Auvergne et and to-you say (I) that between Brinde in Auvergne and cause... Eause... 'I am telling you that between Briude in Avergne and Eause' (Adams, p. 114)

The data under (5) allow us to say that the frequent use of items e and ca in initial position of root clauses in OP is not a syntactical recourse to comply with the verb-second restriction in view of the fact that these items co-occur mostly with other fronted XP constituents: ca assi, e todo contrario, e tanto, e desto, ca no meu mosteiro. However, the construction (7) below seems to

EVIDENCE FOR A VERB-SECOND PHASE IN OLD PORTUGUESE

115

indicate that enton 'then/now' is used as a syntactic recourse to meet the verbsecond restriction (example not included in Table 5.1): (7)

Enton disse San Gregorio... (1.16.37) Then said San Gregorio 'Then San Gregorio said'.

The conclusion which may be reached at this point is that e and ca may virtually appear in front of or outside of any type of matrix clause,8 since such elements may precede SVO or CVS clause (examples not included in Table 5.1) (but cf. below: cf. also SECTION 2.3): (8)

a. E ele non Ihos quis dar (1.28.28) and he not them+it wanted give 'And he did not want to give it to them' b. E tan comprida era a vida (1.2.5) and so complete was the life 'And life was so complete'.

These facts corroborate Mattos e Silva's classification of the element e (and ca) as a conjunction rather than an adverb. Also, we agree with her that the elements e and ca, when at the beginning of a matrix clause, are used to connect a new sentence to the foregoing discourse. Thus, we conclude that they are positioned outside of the sentential syntactic structure, i.e., outside CP. We do not mean by these observations to exclude the possibility of e (and ca) occurring in certain contexts with adverbial value, just as et is found in some 13th-century French structures. A more detailed study of these constructions is required before we can determine whether that is a probable property of e. Notwithstanding, a brief and unsystematic observation of root clauses introduced by e has provided the types of examples below (not included in Table 5.1): (9)

a. E juntaron-se muitos homees... (1.27.3) and gathered REFL a lot of men... 'And a lot of men gathered together...' b. E foronse logo muit'aglnha (1.2.45) and went-3pp REFL soon very fast 'And soon they went out very fast' c. E disse-o logo aos frades (4.6.7) and told-3sp-it soon to the friars 'And he soon told it to the friars' d. E non conta a Escritura que... (2.30.17) and not says the Scripture that... 'And the Scripture does not say that...'.

The examples (9a-c) are all cases of the phenomenon known as Tobler/ Mussafia law in the traditional literature. This phenomenon, found in the

116

THE DIACHRONY OF VERB SECOND

Medieval Romance languages, implies that the clitic should not occur in the sentence-initial position; so, the use of the enclitic pronoun is obligatory whenever proclisis will place the clitic in the sentence-initial position (cf. Cardinaletti and Roberts (1991); Salvi (1990) and Beninca (this volume) for further details on the Tobler/Mussafia law effects). Cardinaletti and Roberts analyse enclisis as a process involving the presence of an inflected verb in C° (the verb undergoes structure-preserving topicalization), the Spec/C position being empty. So, the enclitic use of the pronouns in the examples (9a-c) may be derived from the fact that the verb is the first constituent of the sentence and the Spec/C' position is empty; thus, the conjunction e is outside CP.9 In the example (9d) non would be considered as the first constituent of the sentence. In such a case, in this structure as well, the conjunction e would be outside CP. Of all the strings computed as XPV(S), the ones which are associated with phonologically null subjects are of the type given below: (10) a. D'alguas cousas me calarei (1.5.25) of-some things REFL silent-will-Isp 'I shall be silent on some of the issues' b. Tan aglyha o passaron (1.2.46) so fast passed-3pp by it 'So fast they crossed it'. Adams (1987) and Vance (1988, 1989) point out that null subjects are frequent in matrix clauses in 13th-century French: (11) a. Si firent grant joie la nuit (R.C1.XII) so made (they) great joy that night They celebrated that night' (Adams, p. 44) b

Grant piece parlerent de ceste chose (Q3) great piece spoke-3pl of this thing 'They spoke a great deal about this thing' (Vance, p. 97)

According to Adams, in the construction types under (11), the position of the null subject is postverbal. For her, this is the basic configuration for licensing null subjects: pro is licensed by canonical government of Agr° in C°, in verbsecond contexts. A strict correlation between null subject and context of inversion verbsecond context) is not kept in the 15th century, according to Vance's analysis (cf. also Roberts (1992); Hirschbuhler and Junker (1988); Dupuis (1988)). She also observes that in the 15th century pro (referential/expletive) also appears in both matrix and embedded clauses in non-inversion contexts, namely, in preverbal position. Thus, French changed from a system that only licensed null subjects in a verb-second context (13th century) to a system that licensed null subjects in preverbal position (15th century). This change in the context of pro-licensing follows from the progressive loss of the verb-second property, according to Vance, and from the possibility

EVIDENCE FOR A VERB-SECOND PHASE IN OLD PORTUGUESE

117

of Nominative Case according to Roberts (1992). In Roberts, the realization of pro in postverbal position is related to the parameter of Nominative assignment: pro must be licensed in a configuration of Case assignment, and the possibility of licensing null subject under government is dependent upon Nominative Case being assigned under government. In these terms, in a language where Nominative is assigned only under government, pro may be licensed under government only, which explains the distribution of null subjects in the 13th century (cf. also below) We will not go into the discussion of the mechanism of pro-licensing in OP and what could be the role of C° [+Agr] (Tomaselli (1990)) in the licensing condition for pro. However, we want to offer evidence that referential pro could be licensed in both pre- and postverbal position and it was not sensitive to the root/embedded distinction. Vance (1989) utilizes, as a test to determine the position of pro, a comparison to parallel constructions with lexical pronouns. In her 15th-century data, non-pronominal VS occurs occasionally, whereas pronominal VS does not. The conclusion she reaches, then, is that "verb-fronting has not occurred and that pro is in preverbal position in VX structure" (p. 10). The application of Vance's test to define the contexts for pro in the data from the DSG results in the evidence that pro was licensed in both pre- and postverbal positions, as the examples below illustrate ((12a) and (12b) not included in Table 5.1) (my italics): (12) a. Que pescado cuidas tu ora que ti nos Tragamos what fish care-2sp you now that to-you we bring-lpp naquestes monies? (1.2.7) in the mountains? 'What fish do you think that we can bring to you in these mountains?' b. se queres tu mais saber da obra (1.7.19) if want-2sp you more to know about the work 'If you would like to know more about the work' c. Eu queria de boa mente sempre ouvir...(1.29.3) I wanted of good mind always to-hear... 'Truly I wanted always to hear...' d. e tanto...que chegou ela aas feestras (3.20.3) and so... that arrived she (= water) at-the windows 'and so...that it arrived at the windows'. So, the pronominal subject could appear in both pre- and postverbal positions in matrix clauses (cf. (12c) and (12a)) as well as in embedded clauses (cf. (12a), (12b) and (12d)). In all these context types the subject can be null. Therefore, it may be concluded, based on that test, that pro could be licensed in both pre- and postverbal position in root/embedded clauses. Assuming the licensing conditions for pro discussed in Rizzi (1986) we can expect that the licensing contexts for referential pro in OP and OF are

118

THE DIACHRONY OF VERB SECOND

not the same. Pro can only be formally licensed in contexts of Case assignment, i.e., in a context either under agreement or under government by Agr°. However, when pro is licensed under agreement, rich inflectional verbal morphology is required for its content-licensing or identification (cf. Roberts (1992) and Cardinaletti and Roberts (1991)). So, the possibility of licensing referential pro under agreement in 15th-century French texts (Nominative Case-assignment possible) is restricted to some syntactic contexts (mainly in 2pl null subject contexts) (cf. Hirschbiihler and Junker (1988) and Vance (1989)), since the 12th/ 13th-century OF verbal system was already undergoing erosion. However, the 14th-century OP Agr-system contains person/number features sufficiently rich to recover the content of pro: thus OP verbal morphology is able to license referential pro under agreement. This way, in the 14th-century OP structures pro could be licensed in contexts under agreement or under government by Agr° (Nominative Case assignment possible in both contexts: cf. SECTION 3). 2.7.2 The Completive Clauses In general, verb-second effects have been analysed as resulting from the movement of the verb to C°. Since C° in embedded contexts is already occupied either by a complementizer or by an abstract +Wh feature, there may not be movement of V° to C°. Therefore, verb-second effects typically involve root clauses only. Notwithstanding, some systematic exceptions have been found in languages like German and OF, for example: verb-second is possible in completive clauses with bridge verbs, with or without realization of the complementizer corresponding to 'that', in a CP-recursion structure (cf. Roberts (1992) and Cardinaletti and Roberts (1991)). The data in Table 5.1 show that there should be in OP some type of embedded clauses which were verb-second, since, out of the 56 completive clauses analysed, 19 are SV(C) and 12 are VPV(S). We give examples of the two below (rny italics):

(13) SV(C) a. Entendemos nos que a alma vive (4.4.55) understand we that the soul lives 'We understand that the soul lives' b. Dizemos que a alma recebe peas (4.27.10) say-1 pi that the soul receives sufferings 'We say that the soul undergoes sufferings' c. Acaeceu huu dia que sen padre e sa madre happened one day that your father and your mother fezeron gram jantar (1.2.6) prepared big dinner 'It happened, one day, that your parents prepared a big dinner'.

EVIDENCE FOR A VERB-SECOND PHASE IN OLD PORTUGUESE

119

(14) XPV(S) order a. Soube que en aquela hora morrera (4.6.7) knew-lsg that in that hour died-3sg 'I knew that he had died at that time' b. Nunca leemos que meestre nen huu ouvesse (1.2.25) never read-1 pi that master none had 'We have never heard that he had had a master' c. E por esso diss'el que aqueles juizos de Deus and because-of that said he that those judgements of God pronunciara el que sairan ja da sa hoca (2.16.40) uttered he which came-3pl already from his mouth 'And because of that he said that he had uttered those judgements of God which had just come out of his mouth' d. Acaece que pelas boas obras que o homen fat happen-3sp that by-the good deeds which the man does acrecenta Deus depots a sa grafa e os seus does (1.7.16) adds God after his grace and his gifts 'It happens that by the good deeds which man does, God later adds his grace and his gifts'. The constructions in (14c) and (14d) are clear cases of embedded verb-second, with fronting of a VP complement and a circumstantial XP, respectively. They indicate that OP has verb second in completive clauses. These cases display lexical subjects. The topicalized constituents must occupy the position Spec/C (in a structure involving CP-recursion, under the analysis proposed by Roberts for completive verb-second constructions in OF) and the verb must occupy the head position C°. So, these examples can be analysed as involving selection of C* [+Agr], giving rise to embedded verb second. Cases (14a) and (14b) display null subjects. I think these examples could be treated as cases of embedded verb second as well. There are verb-second languages which do not keep the matrix/embedded asymmetry with those structures. These languages, then, allow verb-second structure in both root and embedded clauses as is the case of Yiddish (cf. Diesing (1988) and Cardinaletti and Robert (1991)) and Icelandic (cf. Roberts (1992) and Cardinaletti and Roberts). The matrix/embedded symmetry has been analysed as being derived from structures which do not include the C° node. Roberts and Cardinaletti and Roberts propose that embedded verb-second structures in languages such as Icelandic, Yiddish and 13th-century French are derived from the possibility of AgrP-recursion. The data from the DSG which we examined are not very clear regarding this fact, as the examples below (not included in Table 5.1) show: (15) a. Aqueles que no moesteiro viviam (1.5.78) those that in-the abbey lived Those who lived in the abbey"

120

THE DIACHRONY OF VERB SECOND

b. Juiz a que se ren non asconde (4.41.12) judge to whom REFL nothing not hides 'A judge from whom nothing can be hidden' c. pera saberen os que nados eran (2.1.36) to know-3pp the-ones that bom were-3pp 'for them to know the ones who were born' d. aqueles que ordiada vidafazem (1.2.19) those that tidy life make 'those that make life tidy' e. For Esso Pedro, non fez Deus senon aquelo for this, Peter, not did God otherwise that que ordinhado tiinha (1.16.24) which ordered had Therefore, P., God did nothing else than what he had determined'. The examples in (15) are good cases of Stylistic Fronting, an existing rule in contemporary Icelandic and in the Medieval Scandinavian languages (Platzack (this volume) discusses some examples of this kind of construction in Icelandic, Yiddish, Medieval English and in OF). This rule fronts some VP constituent (for instance, a participle, an adverbial, a negation or a complement) to a position between C° and the inflected verb; the basic requirement for the Stylistic Fronting operation is that the subject position be phonologically empty, holding a trace of pro (cf. Maling (1980)). Maling claims that this rule is particularly common in subject-relative sentences in Icelandic and OF. Considering Padua's (1960) remark that relative structures as those exemplified in (15) (with verb-final order and omitted subject) are very common in OP, we may conclude that Stylistic Fronting constructions seem to be common also in OP. Thus, in the relative constructions mentioned above the fronted elements are: no moesteiro in (15a), ren in (15b), nados in (15c), ordiada vida in (15d) and ordinhado in (15e). Cardinaletti and Roberts (1991) analyse Stylistic Fronting constructions as double-Agr structures, in VO languages. This way, the examples in (15) can be seen as evidences that a double-Agr system was at work in 14th-century OP. Adverbial sentences like those in (16) seem to be verb-second structures (examples not included in Table 5.1): (16) a. se Ma molher prenhe metessen en huu career (4.1.10) if a woman pregnant put-3pl in a jail 'if they put a pregnent woman in a jail' b. porque ata aqui contei eu os feitos groriosos (3.1.2) because until here told I the deeds glorious 'because until now I have told the glorious deeds'. These examples seem to be embedded Wh-clauses with verb second. latridou and Kroch (1992) state that only semantically vacuous CPs can undergo re-

EVIDENCE FOR A VERB-SECOND PHASE IN OLD PORTUGUESE

121

cursion and that this recursion is limited to environments where the recursive CP is governed by a verb. Thus, there is no generalized CP-recursion (cf. also Cardinaletti and Roberts (1991)). So, being adjunct clauses, the sentences in (16) cannot be treated as cases of CP-recursion. We assume a double-Agr structure to derive the verb-second order in these embedded Wh-clauses. It is worth noticing that embedded verb second in adjunct clauses is not generalized. It seems restricted to limited clauses as those in porque 'why' and se 'if. This situation should be compared to the one in Yiddish mentioned in Cardinaletti and Roberts.

2.2 Verb-Initial Clauses The three matrix VS-clauses and the eight V(C)-clauses, which are constructions that show up in verb-first structures are of the following type: (17) a. Conven, Pedro, que te cales (2.14.3) suits, P., that REFL silent-2ps 'It is convenient that you be quiet, Peter' b. Ide-vos a boa ventura (1.2.44) go-you with good luck 'Go and good luck'. c. Diremos nos ora, padre, que...(1.4.16) say-will-lpp we now, father, that... 'we will say now, father, that...' d. levaron-nos aa pousada homens que hi estavan (1.28.31) took-lpl-us to the lodging men who there were-3pl 'men who were there took us to the lodging'. The first two (17a,b) are directive constructions. These kinds of verb first are root phenomena and the landing site of the V° in these constructions is C°. The last two (17c,d) can be analysed as the kind Hirschbiihler and Junker (1988) call discursive (cf. below). The last one (17d) illustrates "free" subject inversion in the DSG. In verb-second languages, verb-first constructions are restricted to certain environments. Generally verb-first structures occur in yes/no questions, conditionals and in imperatives. Hirschbiihler and Junker identify verb-first constructions licensed, as they say, by a discursive factor. Roberts (1992:133) analyses verb-first structures of the type below (18a,b) as involving a null operator of some sort in Spec/C': (18) a. Tienent

oiseaus por lor

Hold (they) bird

cors

deporter (Le Charroi de Nimes 1.26) for their bodies to disport

b. Voit le li rois (Le Charroi de Nimes 1.58) sees him the king.

122

THE DIACHRONY OF VERB SECOND

In his terms, the verb-first constructions in (18), as well as the ones mentioned above (17), all present the Spec/C position filled in by a phonologically null operator, which allows him to maintain that these are true verb-second structures. All those possible verb-first structures occur in OP; some of them were illustrated in (17). As to the verb-first yes/no interrogative constructions, we note that Padua (1960:94) considers that the placement of the verb in sentenceinitial position constitutes "the major recourse to mark interrogation." In what follows, we give some examples not included in Table 5.1 of this type of construction: (19) a. Acaeceu, padre, ja algua cousa nova por que happened-3sp, father, already some thing new why choras mais que sooes? (1.1.8) cry-2sp more than you are used to 'Father, has anything happened over which you cried more than you use to?' b. Cuidas, padre, que este homen ouve alguu meestre care-2sp, father, that this man had some master que o ensinasse? (1.2.17) that him taught 'Do you think, father, that this man had a master who taught him?' c. Queres, Pedro, que ti conte...? (1.16.38) want-2sp, P., that you-Acc-2sp tell-lsp...? 'Do you want me, P. , to tell you...?' Following Rizzi's (1991) proposal that yes/no questions have an empty operator in Spec/C', these structures can be analysed as involving a Spec/C' position filled in by null operator and the inflected verb in C° positon. The seven embedded VS constructions computed in Table 5.1 are of the following type (my italics): (20) a. Crees que ando eu sen almal (4.4.50) believe-2sp that am I without soul 'Do you think that I have no soul?' b. Non ti semelhas, Pedro, que deven a aver gram not to-you appear-3sp, P., that must to have great vergonha os homees (11.15) shame the men 'Don t you think, P., that men must be honorable?' c. Conta San Gregorio que veo huu principe (1.3.2) say-3sp San Gregorio that came a prince 'San Gregorio says that a prince came'. Although we have grouped them as VS, the constructions above comprise two different syntactic structures: the one in (20b) is of the VCS type or Romance inversion (free-subject inversion); those in (20a) and (20c) are both

EVIDENCE FOR A VERB-SECOND PHASE IN OLD PORTUGUESE

123

unaccusative structures, in which the subject may remain in its basic position inside VP. In this way, there is no evidence of movement of V° to C° in either construction. Due to what was observed in SECTION 2.1.1, on the possibility of pro being licensed in preverbal position, declarative verb-initial structures such as those in (21) below occur (frequently) in the DSG text: (21) a. Fez mui gram chanto (2.8.27) did very great cry 'He cried a lot' b. Acharon as sas maaos e

os seus pees tan ben saos (4.25.13) found-3pl the his hands and the his feet also healthy 'they also found his hands and his feet recovered'.

In the examples above, referential pro occupies Spec/X' position. These examples may be treated as cases of pro-realization in preverbal position.

2.3 Verb-Third Clauses In the corpus analysed, omission of the subject or its allocation to the postverbal position takes place when either a verb-subcategorized complement or a circumstantial occupies the first position in the sentence. According to Mattos e Silva (1989), these construction types are frequent in DSG. This allows us to confirm with a certain amount of confidence that matrix sentences in OP may be characterized as verb-second structures. Vance (1989) notes the occurrence of different types of verb-third structure in 15th-century French. We want to draw a distinction here between those structure types: a) CSV, in which C is a complement subcategorized by the verb; b) XPSV, in which XP is a sentence adverb; c) SXPV, in which XP may be either a complement subcategorized by the verb or a circumstantial. In what follows, I will treat each one of them. In the data examined, no occurrence of matrix structure type CSV was found, which was in accordance with Mattos e Silva's (1989) observation that the order CSV does not occur in the DSG when a direct object or a prepositional complement occupies the matrix sentence-initial position. Padua (1960:84-85) presents the following instance of CSV-type, from the Leal Conselheiro text, dated from the first half of the 15th century: (22) Todas astas cousas as gentes demandam (Leal cons. XXXV) all these things the persons demand 'People demand all these things'. P£dua says this is a construction scarcely used in OP, and that, in a way similar to OF, the placing of the complement in initial position leads, in most cases, to subject-verb inversion. Padua's statement indicates that still in the 15th century the fronting of a complement triggered verb-second structure.

124

THE DIACHRONY OF VERB SECOND

Vance (1988, 1989) remarks that, in Middle French, CSV word order cooccurs with CVS word order in matrix and embedded clauses. Verbsubcategorized complements, which in the 13th century triggered obligatory inversion, as well as sentential adverbs, may appear in the initial position of structures of the form CSV in the 15th century, without triggering inversion. The following examples illustrate this situation: (23) a.

Et lors ilz commencerent a rire (Q64) (Adams (1988:94)) And then they started laughing

b. le petit Saintre les yeux de Madame ne cessoient deregarder... (J55) (Vance, p. 163) the little S. the eyes of Madame not ceased 'Madame's eyes did not cease to look at little Saintre' c. et aussi fis je de par vous (J104) (Vance p. 159) and thus did I from by you 'and I did likewise with respect to you'. The example in (23a) is a construction from the text La Queste del Saint Graal (13th century). It illustrates a rare context of CSV word order in which C is circumstantial; the construction in (23b) exemplifies a CSV structure type from the text Jehan de Saintre (15th century) in a context which, in the 13th century, required the CVS order. In this construction, C is the equivalent of verbal complement. The case in (23c) shows that the CVS order also occurs in the 15th century in a context similar to those found in the 13th century. Although no occurrence of matrix CSV structure type was registered in the data examined, some XPSV constructions did occur, in which XP is not a constituent subcategorized by the verb. Certain adverbial elements did not trigger obligatory subject-verb inversion as shown by the examples below (with the exception of (24e), not included in Table 5.1): (24) a. e

assi o

santo homen defendeu os seus discipulos (1.9.13) and thus the saintly man defended the his disciples 'and the saintly man thus protected his disciples'

b. E enton huu homen siia en sa pousada... (1.2.25) and then a man sat down in his inn 'Then a man sat down in his inn' c. E assi o fez o poder de Deus (1.2.47) and thus it made the power of God 'And the power of God thus made it' d. E pois se juntaron dois homens ou tres (2.9.3) and then REFL gathered together two men or three 'And then two or three men gathered together' e. E enton respondeu o abade santo e disse (1.8.33) and then answered the abbot saintly and said 'And then the saintly abbot answered and said'.

EVIDENCE FOR A VERB-SECOND PHASE IN OLD PORTUGUESE

125

The constructions above show that adverbial elements as such as enton, pois and assi permit two word orders, namely XPadvSV and XPadvVS. The structure of the form XPadvSV indicates that the XP-constituent is in adjunction either to CP or to AgrP. It is widely assumed that matrix declarative clauses are CPs in a verbsecond system and AgrPs in a non-verb-second system (but cf. note 7); and adverbials adjoin more readily to AgrP than to CP. Consequently, the order XPadvSV is impossible in matrix clauses in verb-second languages. However, a small number of adverbials are able to appear in that position, even in a verb-second language like German: (25) ...denn Johann hat gestern das Buch gelesen so J. had yesterday the book read (Roberts (1992:186)) Vance (1989) notes that in Jehan de Saintre (15th century), a large class of adverbial elements appear in initial position, without triggering inversion, as seen in (26) below: (26) Lors la royne fist Santre' appeller then the queen made S. to-call 'Then the queen had Santre called' (Vance, p. 158). She analyses this structure as resulting from the adjunction of the adverbial to AgrP. The choice between the two possibilities of analysis (adjunction to CP, as in the case of German in (25) or adjunction to AgrP, as in the OF case in (26)) is dependent upon various factors, including the frequency of that word order type in the DOS (and the other corpora representative of the 14th century), and the definition of the class of elements able to appear in initial position without triggering inversion. Only after these factors are considered can we get the evidence needed for a decision as to whether or not matrix declarative sentences in OP were always CPs. In the sample analysed for Table 5.1, we registered one occurrence of the SXPV construction type: (27) El con sa mao deu a oferta (2.23.17) he with his hand gave the offer 'He himself presented the offering'. The order SXPV is also attested in embedded clauses in OF, with XP representing different complement types (cf. Vance (1988, 1989)). Below is one example: (28) Je vous dis que vos avec moi venez (Ql) I to-you say that you with me come 'I tell you to come with me'. According to Vance's analysis, the SXPV clauses in 13th-century French are derived via topicalization to Spec/V position of an adverbial or a VPcomplement. She further adds that these structures are found only in embed-

126

THE DIACHRONY OF VERB SECOND

ded clauses.10 Nevertheless, in her analysis, 15th-century matrix SXPV clauses appear as a new possibility of adjunction to AgrP: (29) Madame en sa chambre entra (J 7) Madame into her bedroom entered 'Madame entered her room'. The occurrence of one single root sentence of the SXPV type does not allow us to decide on an analysis for it, at the moment. Thus we noticed that time and manner adverbials may be topicalized in OP matrix clauses (-Wh), occurring in structures of the verb-third type: this possibility does not hold for complements, that is, the order XPSV occurs when XP is adverbial, but not when XP is a complement. Genuine verb-second systems do not allow this word order. Roberts (1992) remarks that a diachronic movement ending up in the loss of OF verb-second property began before the end of the 12th century. The texts of the llth century and of the first half of the 12th century are all "perfect" verb second (Roberts' term), considering they present only a few cases of verbfirst or verb third and the subject does not show an overwhelming tendency to appear in the first position. The late 12th and 13th-century texts already present a slight evidence of verb-second erosion. However, as Roberts himself remarks, the preference to place the subject in first position or to place any constituent out of CP, as seen in the late 12th and 13th-century texts, does not interfere initially with the verb-second property of OF. For him, the technical sense of verb second is correlated with the property of C° containing [+Agr] (cf. also Tomaselli (1990)). The picture we have just painted of the DSG suggests it is not a perfect verb-second text. However, we suggest that the OP grammatical system was verb-second in the technical sense, a system in which C° has the feature [+Agr]. The evidence for this will be presented in 2.4 and will be correlated to the ability of C° to host the clitic complement.

2.4 Cliticization and the Complementizer Head The distribution of the complement pronouns (= clitics) offers one more set of evidence in favor of the analysis that considers OP as a language which has verb-second properties. We shall focus mainly on their distribution in embedded structures, for they most properly highlight the facts to be considered. In completive structures in DSG, the complement clitic pronoun systematically occurs adjacent to C°. Different types of elements can intervene between the clitic pronoun and the verb but not between the constituent C° and the clitic. The examples below are illustrative (examples not included in the analysis presented in Table 5.1):11

EVIDENCE FOR A VERB-SECOND PHASE IN OLD PORTUGUESE

127

(30) a. ...que ti n6s tragamos naquestes montes (gloss = (14a)) b. mandou que o non dissessen a nenguu (1.7.22) ordered-3sp that it not said-3pp to noone 'He ordered them not to say it to anyone' c. e dexia que se lhi non enviassem Basilio (1.5.68) and said that REFL him not send Basilio 'and (he) said they did not send B. to him'. In the construction under (30), the clitic is always adjacent to the complementizer que 'that'. Between the clitic and the verb a pronominal subject occurs in (30a), a negative element in (30b and c). If we assume Tomaselli's (1990) analysis dealing with subject clitics in (standard) German, we may then say that the structures in (30) reflect the property of the head C° being the host for clitics. Only a head C° associated with Agr can host a clitic. Thus, in view of the fact that the property of having Agr in C° (or a pronominal C° in Tomaselli's words) has been analysed as pertinent to the verb-second languages, it may be concluded that C° is characterized by the feature [+Agr] (or [+pronominal]) in OP and that, therefore, OP instantiates a verb-second language type. In a similar way, in embedded structures introduced by a +Wh-element, the clitic object occurs strictly adjacent to that element as the data below illustrate: (31) a. aqueste por que me tu rogas (1.5.61) that for whom me you implore 'the one for whom you implore' b. ata que lhi a alma saisse da came (4.12.13) until when him the soul left the flesh 'until when the soul left his body' c. de

seu cavalode

que

o

primeiramente derribaron (1.2.44) from his horse from where him firstly fell-3pp 'from his horse from which he first fell'

d. ainda que o el primeiramente salvasse (1.7.20) even though him he firstly saved 'even though he firstly saved him' e. don que lhi a el Deus derea (1.5.57) gift that him to him God had given 'Gift that God had endowed him with' In the examples above, the clitic is always adjacent to the +Wh elements which introduce the embedded structures; between the clitic and the verb occur a pronominal subject in (3la), a non-pronominal subject in (31b), an adverbial in (31c), a pronominal subject and an adverbial in (31d) and a prepositional complement (= clitic reduplication) and a non-pronominal subject

128

THE DIACHRONY OF VERB SECOND

in (31e). These data can be explained if it is assumed that C° is abstractly characterized by the feature [+Agr] in OP, attracting the clitic complement to this position.12 The example below clearly illustrates the distribution of complement clitics in matrix and embedded clauses (data not included in Table 5.1): (32) E non lho dissera se o non conhocera (4.30.9) and not to-him+it had-told if him not had-known 'And he would not have told it to him if he had not known him'. If C° is [+Agr] in OP, then the matrix clause above may be analysed as a verbsecond structure, with both the clitic and the verb located in the head C°, and the negative element non located in Spec/C. We may now conclude our discussion on the OP structures. The evidence presented in this section indicates that OP used to observe the restrictions relevant to verb-second languages. In SECTION 2.1 we considered the situation of verb-second structures in root and embedded clauses. We observed that in declarative root clauses the fronting of an XP constituent from VP triggers subject-verb inversion, giving rise to an XP V S structure. We took account of the complement clauses' status and observed the existence of verb-second complement clauses of the same type observed in OF (in CP-recursion structures). Stylistic Fronting clauses are found in Wh-complements (in AgrPrecursion structures). Taking the pro licensing condition into consideration we propose that in OP pro was licensed in pre/postverbal position, i.e., in contexts either of government under Agr° or under Spec/head agreement (agreement between Agr° and pro in its specifier position). In SECTION 2.2 we remarked that verb-first structures are found in yes/no questions, and in directive and discursive constructions. We also observed that pro can be preverbal in verb-first constructions. In SECTION 2.3 we took a look at the status of verb-third clauses and observed that time and manner adverbials such enton, pois and assi admit of two word orders: XPSV (= verb-third structure)/XPVS (= verb-second structure); on the other hand the order XPSV did not appear in the data when XP was equal to verb complement. In SECTION 2.4 we suggested that the complement clitic pronoun is cliticized to C°.

3. The Change It is known that Modern BP does not manifest verb-second structures, that the fronting of an element to clause-initial position does not result in a verbsecond string, and finally that, even in direct interrogative constructions, verbsecond residues started to disappear by the late 19th century. Assuming the change from a verb-second language to a non-verb-second one, we will try to present a sketchy analysis of certain facts that were brought up here in relation to the element which triggered the change. The Principles and Parameters model considers linguistic change as a modification in the setting of certain parameters (cf. Rizzi (1988); Roberts (1992);

EVIDENCE FOR A VERB-SECOND PHASE IN OLD PORTUGUESE

129

Lightfoot (1991), among others). This theory predicts that, if a language loses a basic property linked to a given setting of a parameter, it will also lose the other properties related to that setting, which leads to a restructuring in the language (i.e., a change) According to Rizzi, restructuring is not an instantaneous process and there is a certain amount of inertia to be taken into account when one deals with such a complex system as language. In these terms, an accurate and realistic characterization of the changing process must imagine that "the language in question will enter a phase of instability which will be resolved by eliminating some of the properties related to the abandoned value, and by confining some of them as relics on special stylistic registers or as crystallized construction markers" (Rizzi, p. 17). The verb-second residues of English, French and Portuguese illustrate this point, since they are relics of a general process of moving the inflected verb to a second position in the sentence, which was very productive in these languages at former stages, but which is now confined to a few marked constructions. According to theoretical assumptions children build their grammar from what they hear (primary data) and from what they already possess innately (Universal Grammar); also, they do not have direct access to the grammar of the people around them (cf. Lightfoot (1979, 1981, 1991), among others). This being so, the child's grammar (grammar 2 = G2) and the adults grammar (grammar 1 = Gl) are totally discontinuous: (33) Primary data 1 Primary data 2

Grammar 1 Grammar 2

Because the surface structures may be compatible with more than one grammar and also because children do not know which analysis represents Gl, children may opt for an incorrect hypothesis about Gl; this will result in a grammatical change. Hence, the grammar that children build may differ from the adult's grammar (cf. Lightfoot (1979, 1991)). Adams (1987) argues that the reason why OF lost its verb-second effects was that children reanalysed the derived SVO structure (= verb-second structure) as a basic SVO one. Having also lost the initial accent, typical of verbsecond languages, the surface structures did not present enough evidence for the children to assume OF as a verb-second language, inasmuch as the SVO order was very frequent. I take a position different from Adams, by assuming that the sole reanalysis of a derived order as basic is not a sufficient condition for the loss of verb-second status, even if that is connected to a change in accent pattern. Thus, in the case of OF, Gl analyses SVO as a structure derived by the displacing of S and V; G2 analyses SVO as a basic structure. However, children had enough evidence of verb-second structures, produced by Gl, and these could not be analysed as basic structures. Furthermore, by considering the time period for the loss of verb-second (14th-16th centuries), we have to admit that G2 also generated XPVS structures. Therefore, we believe that a single reanalysis of a derived structure as basic and the loss of initial secondary accent are not enough to explain the loss of verb second in OF.

130

THE DIACHRONY OF VERB SECOND

The grammar of a language encompasses innate principles and a certain number of parameters. Then, it is supposed that the reanalysis of a structure may (but need not) lead to a change in parameter. As Rizzi (1988) observes, aspects of linguistic change may be regarded as modifications in the setting of certain parameters. Much along the same lines Roberts (1992) advocates that the loss of verb-second order in OF can be defined as a change in the way in which Nominative Case was assigned. Koopmann and Sportiche (1991) put forward the proposal that structural Case may be assigned under government, as shown in (a), or agreement, as shown in (b):

Government is defined as a relation between a head and its complement or the specifier of its complement. In the configuration (a) X° governs the YP constituent, its complement, and the NP constituent, the specifier of its complement. Agreement is a structural relation between a head and its specifier. In the configuration (b) there is an agreement relation between the head X° and its specifier. Roberts accepts and elaborates this proposal of Nominative Case assignment. He assumes that the choice between one or the other option (government-only or agreement-only) or of both options (government and agreement) for Nominative assignment shown in the configurations in (34) is a parametric one. He exemplifies the application of this system of Nominative assignment by Agr° in relation to the parametric choices as follows: (35) a. government and agreement — English, Middle French b. agreement, but not government — French, Italian c. neither government nor agreement — Welsh For Modern BP, Roberts (p.32) proposes that Nominative assignment by Agr° follows (35b). In his view, a system that follows (35b) prohibits free inversion and excludes constructions such as those in French and Italian: (36) a. *A Jean pris le livre? *A Gianni preso il libro? 'Has J. taken the book?' b. *Quel film a Jean vu? *Che film ha Gianni visto? 'Which film has John seen?' In the structures above, the movement of Agr° (or V°+Agr°) to C° destroys the configuration of Nominative assignment under agreement.

EVIDENCE FOR A VERB-SECOND PHASE IN OLD PORTUGUESE

131

We agree with Roberts in that present-day spoken BP assigns Nominative under Spec/head agreement. However, 14th-century OP seems to choose the (35a) option. We shall examine the 14th-century OP data based on this hypothesis. Pondering the fact that languages which allow verb-second structures must also allow Nominative assignment via government, it may be concluded that OP allows Case assignment to the subject under government: in the (XP)VS structures Nominative was assigned under government. Thus, in the matrix clauses in (5) and in the embedded one in (14), the subject must appear in the NP position of the configuration in (34a) and the inflected verb must be in the X° position. So, the verb Case-marks the subject from that position, under government. In (37) below there is a compound verbal form involving an auxiliary-like verb and a non-finite verbal form. The occurrence of the subject between the two forms shows that Nominative is assigned under government: (37) a. devemos nos a pousar (3.5.4) must we to lie down 'We must lie down' b. deven os seus discipulos querer os seus boos feitos should the his disciples want the his good deeds asconder (1.17.30) hide 'His disciples should wish to hide his good deeds'. As to the order SV(C) in matrix clauses, such as exemplified in (4), this order does not constitute evidence in favour of Nominative assignment under agreement in a verb-second language since the subject trace may receive Nominative under government (cf. Roberts (1992)), considering that the SVC matrix clauses in verb-second languages involve successive verb raising together with some operation that fronts the NP subject. However, the SV(C) completive clauses may be seen as evidence of Nominative assignment under Spec/head agreement. So, in the examples such as those in (13), the subject must be marked for Nominative Case under agreement. Consequently, it seems that both modes of Nominative assignment were available in OP. The change in the parameter of Nominative assignment entailed the loss of verb second. The reanalysis of the derived form as basic SVO led the child to opt for (35b), that is, for Nominative assignment via Spec/head agreement. Berwick's (1982) Subset Principle states that the child always opts a priori for the most restrictive hypothesis. Therefore, a system in which Agr assigns Nominative only under agreement is preferable to the one in which Agr assigns Nominative under both government and agreement (cf. Roberts (1992)). There are still the XPVS constructions. How could the child have analysed those structures in a system showing Nominative assignment under agreement? Haider and Prinzhorn (1986) say that the fact that one generation projects a subset of a grammar which differs from that of the previous generation will not have, at first sight, drastic consequences because the resulting differences

132

THE DIACHRONY OF VERB SECOND

are disguised by adaptative rules. These rules are necessary only to interpret structures produced by previous generations; these structures. consequently, become obsolete, and this is the final stage in a change. If adaptative rules are restricted to the interpretation of structures, and not used to produce them, the realization of verb-second structures cannot be explained in these terms. We saw in Duarte's (1989) work that direct Wh-questions displayed verb-subject order until the 19th century. It is known that in the Portuguese of the 18th-century and the beginning of 19th-century matrix declarative clauses which showed movement of the auxiliary to a pre-subject position were attested (cf. Kato and Tarallo (1986); Tarallo and Kato (1989); and Berlinck (1989)) involve Nominative assignment under government. Constructions like (38) are not very much used in contemporary spoken BP; they are still found in formal written texts: (38) a. Tivessem eles cumprido o acordo, tudo seria resolvido 'Had they fulfilled the agreement, everything would have been settled' b. Maria acredita terem os meninos saido cedo 'M. believes the boys to have left early' c. Que filme tinha Joao visto? 'Which film had John seen?' d. Tem Joao feito o trabalho? 'Has John done the work?' The fact that these two types of constructions (subject-verb inversion in Whquestion and subject-auxiliary inversion like (38); cf. also the clauses in (2)) were kept in the system until the late 19th century, indicates that the change in the Nominative assignment parameter did not happen abruptly from (35a) to (35b). From a letter written by Padre Leonardo do Vale in Bahia in 1561 (Serafim Leite, 1942), we analysed the first 110 sentences and found that, out of the thirty strings with evident subjects, eleven manifested the VS order, in constructions like: (39) a. polio que vendo elle que... for what seeing he that... 'for this reason he seeing that...' b. primeiramente, foy a viagem muy trabalhosa firstly, was the trip very laborious 'firstly, the trip was very hard' c. assi que, com todos estes trabalhos, teve o Senhor so that with all these tasks, had God por bem de... for the sake of... 'thus, with all these tasks, God decided...'

EVIDENCE FOR A VERB-SECOND PHASE IN OLD PORTUGUESE

133

d. nao podiao os homens cobicosos dos beens deste mundo not. could the men greedy of-the riches of-this world fartar a sede de pecas e fazendas quench the thirst of "goods and wealth" 'the men greedy for riches of this world could not quench their thirst for goods and wealth' e. porque nem agua achavao os caminhantes because not even water found the walkers 'because not even water the walkers found' f.

despedio-se o Padre delles said-farewell-REFL the Priest of-them 'the Priest said farewell to them'

g. que cousa erao Padre nem bautismo what thing were Priest nor baptism 'what thing were Priest and baptism'. The construction in (39a) indicates that Tense assigned Nominative under government; examples (39c), (39d), (39e) and (39f) are evident cases of verb/aux fronting, the NP subject being in Spec/Agr', which favours an analysis of Nominative assignment under Agr° government ((39d) and (39e) may be verbsecond structures with nao/nem in the first position; (39f) is a case of the the Tobler/Mussafia law); though (39b) and (39g) are constructions with unaccusative verbs, they seem to be analysed as verb-second structures; the first triggered by initial adverbial and the second by XP-Wh. Anyway, since the NP subjects are in Spec/Agr' or in adjunction to VP, Nominative is assigned under government by Agr°. Reflecting on those facts and on studies by Kato and Tarallo (1986) and Berlinck (1989), which deal with the VS constructions and present empirical facts concerning their restrictive nature, we came to the conclusion that the parametric choice of Nominative assignment did not bring about an abrupt change from (35a) to (35b), but rather produced something of the type informally formulated as: (40) agreement, and government restricted to certain contexts. We believe it is theoretically possible to admit that the reanalysis to basic SVO structure led the child to opt for Nominative assignment under agreement, and to analyse the XPVS evidence as marked structures involving Nominative assignment under government. The examples in (39) seem to indicate that Nominative assignment under government became restricted mostly to constructions with an auxiliary-like verb (in Aux to C° structures). Even those VS construction types started to disappear from BP in the late 19th century and the beginning of the 20th century. Thus, we sketched in a rough outline a discussion about Nominative-assignment possibilities in OP. What we proposed was that Nominative-assignment

134

THE DIACHRONY OF VERB SECOND

by Agr° was possible either under agreement or under government configurations. We have put forward a preliminary analysis of OP Nominative-assignment possibilities.

4. Conclusion In this work I have discussed the verb-second phenomenon found in the DSG. I propose that this OP document reflects properties of a verb-second language. The verb-second nature of OP in root clauses is illustrated clearly by the examples in (5). We related the verb-second nature of OP to the feature [+Agr] in C°. OP allows embedded verb second under limited conditions. We saw that verb-second is possible in the complements to bridge verbs as in examples (14). The class of bridge verbs in question is the same one which allows verbsecond completives in verb-second Germanic languages and in OF. Verb-second is also possible in causal and conditional clauses, as in examples (16b) and (16a), respectively. Stylistic Fronting is possible in relative clauses (examples in (15)). In terms of current proposals, the inflected verb cannot move to an embedded [+Wh] C° (cf. Rizzi (1991) and Rizzi and Roberts (1989)). So, both the relative and the adjunct verb-second clauses motivate an analysis in terms of a double-Agr structure. Given this analysis of relative/adjunct clauses, we do not have to treat completive verb-second as involving movement of the verb to C°. A double-Agr structure would also suffice to derive these constructions. We noticed that finite verb in initial position is not restricted to yes/no questions, conditional clauses and imperatives. Verb-first declaratives are quite common in the DSG. This may be due to the fact that movement of V° to C° was not required for all matrix declarative clauses. I have also discussed the loss of verb-second properties of OP, based on the approach to Nominative-Case assignment advocated in Koopman and Sportiche (1991) and Roberts (1992). The evidence I presented shows that in OP Nominative Case was assigned both under government and under agreement. The fundamental difference between OP and Modern Portuguese is that the latter does not assign Nominative under government any more. Data from the 16th century indicate that the possibility of Nominative assignment under government was already restricted to contexts with verbs of the auxiliary type.

Notes A first version or this paper was presented in the course "Topicos em Linguistica Historica" given by Mary Kato and Fernando Tarallo for the PhD Program at UNICAMP. I thank Mary Kato, Fernando Tarallo, Giampaolo Salvi and Charlotte Galves for valuable comments and suggestions, and one anonymous reviewer for interesting observations. Special thanks go to Ian Roberts for his invaluable comments on the present version of this paper, and

EVIDENCE FOR A VERB-SECOND PHASE IN OLD PORTUGUESE

135

continuous support and assistance. Special thanks go also to Rosa Virginia Mattos e Silva for the many hours spent with us, discussing the 14th-century Portuguese constructions. I also thank Vicente Cerqueira and Maria Emiliana Passos for the English rendering of this paper and the latter for the careful general revision. Mistakes are, of course, solely mine. 1. The French and English data in (1) and (3) are from Rizzi (1991). 2. The constructions (2) with inversion of subject-verb to verb-subject order are not very much used in contemporary spoken BP: they are generally found in literary and formal style. Though seldom used, these constructions are not judged ungrammatical by all BP native speakers (cf. SECTION 3). 3. Verb-subject inversion is always obligatory in any root Wh-construction in Modern European Portuguese, as the following examples ilustrate: (i)

(O) que estiveste tu a fazer? 'What had you been doing?'

(ii) *(O) que tu estiveste a fazer? (iii) Onde trabalha a Maria? 'Where does Mary work?'

4.

5.

6.

7.

(iv) *Onde a Maria trabalha? Cf. Ambar (1988) for more details about VS order in European Portuguese; cf. also Torrego (1984) on Spanish verb-subject inversion in Wh-questions. Within the limits of study, the hypotheses raised here are both qualitative and quantitatively insufficient if we consider necessary for the investigation of any linguistic change the analysis of individual texts, various dialects and different epochs (Hirschbuhler and Junker (1988)). Mattos e Silva's work (1971) focuses on the analysis of a Portuguese document, named by her "A mais antiga versao portuguesa dos Quatro Livros dos Dialogos de Sao Gregorio" (The oldest Portuguese version of the Four Books of Saint Gregory Dialogues) and dated by internal chronology as before 1385. The linguistic analyses of the 14th-century structures (Mattos e Silva (1989)) build upon that critical/philological reading. This is the corpus on which are based most of my reflections regarding the verb-second phenomenon in OP. OP examples will be cited followed by the book, chapter and line numbers, according to Mattos e Silva's citations. We name "root" the affirmative independent/matrix sentences of Mattos e Silva's study, which include transitive, intransitive and copula constructions, and existential structures as well. The XPV(S) and VS orders indicate that the lexical subject is postverbal, but this is not meant to say that the subject is always postverbal when it is phonetically null (cf. comments in 2.1.1). In the V(C) order we disregarded the verbal complements represented by clitics, which are in general preverbal (but see the observations in 2.4). For Travis (1984) SVO verb-second matrix clauses can be AgrP structures. For Roberts (1992) and Cardinaletti and Roberts (1991) the landing site of the verb in verb-second structures may vary cross-linguistically; the constraints

136

THE DIACHRONY OF VERB SECOND

on verb second could be met either on the CP level or on the AgrP level, in a double-Agr structure. For the moment, we shall analyse the matrix clauses presented in this paper as CP-structures. 8. Vance (1989:91) proposes an analysis along the same lines to explain the occurrence of the element et 'and' introducing sentences in Middle French. 9. Beninca (1989) analyses enclisis as a process involving a clitic+verb in C°; since Spec/C' is empty, the verb must move to this position, resulting in an enclitic structure. Even considering her proposal, the conjunction e is outside CP. 10. In the DSG the SXP V order is common in both adverbial and relative clauses: (i)

as lagrimas que eu cada dia deito dos meus olhos (1.1.10) 'the tears which I drop from my eyes every day'

(ii)

quando eu no moesteiro vivia (1.1.11) 'when I used to live in the monastery'.

11. In Modern BP, constructions like (30), as well as those in (31), have the clitic adjacent to the verb. 12. Cardinaletti and Roberts (1991) analyse the "clitic-second" phenomenon in Germanic and Romance languages. They identify the traditionally recognized Wackernagel clitic position of Germanic languages as Agrl ° and assume Agrl ° also to be the clitic position in the Romance languages which obey the Tobler/ Mussafia law. Their proposal differentiates the clitic position from the inflected verb position. Agrl0 can also be the position for the inflected verb; the possibilities of verb movement from Agr2° to Agrl 0 may vary from language to language, and even within the same language this movement may vary according to the sentence status: matrix or embedded. In this section I put forward an analysis which considers clitic elements to be constituents of C° head in embedded (and matrix) clauses. In view of Cardinaletti and Roberts' proposal that Agrl 0 is the clitic position, how can these two analyses be related as regards the 14th-century OP data? In research in progress I attempt to argue that in OP Agrl 0 is the syntactic clitic position; however, the clitic elements must move to C° in some embedded (and matrix) structures, such as those (30, 31 and 32). It is the agreement (pronominal) feature in C° that makes the clitic-movement/adjunction to this position possible. This analysis also implies that some OP matrix clause types are instances of a double-Agr system (cf. note 7).

References Abraham W., ed. (1983) On the Formal Syntax of the Westgermania. Amsterdam: John Benjamins. Adams, M. (1987) Old French, Null Subjects, and Verb Second Phenomena. PhD Dissertation, UCLA. Adams, M. (1988) "Les effets V2 en ancien et en moyen francais." In Hirschbuhler and Rochette, eds., 13-40.

EVIDENCE FOR A VERB-SECOND PHASE IN OLD PORTUGUESE

137

Ambar, M. (1988) Para uma sintaxe da inversao sujeito verbo em portugues. PhD Dissertation, University of Lisboa. Belletti, A. (1988) "Unaccusatives as Case Assigners." Linguistic Inquiry 19:134. Beninca, P. (1989) "L'ordine delle parole nelle lingue romanze medievali." XIX Congreso Internacional de Linguistica e Filoloxia Romanicas, Santiago de Compostela. Beninca, P. (this volume) "Complement Clitics in Medieval Romance: the Tobler/Mussafia Law." Berlinck, R. de A. (1989) "A construcao V SN no portugues do Brasil: uma visao diacronica do fenomeno da ordem." In Tarallo, ed. Berwick, R. (1985) The Acquisition of Syntactic Knowledge. Cambridge, Mass.: MIT Press. den Besten, H. (1983) "On the Interaction of Root Transformations and Lexical Deletive Rules." In Abraham, ed., 47-131. den Besten, H. and C. Moed van Walraven (1986) "The Syntax of Verb in Yiddish." In Haider and Prinzhorn, eds., 111-135. Cardinaletti, A. and I. Roberts (1991) "Clause Structure and X-Second." To appear in W. Chao and G. Horrocks, eds. Levels, Principles and Processes: The Structure of Grammatical Representations. Dordrecht: Foris. Chomsky, N. (1981) Lectures on Government and Binding. Dordrecht: Foris. Chomsky, N. (1986a) Knowledge of Language. New York: Praeger. Chomsky, N. (1986b) Barriers. Cambridge, Mass.: MIT Press. Chomsky, N. (1991) "Some Notes on Economy of Derivations and Representations." In R. Friedin, ed. Principles and Parameters in Comparative Grammar, 417-454. Cambridge, Mass.: MIT Press. Diesing, M. (1988) "Word Order and the Subject Position in Yiddish." In J. Blevins and J. Carter, eds. Proceedings ofNELS 18, 124-140. Amherst, Mass.: GSLA. Duarte, M.E.L. (1989) "A evolucao da ordem V(erbo) S(ujeito) para S(ujeito) V(erbo) em interrogativas QU- no portugues do Brasil." Ms. UNICAMP. Dupuis, F. (1988) "Prodrop dans les subordonnees en ancien francais." Revue quebecoise de linguistique theeorique et appliquee 7:41-62. Galves, C. (1990) "Two AGR in Portuguese." Ms. UNICAMP. Galves, C. (1991) "V-Movement, Levels of Reprensentation and the Structure of S." To appear in W. Chao and G. Horrocks, eds. Levels, Principles and Processes: The Structure of Grammatical Representations. Dordrecht: Foris. Haider, H. (1986) "V-second in German." In Haider and Prinzhorn, eds., 4975. Haider, H. and M. Prinzhorn, eds. (1986) Verb Second Phenomena in Germanic Languages. Dordrecht: Foris. Hirschbuhler, P. (1990) "La legitimation de la construction Vla sujet nul dans la prose et le vers en ancien francais." Revue quebecoise de linguistique theorique et appliquee 19:32-55.

138

THE DIACHRONY OF VERB SECOND

Hirschbuhler, P. and A. Rochette, eds. (1988) "Aspects de la syntaxe historique du francais." Revue quebecoise de linguistique theorique et appliquee 7, Quebec. Iatridou, S. and A. Kroch (1992) "The Licensing of CP-recursion and its Relevance to the Germanic Verb-Second Phenomenon." Working Papers in Scandinavian Syntax 50:1-24. Kato, M. and F. Tarallo (1986) "Restrictive VS Syntax in Brazilian Portuguese: Its Correlation with Invisible Clitics and Visible Subjects." Ms. Proceedings of the 39th Georgetown Rountable. Kayne, R. and J.Y. Pollock (1978) "Stylistic Inversion, Successive Cyclicity, and Move NP in French." Linguistic Inquiry 9:595-621. Koopman, H. (1984) The Syntax of Verbs: From Verb Movement Rules in the Km Languages to Universal Grammar, Dordrecht: Foris. Koopman, H. and D. Sportiche (1991) "The Position of Subjects." Lingua 85:211-258. Lightfoot, D.W. (1979) Principles of Diachronic Syntax. Cambridge: Cambridge University Press. Lightfoot, D.W. (1981) "A History of NP Movement." In C.L. Baker and J. McCarthy, eds. The Logical Problem of Language Acquistion, 86-120. Cambridge, Mass.: MIT Press. Lightfoot, D.W. (1988) "Syntactic Change." In F.J. Newmayer, ed. Linguistic Theory: Foundations. Linguistics: The Cambridge Survey /, 303-323. Cambridge: Cambridge University Press. Lightfoot, D.W. (1991) How to Set Parameters: Arguments from Language Change. Cambridge, Mass.: MIT Press. Maling, J. (1980) "Inversion in Embedded Clauses in Modern Icelandic." Islenskt mal og almenn mdlfroedi 2:175-193. Reprinted in J. Maling and A. Zaenen, eds. (1990) Modern Icelandic Syntax. Syntax and Semantics 24:71-91. San Diego: Academic Press. Mascaro, J. and M. Nespor, eds. (1990) Grammar in Progress: GLOW Essays for Henk van Riemsdijk. Dordrecht: Foris. Mattos e Silva, R.V. (1971) A mais antiga versao portuguesa dos "Quatro livros dos Dialogos de Sao Gregorio." Phd Dissertation, University of Sao Paulo. Mattos e Silva, R.V. (1989) Estruturas trecentistas: elementos para uma gramatica do portugues arcaico. Lisboa, Estudos gerais, Imprensa Nacional. Padua, M.P. de (1960) A ordem das palavras no portugues arcaico: frases de verbs transitivos. Coimbra: University of Coimbra Press. Platzack, C. (1986) "The Position of the Finite Verb in Swedish." In H. Haider and M. Prinzhorn, eds., 27-47. Platzack, C. (this volume) "The Loss of Verb Second in English and French." Pollock, J-Y. (1989) "Verb Movement, UG, and the Structure of IP." Linguistic Inquiry 20:365-424. Pontes, E. (1987) O Topico no Portugues do Brasil. Campinas: Pontes Editores. Rizzi, L. (1982) Issues in Italian Syntax. Dordrecht: Foris.

EVIDENCE FOR A VERB-SECOND PHASE IN OLD PORTUGUESE

139

Rizzi, L. (1986) "Null Objects in Italian and the Theory of pro." Linguistic Inquiry 17:501-557. Rizzi, L. (1988) "The New Comparative Syntax: Principles and Parameters of Universal Grammar." Ms. University of Geneva. Rizzi, L. (1990a) Relativized Minimality. Cambridge, Mass.: MIT Press. Rizzi, L. (1990b) "Speculations in Verb Second." In Mascaro and Nespor, eds., 375-385. Rizzi, L. (1991) "Residual Verb Second and the WH Criterion." Technical Reports in Formal and Computational Linguistics, 2. University of Geneva. Rizzi, L. and Roberts, I. (1989) "Complex Inversion in French." Probus 21:130. Roberts, I. (1985) "Agreement Parameters and the Development of English Modal Auxiliaries." Natural Language and Linguistic Theory 3:21-58. Roberts, I. (1989) "Case-assignment Parameters and the History of French Inversion." GLOW Newsletter, 22:55-57. Dordrecht: Foris. Roberts, I. (1992) Verbs and Diachronic Syntax. Dordrecht: Kluwer. Said Ali, M. (1957) Dificuldades da lingua portuguesa: estudos e observacoes. Rio de Janeiro: Livraria Academica. Said Ali, M. (1969) Gramatica secundaria da lingua portuguesa. Sao Paulo: Edicoes Melhoramentos. Salvi, G. (1990) "La sopravvivenza della legge di Wackernagel nei dialetti occidentali della Penisola Iberica." Medioevo Romanzo 15:177-210. Serafim Leite, S. I. (1942) Cartas dos primeiros Jesuitas do Brasil (153813). Comissao do IV Centenario da Cidade de Sao Paulo. Taraldsen, K.T. (1986) "On Verb Second and the Functional Content of Syntactic Categories." In Haider and Prinzhorn, eds., 7-26. Tarallo, F. (1990) Tempos linguisticos: itinerario historico da lingua portuguesa. Sao Paulo: Atica. Tarallo, F., ed. (1989) Fotografias sociolinguisticas. Campinas: Pontes Editora. Tarallo, F. and M. Kato (1989) "Harmonia trans-sistemica: variacao intra e inter-linguistica." Preedicao 8, UNICAMP. Tomaselli, A. (1990) "COMP° as Licensing Head: An Argument Based on Cliticization." In Mascaro and Nespor, eds., 433-445. Torrego, E. (1984) "On Inversion in Spanish and Some of Its Effects." Linguistic Inquiry 15:103-129. Travis, L. (1984) Parameters and Effects of Word Order Variation. PhD Dissertation, MIT. Vance, B.S. (1988) "L'evolution de pro-drop en francais medieval." In Hirschbuhler and Rochette, eds., 8-112. Vance, B.S. (1989) Null Subjects and Syntactic Change in Medieval French. PhD Dissertation, Cornell University.

6 Indo-European Origins of Germanic Syntax Paul Kiparsky Stanford University

1. Introduction In accord with the standard analysis I assume that verb-second and verb-first order in Germanic1 is derived by movement of the finite verb to a.vacant C° position (den Besten (1983)).2 I shall argue that in English and in the early Germanic languages the Specifier of CP is a focus position, which hosts whphrases, demonstratives, and negation. This specifier position can in turn be preceded by a topicalized/left dislocated element adjoined to CP; if it is an NP, it must bind a resumptive pronoun in the clause. In addition, constituents can be topicalized by adjunction to S.

140

INDO-EUROPEAN ORIGINS OF GERMANIC SYNTAX

141

In SECTION 2,1 present an analysis of fronting phenomena in Old English, Old High German, and Old Icelandic in support of this claim, and in SECTION 3,1 then argue that the Germanic system developed from the minimally different system of Indo-European in consequence of a more fundamental syntactic innovation. The historical development which I propose is as follows. On the evidence of Vedic, Greek, and Hittite, the Indo-European proto-language had two leftperipheral operator positions corresponding to those in (1) (Hale (1987, 1989); Garrett (1990)). However, it lacked the category of complementizer and had no syntantically embedded sentences. Finite subordinate clauses, including relative clauses and sentential complements, were syntactically adjoined to the main clause, exhibiting "main-clause properties," such as topicalization of constituents to clause-initial position. In most daughter languages, including those of the Germanic family, subordinate clauses became syntactically embedded, taking up argument or modifier positions within the main clause, losing their main-clause properties and becoming headed by C°, which in Germanic was filled by one of a set of new indeclinable complementizers. The introduction of complementizers is a consequence of the shift from adjoined to embedded subordination on the assumption that only CPs can function as sentential arguments and modifiers (Kayne (1982); Taraldsen (1986); and Holmberg (1986)). This restructuring in turn led to several major syntactic characteristics of Germanic, including the rise of V-to-C° movement, triggered by whphrases and other focused elements in Spec-C. "Residual verb second" (Rizzi (1990)) is then the original core of the verb-second system. The modern Germanic verb-second languages have extended the Spec-C position to Topics, and consequently generalized V-to-C° movement. From this comparative perspective Old English preserves the earliest Germanic syntax with remarkable fidelity. Its archaic traits include V-final main clauses, different landing sites for topicalization and wh-movement, residual verb second, the original system of pronominal cliticization, and a version of the Indo-European relative clause system.

2. Germanic 2.1 V-to-C° movement That V-to-C0 movement is obligatory is uncontroversial, though the explanation for this is contested.3 Old English is important in this regard because, alongside verb-second main clauses, it has main clauses in which the finite verb remains inside S, including those having verb final.4 Instead of assuming that V-to-C° movement is optional in Old English, I will argue that the category C itself is optional, where no principle of grammar requires its presence. On these assumptions, the word order patterns are derived in outline as follows. C° is obligatory in subordinate clauses, because they are in argument and modifier position and sentences must be turned by complementizers into CPs in order to function as arguments and modifiers. C° is also obligatory in

142

THE DIACHRONY OF VERB SECOND

sentences with a fronted wh-phrase or other focused constituent: I assume because, for scopal reasons, these elements must be in the Specifier of CP, and that CP, like every Xmax must have a head. Where C° is not required for these or other reasons, its presence or absence is fixed on a language-specific basis. In German it is obligatory, in Old English optional, and in Modern English prohibited — a parametric difference reflected both in the range of permissible main-clause word orders of these respective languages and in whether overt complementizers can head declarative main clauses. In subordinate clauses, the obligatory C° position is usually filled by a lexical complementizer, blocking V-to-C° movement.5 Subordinate clauses lacking a lexical complementizer must, in consequence of the above considerations, undergo V-to-C° movement; this yields the obligatory verb-first pattern seen in bare concessive and conditional clauses, such as (2): (2)

Were se mon on swelcum lande swelce he were be hi ahte, oonne were his wela & his weoroscipe mid him (Boethius, ConsPhil 63.21) 'Whatever land the man who possessed them might be in, his wealth and his dignity would be with him.' (lit. 'were the man in whatever land...')-

The C° position of CP main clauses also may or may not have a lexical complementizer. CP main clauses with vacant C° yield, through V to C° movement, the standard verb-second word order type. I will assume that CP main clauses with lexical complementizers in Old English are instantiated by yes/ no questions introduced by hweoer; these clauses have subordinate clause word order, not verb-second: (3)

a. Hweoer ou nu swelces auht wyrcan mege? (Boethius, ConsPhil 29.22) 'Now can you do anything like that?' b. Hwether he wolde bam forcuoestum monnum folgian? (ibid, 37.15) 'Would it follow the most wicked men?'

However, nothing forces main clauses to have a complementizer, and in Old English they usually do not. I will assume that main clauses without complementizers are simply Ss, rather than CPs with empty heads and specifiers. Combined with the option of filling the Spec-C position, we then have three possible structures for sentences without lexical complementizers: (4) a. cp[XP c [ c [V] s [ . . . ] ] ] b. c p [ c ' [ c [ V ] s [ . . . ] ] ] c. S [ . . . V . . . ] The first two structures are available in all the old Germanic languages. Verbsecond clauses like (4a) are the standard main-clause type. Verb-first clauses like (4b) are obligatory in standard yes/no questions, and they also occur in declarative main clauses in Old English, Old High German, and Old Icelandic:6

INDO-EUROPEAN ORIGINS OF GERMANIC SYNTAX

143

(5) Hefde se cyning his fierd on tu tonumen (Anglo-Saxon Chronicle, A.D. 893) 'The king had divided his army in two' Uuarun tho hirta in thero lantskeffi (Tatian 6) 'At that time there were shepherds in the area' ferr pa Vagn heim suor til Danmerkr (Heimskringla 160.29) 'Then Vagn went home southwards to Denmark'. Verb-final main clauses instantiate the bare S structure of the form (4c):7 (6) He pa his here on tu todelde (Orosius 116.16) 'He then divided his army in two' Her haebne men aerest on Sceapige ofer winter saetun (Anglo-Saxon Chronicle, A.D. 855) 'Here (in this year) heathen men first encamped in S. over the winter'. They are mostly found in Old English.8 I take the absence of sentences like (6) in Old High German and Old Icelandic to mean that C° is syntactically obligatory even in main clauses in these languages, a conclusion for which further support will be given in SECTION 2.4 below. In sum, V-to-C° movement can be considered obligatory for all the languages, and its apparent optionality in Old English is attributable to the parametrically specified optionality of C° itself.

2.2 Topicalization and Wh-Movement Modern Germanic languages are generally assumed to have topicalization and wh-movement to the same landing site. In early Germanic, however, there is evidence that Topics are positioned to the left of wh-elements. This is also arguably the case in Modern English.9 Topicalization in interrogative and negative clauses with Subject-Aux Inversion provides direct evidence for the "Topic-wh" order postulated here. English shows the predicted order of constituents: (7)

a. Tomorrow, where shall we go? b. During all the time John was in Greece not once did he drink ouzo.

Further, the pattern of data in (8) is predicted on the assumption that NP Topics must bind a correlative resumptive pronoun in the clause and that whphrases and other elements triggering Subject-Aux Inversion move into Spec-C: (8)

a. Beans I like. b. Beans, I like them. c. *Beans who needs?

(focused NP in Spec-C) (Topic adjoined to CP) (focusing blocked by wh-phrase in Spec-C)

144

THE DIACHRONY OF VERB SECOND

d. Beans, who needs them? e. *Beans not once did I eat

(Topic adjoined to CP) (focusing blocked by NEG-phrase in Spec-C) f. Beans, not once did I eat them. (Topic adjoined to CP) Assuming movement in English allows preposition stranding, the contrast in (9) indicates that the adjoined Topics are generated in situ: (9)

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

*This shelf, what should we put on? This shelf, what should we put on it? On this shelf, what should we put? This room, never once did John sleep in. This room, never once did John sleep in it. In this room, never once did John sleep.

Spec-C is filled by the wh-element in (9a-c) and bv the negation in (9d-f). So the Topics must all be CP-adjoined and the data follow from the previously stated requirement that NP Topics must be resumed by a pronoun in the argument position. In the modern Germanic verb-second languages, such as German and Swedish, the Spec-C position itself has become capable of hosting topicalized as well as focused constituents. In consequence, the CP adjunction position (the German "Vorvorfeld") has been relegated to a more marginal role in their syntax. Moreover, in contexts which force its use such as wh-questions those languages generally require pronominal resumption even of adverbials, so the equivalents of (7a), (9c,f) are ungrammatical.10 In these languages, then, SpecC has become the preferred site of every type of fronting. It has generally been assumed in the literature that German and Swedish represent the original situation and that English has innovated. I should like to argue that the opposite is the case. The Modern English system has clear antecedents in Old English and beyond. The availability of a CP-adjoined position for Topics as distinct from SpecC position for wh/focus elements is also demonstrable in Old English. First, elements in Spec-C may be preceded by CP-adjoined material; in (10), for example, hu 'how, whether' is preceded by the if-clause in (51): (10) Gif hwa nu bio mid hwelcum welum geweorbod & mid hwelcum deorwyrbum aehtum gegyrewod, hu ne belimpp se weoroscipe ponne to bam pe hine geweoroao? (Boethius, ConsPhil 96.12) 'Now if anyone is endowed with all riches and adorned with all valuable possessions, does not the glory then belong to him who adorned him?' V-to-C° movement in Old English main clauses also diagnoses the difference in Topic and wh-landing sites. The key factor is that there are verb-final declarative sentences, such as (6), but wh-questions always have verb-second order (Allen (1977:48 ff.)). This asymmetry is explained directly by our assumptions. Since C° is optional unless required for the satisfaction of some principle of grammar, declarative main clauses may or may not be CPs, hence

INDO-EUROPEAN ORIGINS OF GERMANIC SYNTAX

145

undergo verb second or not as already discussed. But if wh-phrases must reside in Spec-C, sentences containing them must be CPs, therefore must contain a C°, to which the finite Verb must move. Hence wh-phrases induce verb second.11 On the assumption that topicalization and wh-movement in Old English have the same landing site, as assumed in some analyses on the analogy of Dutch and German (e.g. Kemenade (1987)), these facts are mysterious. From a modern English perspective they are not surprising since they persist to this day in the conditions on Subject-Aux Inversion ("residual verb second"):12 (11) a. When will John arrive from London? b. Tomorrow will John arrive from London. c. Tomorrow John will arrive from London. Maling and Zaenen (1981) noted that in Old Icelandic subordinate clauses, topicalization but not wh-movement triggers verb second. This difference between Topic and wh follows directly if we suppose that the complementizer in embedded wh-clauses is lexically filled,13 and treat subordinate clauses with topicalization as embedded root sentences with an external complementizer.14 On the assumption that adjunction is allowed to maximal projections we would expect topicalization by adjunction to S as well. In Old English, topicalization by adjunction to S must be assumed in main clauses with initial XP and no verb second (12a,b) and in "embedded root clauses" (12c,d):15 (12) a. & him ba Ioseph, rihtwis man, mid godcunde fultume gehealp (Orosius 32.26) 'And him then Joseph, a righteous man, helped with holy aid.' b. Swelcum monnum Dryhten cidde burn bone witgan (CP 27.12) 'Such men the Lord reproached through the prophet.' c. Ic secge past behefe ic eom ge cingce & ealdormannum (AEColl. 150) 'I say that useful I am to the king and the chiefs.' d. Be oam is awriten oaet betera beo se geo1dlega wer oonne se gielpna (CP 217.10) 'Therefore it is written that better is the patient man than the boastful (one).'

2.3 The Distribution of Clitic Pronouns A third argument that wh-movement and topicalization have different landing sites in Old English is based on the position of clitic pronouns, as described for Old English by Allen (1977:49), Mitchell (1985 §3907) and Kemenade (1987, ch.4); a similar situation seems to obtain in archaic Old High German (Behaghel (1932:14)). In these languages, if a wh-constituent is fronted, clitic subject pronouns are placed directly after the verb:

146

THE DIACHRONY OF VERB SECOND

(13) a. Hu begaest pu weolc pin? (AEColl. 23) 'How do you go about your work?' b. uuanan uueiz ih thaz? (Tatian, 2) 'Whence do I know that?' In relative clauses, subject pronouns follow the whole demonstrative + complementizer complex: (14) Ac gif we asmeagao oa eadmodlican daeda oa oe he worhte, ponne... 'But if we consider the humble deeds which that he wrought, then...' But if the sentence begins with a fronted Topic, clitic subject pronouns come directly after it: (15) a. Hefonas he purhfor mid his modes sceawinge (CP 99.23) 'Heaven he traversed with his mind's vision.' b. & ealle pa o5re peoda pe on Crecum waeron he to gafolgieldum gedyde (Orosius 124.6) 'And all the other nations that were among the Greeks he made into tributaries.' c. erino portun ih firchnussu (Isidor 6.2) 'I crash the iron gates.' I assume the theory of cliticization elaborated by Halpern (1992) according to which the placement of clitics arises from the interaction of independent syntactic and phonological constraints. Clitics are syntactically associated with edges of maximal projections, and phonologically required to lean to their left or right on a prosodic constituent whose size is parametrically specifiable as a phonological word or a phonological phrase. When necessary for the satisfaction of the phonological requirement, the clitic is moved by "prosodic inversion" at PF to a position after the first prosodic constituent of its syntactic host (or, if it is a proclitic, before the last). The pattern of cliticization illustrated in (13), (14) and (15) is readily accounted for in accord with Halpern's theory on the assumption that Old English subject clitics are syntactically Sinitial, and must lean leftwards on a phonological phrase. When S is preceded by other material, this phonological requirement is satisfied in situ, and they consequently appear in their basic position at the left edge of S, as in (13, 14). When S is not preceded by any other material, the phonological requirement coerces them into a position after the first phonological phrase of S, as in (15). Schematically: (16) a. cp[XP c .[ c [V] s [clitic . . . ] ] ] b. CP [ C'[c[V]s[clitic . . . ] ] ] c. s[ XP clitic ...} There is independent evidence that the pronouns in (14-15) are positioned by cliticization rather than by syntactic movement: as a syntactic movement rule, it would be unique in Old English in allowing preposition stranding.

INDO-EUROPEAN ORIGINS OF GERMANIC SYNTAX

147

Otherwise pied-piping is obligatory in Old English (Allen (1977:53)).16 But if pronouns are fronted by cliticization. they are moved at PF, and preposition stranding is expected.17 Independent confirmation comes from the behavior of certain adverbs in Old English, particularly pa, ponne 'then' and ne 'not'. Their syntax shows that, unlike most adverbial expressions. which are topicalized by adjunction, they are fronted into Spec-C. When fronted, these adverbs pattern exactly like wh-expressions, both in inducing V-to-C° movement and in not attracting clitic subjects.18 That is in the presence of fronted pa, ponne, and ne, the finite verb is obligatorily moved to C°, as in wh-questions.19 And in exactly those cases the finite verb cannot be separated from the fronted constituent by a subject clitic.20 (17) a. pa ondwyrdon hie him tweolice (Orosius 156.2-3) 'Then they answered him doubtingly' b. Ne maege we awritan ne mid wordum asecgan ealle pa wundra (AELS21.242) 'We can neither write nor express with words all those wonders.' Placement of the finite verbs ondwyrdon, maege in final position in (17) would be ungrammatical (contrast (6)) as would second position of the pronouns hie, we (contrast (15)). On the analysis of cliticization proposed above, both facts follow from the assumption that pa, ponne, and ne must be fronted into SpecC and that V-to-C° is obligatory. Our analysis is again confirmed by the fact that the focused element in SpecC can be preceded by an adjoined phrasal or sentential Topic, such as the though-clause (18): (18) a. peah be we pas ping cwepe, ne tellao we synne weosan gesinscipe (Bede 1.18, p. 82) 'Though we say these things, we do not count wedlock as sin.' b. peah se laerow ois eall smealice & openlice gecyde, ne forstent it himnoht(CP 163.18) 'Though the teacher tells all this carefully and openly, it avails him nought.'

2.4 The Word Order of Conjoined Sentences In Old English, the non-initial conjuncts in a series of coordinate sentences, introduced by such conjunctions as ond, ac 'and', oppe 'or' often have a different verb position than the first conjunct, (Bacquet (1962, ch.III); Mitchell (1985, §1719-1731, 1869, 3934)). They may either have the verb-final word order characteristic of subordinate clauses, as in (19):

148

THE DIACHRONY OF VERB SECOND

(19) a. pa was domne Leo papa, on Rome: ond he hine to cyninge gehalgode, ond hiene him to biscepsuna nam (Anglo-Saxon Chronicle 853 A.D.) 'Then was lord Leo pope in Rome, and he consecrated him king, and adopted him as his godson.' b. Her for se here from Lindnesse to Hreopedune, ond paer wintersetl nam (Anglo-Saxon Chronicle 874 A.D.) 'Here (this year) the army went from L. to H., and took up winter quarters there' or else—an equally surprising breach of parallelism in coordinate structures —they may have verb-initial order: (20) a. Her Ecgbryht cyning forpferde,—-ond feng AEpelwulf Ecgbrehting to Wesseaxna rice (Anglo-Saxon Chronicle 836 A.D.) 'Here (in this year) king E. died, and A. E. came to the Wessex throne' b. Her waes ofslaegen Osric, ...and feng Ceolwulf to bam rice (ibid.) Here O. was killed, and C. came to the throne.' Of course, first conjuncts, like any other main clauses, occasionally have verbfinal and verb-initial order too—recall the discussion of sentences like (6). An example of such a coordinated structure is (20b). But there still is a special tendency of non-initial conjuncts to have those orders, which needs to be explained. And secondly, conjoined yes/no questions and imperative sentences, which are regularly verb-initial, can likewise revert to verb-final order under the same conditions (Mitchell §905, §1871). (21) Eart pu se oe toweard is, obpe we o5res andbidian sceolon (AECHom i.480,6) 'Are you the one who is to come, or shall we wait for another? Lufian we hine ny & his noman myclian (BlHom 13.6) 'let us now praise him and magnify his name.' The occurrence of verb-final word order in non-initial conjuncts is sometimes taken to mean that those conjuncts are subordinate clauses in some sense—as if, mysteriously, "even co-ordinating conjunctions are subordinating," as Campbell (apud Mitchell) puts it. This cannot be right, for several reasons. First, it is only their word order that is peculiar; in every other syntactic and semantic respect they are like main clauses and unlike subordinate clauses. For example, Old English allows certain long-distance wh-dependencies between elements in main and subordinate clauses (Allen (1977:79)), but it does not allow such dependencies into a sentential conjunct, even one with verb-final word order. Secondly, not all conjoined clauses have subordinateclause order: under certain well-defined conditions they must have main-clause word order and in all other cases they may have it, while true subordinate clauses never do.21 And thirdly, this view of things leaves out of the picture

INDO-EUROPEAN ORIGINS OF GERMANIC SYNTAX

149

the fact that non-intial conjuncts have another special word order possibility, namely verb-initial word order, as in (20). Given the phrase structure we proposed for Germanic, the entire range of coordinate-clause word orders can be derived from the assumption that SpecC and C° may be omitted in non-initial conjuncts. This approach gains support if we examine the conditions under which conjoined main clauses must have verb-second word order. Sentences conjoined with ond or oppe have obligatory verb-second order if they begin with one of the following things: (a) a wh-phrase, (b) ba, bonne 'then' or (c) ne 'not' (Mitchell (1985, §1723 ff. 1753, 1870)): (22) a. Hu mceg la se blinda laedan pone blindan? (ALHom 13.124) 'O how can the blind lead the blind?' b. ...andba wearS he oferswiSed (&CHom i. 176.15) 'and then he was overcome' c. ...and ne mceg nan bing his willan wi5standan (JECHom i.10.1) 'and nothing can withstand his will.' But these are the very constituents which, as we concluded in the preceding section, are fronted to Spec-C, for they always induce V-to-C° movement, and fail to attract pronominal subjects. So we can reduce these special cases to the previously established generalization that sentences undergo V-to-C° movement if and only if they begin with a lexically unfilled complementizer position. Since the second conjuncts in (22) have overt wh-elements, they necessarily have a C° head, and therefore show obligatory verb-second position. Now we also have the explanation for why the other Germanic verb-second languages do not allow verb-final word order in conjoined main clauses. In those languages C° is obligatory; it is only in Old English that it is optional. We drew this conclusion from other evidence in SECTION 2.1, on the basis of the fact that Old English is the only language which allows the verb to remain in situ in main clauses, as in (6). The obligatoriness of C° in the other Germanic languages would make the structure (4c) unavailable in them, explaining why they do not have conjoined sentences with verb-final word order as Old English does. Since the other structures, namely (4a,b), are available in main clauses in the other Germanic languages, non-initial conjuncts with verb-initial order of the type (20) should occur in them. And in fact they are well attested in both Old High German and Old Icelandic (Behaghel (1932. §1447, ff.); Heusler (1962:175); Nygard (1905:347)): (23) End fuorun flz sine scalcha in dea uuega" enti kasamn6tun all so huuelthhe so sie funtun, ..., enti uuarth arfullit des bruthlauftes kastuoli (Monsee-Vienna Fragments, Matth. 22.10) 'And his servants went out into the streets and gathered together whoever they found, and the wedding table became filled.'

150

THE DIACHRONY OF VERB SECOND

(24) Magnus konungr f6r um haustit alt austr til landsenda, ok var hann til konungs tekinn um alt land (Heimskringla 518, 34) 'King M. rode in the fall to the easternmost end of the land and he was chosen king over the whole country.' One fact which this anaysis leaves unexplained, however, is that it does not account for the fact that verb-final yes/no questions and commands such as (21) essentially occur only in non-initial conjuncts. In any case, the evidence from coordinate sentences in the older Germanic languages supports the special positional status of wh-elements and of pa, ponne, and ne.

2.5 Relative Clauses Each of the Germanic languages developed one or more indeclinable complementizers, of diverse provenience but with similar syntax (Gothic ei, Scandinavian es/er, en, sem/son OHG, OS the, OE pe, paet). They can in general, head both finite complement clauses and relative clauses. The relative clauses headed by these complementizers are of two types. The first type, which is attested in all old Germanic languages has a bare complementizer, with either a gap or a resumptive pronoun in the clause-internal argument position. In the second type of relative clause, a determiner is moved out of the relative clause and adjoined to the position immediately before the complementizer, resulting in such combinations as Gothic sa+ei (OE se pe), which function much like the ordinary relative pronouns of the modern Germanic languages. This construction is probably of later date, and it is in any case not attested in Old Icelandic or Old High German. But as we shall see, the basis for it lies in the Germanic constituent structure (1).22 The two types of relative clauses differ substantially, in ways which suggest that relative clauses headed by the bare complementizer are not derived by wh-movement, while those with a demonstrative pronoun plus complementizer complex are derived by wh-movement, as shown by Allen (1977) for Old English, and by Maling (1976) for Old Icelandic.23 The most telling differences between the two types of relative clauses are the following:24 Preposition stranding is obligatory in bare-complementizer relative clauses, but does not occur in pronominal relative clauses (Mitchell (1965, §22312248)). This is especially significant in view of the status of preposition stranding as a diagnostic for movement: (25) ...paem burgumpe he on geworhte his wundra 'the cities that he wrought his miracles in' paer barsmiper, es landet eyddesk af 'that kind of fights, that the land would be devastated by.' Subjacency violations occur in bare-complementizer relative clauses, but not in pronominal relative clauses:

INDO-EUROPEAN ORIGINS OF GERMANIC SYNTAX

151

(26) Ac for paem he geneode swioost over pone munt oe he wiste baet Flamineus se consul wiende paet he buton sorge mehte on baem wintersetle gewunian (Orosius 188.30) 'but because he ventured most quickly over the mountain that he knew the Flamineus the consul thought that he might dwell on in winter quarters without care.' Left-branch constraint violations (extraction of genitive modifiers from NPs) occur in bare-complementizer relative clauses, but not in pronominal relative clauses: (27) se god pe bis his beacen waes 'the God that this was his sign' sja maor, er ver segjum nu fra jartegnum, atti marga laerisveina 'the man, that we now tell about the signs had many disciples.' NPs bearing different cases in conjoined relative clauses may be relativized by a single bare complementizer, but not by a single pronoun + complementizer complex: (28) bam freondum pe ic lufige and me lufiao 'to the friends that I love and love me' (Acc. and Nom.) pau Qtto son, er Haraldr konungr ios vatne ok gaf nafn sitt 'they had a son, that king Harald baptized and gave his name' (Acc. and Dat.). That-trace constraint violations occur in bare-complementizer relative clauses but not (or very rarely) in pronominal relative clauses: (29) ne lufige ge oisne middangeard oe ge geseoo oaet lange wunian ne rnaeg (Alc. Th.XL. p. 614) 'do not love this world that you see (that) cannot last long.' Relative clauses with a demonstrative pronoun adjoined tope behave much as the relative clauses of modern English, German, French, etc.25 But, as Allen and Maling (1976) observe, the properties of the other type of relative clause suggest that there is no wh-movement involved. To account for it they propose a rule which deletes the relativized item in situ under identity to the head of a relative clause. Allen considers and rejects the possibility of deriving them by "free deletion" of a pronoun in the relative clause. However, it is possible that the "gaps" are really null resumptive pronouns and that there is no controlled deletion rule.26 One reason why this is an attractive possibility is that overt resumptive pronouns are common in bare-pronoun relative clauses, as in (30), but rare in pronominal relative clauses. For a proposal along these lines, see Kemenade (1987:164). (30) se god be bis his beacen waes 'the God that this was his sign' sa er skilgetenn er faper hans 'he that his father is born in wedlock.'

152

THE DIACHRONY OF VERB SECOND

2.6 The Position of the Finite V in S Typologically oriented discussions of word order change have shed confusion by identifying the rise of V-to-C° with the shift of the V from VP-final to VP-initial position, under the heading of "SOV to SVO." All the Germanic languages, except Gothic, have V-to-C° movement (or at least processes historically related to it) in main clauses, but in Dutch and German the verb is final in the VP, while in the Scandinavian languages, as well as English since ca. 1200, the verb is initial in the VP. The preceding discussion has assumed that the underlying word order of Old English was verb-final, for the reasons laid out in Kemenade (1987) and in earlier works. That this was also the basic Proto-Germanic word order has been accepted by a majority of scholars since Delbruck (1878). Sentences that apparently contradict the assumption of underlying verb-final word order abound on every page of Old English. The reason is that V-toC° movement in main clauses is not the only way in which the underlying verb-final word order can be perturbed. Two important stylistic movement rules in particular have been established in previous work. The first is a rule of Extraposition which moves Noun Phrases and Prepositional Phrases into sentence-final position. Pintzuk and Kroch (1989) argue from metrical evidence in Beowulf that this rule is to be identified specifically with Heavy NP Shift. However, in Old English prose even light NP objects can follow the verb. Pintzuk (1991) argues that Old English at that stage has acquired two competing basic word orders, reflecting the coexistence of left-and right-headed VP and IP. In addition, Kemenade makes a good case that Old English, like Modern German, possesses a "Verb-raising" rule whose effect is to move finite verbs to the left of some subconstituent of the VP. Unlike V-to-C° movement, Verb raising applies to subordinate clauses and can never move the verb to the left of the subject.27 Let us lay out systematically the additional word order possibilities derived in this way. We abstract away from the word order variation introduced by the previously discussed processes of topicalization, wh-movement and cliticization, and consider only CPs with the Subject in initial position. Taking the possible permutations ol the Object, a finite Modal and the infinitive Verb, we predict for main clauses the orders in (31): (31) a. S VfO V b. S V f V i O

and for subordinate clauses: (32) a. b. c. d. e.

S O Vi M S Vi Vf O SOVfV S Vf O V SVfViO

with only *S Vi O Vf ruled out as ungrammatical.

INDO-EUROPEAN ORIGINS OF GERMANIC SYNTAX

153

This agrees exactly with the description of the word orders in main and subordinate clauses in Old English that is presented in Allen (1974). The derivations are obvious. The underlying order is S O Vinf Vf. Of the two mainclause orders, the first arises directly from V-to-C°, the second by the additional application of Heavy NP Shift, or (if we adopt Pintzuk's proposal) from the head-initial base structure. No further rules are applicable. In subordinate clauses, V-to-C° movement is inapplicable and the underlying order is realized directly as (32a). The type (32b) is the crucial case demonstrating that Heavy NP Shift must be assumed in any case, even if head-initial VP/IP is postulated. Verb raising yields (32c) and (32d), and, compounded with Heavy NP Shift (or underlying VO order), (32e). *S ViO Vf is not derivable.

3. From Indo-European to Germanic 3.1 The Constituent Structure of Indo-European In the preceding sections I have provided evidence for the specific phrase structure shown in (1), based on verb movement, cliticization, coordination, and relative clause formation in Old English, Old Icelandic, and Old High German. On the basis of the testimony of these languages, it can be assumed for Proto-Germanic as well. This is an interesting result because our reconstructed Germanic phrase structure differs minimally from that established for Indo-European on the basis of Vedic Sanskrit, Hittite, and Greek evidence (Hale (1987, 1989)). In the remainder of this paper I will attempt to relate the Germanic system historically to the Indo-European one, arguing that the Germanic innovations result from the rise of embedded finite clauses. As in Germanic, Indo-European had two left-peripheral operator positions. As in Germanic, the inner of these positions hosted focal elements, in particular wh-phrases (relative and interrogative) and other focused elements, such as demonstratives (Hock (1989)), and the outer Topic position adjoined to the maximal projection, binds a (possibly null) resumptive pronoun in the argument position (Garrett (1992)). The key difference is that there were no complementizers, and therefore no CP, and no embedding. The basic phrase structure of Indo-European, then, was along the lines of (33):

154

THE DIACHRONY OF VERB SECOND

Hale uses primarily two pieces of evidence to establish that Vedic had such a constituent structure. The first is that exactly one constituent can appear in front of wh-elements, both in main clause questions and in relative clauses: (34) ratham ko m'r avart ayat (RV. 10.135.5) chariot-ACC who down rolled 'Who rolled out the chariot? (35) sahasras'rngo vrsabho yah samudrad ud acarat, tena thousand-horned bull which sea-ABL out rose that-INSTR sahasyena vayam ni janan svapavamasi (RV. 7.55.7) mighty-INSTR we in people-ACC put to sleep 'The thousand-horned bull that rose from the sea, with that mighty one we put the people to sleep.' The second argument, from the placement of clitics, reveals the constituency in a still more interesting way. Hale shows that clitics (all of which are enclitics) are of two distinct types. The first type comprises emphatic clitics, and conjunctions such as va, gha. The second type comprises pronouns. Emphatics and conjunctions are placed after the first constituent of the entire sentence, and clitic pronouns are placed at the beginning of S, unless there is no fronted element, in which case they are placed after the first word of S. See the relative clauses in (36), with va 'or' and te 'to you' positioned respectively before and after the relative pronoun in Focus position: (36) a. s"[ urau va va s'[ ye s[ antarikse madanti ] ] ] (RV. 3.6.8) wide-LOC or who atmosphere-LOC rejoice-3PL 'or who rejoice in the wide atmosphere' b. s"[ idhmam s"[ yas s[ te jabharac kindling-ACC' who you-DAT carry=PerfSubj chasramana'h ] ] ] (RV. 4.12.2) exerting himself 'who, exerting himself, bears the kindling to you." Halpern's (1992) theory of cliticization accounts for the two types of clitics in a simple way. Emphatics and conjunctions belonging to the functional system, are syntactically affiliated with S", and pronominal clitics being arguments, are syntactically affiliated with S. Both require a prosodic host on their left, which is supplied through prosodic inversion if the base position does not provide it. Hale (1989) shows that the Hittite system is strikingly similar. In Hittite, relative clauses whose head is definite, allow the topicalization of one constituent out of the clause into the pre-wh slot. The proposed constituent can be either the head, as in (37a), or some other constituent, as in (37b): (37) a. nu DUMU-an kuin hukkiskimi PRT child-ACC which-ACC treat-magically-lSg 'the child which I treat magically'

INDO-EUROPEAN ORIGINS OF GERMANIC SYNTAX

b.

155

nuza ANA DINGIR.MES kuit arkuwar PRT-REFL to gods which-ACC prayer-ACC iyami make-lSg 'the prayer I make to the gods.'

Thus Hittite appears to have the same phase structure organization as Vedic Sanskrit, but with somewhat different conditions on topicalization.28 Moreover, Hale argues that Homeric Greek clitics are positioned by the same principles, though Greek has already developed an extensive complementizer svstem. This converging evidence justifies reconstructing the system for the proto-language. Hale's demonstration that Topic and wh-elements occupied separate positions in Indo-European meshes with the Germanic evidence presented in SEC TIONS 2.1-2.5. It remains to be explained, however, how (33) was elaborated into the full-fledged CP structure (1) in Germanic. I attempt this in the remaining part of this paper.

3.2 Adjoined Subordinate Clauses A second major characteristic of Indo-European syntax, best preserved in Sanskrit, Hittite, and Old Latin, was that finite subordinate clauses were not embedded but adjoined (Haudry (1973); Watkins (1976); Lehmann (1980); Holland (1984); Calboli (1987); Hettrich (1988); Hock (1989)). By this I mean that subordinate clauses were not internal constituents of sentences in argument or modifier positions, at any level of structure but were rather positioned at their right or left periphery (though there was a rich system of nominalized forms with infinitival and participial function which were deployed in argument and modifier position). Relative clauses could, but did not necessarily have to be, anaphorically liked as adjuncts to a correlative phrase in the main clause, consisting of a demonstrative pronoun and sometimes other lexical material, this phrase constituting the actual argument or modifier of the main clause. So-called "inverse attraction" is simply the result of omitting the correlative. In this respect, Indo-European sentence structure resembled that of Finno-Ugric and many other languages; see in particular Hale (1976) for an analysis of adjoined relative clauses in Warlpiri. There are several reasons for assuming that adjoined clauses in Indo-European were generated in peripheral position, and not simply obligatorily extraposed from sentence-internal argument or modifier positions. First, with certain clause types, such as relative clauses, a correlative pronoun appears obligatorily (albeit subject to pro-drop29) in the argument/modifier position of the main clause. The obligatoriness of a correlative would be unexplained on the extraposition hypotheses, but it is predicted on the assumption that the correlative is in fact the real bearer of the grammatical role.

156

THE DIACHRONY OF VERB SECOND

Secondly, extraposition cannot easily account for cases when the "head" is contained in the relative clause itself, which in fact is statistically the most frequent case for Vedic and is common in the other languages that preserve adjoined subordinate clauses:30 (38) yam yajnam adhvaram vis'va'tah which-ACC sacrifice-ACC service-ACC everywhere-from paribhur asi, sa id devesu gachati (RV. 1.1.4) surrounding are-2SG, that FOCUS gods-LOC go-3Sg 'It is the sacrifice (and) service that you surround on all sides which reaches the gods,' (39) ostiumquod in angiporto est horti, patefeci fores gate which in alley-ABL is garden's open-PerflSg doors-ACC (Plautus. Most. 1046) 'The garden gate which is in the alley I have opened (its) doors.' "Lowering" of the head into an embedded relative clause is ruled out on general grounds, so under the extraposition analysis some kind of reanalysis of the head into the relative clause would have to be assumed for cases like (38). On the adjunction account they are directly generated. Thirdly, there can be no question of extraposition when both clauses have separate lexical heads: (40) a. sa gha viro na risyati, yam mdro that EMPH man not perish-3Sg, which-ACC Indra. brahmanaspatih somo hin6ti martyam (RV. 1.18.4) Brahmanaspati, Soma support-3Sg mortal-ACC' 'The man who does not perish is the mortal whom Indra Brahmanaspati Soma supports.' b. yo martyah sisite aty aktubhir, ma which mortal sharpen-Mid-Sg overly nights-INSTR, not nah sa ripur TSata (RV. 1.36.16) us-GEN that trickster dominate-Subj3Sg 'As for the mortal who makes himself too sharp by night, may that trickster not gain power over us' (41) nuza ANA DINGIR.MES kuit arkuwar iyami PRT-REFLto gods which-ACC prayer-ACC make-lSg, nukan AWATEMES ANA DINGIR.MES anda Sunni (KUBVl 46 11) now

words

to

gods

in

fill'

'As for the prayer I make to the gods, report the words to the gods.' and when the lexical content of the "head" NP is distributed over both clauses:

INDO-EUROPEAN ORIGINS OF GERMANIC SYNTAX

157

(42) ya nah piarad asVina jyotismati which-NOM us-DAT carry-Inj3Sg Asvins-DuVoc bright-Norn tamas tirah, tam asme rasatham darkness-ACC across, that-ACC us-DAT grant-AorSubjMid2Du isam (RV. 1.46.6) refreshment-ACC "O you two ASvins, grant us that bright refreshment, and may it carry us across the darkness!" Finally, that the adjoined clauses must be generated in any case independently of their function in relative/correlative structures is shown by the fact that they can appear even where there is no position in the main clause to extrapose from. In particular, adjoined clauses formally identical to relative clauses may appear unlinked to any argument/modifier position in the main clause, in which case they effectively function as if-clauses: (43) yasyanaksa duhita jatv asa, whose-GEN eyeless-NOM daughter-NOM from birth who-NOM kas tarn vidvam abhi manyate her-ACC knowing-Norn to think-Subj3Sg blind-ACC andham (RV. 10.27.11) 'If someone's daughter is eyeless from birth, who would, knowing her, desire the blind one?' (44) kuis eshar iezzi nu kuit eshanaspit who blood-ACC make-3Sg PRT what-ACC avenger ishas tezzi takku tezzi akuwaras nas aku victim-GEN say-3Sg if say-3Sg die-Nmnl PRT-he die-Imp3Sg takku tezzima Sarnikduwa nu Sarnikdu if say-3Sg-but replace-Imp3Sg-3SgPro PRT replace-Imp3Sg (KUB XI 1 IV 19ff.) 'Who(ever) commits a murder whatever the avenger of the murdered man says, if he says "Let him die," let him die if he says "Let him make restitution," let him make restitution.' (45) ista virtus est...qui malum fert fortiter (PI. As. 323) 'that is courage, who (= if someone) bravely endures misfortune.' Even though complementizers and embedding have developed by the time of earliest Germanic there is still residual evidence for the adjoined structures in these languages. In Old English (unlike what is the case in modern German and Swedish for example) proposed subordinate clauses do not trigger verb second, which suggests that they are adjoined, rather than moved from an argument position to Spec-C: (46) peah hie aer paes ecan lifes orwene wajron, hie synt nu swipe blibe (BIHom 85.27) 'Even though thev were formerly despairing of eternal life they are now exceedingly joyful.'

158

THE DIACHRONY OF VERB SECOND

Secondly, temporal adverbial clauses are often followed by the resumptive element pa 'then', which, as noted in SECTION 2.3, attract the finite verb and are positioned in Spec-C; the clause that precedes it must therefore be adjoined toCP: (47) Ac pa pa Octauianus se casere to rice feng, pa wurdon lanas dura betyneda (Orosius 106.20) 'But when emperor Octavian ascended to the throne, then the doors of Janus were closed.' Third, adjunction is shown by the positioning of nested subordinate clauses: for example, the unless-clause in (48) precedes the complementizer of the clause that it belongs to:31 (48) Ac baem aefstegum is to secganne, gif hie hie nyllao healdan wio oaem aefste, oaet hiw weoroao besencte on oa ealdan unryhtwisnesse oaes lytegan feondes (Boethius, ConsPhil 233.16) 'But the envious are to be told that, if they will not stand fast against envy, they will be plunged into the old unrighteousness of the cunning fiend.' (literally: 'told, if..., that...') We can connect the fact that subordinate clauses were adjoined with the fact that they could undergo topicalization. Vedic and Hittite subordinate clauses, though not embedded, are semantically equivalent to ordinary embedded subordinate clauses of other languages (Hettrich (1988:127ff.)). So their ability to topicalize cannot be due to a lack of the required illocutionary force. Rather, we will assume that since they are functionally adjuncts rather than arguments or modifiers nothing prevents topicalization out of relative-clause postion to the left of the Focus position in (33).32

3.3 The Origin of V-to-C° Let us summarize the account of the origin of V-to-C° in Germanic at which we have arrived. When subordinate clauses changed in status from adjuncts to arguments and modifiers in their own right, they had to become CPs, and hence acquire complementizers. Germanic solved the problem by introducing pure complementizers which either headed embedded clauses with in situ resumptive pronouns or gaps, or were combined syntagmatically with a wh-phrase moved to Spec-C from the argument position. An ordinary relative pronoun or adverbial conjunction in effect covers two functional positions, both the head and the Specifier of CP: compare Jespersen's remark: (49) A relative pronoun really fills two functions: in the first place it serves to connect the clause with the rest of the sentence (generally with some antecedent) and in the second place it has its own

INDO-EUROPEAN ORIGINS OF GERMANIC SYNTAX

159

role to play within the clause as subject, object, or whatever it may be...." (1909-49, III) Germanic linearized these two functions in separate constituents — most transparently in relative clauses, which were formed with a relativized NP plus an indeclinable complementizer (pe, peat, er, som etc.). The immediate cause of the rise of V-to-C° in Germane, according to this scenario, is that Germanic created "pure" indeclinable complementizer slots devoid of nominal function and therefore capable of hosting verbal elements, and the ultimate cause is that subordinate clauses acquired argument and modifier status. At this point it would be appropiate to address the other theories which have been advanced to explain Germanic verb movement. These fall into two more or less clear-cut types. Theories of the first type view Germanic verb second as an historical continuation of some grammatical process inherited from IndoEuropean, the most popular being cliticization and verb-topicalization. Theories of the second type view verb second as arising entirely within Germanic in consequence of other changes that took place within that family. An example would be the proposal of Vennemann (1975) that verb second is functionally motivated by the depauperation of the case system, which causes a drift to SVO order in order to allow grammatical relations to be kept apart. Of course, these ideas are not necessarily incompatible. For example, we could imagine that verb second has roots in Indo-European but was elaborated within Germanic in ways internally motivated within that family. Our own proposal has this character, in that we take the constituent structure that underlies Vto-C° to be substantially inherited from Indo-European, but with V-to-C° itself arising as the result of a crucial modification in this structure caused by developments internal to Germanic. Because of space limitations I will restrict myself to a few comments on the most important representatives of the first type of theory. Since Wackernagel (1892) it has often been suggested that verb second in Germanic originates as a process of cliticization, specifically encliticization of unaccented elements to second position (Wackernagel's Law). The traditional version of this explanation is based on the assumption that finite verbs in main clauses were unaccented in Germanic (as we indeed know they were in Sanskrit). However, this can hardly have been the case because the alliterative conventions of Germanic epic poetry, which was certainly flourishing around the time verb-second word order gained ground in the Germanic languages, reveal that finite verbs (including those in C°) were accented. So even if Sanskrit-type verb deaccenting in main clauses could be projected back into Indo-European, it would have had to be lost in Germanic prior to the change it is supposed to explain. Secondly, this explanation is based on the mistaken premise that all unaccented words are clitics. In particular, although verbs in main clauses are unaccented in Sanskrit, they certainly are not clitics, and there is no reason to think that they ever were. In fact, cliticization of lexical categories such as main verbs does not appear to be attested in any language and has moreover been argued on theoretical grounds to be an impossibility

160

THE DIACHRONY OF VERB SECOND

(Inkelas (1989)). Thirdly, the proposal that verb movement is cliticization to Wackernagel's position fails to account for verb-first order, a conspicuous construction in all the early Germanic languages, and one manifestly cognate with verb second from a structural point of view, and very likely from an historical point of view as well (see the examples in (5) and the discussion in SECTION 2.4). According to the variant of the cliticization hypothesis proposed by Hock (1982), verb second develops as a generalization of a process cliticizing Auxiliaries. Auxiliaries are not lexical categories and so it is a priori not impossible that they were cliticized in early Germanic. However, I am not aware of any actual evidence that they were. Hock's putative evidence shows only that they were unaccented whereas a demonstration that they were clitics would require evidence that they were placed in Wackernagel's position. As far as I am aware, the only topicalization process to which Auxiliaries were subject (other than V-to-C° itself) is the above-mentioned Verb raising rule, but even if we suppose that it was at one time restricted to Auxiliaries (and there is no good evidence that it was), it quite clearly does not have a second-position landing site. The biggest stumbling-block for any attempt to derive verb second from cliticization is that the two processes have quite distinct properties, a fact unfortunately .disguised by the use of non-committal "S VO" terminology. Clearly the verb-second process that arises from the putative generalization cannot be assumed to remain a cliticization rule, now applying to main verbs as well for the reasons stated in the preceding paragraph. If, on the other hand, it is V-to-C° movement (as the syntax of the Germanic languages strongly indicates), then it could not simply be the result of generalizing a cliticization rule to main verbs. In particular, any Aux-cliticization rule (and for that matter, Verb raising) would have had to apply regardless of clause type, so that the explanation for the difference between main and subordinate clauses, which made the original cliticization hypothesis about the origin of verb second so attractive, is lost again.33 Anderson (1992) proposes to connect cliticization and verb second as manifestations of a principle which he formulates as follows: "locate the formal reflection of a linguistic unit's relational properties by reference to a prominent position in that unit." In Germanic, the reference point is the first constituent, the location is directly after it and the elements placed there are either clitics or inflectional features: if they are inflectional features, the verb is moved there as well to support them. This version of the verb second as cliticization story has the signal virtue of obviating the dubious assumption that verbs are unstressed. However, like the others it fails to unify the verbsecond and verb-first clause types as manifestations of verb movement. It also assumes that clitics are positioned by purely morphosyntactic rules, against much evidence that prosodic constituency is decisive (Halpern (1992)); it is also not clear how it would account for the regularities discussed in SECTIONS 2.3 and 3.1 in a principled way. Lenerz (1984, 1985) proposes to relate verb second to another inherited topicalization process, namely movement of the finite verb to initial position.

INDO-EUROPEAN ORIGINS OF GERMANIC SYNTAX

161

He suggests that Germanic verb-second word order is the result of reanalysis of a verb-first word order option which can be assumed to have existed already in Indo-European (Dressier (1969)). In this scenario, verb-first word order, originally derived by "stylistic proposing" or topicalization, came in Germanic to be reinterpreted as arising from V-to-C° movement, and as a result a structural Topic position was introduced before the verb. Schematically,

is reanalyzed as:

The general idea of tracing the rise of the Germanic verb-first/verb-second/ verb-final word order pattern to a reanalysis of phrase structure, specifically to a restructuring of Comp, is attractive, and in fact the present account adopts a version of it. The particular reanalysis posited by Lenerz, though, does not constitute an explanation for the rise of V-to-C° movement, nor can it, as far as I can see, be made part of one: it assumes the wrong syntactic structure for Germanic, fails to tie in Germanic with the Indo-European system, and renders the change isolated and unexplained. The present proposal while much in the same spirit, does seem consistent with current understanding of IndoEuropean syntax, and has the advantage of situating the change in the context of the evolution of the Germanic syntactic system, thereby making possible a structural explanation for V-to-C° movement.

4. Conclusion I have argued that the rise of V-to-C° movement is just one of several reflexes of the introduction of complementizers which itself is just one of several reflexes of the shift from adjoined to embedded clause structure. The most celebrated feature of Germanic syntax turns out to be only the final stage of a complex causal chain of syntactic innovations. Chronologically, too, it is the last of the major Germanic syntactic innovations. The shift from adjunction

162

THE DIACHRONY OF VERB SECOND

to embedding, as well as the associated rise of the C° slot with its indeclinable subordinating complementizer, are pan-Germanic and most likely ProtoGermanic. V-to-C° movement, on the other hand, belongs in a group of innovations that are absent from Gothic (and probably from the earliest Scandinavian of the Runic inscriptions), and which begin to spread shortly before the historical period, with language-specific variations, throughout the rest of the Germanic family.34 The last stage in this process is the elimination of the variation between CP and S main clauses, preserved in the guise of apparent optionality of verb second in Old English. This homogenization went in two opposite directions. In English, the possibility of main clause CP was eliminated, resulting in the loss of verb second (and, concurrently, of main clauses with overt complementizers such as (3)). In the other Germanic languages, CP became obligatory in main clauses, resulting in fixed verb-second word order. The diversity of Old English word order is then an archaic trait, as is its treatment of pronominal clitics and the structure of its relative clauses. If this analysis is right, then the inherited distinction between the Topic and wh landing sites is not only preserved in Old English but ultimately lies behind the Modern English "residual verb second" system as well. The contrast between Where will Max put a book? and On the table Max will put a book has Indo-European roots. The other Germanic verb-second languages have innovated by effectively collapsing the two positions (Wohin wird Max ein Buch legen?, Aufden Tisch wird Max ein Buch legen.) In this respect at least, English is syntactically the most conservative of all the modern Germanic languages.

Notes 1. This is a revised and updated version of a paper read in 1989 at the 8th East Coast Indo-European Conference at Harvard and at a DYANA workshop on parametric variation in Germanic and Romance at Edinburgh, as well as in talks at Stanford, Berkeley, and Helsinki. I would like to thank the discussants on each of these occasions, especially Andrew Garrett and Elizabeth Traugott, as well as an anonymous reviewer, for their most helpful comments. 2. The insight that subordinate-clause word order is basic goes back to Bach (1962), Bierwisch (1963), and Koster (1975); the idea behind the V-to-C° analysis is anticipated in the work of Diderichsen (1941) (on which see Heltoft (1986)) and in Drach (1937). For two recent detailed theories of verb movement, with full discussion of the recent literature, see Viker (1990) and Roberts (1993). Lenerz (1984, 1985), and Kemenade (1987), show that the V-toC° analysis is also supported for the older Germanic languages (excepting Gothic), though there are some differences in detail, as we will see below. 3. The obligatoriness of V-to-C° movement has been ascribed to the ECP (Travis (1984)), to the feature composition of C° (den Besten (1983); Haider (1986); Lenerz (1985); Holmberg (1986); Holmberg and Platzack (1988); Roberts (1993)), to the need to assign Nominative Case to the subject (Platzack (1986)),

INDO-EUROPEAN ORIGINS OF GERMANIC SYNTAX

163

to a universal ban on vacuous quantification (Pollock (1989)), and to a universal principle that shorter derivations are preferred over longer derivations (Chomsky (1991)). See Vikner (1990) for enlightening discussion of the issue in the modern Germanic languages. 4. I defer the question of the order of the finite verb in S to SECTION 2.6. 5. In relative and adverbial clauses with an overt wh-element in Spec-C, the complementizer may be phonologically empty must in any case be assumed to be syntactically present. 6. Bacquet (1962:585-596), Behaghel (1932:27 ff.), Heusler (1962:173 ff.), Mitchell (1985, §3930 ff.), Nygaard (1905:348 ff.), Sigurosson (1985). Many of these authors remark that verb-first sentences have a special stylistic function, being typical of "lively narrative," often linking a sentence to the previous one. The construction is most common with auxiliary-type verbs, but occurs quite often with main verbs as well. 7. See Mitchell (1985, §3914 ff.) and Bacquet (1962:617-629 for discussion and examples. 8. Of course, even a verb in C° can be trivially in final position if all other constituents are moved in front of it, by wh-movement, by topicalization, by pronoun cliticization as discussed in SECTION 2.3 below, or by some combination thereof. Therefore, examples like (15b) below cannot be cited as evidence for the survival of Proto-Germanic verb-final main clause order in Old High German (contra Lenerz (1985:106)). Rare examples of verb-final main clauses do occur in Old High German scholarly texts; Behaghel (1932, § 1132) suggests that they reflect the influence of Latin word order rather than survivals of Germanic verb-final order. 9. This structure has been demonstrated in other languages as well. In the rather elaborate hierarchy of functional operator positions of Hungarian, the position of Topic is clearly external to that which hosts wh-elements and focused constituents (Horvath (1985); Kiss (1987)). Kiss suggests that this ordering is due to the principle that operators must c-command the elements in their scope. 10. With certain speech-act oriented adverbials, such as German aufrichtig gesagt 'frankly', tatsachlich 'in fact', resumption (by a fronted adverb, German so, Swedish sa) in the adjoined structure is optional. Even these speech-act oriented adverbials, however, are probably more commonly deployed in Spec-C position with verb second. See Altmann (1981) and Thim-Mabrey (1988) for discussion of the German data. 11. The fact that yes/no questions are verb-initial (except for the previously mentioned "whether" type) is explained on the assumption that they have an obligatory C° head but no overt wh-element or Topic before it. 12. See Stockwell (1984) for a detailed analysis of the ways in which Modern English Subject-Aux-Inversion reflects the Old English verb-second rule and the ways in which it has innovated. As Stockwell emphasizes, the fact that negation and certain adverbs pattern with wh-constituents also has a counterpart in Old English, as will be discussed shortly below . 13. Either because embedded wh-elements in Icelandic are complementizers (Thrainsson (1986)), or because they are accompanied by a phonetically unrealized complementizer (Platzack (1986)).

164

THE DIACHRONY OF VERB SECOND

14. In fairness it should be noted that this solution was not available to Maling and Zaenen since at that time they were not assuming V-to-C° movement but rather a verb-second surface filter. In fact, they took these data as evidence that wh-movement in Icelandic is adjunction to S to the left of C°, while topicalization is adjunction to S, to the right of C°. 15. In Old English, as well as in Modern English and Mainland Scandinavian languages, topicalization in embedded that-complements is allowed only after a class of non-factive verbs. Wechsler (1990) argues that this is a consequence of the special illocutionary status of these complements as "assertions". These clauses seem to represent the thought or words of the implicit or explicit agent of the containing verb. In Icelandic, however, embedded padclauses are reported to undergo topicalization freely. 16. This statement anticipates the treatment of relative clauses discussed in SECTION 2.5. 17. See Kemenade (1987, chs.4-5) for other arguments that Old English weak pronouns are "syntactic clitics" and for extensive discussion of preposition stranding. 18. See Kemenade (1987:138), Mitchell (1985, §1599 ff., §3922 ff.) for the evidence. Old Icelandicpa 'then' behaves in a similar way (Heusler (1962:174)). 19. For pa, though, things are a bit complicated because, in addition to being an adverb, it can also be a subordinating conjunction, in which case the verb must naturally be in final position. It is sometimes not possible to tell which it is in a given sentence. Moreover, there appear to be some instances where the adverb pa is topicalized rather than focused (Mitchell (1985, §2547 ff.)). 20. Archaic Old English seems to have had a different system in which pronouns were proclitic (Fourquet (1974); Hock (1988)). 21. Setting aside, as before, the possibility of "embedded root clauses" under verbs of saying and believing (fn. 15). 22. Hock (1988) suggests that the Germanic pronoun + complementizer complex arises by rebracketing of the main clause correlative with the indeclinable complementizer of the subordinate clause. This is an interesting possibility, but seems to predict wrongly that the pronoun should be inflected in the case required by the main clause, rather than in the case required by the subordinate clause, and fails to explain why the pronoun + complementizer complex is frequently accompanied by a correlative pronoun in the main clause, as well as the obligatoriness of pied piping. 23. The two types of relative clauses are comparable to the nasalizing and leniting relative clauses of Irish (Hale n.d.; McCloskey (1979); McCone (1980); Sells (1984)), which have been argued to involve deletion (or, more precisely, null resumptive pronouns) vs. wh-movement, respectively (see especially McCloskey (1990)). Since Irish (and Celtic in general) shares also V-to-C° movement with Germanic, there may be a deeper structural parallelism here. 24. In this section, unless otherwise noted, Old English examples are cited from Allen and Old Icelandic examples are cited from Nygaard (1905:256-265) and from Heusler (1962:158-163).

INDO-EUROPEAN ORIGINS OF GERMANIC SYNTAX

165

25. As Allen shows, there is a parallel distinction between relative clauses with no overt subordinatorpe, depending on whether a demonstrative pronoun is present or not. These relative clauses behave in exactly the same way as the corresponding clauses with overt pe. They can be considered as having a phonologically null (but syntactically visible) complementizer functionally identical to pe. 26. A similar proposal was made by Ingria (1980) for Modern Greek, who showed that it has counterparts of both Germanic types of relative clauses. 27. Let us also repeat here that Old English, like Modern English and the Scandinavian languages, might very well have had "root" phenomena in subordinate clauses, which would have constituted another source of apparent exceptions to the basic verb-final order. 28. Particles such as nu belong in a pre-topic position (perhaps as proclitics to S"). 29. That is, the correlative pronouns could be omitted only under the same conditions as pronouns in general could be omitted, namely rather freely when nominative or accusative, and only very marginally in other case forms. Hettrich (1988:546) notes that the conditions are different for non-restrictive relative clauses. 30. Really this is not an autonomous generalization but a reflection of two other tendencies: the relative clause usually precedes the main clause, and whichever clause comes first usually contains the "head". See Hettrich (1988:578, 682) for statistics and discussion. 31. O'Neil (1977) has claimed that relative clauses are "(almost) always at the margins of the main clause" in Old English. Mitchell (1985, §2288 ff.) however, notes that they are "commonly found in the position usual in MnE, viz. immediately after the antecedent or separated from it by a word or phrase which qualifies it." 32. To say that subordinate clauses were adjoined is not to say that their syntax was identical to main clauses. They were still subordinate clauses, linked to a correlative element in the main clause. However, Hettrich (1988) notes that non-restrictive relative clauses in Vedic do have essentially the syntactic properties of main clauses; in particular, the relation between relative and correlative is governed essentially by the same principles that govern anaphoric relations between main clauses. 33. Hock appeals to the Penthouse Principle (Ross (1973)), but this is only legitimate if verb second is assumed to be V-to-C° after all, for the only genuinely syntactic evidence for the Penthouse Principle comes precisely from the sorts of main-clause fronting phenomena which we have been concerned with here (including most prominently topicalization and verb second itself); there is no evidence that anything like a Penthouse Principle constrains syntactic rules in general. Ross also cited a number of special semantic properties characteristic of main clauses, such as applicability of conversational postulates such as that which allows questions to be interpreted as requests, occurrence of pseudo-imperatives (imperatives interpreted as conditionals), and certain idioms, but these are probably consequences of the status of main clauses as

166

THE DIACHRONY OF VERB SECOND

independent speech acts such as assertions and requests. In fact, in so far as such speech acts can be embedded qua assertions, requests etc., they do undergo the same processes (Hooper and Thompson (1973); Andersson (1975); Green (1976); Bolinger (1977)). 34. Thus, the restructuring has nearly the same chronology and distribution as Umlaut and some of the vowel syncope processes.

References Allen, C. (1974) "Old English Modals." University of Massachusetts Occasional Papers in Linguistics 1:89-100. Allen, C. (1977) Topics in English Diachronic Syntax. PhD Dissertation, MIT. Reprinted in 1981, Garland Press. Altmann, H. (1981) Formen der "Herausstellung" im Deutschen. Tubingen: Niemeyer. Anderson, S. (1992) A-morphous Morphology. Cambridge: Cambridge University Press. Andersson, L.-G. (1975) "Form and Function of Subordinate Clauses." Monographs in Linguistics 1. Bach, E. (1962) "The Order of Elements in a Transformational Grammar of German." Language 38:263-269. Bacquet, P. (1962) La structure de la phrase verbale a I'epoque alfredienne. Paris. Behaghel, O. (1932) Deutsche Syntax, IV. Heidelberg: Winter. Besten, H. den (1983) "On the Interaction of Root Transformations and Lexical Deletion Rules." In W. Abraham, ed. On the Formal Syntax of the Westgermania, 47-131. Amsterdam: John Benjamins. Bierwisch, M. (1963) Grammatik des Deutschen Verbs (Studia Grammatica II). Berlin: Akademie-Verlag. Bolinger, D. (1977) "Another Glance at Main Clause Phenomena." Language 53:511-519. Calboli, G. (1987) "Die Syntax der altesten lateinischen Prosa." In A. G. Ramat et al., eds. Papers from the 7th International Conference on Historical Linguistics. Amsterdam: John Benjamins. Chomsky, N. (1991) "Some Notes on Economy of Derivation and Representation." In R. Friedin, ed. Principles and Parameters in Comparative Grammar, 417-454. Cambridge, Mass.: MIT Press. Delbruck, B. (1878) Syntaktische Forschungen HI. Halle. Diderichsen, P. (1941) Scetningsbyningen i Skaanske Lou. K0benhavn: Munksgaard. Drach, E. (1937) "Grundgedanken zur deutschen Satzlehre." Frankfurt. Dressier, W. (1969) "Eine textsyntaktische Regel der idg. Wortstellung. (Zur Anfangsstellung des Pradikatsverbums)." KZ 83:1-25. Fourquet, J. (1974) "Genetische Betrachtungen uber den deutschen Satzbau." In W. Besch, G. Jungbluth, G. Meissburger and E. Nellmann, eds.

INDO-EUROPEAN ORIGINS OF GERMANIC SYNTAX

167

Studien zur deutschen Literatur und Sprache des Mittelalters. Festschrift fur Hugo Moser zum 65. Geburtstag, 314-323. Berlin. Garrett, A. (1990) The Syntax of Anatolian Pronominal Clitics. PhD Dissertation, Harvard. Garrett. A. (1992) "Topics in Lycian Syntax." Historische Sprachforschung 105:200-212. Green, G. (1976) "Main Clause Phenomena in Subordinate Clauses." Language 52:382-397. Haider, H. (1986) "V-Second in German." In H. Haider and M. Prinzhorn, eds. Verb Second Phenomena in the Germanic Languages, 49-76. Dordrecht: Foris. Hale, K. (1976) "The Adjoined Relative Clause in Indo-European." In R.M.W. Dixon, ed. Grammatical Categories in Australia. Canberra: Australian Institute of Aboriginal Studies. Hale, K. (n.d.) "Agreement and Incorporation in Athabaskan and in General." Ms. MIT. Hale, M. (1987) "Notes on Wackernagel's Law in the Language of the Rigveda." In C. Watkins, ed. Studies in Memory of Warren Cowgill. New York/Berlin: de Gruyter. Hale, M. (1989) "Clitics and Constituency." ECIEC 8. Halpern, A. (1992) Topics in the Placement and Morphology of Clitics. PhD Dissertation, Stanford University. Haudry, J. (1973) "Parataxe, hypotaxe et correlation dans la phrase latine." BSL 68:147-186. Held, W.H. (1957) "The Hittite Relative Sentence." Linguistic Society of America, Language Dissertation No. 55. Heltoft, L. (1986) "The V/2 analysis — a reply from the Diderichsen tradition." In 6. Dahl and A. Holmberg, eds. Scandinavian Syntax, 50-66. Institute of Linguistics, Stockholm University. Hettrich, H. (1988) Untersuchungen zur Hypotaxe im Vedischen. Berlin: de Gruyter. Heusler, A. (1962) Altisldndisches Elementarbuch. Heidelberg: Carl Winter. Hock, H.-H. (1982) "AUX-Cliticization as a Motivation for Word-order Change." Studies in the Linguistic Sciences 12:91-101. Hock, H.-H. (1988) "Rebracketing and Relative Clauses in Old English." In E. Antonsen and H -H. Hock, eds. Germanic Linguistics II: Papers from the Second Symposium on Germanic Linguistics. Bloomington. Ind.: Indiana University Linguistics Club. Hock, H.-H. (1989) "Early Indo-European Syntactic Typology: a Different Approach." ECIEC 8. Holland, G.B. (1984) "Subordination and Relativization in Early Indo-European." BLS 10:609-622. Holmberg, A. (1986) Word Order and Syntactic Features in the Scandinavian Languages and English. PhD Dissertation, University of Stockholm. Holmberg, A. and C. Platzack (1988/) "On the Role of Inflection in Scandinavian Syntax." Working Papers in Scandinavian Syntax 42:25-42.

168

THE DIACHRONY OF VERB SECOND

Hooper, J. and S. Thompson (1973) "On the Applicability of Root Transformations." Linguistic Inquiry 4:465-497. Horvath, J. (1985) Aspects of Hungarian Syntax. Dordrecht: Foris. Ingria, R. (1980) PhD Dissertation, MIT. Inkelas, S. (1989) Prosodic Lexical Phonology. PhD Dissertation, Stanford University. Jespersen, O. (1909-49) A Modern English Grammar on Hisorical Principles. London: George Allen and Unwin. Kayne, R. (1982) "Predicates and Arguments, Verbs and Nouns." In GLOW, Paris. Kemenade, A. van (1987) Syntactic Case and Morphological Case in the History of English. Dordrecht: Foris. Kiss, K. E. (1987) Configurationality in Hungarian. Dordrecht: Reidel. Koster, J. (1975) "Dutch as an SOV Language." Linguistic Analysis 1:111136. Lehmann, C. (1980) "Der indogermanische A w i-/k w o-Relativsatz im typologischen Vergleich." In P. Ramat, et al., eds. Linguistic Reconstruction and Indo-European Syntax. Amsterdam: John Benjamins. Lenerz, Jurgen (1984) Syntaxtischer Wandel und Grammatiktheorie. Eine Untersuchung an Beispielen aus der Sprachgeschichte des Deutschen. Tubingen: Niemeyer. Lenerz, J. (1985) "Diachronic Syntax: Verb Position and COMP in German." In J. Toman, ed. Studies in German Grammar, 103-132. Dordrecht: Foris. Maling, J. (1976) "Old Icelandic Relative Clauses: an Unbounded Deletion Rule." NELS7:175-188. Maling, J. and A. Zaenen (1981) "Germanic Word Order and the Format of Surface Filters." In F. Heny, ed. Binding and Filtering, 255-278. London: Croom-Helm. McCloskey, J. (1979) Transformational Syntax and Model Theoretic Semantics. Dordrecht: Reidel. McCloskey, J. (1990) "Resumptive Pronouns, A-Binding and Levels of Representation on Irish." In R. Hendrick, ed. The Syntax of the Modern Celtic Languages, 199-248. New York: Academic Press. McCone, K. (1980) "The Nasalizing Relative Clause with Object Antecedent in the Glosses." Eriu 31:10-27. Mitchell, B. (1985) Old English Syntax. Oxford: Clarendon Press. Nygaard, M. (1905) Norr0n Syntax. Kristiania: Aschehoug & Co. O'Neil, W. (1977) "Clause Adjunction in Old English." General Linguistics 17:199-211. Pintzuk, S. (1991) Phrase Structures in Competition: Variation and Change in Old English Word Order. PhD Dissertation, University of Pennsylvania. Pintuk, S. and A. Kroch (1989) "The Rightward Movement of Complements and Adjuncts in the Old English of Beowulf." Language Variation and Change 1:115-143.

INDO-EUROPEAN ORIGINS OF GERMANIC SYNTAX

169

Platzack, C. (1986) "COMP, INFL, and Germanic Word Order." In L. Hellan and K. Koch Kristensen, eds. Topics in Scandinavian Syntax, 185-234. Dordrecht: Reidel. Pollock, J.-Y. (1989) "Verb Movement, UG, and the Structure of IP." Linguistic Inquiry 20:365-424. Rizzi, L. (1990) "Speculations on Verb Second." In J. Mascaro and M. Nespor, eds. Grammar in Progress: a Festschrift for Henk van Riemsdijk, 375385. Dordrecht: Foris. Roberts, I. (1992) Verbs and Diachronic Syntax. Dordrecht: Kluwer. Ross, J.R. (1973) "The Penthouse Principle." CIS 9. Sells, P. (1984) Syntax and Semantics of Resumptive Pronouns. PhD Dissertation, Amherst. Sigurosson, H.A. (1985) "Subordinate V/l in Icelandic: How to Explain a Root Phenomenon." Working Papers in Scandinavian Syntax 18. Stockwell, R. (1984) "On the History of the Verb-second Rule in English." In J. Fisiak, ed. Historical Syntax. Trends in Linguistic: Studies and Monographs 23:575-592. Berlin: de Gruyter. Taraldsen, K.T. (1986) "On Verb Second and the Functional Content of Syntactic Categories." In H. Haider and M. Prinzhorn, eds. Verb Second Phenomena in the Germanic Languages, 7-26. Dordrecht: Foris. Thim-Mabrey, C. (1988) "Satzadverbien und andere Ausdrucke im Vorvorfeld." Deutsche Sprache 16:52-67. Thrainsson, H. (1986) "V1,V2,V3 in Icelandic." In H. Haider and M. Prinzhorn, eds. Verb Second Phenomena in the Germanic Languages, 169-194. Dordrecht: Foris. Travis, L. (1984) Parameters and Effects of Word-Order Variation. PhD Dissertation, MIT. Vennemann, T. (1975) "An Explanation of Drift." In C.N. Li, ed. Word Order and Word Order Change. Austin, Texas: University of Texas Press. Vikner, S. (1990) Verb Movement and the Licensing of NP-positions in the Germanic Languages. PhD Dissertation, Universite" de Geneve. Wackernagel, J. (1892) "Uber ein Gesetz der indogermanischen Wortstellung." Indogermanische Forschungen 1:333-436. Watkins, Calvert (1976) "Towards Proto-Indo-European Syntax: Problems and Pseudo-Problems." In S.B. Steever et al., eds. CLS Parasession on Diachronic Syntax, 305-326. Chicago: Chicago Linguistics Society. Wechsler, S. (1990) "Verb Second and Illucutionary Force in Swedish." Edinburgh Working Papers in Cognitive Science 6:229-244.

This page intentionally left blank

Part Two Verb Second and the Null-Subject Parameter

This page intentionally left blank

7 On the Decline of Verb Movement to Comp in Old and Middle French* Barbara Vance Indiana University

1. Introduction In the study presented here I examine inverted declarative clauses from several Old and Middle French texts written between the 13th and the 15th centuries and suggest, on the basis of synchronic and diachronic differences within this class of clauses, an explanation for the decline of verb movement to Comp over the period in question. Although it has often been claimed (cf. e.g. Martin (1980); Adams (1987a,b), Lemieux and Dupuis (this volume)) that French remains essentially a verb-second language until the end of the Middle French period (14th-15th c.), my data show clearly that not all inversions are strictly of the verb-second type, despite many surface indications that are consistent with such an interpretation. We will look at two developments that we claim had their origins at least as early as the 13th century, and which came to dominate the grammar in Middle French: the coexistence of "Germanic" verbsecond inversions and Romance "free inversion" (SECTION 2), and the dissociation of the initial constituent from the inverted clause, also responsible for the rise of verb-third orders (SECTION 3). I will claim that the variations observable in 13th-century inversions are the seeds from which the decline of the verb-second constraint in French developed. It is important to emphasize that the focus of this study is the decline, rather than the actual loss, of verb-second inversions in French. The studies of Kroch (1989) and Roberts (1993), for example, show clearly that parameter resetting takes place only after the gradual accumulation of evidence for the new grammar results in a new generation's inability to construct the old grammar out of the linguistic data it receives. In the case of verb-second in French, the crucial generation seems to have lived in the 16th century, well after the time period we will consider. I will assume for present purposes that the gram173

174

VERB SECOND AND THE NULL-SUBJECT PARAMETER

matical property at issue is the movement of the verb to Comp in root clauses (V-to-C), but it should be noted that the resetting of an even more abstract parameter—such as the Nominative-Case assignment parameter of Roberts— is likely to be ultimately responsible for the loss of verb second. What concerns us most here is simply the way in which certain inversion structures came to be attributable to other grammatical properties than V-to-C, reducing the amount of overall positive evidence for this movement in the late Old French/ Middle French period.

2. The Position of the Subject in Inversion In this section we examine the possible positions of the postverbal subject in Old French (OF) and distinguish the syntactic behavior of subject NP's from subject pronouns. We then look at evidence for changes over time in the proportion of pronominal inversions, which we claim are generated only by Vto-C movement, and non-pronominal inversions, which can be generated by a non-verb-second grammar as well. Finally, we see that a rise in the number of unaccusative and passive verbs in inversion constructions supports the notion that Middle French (MidF) inversion is increasingly independent of the verb-second constraint.

2.1 "Germanic" Inversion and "Free" Inversion The verb-second property of OF creates from basic SVO order "inverted" structures of the type in (1) and (2), where an initial non-subject constituent occupies the specifier of CP, the finite verb moves to C°, and the subject (italics) remains in its base position:1 (1) Sor ceste piere edefierai je m'eglise (Queste del Saint Graal (Q) on this rock will-build I my-church 101,31) (2)

Longuement resgarda Perceval 1'ome

(3)

[CP Longuement [c resgarda [IP Perceval [I' t [ VP t l'ome...]]]

qui ou lit

seoit (Q 82,25) long watched Perceval the-man who on bed sits 'For a long time Perceval watched the man who was sitting on the bed'.

(1) and (2) represent the most common type of inversion in OF, where the positions of verb and subject can be entirely explained by the V-to-C movement proposed for Germanic verb-second by den Besten (1983) and adapted for OF by Adams (1987a,b). As predicted by this model, such inversions are impossible in non-root contexts, where the Comp node is not available as a landing site for the verb.2 Instead, the verb moves only as far as I and the subject remains preverbal:

ON THE DECLINE OF VERB MOVEMENT TO COMP

(4)

175

(Et lors li

conte) coment il avoit veu le Saint Graal... (Q 66,19) (and then to-him tells) how he had seen the Holy Grail

Matrix SVO orders in OF, according to this view of verb second, are CP's in which the subject has moved to Spec CP. (On the obligatory nature of the two movements associated with the verb-second constraint, see Hulk and van Kemenade (this volume); Platzack (1987); Roberts (1993); Tomaselli (1989)). All matrix-clause inversion in OF is compatible with a V-to-C analysis of verb movement. Some inversions involving non-pronominal subjects, however, make use of additional possibilities for the placement of the subject, as in (5a-d): (5) a. Si plorerent assez a cest departement thus cried much at this departure cil qui plus cuidoient avoir les cuers et durs et those who most thought to-have the hearts both hard and orgueillox (Q 26,19) proud b. car assez 1'ot eschaufe' li serpenz (Q 95,1) for much him-had angered the serpent (nom.) 'for the serpent had angered him greatly' c. et par ceste parole entra en aus covoitise (Q 103,12) and by this word entered into them covetousness d. car ja seront repeu li verai chevalier de la viande for now will-be fed the true knights of the food del ciel (Q 267,14) of-the heaven. These subjects appear to be in a position at the right periphery of the VP analogous to the surface position of the subject in "free inversion" in Spanish or Italian and "stylistic inversion" in Modern French. Notice that all types of verbs may participate in this construction: the verb of (5a) is an intransitive unergative, that of (5b) is transitive, (5c) is an intransitive unaccusative, and (5d) is passive. Furthermore, the subjects may be phonologically "heavy," as in (5a), or relatively light as in (5b-d). Although our data show some restrictions on what can intervene between the finite verb and the subject in such constructions, these restrictions are similar to those observed for Italian inversion by Calabrese (1992), Rizzi (1991), and Saccon (1993). Hence we feel justified in concluding, contra Adams (1987a,b), that OF had a productive free inversion construction.3 We designate this position as "VP-final" despite the occasional presence of extraposed elements such as the prepositional phrase in (5d). Since they occur further to the right than V°, the subjects in (5) are unaffected by the leftward movement of the finite verb. Hence, such inversions are predicted to occur even in embedded clauses, where the verb moves to I

176

VERB SECOND AND THE NULL-SUBJECT PARAMETER

rather than C. Embedded inversion is rare in the Queste but not impossible; ten cases exist, to my knowledge, in the entire text (but see also note 2). Example (6) is representative: (6)

...quant vint par la volent6 when came by the will (of) chevaliers de la meson le knight of the house (of) the

Nostre Seignor Calogrenant, uns Our Lord Calogrenant a roi Artus... (Q 190,27) king Arthur

An ambiguous inverted word order is thus possible in OF, variably analyzed as the result of simple V-to-C movement or of free inversion: (7)

maintenant s'en ala la damoisele (Q 112,22) now refl-'en' went the maiden 'now the maiden went away'.

Either this clause is directly parallel to (2), where the subject is in Spec IP, or it is an instance of a VP-final subject, as in (5), which happens not to be separated from the finite verb by any other VP material. If OF had obligatory Vto-C movement, as we have assumed, then only the position of the subject is in fact ambiguous. But since these types of clauses are not crucially dependent upon V-to-C, and since we know that French has lost V-to-C movement as a productive rule of grammar, we are led to examine (5a-d), and (7) as possible sources of diachronic change. In so doing, we extend and unify suggestions made elsewhere (see Vance (1988) with respect to unaccusative and passive declaratives and Roberts (1993) primarily on interrogatives) that the possibility of interpreting certain CVS structures as either free inversion or Germanic inversion can trigger grammatical reanalysis over time. It is important to note that pronominal subjects are not possible in the inversions of either (5a-d) or (6). Vance, largely following Kayne (1983), accounts for both facts by proposing that subject pronouns cannot occur lower in the tree than Spec IP, and that they may in fact be enclitic on Comp when postverbal. Roberts (section 2.2.2) invokes two additional arguments for this condition on subject pronouns which suggest that encliticization to Comp may be a general property of subject pronouns in OF. In inverted matrix clauses, a pronominal subject must be immediately postverbal; in embedded clauses, it must be preverbal, immediately to the right of the subordinator. The hypothetical pronominal versions of (5a) and (6) are then (5a') and (6'), respectively. These hypothetical sentences correspond exactly to the well-represented sentence types illustrated in (1) and (4) above. (5a) Si plorerent assez a cest departement thus cried much at this departure cil qui plus cuidoient avoir les cuers et durs et those who most thought to-have the hearts both hard and orgueillox (Q 26,19) proud

ON THE DECLINE OF VERB MOVEMENT TO COMP

177

(5a') si plorerent ils assez... thus cried they much (6)

...quant vint par la volente" Nostre Seignor Calogrenant, uns when came by the will (of) Our Lord Calogrenant a chevaliers de la meson le roi Artus... (Q 190,27) knight of the house (of) the king Arthur

(6') quant il vint... when he came We note, then, that pronominal inversions such as (5a') and (1) are not potentially ambiguous in the way that (7) is; they can only be generated by verb movement to Comp in French.

2.2 The Diachrony of Inversion Before beginning our study of the evolution of inversion, let us clarify some assumptions about the position of postverbal subjects. According to Burzio (1986), the subjects of unaccusative (ergative) and passive verbs in Italian may remain in their base-generated direct object position, where they receive Nominative Case by special mechanisms. In such cases, then, there is no true "inversion" at all despite the appearance of a postverbal subject. Other inverted subjects in Italian, Burzio assumes, become postverbal via NP movement. More recently it has been proposed, alternatively, that the subject in Italian and Spanish is always base-generated in a VP-final position (e.g. Contreras (1987); Groos and Bok-Bennema (1986)); Roberts (1993) assumes the same is true of French. I follow Burzio's account here in order to maintain a distinction between the position of unaccusative and passive subjects and other subjects, a contrast that is of potential importance but is more difficult to formulate under the newer hypothesis. We noted above that in OF the postverbal options for the surface position of the subject may simply be superimposed upon the basic verb-second structure; that is, V-to-C may apply obligatorily in matrix clauses even if it is not alone responsible for the presence of a postverbal subject. There is already some evidence in the 13th century, however, that VP-final subjects account for a growing, rather than a static, proportion of the total inversions. Although I have at present no figures on inversion before the 13th century, a survey of several early texts suggests that in some texts sentences of the type in (5) are virtually unknown. In the Queste del Saint Graal (1225), on the other hand, they account for 16% of the inversions in our sample. This proportion increases to 34% in our latest text, Commynes' Memoires (1491). Table 7.1 gives more complete information on the distribution of word-order types within inversion constructions for these two texts, as well as for La Vie de Saint Louis of Joinville (1306) and Antoine de la Sale's Jehan de Saintre (1456). (I follow philological tradition in abbreviating: C = initial constituent, V = finite verb, S = subject, Sp = subject pronoun, Sn = non-pronominal subject, X = intervening material.)

178

VERB SECOND AND THE NULL-SUBJECT PARAMETER

Table 7.1 Pronominal vs. Non-pronominal Inversion in Old and Middle French4

Queste (1225) pp. 1-41,71-115 Joinville (1306) pp. 104-125 Saintre (1456) pp. 239-294 Commynes (1491) pp. 58-97

Total

CVSp

CVSn

CVXSn

419

212 (.51)

138 (.33)

69 (.16)

74

24 (.32)

43 (.58)

7 (.10)

44

15 (.34)

23 (.52)

6 (.14)

93

11(.12)

50 (.54)

32 (.34)

Although the total numbers of inversions in each text are not directly comparable, due to the difference in the lengths of the passages, it is clear that subject inversion, while still a prominent word order in MidF, becomes less frequent after the 13th century. (Compare 93 inversions per 40 pages in Commynes, the MidF text which most favors inversion, to roughly 210 inversions in 40 pages of the OF text.) Furthermore, within the class of inverted word orders we find some interesting diachronic patterns. The proportion of pronominal inversions (CVSp) to other types of inversion decreases steadily over time. The percentage of CVXSn inversions appears to remain stable until the late 15th century, whereas CVSn jumps to 50% of inversions by the beginning of the 14th century and remains there. Recall, however, that the category CVSn artificially groups together both simple V-to-C inversions, identical in structure to CVSp, and ambiguous clauses such as (7), which may be either simple V-to-C clauses or examples of CVXSn in which the subject is VP-final but X=0. In order to interpret properly the proportion of non-pronominal inversions in the data, then, we must divide this category further into ambiguous and unambiguous clauses. In establishing criteria for identifying ambiguous clauses, I will once again assume that Spec IP is the normal case position for subjects. In a clause such as (2), then, the subject is unambiguously in Spec IP, to the left of the object in VP, NP. Also unambiguous are instances in which the subject separates the finite verb from a non-finite verb, as in (8) (the object pronoun vos is a clitic on the finite verb): (8)

autresint vos a Nostre Sires esleu (Q 38,16) in the same way you has Our Lord elected.

On the other hand, the ambiguous clauses—those in which the subject may in fact be VP-final—have a simple verb and no postverbal object. They may, however, have extraposed PP's analogous to the one in (5d): (9)

ainz demorront li plusor en ceste Queste (Q 17,6) rather will-remain the most in this Quest.

ON THE DECLINE OF VERB MOVEMENT TO COMP

179

The case of passives is rather different, however. Since passives by definition have at least two verbs, all of the passives included in the CVSn count pat-

tern on the surface like (8): (10) et lors seront tes plaies garies (Q 86,2) and then will-be your wounds healed. It is not necessary to conclude, however, that the underlying direct object tes plaies has moved to Spec IP. The occurrence of the direct object before, rather than after, the past participle is common in Old and Middle French i inverted as well as non-inverted clauses. In such cases, the object has p sumably moved to Spec VP: (11) Lors ot Eve virginite" perdue (Q 214,5) then had Eve virginity lost (11') [CP lors [c' ot i [ IP Eve [I' ti [VP virginitej [v, perdue [NP tj]. The underlying object tes plaies in (10), then, might be located either in Spec IP, parallel to the subject Eve of (11), or in Spec VP, parallel to the object virginite. I thus conclude that the passives included under the CVSn column are not obligatorily interpreted as simple V-to-C constructions and so the position of their subjects is, as in the case of (7) and (9), ambiguous.5 Table 7.2 divides up the category CVSn into simple V-to-C clauses and clauses in which the subject is VP-final, according to these guidelines. Clearly, it is the ambiguous inverted clauses that account for most of the CVSn class, even in the earliest text. As time goes on, inversions that could be produced without V-to-C movement increase from 43% of total inversions to 64% to 85%. Seen from this angle, the figures for Jehan de Saintre fall into place; although this text has relatively few CVXS clauses, this situation is due simply to a small number of VP-final subjects actually separated from the verb by phonetic material and not to a shortage of VP-final subjects in general. Table 7.2 Simple V-to-C and VP-final Subject Clauses in Old and Middle French Simple V-to-C Clauses

VP-final Subject Clauses CVSn

CVSp

CVSnV(-fin) CVSn# CVSnO

CVXSn

Total VP-final

Q J

212 (.51)

27 (.06)

111 (.26)

69 (.16)

180 (.43)

24 (.32)

6 (.08)

37 (.50)

7 (.10)

44 (.60)

S C

15 (.34)

1 (.02)

22 (.50)

6 (.14)

28 (.64)

11 C12)

3 (.03)

47 (.50)

32 (.34)

79 (.85)

180

VERB SECOND AND THE NULL-SUBJECT PARAMETER

Another clue to the changing nature of inversion in late OF and MidF comes from the high proportion of unaccusative and passive verbs in VP-final inversions. Since the subjects of such verbs are underlyingly direct objects, it might be supposed that they favor word orders in which movement out of the VP is not necessary. Table 7.3 contrasts the percentage of unaccusatives and passives in pronominal inversions to that in non-pronominal inversions. Although some of the samples are too small to provide reliable percentages, the general trends are clear. At each stage of the language, unaccusative and passive verbs occur in a relatively small percentage of CVSp clauses which, it will be recalled, are merely the result of V-to-C and should take place without respect to the type of verb involved. A larger portion of CVSn clauses, and even more of the CVXSn clauses, contain unaccusative and passive verbs. Furthermore, the percentage of unaccusatives and passives in non-pronominal inversions increases over time. These observations are consistent with the notion that such inversions rely more and more on options other than verb-movement, e.g. the availability of the direct object position as a location for a surface subject. It should be pointed out that the proportion of unaccusatives and passives in CVSp clauses rises as well. These figures are unexpected; if CVSp represents the conservative inversion type, as we have argued, then its character should not change over time. A possible explanation is that the overall number of unaccusatives and passives in the texts increases, and that all categories of inversion reflect that increase. If this account is correct, then Table 7.2 gives evidence of synchronic differences between inversion types, as we have assumed, but not of diachronic change. To verify this possibility we would need figures on the rate of unaccusatives and passives in a general sample from each text, to compare with the rate in inversions. Such figures are available at present only for the Queste, where the overall percentage of unaccusatives and passives is 40% (greater than pronominal inversions but less than non-pronominal inversions). Although more work remains to be done, it seems unlikely that the overall distribution of verb types in French should change over time. It is more plausible that the low number of CVSp clauses in the later texts simply makes the percentages less reliable, and therefore not directly comparable with, that of the Queste, where CVSp accounts for fully half of the inversions. Table 7.3 Percentage of Unaccusative and Passive Verbs in Old and Middle French Inversions CVSp Queste 42 of 208, 20% pp. 1-41, 71-115 Joinville 2 of 24, 8% pp. 104-125 Saintre 5 of 15, 33% pp. 239-294 Commynes 5 of 11, 45% pp. 58-97

CVSn

CVXSn

67 of 137, 49%

76 of 121, 63%6

24 of 43, 56%

3 of 7, 43%

18 of 26, 69%

5 of 6, 83%

39 of 50, 78%

25 of 32, 78%

ON THE DECLINE OF VERB MOVEMENT TO COMP

181

Let us summarize the diachrony of inversion in medieval French as we understand it thus far. In a verb-second language of the German type, in which V-to-C obligatorily applies in matrix clauses and Spec CP is obligatorily filled, inverted clauses have the representation (3), repeated below: (3)

C [ P Longuement [c' resgarda [IP Perceval [I'' t [ VP

l'ome...]]]

OF additionally allows free inversion of the Italian type. In Italian, such clauses have the structure of (12a) (NP-movement) or (12b) (unaccusatives and passives). (We follow here Rizzi (1982); Burzio (1986); and Pollock (1986).) (12) a. [IP pro [r V [VP VP subject]]] b. I[P pro [I'V [VP ...[NP subject]] The verb-second character of OF, however, prevents surface verb-first structures like (12) from occurring; instead, the verb-second constraint and the free inversion possibility work together to produce the hybrid constructions (13a) and (13b), illustrated by (5) and (7) above: (13) a. [CP XP [c' V [IP pro [I' t [VP...] subject]] b. [CP XP [c' V [IP pro [I' t [VP... [NP subject]]. The numbers we have seen in Tables 7.1, 7.2 and 7.3 suggest that the structure in (13) is, however, unstable. Speakers of OF are already forced to posit inversions within IP for sentences like (6): (14) quant [IP pro [I' vint [VP t par la volente" Nostre Seignor [NP Calogrenant]]. If Spec IP is open to non-subjects, then hybrid verb-second clauses like (13) could be reanalyzed as matrix IP's in which free inversion, but not Germanic inversion, has taken place. There is indeed evidence that in the 13th century Spec IP may be an A' position. This evidence comes from a very few embedded clauses, where we can be sure that an initial non-subject is in Spec IP rather than Spec CP: (15) quant en si haute bonte et en si haute chevalerie when in such high goodness and in such high knighthood seroit fichiee la bosne de son lignage (Q 221,13) would-be fixed the end of his lineage 'when the last of his descendents would be fixed in such great goodness and chivalry'. It seems plausible, then, that at least some speakers of late OF interpreted inversions of the type (5a-d) and (7) as IP's rather than CP's. Hence (13a-b) are simplified to (16a-b): (16) a. I[P X P [ I ' V [yp...] subject]] b. I[P XP [I'V [VP... [NP subject]]. Another way of putting this observation is that the data available to speakers does not force the conclusion that V-to-C is obligatory in all matrix clauses.

182

VERB SECOND AND THE NULL-SUBJECT PARAMETER

But if speakers then formulate grammars in which V-to-C no longer applies automatically in every inverted clause, then the number of environments in which postverbal pronoun subjects are possible, is reduced. This state of affairs is what we see reflected in Tables 7.1 and 7.2, where ambiguous nonpronominal inversions—i.e. those that may be generated within IP's as well as CP's—increase at the expense of CVSp, which can be generated only if a CP is formed. Table 7.3 provides additional support for this conclusion by showing that the types of verbs most likely to have VP-final surface subjects, unaccusatives and passives, are disproportionately represented in the non-pronominal inversions of MidF. Roberts (1993, sections 2.3.1 and 2.3.2), in part following Adams (1987a,b), discusses a parallel reanalysis in Medieval French: that of matrix SVO clauses. I will not go into the details of Roberts' account, which suggests a principled interpretation of the way in which structures may be simplified diachronically. I point out, however, that SVO clauses and the free inversions we have described here have in common the property that they have identical surface structures in both matrix and embedded clauses in OF. The only clear evidence for asymmetric verb movement—V-to-C—in OF, therefore, comes from simple V-to-C structures of the type (1) and (2). The ambiguity of both SVO and many inverted clauses in OF may provide an additional clue to the question, addressed by both Roberts and Adams, of why OF loses verb second while the mainland Scandinavian languages, which also have SVO in both matrix and embedded clauses, have retained it. Since free inversion is not available in Scandinavian, all inverted clauses give uncontrovertible evidence for verb second. There is no occasion for the type of instability observed in OF to arise. We return to these issues in SECTION 3, where we examine in more detail the role of the initial constituent in inversion constructions.

3. Initial Non-Subject Constituents We have seen evidence that CVSp, a construction that can only be produced by V-to-C movement, declines dramatically in MidF when compared to other types of inversions. In addition, the types of inversions that come to dominate in the 15th century are particularly amenable to an explanation that does not involve verb movement. We interpret these facts to mean that verb-fronting to Comp is no longer obligatory in declaratives in MidF and that it becomes less and less frequent over time. Furthermore, the decline of V-to-C may have already been in progress at the beginning of the MidF period, triggered by a late OF system in which a large number of inversions, as well as SVO clauses, are ambiguous between a CP and an IP interpretation. We must now address a second question, that of the role of the initial nonsubject constituent. We first see evidence, once again, of synchronic differences among inversions in OF, focussing this time on topicalization constructions. We then move on to MidF to examine the advent of a new option for fronting a non-subject constituent, prepared by the OF system: adjunction

ON THE DECLINE OF VERB MOVEMENT TO COMP

183

to the left of the clause. Finally, we note the complete breakdown of the former verb-second system in late MidF, where inversion may occur independently of the presence of a fronted constituent. Our data in this section are slightly more complete than in SECTION 2 and help confirm the trends noted there.

3.1 Topicalization in 13th-Century Old French 13th-century OF, as represented by the Queste and similar prose texts, is a strict verb-second language in both the descriptive and the theoretical senses of the term. That is, there are very few examples in the texts where the finite verb is in first or third position, and even those can usually be explained as underlyingly consistent with the principles of verb-second syntax. Hence the vast majority of initial non-subject constituents are followed immediately by the finite verb and, optionally, a postverbal subject. The remaining initial constituents, which may be followed by either SVO or CVS order, are for the most part sentential adverbs or adverbial expressions belonging to a restricted class.7 It is interesting to note that membership in this class generally excludes an adverb from participating in inversion at all; it appears rather that each potentially fronted constituent is marked to occupy either Spec CP or a position to the left of CP.8 Table 7.4 shows the types of initial constituents that occur in inverted clauses in two 40-page passages from the Queste del Saint Graal. Percentages are given as decimals. Clearly, the two most well-represented categories are adverbs and adjunct prepositional phrases. Together they account for 75% of the initial constituents in CVSp clauses, for 93% of the CVSn clauses, and for 93% of the CVXS clauses. The initial position in 13th-century inverted clauses is thus not reserved for the topicalization of complements of the verb phrase, although such topicalization is possible, but is more typically the site of adverbial material Table 7.4 Types of Initial Constituent Adverb PP (adjunct) Dir.Obj. Clause Inf Past Participle >1 Constituent Adjective Misc Subcategorized

CVSp 84 (.42) 66 (.33) 23 (.12) 19 (.10) 3 1 3

199

CVS 101 (.73) 28 (.20) 6 (.04) 0 0

CVXS 47 (.68) 17 (.25) 4 (.06) 0 1

3 1 139

69

184

VERB SECOND AND THE NULL-SUBJECT PARAMETER

which may or may not be focussed. Furthermore, when direct objects, past participles, or infinitives are fronted, they are more than twice as likely to occur with postverbal Sp as with postverbal Sn. This situation hints at a contrast between the syntactic behavior of true topics—those constituents fronted from their normal position within the VP—and adjuncts or adverbs which happen to occur clause-initially. The true topics occur preferentially in clauses obligatorily interpreted as CP's, while those clauses open to an IP interpretation host topics more rarely. In fact, the evidence for a split between CVSp and CV(X)Sn clauses with respect to topicalization is more significant than can be seen in Table 7.4. As initial constituents the adverbs si and lors are very frequent: these adverbs may have little semantic content, as example (5a) above shows where the translation "thus" implies no particular causality. But these "meaningless" adverbs are more frequent in the non-pronominal inversions. The overall percentage of si and lors combined within the adverb class for each of our inversion contexts is given in Table 7.5. These two adverbs are each much more frequent than any other single adverb. Here our pronominal examples can be seen to make use of these adverbs approximately half as frequently as the non-pronominal examples. This contrast is even more striking when we consider that a large number of the instances of si in CVSp constructions are in fact topicalization constructions. When si triggers the inversion of a subject pronoun, it usually has a contrastive function linking it to the previous clause:9 (17) ainz vos fust avis, se vos les veissiez, qu'il en rather to-you would-be mind, if you them would-see, that-they 'en' fussent trop lie et si estoient il sanz faille. (Q 25,14) would-be very happy and so were they without fail 'rather you would have thought, if you had seen them, that they were very happy about it, and so they were indeed'. The observations just made complement nicely the evidence that Roberts (1993, section 2.3.1) offers for the reanalysis of matrix SVO clauses as IP's. Pointing out that in OF the expletive subject pronoun il, which cannot be a topic, could occur in matrix SVO constructions, Roberts (who also cites work by Cardinaletti) hypothesizes that speakers interpret the initial position in SVO clauses as an argument position rather than a topic position. They are then led to assume that the subject is in Spec IP rather than Spec CP. Parallel to Table 7.5 Si and Lors Within the Adverb Class CVSp

CVS

CVXS

si lors

17 (.20) 7 (.08)

27 (.27) 24 (.24)

16 (.34) 6 (.13)

Total Adv %si+lors

84 24/84=.29

101 51/101=.50

47 22/47=.47

ON THE DECLINE OF VERB MOVEMENT TO COMP

185

the ambiguity that Roberts describes, then, is an ambiguity in the interpretation of the initial constituent of CVSn clauses. The presence of the "meaningless" adverbs lors and si in a large number of 13th-century non-pronominal inversions, like the presence of expletive il, could lead speakers to assume that the initial position is a place-holder to ensure verb-second order rather than a topicalized position. If Spec CP, but not Spec IP, is the normal position of topicalization, then the reinterpretation of CVSn clauses as IP's is expected.

3.2 Non-inversion After an Initial Non-subject Constituent in Middle French In contrast to the strict verb-second order of OF, MidF has a productive verb-third construction, which we will abbreviate CSV, possible with either pronominal or non-pronominal subjects and in matrix or embedded clauses, and in which the initial constituent is from an increasingly unrestricted class: (18) Lors il retourna et parla a pluseurs chevaliers... (Saintre 102,25) then he returned and spoke to several knights (19) se par le plaisir

de Dieu fortune venoit en vostre ayde, ... (Saintre 75,22) if by the pleasure of God fortune came to your aid.

We saw in the last section that the vast majority of initial non-subject constituents in 13th-century prose are adverbs and adverbial PP's; the first constituent of the new CSV clause carries over this tendency toward nonsubcategorized material. The competition between inversion and non-inversion becomes even clearer when we note that the same lexical items that once required inversion now alternate between inversion and non-inversion (see Vance (1988, section 4.1.2) for further examples and discussion). In some individual cases (e.g. lors) there is already a marked preference, in early MidF, to avoid inversion. Table 7.6 Inversion vs. Non-inversion After Initial Non-subject Constituent CSV % Inv CV(X)S 97% 219 7 Queste 1225 74 30 Joinville 71% 1306 48 38 56% Froissart c. 1375 44 41 52% Quinze Joies 1420 Jehan de Saintrd10 58 40% 39 1456 Commynes 89 91 51% 1491

186

VERB SECOND AND THE NULL-SUBJECT PARAMETER

Table 7.6 above shows the ratio of inversion to non-inversion after an initial constituent for subject pronouns and subject NP's combined. The figures for the Queste del Graal are slightly misleading because they include sentential adverbs from the class we have recognized as unable to trigger inversion. If such adverbs are not included, the figure for inversion after a non-subject constituent is virtually 100%. The text Jehan de Saintrt appears to be slightly anomolous in preferring non-inversion to a greater extent than the other MidF texts. It is clear, however, that at some point between 1225 and 1306 an important change took place in French with respect to word order. Whereas in the 13th-century verb-third orders were essentially unknown, verb-third clauses abound in the 14th and 15th centuries. From the numbers in Table 7.6 it would appear that, after an initial decline, inversion stabilized over the MidF period. A slightly different situation is suggested by Table 7.7, which breaks down the statistics in Table 7.6 according to pronominal and non-pronominal subjects. In this table we observe a steady decline in pronominal inversion after 1225, from nearly 100% to just 15% over a century and a half. The NP subjects, on the other hand, decline initially and then vacillate between 50% and 73% in MidF. The fluctuation of the percentage of CVSn clauses, I claim, reflects the fact that only a subset of such clauses are directly affected by the more abstract grammatical change in progress. Another way of looking at this is the following: because both CSnV clauses and many of the CVSn clauses may be interpreted as IP's, the contrast between them is in part superficial. The contrast between CVSp and CSpV, on the other hand, is real, since if V-to-C fails to apply in the presence of an initial constituent only the latter is possible. Table 7.7 Inversion vs. Non-inversion of Pronominal and Non-pronominal Subjects After XP

Queste 1225 Joinville 1306 Froissart c. 1375 Quinze Joies 1420 Jehan de Saintre 1456 Commynes 1491

CVSp

CSpV

% inv

CVSn

CSnV

%Inv

97

3

97%

122

4

98%

24

17

59%

50

13

79%

15

26

37%

33

12

73%

17

28

38%

27

13

68%

9

28

24%

30

30

50%

11

60

15%

80

29

73%

ON THE DECLINE OF VERB MOVEMENT TO COMP

187

Kroch (1989) discusses, among other examples, the competition at issue here, i.e. between inversion and non-inversion after an initial constituent in the history of French. His figures, drawn from Fontaine (1985), are detailed only from 1500 onward but show a continual decline in inversion for both pronominal and non-pronominal subjects. Furthermore, Kroch shows that the slopes of the two declines, as calculated by the logistic transform of frequency, are identical not only to each other but to the slopes of two other declines presumed to be related; the loss of pro (vs. overt subject pronouns) and the loss of inversion after a direct object (vs. the "reprise" construction discussed in Priestley (1955)). These results have great potential to clarify our understanding of diachronic change; when added to similar evidence from syntactic change in other languages, they lead Kroch to formulate the "constant rate hypothesis": the various surface reflexes of a larger syntactic change in progress all change at the same rate, despite varying degrees of preference for the new form over the old at any given time. Although much further research is required in order to take full advantage of the implications of Kroch's study, I presume that the basic reason for the lack of a parallel decline between Sp and Sn in Table 7.7 is, as I have claimed, that the figures for Sn are influenced, at least in the time period in question, by other factors than the loss of V-to-C. These factors may turn out to be insignificant in the overall decline of CVSn in matrix clauses over a longer period of time and as seen through sophisticated statistical methods. What seems to be reflected here is that MidF appears to be a verb-second language only because the rise of "free inversion" masks its decline.11 We will look more closely at the internal structure of MidF CVSn clauses in the next section. It is not entirely clear how the competition between CVS and CSV began. At what point does the decline in V-to-C begin to produce evidence against verb second rather than simply ambiguous structures, and how do these verbthird orders arise? It has frequently been claimed that the cliticization of preverbal subject pronouns to the verb produced apparent verb-third clauses within the confines of a verb-second grammar, and that this word order then spread to non-pronominal subjects (cf. Zwanenburg (1978); Adams (1987b); Hulk and van Kemenade (this volume); Platzack (this volume)). According to this account, then, the appearance of CSV clauses marks the beginning point of the decline of V-to-C rather than a step along the way. Vance (1988) argues against this account on the basis of evidence that preverbal subject pronouns are still full NP's at the syntactic level in MidF. The cliticization that becomes apparent in MidF is phonological and therefore can occur only after the syntax produces CSV order.12 Furthermore, the late OF and early MidF texts I have surveyed informally reveal no evidence of a stage in which CSpV, but not CSnV, was possible. The first texts to show significant numbers of CSpV clauses also have CSnV, as shown by the figures for Joinville in Table 7.7. While it is true that CSnV occurs less frequently than CSpV in that text, this difference in percentages is compatible with the account proposed here, namely that verb movement to C, already in decline, affects pronominal subjects more directly than non-pronominal ones. It is thus difficult to support

188

VERB SECOND AND THE NULL-SUBJECT PARAMETER

empirically the hypothesis that an early wave of preverbal subject pronoun cliticization was the origin of the order CSV.13 An appealing approach, although one not without problems, is that of Roberts (1993, section 2.3.1). Suppose that adjunction to IP was theoretically available all along but failed to occur in OF because all matrix clauses were CP's. Then the emergence of CSV orders in early MidF can be attributed to the earlier reanalysis of matrix SVO clauses as IP's. If such were the case, however, we would expect to find evidence of adjunction to IP in embedded clauses in OF, as in Modern German (cf. Schwartz and Vikner (1989)). This prediction is not fulfilled; the only adjunctions possible in 13th-century OF are those of the limited adverbial class discussed above, and these are variably to CP or IP. It is possible, of course, that adjunction to IP is admissible in embedded clauses in OF but does not occur in the narrative contexts we are considering. The prevalence of the construction in MidF narrative, however, as illustrated in (19) above, suggests that this gap is not an accident of the data. Furthermore, (as Roberts also points out), a few of the adjunctions found in MidF are to CVSp clauses and thus cannot be to IP.14 A close look at the data, then, suggests that the leftward adjunction that characterizes MidF is adjunction to any clause, rather than specifically to IP. What changes over time is not the type of adjunction but the range of items that may participate in the adjunction, which increases greatly in MidF. This characterization of the phenomenon at issue nevertheless leaves unresolved the question of the origin of the change. A partial solution may be found in the evolution of the rhythmic structure of French, as discussed in Kroch (1989) (see also Adams (1987c)). As the phrasal stress shifts to the end of the clause, perhaps in conjunction with the rise of the free inversion constructions we have been studying, it becomes more difficult to prepose long, complex constituents to Spec CP. In the Queste, entire clauses may trigger pronominal inversion: (20) Et por ce que tu n'ailles sels, voil ge que tu meines and for this that you neg-go alone want I that you take o toi Perceval et Boorz. (Q 271,13) with you Perceval and Boorz 'and so that you don't go alone I want you to take Perceval and Boorz'. Over the MidF period, however, pronominal inversion triggered by a clause becomes extremely rare. Fronted clauses may, instead, precede SVO clauses, inverted clauses introduced by si, and non-pronominal inversions; in each of these environments, movement of the clause to Spec CP is either excluded or unnecessary. Moreover, despite examples such as (20), even in the 13th century the majority of proposed clauses—and especially those introduced by quant— fail to trigger pronominal inversion. Such clauses have not, in fact, been counted as constituents at all in the statistics presented herein. This early tendency to detach fronted embedded clauses from the following main clause may reflect the creation of a new stressed position to the left of the clause which eventually draws shorter fronted constituents as well.

ON THE DECLINE OF VERB MOVEMENT TO COMP

189

To account completely for CSV clauses, however, we must allow as well for a parallel tendency, from the earliest MidF texts on, for certain common short adverbs such as lors to occur in an adjoined position rather than in Spec CP. It is possible that the decline of V-to-C in inversion directly forces CSV structures into existence. If V-to-C fails to apply, then Spec CP is no longer available as a landing site for an initial topicalized constituent. But Spec IP is also unavailable, if our earlier hypothesis that it accepts only non-topicalized constituents is correct. Under these circumstances, the only recourse is to a stressed, adjoined position on the left of the subject in Spec IP. In support of this idea, I note that the adverbs of early CSpV clauses carried the same topicalized information as the competing CVSp. I have not yet studied the semantic behavior of the adverbs paired with NP subjects. It is likely that some combination of all three proposed explanations will provide the correct analysis of the rise of verb-third clauses. What is essential to our analysis is that the advent of CSV was prepared by earlier changes in the language, either purely syntactic or both syntactic and rhythmic. The appearance of CSV is one of the intermediate steps in the decline of V-to-C and in no way marks the beginning point of the loss of verb second.

3.3 Verb-Initial Inversions We have seen that whereas in OF the initial constituent occupies a position essential to the syntax of the verb-second clause, in MidF this element may be dissociated from inversion, occurring in an adjoined position that has no influence on the syntax of the clause and is fully omissible. This dissociation results not only in the CSV clauses we saw above, but also in some new inversion constructions characterized by apparent verb-first order. These are illustrated in (21) and (22): (21) et

luy

fut adoubee sa playe qu'il

avoit au col. (Commynes, 63,20) and to-him was dressed his wound that-he had in-the neck

(22) Fut le duc de Brunsvich pour 1'empereur (Saintre, 207,28) was the duke of Brunsvich for the-emperor 'The duke of Brunswick was there for the emperor'. The rise in these types of structures is in part possible because of a change in the licensing of pro in MidF discussed in Vance (1988, 1993), Roberts (1993), and others. There is independent evidence that pro occurs in preverbal position in MidF, for example in initial position in embedded clauses: (23) et

que pro voyez (Saintre, 246,24) of-the surplus such as pro see-2pl.

grant partie du

and great part

surplus tel

Presumably, the clauses in (21) and (22) also have preverbal null subjects which occupy the first position in the clause, Spec IP. A number of cases of

190

VERB SECOND AND THE NULL-SUBJECT PARAMETER

apparently verb-second inversion in MidF may then in fact have an adjoined initial constituent and a preverbal pro: (24) Le lendemain, a 1'aube du jour, passerent les ducs de the following-day at the-break of-the day passed the dukes of Berry et de Bretagne... (Commynes, 74,3) Berry and of Bretagne (25) Des

qu'il fut

dresse, vint

un officier d'armes du roy (Commynes 88,13) as-soon as-it was raised, came an officer of-arms of-the king

It is highly unlikely that these are clauses in which V-to-C has applied, even though they have exact counterparts in OF. First, recall that our statistics show that unaccusative verbs such as passer and venir are typical of such clauses in MidF and that the postverbal subjects may in fact be underlying direct objects that have not moved from their base positions. Second, pronominal inversion after a clause is to my knowledge not found in MidF; this fact implies that an IP structure is at issue. Finally, such clauses have exact counterparts in formal varieties of Modern French. The following examples were taken from contemporary scholarly writing. (26) is a verb-initial inversion with an unaccusative verb, and (27) shows inversion with an unaccusative reflexive with the finite verb in second position. The VP-final subjects are italicized. (26) 'N'entraient, naturellement, en ligne de compte ni les paysans et leur patois, ni la plebe des faubourgs,' 'Neither the peasants and their dialects nor the plebeians of the suburbs entered into consideration' (27) Depuis la deuxieme guerre mondiale, s'est accentuee une tendance, qui s'esquissait deja dans les annees trente, a la confusion des deux a au profit du /a/ d'avant. 'Since the second world war, there has increased a tendency, which was already developing in the 30's, toward the confusion of the two a's to the advantage of the front /a/.' (Andre Martinet, "La variete des usages dans la phonie du francais") One further type of clause supports our claim that initial constituents may be adjoined to IP in inverted as well as SVO clauses in MidF: (28) toutefois nevertheless este caches been hidden

il revenoit beaucoup de gens, qui avoient there returned many of people who had es bois (Commynes 65,10) in-the woods

The expletive pronoun il is clearly not required, since its absence would simply result in structures of the type in (24) and (25). The fact that it is sporadically inserted suggests that speakers of late MidF conceive of such clauses

ON THE DECLINE OF VERB MOVEMENT TO COMP

191

as having an expletive subject and an optional preverbal constituent rather than as having a verb-second construction. Clearly, the grammar of V-to-C is in full decline.

4. Conclusion The data presented here show that a large portion of the inversion structures that occur in the MidF period can be attributed to a non-verb-second grammar. The crucial clause-type is CV(X)Sn, a hybrid "verb second free inversion" construction, which is consistent with both a grammar in which obligatory V-to-C applies and with one in which it does not. My claim is that the rise in this construction over the course of the 14th and 15th centuries reflects the growing generalization of alternative means of producing inversion structures, to the detriment of CVSp, produced strictly under V-to-C movement. We have seen, furthermore, that the initial constituent of the clause is gradually reanalyzed as being unrelated to inversion. Both changes in the grammar of French had their origins at least as early as the 13th century, which is not true of the cliticization of preverbal subject pronouns, often claimed to be a source of the loss of verb second. This shift in the proportion of pronominal and non-pronominal inversions in MidF has the effect of masking, on the surface, the fundamental decline in verb second that is taking place. Insofar as it views inversion structures as a source of grammatical reanalysis (of CP's as IP's) in the history of French, my account is partially similar to that of Hulk and van Kemenade (this volume). It also resembles in part that of Lemieux and Dupuis (this volume), who see inversion structures as IP's even in OF. Both these studies, however, consider IP inversions to be true verb-second constructions. Under my account, the types of inverted word orders produced in IP's and those produced in CP's are fundamentally different. Specifically, CP inversions may—and, in the case of subject pronouns, must—have immediately postverbal subjects. They are of course restricted to matrix clauses, and they have an obligatory preverbal non-subject constituent. IP inversions, on the other hand, generally have VP-final subjects, strictly exclude pronouns, and may occur in embedded contexts. Since they are essentially "free inversions" and not verb-second structures, they may in fact be verb-initial in MidF. It is the coexistence of these two underlyingly different inversion constructions, I claim, that sets in motion the decline of verb movement to Comp in the history of French. The notions of language change that I have relied upon here are quite general and should be applicable to other cases than the history of French. Specifically, my analysis makes a broad prediction which requires further refinement: that SVO languages in which free inversion and verb-to-Comp inversion coexist, should develop in the same direction as French, i.e. toward the elimination of verb second characteristics. Many Northern Italian dialects seem to have followed a similar progression (see for example Vanelli, Renzi, and Beninca (1985) and Poletto (this volume)). Old Spanish offers another

192

VERB SECOND AND THE NULL-SUBJECT PARAMETER

example of a language which lost verb movement to Comp in declaratives (Fontana, in preparation) and retained free inversion. To my knowledge, the only medieval verb-second Romance language that has not changed in this way is Swiss Rhaeto-Romance, which retains verb second. Research in progress based on Sprouse and Vance (1993) is devoted to the isolation of the factors that may separate this case from that of related languages.15

Notes *

I am grateful to Deborah Arteaga, Anthony Kroch, Ian Roberts and Rex A. Sprouse for helpful discussion and to the following contributors to this volume who shared drafts of their papers with me: Paola Beninca, Paul Hirschbuhler, Aafke Hulk and Ans van Kemenade, Monique Lemieux and Fernande Dupuis, David Lightfoot, and Christer Platzack. I would also like to thank Ian Roberts for access to the manuscript of his forthcoming book, which has now appeared as Roberts (1993).

1. For the sake of simplicity, I assume here that the base position of the subject is Spec IP. A great deal of evidence exists, however, to place that position within the VP or Vmax projection (see e.g. Zagona (1982) and Koopman and Sportiche (1986, 1991)). The circumstances under which the subject moves to the Spec of IP (or analogous position), presumably to acquire Nominative Case, are understood differently by different authors. If we assume, as I have here, that the subject normally occurs in Spec IP at S-structure even if it originates in the VP, we predict that Old French does not allow verb-second inversions in embedded clauses. If, on the other hand, OF allows the subject to remain in a position to the left of the verb in VP, as claimed by Adams (1987c), Dupuis (1989), and Lemieux and Dupuis (this volume), it is predicted that OF is a symmetric verb-second language allowing inversions in both main and embedded contexts. This matter is an empirical question of some importance. Although early Old French may have had a symmetric verb-second system, I interpret prose texts from the 13th century on to show little if any evidence of embedded clause inversion other than free inversion (discussed below) and the type of inversion noted in note 2. On the other hand, an analysis in which the VP-internal position of the subject is to the right of V in OF, as in Roberts (1993), is compatible with the present account since it predicts that embedded clause postverbal subjects will be VP-final rather than immediately postverbal. 2. The occurrence of inverted clauses after the conjunction que is widely recognized to be due to the embedding of an entire root CP under a second CP headed by a complementizer, a phenomenon found in many verb-second languages (cf. Adams (1987a,b); Platzack (1983); Sigurosson (1990); Vance (1988)). Such inversions are considered to be root phenomena and will not be treated separately from matrix clauses in this paper.

ON THE DECLINE OF VERB MOVEMENT TO COMP

193

3. The most salient of the restrictions on "free" inversion in Old French is on non-pronominal direct objects, which may not separate the finite verb from the subject in simple tenses or the past participle from the subject in complex tenses. Bare NP objects (those in which no article is present) occurring with simple tenses, however, do not fall under this restriction. For an approach to similar facts in Italian, see Rizzi (1991), who proposes a linear adjacency condition (for purposes of Nominative Case assignment) between the subject and the lowest functional head above VP, i.e. the head of the tense projection in simple tenses or the participial morphology in complex tenses. To implement Rizzi's suggestion for Old French would require a more detailed articulation of functional projections than I have assumed here. 4. These figures do not include Middle French inversions of the type discussed in SECTION 3.2.2, which are not accompanied by a preverbal non-subject constituent. 5. It is difficult to find syntactic evidence that speakers of Old and Middle French were forced to interpret the position of postverbal subjects of passives and unaccusatives differently from the postverbal NP subjects of other verbs. In most cases, the surface position of the NP's is the same in both types of constructions, and the postverbal subjects of unaccusatives and passives are usually amenable to the focussed interpretation typical also of moved NP's. Whether unaccusatives and passives remain in direct object position or move to a common postverbal subject position is irrelevant to our analysis, except in the case of the split passive verbs in (10). A 3-verb passive from the Queste provides interesting support for the syntactic nature of unaccusative and passive "inversion": (i)

(il troveroitque) au jor d'ui doit estre cist sieges aempliz (Q 4,15) (he would-find that) today must be this seat filled.

Cist sieges is not VP-final, nor can it be in Spec IP, since it occurs after the infinitive. It can, however, be analyzed as an underlying direct object which precedes its past participle exactly as in (10). If this is the only interpretation of (i) available to speakers of OF, then it seems that unaccusatives and passives must play a syntactic role distinct from that of other verbs. This in turn lends extra support to our analysis of (10) as ambiguous between a reading where the subject is within the VP and one in which it occupies Spec IP like ordinary subjects. On the other hand, this example may reflect the sort of movement of nonfinite verbs to Spec VP proposed by D6prez (1989), where cist sieges would be base-generated in the VP of aemplir and where estre moves to the specifier of that VP. Such movement would not depend on the type of the verb. Although Ddprez's analysis appears in most instances to introduce unnecessary complexities into the grammar, there are two examples in the Queste of a con-struction that resists most other interpretations. One of them is given in (ii): (ii) bien nos a tenu ceste damoisele convenant (Q 275,20) well to-us has kept this maiden covenant

194

VERB SECOND AND THE NULL-SUBJECT PARAMETER

As in (i), the subject ceste damoisele is neither VP-final nor immediately to the right of the finite verb, but here the unaccusative/passive explanation is unavailable. Either the past participle tenu has moved to the left and the subject is within the VP of tenir, as Deprez's account would have it, or both the subject and the object convenant have been moved, the latter even further to the right than the former. A completely satisfactory explanation will have to account additionally for the fact that the subject and object end up in the order found obligatorily in simple V-to-C clauses such as (2); i.e. subject first, object second. In this respect, at least, the approach of Rizzi (1991) mentioned in note 3 offers a promising area for further research. 6. This figure is based on all the CVXS clauses in the text; all others figures are based only on the passages indicated. 7. Those found to occur regularly in the Queste del Saint Graal are: par mon chief'by my head', sanz faille 'without fail', certes 'certainly', neporquant or neporec 'nevertheless', ja (mes) or onques 'never'. 8. A regular exception is the case ofja mes and onques, which trigger omission, if not clearly inversion, of the subject pronoun. There are also a few random examples in the Queste where an adverb that generally triggers inversion fails to do so. 9. For further discussion see Vance (1981) and the summary thereof in Vance (1988:178ff). The basic observations have been made repeatedly in the philological literature; cf. Price (1966); Offord (1971), and references noted there. 10. The count for Jehan de Saintre in Tables 7.6 and 7.7 is based on a different sample than for the same text in Table 7.1. 11. Note that in Old French the situation is reversed: cf. Roberts (1993, section 2.1.2), who suggests that free inversion is not as prevalent in Old French as in Modern Italian because it is "frequently disguised by verb second" in the former. 12. The discussion in Vance (1988:197ff.) is based largely on Kayne (1983), Rizzi (1986), and Burzio (1986), who consider (contra Safir (1985) and others) that preverbal subject "clitics" even in Modern French are full NP's at the syntactic level, cliticizing only at PF. See also Roberts (1993, sections 2.2.2 and 2.3.1) for further discussion. In contrast, postverbal subject pronouns, as we have argued above, are syntactic clitics in Old French as well as Modern French. An anonymous reviewer points out that later generations of speakers may have reinterpreted phonological cliticization as syntactic. Work by Lambrecht (1981) and Auger (1992), among others, indicates that syntactic cliticization of subject pronouns is in progress in Modern Popular French. We have no evidence, however, that this change began as early as Middle French, since it is still possible at that time to separate the atonic preverbal subject pronoun from the verb by non-clitic material. Alternatively, it is possible that Modern Popular French is descended from a dialect significantly different from that of our medieval texts; however, there appears to be no way to determine how subject pronouns might have behaved in that dialect.

ON THE DECLINE OF VERB MOVEMENT TO COMP

195

13. It should be pointed out here that there is a slight preference in Middle French for CSpV over CSnV when the initial constituent is a direct object. Proponents of the cliticization account interpret this fact as evidence for their position, supposing that true VP complements, as opposed to adverbials, must appear in Spec CP and thus can only co-occur with preverbal subjects if the latter are clitics. The evidence in question, however, is not available prior to the decline observed in Table 7.7 and so cannot be used to argue that cliticization is a source of that decline. The discussion in Vance (1988) to the effect that preverbal subject pronouns could be phonological but not syntactic clitics in Middle French is consistent with the data examined here as well as with the main thrust of the account in Kroch (1989), where rhythmic factors are held to be ultimately responsible for the loss of verb second. Furthermore, Kroch's account of the rise of the "reprise" construction in French makes the importance of the behavior of objects in CSV constructions questionable; both OVS and OSV are being replaced by a new structure that makes use of left-dislocation and a resumptive pronoun. 14. In fact, SVO clauses may have been IP's in Old French as well as in Middle French. If the grammar of the texts is representative, there are no data available to speakers of Old French that would force a CP interpretation of SVO clauses. Under these circumstances, given the principle of economy of Chomsky (1991) and Roberts' (1993, section 2.3.2) formal interpretation of it, we would expect that the least complex possible structure, here IP, would be posited from the start. I will not explore here the consequences of such an account, which also forms part of the analysis suggested in Lemieux and Dupuis (this volume), but the reader is referred to Travis (1984) and Zwart (1991) for broader theoretical discussions of this view of verb second. 15. The preliminary results of this research, which require verification, indicate that in at least one dialect of Rhaeto-Romance the status of free inversion is different from that of French and Italian. From the earliest texts available (16th century) to the present day, the Surselvan dialect appears to have VPfinal inversion only with unaccusative and passive verbs; even French-style stylistic inversion with other intransitive verbs is prohibited. Such inversion, however, has been observed in the Engadine dialects.

References Primary Adams, M. (1987a) "From Old French to the Theory of Prodrop." Natural Language and Linguistic Theory 5:1-32. Adams, M. (1987b) Old French, Null Subjects, and Verb Second Phenomena. PhD Dissertation, UCLA. Adams, M. (1988) "Embedded pro." In J. Blevins and J. Carter, eds. Proceedings ofNELS 18, 1-21. Amherst, Mass.: GLSA, University of Massachusetts.

196

VERB SECOND AND THE NULL-SUBJECT PARAMETER

Auger, Julie (1992) Pronominal Clitics vs. Agreement Affixes: On the Verb Morphology of Colloquial Quebec French. Dissertation proposal, University of Pennsylvania. Besten, H. den (1983) "On the Interaction of Root Transformations and Lexical Deletive Rules." In W. Abraham, ed. On the Formal Syntax of the Westgermania, 47-131. Amsterdam: John Benjamins. Besten, H. den (1985) "The Ergative Hypothesis and Free Word Order in Dutch and German." In J. Toman, ed. Studies in German Grammar, 23-64. Dordrecht: Foris. Burzio, L. (1986) Italian Syntax: A GB Approach. Dordrecht: Reidel. Calabrese, A. (1992) "Some Remarks on Focus and Logical Structures in Italian." Harvard Working Papers in Linguistics 91-127. Chomsky, N. (1991) "Some Notes on Economy of Derivations and Representations." In R. Friedin, ed. Principles and Parameters of Comparative Grammar, 417-454. Cambridge, Mass.: MIT Press. Contreras, H. (1987) "Small Clauses in Spanish and English." Natural Language and Linguistic Theory 5:225-243. Deprez, V. (1989) "Sujets inverses dans Vigneulles." Ms. MIT and Universite d'Ottawa. Dupuis, F. (1989) L'expression du sujet dans les propositions subordonnees en ancien francais. PhD Dissertation, Universite de Montreal. Fontaine, C. (1985) Application de methodes quantitatives en diachronie: I'inversion du sujet enfrancais. MA Thesis, UQAM. Fontana, J.-M. (1993) Phrase Structure and the Syntax of Clitics in the History of Spanish. PhD Dissertation, University of Pennsylvania. Foulet, L. (1930) Petite syntaxe de l'ancien francais. 3rd ed. Paris: Honore Champion. Giorgi, A. and G. Longobardi (1991) The Syntax of Noun Phrases: Configuration, Parameters and Empty Categories. Cambridge: Cambridge University Press. Groos, A. and R. Bok-Bennema (1986) "The Structure of the Sentence in Spanish." In I. Bordelois, ed. Generative Studies in Spanish Syntax, 6780. Dordrecht: Foris. Hirschbuhler, P. (this volume) "Null Subjects in Verb-First Embedded Clauses in Phillippe de Vigneulles' Cent Nouvelles Nouvelles." Hulk, A. and A. van Kemenade (this volume) "Verb Second, Pro-drop, Functional Projections and Language Change." Kayne, R. (1983) "Chains, Categories External to S, and French Complex Inversion." Natural Language and Linguistic Theory 1:107-139. Kayne, R. and J.-Y. Pollock (1978) "Stylistic Inversion, Successive Cyclicity, and Move NP in French." Linguistic Inquiry 9:595-621. Koopman, H. and D. Sportiche (1986) "A Note on Long Extraction in Vata and the ECP." Natural Language and Linguistic Theory 4:357-373. Koopman, H. and D. Sportiche (1991) "The Position of Subjects." Lingua 85:211-258.

ON THE DECLINE OF VERB MOVEMENT TO COMP

197

Kroch, A. (1989) "Reflexes of Grammar in Patterns of Language Change." Language Variation and Change 1.3:199-244. Lambrecht, K. (1981) Topic, Anti-Topic, and Verb Agreement in Non-Standard French. Pragmatics and Beyond II:6. Amsterdam: John Benjamins. Lemieux, M. (1991) "Le role de el dans les constructions a sujet nul et a sujet postpose" en moyen francais." Talk given at 7th International Colloquium on Middle French, Ghent. Lemieux, M. and F. Dupuis (this volume) "The Locus of Verb-Movement in Non-Asymmetric Verb-Second Languages: The Case of Middle French" Lightfoot, D. (this volume) "Why UG Needs a Learning Theory: Triggering Verb-Movement." Maling, J. (1980) "Inversion in Embedded Clauses in Modern Icelandic." Islenskt Mal 2:175-193. Reprinted in J. Maling and A. Zaenen, eds. Modern Icelandic Syntax, Syntax and Semantics 24:71-91. San Diego: Academic Press. Maling, J. and A. Zaenen, eds. (1990) Modern Icelandic Syntax. Syntax and Semantics 24. San Diego: Academic Press. Marchello-Nizia, C. (1979) Histoire de la langue francaise aux XlVe et XVe siecles. Paris: Bordas. Marchello-Nizia, C. (1985) Dire le vrai: I'adverbe "si" enfrancais medieval. Geneve: Droz. Martin, Robert (1980) "De 1'ordre des mots dans le Jehan de Samtre." In M. Wilmet, ed. Semantique lexicale et semantique grammaticale en moyen francais, 305-336. Brussels: VUB Centrum voor Taal- en Literatuurwetenschap. Offord, M. (1971) "The Use of the Personal Pronoun Subject in Post-Position in 14th Century French." Romania 92:37-64, 200-245. Platzack, C. (1983) "Germanic Word Order and the Comp/Infl Parameter." Working Papers in Scandinavian Syntax, 2. University of Trondheim. Platzack, C. (1987) "The Scandinavian Languages and the Null Subject Parameter." Natural Language and Linguistic Theory 5:377-402. Platzack, C. (this volume) "The Loss of Verb Second in French and English." Poletto, C. (this volume) "The Diachronic Development of Subject Clitics in North Eastern Italian Dialects." Pollock, J.-Y. (1986) "Sur la syntaxe de en et le parametre du sujet nul." In M. Ronat and D. Couquaux, ed. La Grammaire Modulaire, 211-246. Paris: Editions du Minuit. Pollock, J.-Y. (1989) "Verb-Movement, UG, and the Structure of IP." Linguistic Inquiry 20:365-424. Price, G. (1966) "Contribution a I'etude de la syntaxe des pronoms personnels sujets en ancien francais." Romania 87:176-504. Priestley, L. (1955) "Reprise Constructions in French." Archivum Linguisticum 7.1:1-28. Rizzi, L. (1982) Issues in Italian Syntax. Dordrecht: Foris. Rizzi, L. (1986) "On the Status of Subject Clitics in Romance." In O. Jaeggli and C. Silva-Carvalan, eds. Studies in Romance Linguistics, 391-419. Dordrecht: Foris.

198

VERB SECOND AND THE NULL-SUBJECT PARAMETER

Rizzi, L. (1991) "Residual Verb-Second and the Wh-Criterion." Technical Reports in Formal and Computational Linguistics, 2. Universite de Geneve. Rizzi, L. and I. Roberts (1989) "Complex Inversion in French." Probus 1:1-30. Roberts, I. (1992) Verbs and Diachronic Syntax. Dordrecht: Kluwer. Saccon, G. (1993) Postverbal Subjects: a Study Based on Italian and its Dialects. PhD Dissertation, Harvard University. Safir, K. (1986) "Subject Clitics and the Nom-Drop Parameter." In H. Borer, ed. The Syntax of Pronominal Clitics, 333-365. San Diego: Academic Press. Santorini, B. (1989) The Generalization of the Verb-Second Constraint in the History of Yiddish. PhD Dissertation, University of Pennsylvania. Schwartz, B.D. and S. Vikner (1989) "All Verb-Second Clauses are CP's." Working Papers in Scandinavian Syntax 43:27-49. Sigurosson, H.A. (1990) "V1 Declaratives and Verb-Raising in Icelandic." In J. Maling and A Zaenen, ed. Modern Icelandic Syntax. Syntax and Semantics 24:41-69. San Diego: Academic Press. Sprouse, R.A. and B. Vance (1993) "Null Subjects in Surselvan." Paper Presented at the Annual Meeting of the Linguistic Society of America, January 8. Tomaselli, A. (1989) La sintassi del verbo finito nelle lingue germaniche. PhD Dissertation, University of Pavia. Travis, L. (1984) Parameters and Effects of Word Order Variation. PhD Dissertation, MIT. Vance, B. (1981) A Syntactic and Semantic Study of Subject Pronoun Usage in Old French. MA Thesis, Cornell University. Vance, B. (1988) Null Subjects and Syntactic Change in Medieval French. PhD Dissertation, Cornell University. Vanelli, L., L. Renzi, and P. Beninca (1985) "Typologie des pronoms sujets dans les langues romanes." Actes du XVHe Congres International de Linguistique et Philologie Romanes 3:164-176. Zagona, Karen (1982) Government and Proper Government of Verbal Projections. PhD Dissertation, University of Washington. Zwanenburg, W. (1978) "L'ordre des mots en francais medieval." In R. Martin, ed. Etudes de syntaxe du moyen francais, 153-171. Paris: Klincksieck. Zwart, C.J.W. (1991) "Clitics in Dutch: Evidence for the Position of Infl." Groninger Arbeiten zur Germanistischen Linguistik 33:71-92. Editions La Queste del Saint Graal. Ed. A. Pauphilet. Classiques Francais du Moyen Age. Paris: Champion, 1984. (c. 1225) Joinville, Jehan de. La Vie de Saint Louis. Ed. Noel Corbett. Sherbrooke: Editions Naaman, 1977. (c. 1306)

ON THE DECLINE OF VERB MOVEMENT TO COMP

199

Froissart, Jean. Chroniques. Ed. Leon Mirot. Societe de 1'histoire de France, tome 12, 1931. (1375-1400) Les Quinze Joies de Manage. Ed. Jean Rychner, Textes Litte'raires Francais. Geneve: Librairie Droz, 1967. (c. 1420) de la Sale, Antoine. Jehan de Saintre. Ed. Jean Misrahi, Textes Litte'raires Francais. Geneve: Librairie Droz, 1978. (c. 1456) Commynes, Philippe de. Memoires sur Louis XL Ed. Jean Dufournet. Paris: Gallimard, 1979. (c. 1491)

8 The Loss of Verb Second In English and French Christer Platzack University of Lund

1. Introduction In this paper I will attempt to give a unified explanation of the loss of verb second in English and French during the Middle Ages. Prior to the loss of verb second, both languages had developed a grammar with verb second SVO, and subject clitics; in such a grammar, there are very few signs which unambiguously indicate verb second. The hypothesis I will defend in this paper is that the existence of such a stage triggered the loss of verb second, and that certain differences between Modern French and Modern English resulted from the loss of verb second in combination with the fact that Infl was "weak" in English but not in French at the time of the word order change. Scholars studying the diachronic development of English and French seem to agree that the medieval stages of these languages displayed verb-second phenomena which were lost in the 14th and 15th centuries, whereas they may have different opinions concerning the reason why verb second was lost. Consider e.g. van Kemenade (1987) and Weerman (1989) for English, and Adams (1987) and Vance (1989) for French. To my knowledge there is no attempt to search for a common reason why both English and French lost verb second.1 Naturally there does not have to be a common reason; however, since verb second seems to be quite a stable phenomenon once it has entered a language (as evidenced by the bulk of the Germanic languages, both in the OV and the VO variants); it appears to be a good research strategy to try to find a unified explanation before attempting language particular solutions. My discussion of the loss of verb second in English and French will be conducted in the light of ideas which have developed in close contact between me and Anders Holmberg during the last couple of years. Different versions of this concept have been presented in Holmberg and Platzack (1988), Platzack 200

THE LOSS OF VERB SECOND IN ENGLISH AND FRENCH

201

and Holmberg (1989) and Holmberg and Platzack (forthcoming); unless there is specific reason to refer to the earlier papers, I will confine myself to refer to the last mentioned work, which contains the most recent and explicit presentation of our ideas. From the point of view of this paper, the main hypothesis developed in the works referred to in the last paragraph is that verb-second languages are languages where the tense affix and the finiteness feature [+F] are in different functional heads (I° and C°, respectively), whereas non-verb-second languages have tense and [+F] in the same head, I°. In terms of this description, the loss of verb second in English and French is a change of position for [+F], from C° to I°. Assuming that the head where the finiteness feature is realized must be lexicalized in order to license Nominative Case, this description accounts for the need to move the tensed verb to C° in main clauses of verbsecond languages, producing the verb-second effect, and the need to lexicalize I° in non-verb-second languages (cf. Pollock (1989)). In subordinate clauses of verb-second languages, movement of the tensed verb to C° is blocked by the presence of a complementizer, leading to a word order difference between main clauses and subordinate clauses.2 No such difference is predicted for non-verb-second languages, where the presence of a complementizer in C° does not interfere with the lexicalization of the head carrying [+F].3 My paper is organized in the following way. In SECTION 2 some of the leading ideas of Holmberg and Platzack are presented, followed by a survey of recent attempts to explain the loss of verb second in English and French. SECTION 3 is devoted to a presentation of the present description, whereas SECTION 4 contains a discussion of some consequences of the proposed explanation. SECTION 5 is the summary.

2. Background 2.1 The Theoretical Framework According to Holmberg and Platzack, verb-second languages differ from non-verb-second languages with respect to where the concept finiteness is represented. Finiteness is a category related to tense and mood; the function of the finite form of the verb seems to be to indicate the existence of a predication at the time of the utterance, disregarding whether or not this predication is accidental or permanent. Thus, in a way, finiteness is a prerequisite for tense and mood: unless a predication is related to the time of the utterance via the concept finiteness, we have no basis for expressing the relative position in time of the situation expressed by the predication vis-a-vis the utterance, and we cannot relate the attitude of the speaker to this situation. It is to be noticed that a finite verb is not identical with a tensed verb: finite verb forms may have tense or lack it (the imperative is a finite verb form without tense, at least in some languages), and the same applies to non-finite verb forms (participles are present or past, for instance).

202

VERB SECOND AND THE NULL-SUBJECT PARAMETER

Like Holmberg and Platzack, I will express the category finiteness as a feature [F], realized in some functional category. There are two values of this feature: [+F], which is used to express finite categories, and [-F], which is used to express non-finite categories. A unitary verb-second parameter may now be formalized as in (1), expressing the hypothesis that the feature [+F] is in C° in verb-second languages, in I° in non-verb-second languages: (1)

The Verb-Second Parameter ±([+F] is located in C°)

According to the principle of full interpretation, every element in a structure must be licensed. Since the finiteness feature has the function of indicating presence of a predication, it is conceivable that it should be interpreted as an element relating a subject to its predicate; a way to implement this idea is to assume that [+F] has to govern Nominative Case to be licit, according to the condition in (2): (2) Licensing Condition for the Finiteness Feature [+F] An occurrence of the feature [+F] is licit if and only if the head hosting it governs a phonetically realized element bearing Nominative Case, or the trace of such an element. Furthermore, following Holmberg and Platzack, I will assume that subjectverb agreement is represented by the feature Agr, realized in I°, and that Agr may be either a nominal element with certain syntactic consequences, or just a form of the tense affix. A similar idea is developed in Zagona (1988, e.g. p. 168) to account for differences between English and Spanish. In Platzack and Holmberg (1989) it is argued that both Old French and Middle English (henceforth OF and ME, respectively) belong to a group of verb-second languages where Agr is a nominal element; I will adopt this idea in the present paper. When Agr is nominal, it must bear Case: Holmberg and Platzack assume that nominal Agr is inherently Nominative. As such it must be licensed according to the same rules as other Nominative categories. The licensing conditions for structural Nominative Case, assumed by Holmberg and Platzack are given in (3):4 (3)

Nominative is licit iff a. it is head governed by a head with lexical features hosting [+F] (direct licensing), or b. it is governed by a member of a chain, the head of which is licensed by virtue of (3a) (indirect licensing).

Together, (2) and (3) express a mutual dependency relation between [+F] and Nominative Case: just as every instance of Nominative Case must be licensed, directly or indirectly, by the finiteness feature, every example of this feature must have an instance of Nominative Case to license. Since (2) and (3) are formulated in terms of government, we must indicate how government is defined. Following Rizzi (1990) we assume two kinds of government: Head Government and Antecedent Government;5 the definitions given in (4) and (5) are taken from Rizzi (1990:25) unless otherwise stated:

THE LOSS OF VERB SECOND IN ENGLISH AND FRENCH

203

(4) Head Government: X head-governs Y iff (i) a. X is a head b. X m-commands Y (ii) X = {[±N], [±V], [+F]}6 (iii) a. no barrier intervenes b. Relativized Minimality is respected (5) Antecedent Government: X W-antecedent governs Y (W = {A, A', X0})7 iff (i) a. X is in a W-position b. X c-commands Y (ii) X and Y are coindexed8 (iii) a. no barrier intervenes b. Relativized Minimality is respected. The concept Relativized Minimality is understood according to Rizzi (1990:2): Relativized minimality "makes the blocking effect of an intervening governor relative to the nature of the government relation involved [...] if Z is a potential governor of some kind for Y, it will block only government of the same kind from X." To illustrate how the mechanism outlined in (2)-(5) is supposed to work in a verb-second language with Nominative Agr, i.e. a verb-second language of the OF/ME type, consider the structure given in (6). I will only discuss the licensing of Nominative outside of VP in this context. See Holmberg and Platzack for a detailed discussion of the licensing of Nominatives in other positions. In (6) I have indicated the potential Nominative positions with nomsubj for Spec-IP, nomAgr for the Nominative of Agr. The examples in (6i) are OF, taken from the 13th-century prose text La Queste del Saint Graal, (Vance (1989)), the examples in (6ii) are ME, taken from van Kemenade (1987) (information about the texts from which the examples are taken is indicated in parentheses). From its position in C° [+F] head governs both nomsubj and nomAgr. Assuming there is a bi-unique relation between Case assigners and Case assignees (consider e.g. Rizzi and Roberts (1989)), [+F] cannot, however, license and be licensed by both Nominative elements at the same time. Only the alternative, where [+F] is licensing and being licensed by nomAgr, will produce a well-formed result. In this case, nomAgr will be licensed according to (2b) (a specifier position is head governed by its head). If alternatively [+F] were licensing and being licensed by nomsubj, there would be no way to license nomAgr, since a specifier cannot antecedent govern its head (consider (5): the two positions are of different types, A and X°, respectively, hence they cannot form a chain). Consequently, under this alternative we are left with an unlicensed Nominative, i.e. the derivation is blocked.

204

VERB SECOND AND THE NULL-SUBJECT PARAMETER

i.a.

I1; regardeji e Cj ei Cj 1'enfant he looks-at the child i.b. Lors oi'rentj ils; Cj e then heard they to-come a clap of thunder ii.a. The beei hasj ei ej ei ej thre kyndis (RR,9 14th C.) the bee has three (distinctive) features ii.b. pa namenj hii ej ei ej pa men... (PC,10 12th C.) As mentioned above, since ME and OF were verb-second languages, they must have [+F] in C°. After the loss of verb second [+F] is in I°, not in C°. Since lexicalization of [+F] is a prerequisite for [+F] to participate in the assignment of Nominative Case, as mentioned above, C° must be lexicalized prior to the change, producing verb-second languages, whereas I° must be lexicalized after the change. The change of position of [+F] leads to changes in the properties of the functional heads C° and I°. A head hosting [+F] counts as a head governor, see (4ii) above. In ME and OF, where [+F] is in C° and where Agr in I° is nominal, both C° and I° are head governors (nominal Agr is marked [+N]). After the loss of verb second, when [+F] is in I°, only I° is a governor, independent of the status of Agr. The description outlined in Holmberg and Platzack furthermore predicts that verb-second languages with nominal and Nominative Agr may have empty subjects. This is the case in OF and ME, as shown by the examples in (7):

THE LOSS OF VERB SECOND IN ENGLISH AND FRENCH

(7)

205

a. Par desus seelerent pro une pierre (OF) on top fixed (3pl) a stone b. so mochte water pei wepten pat pro made the so much water they wept that made the forseyd lake (ME) before-mentioned lake

In languages with Nominative Agr there is always a Nominative element governed by the host of [+F], be it C° or I°. As a consequence, there is no need for a Nominative element in Spec-IP. The empty element in Spec-IP may be small pro: Assuming with Rizzi (1986) that pro has to be both licensed and identified by the same head, Platzack (1992) claims that pro in Spec-IP is both licensed and identified by C in a verb-second language. As I will argue below, English and French have developed differently with respect to Agr, something which is also detectable from the morphological development of subject-verb agreement: whereas Agr in French was nominal in nature for at least a century after the loss of verb second, English Agr has lost its nominal status by the time English ceases to be a verb-second language. We will return to a discussion of these matters in SECTION 4.2 below.

2.2 Earlier Attempts to Explain the Loss of Verb Second in English and French Recently, several scholars have attempted to answer the question why English and French gave up verb second at the end of the medieval period. Usually, however, the scholars working with the loss of verb second in English do not consider the loss of verb second in French, and vice versa (but cf. Roberts (1992)). In this section, I will review some of these attempts, trying to show that none of the proposed explanations is sufficient. Independent of each other, van Kemenade (1987) for English and Adams (1987) for French have suggested that the main reason for the loss of verb second in these languages is the underlying SVO order. Whereas SVO is basic in subordinate clauses, it is derived in main clauses (V-to-I-to-C, followed by the fronting of the subject NP to Spec-CP, as indicated in (6)). Adams and van Kemenade assume that the language learners begin to interpret SVO in main clauses as basic as well, probably under the influence of the high frequency of this word order and the fact that such an interpretation requires a less complicated derivation. Both Adams and van Kemenade state that such a change would be unlikely in SOV languages like German or OE, since in these languages the main clause word order is not the same as that of the embedded clause. (See also note 2 above.) Although the explanation proposed by Adams and van Kemenade might seem attractive, it faces certain problems that indicate that it is not a sufficient explanation. First of all, several scholars, including Weerman (1989:186), Roberts (1992) and Platzack (1989), have pointed out that this account is not

206

VERB SECOND AND THE NULL-SUBJECT PARAMETER

compatible with the fact that the Scandinavian languages, which are SVO, have managed to uphold verb second. Secondly, as Weerman (1989:186) remarks, main clauses with SVO order are at best ambiguous between the verb-second interpretation and the interpretation that SVO is the basic word order. Hence, even if the language learners of English and French at the time of the loss of verb second experienced a majority of SVO main clauses, the structural ambiguity of these clauses must have made them inappropriate as indicators for or against a verb-second grammar. At the same time, the language learners must also have met examples which explicitly signified the presence of verb second, e.g. all main clauses where a constituent other than the subject is fronted and the subject is in third position, following the fronted element and the tensed verb. We thus have a situation where the language learner must have experienced a certain number of sentences11 which unambiguously indicated the presence of verb second, and a bulk of sentences which were structurally ambiguous between a verb-second interpretation and a basic SVO interpretation. It is unclear why the language learners should ignore these unambiguous cases in favour of a particular interpretation of the ambiguous ones. The conclusion must be that although Adams and van Kemenade may be on the right track, their explanation has to be supplemented to account for the loss of verb second in French and English. Consider next the account of the loss of verb second in French proposed by Vance (1989). Following Travis (1986), Vance assumes that verb-second languages are languages which do not tolerate adjunction to IP of fronted elements: hence, the loss of verb second in French is described as the introduction of the possibility to adjoin fronted elements to IP. Since the loss of is not in focus for Vance, she just states that there has been a resetting of the parameter of adjunction to IP for adverbs and VP complements from negative to positive. There is no attempt in Vance (1989) to explain how this resetting was triggered. Turning to English, we find a serious attempt to explain the loss in English by Weerman (1989:235~240),12 who claims that the loss of verb second is due to a change in his proposed system of S-identification, triggered by the emergence of blockades (mainly the negation not) which prevented the finite features of V from reaching Vmax. According to Weerman (1989:78ff.), both verbal projections and nominal projections must be identified at S-structure: for nominal projections, S-identification is syntactic case, for verbal projections, S-identification is conjugation (i.e., tense and mood). In the same manner as for Case, conjugation can be assigned either inherently or structurally (via C°, governing a finite Vmax; it is to be noticed that there is no I-projection in Weerman's system). The contrast between structurally and inherently assigned conjugation is the basis for the contrast between verb-second languages and non-verb-second languages, according to Weerman: in verb-second languages, C must be lexicalized in order to S-identify Vmax, whereas in non-verb-second languages, S-identification can take place within Vmax, e.g.

THE LOSS OF VERB SECOND IN ENGLISH AND FRENCH

207

via morphologically realized mood indicators. During the ME period, there

are two partly conflicting tendencies, one aiming at structural S-identification (i.e., verb second) the other at inherent S-identification. The emergence of blockades forced English to choose the second alternative, which ultimately (after the loss of the subjunctive) led to the development of a certain class of auxiliaries. There are at least three problems with Weerman's explanation. First it does not seem to hold for French: it is not clear that there emerged any blockade of the English type in French during the Middle French period, when verb second was lost.13 Secondly, the link between the loss of subjunctive and verb second seems somewhat weak, considering the fact that Modern Icelandic, which is a true verb-second language, upholds a morphological distinction between indicative and subjunctive. Finally, there are certain theoretical problems with Weerman's way of implementing the idea that both verbal and nominal categories must be S-identified, as pointed out in Platzack (1990). Summing up, I have shown that none of the attempts to explain the loss of verb second in English and French, reviewed above, is wholly acceptable.14 In the next section I will propose a new explanation, combining the hypothesis of Adams (1987) and van Kemenade (1987) that the ambiguity of SVO sentences is of importance for the loss of verb second with the hypothesis that the loss of verb second is further supported by the presence of structurally ambiguous sentences with weak subject pronouns occurring between the topic and the tensed verb. As I will show, both types of ambiguous sentences are found in English and French prior to the loss of verb second.

3. A Stage with Few Verb-Second Triggers 3.1 The Explanation In this section I will propose an explanation for the loss of verb second which may be applied both to English and French. For a syntactic change to take place there needs to be, at a certain stage in the historical development of the language, structures which the younger generation will interpret differently from the older generation. As pointed out above in connection with my review of the explanation for the loss of verb second proposed by Adams (1987) and van Kemenade (1987), most sentences which the language learner met at the time of the loss of verb second in English and French were probably sentences with a word order ambiguous between a verb-second interpretation and a basic SVO-order interpretation. The structural ambiguity of such sentences is illustrated in (8), where (8a) shows the verb-second interpretation of the ME sentence The bee has thre kyndis 'the bee has three distinct features', and (8b) shows the non-verb-second interpretation of this sentence.

208

VERB SECOND AND THE NULL-SUBJECT PARAMETER

(8)

a. Verb-second interpretation of an SVO sentence.

b. Non-verb-second interpretaion of an SVO sentence.

As pointed out in the previous section, the presence of sentences with ambiguous word orders, as illustrated in (8), cannot be the sole explanation for the loss of verb second. However, at the time when verb second was lost in English and French, there was also another type of ambiguous sentence present:

THE LOSS OF VERB SECOND IN ENGLISH AND FRENCH

209

a sentence with a weak subject pronoun between the topic and the tensed verb. An English and a French example are given in (9): in these examples the weak pronoun is in bold face: (9)

a. Certis pei ben opyn foolis, and... (Wycliffe, late 14C15) Certainly they are open fools b. En verite, il a este et est bon valeton (Jehan de Saintre", in truth he has been and is a good valet-DIM mid 15C16) 'In truth he has been and is a good little valet'.

There are two different structural interpretations of sentences like (9): the pronoun may be in Spec-IP, as ordinary subjects, or it may be cliticized to the tensed verb. These two options are illustrated in (10), where '+' indicates the clitic attachment of the pronoun to the tensed verb in (l0b): (10) a. Certis [IP pei i [ r benj ] [VP ei [v ej ] opyn foolis]]

b. Certis [c peii+benj]k [IP e [I ek] [VP ei [v ej ] opyn foolis]] Consider first (10b), where the subject pronoun is cliticized to the tensed verb. I have not indicated whether this cliticization takes place in VP or in IP; for the sake of my argument this does not matter. The important thing to notice is that the pronoun is cliticized to the verb prior to the movement from I° to C°. Therefore (lOb) counts as a verb-second structure, and can be produced by a person with a verb-second grammar. Consider next the structure outlined in (l0a), where the pronoun is not interpreted as a clitic element. This structure violates the verb-second requirement that the tensed verb must be in second position, hence such a structure cannot be produced by a person with a verb-second grammar. Obviously, sentences like (9) are structurally ambiguous; thus, together with the SVO sentences, they constitute possible triggers for the loss of verb second.17 Summarizing, at the time of the loss of verb second in English and French, we seem to have a situation which can be described as follows. Most utterances which the children met had a word order which was ambiguous and thus did not enable them to determine the position of the finiteness feature: these sentences could be produced by parents having either the plus or the minus value of the Verb-Second Parameter (2). Only sentences with inversion (direct questions, wh-questions and cases with topicalization) unambiguously indicated that the finiteness feature was in C°. The possibility of using examples like (9) with a weak pronoun between the topic and the tensed verb constituted a further ambiguity. Together this led to a situation where true verbsecond indicators were few in number. In spoken language, and especially in the language spoken to children, the number of sentences unambiguously indicating a verb-second grammar must have been quite small: we know from studies of modern verb-second languages that the number of sentences with non-pronominal subjects is low,18 and we can infer that the number of sentences with inverted non-pronominal subjects must have been even lower. In such a situation it is conceivable that many children selected the minus value

210

VERB SECOND AND THE NULL-SUBJECT PARAMETER

of the verb-second parameter, realizing the finiteness feature in I° instead of in C°. To be able to uphold the hypothesis that the loss of verb second in English and French can be understood as a result of the situation described above, where the number of true verb-second indicators was quite low, I must show that there is nothing in the timing of such a stage in the history of both French and English which makes this explanation less plausible. I will start with the ME facts. 3.2 Middle English According to van Kemenade (1987), both subject and object clitics are present in English in the OE period. The subject clitics appear to the left of the tensed verb in clauses where the first constituent is a topic, to the right of the tensed verb when the first constituent is pa, a w/z-word or ne, and to the right of the complementizer in clauses with a base-generated complementizer. OE was an OV-language, hence main clauses introduced by the subject were not structurally ambiguous. However, as soon as the underlying word order changed from OV to VO (a change which was completed around 1200, according to van Kemenade (1987:177)), the potential ambiguities found in SECTION 3.1 above necessary for the loss of verb second were present. Still, as long as there were clear indications that weak pronouns should be interpreted as clitics, the children got enough information for setting the verb-second parameter in such a way that the finiteness feature was realized in C°. During the ME period, the number of object clitics rapidly faded, whereas the patterns for clitic subjects remained stable up to about the third quarter of the 14th century (van Kemenade (1987:197)). Since the loss of object clitics must have introduced the possibility to use strong and weak forms of ordinary pronouns (prior to the loss of object clitics, all weak pronouns could be analyzed as clitics), the status of subject clitics must have been influenced as well. Thus, we could imagine a gradual decrease in the security with which the younger generation was able to interpret weak subject pronouns as clitics. Of most interest to us is the case where we have a subject pronoun to the left of the tensed verb in clauses where the first constituent is a topic: the following exabmples from the 14th century are taken from van Kemenade (1987:197-198): (11) a. An haste he yarn to be gerniere in haste he ran to the storehouse b. Certis ei ben opyn foolis, and don pleynly agenst Certainly they are open fools, and act plainly against Cristis Gospel Christ's gospel. Following van Kemenade (1987:220 ff.), we notice that different dialects (or texts) behave differently with respect to the word order of sentences with a topicalized phrase. Both Rolle (Yorkshire, c. 1340) and Wycliffe (SW Mid-

THE LOSS OF VERB SECOND IN ENGLISH AND FRENCH

211

lands, c. 1380) have optional inversion with nominal subjects, i.e. there are cases in the writings of these two authors where not only a subject pronoun but a full subject NP is found in the position between the topic and the tensed verb. Rolle has optional inversion also with pronominal subjects, indicating that there are no subject clitics in his language (van Kemenade 1987:220), whereas Wycliffe always has subject pronouns in the clitic position between the topic and the tensed verb. The language of Chaucer (E. Midland/London, c. 1390), on the other hand, is consistently verb second both with nominal and pronominal subjects; there is no trace of an ongoing change. Hence, although the text of Chaucer is the youngest one of the three texts investigated by van Kemenade, it is the most conservative with respect to verb second. Discussing the texts of Rolle and Wycliffe, van Kemenade actually proposes that the use of subject clitics may be a reason for the loss of verb second, arguing more or less as I have done above. However, taking into consideration the regular verb second found in Chaucer, she concludes that "the behavior of pronominal subjects cannot have provided the basis for a reinterpretation" (p. 222), settling for the insufficient explanation of the loss of verb second discussed in the previous section. In my view this is a mistake. The alternative seems to be to treat the language/dialect of Chaucer as representing a blind alley in the development of English syntax. There are certain properties of the syntax found in Chaucer's prose that might support the assumption that Chaucer's language is not representative for the development of English at the end of the 14th century. For example, there are very few examples of periphrastic do in the texts of Chaucer, who seems to prefer gin, which was lost in the early part of the Modern English period (Mustanoja (1960:614)), although one of the first cases of do in a question occurs in Chaucer's poetry (Mustanoja (1960:607)). Another case where Chaucer's language is not representative for the development of English is found in his use of overt subjects with infinitives, as in (12) (cf. Einenkel (1887:80ff.)): (12) And I to ben youre veray humble trewe and I to be your very humble faithful. Summing up, provided it is correct to consider Chaucer's language somewhat exceptional with respect to verb second,19 there seems to be a period at the end of the 14th century when the use of subject clitics in ME, together with the general SVO word order, paved the way for the loss of verb second. We have a situation where the language learners could interpret an overwhelming majority of the utterances they met as indicating a non-verb-second grammar, whereas their parents produced these utterances with the help of a verb-second grammar.

3.3 Old and Middle French The situation in French differs from the English situation described above in two ways. Firstly, there is no word order change from OV to VO in French:

212

VERB SECOND AND THE NULL-SUBJECT PARAMETER

like Modern French, OF is a VO language. Secondly, unlike English, subject pronouns in OF cannot be interpreted as clitics; consider Adams (1987:5 fn.). Hence, OF is a verb-second language with SVO word order, lacking subject clitics. Thus, at this stage there is not sufficient ambiguity in the language to cause the loss of verb second. However, at the beginning of the Middle French (henceforth MF) period, there seems to be a change in the status of subject pronouns: according to Adams (1987) and Zwanenburg (1978), pronominal subjects between the topic and the tensed verb may now be considered as (optionally) cliticized to the tensed verb.20 When this happens, the situation is just like the ME one described above: we have a situation where most utterances that the language learners experienced did not tell them how to set the verb-second parameter. Regarding the hypothesis put forward in the present paper, the crucial point is whether it is possible to find a period of time in OF/MF when the parents used pronominal subjects but not nominal subjects between the topic and the tensed verb (i.e. a stage similar to the one found in the writings of Wycliffe (cf. above)). It is not possible for me to answer this question with any degree of certainty. However, the data given by Vance (1989:152-164), especially the observation that pronominal subjects between the topic and the tensed verb are more prevalent than nominal ones when the topic element is a subcategorized complement of the verb phrase, indicate that such a stage may have existed. According to Vance (1989:153 ff.) there are two types of exception in OF to the general observation that the subject cannot occur between an initial constituent other than the subject and the tensed verb: cases where the sentence is introduced by a subordinate clause, and cases where the sentence is introduced by a restricted set of adverbials, including, for instance, sanz faille 'without fail', neporquant, nequedant, neporec 'nevertheless', certes 'certainly'. Vance assumes that the cases with introductory clauses are only apparent exceptions, "because they occur at all stages of French and are to be analyzed as SVX rather than CSV clauses" (p. 153). The cases with introductory adverbials are considered to be real exceptions. This is a doubtful conclusion, however, since similar cases are found in a true verb-second language like Modern Swedish. Compare the OF (13) with its Modern Swedish counterpart (14):21 (13) et sanz faille ele estoit de trop grant biaute pleinne and without fail she was of very great beauty full 'And surely she was full of very great beauty.' (14) Utan tvekan: hon var mycket vacker. without doubt: she was very beautiful It is usually assumed that initial adverbials of the type shown in (14) are situated outside the C-projection (eventually, they may be analyzed as adjoined to CP), hence to the left of the topic position (Spec-CP).22 A similar analysis might be suitable for the OF examples. If so, both types of exceptions to verb second in OF discussed by Vance may be considered superficial.

THE LOSS OF VERB SECOND IN ENGLISH AND FRENCH

213

In MF, there is a great increase in the number of constituents which can appear in initial position without triggering inversion. Consider Vance (1989:157). The initial position in the non-inversion cases is no longer restricted to clauses and adverbials, we also find in this position complements for which the verb subcategori/es. Due to the uncertainty introduced by the possible analysis of examples like (13), mentioned above, the latter type is the only one which really gives evidence for the absence of verb second. For our purposes, then, it is important to determine if the non-inverted subject in these cases might be a clitic, or if it must be analyzed as a true argument, hence if the absence of verb second is real, as Vance suggests, or just apparent. Considering cases where the initial element is a subcategorized complement of the verb phrase, Vance (1989:162) notices that pronominal subjects are more prevalent than nominal ones, although both are possible.23 She is fully aware of the fact that this asymmetrical distribution of pronominal and non-pronominal subjects might support the hypothesis that the pronouns are syntactic clitics at this stage of MF (Vance, p. 187), but prefers to account for the low frequency of nominal subjects in this context in terms of non-syntactic factors. However, as far as I can see, nothing prevents us from taking these frequency facts as support for our hypothesis that there has existed a period of time in OF/MF when the parents used pronominal subject clitics but not nominal subjects between the topic and the tensed verb. Together with the presence of the general SVO word order, the occurrence of subject clitics lead to the loss of verb second. As in English, we have a situation in French where the language learners could infer a non-verb-second grammar on the basis of an overwhelming majority of the utterances they encountered, notwithstanding the fact that their parents produced these utterances with the help of a verb-second grammar. 3.4 Conclusion Concluding this section, we have found that different developments in English and French led to a situation where both languages had subject clitics, were verb-second languages and had an underlying SVO word order. In such languages there are very few indications of how to set the verb-second parameter. As I have argued, language learners of both English and French reacted in the same way to this situation: they gave up verb second. In my terms, this means that they changed the position of the finiteness feature, from C° to I°. In the next section I will consider certain syntactic consequences of this change.

4. Consequences of the Change of Position of [+F] As mentioned above, it follows from the system developed in Holmberg and Platzack that the node hosting the finiteness feature [+F] must be lexicalized in order to license Nominative Case on the subject. We have seen

214

VERB SECOND AND THE NULL-SUBJECT PARAMETER

that this requirement is the reason for the verb to move to C° in verb-second languages. When verb second is lost in French and English, and [+F] is in I°, there is a similar demand for I° to be lexicalized. French and English have reacted differently to this lexicalization requirement, due to the different status of Agr in I° at the time of the loss of verb second. Since Agr in I° is nominal in nature and hence marked with the feature [+N] in both OF/MF and ME, I° is a head containing [+N] and consequently a head governor in both languages; cf. the definition of Head Government in (4) above. When [+F] is placed in I°, the governor status of I° becomes independent of the categorial status of Agr. In present day English and French, Agr has lost its categorial status, as is evident from the impossibility of using empty subjects in the modern versions of these languages (with respect to the correlation between nominal/Nominative Agr and empty subjects, cf. SECTION 2 above).24 However, whereas this loss of nominal/Nominative Agr seems to take place more or less at the same time as the loss of verb second in English, the loss of nominal/Nominative Agr is much later in French. The two factors mentioned above, i.e. the change of position of [+F] and the different developments of Agr, are responsible for several syntactic differences between French and English which begin to be visible in the first half of the 15th century. In this section I will discuss these differences, starting in SECTION 4.1 with a discussion of the differences with respect to how the two languages meet the demand of lexicalizing [+F], following this with a discussion of null subjects in SECTION 4.2.

4.1 The Lexicalization of [+F] Due to the Head Movement Constraint (see Travis (1984); Baker (1988)), which says that a head must move to the position of its closest governing head, the verb has to move to I° before it can reach C°. As noticed above, prior to the loss ofverb second the head hosting [+F], i.e. C°, was lexicalized by means of V-to-(I-to-)C. We could expect that both English and French would use Vto-I to lexicalize the host of [+F] after the loss of verb second. However, facts indicate that only French has taken this step: compare the following sentences of modern French and English: (15) a. Jean embrasse souvent Marie. / *Jean souvent embrasse Marie, b. *John kisses often Mary. / John often kisses Mary. If adverbs like souvent/often are adjoined to VP, the presence of V-to-I in French and its absence in English would automatically account for the word order difference illustrated in (15). Hence, whereas English seems to meet the requirement that the host of [+F] must be lexicalized by establishing a certain category of auxiliaries in I° (see below), French uses head movement of the (first) verb to accomplish this lexicalization, both prior to and after the loss of verb second. Lacking a better solution, I will follow Pollock (1989) and assume that the reason why English gave up V-to-I is a direct consequence of the status of Agr: "Agr in English, unlike Agr in French, is not 'rich' enough

THE LOSS OF VERB SECOND IN ENGLISH AND FRENCH

215

morphologically to permit transmission of the verb's 0-role" (Pollock (1989:385)).25 This assumption is compatible with the fact that auxiliaries in English, which do not have 6-roles, are allowed in I°, meeting the demand of making the host of [+F] detectable. (16) a. John has often kissed Mary. b. John must often kiss his wife. c. John did not kiss his secretary. If the description outlined so far is correct, the crucial question is how the lexicalization requirement of [+F] is met in English examples like (15b), where no V-to-I has taken place. To answer this question, we must first consider the system of do-support in English. In a case like (16c), [+F] is lexicalized with the help of a dummy auxiliary do. Pollock (1989) assumes that the auxiliary do is a substitute verb which copies the 0-role of the main verb. Returning to cases like (15b), I will elaborate on the suggestion made by Pollock (1989:404) that English has a phonologically unrealized counterpart of the auxiliary do, i.e. 0, which shares all its defining properties except its phonological form.26 This phonologically empty auxiliary is supposed to lexicalize I° in cases like (15b). Since 0 does not have a phonological form, it cannot host the tense ending. On the other hand its presence in I° blocks V-to-I. Therefore, Infl-lowering must apply, attaching the tense ending to the main verb in VP. Still following Pollock, I assume the presence of a Neg Phrase in cases like (16c),27 which acts as a barrier for Infl-lowering. This accounts for the ungrammaticality of examples like (17): (17) *John not kissed his secretary. In cases like (17), the tense ending cannot reach the verb, and the sentence is blocked, since affixes must attach to phonologically realized heads. Pollock's description of the difference between Modern English and modern French suggests the establishment in English of two processes not present in French: do-support and Infl-lowering. As I have interpreted these processes above, it is clear that they are both the result of the combination of the change of position of [+F] and the weakening of Agr in ME. To support this hypothesis, I must show that the weakening of Agr and the emergence of both auxiliary do and Infl-lowering take place more or less at the same time as the loss of verb second. With respect to Agr, we notice a considerable weakening of subject-verb agreement during the ME period. In the 14th century, the endings for person and number are lost in the preterite indicative, as well as the endings for the first person present singular and present plural (van Kemenade (1987:204)). Since it is not inconceivable that there is a strong correlation between overt subject-verb agreement and the presence of a syntactically active Agr (see Platzack and Holmberg (1989)), I conclude that Agr is weakened during the 14th century, i.e. at the same time that verb second is lost. Turning to the emergence of do-support, it seems clear from studies by Ellegard (1953) and Visser (1963-84:1488-1568) for example, that the peri-

216

VERB SECOND AND THE NULL-SUBJECT PARAMETER

phrastic auxiliary do came into existence in the late 13th century. One of the factors playing an important role for its regulation was actually the loss of verb second, according to Ellegard (1953:209).28 It is also possible to trace the emergence of Infl-lowering in English to the time when verb second was lost. At this time there occur in English instances which seem to indicate some kind of tense copying or tense agreement within the single clause, cases where tense seems to be realized both in I° and on the verb in VP. Such examples provide evidence for the presence of Infl-lowering, since there is no other way for tense to be realized within VP. As noticed by Ellegard (1953:123), discussing the use of semi-auxiliary do + "explanatory" verb, before 1400 "the explanatory verb was generally in the same form as do: finite if do was finite, infinite if do was infinite." An example is given in (18): (18) Thalestris ... did wroot to kyng Alexandre T. did wrote to king A. In (18) there are two tense affixes: on the auxiliary do, and on the verb in VP. The empty auxiliary 0, suggested by Pollock, may be seen as a remnant of the overt do in cases like (18). In conclusion, we have found that the loss of verb second in English occurs at the end of the 14th century, more or less simultaneously with the emergence of auxiliary do, Infl-lowering, and the weakening of Agr. The hypothesis that these changes are related to the change of position for [+F] thus seems to be supported.29

4,2 The Licensing of Small Pro French word order data do not provide us with any clear evidence for the hypothesis that the loss of verb second implies a change in the status of I°. To find such evidence, we must consider another aspect of the finiteness feature: its role as a licenser of small pro. As mentioned in SECTION 2, Holmberg and Platzack argue that small pro in Spec-IP is licensed by the host of [+F] and identified by the (p-features of Nominative Agr. Consider the discussion in Holmberg and Platzack, where the following licensing and identificational conditions for pro are suggested, mainly following Rizzi (1986):30 (19) a. Formal Licensing of Small Pro: Pro is head governed by a Case-licensing head Xy°. b. Identificational Condition of Small Pro: (Rizzi 1986:520) Let X be the licensing head of an occurrence of pro. Then pro has the grammatical specification of the features on X coindexed with it. Since the Case-licensing head of Spec-IP is C° in verb-second languages and I0 in non-verb-second languages, it follows from (19) that, pro is licensed from different heads in these two types of languages. Hence the loss of verb sec-

THE LOSS OF VERB SECOND IN ENGLISH AND FRENCH

217

ond leads to a change with respect to the licenser of small pro: prior to the change, the licenser of pro is different from the head hosting Agr, after the change, the licenser of pro is identical to the head hosting Agr. Following Rizzi (1986:520) Holmberg and Platzack assume that the different interpretations of pro depend on which (p-features pro is associated with, according to the specification in (20); see Rizzi (1986:543):31 (20) a, referential pro: [+number, +person] b. quasi-argumental pro: [+number, -person] c. true expletive pro: [-number, -person] It follows from (19) that pro has the grammatical specification (i.e. '(o-features) of its licensing head. Hence a change of position for [+F] does not only change the licensing head of small pro, it also changes the way in which pro is identified. Whereas it is clear that I° contains o-features due to the presence of Agr in I°, it is not evident that there are any o-features in C°. To understand how pro is identified in verb-second languages, i.e. in cases where the licensing head (C°) is not automatically associated with any cp-feature, we must consider how pro might be interpreted in such languages.32 Obviously verb-second languages with syntactically active Agr differ in their abilities to represent (p-features in C° and hence to identify small pro. Modern German seems to lack (p-features in C° altogether: the only possible identification of pro in such a case is that pro is a true expletive. Thus, an example like (2la) with expletive pro is well formed in modern German, whereas examples like (21b,c) are not well formed, since C° without any (p-features cannot identify either referential pro or quasi-argumental pro: (21) a. Gestern wurde pro getanzt. yesterday was danced b. (Sein Buch ist sehr interessant.) *Leider hat pro sehr schlechte Bilder. his book is very interesting. Unfortunately has very bad pictures c. * Gestern hat pro geregnet. yesterday has rained On the other hand, modern Icelandic differs from German in allowing quasiargumental pro: consider the well-formed example in (22): (22) Rigndi pro i gaer? rained yesterday Since C° is the identifying head of pro in Icelandic as in German, the contrast illustrated by (21c) and (22) indicates that C° may be marked for [+number] in Icelandic, but not in German. Consider (20) above. There is no morphological support for this difference: in both (21c) and (22), the tensed verb in C° is in its 3sg form. Furthermore, the facts are the same in embedded clauses: although neither German nor Icelandic complementizers show agreement, quasi-argumental pro is possible in Icelandic embedded clauses, but not in German embedded clauses:

218

VERB SECOND AND THE NULL-SUBJECT PARAMETER

(23) a. Hann sagoi ao hefoi rignt mjog allan daginn i gaer. (Ice.) he said that had rained much whole day yesterday b. * Er sagte daB gestern

hatte den ganzen Tag vieles geregnet. (Ger.) he said that yesterday had the whole day much rained.

To account for the difference between Icelandic and German with respect to the interpretation of pro, Platzack (1992) proposes that Icelandic C° differs from German C° in respect to being marked with the feature [+number]. As indicated by the examples in (22) and (23a), the ability of C° to license quasi-argumental pro seems to be the same in main clauses as in subordinate clauses, notwithstanding the fact that the agreement morpheme realized on the tensed verb is actually occurring in C° in main clauses, but not in subordinate clauses. This observation supports the assumption that Agr is not necessarily to be identified with the verbal inflection: Agr is a property of I°, which may or may not be interpreted as a property of the inflected verb. We conclude that languages seem to differ with respect to whether or not (p-features may appear in C°.33 Whereas modern Icelandic, as we have seen, does not allow referential pro, and hence does not allow the (p-feature [-t-person] of Agr to be realized in C° even when the inflected verb is moved to C, the situation is different in Old French: as, for example, Adams (1987:2) points out, empty referential subjects typically occur only in main clauses with inversion in OF. Hence small pro can be identified as referential in OF in case the p-feature [+person] is allowed to follow the verb to C°.34 Now, when verb second is lost, small pro is identified by I°, not by C°, as mentioned above. In a situation where Agr has the same status before and after [+F] changed its position, we expect this change to lead to an extension of the distribution of pro: after the change, the interpretation of pro should no longer be dependent on verb movement to C°.35 This is exactly what we find in French: in 15th-century texts, i.e. texts produced after the loss of verb second, null subjects are found not only in main clauses, but also in subordinate clauses. Consider the following example, taken from Vance (1989:3): (24) Puet bien estre que n'en avez point may (3sg) well be that neg-en have neg 'It may well be that you have none'.

(Jehan de Saintre)

Since the increased distribution of small pro in French appears after the loss of verb second we have perfect timing: the change of position of [+F], which is the theoretical consequence of the loss of verb second, has visible effects on the grammatical output.36 Turning to English, null-subject data do not give us a similar support for the change of position of [+F]. At the time of the loss of verb second, English and French differed considerably with respect to subject-verb agreement: whereas both OF and MF have rich verbal inflection (although Vance (1989:267) notices that this inflection is eroding in MF), the ME system of

THE LOSS OF VERB SECOND IN ENGLISH AND FRENCH

219

subject-verb agreement is considerably weakened, as we saw in SECTION 4.1. In that section, I argued that Agr in English ceased to be syntactically active more or less at the time of the loss of verb second. Around this time, we still find non-referential pro, quasi-pro and now and then cases of referential pro in main clauses; the situation seems to be similar to the OF situation. However, the use of overt pronominal subjects was the rule already in OE, as Visser (1963:4) notices, and the last remnants of null subjects are found during the 16th century. This is in accordance with my description: we do not expect any expansion of the use of null subjects in English after the loss of verb second, provided that Agr lost its categorial features at this time.

5. Conclusion In this paper I have tried to show that the loss of verb second in French and English during the 14th and 15th centuries could be given a common explanation: at the time of the loss of verb second, both languages had a predominant use of SVO word order, and in addition both languages had developed subject clitics. The SVO word order can be produced both by a grammar with verb second and by a non-verb-second grammar where SVO is the underlying word order. The same is true of sentences with a pronominal subject between the topic and the tensed verb. Hence for both French and English there is a period when an overwhelming majority of the sentences uttered could be given two different structural interpretations. Such a situation is a necessary prerequisite for a syntactic change to take place. According to Holmberg and Platzack, who provide the theoretical framework for the present study, the loss of verb second is described as a change in position of the finiteness feature [+F]: in verb-second languages, this feature is in C°, in non-verb-second languages it is in I°. This change in position has different consequences for English and French. For English the consequences involve the development of do-insertion and Infl-lowering, for French a change in the distribution of null subjects. The difference between English and French in this respect has to do with the status of Agr: whereas Agr is nominal and Nominative in French both prior to and after the loss of verb second, Agr is losing its nominal status in English more or less at the same time as verb second is lost.

Notes A preliminary version of this paper was read at the First Generative Diachronic Syntax Conference in York, England, April 1990, whereby I got several valuable comments by the audience. Thanks also to Lars-Olof Delsing, Cecilia Falk, Anders Holmberg, Ans van Kemenade, Beatrice Santorini and Halld6r Sigur5sson for their reactions to the preliminary version, and to Ian Roberts for letting me read an early draft of Roberts (1992), a book which has been of

220

VERB SECOND AND THE NULL-SUBJECT PARAMETER

great importance to me for the revision of this paper. Naturally I am solely responsible for all errors. 1. there seems to be such an attempt in Roberts (1992): however, at present I have only had access to that part of the manuscript which deals with French. 2. This difference is most obvious in verb-second languages with a basic OV order, where the tensed verb ends up in final position in subordinate clauses and in second position in main clauses. Also verb-second languages with a basic VO order which lack Agr (e.g. the Mainland Scandinavian languages: Danish, Norwegian, Swedish) display a clear word order difference: in these languages the tensed verb occurs in third position in subordinate clauses with sentence adverbs, and in second position in main clauses. The difference between main clauses and subordinate clauses for verb-second languages with Agr in I° and a basic VO order is not obvious in the same way. The three different types of verb-second languages are illustrated in (i)-(iii) with German, Swedish and Icelandic examples, respectively: (i)

a. daB Karl das Buch nicht kaufte (German) that K. the book not bought b. att Ulf inte kopte boken that U. not bought the-book

(Swedish)

c. a5 Jon keypti ekki bokina that J6n keypti ekki bokina

(Icelandic)

(ii) a. Karl kaufte das Buch nicht. K. bought the book not b. Ulf kopte inte boken. U. bought not the-book

(German) (Swedish)

c. Jon keypti ekki bokina. (Icelandic) J. bought not the-book The apparent lack of a root-subordinate distinction in languages like Icelandic and Yiddish has led several scholars to question the virtue of attempting to give a unified account of the verb-second phenomenon. Consider Rognvaldsson and Thrainsson (1990), Diesing (1990), and Santorini (1989), who all agree in claiming that verb second in Icelandic and Yiddish is partly different from verb second in other Germanic languages. See also the contributions by Santorini and Lemieux and Dupuis in this book. According to these scholars, the tensed verb ends up in C° in main clauses of languages like German, Dutch, Danish, Norwegian, and Swedish, whereas it ends up in I0 in main clauses of Icelandic and Yiddish. Since both Middle English and Old French were languages of the same type as Icelandic and Yiddish (consider e.g. Platzack and Holmberg (1989)), this description should apply to these languages as well. Consider also Travis (1984), who claims that subject-first clauses have the tensed verb in I°, whereas clauses where some non-

THE LOSS OF VERB SECOND IN ENGLISH AND FRENCH

221

subject element is fronted have the tensed verb in C°. Recently, Vikner and Schwartz (1991) have given empirical support to the assumption that the tensed verb is always outside of IP in verb-second clauses. 3. This account of the verb-second effects has much in common with the standard V-to-Comp analysis which was first suggested by den Besten in an unpublished paper from 1977, later printed as den Besten (1983) (see also den Besten (1989:94)). 4. Compare also Borer (1986). 5. Consider especially Rizzi's (1990:76-80) arguments against the existence of Theta Government. 6. Rizzi (1990:25) formulates this condition as follows: X = { [±N], [±V], Agr, T}. The difference between the formulation in (4ii) and Rizzi's formulation is not dramatic: for most non-verb-second languages, [+F] is situated in the same position as T, hence for these languages the difference is just notational. With respect to Agr, Holmberg and Platzack claim that only Agr which is nominal, i.e. containing the feature [+N], qualifies as a head governor. Hence most cases where Rizzi explicitly refers to Agr are handled by the presence of the feature [+N] in (4). The only case where the definition in (4) and the definition in Rizzi (1990) differ is for languages where Agr is not nominal and not in the same position as [+F], i.e. verb-second languages with nonnominal Agr. Such languages should have no V-to-I in spite of having subject-verb agreement. There are some MSc. dialects of this kind, e.g. the dialect of Hallingdalen in Norway (see Trosterud (1989)). 7. Rizzi uses this formulation in order to specify the three subcases of antecedent government: chains formed by NP-movement, wh movement, and head movement. 8. Rizzi (1990:92) reformulates this condition as: (i) X and Y are non-distinct. 9. Richard Rolle of Hampole, The Bee and the Stork. From English Writings of Richard Rolle Hermit of Hampole, ed. by Hope Emily Allen, Oxford 1931. 10. The Peterborough Chronicle, ed. by C. Clark. 2nd Edition, Clarendon Press, Oxford 1970. 11. The number of sentences which unambiguously indicated verb second was probably quite low, however. For Chaucer's Treatise on the Astrolabe, Macleish (1969) has recorded 83% SV order in main clauses, leaving just 17% for S-V inversion. For the first 135 lines of the OF prose text La Queste del Saint Graal, Vance (1989:35) reports a frequency of 75% SV order in main clauses (only clauses with overt subject are taken into consideration); the frequency of S-V inversion of full NPs is 13%. 12. This is a simplification of Weerman's position: according to Weerman (p. 234), English never was verb-second language in all respects. 13. Naturally this is a problem only if you try to find a common explanation for the loss verb second in French and English. It should be pointed out that Weerman is only concerned with the loss of verb second in English. 14. As I have mentioned several times by now, Roberts (1992) might be an exception. In the part of his book that I have had access to, Roberts gives a detailed and very interesting account of the loss of verb second in French,

222

VERB SECOND AND THE NULL-SUBJECT PARAMETER

which in its essence seems to be quite close to the account I will present in SECTION 3 below. In particular, it can be noticed that Roberts derives the loss of verb second from a change in the way Nominative Case is assigned. In my description, verb second is a consequence of the position of the finiteness feature, which in its turn has an important role to play for the licensing of Nominative Case. However, since I have not had access to the final version of his book, I will abstain from making a detailed comparison of our approaches. 15. The example is found in van Kemenade (1987:200). 16. The example is found in Vance (1989:200); notice that Vance does not consider il in this example to be a true syntactic clitic. 17. A third factor might have been the word order of subordinate clauses. As noticed in note 2 above, both OF and ME belong to a type of verb-second languages with few differences between main clauses and embedded clauses. 18. Einarsson (1978:143) reports that 45% of the subjects in written Swedish are pronouns (an average of different genres), whereas 82% of the subjects in modern spoken Swedish are pronouns (once again this figure is an average). In this material there are unfortunately no frequency figures for the amount of subject pronouns in the language of adults talking to young children. 19. Van Gelderen (1989) notices that verb second is optionally used in Chaucer's Prologue to The Wife of Bath, from which she quotes the following examples: (i)

a. An housbande I wol have... a husband I will have

b. In wyfhod I wol use myn instrument In wifehood I will use my instrument. However, since the Prologue is poetry, not prose, examples like these should most probably not be taken into consideration. 20. Both Vance (1989:183ff.) and Roberts (1992) argue that there is no syntactic cliticization in French at this time, only phonological cliticization: hence, according to Vance, the subject pronoun is not syntactically adjoined to the tensed verb, meaning that cases where the subject pronoun occurs between the topic and the tensed verb show that MF is not a verb-second language. 21. Another type of main clause lacking verb second found in all Scandinavian languages, is introduced by adverbs corresponding to the English maybe, for instance Swedish kanske, Icelandic kannski. Illustrating examples are given in (i): (i) a. Kanske jag kommer. b. Kannski eg komi. maybe I come. The position of the adverb in examples like (i) has been debated. Platzack (1986) notices that Swedish kanske can appear in the positions which are normally reserved for the tensed verb, i.e. in C° in main clauses. It is, for example, found in second position in declarative clauses, as shown in (ii): (ii)

a. Allan kanske redan har fatt jobbet. Alan maybe already has got the job

THE LOSS OF VERB SECOND IN ENGLISH AND FRENCH

223

b. Igar kanske Lena kopte en ny bok. Yesterday maybe Lena bought a new book Another possible analysis, suggested by the fact that kanske/kannski in examples like (i) may be followed by a complementizer (kanske attjag kommer/ kannski ad eg komi), is that the adverb is in Spec-CP, and that it for some reason triggers the presence of a complementizer in C°. 22. Modern Swedish adverbials of this kind often trigger the use of sa 'so' as a filler of Spec-CP, as in (i): (i)

Utan tvekan sa var hon mycket vacker. without doubt so was she very beautiful It is to be noticed that the construction with sa differs in certain respects from ordinary cases of Left Dislocation or Constituent Dislocation; consider the discussion in Ekerot (1990). 23. According to Roberts (1992), the sequence XP - non-pronominal subject tensed verb is never found with topicalized complements. 24. Notice that there is virtually no indication of subject-verb agreement in Modern spoken French. Holmberg and Platzack take this fact to indicate that Agr does not play any syntactic role in Modern French. 25. The lack of overt subject-verb agreement in Modern Spoken French is embarrassing for this idea, especially if Platzack and Holmberg (1989) are right in claiming that overt subject-verb agreement is a necessary prerequisite for syntactically active Agr. 26. Pollock uses the term "its lexical character." However, I will claim that 0 is a lexically realized element lacking phonological form, hence the formulation in the text. 27. Compare the review of Weerman (1989) in SECTION 2.2 above. 28. Ellegard (1953:209) claims that "[t]wo sorts of changes in the language structure that were of decisive importance for the regulation of the use of do were taking place in the 15th and 16th centuries. One was the movement of the more lightly stressed adverbs towards the position between the subject and the main verb, and the other was the virtual disappearance of inversion for other than auxiliaries and intransitive verbs." 29. It should be noticed that different systems seem to be available during the 15th and 16th centuries. According to Roberts (to appear) we find both examples with Infl-lowering and examples with V-to-I in the late 16th century: (i)

a. I spoke not. (V-to-I, not after c 1600) b. I not spoke. (Infl-lowering, 1550-1650) 30. The formulation proposed by Holmberg and Platzack differs minimally from the formulation in Rizzi (1986), the only difference being that Rizzi claims that pro must be Case marked, whereas Holmberg and Platzack only state that pro should be head governed by a Case-licensing head. 31. The remaining combination of features, [-number, +person], might perhaps be used to characterize arbitrary pro. 32. Similar investigations are found in Tomaselli (1990), Cardinaletti (1990), and Falk (1990).

224

VERB SECOND AND THE NULL-SUBJECT PARAMETER

33. Naturally these data might also be interpreted to show that Rizzi (1986) is wrong when he claims that pro is both licensed and identified by the same head. If that restriction is lifted, we could say that pro is identified by Agr in either C° or I°. In the lack of any independent reasons to perform such a weakening of the theory, I will stick to the more restricted version given in (19). See Platzack (1992) for a discussion of the consequences of such an approach. 34. Roberts (1992) extensively discusses cases with null subjects in embedded clauses in OF, showing that null subjects are found only in cases where embedded verb second is possible. 35. Independent evidence for this hypothesis is provided by the development of Romance varieties in Northen Italy and Southern France, where null subjects were generalized from verb-second contexts to all contexts with the loss of verb second. See references in Roberts (1992). 36. Consider Vance (1989) and Hirschbuhler (this volume) for detailed investigations of the exact contexts inwhich null subjects are possible in MF texts.

References Abraham, W. and E. Reuland, eds. (1991) Issues in Germanic Syntax. Berlin/New York: Mouton de Gruyter. Adams, M. (1987) "From Old French to the Theory of Prodrop." Natural Language and Linguistic Theory 5:1-32. Besten, H. den (1983) "On the Interaction of Root Transformations and Lexical Deletive Rules." In W. Abraham, ed. On the Formal Syntax of the Westgermania, 147-131. Amsterdam: John Benjamins. Besten, H. den (1989) Studies in West Germanic Syntax. Amsterdam/Atlanta: Rodopi. Borer, H. (1986) "I-Subjects." Linguistic Inquiry 17:375-416. Cardinaletti, A. (1990) Impersonal Constructions and Sentential Arguments in German. Padova: Unipress. Chomsky, N. (1986) Barriers. Cambridge, Mass.: MIT Press. Diesing, M. (1990) "Verb-Second in Yiddish and the Nature of Subject Position." Natural Language and Linguistic Theory 8:41-79. Einarsson, J. (1978) Talad och skriven svenska. Lund: Walter Ekstrand. Einenkel, E. (1887) Streifuge durch die mittelenglische Syntax unter besonderer Berucksichtigung der Sprache Chaucer's. Munster: Verlag von Heinrich Schoning. Ekerot, L.-J. (1990) "The Swedish sa-Construction in a Functional and Contrastive Perspective." Paper presented at the Seventh International Conference of Nordic and General Linguistics, Torshavn 1989. Ellegard, A. (1953) The Auxiliary Do. The Establishment and Regulation of its Use in English. Stockholm: Almqvist and Wiksell. Falk, C. (1990) "Pro-drop in Early Modern Swedish." Paper read at the llth Groningen Grammar Talks, November 1990.

THE LOSS OF VERB SECOND IN ENGLISH AND FRENCH

225

Gelderen, E. van (1989) "The Historical Rationale Behind Split Infinitives and Kindred Constructions." Archivfur das Studium der neueren Sprachen und Literaturen 26:1 -18. Hirschbuhler, P. (this volume) "Null Subjects in Verb-First Embedded Clauses in Philippe de Vigneulles' Cent Nouvelles Nouvelles." Hjartard6ttir, Th. (1987) Getio i eyournar. Um cyour fyrir frumlog og audlog i eldri islensku. MA Thesis, University of Iceland. Holmberg, A. and C. Platzack (1988) "On the Role of Inflection in Scandinavian Syntax." Working Papers in Scandinavian Syntax 42:25-42. Holmberg, A. and C. Platzack (1988) On the Role of Inflection in the Syntax of the Scandinavian Languages. Oxford: Oxford University Press. Kemenade, A. van (1987) Syntactic Case and Morphological Case in the History of English. Dordrecht: Foris. Macleish, A. (1969) The Middle English Subject-Verb Cluster. The Hague: Mouton. Mustanoja, T. (1960) A Middle English Syntax. Part 1: Parts of Speech. Helsinki: Societe Neophilologique. Platzack, C. (1986) "COMP, INFL, and Germanic Word Order." In L. Hellan and K. Koch Christensen, eds. Topics in Scandinavian Syntax, 185-234. Dordrecht: Reidel.. Platzack, C. (1989) "Review of Ans van Kemenade: Syntactic Case and Morphological Case in the History of English." Linguistics 27:948-953. Platzack, C. (1990) "Review of Fred Weerman: The V2 Conspiracy. A Synchronic and a Diachronic Analysis." Nordic Journal of Linguistics 13:77-88. Platzack, C. (1992) "Complementizer Agreement and Argument Clitics." Working Papers in Scandinavian Syntax 50:25-54. Platzack, C. and A. Holmberg (1989) "The Role of AGR and Finiteness in Germanic VO Languages." Working Papers in Scandinavian Syntax 43:1-76. Pollock, J.-Y. (1989) "Verb Movement, UG and the Structure of IP." Linguistic Inquiry 20:365-424. Rizzi, L. (1986) "Null Objects in Italian and the Theory of pro." Linguistic Inquiry 17:501-557. Rizzi, L. (1990) Relativized Minimality. Cambridge: MIT Press. Rizzi, L. and I. Roberts (1989) "Complex Inversion in French." Probus 1:130. Roberts, I. (1989) "Case Assignment Parameters and the History of French Inversion." Glow Newsletter 22:55-57. Roberts, I. (forthcoming) "Object Movement and Verb-Movement in Early Modern English." To appear in H. Haider, S. Olsen and S. Vikner, eds. Proceedings of the Seventh Comparative Germanic Syntax Conference. Dordrecht: Kluwer. Roberts, I. (1992) Verbs and Diachronic Syntax. Dordrecht: Kluwer. Rognvaldsson, E. and H. Thrainsson (1990) "On Icelandic Word Order Once More." In J. Maling and A. Zaenen, eds. Modern Icelandic Syntax.

226

VERB SECOND AND THE NULL-SUBJECT PARAMETER

Syntax and Semantics 24:3-40. San Diego: Academic Press. Santorini, B. ( 1989) The Generalization of the Verb-Second Constraint in the History of Yiddish. PhD Dissertation, University of Pennsylvania. Sigur5sson, H. (1989) Verbal Syntax and Case in Icelandic in a Comparative GB Approach. PhD Dissertation, University of Lund, Sweden. Tomaselli, A. (1990) "COMP° as a Licensing Head: an Argument Based on Cliticization." In J. Mascaro and M. Nespor, eds. Grammar in Progress. GLOW Essays for Henk van Riemsdijk, 433-445. Travis, L. (1984) Parameters and Effects of Word Order Variation. PhD Dissertation, MIT. Travis, L. (1986) "The Parameters of Phrase Structure and V2 Phenomena." Ms. McGill University. Trosterud, T. (1989) "The Null Subject Parameter and the New Mainland Scandinavian Word Order: A Possible Counterexample from a Norwegian Dialect." In J. Niemi, ed. Papers from the Eleventh Scandinavian Conference of Linguistics 1:87-100. University of Joensuu, Finland, Vance, B. (1989) Null Subjects and Syntactic Change in Medieval French. PhD Dissertation, Cornell University. Weerman, F. (1989) The V2 Conspiracy. A Svnchronic and a Diachronic Analysis. Dordrecht: Foris. Vikner, S. and B. Schwartz (1991) "The Verb Always Leaves IP in V2 Clauses." Ms. Visser, F. Th. (1963-84) An Historical Syntax of the English Language. 1111. Leiden: EJ. Brill. Zagona, K. (1988) Verb Phrase Syntax: A Parametric Study of English and Spanish. Dordrecht: Kluwer. Zwanenburg, W. (1978) "L'ordre des mots en francais medieval." In R. Martin, ed. Etudes de syntaxe du moyen francais, 153-171. Paris: Kliencksiek.

9 Verb Second, Pro-drop, Functional Projections and Language Change1 Aafke Hulk University of Amsterdam Ans van Kemenade Free University Amsterdam

In this article we present an account of the history of verb second and prodrop in French and English. We shall identify two types of verb second languages, which we call C-oriented or CV2 languages, and I-oriented or IV2 languages, and the specific properties correlating with this distinction. Next, we shall see how the distinction as drawn, in conjunction with certain assumptions concerning the licensing of functional properties, allows a very articulate account of the history of French, in terms of a development from CV2 to IV2 to non-V2, and sound theoretical motivation why English developed from CV2 to I-non-V2 and did not move through the IV2 stage. The similarities and differences between the diachrony of verb-second and pro-drop in French and English are thus analyzed in a surprisingly uniform way. The paper is organized as follows: SECTION 1 presents the theoretical background to our account of the history of verb second and pro-drop in English and French; 1.1 contains a discussion of verb second and C/I orientation; 1.2 presents the outline of a theory of licensing that we develop in more detail elsewhere (Hulk and van Kemenade (1993)); in 1.3 the role of functional projections in language change is considered. In SECTION 2 an analysis is given of the history of verb second and pro-drop in French and English. In SECTION 3 we summarize our conclusions.

1. Theoretical Background 1.1 Verb Second and C versus I-Orientation Following the characterization of verb-second in the introduction of this volume, we assume that verb second is a process that takes place regardless 227

228

VERB SECOND AND THE NULL-SUBJECT PARAMETER

of the basic sentence structure, and fronts the finite verb (Vf) to presentential position in root clauses. We analyze this as in the structure (1), adapted on the basis of Chomsky (1986) and Sportiche (1988).

One of the core characteristics of verb second in this sense is the asymmetry between root and non-root clauses with respect to verb-fronting. Fronting of Vf is found in root clauses only, cf. early work on verb-second in den Besten (1983). This is the main reason why verb second is widely analyzed as movement of the finite verb/I to Comp, which is blocked in non-root clauses by the presence of a base-generated complementizer. Thus, in verb-second languages, movement of Vf in non-root clauses can take place to I, but never to C. Evidence for this root/non-root asymmetry is the fact that apparent embedded verb-second with topicalization takes place only after bridge verbs, as exemplified for modern Dutch in (2): (2)

a. Hij zei hij heeft hem gisteren gezien he said he has him yesterday seen b.

Hij zei gisteren heeft hij hem gezien he said yesterday has he him seen

c. Hij zei dat hij hem gisteren gezien heeft he said that he him yesterday seen has d. Hij zei dat gisteren heeft hij hem gezien he said that yesterday has he him seen

VERB SECOND, PRO-DROP, FUNCTIONAL PROJECTIONS

229

e. *Hij zei dat heeft hij hem gisteren gezien he said that has he him yesterday seen 'he said he saw him yesterday'. After bridge verbs, embedded clauses as in (4) must be analyzed as root clauses (i.e. CP). For further discussion, see, for example, Weerman (1989). In terms of the structure (1), verb-second languages (the Scandinavian languages, Modern German, Modern Dutch, Old and Early Middle English, Old French) are analyzed as follows: in (1) I and C are both heads. Verb-second languages have movement of Vf to I, and subsequent movement of V/I to C in root clauses. Movement of V to I and V/I to C is subject to the Head Movement Constraint, and is triggered by the necessity for C to be lexically realized. In root clauses this is done by movement of V/I; in non-root clauses by a base-generated complementizer.2 The notion that C need be lexically realized is found in various guises in recent proposals for the analysis of verbsecond.3 We will come back to this in the next section. Non-verb-second languages (Modern English and Modern French) are characterized by movement of Vf to I (or vice versa4), and no further movement to C, except under restricted circumstances such as in Modern English interrogative contexts and Modern French subject-clitic-inversion constructions. In languages that are not CV2, one can still find verb-second word orders, as Santorini (1989) makes clear with respect to the history of Yiddish. On the assumption that in the structure (1) above, the Spec,I' position can be an XP position rather than an NP (subject) position as in the standard analysis, an initial XP position followed by Vf may reflect an IP rather than a CP.5 This is discussed in detail in Santorini. It is crucial, though, that in such instances there is no root/non-root asymmetry. Thus, in such languages, XP-Vf-subject sequences occur freely in both root and non-root contexts. However, these are to be analyzed as instances of V to I. The asymmetry between root and non-root contexts, which reflects movement of V/I to C, can only emerge in a CV2 language. The essential difference between verb-second (CP) and non-verb-second languages then is that in verb-second languages Comp is always lexically realized, whereas this is not the case in non-verb-second languages. We take it that this is a reflection of the fact that in verb-second languages C has a number of crucial head properties that in non-verb-second language are characteristic of I. For instance, in a number of verb-second languages we find morphological agreement between Comp and the finite verb.6 More evidence comes from West Flemish (WF), another verb-second language, where object pronouns can be cliticized onto Comp, cf. Bennis and Haegeman (1983).7 We conclude that in verb-second languages, C is in some sense a more prominent head than I, whereas in non-verb-second languages the reverse seems to be the case. We will henceforth call this C-orientation (verb-second languages) and I-orientation (cf. also Hulk and van Kemenade (1988); Tomaselli (1990)). In what follows we will see that this distinction, interacting with other independent phenomena, underlies a variety of other differences between the two "types" of languages.

230

VERB SECOND AND THE NULL-SUBJECT PARAMETER

1.2 Licensing Theory In this section we outline a theory of licensing for functional properties. By functional properties, we mean: the identification of functional heads such as C and I, case-marking properties and subject-verb agreement. For functional heads such as C and I we assume the following licensing condition, given the distinction between C and I-orientation of the previous section: (3)

the dominant functional head must be lexicalized

(3) will trigger, in a C-oriented language, movement of V/I to C in the absence of a base-generated complementizer; in an I-oriented language such as Modern French, it will trigger V to I movement.8 We further assume the following condition for the functional properties of both lexical and functional heads: (4) functional features must be licensed The functional features Case and agreement must be identified structurally or morphologically.9 Let us consider in some more detail how (3) and (4) are instantiated. 1.2,1 NP Licensing: Case NP is typically licensed by being Case-marked. Case on NP is realized either syntactically (i.e. structurally) or morphologically. In this article, we concentrate on Nominative Case. We assume that structural Case is assigned in the following configuration: (5) structural Case

In (5), a Case-marking head X assigns structural Case under government to the dependent NP, either to the left or to the right. Nominative is considered a structural Case and is thus an instantiation of the structure in (5). We assume the definition (6): (6) NP -> Nominative if it is governed by the dominant head C or I According to (6), Nominative Case is assigned by C to Spec,IP in C-oriented languages, and by I to Spec, VP in I-oriented languages. Within this view, the movement of the subject from Spec.VP to Spec, IP in I-oriented languages cannot be motivated by Case-theoretic reasons. We suggest in the following section that the trigger for this movement is (p-feature agreement.

VERB SECOND, PRO-DROP, FUNCTIONAL PROJECTIONS

231

7.2.2 Licensing of the Functional Properties of IP and VP Beside the requirement of lexicalization, where relevant imposed by condition (3) above, (4) demands identification of the functional features of I(P). Functional features are agreement features for person, number, gender which we shall henceforth call (p-features. O-features, like Case, can be identified syntactically or morphologically. We propose the following parametric options: (7) I licensing; (p-features must be identified a. syntactically: Spec-head agreement with NP in Spec,IP b. morphologically: (p-features are realized morphologically c. default licensing: impersonal form Only in cases (7a) and (7b) is I what we call "fully licensed". (7c) represents a special case of licensing that we explicate below. Licensing of the functional features of V, i.e. its structural Case-marking properties, depends crucially on I; we propose the following condition for licensing: (8) V in VP can assign structural Case iff it is governed by a fully licensed I Let us consider how (7) and (8) work and interact in specific instances. (7a) represents subject-Infl agreement, the (p-features of I are identified by means of local agreement with NP in Spec.IP position.10 We suggest that this is the trigger for movement of the subject NP to Spec.IP in I-oriented languages. (7b) defines morphological identification: inherent or morphological (p-features represents the case of languages such as Modern Italian where agreement morphology is sufficiently rich to make (p-features inherently visible. Default licensing (7c) is a special case of licensing and typically represents the case of impersonal sentences: there is no nominative subject; there is no verbal agreement as I assumes a default form, third person singular; thus I is not fully licensed and V in such cases cannot assign structural case. It follows that V in such examples has NP complements only in languages that have morphological licensers: Case endings, morphological (p-features. Let us see how this works with a few examples from Modern Dutch and Old English: (9) a. op het plein werd gedanst in the square was danced 'there was dancing in the square' b. hine (A) nanes pinges (G) ne lyste on pisse worulde him nothing not pleased in this world 'nothing in this world pleased him'. (Boeth. 102, 9) The Modern Dutch sentence (9a) is an impersonal passive. Thus, there is no thematic subject, and the finite verb werd has the default third person singular form. I is then not fully licensed, as there is no structural or morphological licenser. Therefore V gedanst cannot assign structural Case and thus cannot have any direct object as V cannot assign structural Case. Hence, only a PP

232

VERB SECOND AND THE NULL-SUBJECT PARAMETER

op de tafel can complement it, in the absence of morphological case in Modern Dutch, as a potential morphological licenser for NP. In contrast, the Old English sentence (9b) does have verbal complements, but note that these are identified by overt morphological marking. 1.2.3 Licensing of Pro: Pro-drop For the analysis of pro-drop, a distinction between full pro-drop and expletive pro-drop is customary. Full pro-drop appears in languages such as Modern Italian, where the subject pronoun can remain lexically null, even when it has a thematic role. This is exemplified in (10) with some sentences from Italian, but the phenomenon also occurs in Spanish and Rumanian, and in Old French in a restricted form. (10) a. hanno parlato troppo have talked-3pl too much 'they talked too much' b. ti conosco you know-lsg 'I know you'. We define full pro-drop then, as "omission" of a subject that has a thematic role, as opposed to expletive pro-drop, which involves "omission" of a nonthematic subject, as in impersonal sentences such as those in (9) above. Expletive pro-drop is found in a number of Germanic languages. Examples from Modern Dutch and Old and Middle English are those in (9) above and those in (11): (11) a. and swa miclum sniwde swelce micel flys feolle and 0 so heavily snowed as if much fleece fell 'and it snowed so heavily, as if a lot of fleece were falling' (Epist. Alex. 159, 538) b.

... oaette foroy to ungemetlice ne sie gliood that therefore 0 too greatly not be mitigated oaem scyldgan the guilty 'that therefore it must not be mitigated too greatly to the guilty' (CP, 151, 2)

c. penne scheomep me (obj) perwip then shames me with that 'then I am ashamed of that' (St. Marh, 34, 30) d.

him (obj) wile sone longe parafter him wile soon long after that 'he will soon long for that' (Trin., 148, 19)

VERB SECOND, PRO-DROP, FUNCTIONAL PROJECTIONS

233

Similar examples can be adduced from Middle Dutch, Modern German as well

as older stages of German and Modern Icelandic, but we will not do this here for reasons of space. We now discuss how the various types of pro-drop should be analyzed. Our analysis of pro-drop essentially follows up the classical analysis of full pro-drop as in Italian (Taraldsen (1978); Chomsky (1982); Rizzi (1982, 1986)) and of expletive pro-drop as in the Scandinavian languages analyzed by Platzack (1987). The option of having an empty subject at all depends on the possibility of identifying it. Following Rizzi (1982, 1986) we assume that a pro-subject must satisfy two conditions: it must be formally licensed, and its content (i.e. (p-features) must be identified. Rizzi (1986) defines these conditions as follows: (12) pro-licensing: a. pro must be formally licensed under government by a designated Case-assigning head b. the content of pro ((p-features) must be identified. Notice that (12a) in fact imposes two requirements on the governing head: first of all that it be a "designated" head (in Rizzi (1982) the head is designated by virtue of a feature specification [+ pronominal]); secondly that the head be a Case-assigner. The term "designated" indicates that such a head is marked for pro-licensing language-specifically, i.e. presumably each language has an inventory of heads that are termed designated; it is even quite possible that languages lack such an inventory entirely. The latter seems to be the case in Modern English. Rizzi (1986) considers that the licensing conditions for pro (12a-b) should be satisfied by one and the same head. We propose that the designated Case-assigning head in (12a) coincides crucially with the dominant head for C/I orientation in terms of our theory. In Coriented languages, C is the relevant governor/Case marker, and can be listed as the designated head; in CV2 languages where C is a designated head, it formally licenses pro in Spec,IP. Modern Dutch is an example of such a language. In I-oriented languages, I, if listed as a designated head, is the relevant governor and can license pro in Spec.VP. Modern Spanish and Modern Italian are cases in point. We furthermore recall the definition for Nominative Case ((6) above), which states that Nominative is assigned under government by the dominant head C or I. Finally, we propose that content-licensing of pro (12b) is by the (p-features of I, if these are morphologically realized. With these theoretical preliminaries in mind we can consider the question of what it is that determines whether a language has full or expletive pro-drop characteristics. This is related to the fact that a pronominal I, but not a pronominal C, may contain (p-features: [±person, ±number, ±gender]. We follow Rizzi (1982:42) in assuming that it is the specification [+person] that is necessary to license referential pro-drop. If a pronominal I does not have the [+person] specification, only expletive pro-drop is possible. In I-oriented languages, a pronominal I may optionally be specified [+person]. Therefore, we

234

VERB SECOND AND THE NULL-SUBJECT PARAMETER

find both referential pro-drop (13) and expletive pro-drop (14) in these languages: (13) hanno parlato have spoken 'they have spoken' (14) Sembra che gli amici sono arrivati seems that the friends have arrived "It seems that the friends have arrived'. When I or C is a designated head, and there are no p-features under I, we find expletive pro-drop. Pro can be formally licensed, but there are no (p-features that must be identified, hence formal licensing is sufficient. C-oriented languages for which this is relevant, are Modern Dutch and earlier stages of English, as exemplified above. We will discuss the latter in more detail below. The possibilities for pro-drop discussed are summarized in (15), with mention of some of the core instantiations of the various options.11 (15) MoDu/OE/ME Cont. Scand. Ital/Span MoFr/MoE Icelandic

V2

Comp

Infl

+ + +

+DH -DH -DH

-phi - Phi +DH+phi -DH-phi +DH-phi

expl.pro-drop no pro-drop full pro-drop no pro-drop expl.pro-drop

The table in (15) is not exhaustive. We shall see below that Old French, which cannot be accommodated in (15) in any obvious way, is an interesting mixture of the COMP and INFL properties listed. We discuss this in SECTION 2.1.

1.3 Functional Categories and Language Change Our starting point is the principles and parameters approach to language change that has become current in generative grammar, subsequent to the pioneering work of Lightfoot (1979). In the vein of this work syntactic change has often been analyzed as a reanalysis from one parameter value to the other. The scenario of a particular syntactic change is then one that involves a sharp turning point. There is one specific problem with this kind of scenario, noted among others by Weerman (1989), viz. that it is not easily reconciled with the gradual picture that is usually presented by the factual situation in language change. Weerman's alternative, which we do not wish to discuss further here, involves an analysis of the diachrony of verb-second in the Germanic languages that makes crucial use of one functional projection CP as a matter of principle. So far, we have assumed two types of verb-second language: CV2 languages that have a root/non-root asymmetry with respect to verb second and IV2 languages which do not have this asymmetry. The analysis that we present in SECTION 2 of the history of French and English crucially relies on

VERB SECOND, PRO-DROP, FUNCTIONAL PROJECTIONS

235

these two types of verb-second language. We shall argue that French changed from a CV2 language with full pro-drop in root contexts through an IV2 stage with postverbal subjects and pro-drop in both root and non-root contexts, to non-verb second, i.e. Modern French. This scenario involves a number of phenomena that will be difficult to account for in a framework with only one functional projection CP. One may even want to carry this further and hypothesize that functional projections may yield an interesting account of synchronic variation. The theoretical framework we assume also gives the motivation for why English did not go through this intermediate stage and changed from CV2 to non-verb second. Thus, we will see that the analysis we present in the following sections provides crucial support for distinguishing at least two functional projections above VP.

2. Two Cases of Language Change In this section we consider the history of French and English with respect to verb second and to pro-drop. We shall argue, following up work by Adams, Vance, Hirschbuhler, Dupuis, and Lemieux on Old French (OF), and by van Kemenade on Old English (OE), that both OF and OE are C-oriented languages with verb second. In this respect the two languages are very similar. They differ however, in that OF has referential pro-drop in root contexts only and expletive pro-drop that is not limited to root contexts, whereas OE has only expletive pro-drop, no referential pro-drop. While the functioning of verb second is the same in the two languages, this difference, we argue, is to be attributed to the fact that in OF, unlike in OE, I has the appropriate morphological characteristics to identify a pro subject. 2.1 The History of French 2.1.1 Old French Recently, there has been an upsurge of interest in OF from the perspective of GB-theory, no doubt instigated by the work of Marianne Adams (1987, 1988a, 1988b), who gives a systematic account in generative terms of a number of observations that were earlier made by traditional grammarians. Adams analyzes OF as a verb-second language with referential pro-drop, but interestingly, referential pro-drop is restricted to root clauses with XP-Vf-subject contexts. Some of the core examples for this pattern are given in (16) and (17): (16) Verb second in root clauses a. Ensi fut Joseph perdus une grant piece Thus was Joseph lost for a long while 'Thus Joseph was lost for a long time' (R.Gr. 27)

236

VERB SECOND AND THE NULL-SUBJECT PARAMETER

b.

Messe e matines at li reis escoltet. Mass and matins has the king heard 'the king attended mass and matins' (Rol. 670)

(17) Verb-second root clauses with referential pro-drop: a. or fait senblant con s' ele plore now makes (she) seem as if she cries 'now she pretends to cry' (Tristan, 1.8) b. Einsi corurent par mer tant que il vindrent a Cademelee thus ran (they) by sea until they came to Cadmee 'thus the ran by the sea until they got to C.' (Vil. XXV) The examples are all from the secondary literature quoted. The example (17b) is particularly interesting in that it has a null subject in the root clause and a lexically realized (co-referential) subject in the non-root clause. Adams analyses OF crucially as a verb-second language with SVO basic order that is, in our sense, a C-oriented language. Vf moves to I, V/I moves to C in root clauses in the absence of a base-generated complementizer. The fact that referential pro-drop is restricted to root contexts with verb second is explained in terms of directional government: SVO languages like OF have directional government from left to right, hence the licensing of the (p-features of pro is also from left to right, by Vf in C, hence we only find pro-drop in root clauses in contexts to the right of Vf. There are various problems with this analysis, as subsequent research by Adams and others has shown. For one thing, the directionality parameter in its strictest sense cannot be maintained, as Old High German (Lenerz (1989)) is an SOV language (thus with government by V from right to left in Adams's terms) with roughly the same null subject facts as OF. OF is a language that has a root/non-root asymmetry with respect to verbsecond and with respect to referential pro-drop. Let us first consider how this basic fact can be captured in our theory. OF is a C-oriented language with movement of Vf to I, and V/I to C. Movement of V/I to C is derived by condition (3) above, that states that the dominant functional head in any language must be lexicalized. Given the fact that OF is CV2, and given our licensing conditions for pro (12), we predict that C is the designated head that acts as a formal licenser for pro. But C cannot license the (p-features of pro, as these are always generated under I. However, in root clauses, i.e. in precisely those contexts where we find referential prodrop, I moves to C. As a result, C (+I+V) has p-features, and thus licenses the content of pro. In other words, our theory predicts that a C-oriented verbsecond language where C can act as a formal licenser for pro and I has (pfeatures, will have referential pro-drop in root contexts only. It also predicts that there is no root/non-root asymmetry with respect to expletive pro-drop, as content-licensing for pro is irrelevant there, cf. Vance (1989:119ff.). (18) shows that this is correct with some examples of expletive pro-drop in nonroot clauses:

VERB SECOND, PRO-DROP, FUNCTIONAL PROJECTIONS

237

(18) a. quant vos plera, ge m'en irai when (it) you pleases, I there will go 'I will go there when it pleases you' (Artu 158038) b. quant vint a eure -de midi when (it) comes to hour of noon 'when the hour of noon comes'. (Vance (1988)) We will see below that this analysis, though basically correct for OF, will have to be refined considerably if we consider more details of OF, and details of the further history of French. After all, we do not just want to give an account of OF. We want to see also how the further developments in the history of French can be accounted for in a plausible way; thus we want to explain what changes took place and how they could take place. Since Adams (1987) a number of proposals for OF and its further history have been advanced that urge some modifications in the scenario outlined by Adams (1987). Following, we will discuss the facts pertaining to this. Under rather more exceptional conditions, we find embedded verb-second phenomena in OF: the finite verb is preceded by a constituent other than the subject and the subject is either absent or in postverbal position. According to Adams (1987); Vance.(1988) and Hirschbuhler and Junker (1988) this is found in embedded'sentences introduced by the complementizer que, after bridge verbs. This situation arises also in other verb-second languages, e.g. Icelandic and Old English, as discussed below: (19) a. carjesai bien que 1'aventure acheveroiz vos legierement as I know well that the adventure end you easily 'as I know well that you will end the adventure easily' (Queste) b. il respondirent que de ceste nouvele sont il moult lie they answered that of that news are they very happy 'they anwered that they were very glad of that news'. (Artu 45) In the literature two different analyses of embedded verb-second phenomena have been proposed. Here we will adopt the one defended for OF by Vance (1988) who follows up work on Icelandic by Sigur5sson (1989) and Platzack (1983). We assume that in the embedded clause the complementizer C may exceptionally be followed by a complete CP. This permits us to extend the analysis of verb-second in main clauses to embedded clauses: the finite verb first moves to I and subsequently V+I move to C.12 The subject moves to Spec.IP and the subject or another constituent moves to Spec.CP. (20) [cque [CP1'a venturei [cacheveroizj [IP vos [ I t j [ t j t i legierement]]]]]] Another analysis of embedded verb-second phenomena was proposed for OF by Hirschbuhler and Junker and Lemieux and Dupuis (this volume). They follow Diesing's (1990) analysis of Yiddish and suggest that in embedded clauses the subject remains in its base position in Spec,VP and another constituent moves to Spec.IP which is then viewed as an XP position. Thus, in our terms, they analyze OF as an IV2 language.

238

VERB SECOND AND THE NULL-SUBJECT PARAMETER

(21) [cque [IP[l'aventure]j [I [vacheveroiz]i[vpvos [v. t i t j legierement

]]]]]

This analysis, however, raises some major problems (see also den Besten (1989)). We have argued above that OF is a C-oriented language in which Nominative Case is assigned under government by [+tensed] C. It is unclear, however, how the base-generated subject in (21) will get its Nominative Case: in the Spec,VP position it is not (minimally) governed by C. It cannot receive Nominative Case under government by I, since then: - the root/non-root asymmetry is inexplicable - the insight that embedded verb-second is exceptional is lost - the difference between C-orientation and I-orientation becomes rather unclear - the differences between OF (a C-oriented language) and Modern Italian (an I-oriented language), with respect both to inversion and pro-drop, cannot be accounted for. Summarizing, such an analysis is not compatible with the C-oriented character of OF and must therefore be rejected.13 Our analysis predicts that referential pro-drop is impossible in non-root clauses in OF, as the finite verb moves only to I and therefore C has no 9features. SpecJP is governed by lexical C and receives Nominative Case. Pro in Spec,IP can only be licensed formally, since it is not governed by an element containing p-features, and the option of identification under Spec-head agreement in IP is not available in OF. There are two classes of apparent counter-examples to this. First, we do exceptionally find referential pro-drop in embedded contexts, i.e. those contexts which were analyzed in the previous subsection as CP embedded under C. Given this analysis, pro-drop is expected rather than surprising. An example is given in (22): (22) a. et disoies que ja en ceste maleurte" ne charroies and said that never in this misfortune neg will fall-2sg 'and said that you will never fall into this misfortune' (Queste 123) b. mes puis que je vois que a fere le covient but since that I see that to do it is fitting-3sg 'but since I see that it is fitting to do it'. (Queste 25) Our analysis in terms of lexicalization of the dominant functional head, and movement of I with morphological (p-features to C in root clauses is different from that of Adams (1987) in several respects, but enables us to maintain the basic insight that there is a root/non-root asymmetry with respect to verb second and referential pro-drop. By lexicalizing C, verb second creates the context for formal licensing of pro in Spec,IP position. By moving I (with (p-features) to C, verb second creates the context for content-licensing of the p-features of pro. This is the usual situation in root clauses. By regarding CP embedded under a bridge verb as a root clause too, a set of apparent counterexamples is explained. The second set of counter-examples is one that requires closer inspection. We will examine this in more detail separately.

VERB SECOND, PRO-DROP, FUNCTIONAL PROJECTIONS

239

Although Adams (1987) assumes that pro-drop was impossible in embedded IP in OF, subsequent work by Adams (1988a, 1988b), Vance (1989), Dupuis (1988), Hirschbuhler (1990, this volume) has shown that this assumption cannot be maintained. According to Dupuis null subjects in a position immediately preceding the finite verb (a non-verb-second context) were possible in specific genres even in OF (23). Others (Vance 1988) relate this possibility in OF to the type of complementizer ("lexical" versus que) (24). (23) ainz que m'en aille en France such that (I) away went to France 'such that I went away to France' (Aymeri de Narbonne 204) (24) 1'espee dont s' estoit ocis the sword with which (he) himself had stabbed 'the sword with which he had stabbed himself. (Chast. 913) Hirschbuhler and Junker (1988) and Vance (1988) assume that this type of prodrop was extremely rare in OF and became more and more frequent in the 14th and 15th century, not only in non-root clauses but also in root clauses. However, by that time several other changes had already taken place. We will discuss this further below. Some examples from Middle French are given in (25). (25) a. puet bien estre que n' en avez point can well be that not-of-it have (you) at all 'it may well be that you have none of it' (Jehan de Saintre 8) b. mais que soions en la chambre but that are (we) in the room 'but that we are in the room' (Jehan de Saintre 7) c. ...que 1' eusse reconneu that him-had (he) recognized 'that he had recognized him'. (CH 329/30) Referential pro-drop of this type is difficult to account for in the present analysis, as we propose that lexical C is the formal licenser for pro, but the content-licenser only in those contexts where V/I has moved to C, i.e. only in root clauses. The data in (25) suggest that I in IP can license the content of pro, as there is no movement of V/I to C and (p-features are generated under I. We have to show then how these facts can be made compatible with our analysis, though doing justice to the fact that they are rare. We could hypothesize that, though they are of a different nature, the two licensing conditions for pro—formal-licensing and content-licensing as formulated in (12)—are satisfied by the same head in the unmarked situation. This is reflected in a language like Modern Standard Italian: both conditions are satisfied by I. This is reflected also in the vast majority of cases in OF: both conditions are satisfied by C. However, we hypothesize that in OF we also, infrequently, find the marked situation: pro in Spec,IP is formally licensed

240

VERB SECOND AND THE NULL-SUBJECT PARAMETER

under government by C, but its p-features are licensed by I. This may present an interesting starting point for the developments seen in Middle French (MidF), which we deal with in the next subsection. 2.7.2 Middle French If we compare OF to the MidF situation with respect to verb second and pro-drop, we consider that there are two important differences: 1) the increase of verb-third structures and 2) the increase of embedded verb-second stuctures. In the early MidF period we find an increase in "verb third" structures in root clauses, indicating that topicalization of XP no longer goes hand in hand with fronting of Vf to C (cf. Adams (1988a,b); Kroch (1989); and Roberts (1992), among others). (26) a. et lors ils commencerent a rire and then they began to laugh (Q 64) b. car encors vous n'avez point... for yet you not-have at all... 'for as yet you don't have any...'. (Jehan de Saintre 58) Fronting was no longer necessarily movement to Spec,CP, but could also be realized by adjunction to IP. For quite some time these two fronting operations co-existed (cf. Vance (1988, 1989)).14 Adams (1988a,b) notes that in verb-third structures subject pronouns were far more frequent than subject NPs. The status of the subject pronouns is one of the things that changed in the period under consideration (late OF/early MidF). In OF subject pronouns were independent elements, comparable to NPs. Gradually, however, they lost their independent status and became clitics.15 We suggest (but see also Adams (1987)) that the gradual cliticization of subject pronouns resulted in ambiguity: the sequence XP-subject pronounVf could be interpreted in two ways: as having an XP fronted to CP and cliticVf fronted to C, or as having XP adjoined to IP, with Vf fronted to I. This structural ambiguity between a CP or IP interpretation for verb-third structures with subject pronouns paved the way for verb-third structures with NP subjects. The latter could only be interpreted as IP. This, we claim, was the major trigger for the loss of verb second, which we analyze as a shift from C-orientation to I-orientation taking place in the course of the MidF period. We will now consider this more closely. Whereas in OF verb second in non-root clauses was exceptional, as discussed above, there was an increase in these structures in the MidF period, cf. Lemieux and Dupuis (this volume). Hirschbuhler and Junker (1988) report that in Vigneulle' Les Cent Nouvelles Nouvelles, verb second occurs fairly freely in embedded clauses, after bridge and non-bridge verbs. Consider the examples in (27) (from Hirschbuhler and Junker):

VERB SECOND, PRO-DROP, FUNCTIONAL PROJECTIONS

241

(27) a. quand elle veit qu' aultre reponse ne pouvait avoir de luy (Vigneulles099112) when she saw that other answer not could have from him 'when she realized that she would not get any other answer from him' b. promist que ainsi feroitilz (Vigneulles 077041) promised that thus would-do they 'promised that they would act like that'. Recall that these patterns already existed in OF, though with an unambiguous CP interpretation. The extension of verb second to clauses embedded under non-bridge verbs should be interpreted as a shift to IV2, we would argue. As noted above, the causal factor responsible for the increase was the cliticization of subject pronouns, resulting in ambiguity between a CP and an IP interpretation; as an IP interpretation for the verb-second phenomenon became available, its increase in non-root clauses is expected. Note that this constitutes a reinterpretation of the infrequent CP pattern of OF as an IP pattern. At the same time, the increase of verb second in non-root clauses further weakens the evidence for a root/non-root asymmetry with respect to verb second. During the same period an increasing number of verb-first sentences are reported (Vance (1988); Dupuis (1988)); some examples: (28) a. et se retira le roy a Corbeil and himself withdrew the king to Corbeil 'and the king retired to Corbeil' (Commynes 62, 2) b. et luy fut adoubee sa playe qu'il avoit au col and to-him was dressed his wound that-he had in the neck 'and his neck wound was dressed'. (Commynes 63, 20) These two patterns—verb-first structures and the loss of the root/non-root asymmetry—are characteristic of languages which are typically analyzed as IV2, see for instance Santorini's (1989) analysis of the history of Yiddish. What we propose then is that while OF was a CV2 language with a root/ non-root asymmetry with respect to verb second and pro-drop, the cliticization of subject pronouns, resulting in ambiguity between a CP and IP interpretation, triggered a shift to IV2. Verb-second structures in MidF are then analyzed as follows: (29) IP [ XP [I Vf [VP NP [v, Vj ]]]]

Vf moves to I; XP moves to Spec,IP; the subject remains in its base-generated position. Evidence for this VP-internal position is given by Lemieux and Dupuis (this volume) who cite examples in which a VP-adjoined adverb intervenes between Vf and the postverbal subject:

242

VERB SECOND AND THE NULL-SUBJECT PARAMETER

(30) Et la and then canons canons 'and then

commenca fort le trait d' arbalestres et de started violently the firing of crossbows and of (Melusine, p. 89) started violent firing of crossbows and canons'.

This shows that the subject can get Nominative Case in its base-generated position. We will come back to this below. The Spec.IP position is no longer the position to which the subject must move in order to get Nominative Case. It is not a 9-position either. We therefore suggest that Spec.IP has become an optional position; it can be the landing site for a preposed XP, as in (29) or it may just be omitted, resulting in a verb-first structure. An exception to this is a preposed XP that is a w/z-phrase; in constructions with this type of operator, Spec.CP is the landing site. In the MidF period, both referential and expletive pro-drop were possible, both in verb-second and verb-first contexts (the latter mainly in embedded clauses). With respect to the analysis of pro-drop this suggests that, first of all, I still had inherent p-features. Although the fact that referential pro-drop in verb-first structures appeared to be restricted to nous/vous drop in embedded clauses whereas expletive pro-drop was possible both in matrix and in embedded clauses suggests that the p-features of I were gradually becoming less "rich".16 Above, we have proposed to relate null-subject licensing indirectly to Nominative Case-marking (recall that formal licensing of pro under (10) is subject to government by a designated Case-marking head, cf. Rizzi (1986)). We suggested that in OF Nominative Case was assigned to Spec,IP under government by a lexical C. Consequently null subjects were also licensed in that position under government by C. We argued that MidF is an I-oriented language in which the subject can receive Case in its base-generated position. Under the definition of Nominative Case-marking, (6) above, I in an I-oriented language can assign Nominative Case under government or under Spechead agreement in IP. This implies that at least the first option must have been available in MidF, in order for the postverbal subject to receive Case, (cf. also Roberts (1992)). If this is so, we predict that in this position null subjects could be licensed too, as they were formally licensed by I as the designated Casemarking head, and content-licensed by the same I. This is schematized in (31).

VERB SECOND, PRO-DROP, FUNCTIONAL PROJECTIONS

243

In fact, this appears to be borne out: Hirschbuhler and Junker (1988) give evidence that the position of pro in, for example, Vigneulles is postverbal, as it alternates only with postverbal lexical subjects. Recall, however, the exceptional cases of referential pro-drop in non-root clauses in OF. In OF pro is formally licensed by a lexical C, and we suggested that OF had a marked option of (p-identifying pro under Spec-head agreement in IP. Since this option was already marginally available in OF, it seems plausible that it remained available in MidF, the more so because in MidF I had become the formal licenser. However, licensing of p-features under Spec-head agreement could never be the only option, as postverbal subjects in Spec.VP were possible too.17 So in fact null subjects in MidF could be found in two positions: in Spec.IP and in Spec.VP. In both these positions pro is formally licensed by I under government;18 in Spec.IP the p-features are identified under Spec-head agreement; in Spec.VP the p-features are identified under government.19 Summarizing, our analysis for OF and MidF is as follows: (32) Old French: C-oriented; Verb movement: V to I, V/I to C; Nominative Case: government by C Fro-licensing: formal: government by C content: government by V/I/C in roots Middle French: I-oriented; Verb movement: V to I Nominative case: government by I Fro-licensing: formal: government by I content: government by I/Spec-Head Agreement in IP 2.7.3 From IV2 to Modern French At the IV2 stage of French which allows (referential) pro-drop I is evidently still licensed inherently, by its morphological "richness". Gradually however, I loses this inherent morphological characteristic and consequently licensing by Spec-head agreement in IP becomes obligatory, hence a subject NP in Spec.IP becomes obligatory. The obligatoriness of Spec-head agreement forces movement of the subject from Spec.VP to Spec.IP. Movement of XP to Spec.IP is ungrammatical because only NP with appropriate p-features can license I. As the subject comes to surface in Spec.IP always, Nominative Case comes to be assigned exclusively under Spec-head agreement in IP.20 Conversely, as inherent p-features are lost, null subjects tend to disappear: since I no longer has (p-features, it can no longer license a null subject in Spec.VP or in Spec.IP. To conclude, we argue for essentially a two-step development in the history of French where verb second and pro-drop are concerned: first, French changed from a C-oriented to an I-oriented language, thereby loosing the root/ non-root asymmetry with respect to verb second and pro-drop; second, several centuries later French lost the ability to license null subjects, which is essentially due to the loss of morphologically realized p-features. This see-

244

VERB SECOND AND THE NULL-SUBJECT PARAMETER

nario makes crucial use of two functional projections CP and IP, which both allow verb-second facts, given certain Infl properties. This distinction explains the contrast between Old and Middle French with respect to verb second and pro-drop: the root/non-root asymmetry in OF versus the development towards symmetry in MidF . Our assumptions about the licensing of p-features explain why the erosion of agreement morphology on V/I led to the loss of pro-drop and the obligatoriness of NP subject in Spec,IP. Our analysis thus argues for at least two functional projections.

2.2 The History of English 2.2.1 Verb Second, Pro-drop and Nominative Case in OE The history of verb second in English is treated in considerable detail by van Kemenade (1987).21 OE is analyzed as SOV underlyingly, with V-fronting in root clauses. As an initial illustration, we give some examples in (33):22 (33) a. hwi wolde God swa lytles binges him forwyrnan (AHTh, I,14) why would God so small thing him deny 'why should God deny him such a small thing?' b. for baes wintres cyle nolde

se asolcena erian (AEHom, 17,116) for the winter's cold not-wanted the layabout plough 'the layabout didn't want to plough because of the cold'.

It seems clear that with respect to verb second, there is a root/non-root asymmetry in OE; as far as we are aware, the pattern XP-Vf-Subject-(XP)-V-(XP) only occurs in root clauses or after bridge verbs, i.e. in embedded root clauses.23 An example of the latter is given in (34): (34) Gregorius se trahtnere cwaeobaet foroi wolde drihten Gregory the interpreter said that therefore wanted God getrahtnian burn hine sylfne baet bigspel oe... (AECHom II,549.219) interpret through himself the parable that... If OE has a root/non-root asymmetry with respect to verb second, the immediate consequence is that OE is a C-oriented language, given our theory. This means that Nominative Case is assigned under government by C. We saw above that OE also has expletive pro-drop. We repeat some examples here for convenience. (35) a. hine (A) nanes binges (G) ne lyste on pisse worulde him nothing not pleased in this world 'nothing in this world pleased him' (Boeth., 102, 9) b. and swa miclum sniwde swelce micel flys feolle and 0 so heavily snowed as if much fleece fell 'and it snowed so heavily, as if a lot of fleece were falling' (Epist. Alex. 159, 538)

VERB SECOND, PRO-DROP, FUNCTIONAL PROJECTIONS

245

c. ...oaette foroy to ungemetlice ne sie gliood that 0 therefore too greatly not be mitigated oaem scyldgan (dat) the guilty 'that therefore it must not be mitigated too greatly to the guilty'. (CP, 151, 2) (35a) is an example of expletive pro-drop with a psych-verb; (35b) with a weather verb; (35c) with an impersonal passive. The phenomenon of referential pro-drop does not occur in OE. Given our theory, we analyze the OE situation as follows: the basic structure (36) obtains:

OE is a C-oriented language, hence C must be lexical. Nominative Case is assigned by C under government. This forces movement of the subject NP, base-generated in the specifier of V, to Spec.IP, as this is the only way in which it can receive Case. C counts as a designated head for the licensing of pro, hence we expect to find expletive pro-drop in root clauses as well as non-root clauses. This is borne out by the facts. To emphasize this point, we give a few extra examples in (37): (37a) is a root clause; (37b) a non-root clause: (37) a. bonne ofpyncp him (D) baeg ilcan (G) be he aer forbaer then regrets him the same that he before endured 'then he regrets what he endured before' (CP 225,19) b. gif us (D) ne lyst oaera aerrena yfela (G) 5e we if us not pleases the earlier evil that we aer worhton earlier wrought 'if the evil that we first wrought displeases us'. (CP, 445, 29)

246

VERB SECOND AND THE NULL-SUBJECT PARAMETER

The absence of referential pro-drop in OE indicates that I cannot be inherently licensed, hence it must be licensed under structural conditions by specifier-head agreement in IP, i.e. agreement with the subject. This approach implies that the position of the subject in OE is redundantly derived by two separate principles: Nominative Case-marking and licensing of the I-projection. We will now consider how default licensing is instantiated in OE. The examples involving expletive pro-drop (37) are all examples where I is not fully licensed: there is no morphological licensing (no inherent p-features) or syntactic licensing (no NP that ensures Spec-head agreement in IP). Our theory predicts that in such instances licensing of VP is impossible, hence V cannot assign structural Case. Complements of V in this construction should therefore only occur with morphological case-marking; the morphological option of NP-licensing. This is in fact what we find: in (37) the complements bear dative and genitive Case respectively. Thus we see that with respect to OE, our theory makes interesting predictions that are borne out in detail. We will now consider the developments in these constructions subsequent to OE. 2.2.2 Middle English Developments It is well known that English changed from an SOV language to an SVO language. This change in the underlying structure should be dated around 1200 (cf., Canale (1978); van Kemenade (1987)). This change in basic word order, however, did not affect the verb-second status of the language. SVO order with fronting of the finite verb and topicalization of some constituent persisted well into the 14th century. Some examples of this are given in (38): (38) a. On bis gear wolde pe king Stephne taecen Rodbert,... in this year wanted the king Stephen seize Robert 'in this year king Stephen wanted to seize Robert' (PC 1140, 1) b. Alswa scal be laroeu don pe oet... also shall the teacher do who that... 'likewise the teacher who...shall do' (OEH, 95) c. Thus may thine instrument last perpetuel thus may your instrument last perpetually 'thus your instrument can last perpetually' (EP, 42) d. and now is my sonne gon to reste as for that Saturday and now is my sun gone to rest as for that Saturday 'and now as for that Saturday my sun has gone to rest' (Ch.TA, II, 12, 29) e. Ofte schal a womman have thing which... often shall a woman have things which... 'often a woman must have things which... (GowerCA, 3206, 123) Given our theory, we expect the development concerning verb second to go hand-in-hand with the development concerning expletive pro-drop. Our data

VERB SECOND, PRO-DROP, FUNCTIONAL PROJECTIONS

247

show that we find expletive pro-drop until the early 15th century; this is borne out by such data studies as van der Gaaf (1904) and Elmer (1981);24 some ME data are given in (39): (39) a. benne scheomep me (Obj) berwib then shames me with that 'then I am ashamed of that' (St. Marh, 34, 30) b. him (Obj) wile sone longe barafter him will soon long after that 'he will soon long for that'. (Trin., 148,19) With respect to constructions like those in (39), it is important to realize that the emerging obligatorinesss of a Nominative subject is only part of the story. While in OE the range of complements and cases available to subjectless verbs was a lot wider than, for example, in the 13th century, due to the loss f morphological case, these verbs could appear with complements, as (3 ) shows: in both sentences we find an object pronoun and a PP complement. The most important observation with respect to the loss of expletive pro-drop however, is the obligatory presence or not of a Nominative subject. The Nominative subject became obligatory in the early 15th century. From the middle of the 14th century, verb second declines. As to the primary cause for this, we have to go back to a peculiarity of verb second in earlier times. While in OE verb second, as analyzed here, is fairly regular given that patterns like XP-Vf-subject NP-(...)V are the norm, this pattern is not followed if the subject is a personal pronoun. In that case the subject is found at the left of the finite V. Van Kemenade (1987) analyzes this as a process of cliticization of the pronominal subject to V/I. We do not wish to go into the nature of this process here. It suffices for our purposes that there is a systematic discrepancy between the position of the nominal subject and the pronominal subject with respect to the position of the finite verb in root clauses. Some examples are given in (40): (40) a. For5on we sceolan mid ealle mod and masgene to Gode therefore we shall with all mind and power to God gecyrran turn 'therefore we must turn to God with all our mind and power' (Blickling 97) b.

Be oaem we magon suioe swutule oncnawan oaet by that we may very clearly perceive that... 'by that we can perceive very clearly that...' (CP, 181,16)

This discrepancy gradually vanishes in the course of the second half of the 14th century and first quarter of the 15th century. This, we propose, is the main trigger for the loss of verb second. Pronominal subjects regularly appeared on the left of verb second in root clauses, while nominal subjects regularly "inverted" (cf. the examples (38) with (40)). As this discrepancy between nominal and pronominal subjects got lost, nominal subjects adopted the pro-

248

VERB SECOND AND THE NULL-SUBJECT PARAMETER

nominal pattern.25 We can show this with some examples from the Wycliffite writings: (41) a. wip newe wenchis is Crist now weddid and with new wenches is Christ now wedded and on newe maner he kepte his furste matrimonye in new manner he kept his first matrimony (The Wycliffite sermons, ed. Hudson, p. 361, 1.42) b. And by pis same skyle hope and sorwe schulle iugen us and by this same skill hope and sorrow shall judge us (The Wycliffite sermons, ed. Hudson, p. 372, 1.97) c. for more ioyse pei myhte not haue for more joy they could not have (The Wycliffite sermons, ed. Hudson, p. 382, 1.106) We regularly find these three patterns: inversion of a subject noun (41a); noninversion of a subject noun (41b); and non-inversion of a subject pronoun (41c). There are very few examples of inverted subject pronouns. Inversion of a subject pronoun is very infrequent, except with an operator type of first constituent like a w/z-element. What these examples indicate is that in this dialect, which there is some reason to assume played a role in the rise of the standard language, the nominal subjects tended to adopt the pattern of pronominal subjects. The result of this was that fronting of XP to Spec.CP no longer went hand-in-hand with V/I to C. Hence verb second declined and English changed from a C-oriented language to an I-oriented language. Let us consider the relationship between the loss of verb second and the loss of expletive pro-drop more closely. At the stage when English was still verb-second with expletive pro-drop, say c. 1250, the basic sentence structure is as in (42):

VERB SECOND, PRO-DROP, FUNCTIONAL PROJECTIONS

249

C is the dominant head, hence it must be lexicalized, either by a complementizer or by a finite V. According to the definition of Nominative Case-marking (6), Nominative Case is assigned by C under government. C is also the designated Case-marking head that acts as a formal licenser for pro. I does not have inherent p-features, hence referential pro-drop is excluded; only expletive pro-drop is possible. The p-features of I must therefore be licensed by movement of the subject to Spec.IP to ensure licensing of I under Spec-h d agreement in IP. In the course of the latter half of the 14th century, s jects come to appear more and more frequently to the left of the finite verb, verb second declines. It is at this time that English changes from a C-oriented language to an I-oriented language. I is then the dominant head and must be lexicalized. Vf moves to I; V/I to C is lost. In this process, the dominant head loses its characteristic of being able to formally license pro, hence expletive pro-drop is lost. Observe that the loss of verb second and the loss of expletive pro-drop are, though historically related, not directly causally related. That is, our theory does not force the loss of expletive pro-drop when verb second is lost. However, there is presumably some indirect causation involved here. At the stage in the history of English where these developments occur, the subject is always in Spec,IP (to license p-features), and even expletive pro-drop, though possible, is not really frequent; insertion of it or there in the position of non-thematic subjects occurs on a large scale. The evidence for expletive pro-drop, hence for C (or I) as a formal licenser of pro, was not particularly robust then, and is likely to be reinterpreted when a major shift from C-orientation to I-orientation takes place. As the subject always appears in Spec.IP positio (to license I), Nominative Case comes to be assigned under Spec-head agr ment in IP.26 The loss of verb second and the loss of expletive pro-drop coincide historically.

4. Conclusions Let us conclude by comparing the scenario for the history of English with that for the history of French. OF and OE are both CV2 languages. The difference between OF and OE is that in OF I has inherent p-features, whereas I in OE does not. As a result OF has referential pro-drop in root contexts and expletive pro-drop in both root and non-root clauses. OE on the other hand allows expletive pro-drop everywhere; since I has no p-features, a referential subject is obligatorily in Spec-head agreement relation in IP. Since a referential subject could never be dropped, English, when it lost its CV2 characteristic, could not become an IV2 language, since Spec.IP was a subject position, not an XP position. In contrast, when French lost its CV2 characteristic, I still had inherent p-features and a subject, whether lexical or null, could be licensed in its base-generated position in Spec.VP in the structure (1). As a result Spec.IP can be an XP position and the IV2 characteristic obtains. The final step in the development of French was that inherent p-features were lost. As a result, I came to be licensed by Spec-head agreement with NP in Spec.IP.

250

VERB SECOND AND THE NULL-SUBJECT PARAMETER

The CV2 stage that we argu d for in Old French, and the IV2 stage that we identified in Middle French argue for at least these two functional projections.

Notes 1. This paper was presented at the First Diachronic Generative Syntax Conference in York, April 1990. We thank the audience, the editors of this volume and an anonymous reviewer for comments. We would like to thank in particular Barbara Vance for her constructive criticisms of various versions of the material on French. The first draft of this paper was circulated in 1990. Since then, the theoretical work on which this paper is partly based, and the historical work reported here, have led partially separate lives. Although the theoretical preliminaries of this paper are closely related to Hulk and van Kemenade (1993), the theory in the latter has evolved much since this work was written up. 2. Sometimes root clauses have base-generated complementizers, hence no verb fronting, as in, for example, the Modern Dutch exclamation: (i)

dat -ie dat zo gedaan heeft! that he that so done has! 'I am surprised he has done it is way'.

3. For instance, Evers (1982) proposes at tense is an operator in Comp and must be lexical; Koopman (1984) proposes that lexicalization is triggered for Nominative Case assignment; de Haan and Weerman (1986) suggest that Comp (containing Agr) has categorical features and must therefore be lexical; Weerman (1989) assumes that C assigns a modal role to the proposition, and that C needs to be lexical to be visible as an assigner of modal role. 4. For the moment we do not wish to go into the details of the relation between V and I in these languages. At least for Modern English there is evidence, for example, from Neg placement facts (see Pollock (1989); Roberts (1985)) that V does not move to I. These facts also underlie the pie-Barriers idea that in Modern English I is lowered to V, see Lasnik (1981). We will come back to this below. 5. We will come back below to the theoretical status of this assumption. 6. Consider for instance the following Southern Dutch sentences (from den Besten (1979)). For related facts in German, see Tomaselli (1990). (i) a. date Jan en Rees morgen zullen komen that+pl John and Bill tomorrow shall +pl come 'that John and Bill will come tomorrow' b. dat(*e) Jan morgen zal komen that(*+pl) John tomorrow shall -pl come 'that John will come tomorrow'. These examples show that C has number agreement with Vf.

VERB SECOND, PRO-DROP, FUNCTIONAL PROJECTIONS

251

7. Consider the examples (i): (i)

a. Ik peinzen danze-t-ze zunder gezeid heen I think that-it-her they said have "I think that they said it to her' b. Gisteren heet-t-ze Jan gegeven Yesterday has-it-her John given 'Yesterday John gave it to her'.

Syntactic cliticization is generally accepted to be cliticization of a dependent element on a syntactic head, in the case of personal pronouns usually a Casemarking head. This would seem to indicate that C in WF has Case-marking properties. The choice for examples of object cliticization on C is deliberate, since the fact that the clitics are removed from their base-generated position in VP shows that they are syntactic clitics, and can therefore be reasonably supposed to have cliticized onto a syntactic head. This is not clear in cases of cliticization of a subject pronoun on C, since there the cliticization may be just phonological. 8. One might expect a similar situation for Modern English and for Modern French. However, it seems that Modern English has no V to I for lexical verbs (though for have and be it does). We follow standard treatment in assuming a rule of I-lowering to V for Modern English. 9. We leave tense out of consideration here. While tense too, is a functional feature, it does not bear in any obvious way on the relationship between heads and dependent NP's, like agreement and Case do. 10. The way we have formulated Nominative Case-marking and subject-verb agreement here, in conjunction with our licensing conditions for pro below ((12)), suggest that Nominative Case and agreement are different notions. We argue in Hulk and van Kemenade (1993) that this is in fact correct. However, this does not bear directly on the issues considered in this article. 11. (15) suggests that Icelandic is a IV2 language. In Hulk and van Kemenade (1993), we elaborate on this. 12. This analysis presupposes that this C also has the relevant Nominative Case-assigning properties, if lexicalized by movement of V+I. 13. As noticed by den Besten (1989, ch. 3), among others, embedded V-to-C creates an island for extraction, whereas embedded V-to-I does not. This makes a clear prediction for OF, which, if evidence of such a detailed nature can be found, needs to be checked in future research. 14. Interestingly, a similar situation is found in the interlanguage of Dutch native speakers who learn (Modern) French as a second language. These learners also show a transitional stage in which they seem to have verb second both the Dutch system, verb second with fronting to Spec.CP, and the French one, with V-to-I and fronting as IP-adjunction, see Hulk (1990): (i)

hier mangeait Jean les fraises yesterday ate Jean the strawberries

(ii) hier Jean mangeait les fraises yesterday Jean ate the strawberries

252

VERB SECOND AND THE NULL-SUBJECT PARAMETER

15. Interestingly, Adams relates the cliticization of French subject pronouns to the change in prosodic patterns. She argues that originally OF had a Germanic stress pattern, that gradually changed to the Modern French stress pattern, which, among other things, lacks the initial stress that is characteristic of Germanic. 16. The licensing of null subjects does not necessarily coincide with overt rich morphology. It seems that overt rich morphology is a necessary, but not a sufficient condition for morphological I-licensing. Some further abstract property is needed. Roberts (1992) adopts a different metaphor (+/- m) for the same abstract property. 17. This predicts that a preverbal subject, whether null or lexical, is always chainrelated to the base-generated position in Spec.VP. This is unproblematic for Nominative Case and formal licensing; these two properties can be transmitted by chain. We assume that p-identification cannot be transmitted by chain. However, asp-identification can also take place under Spec-head agreement, there is no problem in this respect. This is discussed further in Hulk and van Kemenade (1993). 18. Vance (1988) observes that sentences like (i)

lors voit le roi then saw the king 'then he saw the king' are ambiguous with respect to the position of pro: the position can be preverbal or postverbal. 19. In Hulk and van Kemenade (1993) we give a more elaborate analysis that provides an explanation for the choice between content-licensing under spechead agreement in IP and content-licensing under government by I. 20. Obviously a lot more needs to be said about this transition. For instance, it seems probable that I first loses its content-identifying properties and afterwards its formal licensing properties. The precise relationship with Nominative Case marking to postverbal subjects (which persists well into the 17th century if Clifford (1973) is correct) deserves some closer consideration, as well as the relationship with Modern French stylistic inversion. These are very detailed considerations that we leave for future research. 21. Van Kemenade (1987) analyzes Old English as a verb-second language in the sense understood in this article. Recently, Allen (1990) states that this analysis is based on unfounded factual claims, as V-fronting in root clauses is not completely obligatory, and S-Vf-inversion following a PP topic is far from obligatory (Allen (1990:150): "Sentences of the type PPSV are very common indeed."). It is true that van Kemenade does not take full account of these facts. But Allen's criticisms are not really to the point. It is true that V-fronting is not completely obligatory in OE, but a proviso of this sort is easily incorporated in the analysis, and given this, the patterns that Allen presents as "counterevidence" are not appropriate counterevidence, as V-fronting is only properly exemplified by sentences with two verbs (as it can be shown conclusively only in these cases that the position of Vf is different from that of the

VERB SECOND, PRO-DROP, FUNCTIONAL PROJECTIONS

253

infinitival verb, given further movement rules operative in OE; in fact all van Kemenade's examples are sentences with two verbs). A thorough look at such constructions bears out van Kemenade's system, with two provisos. First, Vfronting is not completely obligatory; and second, the position preceding the finite V in root clauses is sometimes occupied by two constituents. This, however, is not counterevidence to V-fronting, as it can still be shown that there is fronting. Allen's criticisms seem to be intended as refutation. As such, they mislead the reader about the status of the counterevidence, and appear to stem from a basic misappreciation of the grammatical status of the phenomenon. 22. These examples are suggestive of a fairly homogeneous word order in OE, but in fact this is far from being the case. The exceptions are explained by van Kemenade (1987) as (heavy) NP shift to postverbal position; a relatively free process of verb-raising as attested in various West Germanic languages, and a clitic analysis of personal pronouns. Pintzuk (1991) analyzes OE as variable between I-medial and I-final word order. This does not bear on the issue whether OE verb-second is movement to I or to C. 23. Only a pattern with a fronted XP is valid as evidence here, as examples with Subject-Vf-(XP)-V-(XP) occur quite frequently in non-root clauses. The latter may be analyzed either as an instance of a variant pattern with I-medial order in OE, or as an instance of verb-projection raising. 24. Apart from these two data studies, Allen (1986) adduces interesting evidence that in constructions with like, Nominative subjects are favoured as of the 15th century. 25. Actually this is subject to some dialectal variation (cf. van Kemenade (1987)). It seems that in the prose of Chaucer, pronouns adopted the nominal subject pattern. For instance, in one of Chaucer's prose works, the Treatise on the Astrolabe, verb-second inversion is completely regular. This is in marked contrast to the dialect of Wycliffe, who is a near contemporary of Chaucer. Interestingly, there is good reason to assume that the Wycliffite sermons represent an important literary standard, see Warner (1982). Allen (1990) criticizes van Kemenade's (1987) observations on the dialects of Wycliffe's and Chaucer's prose as being highly inaccurate. It is unclear how she arrived at this remark. Detailed further investigations have shown that these observations are pretty much correct (see van Kemenade (1990); Kroch (1989)). 26. At a later stage, even V to I is lost. This change does not concern us at present, but should be dated as later than the general loss of verb second. Roberts (1985) dates the loss of V to I in the 16th century.

References Abraham, W. ed. (1983) On the Formal Syntax of the Westgermania. Amsterdam: John Benjamins. Adams. M. (1987) "From Old French to the Theory of Prodrop." Natural Lanugage and Linguistic Theory 5:1-32.

254

VERB SECOND AND THE NULL-SUBJECT PARAMETER

Adams, M. (1988a) "Embedded Pro." In J. Blevins and J. Carter, eds. NELS 18, 1-21. Amherst, Mass.: GLSA. Adams, M. (1988b) "Les effets V2 in ancien et en moyen francais." Revue quebecoise de linguistique theorique et appliquee 7:13-39. Allen, C.L. (1986) "Reconsidering the History of like." Journal of Linguistics 22:375-409. Allen, C.L. (1990) Review of van Kemenade (1987). Language 66:146-152. Bennis, H. and L. Haegeman (1983) "On the Status of Agreement and Relative Clauses in West Flemish." In W. de Geest and Y. Putseys, eds. Sentential Complemetation, 33-55. Dordrecht: Foris. Besten, H. den (1983) "On the Interaction of Root Transformations and Lexical Deletive Rules." In W. Abraham, ed., 47-131. Besten, H. den (1989) Studies in West Germanic Syntax. Amsterdam: Rodopi. Canale, W. (1978) Word Order Change in Old English. PhD Dissertation, McGill University. Chomsky, N. (1981) Lectures on Government and Binding. Dordrecht: Foris. Chomsky, N. (1982) Some Concepts and Consequences of the Theory of Government and Binding. Cambridge, Mass.: MIT Press. Chomsky, N. (1986) Barriers. Cambridge, Mass.: MIT Press. Clifford, P. (1973) Inversion of the Subject in French Narative Prose from 1500 to the Present. Oxford: Blackwell. Coopmans, P. and A. Hulk, eds. (1988) Linguistics in the Netherlands 1988. Dordrecht: Foris. Diesing, M. (1990) "Verb Movement and the Subject Position in Yiddish." Natural Language and Linguistic Theory 8:41-79. Dupuis, F. (1988) "Pro-drop dans les subordonnees en ancien francais." Revue quebecoise de linguistique theorique et appliquee 7:47-62. Elmer, W. (1981) Diachronic Grammar: the History of Old and Middle English Subjectless Constructions. Tubingen: Niemeyer. Evers, A. (1982) "Twee fonctionele principes voor de regel 'Verschuif het verb'." GLOT5:ll-30. Gaaf, W. van der (1904) The Transition from the Impersonal to the Personal Construction in Middle English. Amsterdam: Swets and Zeitlinger. Haan, G. de and F. Weeman (1986) "Finiteness and Verb Fronting in Frisian." In H. Haider and M. Prinzorn, eds. Verb Second Phenomena in Germanic Languages, 77-110. Dordrecht: Foris. Hirschbuhler, P. and J. Junker (1988) "Remarques sur les sujects nuls en subordonnees en ancien et en moyen francais." Revue quebecoise de linguistique theorique et appliquee 7:63-84. Hornstein, N. and D. Lightfoot, eds. (1981) Explanation in Linguistics. London: Longman. Hulk, A. (1989) "La construction impersonelle et la structure de la phrase." Recherches Linguistiques de Vincennes, 18. Hulk, A. (1990) "Parameter Setting and the Acquistion of Word Order in L2 French." Ms. Free University of Amsterdam.

VERB SECOND, PRO-DROP, FUNCTIONAL PROJECTIONS

255

Hulk, A. and A. van Kemenade (1988) "Nominative Identification in Germanic and Romance Languages." In P. Coopmans and A. Hulk, eds. Linguistics in the Netherlands, 69-78. Dordrecht: Foris. Hulk, A. and A. van Kemenade (1993) "Subjects, Nominative Case, Agreement and Functional Heads." Lingua 89:181-215. Kemenade, A. van (1987) Syntactic Case and Morphological Case in the History of English. Dordrecht: Foris. Kemenade, A. van (1990) "The Loss of Subject-Verb Inversion in Late Middle English." Paper given at Manchester University, April 1990. Koopmann, H. (1984) The Syntax of Verbs. Dordrecht: Foris. Kroch, A. (1989) "The Loss of the Verb Second Constraint in Middle English and Middle French." Talk given at the University of Amsterdam. Lasnik, H. (1981) "Restricting the Theory of Transformations: A Case Study." In Hornstein and Lightfoot, eds. Lemieux, M. and F. Dupuis (this volume) "The Locus of Verb Movement in Non-Asymmetric Verb-Second Languages: The Case of Middle French" Lenerz, J. (1989) "Pro-drop in Old High German." Unpublished paper given at Utrecht. Lightfoot, D. (1979) Principles of Diachronic Syntax. Cambridge: Cambridge Unversity Press. Mascaro, J. and M. Nespor, eds. (1990) Grammar in Progress: GLOW Essays for Henk van Riemsdijk. Dordrecht: Foris. Pintack, S. (1991) Phrase Structures in Competition. PhD Dissertation, Universtiy of Pennsylvania. Platzack, C. (1983) "Germanic Word Order and the COMP/INFL Parameter." Working Papers in Scandinavian Syntax, 2. University of Trondheim. Platzack, C. (1987) "The Scandinavian Languages and the Null Subject Parameter." Natural Language and Linguistic Theory 5:377-401. Pollock, J.-Y. (1986) "Sur la syntaxe de en et le parametre du sujet mil." In M. Ronat and D. Conquaux, eds. La Grammaire Modulaire. Paris: Editions du Minuit. Pollock, J.-Y. (1989) "Verb Movement, UG, and the Structure of IP." Linguistic Inquiry 20:365-424. Rizzi, L. (1982) Issues in Italian Syntax. Dordrecht: Foris. Rizzi, L. (1986) "Null Objects in Italian and the Theory of Pro." Linguistic Inquiry 17:501-557. Roberts, I. (1985) "Agreement Parameters and the Development of English Modal Auxiliaries." Natural Language and Linguistic Theory 3:21-58. Roberts, I. (1992) Verbs and Diachronic Syntax. Dordrecht: Kluwer. Safir, K. (1986) "Subject Clitics and the Nominative Drop Parameter." In H. Borer, ed. The Syntax of Pronominal Clitics, 333-365. San Diego: Academic Press. Santorini, B. (1989) The Generalization of the Verb-Second Constraint in the History of Yiddish. PhD Dissertation, University of Pennsylvania. Sigur5sson, H.A. (1989) Verbal Syntax and Case in Icelandic in a Compa tive GB Approach. PhD Dissertation, University of Lund.

256

VERB SECOND AND THE NULL-SUBJECT PARAMETER

Sportiche, D. (1988) "A Theory of Floating Qualifiers and its Corollaries for Constituent Structure." Linguistic Inquiry 19:425-449. Taraldsen, K.T. (1978) "On the NIC, Vacuous Application and the That-trace Filter." Unpublished Ms, MIT. Thrainsson, H. (1986) "VI, V2, V3 in Icelandic." In Haider and Prinzhorn, eds., 169-194. Tomaselli, A. (1990) "COMP as a Licensing Head: An Argument Based on Cliticization." In Mascaro and Nespor, eds., 433—445. Vance, B. (1988) "Uevolution de pro-drop en frangais me'di6val." Revue quebecoise de linguistique thlorique et appliquee 7:85-109. Vance, B. (1989) Null Subjects and Syntactic Change in Medieval French. PhD Dissertation, Cornell University. Warner, A. (1982) Complementation in Middle English and the Methodology of Historical Syntax. Pennsylvania State University Press. Weerraan, F. (1989) The V2 Conspiracy: A Synchronic and a Diachronic Analysis of Verbal Positions in Germanic Languages. Dordrecht: Foris.

10 Null Subjects in Verb-First Embedded Clauses in Philippe de Vigneu es' Cent Nouvelles Nouvelles Paul Hirschbiihler Universitt d'Ottawa

"C'est un syst&me bien dangereux, monsieur Fred, bien dangereux, que celui qui consiste a partir de Fid6e qu'on se fait de 1'assassin pour arriver aux preuves dont on a besoin!" (Rouletabille)

The focus of the present study concerns the licensing of null subjects in embedded verb-first constructions in a late Middle French text, the Cent Nouvelles Nouvelles (from now on, the CNNV) by Philippes de Vigneulles.1 More precisely, our goal is to characterize as precisely as possible the conditions under which these null subjects are found, hoping that some of the ideas discussed will find their way in a more principled account than the one that is offered here. One point that should be clear from the start is that I do not claim that all the variables that appear to be required in characterizing the data examined here are necessarily relevant for other Middle French texts. Some of the facts are unexpected given previous literature. It may therefore be tempting to consider them as accidental patterns rather than considering them as significant. The second position will be adopted here, in the hope that close study of other texts will reveal similar patterns.2 This paper is organized as follows: I first very briefly review the main points of the grammar of Middle French with respect to main clauses that are relevant for the discussion of embedded clauses; then I present the main results of my analysis of null subjects in verb-first embedded clauses; finally I turn to the data and discuss it in detail.

257

258

VERB SECOND AND THE NULL-SUBJECT PARAMETER

1. Overview of the Analysis 1.1 (S)VO in Main Clauses In Middle French, assertive root clauses are of two main types. Verb-second sentences, which are analyzed as CPs, and SVO clauses which, following Vance (1988), we take to be IPs or, in a more ramified view of clausal structure, AgrPs.3 Among these SVO root clauses, there is a large enough number of examples where the subject is null to support the idea that for all persons, Agr, perhaps in combination with Tense, can license both the position and the content of pro subject to its left, i.e. under Spec-Head agreement.4 Examples are given in (1) below. Generally, these sentences do not strictly start with the verb (except for advint que '[it] happened that'), but the initial element does not normally induce verb-second effects in Vigneulles; this initial element may be a coordinating conjunction, the negative element ne or phrases that are analyzed as adjoined to IP (see Vance (1988, sections 4.1.3, 4.2.1) and Roberts (1992, section 2.3.1) for detailed arguments in favor of adjunction to IP in cases of that sort). (1)

a. Tu jeuneras demain et apres et aprez. Et 01 te ordonne que ainsi sefasse... 002042 You will fast tomorrow and after and after. And [I] order you that [it] so be done...5 b. Et aultre responce n'en peust avoir ladicte femme. 03] Ne demeura guiere apre~s qu'elle fust morte. 0t Ne scay s'elle en alia... 062039 And another answer could not receive the mentioned woman. [It] didn't last long before she was dead. [I] don't know if she went... c. ...pour tant que 1'une des fois tu met la cornette d'ung costez, ...aucunes fois tu la fais aller devant, aucunes fois 02 la fais aller derrier, et 03 va et vient de tous costez... 044043 ...as sometimes you put [your hat's] horn on one side, ...some times you put it in front, sometimes [you] make it go behind, and [it] goes in all directions. d. Puis, cela faict, 03 print de l'ancre...Cela faict, 03 s'en retourna avec les autres, et quant temps fut de partir, 06 prindrent congie et0 6s'enallerent... 001019 Then, this done, [he] took some ink.... This done, [he] went back with the others, and when time was to leave, [they] took their leave and left... e. 031 Advint un temps que Ton se doubtoit a Mets pour aucuns gens d'armes... 061018 [It] came a time when people were afraid in Mets because of soldiers... f. Je la te venderes une somme qui sera dicte, et 04 metterons les vins a bon pris et ainsi 0 4 buverons de la marchandise. 091048

NULL SUBJECTS IN CENT NOUVELLES NOUVELLES

259

I will se it to you for an amount that will be said, and [we] will put the ines at a good price and so [we] will drink from the trade. g. .... j'en prendra la chose sur moy, et 05 direz que je le vous ay fait faire. 015089 I will take it upon myself, and [you] will say that I had you doit. We now turn to our gene l claims concerning null subjects in verb-first embedded clauses in the CNNV.

1.2 Null Subjects in Verb-First Embedded Clauses In this section, we summarize the results of our examination of null subjects in verb-first embedded clauses in the CNNV and sketch an analysis of the facts. In the licensing of null subjects, two factors must be taken into consideration: formal licensing and identification (Rizzi (1986); Adams (1988); Vance (1988); Roberts (1992)). Formal licensing of SpecIP by Infl or an element in Comp and identification of the content of the null subject by Infl/Agr are discussed in the next four subsections. 1.2.1 Subordinate Clause Introducers In the CNNV, it appears that for the purpose of accounting for verb-first embedded clauses, either a three way distinction is required among the embedded clause introducers, or, if a two way distinction is maintained, as in Hirschbuhler and Junker (1988), Adams (1988) and Roberts (1992), it differs to varying degrees from the division advocated in these works. As we will see in SECTION 3, it is necessary to make the following descriptive distinctions: 1.

2.

3.

Non-licensing subordinators: the conjunction que and perhaps all of the (simple and complex) conjunctions: these never allow verb first, whether with null subjects (except when the verb is in the 2 p.pl.) or with postverbal nominal subjects. Partially licensing subordinators: the que that introduces relative and comparative clauses as well as indirect questions (wh-que). Que here does not allow verb first with null subjects, but allows verb first with postverbal nominal subjects. Licensing subordinators: lexical wA-phrases (excluding those instances where wh-que is a real wfc-word), that allow verb first with null and postverbal nominal subjects.

In order to account for some aspects of this situation, recent suggestions by Roberts (1992, section 2.3.5) will be adopted. Although in Vigneulles evidence from main clauses shows that Infl (or Agr) has the capacity to formally license a null subject by Spec-Head agreement, we consider that in

260

VERB SECOND AND THE NULL-SUBJECT PARAMETER

embedded clauses, formal licensing (or not) of an empty position adjacent to an element dominated by C or a projection of C depends on that element and not on Infl. This might be looked at as the resolution of a conflict between potential licensers: if among two competing potential licensers (e.g. one in Infl and one in C or SpecCP), one is in a position to license formally under government while the other is in a position enabling it to license formally under Spec-Head agreement, the government variant mode of formal licensing prevails.6 Considering then the three classes of subordinators and the case of null subjects, we would say that the first two types of subordinators mentioned are not formal licensers (do not participate in formal licensing) while those in the last group are. The grouping of partially licensing subordinators (on which we come back in SECTION 1.2.4) with non-licensing subordinators rather than with licensing subordinators is important because in previous studies that I am aware of, wh-que was grouped with the latter type, either because the facts required this or because they were not looked at in enough detail (as, e.g., in Hirschbuhler and Junker (1988)). 7.2.2 Agr Features As for the identification of the content of the empty category by features of Agr, it appears that, here again, a three way distinction is necessary. Following Roberts (1992, sections 2.2.3-4, 2.3.5), we consider that while Middle French has a formally rich inflectional paradigm (i.e. the paradigm is morphologically uniform, with a slot for agreement affixes available for each person), this paradigm is functionally weak (in that there is more than one syncretism). In principle, either type of richness is enough to identify the content of pro subject in the appropriate structural configuration. We will see that in addition to this, in the CNNV, strength of individual person affixes needs to be added to the variables playing a role in the distribution of null subjects when the content of pro is identified in the configuration of Spec-Head agreement with Infl. Thus we distinguish between: 1.

2.

Strong Agr. 2nd. p.pl. (i.e. person 5) allows verb first with null subjects in embedded clauses of all types. That is, the 2nd p.pl. is able to formally license SpecIP (as well as identify its content) under SpecHead configuration in embedded clauses despite a filled C° or SpecCP. One would expect 1st. p.pl. (i.e. person 4) to belong to this class, if "rich" is directly related to transparency of morphological indication of person and number. As the data will suggest, it is not clear that the relation is so direct. Healthy Agr. 3rd singular (expletive, i.e. person 31, and non-expletive, i.e. person 3) and 3rd p. pi. (person 6). These are able to identify the content of pro in a configuration of Spec-Head agreement in verb-first embedded clauses provided that the position is formally licensed by a licensing subordinator.

NULL SUBJECTS IN CENT NOUVELLES NOUVELLES

3.

261

Weak Agr. 1st and 2nd singular. With some exceptions that will be discussed, they are not found in a verb-first construction in any of the embedded clause types. One way to describe their behaviour would be to say that their ability to identify the content of pro in a configuration of Spec-Head agreement is parasitic on whether or not they also actually formally license the subject pro position. Given that in embedded clauses any element in C or SpecCP, when adjacent to SpecIP, is a potential formal licenser, weak Agr is never in a position to license pro in that position.

1.2.3 Non-Licensing Subordinators and Licensed Null Subjects Support for the analysis proposed thus far is provided by XP (S)VO subordinate clauses, which are discussed in SECTION 4. There are "XP (S)VO" subordinate clauses just as there are main clauses of that type. When the subject is null, they resemble verb-second constructions, but there are clear indications that these are constructions with XP adjoined to IP. In this situation the subject may be null under Spec-Head agreement, not only in main clauses, but in embedded clauses too, even when the subordinator is of the non-licensing type. The intervening XP prevents C from governing SpecIP, i.e. the complementizer system does not provide a potential licenser. As a result, Infl (or Agr) is the only potential formal licenser for SpecIP, and formally licenses it. And, as might be expected, given that Infl formally licenses SpecIP, it (or Agr) is again able to identify its content, and null subjects are found for persons 1 and 2. 1.2.4 Stylistic Inversion vs Null Subjects Recently, it has been suggested that the so-called Stylistic Inversion construction should be analyzed, whether for contemporary French (e.g. Deprez' (1988) interesting analysis of Triggered Inversion) or previous periods of the language (Vance (1988); Roberts (1992)), by appealing to the presence of an expletive pro in the highest SpecX" node subjacent to C. In Vigneulles, there are 85 cases of verb first with a postverbal nominal subject in embedded clauses. All of the examples are w/z-constructions (or operator constructions) of one type or another. What is of interest to us is that the subordinate clause introducer may be que, as shown in (2), contrary to what we found was the case for null subjects, even for impersonal (uses of) verbs. The examples with que are divided as follows: ten comparatives, five indirect questions, eighteen relatives, and one topicalization. (2)

a. car il estoit mieulx l'homme pour le despendre que n'estoit ledit cure, ... 011051 because he was more the man to spend it than was the priest (comparative) b. Ores maintenant je vous dires que fist ung aultre fol... 099135 Now I will tell you what did another fool... (indirect question)

262

VERB SECOND AND THE NULL-SUBJECT PARAMETER

c. Et encor n'est rien au regard des peines et tourmens qu'a endurez nostre doulx Sauveur Jhesu Crist pour nous. (relative) 059028 And [it] still is nothing compared to the pains and torture that suffered our kind Savior Jesus Christ for us. d. ...je ne m'y suis pas fain; aussi croy je que n'a pas fait le Ribauldez, mon compaignon. 058090 ...I did not deprive myself of it; neither I do believe that did the bad man, my friend. (topicalization out of complement clause). ['aussi' is complement of 'faire'; it has the meaning of<<non plus>> (cf. Martin et Wilmet, 1980, §12)] Given that Stylistic Inversion is found in verb-first que clauses with a whgap, while impersonal null subjects are not, it would seem that the expletive of Stylistic Inversion constructions, if there is one, should be treated differently from the null expletive of impersonal constructions. Impersonal null subjects have a content that needs to be identified by Agr, the content being the features of person and number. In the CNNV, it is apparent that wh-que does not formally license the position of an impersonal null subject. This indicates that wh-que is not strong enough to license the SpecIP position when Infl (or Agr) is grammatically active with regard to SpecIP (which is the case here, in the sense that identification of the content of SpecIP can only come from Agr). This is so even when Agr is itself rendered inoperative as a formal licenser for SpecIP by the presence in C or SpecCP of a potential licenser governing it. So, the presence of que in C or SpecCP prevents Agr from acting as an formal licenser, and the relation of Agr to SpecIP, required for content identification, contributes to wh-que not being an actual formal licenser for that same position. The situation is quite different in the case of Stylistic Inversion. Here, SpecIP can be thought of as radically empty (i.e. as not containing an impersonal pro) or as associated via some type of coindexation to the postverbal subject. In either case Agr is not active with regard to SpecIP; it is active with regard to the postverbal subject, assigning it Nominative Case. Not being active with regard to SpecIP, Agr is not a potential formal licenser for that position, resulting in wh-que fulfilling that role. As a more detailed examination of Stylistic Inversion is not required for the goals of this paper, it will not be discussed in more detail below. These are in a nutshell the main claims of this paper. Keeping them in mind will allow us to have an easier time looking at the data in the CNNV in more detail, to which we now turn. In SECTION 2, we give a general overview of the data. In SECTION 3, we turn to a detailed discussion of null subjects in embedded verb-first constructions. In SECTION 4, we examine the cases where a constituent intervenes between the embedded clause introducer and the verb.

NULL SUBJECTS IN CENT NOUVELLES NOUVELLES

263

2. Data Following a standard view, the null subject (Spo) is analyzed as pro. In what follows, its distribution is compared to that of the lexical pronominal subject (Spr). Both types of pronominal subjects together are designated by Sp. Spr and Spo have been divided in several groups to determine the possible role of various factors in the licensing of Spo. The following potential factors have been examined: the type of subordinate clause as well as the specific subordinator used the verb position: does it immediately follow the subordinator or not? the person and number of the verb the particular verb used The distribution of Spo in subordinate clauses of the type "Sub (XP) V..." is summarized in Table 10.2 and Table 10.15 below. Table 10.3 and Table 10.14 give us the same information for subordinate clauses with a preverbal lexical pronominal subject Spr. The tables are divided into three parts, the two main categories corresponding in part to the divisions discussed in previous studies: Hirschbiihler and Junker (1988), Vance (1988, 1989), Adams (1987a,b, 1988) and Dupuis (1988, 1989). In part A, we have grouped together the conjunctive clauses introduced by que or by a conjunctive phrase ending with que, as well as wh-clauses introduced by que: cleft sentences, restrictive relative, comparative and superlative clauses, and indirect questions. Part B contains temporal and conditional clauses, as there were reasons to expect that they might pattern with category C as well as A, given some of the existing literature (Hirschbiihler and Junker (1988); Adams (1988)). Finally, part C of the tables contains whclauses (indirect questions, relatives, comparatives), except for those introduced by que. Before turning to these tables, we shall give a brief overview of the distribution of null and pronominal subjects in the CNNV. The total number of null subjects in the text is 2403. This number excludes (1) direct questions and (2) the second (third, etc.) of a series of coordinate clauses whose null subject is interpreted as identical to the subject of the preceding coordinated clause. The null subjects can be further subdivided as in Table 10.1. As can be seen, the 513 examples of null subjects in subordinate clauses represent 21.35% of the total number of null subjects in the text. Table 10.1 Distribution of Null Subjects (Spo): Person Main Clause Subord. Clause Total

2

3

31 7 ce

4

127

12

927

415 28

28

7

6

124

207

134

18

1051

12

4

622 32

5

6 Total

53 300

7 127

31

35 180 331

%

1890

78.61

514

21.38

2404 100.00

264

VERB SECOND AND THE NULL-SUBJECT PARAMETER

The total number of embedded lexical pronominal subjects (Spr) is 3981. This excludes on 'one', for which there is no null variant. It includes: 3884 72 25

preverbal Spr immediately the subordinate clause introducer (cf. Table 10.3), preverbal Spr separated from the subordinate clause introducer by some phrase (cf. Table 10.14), (to which nine examples of on might be added), and postverbal Spr.

The 514 embedded clauses with Spo represent 11.43% of embedded Sp, i.e. of the total of Spr + Spo (4495). As mentioned above, this excludes coordinations that could be interpreted as coordinations of VPs, as in (3a) below, but it does not exclude conjoined clauses whose implicit subject is distinct from that of the preceding clause, as in (3b): (3) a. ..., advint qu'il print ung bachet et 0 le rua aupres d'un mullon defoin... 055058 ..., [it] happened that he took a pike and [that he] threw it on a stack of hay b. ...1'enfant luy confessa toute la maniere et comment son pere avoit tuez sa vaiche et 1'avoit sallez et 0 en avoient plussieurs fois mangez. 008066 ...the child confessed to him how (= that) his father had killed his cow and had pickled it and [how they] had several times eaten of it. Contrary to null subjects, lexical pronominal subjects appearing after a coordinating conjunction have been included in the counts. Thus, the proportion of Spo vs Spr is bent towards Spr.

3. Sub 0V We first examine the cases where no constituent intervenes between the subordinator and SpecIP, leaving to SECTION 4 the study of examples with an intervening XP, Table 10.2 summarizes the facts for null subjects in embedded verb-first constructions. There are 179 such Spo (second column). The column "Total 0" gives the sum of the embedded Spo in verb-first (i.e. Sub 0 VX, Table 10.2) with those in verb-second (i.e. Sub XP VX, Table 10.15) constructions. The 179 Spo examples in Table 10.2 are to be compared to the 3884 Spr examples of Table 10.3; they represent 4.40% of the total Sp in verb-first constructions (4063). Null subjects are excluded in category A (except for the Strong Agr fifth person and, to some extent the fourth). If only categories B and C are considered, Spo represents 7.41% of the total of Spo + Spr (6.51% for category B and 8.06% for category C). The embedded verb-first construction with Spo is therefore not that frequent compared to the embedded verb

NULL SUBJECTS IN CENT NOUVELLES NOUVELLES 265

second with preverbal Spr, but it cannot be ignored, since its distribution appears to be subject to restrictions related to the type of subordinate clause, the person of the verb, and the personal vs impersonal distinction, as is discussed below. Table 10.2 Subordinator 0 V X (179/514 = 34.82% of Embedded Spo) A. Que

Total 0 Sub

0V

purpose cause subject/object misc. concessive result optative comp. que R. Rel.: que I.Q.: wh-que

7 16 114 13 7 47 2 6 28 0

5 6 22 5 3 2 0 2 19 0

Total A

240

64

B. Spec. Cases

Total 0 Sub 0 V

1

2

3

31

4

5

1

2

4 6 22 5 3 1 0 2 17

3

60

4

5

1

1 1

2 ce

quant 'when' Temporal phrase conditional si

13 4 63

5 3 34

1

Total B

80

42

1

1

Total 0 Sub 0 V

1

C. Wh # que

ce

3

1

31

4

0

29

4

3

4

29

2 ce

3

31

4

6 5 42 26 7 108

3 3 36 11 3 17

2 2 11 5

Total C

194

73

20

31

1

Total A, B, C

514

179

24

35

1

6

4 2 1

I.Q.: si I.Q.: wh # que NP Appos. Rel. CP Appos. Rel. R.Rel. # que comme 'as'

1

6

4

5

6 1

1 22 4 2 2

1

2 2 1 15

0

16

5

4 105

5

266

VERB SECOND AND THE NULL-SUBJECT PARAMETER

Table 10.3 Subordinator Spr V X (3884) A. Que

Sub Spr V

1

2

ce

3

31

4

5

6

purpose cause subject/object misc. concessive result optative comparative / superlative: que Rest. Rel. : que I.Q.: wh-que

30 143 1027 62 57 379 10

2 14 86 1 6 11 2

1 4 32 3 2 9 0

0 3 108 2 3 10 1

20 79 554 40 34 246 5

1 10 103 4 5 42 0

0 10 16 1 0 3 1

1 3 38 2 4 3 0

5 20 90 9 3 55 1

106 506 129

6 72 12

4 12 2

1 3 24

67 283 37

1 29 27

3 14 5

8 21 7

10 72 15

Total A

2449

212

69

155

1365

228

53

87

280

B. Spec. Cases Sub Spr V 1

2

ce

3

31

4

5

6

1

quant 'when' Temporal phrase conditional si

267 115 221

8 10 26

8 2 19

24 2 8

166 60 97

9 8 17

4 4 4

2 33

47 27 17

Total B

603

44

29

34

323

34

12

36

91

1

2

ce

3

31

4

5

6

C. Wh # que

Sub Spr V

I.Q.: si I.Q.: wh # que NP Appos. Rel. CP Appos. Rel. Rest.Rel. # que comme 'as'

66 122 87 99 183 275

2 17 15 12 38 38

2 2 I 0 1 7

2 11 0 4 2 0

42 67 55 63 98 78

6 0 4 8 11 39

0 1 3 0 14 1

3 7 4 0 5 90

9 17 5 12 14 22

Total C

832

122

13

19

403

68

19

109

79

3884

378

111

208

2091

330

84

232

450

Total A, B, C

3.1 Category A: que Clauses Part A of Table 10.2 contains the type of subordinate clauses which exclude null subjects in a verb-first construction, except for person 5 and perhaps person 4, where person-number morphological marking is transparent. The presence of null subjects of person 5 is striking, as there are 60 Spo subjects against 87 Spr subjects, i.e. 40.81% of Spo subjects. The licensing of Spo subjects is often related to the transparency of person-number inflectional marking on the verb; in this respect, one would not be surprised if person 4

NULL SUBJECTS IN CENT NOUVELLES NOUVELLES

267

were to behave very much like person 5. But there are only three examples of a Spo subject against 53 cases of a Spr subject for person 4, i.e. only 5.35% of Spo subjects. If a total of 56 cases is enough to feel confident that these 5.35% represent a regular syntactic phenomenon, but of low frequency, factors separate from morphological distinctness will have to be found to explain the difference between person 4 and person 5.1 have nothing to contribute on this aspect of the data at the present time. A comparison of Table 10.2 with the parallel section of Table 10.3 shows that the absence of null subjects from a person distinct from 5 or 4 is significant beyond any doubt, as there is only one null subject in that case against 2309 examples with a Spr subject. This unique example is of person 3 (against 1365 Spr examples of person 3). Thus the subordinators of category A do not formally license null subjects in SpecIP and prevent Infl from doing it. (3)

...et se print a luitter encontre celle tendre bergiere par tel maniere et par tel force que la ruyst au bas.... 069015 ...and [he] started to fight with this shepherdess in such a way and with such force that he threw her down...

One more comment needs to be made regarding category A of Table 10.2. Given their behaviour with respect to null subjects, comparatives-superlatives, restrictive relatives and clefts (grouped under R.Rel. que), as well as indirect questions headed by que (absent from Table 10.2 from lack of examples) have been grouped with conjunctive clauses introduced by que rather than with the wh-constructions found under category C. When we take into account only persons 3, 31 (expletive) and 6 (excluding persons 1 and 2, given that they are never represented by null subjects in category C, as well as person 4 and 5 because they could allow a null subject in SpecIP whatever the type of the embedded clause), there are 486 examples of wh-clauses introduced by que with a pronominal subject against none with a null subject. By comparison, the percentage of Spo as opposed to Spr is 34.54% for person 5, and 8.33% for person 4 (but again, there are only two examples of a null subject in this last case). Thus, these wh-clauses headed by que behave just as regular nonwh-que clauses. This is important since in other cases they have been assumed or shown to behave like wh-clauses in general (see for example, Adams (1988); Hirscbiihler and Junker (1988); Vance (1988); Roberts (1992)), the distinction between the two types of clauses headed by que being attributed then to the presence of a +wh feature in the que relatives. In the present case, one is led to conclude that in the CNNV, a wh-feature associated with que is not enough to formally license a SpecIP position that dominates pro. As we suggested in SECTION 1.2.4, the situation just described could be described by saying that when Infl/Agr is grammatically active with respect to SpecIP as the identifier of its content, it counts as a potential formal licenser for it, but not as an actual one, because of the filled embedded C or SpecCP. Competition with another potential formal licenser prevents +wh-que from formally identifying SpecIP. This concludes our survey of category A with verb-first embedded clauses. We now turn to categories B and C which will be discussed at some length.

268

VERB SECOND AND THE NULL-SUBJECT PARAMETER

3.2 Categories B and C A glance at categories B and especially C of Table 10.2 reveals that numberwise null subjects are nearly absent for person 1, 2, and 4. This leads to the tentative hypothesis that in the context of embedded clause introducers which (at least for a number of them) do not exclude the possibility of null subjects in the verb-first construction, certain persons of the conjugation are unable to identify null subjects in SpecIP. This hypothesis is based on a comparison between Spo and Spr for each person in Table 10.4 below. The percentages of Spo for each person with respect to the sum of Spo (Table 10.2) and Spr (Table 10.3) show that only persons 31 (the impersonal (uses of) verbs) and 5 have a high percentage of null subjects in the B and C embedded clause types. As for person 3, the substantial number of examples of that person, in particular in category C, allows us to conclude that the percentage of null subjects of person 3 reflects a regular syntactic phenomenon, simply of low, but not insignificant, frequency.8 Ce will be commented upon later. As there is only one example of a null subject of person 4, the higher percentage of this person compared to person 3 may not be taken as reflecting a regular phenomenon. Here, as in category A, person 4 differs from person 5 in that the clear morphological marking of person and number on the verb is not correlated with a high percentage of null subjects. Considering categories A, B and C, null subjects represent only 4.54% for person 4, while they represent 31.15% for person 5! This difference cannot be imputed to a formal difference in the morphological distinctness, as both these persons are equally morphologically distinct from all the others. Table 10.4 Proportion of Null Subjects with Respect to the Total of Sp Category B

42 603

Total (T2a) Spo Total (T 2b) Spr

2

ce

3

31

4

5

6

1 44

1 29

4 34

3 323

4 34

0 12

29 36

0 91

2.222 3.33i 10.552 0.9:2 10.52 0

6.9' 6

% Spo

2

ce

3

31

0 122

0 13

0 19

20 403

31 68

0

0

0

Sub Sp((3)V 1

Categ;ory C Total (T 2a) Spo Total (T 2b) Sp

73 832

8.06

% Spo Total B+C

1

Sub Sp(o)V

44.6il 0 5

6

1 16 19 109

5 79

4

4.72 31.31 5

12.8

5.95

Sub Sp(o)Vl

2

ce

3

31

4

5

6

1 166

1 42

4 53

23 726

35 102

1 31

45 145

5 170

Spo 115 Spr 1435 % Spo

7.35

0.59 2.32 7.01

3.16 25.54 3.12 23.68

2.85

NULL SUBJECTS IN CENT NOUVELLES NOUVELLES

269

3.2.1 Category B: Temporal and Conditional Clauses Three types of clauses have been considered here: clauses introduced by quant 'when', other temporal clauses, and conditional clauses. They will be examined in that order. A priori, one may be tempted to analyze quant as an item of category C, that is as a Wh-phrase. Although null subjects represent quite a small percentage (1.83%), one might want to rely on the mere existence of such examples as supporting the status of quant as a formal licenser. However, an examination of the data shows that quant does not appear to regularly license an empty element in SpecIP, and that it should therefore be grouped with elements of category A. A first argument comes from a comparison with some OF data. Old French texts like La Mort le Roi Artu and La Queste del Saint Graal exceptionally allowed cases of verb-first constructions with null subject in embedded clauses introduced by quant (as well as in conditional clauses, to which we come below), contrary to what was the case for other types of subordinate clauses. But, upon examination (cf. Hirschbiihler (1990)), the only unequivocal examples of null subject verb-first clauses introduced by quant noted in these texts were all variants of the same impersonal expression, quant vint... 'when [it] came to...', which alternates with quant ce vint... (cf. Zink (1987); Martin et Wilmet (1980, §325,2)). In the CNNV, the parallel expression is always realized with an expressed pronominal subject (quant se/ce vint...), which is unexpected if quant participated in formal licensing of SpecIP. Consider then the four examples of a null subject with an impersonal verb. They consist of two expressions: quant est de may 'as far as [it] is of me' (i.e. 'as far as I am concerned'; 026016; 098206), quant est a nous 'as far as we are concerned' (090277), and quant luy en souvint 'when [it] came back to him' (i.e. 'when he remembered'; 062017). In this last case, luy could actually be a strong form, i.e. it could start an embedded verb-second construction, or a clitic form, but there is no example with moi (strong) or me (clitic) that would allow us to decide. These types of examples could represent set expressions, relics of an older stage of the language. Importantly, there is no example with a null subject involving the type of impersonal expression that is found in the nine examples with an Spr subject (one of these cases, 091279, omitted below, is an example where quant is taken up by que, and so one might want to classify it with que): (4) a. ...car, quant il fait beau, je le vueil bien, et quant il pleust, aussi fais je. Quant il faict froit ou quant il faict chault, ...034033/34 ...since, when it is nice weather, I am happy with it, and when it rains, also am I. When it is cold or when it is warm, it is always all the same to me b. ...quant il vous plaira, (...when it it will please you)

048024

c. ...quant il tonneroit. (...when it it will thunder)

083102

270

VERB SECOND AND THE NULL-SUBJECT PARAMETER

d. ...quant il estoit question de paier taille au prince... ...when it was the moment to pay tax to the prince

091018

e. ...mais quant il failloit paier les feuz ou gabelle, 091065 ...but when it was necessary to pay for each house or for the salt tax. The final example of quant is not impersonal; it is of person 2. (5)

...quant 0 n'a seullement la puissance de saillir une Ms 039081 ...when [you] don't even have the strength to make love once

This is the only example of null person 2 in a verb-first subordinate clause in the whole text, and the only null personal subject with quant, as opposed to 258 pronominal personal (i.e. excluding the impersonal) subjects. Here the verb is preceded by the negative marker ne. It has often been noted that ne was present in the case of a null subject in contexts where a null subject was otherwise normally excluded. It is therefore possible that here also, the null subject in SpecIP should not be specifically related to the presence of quant. All this suggests that quant should be assimilated to conjunctions of category A as far as null subjects in an embedded verb-first construction are concerned, despite the fact that it might be analyzed as a temporal-free relative. There are three cases of a null subject distinct from person 5 in a temporal clause introduced by a complex phrase (avant que, depuis que, devant que). Two are of person 3 and one is of person 1 (the only case of that person in an embedded verb-first clause in the text). (6)

a. Mais avant que 03 ce partist, ... But before that [he] refl-left, ...

011008

b. ...je me suis cuidiez lassez morir depuis que 0l partys de ceans. 017056 ...I thought that I would nearly die since [I] left from here... c. ...et dit qu'il courrerait une lance devant que 03 entre au four... 096092 ...and he said that he would make love once before that [he] enter the oven I am not sure what to say here, given the small number of Spo against the 115 corresponding examples with a preverbal Spr in Table 10.3 (2.54% Spo), and in particular against the 60 examples of Spr of person 3 against only two Spo (3.22% Spo). Perhaps, as the possibility was raised in the case of quant, temporal clauses could be analyzed as involving wh-movement (operator movement). The difference between quant and these complex subordinate phrases could then be that quant would be treated idiosyncratically in the same way as wh-que, while the expressions with more phonetic weight would be treated in the same way as the wh-subordinateclause introducers of category

NULL SUBJECTS IN CENT NOUVELLES NOUVELLES

271

C as far as formal licensing of SpecIP is concerned. This would not, however, be enough to account for (6c) with a null subject of person 1, which is a weak Agr (see 1.2.2 and below). At this point, I consider these examples as potentially problematic. Consider finally the case of conditional clauses. In a total of 34 examples with Spo, 29 are of person 5 and the five others represent ce in four cases and person 3 in one case (against 118 examples with Spr for ce, person 3 and the impersonal, i.e. Spo represents 4.06% of all types of 3rd person singular). Given these last five examples, should conditional clauses be grouped together with the embedded constructions in category C, whose introducing element participates in licencing a null SpecIP? Probably not. It should be pointed out that in the Old French texts Mort Artu and Queste du Saint Graal, conditional clauses represented the second type of verb-first embedded clauses with null subject (the other one being the sentences in quant, as noted above). But all the examples were cases of the unique expression se ne fust..., i.e. si ce n'etait pour... 'if it were not for' (Hirschbiihler (1990)).9 In the CNNV, three or four examples are of this type: (7) a. ...et se 0 n'eust este qu'il tenoit sa chemise endroict le trou, ... 081129 ...and if [it ]weren't that he was holding his shirt against the asshole,... b. Et se 0 ne fut cela,... And if [it] weren't for that,...

096027

c. ...comme se 0 fut este toutes les punasies du monde,... 028135 ...as if [it] had been all the stench in the world,... Example (8) is possibly of that type, as the version with a pronominal subject would most probably be ce rather than il: (8)

...tous les gens des villaiges circonvoisin y venoient pour ouyr son sermon comme se fust estez Dieu,... 034005 ...all the people from the surrounding villages were going there to listen to his sermon, as if he had been God,...

In all these cases, given that omission of ce is otherwise exceptional in the text, the absence of ce here might be favored in order to avoid sequences like se ce..., in a manner reminiscent deletion of partitive articles du, de la, des after the preposition de in examples like je parle de chevaux 'I am talking about horses', from underlying je parle de des chevaux. This interpretation is however weakened by the fact that sequences like se ce are not categorically excluded: (9)

...se ce ne fut de peur de gaster mon lit,... ...if it was not for fear of damaging my bed,...

041067

Considering finally example (10), one notes that it is introduced by si rather than the more usual se. Se, as opposed to si, introduces the conditional clause in three quarters of the cases when a pronominal subject is present; in the case

272

VERB SECOND AND THE NULL-SUBJECT PARAMETER

of a null subject (with or without an XP appearing between the conjunction and the verb), the conjunction appears 60 times as se and only twice as si, one of those two cases being example (10). This example could therefore easily be reinterpreted as s'i' Veust, with a pronominal subject. (10) et n'y avoit homme, si 1'eust congneu et il 1'eust veu en cest estat,... 014048 and there wasn't [any] man, if [he] had seen him in this state,... So, provisionally perhaps, I conclude that it is doubtful that conditional si contributes in a productive way to the licensing of null subjects in the present text, and that the examples above are either fixed expressions, orthographic confusions, or, in the case where the null subject would correspond to ce, possibly the result of a "stylistic" deletion rule. The general conclusion is that there is no clear indication that the embedded clause introducers of category B are involved in the regular licensing of null subjects in the CNNV. We will see that things are different in the case of a number of wh-clauses.

3.3 Category C We will go into the details of the various clauses of part C of Tables 10.2 and 10.3. But first, let us have a brief overview of the main facts regarding clauses of this category. These are summarized in Table 10.5 (a subpart of Table 10.4). We may first note that there is not a single example of a null subject of person 1 or 2, while there are examples of null subjects for all the other persons. The facts are clear for person 1, as there are 122 examples with a pronominal subject of this person in clauses of type C; they are perhaps less clear for person 2, given that there are only thirteen examples with a pronominal subject. The absence of null subjects of persons 1 and 2 in clauses of type C is what led us to suggest in SECTION 1.2.2 that for persons 1 and 2, the capacity to identify the content of SpecIP via Spec-Head agreement is available only if Infl formally licenses the same position simultaneously. As in subordinate clauses the power to formally license SpecIP is normally reserved to the C system, realization of persons 1 and 2 by a pro subject will be excluded in the environment discussed. The only examples which go against the general ban against null subject of person 1 or 2 in a verb-first embedded clause are (5) and (6b), discussed above. There are two possible analyses for these. Either the complementizer licenses SpecIP, in which case identification by person 1 or 2 is exceptionally not "parasitic," or Infl acts exceptionally both as a licenser and as an identifier for SpecIP in a verb-first embedded context. In contrast to persons 1 and 2, the impersonal and person 5 are frequently null, although person 5 clearly not as much. One unexpected fact is that person 5 is found more frequently in category A of Table 10.2 than in category C (40.81% vs. 12.80%). For the other persons, there are only between 4% and 6% of null subjects.

NULL SUBJECTS IN CENT NOUVELLES NOUVELLES

273

Table 10.5 Spo vs Spr in Subordinate Clauses of Type C Person

Spo Spr

% Spo

1

2

0 122

0 13

0

0

ce

3

31

4

0 20 19 403

31 68

1 19

0

4.72

31.31

5

56

16 109

6

Total

5 79

73 832

12.80 5.95

8.,06

One factor that we did not discuss in SECTION 1 is the opposition between impersonal and non-impersonal verbs or uses of verbs in clauses of type C. In these clauses, the null subject is more frequent with impersonal uses than with person 3. The contrast may be attributable to discourse conditions or it may simply reflect a weaker requirement on local identification for the impersonal, i.e. person and number agreement could be a process independent from transmission of a referential index. We have no definite idea on this aspect of the data presently. These general considerations being made, let us take a closer look at the data of category C. This will provide more justification of what was said about wh-que in SECTION 1.2.4 as well as allow additional observations. In particular, we will see that there is no clear indication that null subjects are regularly allowed in indirect questions or in clauses introduced by comme. This would leave only the introducer of relative clauses as a regular formal licenser of SpecIP. If other texts display the same characteristics, it would show either that the licenser of the empty category in subject position is not simply the wh-word introducing the clause, or that other factors, still to be discovered, play a role. This detailed examination, some aspects of which perhaps may not be of immediate theoretical import for the analysis presented in this paper, will allow us to get a clearer picture of the grammar of the CNNV and may become important if new facts or new theoretical frameworks raise new questions. 3.3.1 Indirect Questions In the case of indirect questions, although null subjects are found in the CNNV, a close examination of the examples raises doubts as to whether they are productively licensed in this construction. Yes-no questions (introduced by si) and wh-questions will be examined separately. The main facts are summarised in Table 10.6. First, for the si of indirect questions, the percentage of null subjects is low (4.34%) and is based on only three occurrences of a null subject. As these null subjects represent persons 3 and 6, we might want to give them a lot of weight, more than if they were of person 31 or 5, for instance. But on the opposite side, we may also wonder if si here does not stand for s'i (sg. and pl.), in which case there would be no example of a null subject. Given previous work, we actually would not expect null subjects here, as the si complementizer does not belong to the category of wh-words.

274

VERB SECOND AND THE NULL-SUBJECT PARAMETER

Table 10.6 Spo vs Spr in Indirect Questions (excluding que) Category C

Q.I.: si Spr Q.I.: si Spo % Spo

66 3

1

2

ce

3

31

4

5

6

2 0

2 0

2 0

42 2

6 0

0 0

3 0

9 1

4.34

Q.I.: wh Spr Q.I.: wh Spo % Spo

Sub Sp V

122 3 2.4

—

4.54 17 0

2 0

11 0

67 2

0 1

1 0

10 7 0

17 0

2.89

(11) a. .. .luy demanda s'il.. .et si 03 seroit et demeureroit tousjours aussi asne et beste. 013036 ...[he] asked him if he...and if [he] would be and remain always as stupid and silly. b. je ne scay si 03 se doubta de ce qu'estoit... I don't know if [he] suspected what was going on...

041056

c. ...Dieu sail si 0 6 le raillerent... ...God knows if [they] scoffed at him...

052056

In the case of wh-indirect questions, there are again only three examples of null subjects out of a total of 125 (2.4%). These three examples are given in (12): (12) a. ... luy demanderent pour quoy 03 s'en estoit allez. ... [they] asked him why [he] had gone away

031029

b. Ores escoute's de quoy 03 se advisa. (I.Q rather than free rel.) 092023 Now listen to what he thought. c. ...je leurs diroie si bien qu'ilz m'entenderoient, assavoir mont de quoy ilz se meslent ne...ne de quoy 031 leur en appartient. 044052 ...I would tell them so well that they would pay attention to me, that is what they are meddling and...and what [it] matters to them. Although the percentage of Spo is small (close to 3% for person 3 and a little above 4% for 3 and 31 put together), these examples do not look problematic in any respect and we take them to be the first clear case where the choice of subordinate clause introducer participates in the licensing of a null subject. Indirect questions introduced by que display very different properties, as they are followed 129 times by a pronominal subject (Table 10.7) and never by a null subject.

NULL SUBJECTS IN CENT NOUVELLES NOUVELLES

275

Table 10.7 I.Q. Introduced by que with Spr Preverbal Subject Category A I.Q.: wh-que

Sub Sp V 129

1

2

ce

3

31

4

5

6

12

2

24

37

27

5

7

15

Indirect questions are the clearest case where que introducing an embedded clause is a wh-word; we consider it to occupy SpecCP, just as it would in a main clause. The absence of null subject is not only an indication that whque does not formally license a null SpecIP, it also shows that the verb does not move to C in the context under consideration, as otherwise we would expect a number of postverbal null subjects. Particularly strong support for the view that indirect question que is not a licenser comes from impersonal verbs, which represent the category most easily found with a null subject in relative clauses. Their total absence with a null expletive subject in que indirect questions contrasts with the high number (i.e. 27) of impersonal verbs with a pronominal expletive subject. (13) a. Et luy, qui..., ne sentit rien et ne savoit qu'i luy fut advenu, 091288 And he, who..., didn't feel anything and didn't know what it happened to him b. ...je te dires qu 'il te fault faire. ...I will tell you what it is necessary for you to do

094055

c. ...et veit bien quelle heure qu'il estoit.10 ...and [he] saw very well what time it was

091111

The variety of the verbs involved (advenir (15), falloir/failloir (7), sembler d (2), y avoir (2), etre (1)) reduces the possibility of an accidental gap in the data. 3.5.2 Relatives The relative clauses have been divided into appositive relatives with a nominal antecedent, appositive relatives with a CP antecedent, and restrictive relatives. If the three types of relative clauses are put together, it can be seen that the proportion of null subjects in this construction compared to that of Spr subjects is relatively important for some persons and that the number of examples may be high enough for the percentages to be reliable, in particular for persons 31 and 3, and perhaps 6. Table 10.8 Relatives not introduced by que Type Relatives Spo Relatives Spr % Spo

Total 1

50 363

0 65

11.89 0

2

3

0 16 2 216 0

6.89

31

28 23 54.90

4 5 6 Total

1 17

1 4 9 31

5.55 10 11.42

22 341 6.06%

276

VERB SECOND AND THE NULL-SUBJECT PARAMETER

Table 10.9 Spo vs Spr in Appositive Relatives

Category C Sub (Sp) V 1 ce 3 31 4 R.A. Spo R.A. Spr %Spo

36 87

0 15

0

1

0 0

29.26

0

0

0

11

5

6

55

22 4

1 3

0 4

2 5

16.66

84.61

25

0

28.57

Table 10.8 clearly shows that relatives not introduced by que allow null subjects, especially in the case of person 31, and to a lesser extent 3 and 6. Persons 1 and 2 seem to be categorically excluded. Some of the mechanisms involved have been discussed in SECTION 1.2. The discussion there is however incomplete as we refrain from being concrete as to what the property of a subset of relatives and indirect questions (those not introduced by que) is that leads to formal licensing of SpecIP. But see Santorini's approach in note 5. Appositive relatives have been divided in two groups, those with NP antecedent and those with CP antecedent. They are examined in Tables 10.9 and 10.12 respectively. NP appositive relatives have a large proportion of null subjects: 29.26% Spo against 70.73% of Spr. (To the examples of Spr, one may add three examples with on which have not been included in the table, as on has no null variant.) It is important to note the weight of the impersonal: of 36 null subjects, there are 22 examples of expletive ones; this is to be compared to only four examples of impersonal verbs with an expressed pronominal subject. Null subjects appear to be the rule in the case of the impersonal. The proportio of null subjects is high for person 3. It is higher for person 6, but here th

total number of occurrences of Sp (7) is not high enough to make definitive claims regarding that person. Attention will therefore be focused on persons 3 and 31. First of all, a closer look at the examples seems to justify a tentative division of NP appositive relatives in two classes. The first one, with 13 Spo examples, is characterized by the fact that the relative is introduced by a wh-word acting as a determiner of a noun or by a wh-word starting a new sentence after a full stop (or both). Table 10.10 Appositive Relatives of Type 1 Category C

Sub (Sp) V 1

2

ce

3

31

4

5

6

R.A. 1 Spo R.A. 1 Spr

13 6

0 0

0 0

0 0

3

10 1

0 0

0 0

0 1

68.42

—

—

— 42.85

%Spo

4

90.90 —

— 0

NULL SUBJECTS IN CENT NOUVELLES NOUVELLES

111

In all the examples of this type, the relative is preceded by a preposition, as in (14): (14) a. ...mais qu'elle volcist...obeir a sa requeste. A laquelle requeste 03 ne volut oncque obeir... 039019 ...as long as she would...obey his request. To which request [he] would never obey... b. Et premier arriva en Ytalie, auquel pays 03 trouva pluseurs choses estranges, 100055 And first [he] arrived in Italy, in which country [he] found sev eral strange things... c. et en icelluy prey y avoit pluseurs belles fleurs,..., entre lesquelles fleurs 031 y avoit de ces belles grandes marguerites. 043038 ...and in this meadow [there] were several pretty flowers,..., among which flowers [there] were [a number] of these nice and tall daisies. The ten examples with an impersonal all involve (y) avoir, which is never found with a pronominal subject in this group. These ten examples should perhaps therefore have no more weight than the unique occurrence of an impersonal verb with a lexical expletive subject. (15) Et advint ung jour et asses tost apres que les Rogacions ... vindrent, auquel jour il fault aller en pourcession trois jours durant... 023008 And [it] happened one day and soon after that the 'Rogations'...came, in which day it is necessary to walk in procession for three days. Although there are not enough data here to come to a conclusion, one might have the impression that null subjects here are restrcted to cases where the initial wh is a PP, as in the four cases where it is an NP, the subject is Spr. However, as we will see, data from the second class of NP appositive relatives militate against that impression. Let us turn to the second class of NP appositive relatives (Table 10.11), which simply consists of all those which do not fit the description of the first class. Table 10.11 Appositive Relatives of Type 2 Category C Sub (Sp) V

1

2

ce

3

31

4

5

6

R.A,.2 Spo R.A,.2 Spr

23 81

0 15

0 1

0 0

8 51

12 3

1 3

0 4

2 4

22.11

0

0

0

13.55

80

25

0

33.33

%Spo

278

VERB SECOND AND THE NULL-SUBJECT PARAMETER

In this category, null subjects are not limited to person 3 and 31. There is also a large variety of verbs: se pouvoir emerveiller 'to be able to be amazed', souvenir (2) 'to remember', y avoir (6) 'to be', etre compte 'to be counted', prendre devotion 'to take devotion', se doubter 'to be sure', penser 'to think', tarder 'to be late', etre congneu 'to be known', prendre 'to take', venir 'to come', donner 'to give', pouvoir aller 'to be able to go', s'accompagner 'to be accompanied', parler 'to speak', avoir 'to have', se trouver 'to be situated' (once for each verb). The fact that null subjects are well distributed among persons and that it is not restricted to a few particular verbs indicates that the verb-first construction with Spo reflects a productive grammatical phenomenon in appositive relatives. With 80% of Spo and the bulk of the cases, the impersonal is responsible for the high proportion of null subjects in the appositive relatives of type 2. These 80% have to be qualified by the fact that half of the cases of Spo involve y avoir. Even if we take out the examples involving this expression, we still have 66.66% of expletive Spo. The examples involve souvenir a quelqu 'un 'to remember' twice, tarder a quelqu'un 'to be delayed', prendre devotion a quelqu'um 'to honour someone', venir quelqu'un 'to come' and etre conte de quelqu'un par quelqu'un 'to be talked about by someone to someone', i.e. they involve obligatory impersonal verbs, unaccusative ones, and one impersonal passive. With 17.18% from a total of 64 Sp, person 3 confirms that null subjects are not an accident in NP appositive relatives. A few examples are given in (16). (16) a. ...une grainge joindant de leur maison, en laquelle 03 pourroit aller facillement et secrettement de nuyct... 047012 ...a barn joining their house, in which [she] could easily and secretly go at night... b. .. .il demanderoit 1'ausmone, laquelle 03 ne pensoit jamais qu'elle luy deust refuser... 018030 ...he would ask for charity, which [he] never though that she would refuse him... c. Symonnat, a qui 031 ne souvenoit plus de 1'an passes, vint a confesse. 002036 Symonnat, to whom [there] was no recollection of the previous year, went to confession... d. ...il arivait en 1'abayee aucun gentil homme de Lorraine, auquel 031 fut comptez des fais de messire Jehan Pare... 005067 ...there came to the abbey a gentleman from Lorraine, to whom [it] was told of the deeds of master Jehan Pare... e. ...je vueil icy compter ceste presente adventure laquelle 031 n'a guiere qu'elle est advenue a Mets. 041002 ...I want to tell here this present event which [it] is not long ago that it happened in Mets. [with resumptive pronoun]

NULL SUBJECTS IN CENTNOUVELLES NOUVEULES

f.

279

...nous avons trouvez ung homme mort et murtry...lequel 04 avons prins... 020189 ...we have found a man dead and killed..., which [man] [we] took...

g. .. .la ou on ce devisoit de pluseurs besongnes, entre lesquelles 06 parloient des divers marchez qui se tenoient en divers lieux par lescitez... 020007 ...where people were talking about different things, among which [they] talked about the different markets that were held in different places in the towns... Rega ing the hypothesis (suggested by the first class of appositive relatives) that Spo might be excluded if the fronted wh-phrase is an NP rather than a PP, the three examples with an NP that are found in the second class of NP appositive relative go against it. In SECTION 1.2.1, we adopted the often taken position that the wh-phrase itself played a role in formally licensing SpecIP. There is however another very natural hypothesis which deserves a few words. One might be tempted to associate the examples of NP appositive relatives of type 2 with Spo subjects to verb-second constructions in main clauses: the relative here can often be paraphrased by et 'and' followed by a non-wh-fronted phrase, which would normally induce V-to-Comp. At the present time, we reject that approach to the data, given the absence of Spo of person 1, as these are well attested in a postverbal position in the text. In CP appositive relatives, Spo makes up 9.48% of the total, which is just under half of what we saw for the second type of NP appositive relatives. There are enough cases to be sure that Spo is a regular phenomenon here. Null subjects are found for person 3, 31 and 6. There is no example of either Spo or Spr subject for person 2, 4 and 5. In the context of the absence of null subjects of person 1 in category C, we take the absence of Spo of person 1 here against twelve Spr as significant. The examples are characterized by the variety of verbs used with Spo (respondre 'to answer', ne pouvoir faire 'not to be able to do', se prendre & 'to start to', ne savoir 'not to know', estre mande' 'to be commanded', estre commande 'to be commanded', avoir ete possible 'to have been possible', estre ri 'to be laughed about' (2), ne vouloir souffrir 'not to want to suffer', dire 'to say'), which provides independent confirmation for the regularity of the construction. Some examples are given in (17). Table 10. 12 CP Relatives Category C Total R. Ph. Spo R. Ph. Spr

% Spo

11 99 9.48

ce

3

31

0 12

0 4

5 63

4 8

0

0

1

2

7.35

33.33

4

5

6

2 12 14.28

280

VERB SECOND AND THE NULL-SUBJECT PARAMETER

(17) a. Si luy demande le seigneur devant-dit qu'il voulloit faire de ses escailles d'oefz. A quoy 03 respont qu'i les voulloit vendre... 02002 The Lord asked him what he wanted to do with the shells of the eggs. To which [he] answers that he wanted to sell them... b. Ores advint une nuyt qu'il avoit permis d'aller veoir et visiter icelle saincte dame, laquelle chose 03 ne pouvoit bonnement faire... 048006 There came a night when he got permission to see and visit this holy woman, which thing [he] could not really do... c. ...il luy souvint du verlet au barbier et de son beau membre, par quoy 03 se print si tres fort a rire que... 093041 ...he remembered the barber's servant and his nice organ, by which [he] started to laugh so much that... d. .. .il y eust ung Cordellier preschant la Resurrection a qui on avoit comptez 1'hystoire nouvellement advenue, qui recita ce compte audit sermon et en plaine predicacion, de quoy 031 fuit asses ris. 072083 ...there was a Cordelier..., who recited this story in the sermon and in the midst of the predication, as a result of which [there] was a lot of laughter. There are only three cases of Spo as opposed to 183 cases with Spr, i.e. 1.61%. The percentage goes up to 2.04% if person 1 and 2 are excluded, on the basis of the fact that they are never realized as Spo in category C. In the case of person 3 there are 98 cases of Spr whereas there are none of Spo. Impersonal verbs fare better with two Spo against eleven Spr, i.e. 15.38%. The one case of Spo for person 5 against five with Spr tells us nothing, given that person 5 in principle allows Spo in any sort of subordinate clause. The three examples are given in (18): (18) a. .. .elle pensa a demander le nom du recepveur a qui 031 luy failloit parler,... 082023 ...she though of asking the name of the tax collector to whom [it] was necessary that she talked,... b. ...car ledit prebtre fist faire ung gros pastez...auquel 031 y avoit dedans des bons oysellet... 048031 ...because the said priest had someone made a large pie...in which [there] was in-it good birds c. ..., car avec ceste hystoire de Braie laquelle 05 aves icy devant ouy... 083064 ...because, with this story about Braie that [you] just heard... The general conclusion is that if it were not for the two cases of impersonal Spo, we would say that characteristics of the CP system do not provide for formal licensing of SpecIP in restrictive relatives in the CNNV.

NULL SUBJECTS IN CENT NOUVELLES NOUVELLES

281

As we saw in Table 10.8, when the three types of relative clauses are put together, the proportion of null subjects in this construction compared to that of Spr subject is relatively important for some persons. The number of examples may perhaps be high enough for the percentages to be significant, in particular for person 31 (54.90% on 51 examples), 3 (6.89% on 232 examples) and perhaps 6 (11.42% on 35 examples). However, looking more closely, it is clear that the vast majority of examples with Spo come from NP and CP appositive relatives, and that restrictive relatives not introduced by que rarely contribute an Spo example. It is therefore clear that some factor (either grammatical, discourse or other) should allow us to distinguish between the two types of relatives. One hypothesis that we briefly considered was that in NP appositive relatives: the same hypothesis could have been made for CP appositive relatives, the fronted wh-phrase was taken as a topicalized constituent triggering V-to-C, exactly in the way that any non-wh-phrase fronted to SpecCP results in V-to-C. We do not adopt that approach, given the 27 cases of Spr in person 1, vs none with Spo, in NP and CP appositive relatives. As made clear earlier, we have not identified the particular factor responsible for formal licensing of a null subject in SpecIP in some wh-clauses. As a result, it is useless to speculate at this point on what the relevant difference between NP and CP appositive relatives and restrictive relatives—and for that matter, wh-indirect questions perhaps—is. Clearly, this area is open for further research. 3.3.3 Comme In order to be complete, one last type of clause must be considered, i.e. clauses introduced by comme 'as'. In this case, apart from person 5, where null subject is licit independently of the complementizer, there are only two examples of a null subject, both impersonal. In the fifteen cases of person 5, we find the same verb in the same formula fourteen times (comme aves ouy 'as you have heard' thirteen times and comme aves cy devant ouy 'as you have here before heard' once), and in a variant of the formula once (comme aves tousjours dit 'as you have always said'), which results in the percentage for Spo being very low if the one expression commonly used with Spo is discarded. If, as should be the case, person 5 is not taken into account, facts are very similar to what was seen in restrictive relatives not introduced by que, i.e. the remaining null subjects are limited to the impersonal. Table 10.13 comme Category C comme Spr comme Spo

% of Spo

2

ce

3

31

4

5

6

38 0

7 0

0 0

78 0

39 2

1 0

90 15

22 0

5.82 0

0

—

0

0

14.28

Sub Sp V 1 275 17

4.87

0

282

VERB SECOND AND THE NULL-SUBJECT PARAMETER

(19) a. ...tout fut perdus comme 031 en est de nous,... 078164 ...everything was lost, as far as we are concerned b. ...je volroie des poullaine a grant soulez comme 031 se pourtoit au vy temps. 091099 ...I would like to have pointed shoes as [it] was the fashion in the old days. It is not inconceivable that the rarity of the verb-first construction here is simply the result of a stylistic preference resulting in impersonal verbs being preceded by a complement in initial position of IP if there is no formal subject expressed. This would be supported by the large number of verb-second constructions with a non-subject in initial position embedded under comme, i.e. there are 54 occurrences of comme dit est 'as is said' and 37 other cases, among which 22 are with an impersonal expression and 15 with a personal use of the verb. Examination of the 22 impersonal verbs involved in the verbsecond construction, the two examples in (19), and the 39 examples with the expletive il appearing before the verb did not lead us to see anything distinguishing the cases with a fronted complement from those with the expletive il.

3.4 Summary on Spo in Embedded Verb First We have looked in minute detail at the different types and subtypes of constructions, in order to ascertain whether there were finely grained differences between them. Excluding the examples with person 5, it appeared that null subjects in a verb-first construction existed to some degree in wh-constructions, but with unexpected differences. First of all, it seems that in the text considered, wh-clauses introduced by que, whatever their type, do not allow the construction. The construction is also extremely rare in wh-indirect questions, in the remaining restrictive relatives, as well as in the examples introduced by comme. The construction seems to be relatively common in appositive relative clauses not introduced by que. Under the approach described in SECTION 1, where it is assumed that some property of C or SpecCP (whose particular nature I have not committed myself to) is associated to formal licensing of a null SpecIP, we do not expect those differences. As far as I can tell, although it may very well be the correct explanation for the data for which it was originally put forward, the account in terms of a feature [+finite] located in C does not seem to provide more light on the data considered here than saying that a feature associated with wh-phrases is transferred from SpecCP to C, which feature is responsible for formal licensing of an empty SpecIP. There are also important differences according to the person considered in wh-constructions. There is no example of a null subject for persons 1 and 2, and there seem to be enough examples of a pronominal subject for person 1

NULL SUBJECTS IN CENTNOUVELLES NOUVELLES

283

(122 with Spr in category C, and 27 in NP and CP appositive relatives) for it not to be an accidental gap. There are examples for all the other persons, and the numbers are quite high in the case of the impersonal. Although the percentages are not very high in the case of person 3, it reaches 6.89% in relative clauses not introduced by que, and goes up to 11.94% (i.e. sixteen Spo against 118 Spr) in the case of appositive relatives (NP+CP). They are high enough, and they involve enough examples, to be confident that they reflect a real grammatical feature of the language. The same conclusion is probably correct for person 6, but the total number of relevant examples (5.95% Spo of a total of 84 examples for all wh-clauses not introduced by que, and 11.42% of a total of 35 examples for relative clauses not introduced by que) may make the conclusion weaker. Person 5 is well represented in category C (12.80%), but not as much as in que clauses, something for which I have no explanation. And despite its morphological characteristics, person 4 is represented less than might have been expected, with 5%, including only one example of Spo. As for the impersonal, the contrast that exists between category C (and in particular appositive relative clauses not introduced by que) and category A, as well as the fact that, as we will see, Stylistic Inversion, resulting in a verbfirst construction, is possible in wh-clauses introduced by que while the impersonal verb-first construction is excluded from this context, indicates that there is a null subject (pro) that needs content identification in standard impersonal uses, while there is no such null subject in SpecIP in the Stylistic Inversion construction (or that if there is, it does not need identification). Since Infl does not have an active relationship with SpecIP in this case (as it relates to the expressed postverbal subject), the +wh complementizer (which is not phonetically realized in the case of indirect questions, que being a wh-phrase in SpecCP, and which is realized as que in C° in the other constructions) is able to license formally the SpecIP position. This concludes our discussion of Spo in the embedded verb-first construction. In the next section, we briefly examine cases where the embedded clause introducer is not immediately followed by the verb.

4. Sub XP (Subject) V One position adopted at the beginning of this paper, following insights from Roberts, is that, in embedded clauses, whether or not SpecIP is formally licensed depends on properties of C, and not on Infl. This was discussed in relation to the embedded verb-first construction. In the analysis defended here, the verb-first construction is an [lPproVX] construction. The SVX construction may be part of a larger construction where a first constituent (or a series of such constituents) is adjoined to IP, as in the following examples: (20) a. ...mais d'une chose je vous prie... ...but one thing I ask you...

002045

b. Et ce fait, il bouta ce maistre tuppin dedans le mur... 011025 And this being done, he put this large pot in the wall.

284

VERB SECOND AND THE NULL-SUBJECT PARAMETER

We saw that this construction allows for null subjects, in other words, Infl or Agr licenses and identifies a pro subject on its left in SpecIP, as in (21), which is similar to (Id) above. As the participial phrase is never found with a postverbal pronominal subject, it is clear that the null subject is to the left of the tensed verb: (21) Et cela faict, marchanda audit poinctre de luy faire ung sepulchre And this done, [he] made a deal with the painter that he would paint a sepulchre. 011032 The same type of constructions are found in embedded clauses, i.e. with Spr or Spo, as clearly shown in (1c) above where Spr and Spo alternate. The distribution of each construction is given in Table 10.14 and Table 10.15 with examples following each table: Table 10.14 Sub XP Spr V (72+9=81) A: Que Sub XPSp V 1 purpose cause 5 27 complement 1 misc concessive result 19 optative 1 comp.-superl. 2 Rest.ReL: que

56

1 1 2

2

3

31

1 1

2 12

3

ce

4

1 3

4

1

13

1

1

1

1

2

5

6

1 2 2

28

4

1

4

B: Spec. Cases Sub XP Sp V 1

2

3

31

ce

4

quant 'when' temporal phrase conditional si Total B

Total A, B, C

2 5

81

5

6 on

1 1

1 1

5

6 on 1

3

3

C: Wh # que Sub XP Sp V I Q.I.: si 0 Q.I.: wh # que 6 NP Appos. Rel. 5 CP Appos. Rel. 6 R.Rel. # que 2 comme 'as' 20 Total C

6 on

1 2

6

Total A

5

3 1

2

31

ce

4

4 4 3

1 2 1

1 1 1

11

4

10

3

2

42

4

1

4

1 1

1

3

2

7

9

NULL SUBJECTS IN CENTNOUVELLES NOUVELLES

285

Table 10.15 Sub XP V X (with null subject)11 (335/514 = 65.17% Embedded Spo )

A: Que purpose cause complement misc. concessive result optative comp.-superl Rest.Rel.: que

Total Sub X V 0 2 7 10 16 92 114 13 8 4 7 47 45 2 2 4 6 28 9

240

Total A B: Spec. Cases quant 'when' temporal phrase conditional si Total B

Total

13 4 63 80

C: Wh excl. que Total

1

3

1

5 36 4 2 28

1 1

1

1 1

1

1

4

4

Sub X V 0 1

2 1

176

8 1 29 38

0

1

Sub X V 0 1

2

6 Q.I.: si 5 Q.I.: wh # que 42 NP Appos. Rel. 26 CP Appos. Rel. 7 R.Rel. # que 108 comme 'as' comme dit est comme XP V Total C 194

54 37 121

2

514

335

6

Total A, B, C

2

3 2 6 15 4

2

31 1 4 43 2 1 8

4

5

6 1

1

4

7 1 1 7

1

2 6 83

61

1

5

18

3

31

4

5

6

4 4

7 1 15 23

8

2

0

8

2

31

4

5

6

3 1 1 4 4 2

2 1

1 1 2 7

1

2 1

1

0

2 14

54 23 88

1 2

9 9

2 6

5

101

172

3

22

26

(22) a. ...vecy toutes les chandeilles que oncques en ma vie je acheta ...here are all the candles that ever in my life I bought 091125 b. Je ne s§ay si depuis ilz en joyrent. I don't know if since then they enjoyed them.

047061

c. ...comme cy apres vous oyres. ...as thereafter you will hear.

020000

(23) a. ...puis que aultrement 03 n'en pouvoit faire... ...as differently [he] could not do...

016042

286

VERB SECOND AND THE NULL-SUBJECT PARAMETER

b. .. .demanderent au painctre qu'i luy failloit que tant de fois 03 les rappelloit... 087055 ... [they] asked the painter what he was missing so that so often [he] would call them back... c. ...comme cy apres 05 ones... 091008 /...comme cy apres 031 s'ensuit. 092005 ...as hereafter [you] will hear/...as hereafter [it] follows d. Je ne sgay se depuis 06 en eurent quelques parolle I don't know if since [they] talked about it.

065111

e. la journee vint que une fois 03 se trouva leans aveques ses desirees amours... 041023 the day came when once [he] found himself there with his beloved f. Laquelle.. .luy fist meilleur semblant que les aultres fois 03 n'avoit faict... 068012 Which (one)...was nicer to him than the other times [she] had been... g. Et fut long temps en ceste peine sans ce que jamais 03 osait dire son piteux cas a son mary 045101 And [she] was for a long time in this pain without that ever [she] dared tell her pitiful situation to her husband. A look at these examples and at the additional ones in the CNNV shows that the construction is found in all types of embedded clauses, i.e. complement clauses, relatives, comparatives, clauses introduced by comme, etc. This is important because, in contrast to this situation, all the 25 examples of postverbal pronominal subjects in embedded clauses are found only in complement clauses, as in (24):12 (24) a. ...il affermoit...que se n'estoit il point. ...he was swearing...that that this was he not

005025

b. Les autres moinnes ont dit que voirement avoient il une geline... 005126 The other monks said that for sure had they a chicken... c. Je te promes...que tout ainsi luy coupperez je la teste comme... 043052 I promise you...that exactly in this way will I cut his head off as... d. Luy...voyant...que, s'il se courroucoit, encor pourroit il estre batus... 053088 He...seeing...that, if he became angry, in addition could he got beaten up

NULL SUBJECTS IN CENTNOUVELLES NOUVELLES

287

Along the lines of Adams (1988), examples of that type can be analyzed as embedded CPs, with the first phrase after the subordinator in SpecCP and the verb in C. These examples, with the verb governing SpecIP, allow for a null subject, but as I indicated just above, examples like those in (23) are not of the verb-second type, since they are found in all types of embedded clauses, in particular clauses in which a postverbal Spr is excluded. In addition, they are often introduced by phrases that are not found with a postverbal pronominal subject, i.e. that do not trigger V-to-C and are attested in conjunction with a preverbal pronominal subject, as can be seen by comparing some of the examples in (22) and (23). The array of facts discussed suggests that null subjects in examples of the type in (23) must be accounted for exactly like the parallel cases in main clauses, i.e. Infl is responsible for both formal licensing and identification, despite the presence of a subordinate clause marker. I assume that here Infl is able to act as the formal licenser of the embedded SpecIP because it is the only potential formal licenser for SpecIP. The first C up is not a potential formal licenser of SpecIP because it does not govern it, as the result of the intervention of the XP adjoined to IP. Given that Infl is the formal licenser of SpecIP in the construction discussed, we expect null subjects of person 1 and 2 to be found. There are six examples in the text of person 1 (25) and five of person 2 (26) (the fifth example is in (lc)): (25) a. je me repens fort que jamais 0l vous feis venir ceans. 018137 I regret very much that ever [I] made you come here b. nonobstant qu'en ma vie plus chandeilles 01 n'acheta 019128 despite [the fact] that in whole my life I had not bought more candles c. ...de la mesure que ainsi 0l as oublies ...of the measure that so [you] forgot

029023

d. va a tous les deable, que jamais plus 01 ne te voie. go to the devils, that never more [I] see you

066046

e. par quoy a ceste heure 01 vous en vueil dire aucune chose 080001 as a result of which at this hour [I] I want to tell you something about it f. Par quoy, en acquitant ma promesse, 0l vous dir6s de celuy chareton 082069 By which, by accomplishing my promise, [I] will tell you of this carter (26) a. Ne quel affaire a tu eu que au moins 02 ne t'es venu confesser 013019 And what happened to you that at least [you] didn't come to confess yourself

288

VERB SECOND AND THE NULL-SUBJECT PARAMETER

b. ... quant a moy 02 ne peust touchier. ...when me [you] can not touch

039083

c. et me dis...comment celle faulte que envers moy 02 as commis t'est advenue 094044 and tell me...how this misdeed that against me [you] committed happened to you d. ...je te le pardonne, mais que plus 02 n'y renchiesse 094046 ...I forgive you..., as long as [you] don't fall again in it.

5. Conclusion Most of the main points having been stressed several times in the text, this conclusion will be brief. First of all, we saw that it is necessary to establish a tripartition between the various types of subordinate clause introducers according to whether they contributed to the formal licensing of SpecIP, or not. The division is not simply between elements that do and elements that do not formally license SpecIP, since we have seen that wh-que did not license SpecIP when it was connected to Infl (in the sense that this category was the source of the potential identification of pro in SpecIP), while it did in the Stylistic Inversion construction (where we took the position that SpecIP is not associated with Infl). Second, we saw that it is necessary to establish a tripartition between the various persons. Person 5 is always allowed, persons 1 and 2 never (when adjacent to a complementizer), and persons 3, 31 and 6 are allowed in whclauses. Moreover, a detailed examination of the various types of wh-clauses has shown that a null subject in a verb-first construction is not equally well represented in all of them. At this point, we have no clear idea as to why this is so, in particular as to whether this is to be attributed to the way formal licensing works, or whether it follows from different considerations, having to do with discourse, for example, or with the author's stylistic preferences. As a final note, I hope that the observations that have been made here have shown that the discovery of some regularities—as well as the discovery that some facts which appear at first sight to go in one direction are, upon close examination, cases of unproductive formulas—requires much more detailed descriptions than those which are usually found in the literature, however tedious this may be at times.

Notes *

This research has been made possible thanks to grants 410-87-0332 and 41089-1131 from the Conseil de Recherches en Sciences Humaines du Canada.

1. Philippe de Vigneulles was a draper from Metz. He wrote his CNNs between 1505 and 1515. He has been described as a semi-literary author.

NULL SUBJECTS IN CENTNOUVELLES NOUVELLES

289

2. For a comparison between the CNNV and the anonymous Cent Nouvelles Nouvelles (c. 1462), see Hirschbiihler (1992). 3. For the sake of clarity, we will tend to use IP to refer to the highest maximal projection immediately dominated by CP. 4. There is no example of null subjects of the 2nd person singular in a verb-first construction in main clauses, but we take this not as the reflection of a prohibition against that construction with a 2nd person sing., but rather as being related to the fact that there are only a handful of examples (i.e. 12) of null subjects of the 2nd person sing. in assertive main clauses in this text. 5. The interlinear translations are literal translation for the most relevant parts, and more idiomatic English glosses for the rest. The first three numbers after an example correspond to the number of the short story, the next three to the line in that story. So, '002042' means story number 2, line 42. A null subject will be indicated by "0"; when present, a numeric subscript indicates the person. 6. Santorini (1992) develops an analysis of Yiddish based on Platzack and Holmberg (1989). She suggests that a finiteness operator [+F] is located in Comp (for verb-second languages) or Infl (for non-verb-second languages). [+F] licenses Nominative Case under government (strict c-command) and/or agreement configurations, depending on the language. In addition, she defends the idea that empty expletives must be (i) licensed by being head-governed by a case assigner, and (ii) identified by being co-indexed with Agr. The analysis sketched in the text could be reformulated by saying that in whclauses not introduced by que, C° contains the [+F] feature, while this feature would be absent from the other embedded clauses. It could then be said that licensing in Vigneulles is restricted to head government by a case-assigner when there is a configuration of head-government at hand. The "when-clause" is to ensure that licensing under Spec-Head agreement is possible in Vigneulles when there is no potential head governor, i.e. in verb-first assertive main clauses of the SVO type (whether they be analyzed as CPs or IPs). Considering embedded clauses, an empty SpecIP could not be licensed by a [+F] feature in Infl given that there is a higher position (Comp) where a [+F] feature might have appeared. One attractive aspect of the analysis is that it identifies more precisely than in the approach adopted here the particular grammatical property responsible for the licensing of SpecIP under government. I have not yet committed myself to this analysis because I do not see why in embedded clauses the finiteness feature would be located in C0 only in those whclauses not introduced by que (as I consider that the interrogative que is a whword in SpecCP, it could not be that the [+F] feature is present in C only when C is devoid of lexical material); I also have difficulty conceiving a finiteness operator distinct of Infl (or Tense). 7. '31' represents impersonal subjects. 8. According to Santorini (1992, n. 10), 1% is the level at which a phenomenon does not reflect a regular feature of the language: "On the basis of detailed quantitative work of my own and others, it appears that it is common for wellestablished generalizations in a language to be violated in naturally-occur-

290

VERB SECOND AND THE NULL-SUBJECT PARAMETER

ring usage at a low, relatively constant rate of about 1%. For instance, the relative frequency of resumptive pronouns in English in non-island environments is around 1% (Anthony Kroch, p.c.)." 9. The null subject is parallel to ce rather than il, and the omission of ce could therefore be due to avoidance of strings like se ce. See the discussion in the text to follow. (i)

...et vous dis bien, ..., que se ce ne fut de peur de gaster mon lit,... 041066 ... and I tell you, ..., that if it were not for fear of spoiling my bed,... 10. Wh-phrase quelle heure is followed by the complementizer que, the example has been classified with the I.Q. introduced by que. 11. Some of the subjects are preverbal, others postverbal. See Vance (1988) for a discussion on how preverbal null subjects may have arisen in Middle French as a result of the coexistence of "XP Spr V," "XP V Spr" and "XP V" with null subject. 12. These examples are also distinct from those with a null subjects in that the types of initial XPs are much more restricted. The XP preceding the verb may be an adverb or an adverbial phrase: si (twice), ainsi (twice), aussi (twice), a peine (5 times), voirement (3 times), tout ainsi, en mal an, au moins, encore au debout de quinze jours; an object as se (= ce) (2 fois) and moult de telles femmes; or a combination of several phrases: aufait de bien mentir, a cela, par Dieu, voirement ands'il se courroucait, encor.

References Adams, M. (1987a) "From Old French to the Theory of Prodrop." Natural Language and Linguistic Theory 5:1-32. Adams, M.(1987b) Old French, Null Subjects, and Verb Second Phenomena. PhD Dissertation, UCLA. Adams, M. (1988) "Embedded Pro." In J. Blevins and J. Carter, eds. NELS 18, 1-21. Amherst: University of Massachusetts, GLSA. Deprez, V. (1988) "Stylistic Inversion and the Structure of COMP." Escol Proceedings. Dupuis, F. (1988) "Pro-drop dans les subordonnfes en ancien francais." Revue quebecoise de linguistique theorique et appliquee 7:41-62. Dupuis, F. (1989) L'expression du sujet dans les propositions subordonnfes en ancien francais. These Doctorale, UniversitS de Montreal. Hirschbiihler, P. (1990) "La legitimation de la construction Via sujet nul en subordonn6e dans la prose et le vers en ancien frangais." Revue quebicoise de linguistique theorique et appliqute 19:32-55. Hirschbiihler, P. (1992) "Uomission du sujet dans les subordonnees V1: Les Cent Nouvelles Nouvelles de Vigneulles et les Cent Nouvelles Nouvelles anonymes." Travaux de Linguistique 25:25-46.

NULL SUBJECTS IN CENTNOUVELLES NOUVELLES

291

Hirschbiihler, P. and M.O. Junker (1988) "Remarques sur les sujets nuls en subordonnee en ancien et en moyen francais." Revue quebecoise de linguistique theorique et applique'e 7:63-84. Martin, R. and M. Wilmet (1980) Syntaxe du moyen francais. Bordeaux: Sobodi. Platzack, C. and A. Holmberg (1989) "The Role of AGR and Finiteness in Germanic VO Languages." Working Papers in Scandinavian Syntax 43:51-76. Rizzi, L. (1986) "Null Objects in Italian and the Theory of pro." Linguistic Inquiry 17:501-557. Roberts, I. (1992) Verbs and Diachronic Syntax. Dordrecht: Kluwer. Santorini, B. (1992) "Variation and Change in Yiddish Subordinate Clause Word Order." Natural Language and Linguistic Theory 10:595-640. Vance, B. (1988) Null Subjects and Syntactic Change in Medieval French. PhD Dissertation, Cornell University. Vance, B. (1989) "The Evolution of Prodrop in Medieval French." In J. de Cesaris and C. Kirschner, Studies in Romance Linguistics, 413-441. Amsterdam: John Benjamins. (Appeared also as "L'evolution de Prodrop en francais medieval," Revue quebecoise de linguistique theorique et appliquee 1 (1988):85-109. Zink, G. (1987) "«Quant ce vint au congietprendre». De ce anaphorique a ce auto-referentiel en ancien frangais." In Etudes de linguistique generale et de linguistique latine offertes en hommage a Guy Serbat. Bibliotheque de 1'Information Grammaticale, 417-426. Paris.

This page intentionally left blank

Part Three Clitics and Verb Second

This page intentionally left blank

11 The Diachronic Development of Subject Clitics in North Eastern Italian Dialects* Cecilia Poletto Universities of Venice and Padua

Introduction The aim of this paper is to provide some insight into the evolution f subject clitics of Northern Italian Dialects from the Renaissance period to the present stage. It will be shown that subject clitics are strictly related to the head of the Agreement projection. In particular it will be argued that subject clitics have been reanalyzed as heads that take on functions normally related to the Agr head in Standard Italian. From a relatively homogeneous stage Northern Italian Dialects have developed different systems in which subject clitics have specialized as the prodrop licenser head, or the Nominative Case-assigning element, or can even occupy an additional Agr head. In SECTION 2 and 2.1 the Renaissance Veneto dialects will be shown to be exactly parallel to Renaissance French, both in the treatment of subject clitics as well as subject DPs and in the pro-drop system which is activated by the presence of a "strong" Agr or C head (where strong is defined as carrying a particular grammatical feature). SECTION 3 deals with the Veneto dialect of the 16th century which presents the system of a full pro-drop language in which subject clitics are specified as pro licenser heads. Subject clitics can specialize not only as pro-drop licensers, but as Nominative Case-assigners, also, as is the Case for some modern varieties (cf. Poletto (1993a)). In the last two sections it will be shown that the spectrum of the functions played by subject clitics can be even wider: a special series of subject clitics will be shown to appear only with auxiliaries, in order to lexicalize a higher Agreement projection available only to auxiliaries as verbs that do not assign theta-roles. The most advanced variety, namely Friulano, shows a very widespread use of subject clitics which signal the presence of another Agreement head that attracts clitics. 295

296

CLITICS AND VERB SECOND

Northern Italian dialects seem to have reanalyzed subject clitics as a competing head that replaces some of Agreement's syntactic functions. The analysis of this phenomenon can thus help us to define the mechanisms that are involved in the syntactic mapping of the relation between a subject and its predicate.

1. Subject Clitics as Heads Subject clitics of the Northern Italian Dialects (cf. Brandi and Cordin (1989) and Rizzi (1986a)) are considered in the literature as the realization of morphological agreement features placed under the head of the syntactic Infl node, and not as true subject DPs which appear in the Spec of IP. On the other hand, French subject clitics are considered to be in the same position as subject DPs. Adopting Belletti's (1990) hypothesis about the order of the functional projections, this analysis of subject clitics can be rewritten as in (1) where subject clitics appear in an adjunct position to the head of AgrP:

In (1) the head of AgrP assigns Nominative Case to the subject, which is placed in its specifier position. This is the position in which a null subject is licensed in Standard Italian. Northern Italian Dialects (from now on NIDs) are pro-drop languages just as Standard Italian is. In (1) in fact, a pro is licensed in the SpecAgr position. Nevertheless the Agreement structure of NIDs is more complex: a subject clitic appears adjoined to the head of AgrP, where the verb has moved from

THE DIACHRONIC DEVELOPMENT OF SUBJECT CLITICS

297

the V position through T in order to incorporate the Tense and Agreement morphemes placed respectively under T and Agr.1 We will briefly review the arguments used by Rizzi (1986) and reported by Brandi and Cordin (1989) in order to show that subject clitics of the NIDs are heads, because these tests will be important for the following discussion about their development from the Renaissance period to their present status. Subject clitics are considered to be heads because they appear to the right of the preverbal negative marker, while subject DPs and French subject clitics appear on the left: (2)

To mama no vien Your mother not comes

Venetian

(3) Elle ne vient pas She not comes not

French

(4)

Venetian

No la vien Not she comes

As the position of NID subject clitics is to the right of the preverbal negation marker, while subject DPs (and French subject clitics) appear on its left, (cf. (2)-(4)), we cannot assume that NID subject clitics occupy the same position that DPs fill at S-structure.2 Another test that reveals the status of NID subject clitics as heads adjoined to Agr is Agr' coordination. It is a fact that NID subject clitics have to be repeated in coordinate structures, while subject DPs and French subject clitics can be omitted in the second conjunct of the coordination: (5) Nane lese el giornale e fuma un toscan John reads the newspaper and smokes a cigar

Venetian

(6) II lit le journal et fume un cigare He reads the newspaper and smokes a cigar

French

(7) El lese el giornal e *(el) fuma un toscan Venetian He reads the newspaper and *(he) smokes a cigar In (5) the subject DP Nane can be omitted in the second member of the coordination, the same is possible for French subject clitics as (6) shows, but in NIDs this is excluded. In (7), in fact, the sentence is grammatical only if the subject clitic is repeated. This contrast can be explained only accepting that the subject clitics in the NIDs are structurally closer to the inflected verb than a normal subject DP, and precisely that subject clitics occupy a position under Agr', while subject DPs occupy the SpecAgr position. On the contrary, French subject clitics occupy a DP position, namely SpecAgr, and cliticize to the inflected verb only at PF (cf. Kayne (1975)). In purely structural terms, it can be assumed that subject clitics in NIDs are adjoined to the head of Agr as in (1). According to Belletti (1990), the inflected verb moves up to the Agr head position in order to incorporate the Agreement morpheme.

298

CLITICS AND VERB SECOND

As (I) shows, the subject clitic is adjoined to this head and this explains why subject clitics cannot be separated from the inflected verb by any other element than other clitics. In the dialects studied by Rizzi (1986a) and Brandi and Cordin (1989) the subject clitic is always obligatorily expressed, even if a subject DP is present: (8)

La Maria la magna The Mary she eats

Trentino

(9) La Maria magna The Mary eats Also the contrast between (8) and (9) suggests that the subject clitic is not a true subject but a sort of morphological specification that is always expressed on the head of Agreement, independently of the element that is realized in SpecAgr which can be a null subject or a phonetically realized DP. In order to avoid the possibility of interpreting (8) as an instance of left dislocation of the subject DP, Rizzi observes that the subject clitic is obligatory even when the subject DP is a Quantifier phrase, which cannot be left dislocated:3 (10) Tut *(l)'e capita de not Everything it is happened by night

Trentino

(11) Tout (*il) s'est passe' dans la nuit

French

While the Trentino data in (10), show that the subject clitic has to co-occur with a quantifier subject, this is not possible in the French example (11). Subject clitics of the NIDs are thus a part of the Agreement morphology and not true subject pronouns. NIDs correlate typologically with French, because they have subject clitics, but their structure is similar to Standard Italian because they are pro-drop languages. This assumption also explains why the series of subject clitics is not complete for all persons in most NIDs, while it is complete in French, where subject clitics behave as subject DPs with respect to the tests presented here. A closer examination of the distribution of subject clitics in other NIDs shows that not all subject clitics have the distribution described by Rizzi (1986a) and by Brandi and Cordin (1989). In particular, the tests in (4) and (7) are valid also for the subject clitics of Veneto that we will examine here, suggesting that they are all heads. On the contrary, the distribution of subject clitics can vary with respect to subject DP. Not all subject clitics can appear when there is a phonetically realized subject DP in the sentence. As proposed in Poletto (1991), I will assume that subject clitics in NIDs can be distinguished on the basis of a movement versus basegeneration analysis. As proposed by many authors (cf. in particular Koopman and Sportiche (1991)), I will assume that the subject is generated inside the VP, and precisely in the SpecVP position, where it gets its theta-role assigned and it is raised successively to SpecAgr in order to get Nominative Case. I will refer

THE DIACHRONIC DEVELOPMENT OF SUBJECT CLITICS

299

to this subject position inside the VP as the basic argumental subject position. When a subject clitic is generated in the basic argumental position inside the VP, it gets the subject theta-role, which is assigned in that position, and then moves to Agr. No other subject can occur in this structure because the basic subject position is occupied by the trace of the subject clitic. If the subject clitic on the contrary is base-generated in its surface position in Agr, it is an expletive, deprived of the subject theta-role. The subject thetarole is in fact assigned into the lower position in the sentence structure inside the VP. As the basic subject position inside the VP is empty, it can be filled by another DP which absorbs the subject theta-role. So, subject clitics that are generated inside the VP and then moved to Agr can receive the subject theta-role and are argumental clitics, while subject clitics base generated in Agr are expletive elements, as they do not have any theta-role. The tests that permit us to distinguish between argumental and expletive clitics are the following: (12) a. 1'ha paria qualcheduni cl has spoken somebody b. *E1 parla qualcheduni cl speaks somebody In (12a) the subject clitic can co-occur with a subject DP which is realized in the postverbal subject position, while the clitic in (12b) cannot.4 So the subject clitics described in (10) for Trentino can be assimilated to the expletive clitic in (12a) because they are compatible with a subject DP in argumental position. Subject clitics of the type of el cannot appear if the subject DP has been moved through wh-movement (as for instance restrictive relatives, topicalization or clefting), while subject clitics of the type of l can:5 (13) a. El puteo che (*el) vien vanti... The boy that (*he) comes along b.

Ti che *(te) vien vanti You that *(you) come along

(14) a. NANE, che (*el) vien vanti... JOHN, that (*he) comes along b.

TI, che *(te) vien vanti YOU, that *(you) come along

(15) a. Ze Nane, che (*el) vien vanti Is John, that (*he) comes along b. Te si TI, che *(te) vien vanti You is YOU, that *(you) come along.

Veneto

300

CLITICS AND VERB SECOND

(13), (14) and (15) represent respectively cases of restrictive relative clause, topicalization and clefting. In all these cases the third person subject clitic cannot co-occur with the variable trace, while the second person singular subject clitic can (indeed it must). The explanation for the contrasts in (12), (13), (14) and (15) is that, as mentioned above, argumental subject clitics leave a trace in the basic subject position through which the subject theta-role is transmitted. Hence they cannot co-occur with another subject, which would occupy the position of the trace. Non-argumental subject clitics on the contrary are base-generated in their superficial position, leaving the basic position free for another subject, which is the OP nisuni in (12) and the variable trace of wh-movement in (13b), (14b) and (15b). Some dialects have both expletive and argumental clitics; the Veneto variety that we used for the examples above is precisely this kind. Other varieties realize only one of the two possibilities. From a diachronic point of view, it is interesting to investigate how subject clitics of the NIDs have developed to reach their present status. Have they always been heads like today, or were they similar to French in some previous stages of evolution? Renzi (1989) has shown that Fiorentino of the 18th century was like Modem Standard French with respect to the distribution of subject clitics. If this is true, the same could be valid for North Eastern Italian dialects too, in particular for Veneto (cf. Vanelli (1987)). In the following section the tests presented here will be applied to Veneto of the Renaissance in order to determine which syntactic status subject clitics have in this period.

2. The Veneto Variety in the Renaissance The subject clitic system of Veneto of the 15th century was complete for all persons as the schema in (16) illustrates:6 (16)

1. a/e

2. te/ti

3. m. el f. la

1pl. a/e

2pl. a/e

3pl. m. i f. le

expl. 1

As Vanelli (1987) notes, these subject clitics do not present any of the features that induced Rizzi (1986), and Brandi and Cordin (1989) to characterize subject clitics as heads and not as maximal projections. In other words the position of subject clitics of Renaissance Veneto (henceforth RVe) does not correspond to (1): subject clitics do not form a cluster with the inflected verb within the head of AgrP. They seem to behave as true subject DPs as Modern French subject clitics do. In fact they can be left out in a conjoined structure and never appear after the negative marker (cf. SECTION 1): (17) El m'ha lago le cavale (...) e si _ando in la (Ruzante p. 78) RVe He to me has left the mares and so _went away (18) a. E no podeva tior.... (Calmo p. 66) I not could take...

THE DIACHRONIC DEVELOPMENT OF SUBJECT CLITICS

301

b. Che te no vissi ma (Ruzante p. 91) That you not see never c. La no vaga a mio conto (Calmo p. 79) She not goes on my count d. El no puol eser altrimenti ca benedeto (Calmo p. 94)7 He not can be other than blessed e. E no se inganemo (Calmo p. 66) We not ourselves mistake f. Ch'un passo i non fare (Ruzante p. 74) That a step they not make (+future). (17) and (18) show that subject clitics of this period are independent items that appear in the position that DPs fill, namely SpecAgr, and as such they can be left out in a coordinated structure. At this stage subject clitics do not seem to be different from Modern French subject clitics. Subject clitics are argumental clitics in the sense that they start out from the basic position of the subject inside the VP and absorb the subject thetarole. In fact they are incompatible with a QP in the subject position, as (19) shows, and they never co-occur with a variable trace of the subject as in questions or in relative clauses:8 (19) a. Ognon vora acomodarse de si bela stampa (Calmo p. 66) RVe Everyone will take for himself this beautiful picture b. Chi volesse formar un teatro de bontae (Calmo p. 96) Who would like to be a theater of goodness c. Quante persone che vedera ste cossete stampae (Calmo p. 66) How many persons that will see this little things printed The subject clitic does not normally appear even when the subject is a DP: (20) Un'arma longa fa sta indrio el so nemigo (Calmo p. 96) RVe A long weapon 'makes stay behind the enemy The same is true if the subject is a tonic pronoun: (21) Mi ve adoro (Calmo p. 128) RVe I (+stress) you adore As we are examining a dead language, it is impossible to determine for sure if the sequences quantifier-subject clitic or wh-subject clitic are ungrammatical. The only negative proof that can be given is the absence of such a sequence in the corpus examined, which consists of the first 100 pages from a play by the author Ruzante for the Paduan variety and of the first 100 pages from a letter collection by the author Calmo for the Venetian variety. From the fact that they alternate with the subject DP in SpecAgr (cf. (20) and (21)), we can conclude that subject clitics of the Veneto varieties of this period are not Agreement morphology in the sense that they are not always obligatorily

302

CLITICS AND VERB SECOND

realized, as verbal agreement morphology is, independently from the element that appears in the preverbal subject position SpecAgr. On the basis of the examples regarding coordination and the position with respect to negation in (17) and (18) we can conclude that subject clitics of RVe are not heads that adjoin to the head of AgrP where the inflected verb is. On the basis of the distribution of subject clitics with respect to a subject QP or to a subject variable trace, we can assume that subject clitics are true arguments in RVe, (cf. (19)) because they absorb the subject theta-role. In RVe, as in the Fiorentino variety of the 18th century studied by Renzi (1989), subject clitics are not yet reduced to heads adjoined to Agreement, they are independent syntactic DPs like Modern French subject clitics are. The fact that subject clitics in RVe are similar to their Modern French counterparts does not entail that RVe is a non-pro-drop language like Modern French. On the contrary, it is quite common to find examples of null subjects.9 Nevertheless their distribution is complicated by the fact that the possibility of a null subject seems to vary with respect to the main versus embedded character of the sentence. In the following discussion we will consider separately main and embedded clauses. As Vanelli (1987) noted, null subjects are more numerous in embedded clauses than in main clauses. In particular, they are found in embedded sentences when an element like si 'if, a wh-operator or a subjunctive complementizer occupies the head of the Comp projection. In the literature there are some well-known cases of asymmetry between main and embedded sentences, as for instance the verb-second phenomenon, and they are all treated as a function of the difference between the C head of a main clause, which is not selected and in some cases just not present, and the C of an embedded clause, which is in some intuitive sense the head of a clausal argument. Then it seems reasonable to treat the difference noted with respect to null subjects in RVe as a function of the head C. Let us first consider the data. Expletive null subjects of verbs that do not assign a theta-role to the subject are possible in both main and embedded clauses: (22) a. E certo che... (Calmo p. 97) RVe Is sure that b. ...manco mal _ sarave a dir (Calmo p. 74) ...luckly (it) means that... Nevertheless, null subjects are not obligatory: it is possible to find examples of expletive subject clitics realized in both main and in embedded contexts: (23)

El me par che' 1 sarave cossa giusta (Calmo p. 1ll) RVe It to-me seems that it would be right thing

As in (23) the preverbal subject position is occupied by a subject clitic both in main and in embedded contexts, we have to state that RVe pro-drop is in some sense "weaker" than that of Modern Italian. In Italian the expletive element that occupies the preverbal subject position can only be a null element, while RVe has the choice between the two possibilities.

THE DIACHRONIC DEVELOPMENT OF SUBJECT CLITICS

303

It is interesting to note that there is a difference between the distribution of expletive subject clitics in the case of a verb which does not assign a thetarole to its subject and cases of expletive subject clitics with postverbal subjects. An expletive clitic with a postverbal subject can only be omitted in embedded sentences if the element in Comp is a wh-item, si 'if or a subjunctive complementizer and never appears in main clauses, as (24) shows: (24) a. L'e pur una dolce cossa (Calmo p. 99) RVe It is indeed a sweet thing b. Si _ no resta altro (Calmo p. 94) If _ not remains (anything) else In (24a) the expletive element is a subject clitic which occupies the SpecAgr position. No null subject is licensed in this structure. A pro subject can in fact only be licensed in embedded clauses with a particular type of complementizer, as (24b). The contrast between (24a) and (22a) indicates that there must be a difference between an expletive subject which does not get any theta-role and an expletive subject which is coindexed with a postverbal thematic position. The difference noted between an expletive pro which is connected to a postverbal subject and a expletive subject of a verb which does not assign a theta-role to its subject is the same that we find among the persons of the verb. Second person singular, and third person singular and plural argumental null subjects can only be realized in embedded sentences, if the Comp projection is filled by a wh-item, si 'whether' or a subjunctive complementizer. In main clauses a second person singular or third person singular and plural subject is always realized as a subject clitic, never as a null element. (25) a. ...Com fa 1'orsa quando _ se guz gi ongi (Ruz. p.105) RVe As does the bear when _ sharpens her claws b. Dire a Ser Zuan che _ la guarda ben (Ruz. p. 107) (You) will say to Sir John that (he) look (+subj) at her well (26) a. ...Che tuta la zente, co _ li vede, se ghe inchina (Calmo p. 75) ...That all the people, when (they) them see, bow b. Si farae megio... (Ruz. p. 102) Whether (they) would do better to... In other words, the possibility of a pro depends on the features realized in C. In a main clause, C is not realized at all. Hence it cannot license anything, because it is not present. In an embedded clause, C is always realized, because it contains the selectional features assigned by the matrix verb. Nevertheless, not every C is able to license a null subject. Only a C marked by some feature, as for instance the feature +wh, is strong enough to license a null subject. If C does not contain any particular feature, it cannot license the null element, then the subject has to be phonetically realized as in main clauses. This entails that the normal subcategorization features assigned by the matrix verb to the embedded clause and which are supposed to be realized in C do

304

CLITICS AND VERB SECOND

not count for C to be a pro licenser. The intuition is that C counts for the prodrop theory only if it is "visible" in some sense to be defined. On the other hand, the distribution of argumental null subjects of first person singular and plural and second person plural does not seem to be dependent on any feature in C. There are examples of null subjects of first person and second person plural both in main and embedded sentences: (27) a. Ve suplico (Calmo p. 72) RVe (I) pray you b. Havemo buo notita che.. (Calmo p. 129) (We) have had news that... c. Dire a Ser Zuan che... (Ruz. p. 107) (You+plur.) will say to Sir John that... (28) a. Co avesse ben dissenao (Calmo p. 1ll) When (I) had well dined b. Quando aspetemo suto, ... (Calmo p. 73) When (we) await dry weather,... c. Si vole scambiar tuto... (Calmo p. 94) If (you+plur) want to exchange everything... (27) shows that a first person singular and plural and a second plural null subject is possible in a main clause. Hence, a particular type of Comp (such as a +wh or a +subjunctive one) is not relevant for the licensing of the null subject. The relevant head that licenses and identifies the contentive features of the null subject must then be the head of the Agreement projection. At this point we have two classes of null subjects. True expletives and first person singular and plural and second person plural null subjects can be licensed both in a main and in an embedded context. On the other hand, expletives coindexed with an argumental subject position, second person singular, and third person singular and plural null subjects are sensitive to the type of element which is realized in the Comp position: only a +wh or a +subjunctive Comp can license this type of pro. The situation is summarized by the following schema: (29)

MAIN CL

expletive pro -theta l.person + sing.pro l.person +

EMBEDDED CL. EMBEDDED CL -wh /subjunct. +wh /+subjunct. + + + +

+

+

+

+

+

plur.pro

2.person plur.pro

+

THE DIACHRONIC DEVELOPMENT OF SUBJECT CLITICS

2.person sing.pro 3.person sing.pro 3.person

305

-

-

+

-

-

+

-

-

+

-

-

+

plur.pro

expletive pro+postv.DP

Null subjects can thus be divided into two groups. We will refer to the first group of null subjects which are not sensitive to the type of Comp as "extended pro-drop." The second group of null subjects which can only be licensed if the Comp projection has a particular type of feature (+wh or +subjunctive) will be termed restricted pro-drop. Looking at the distribution of expletive subjects and argumental subjects in RVe, it is evident that the pro-drop conditions in RVe are strongly reminiscent of the situation in Renaissance French (henceorth RFr) type of pro-drop studied in Roberts (1992) (see also references quoted there). In RFr the distribution of the null subjects as described by Roberts (1992) can be resumed as follows: expletive subjects, first person plural and second person plural null subjects can be found in both main and embedded clauses. On the other hand, first person singular, second person si gular, and third person singular and plural can only be licensed in embedded contexts and only if there is a +wh-item in the Comp projection of the sentence. The distribution of null subject in RFr is thus the following: (30) expletive pro l.person plur.pro 2. person plur.pro l.person sing.pro 2.person sing.pro 3.person sing.pro 3.person plur.pro

MAIN CL.

EMBEDDED CL. -wh/-subjunct.

EMBEDDED CL. +wh/+subjunct.

+

+

+

+

+

+

+

+

+

-

-

+

-

-

+

-

-

+

-

-

+

The table in (30) shows exactly the same partitioning of table (29) between extended and restricted pro-drop.

306

CLITICS AND VERB SECOND

If we compare the distribution of null subjects in RFr with the distribution of null subjects in RVe, the similarity is striking: in RFr only first person plural and second person plural null subjects are admitted both in main and embedded clauses independently of the features of C. In RVe only first person singular and plural and second person plural null subjects are admitted both in main and embedded clauses. The only difference concerns the first person singular, which behaves like a restricted pro-drop in RFr, while in RVe it behaves as an extended pro-drop. Once we have stated that RVe and RFr share the same double system of extended versus restricted pro-drop, let us examine how the system can be formalized within the context of the theory of prodrop elaborated in Rizzi (1986b) that we are assuming here. One simple observation concerns the head that licenses a pro. Both C and Agr can be pro-drop licensers. Hence we have to formulate the pro-drop parameter for RFr and RVe as containing two licensing heads, namely C and Agr. In both languages it seems that only a head marked with some special feature is able to license a pro. This observation is not only valid for C, but also for Agr. In fact, only a morphologically strong Agreement, like, for instance second person plural is visible for the pro-drop licensing condition, but a weak one, like for instance third person, is not. Let us assume that only if Agr or C are "strong" can they license a null subject. The definition of "strong" must include both a distinct phonetically realized morpheme as is the case for Agr and a particular feature like +wh or +subjunctive, as is the case for C. So, if C is strong, as in +wh and +subjunctive embedded clauses, null subjects are licensed for every person. In main clauses, where C is not active, only a strong Agr can license a null subject: given that only first person and second person plural are strong, null subjects are possible only for these persons. Formalizing this idea we obtain: (31) a. C is a pro-drop licenser if it is strong b. Agr is a pro-drop licenser if it is strong (32) a. C is strong when it contains a +wh or +subjunctive feature b. Agr is strong when it contains a morphologically realized +person and +number feature. A system like that described in (31) and (32) generates the split between extended and restricted pro-drop that we have seen in (29) and (30) for RVe and RFr. So the difference between extended and restricted pro-drop derives from the fact that C is not always marked with a strong feature, while Agr, once it selects a strong feature that includes both number and person, must always realize it. The fact that C can be strong is thus determined by the syntactic environment, while this is not the case for Agr. If the systems of RVe and RFr are really parallel, how is it that Veneto has developed following a different evolutionary line with respect to French? Why has Veneto become a pro-drop language where subject clitics are heads in Agreement, while French has developed into a non-pro-drop language? I do

THE DIACHRONIC DEVELOPMENT OF SUBJECT CLITICS

307

not think that verbal morphology is of such a great importance in this matter that we can attribute the different evolution of these two languages only to the difference in the number of the morphological distinctions on the inflected verb. In other words, the richer morphological paradigm of the Veneto variety is not the only factor that has determined the evolution of this language into a pro-drop language. I would like to connect the different evolution of French and Veneto not only with the number of morphological specifications on the verbal head, but with the relative balance between verbal morphology and the paradigm of the subject clitics. The different evolution of RFr and RVe is a particular case of a generalization formulated by Renzi and Vanelli (1983), which states that the subject person and number features must always be phonetically expressed by Agreement or by the subject itself. In other words there must always be at least one element, verbal morphology or the subject pronoun itself, that expresses the number and person features of the subject. This seems to be true for all Romance dialects examined by Renzi and Vanelli. Both RFr and RVe have a restricted system of pro-drop and six subject pronouns which appear in the SpecAgr position. But, in RVe the subject clitics of first person singular, plural and second person plural have the same form a or e depending on the variety (cf. (16)). In RFr the series of subject clitics has a distinct element for all persons of the verb. Hence even in RVe, not only in Modern Veneto, the morphologically realized inflectional features are the only elements able to identify the number and the person of the subject. Even if the subject clitic is in SpecAgr, it has no features that could convey information about the subject. As in RVe subject clitics are not always distinguished for person and number, so RVe has no other choice than to maintain the person and number features on Agreement, reinterpreting subject clitics as heads for the persons that are not fully specified by verbal morphology. French on the other hand, having a full discrete series of subject clitics, has been able to keep them as true DPs, further limiting the role of Agreement as pro-drop licenser. Hence, the factor that has determined the split between French and the Veneto variety (and probably other Northern Italian dialects as well) is not only the different number of morphological specifications on the verb. It is the relation between the number of morphological specifications on the verb and the number of morphological specifications on the subject clitics. It is interesting to note, however, that both languages have evolved in a way that respects Renzi and Vanelli's generalization: the person and number features of the subject are realized at least once in both languages. The necessity of expressing these features can thus be considered not only as a synchronic property of Romance dialects in general, but also a diachronic tendency to maintain a sort of balance between the features expressed in Agr and in its Spec position. I will now examine a problem which is closely connected with the pro-drop system and the distribution of subject clitics, namely postverbal subjects. There is another quite interesting problem that is connected with the facts discussed up to now, namely the free inversion cases in RVe noted by Vanelli (1987) with a subject clitic in preverbal position.

308 CLITICS AND VERB SECOND

Vanelli observes that examples like (33) constitute a puzzle for Case theory, given the hypothesis that subject clitics are true subjects in RVe (cf. SECTION 2): (33) a. El viene quel so fraelo (Ruz. p. 94) RVe Cl comes that his brother b. L'e sta suspeso le prediche al Sior Geronimo (Calmo p. 15) Cl is been suspended the sermons to Mr. Geronimo (33a) presents a case of postverbal definite subject with an ergative verb and a subject clitic which is realized in preverbal position. On the basis of the discussion about the position of subject clitics it is clear that they cannot be considered as morphological affixes at this stage of evolution. They are true DPs which absorb the Case of the subject. The problem for the theory is presented by the fact that the definite subject in the postverbal position needs a Case, too. It is generally assumed that two phonetically realized elements cannot be assigned the same Case (cf. Kayne (1983)). So, in this structure we need two distinct Cases, one for the subject clitic and one for the postverbal subject DP.10 Looking at verbal Agreement it seems that the Nominative Case is assigned to the subject because the verb agrees with the clitic and not with the subject DP. In (33b) the postverbal DP is feminine plural, but the verb is marked as masculine singular on the past participle and as singular on the auxiliary. We will thus assume that the subject clitic in preverbal position absorbs the Nominative Case, as the verbal morphology indicates. What about the postverbal DP? The Case assigned to the postverbal DP cannot be Accusative, because the verb is an ergative one. It cannot be the Partitive Case postulated in Belletti (1988) either, because Partitive is assigned only to indefinite DPs and the DPs in (33a,b) are both definite. So the Case assigned to the postverbal DP can be neither Nominative through Spec-head agreement with the head of AgrP nor Partitive. In order to solve this problem, we have to consider how Nominative Case is assigned. I will assume Roberts' (1992) idea that Nominative Case can be assigned in two different configurations: Spechead Agreement with the head of AgrP and government by the head of TP. The possibilities of Nominative Case-assignment correspond thus to (34) (cf. Roberts (1992:29 ff.)): (34) a. Agr assigns Case through Spec-head agreement, b. T assigns Case through government. Such a parameter of Nominative Case-assignment has been proposed by Roberts in order to explain the difference between languages such as French and Welsh. In French the subject appears in the preverbal subject position and it triggers morphological agreement of person and number with the verb. Following Roberts' hypothesis, French exploits the possibility expressed by (34a). Hence the subject DP moves from its base position inside the VP to the SpecAgr position, where it is assigned Case and it triggers morphological agreement of number and person.

THE DIACHRONIC DEVELOPMENT OF SUBJECT CLITICS

309

On the other hand, in Welsh the subject appears after the inflected verb and it does not trigger morphological agreement in person and number. This means that Welsh adopts (34b): the subject DP does not need to move to SpecAgr, on the contrary it must remain in situ, in order to get Nominative Case assigned by the head of TP. Given that that there is no Spec-head agreement relation between the subject DP and the head of AgrP, there is no morphological agreement of number and person. Roberts further assumes that in the Romance languages the subject can be in the postverbal position because both options in (34) can be selected: Nominative Case can be assigned both by Spec-head agreement with the head of AgrP or by government from the head of TP. Nevertheless, languages like Standard Italian always show morphological agreement of person and number between the subject DP and the verb, while Welsh never does. Roberts explains this difference on the basis of the observation that in Welsh AgrP is never active in Nominative Case-assignment, while it is in Romance. On the basis of this difference, a rule of cosuperscripting between the heads of AgrP and TP applies in Romance, but not in Welsh. (35) Coindex Agr and T A rule like (35) will thus be active in the Romance languages because both Agr and T are able to assign Nominative, but it will fail to apply in Welsh, because Agr in Welsh is inert with respect to Nominative Case-assignment. This cosuperscripting determines the passage of morphological agreement features of person and number so that the verb and the postverbal subject agree in person and number in Romance. Let us now consider the structure of sentences like (33):

310

CLITICS AND VERB SECOND

In (36) the subject clitic el is realized in SpecAgr, while the postverbal subject is inside the V governed by the head of TP. It is possible to think that in RVe, as in other Romance languages, both mechanisms of Case assignment can be exploited, namely that the head T can assign Case to the postverbal subject DP through government and the head of AgrP can assign Nominative through Spec-head agreement. It is interesting to note, however, that in RVe (as in Modern NIDs) no agreement of person and number between the verb and the subject DP appears to be active. In other words, RVe is more similar to Welsh than to Standard Italian and other Romance languages. We have to assume that the rule of cosuperscripting postulated in (35) for Romance languages does not apply here, but why? Also in RVe there are preverbal subjects that trigger morphological agreement of person and number with the verb. Hence also in RVe the AgrP projection is active for Nominative Case-assignment exactly as in other Romance languages. Is the fact that the rule of cosuperscripting fail to apply a mere coincidence or not? And, if it is not, is it connected with other selectional choices that the grammar of the dialect in question makes? It seems plausible to think that the fact that rule (35) does not apply in RVe is somehow connected with the particular type of postverbal subjects observed in this dialect. In other words, the fact that there is no cosuperscripting must be related to the problem of Case assignment to the postverbal subject in a structure like (36). We already excluded the possibility that the postverbal subject DP receives Partitive Case, because it is a definite DP. It cannot receive Nominative Case through a chain with the expletive, because the expletive is a phonetically realized element, and it needs a Case of its own. Now, Kayne (1984) proposes that two phonetically realized elements cannot be in the same chain and share the same Case, while an overt and a silent element can. Considering the Nominative Case-assignment possibilities expressed in (34), we can make the hypothesis that Case is assigned to the postverbal subject DP by the head of TP, while the expletive in preverbal position receives Nominative through Spec-head agreement with the head of AgrP. At first sight, it might seem strange to assume that two Nominative Cases are assigned at the same time, even in different structural configurations and by different structural configurations and by different heads. Note however, that a system of Nominative Case-assignment like (34) does not specify anything about the possibility that both heads assign Case at the same time. In a language that selects both heads Agr and T as Nominative Case-assigners it could be the Case that the two heads are both active, and that two different DPs get Nominative Case, one through Spec-head agreement with the head Agr and one through government by the head T. This double mechanism of Case assignment is restricted by theta theory that admits only one DP for each thematic role assigned by the verb. So, even if there are two possible Nominative Cases available, only one of the two will be realized, because there is only one subject theta-role. If both Nominative Cases are assigned to two distinct DPs, one of the two will be left without a thematic role, violating the theta criterion. There is only one situation in which a DP can be left without a thematic role, namely where DP is an expletive.

THE DIACHRONIC DEVELOPMENT OF SUBJECT CLITICS

311

A structure with double Case assignment is thus possible only when one of the two elements is an expletive. Furthermore, the expletive element must be the higher one, because SpecAgr is the non-thematic position. If the expletive were realized in the postverbal position and the subject DP in the preverbal one, it would be impossible for the subject DP to receive the subject thetarole, which is assigned inside the VP. The only situation in which the two Nominatives can be assigned thus corresponds to a structure like (36) which does not violate the theta criterion. The subject clitic in SpecAgr is in fact an expletive, and as such it does not absorb the subject theta-role, while the postverbal subject DP does. Hence, RVe has the possibility of exploiting both options expressed in (34) at the same time. Moreover, it must do so, otherwise one of the two elements would remain without a Case. This kind of analysis does not seem necessary for languages such as Standard Italian or Standard French. In Standard Italian in fact there is a null element in preverbal position, and not a phonetically realized one. In this case Kayne's restriction about the presence of two elements sharing the same Case does not apply, because one of the two is silent. Hence, Standard Italian does not need to exploit both options of Nominative Case-assignment described in (34) at the same time. The same is true for French postverbal subjects in the case of Stylistic Inversion: in the preverbal position a pro is licensed probably by a +wh C (see Kayne and Pollock (1978)), and the subject DP receives Case directly from the head of TP. Once we have seen how the mechanism of Case assignment works in a structure like (36), we can go back to the hypothesis that it may be connected to the difference that we noted before with respect to morphological agreement of person and number. In Standard Italian and French postverbal subjects trigger morphological agreement with the verb, while in RVe this is not so. In order to explain this fact, we assumed Roberts' cosuperscripting rule between the heads of AgrP and TP to be active in Romance but not in RVe. It seems that when the cosuperscripting rule applies, the two heads of AgrP and TP are treated as one, both with respect to the morphological features of person and number and with respect to Case assignment. We can thus assume that the rule of cosuperscripting blocks independent Case assignment by the two heads that are able to assign it. Hence the double head constituted by Agr+T can only assign Case once: through government or through Spec-head Agreement. On the contrary, when the cosuperscripting does not apply, the two heads are considered as distinct elements by the grammar: they do not share morphological agreement features and can both assign Case independently. In RVe the rule of cosuperscripting cannot apply, otherwise one of the two Nominatives would get lost and the lexical expletive or the postverbal subject DP would remain without a Case. Moreover, a structure like (36) is the one in which the possibility of a double Case assignment is realized, because it is the only situation which is not blocked by the theta criterion. If the rule in (35) does not apply, no sharing of the morphological features between Agr and T is possible: hence the verb must agree with the preverbal expletive clitic and not with the postverbal DP. In particular, we expect that

312

CLITICS AND VERB SECOND

there will be no examples of a lexical expletive in free inversion structures in which the verb agrees with the postverbal subject. A structure like (37) should never be found: (37) *L'e vegnudi i to fradei Cl are come+ plur. agr. your brothers This seems to be true, in particular in the case of RVe, as far as I could test. As L. Vanelli pointed out to me, this fact seems to be general in NIDs. The solution that we propose here for RVe inversion could possibly be adopted also for other languages, as for instance the Occitan varieties, the Fiorentino variety of the 18th century and Modern Popular French studied by Renzi (1989). The situation in Fiorentino seems to be more or less parallel to RVe, as Renzi has shown. Our prediction seems to be correct at the present state of knowledge concerning these languages. There is another important consequence that derives from this analysis that deserves some brief comments. The solution presented here in fact does not directly connect pro-drop and free postverbal subjects as consequences of the same parameter. This hypothesis seems to be confirmed by other Romance languages, as for instance Portuguese, which has the possibility of null subjects but does not show the possibility of free postverbal subjects. This seems to be correct also on the basis of languages such as Occitan, and Modern Popular French which do not show null subjects but admit free inversion. However, the mechanism exploited by RVe in structures like (36) must be a more marked choice in the grammar because two heads, which are normally very closely connected, are compelled to be kept separate and are independently active in assigning Case at the same time. In other words, it is probable that the unmarked choice for Romance corresponds to the cosuperscripting between Agr and T, given the generalized movement of the inflected verb up to both heads, and that a coalescence of morphological endings of Tense and agreement is quite often observable in this group of languages. Therefore, the languages that exploit the mechanism described for RVe must be less numerous than the languages that exploit the Standard Italian system, in which there is only one Case for the chain, given that one of the two elements is empty.

3. Veneto of the 17th Century In this section I will consider how subject clitics and the pro-drop system of RVe further developed into a variety which still survives in some very conservative areas. The text examined is the Oda Rusticale (see Tuttle (1983)), which dates from 1688, about one century after the Calmo and Ruzante texts examined in SECTION 2. During this period subject clitics have developed one stage further, from phonological clitics to syntactic clitics. They have become clitic heads like their modern counterparts and not subject DPs like RVe subject clitics. The tests that reveal this changes are those used in SECTION 2 for RVe: the order with respect to the preverbal negative marker and coordination of two VPs when the subject pronoun is deleted. In the Veneto variety of the 17th

THE DIACHRONIC DEVELOPMENT OF SUBJECT CLITICS

313

century (henceforth SVe) some subject clitics appear at the right of the preverbal negative marker: (38) Perche no la pole (Oda p. 441) SVe Because not she can (39) No i te fa male (Oda p. 443) Not they to-you do harm In a sample of 145 sentences there are no cases of coordinated structures, so the second test cannot apply. We are thus compelled to base our analysis only on the fact that subject clitics appear at the right of the preverbal negative marker, and for this reason they are to be considered heads at S-structure. As discussed in SECTION 1, the fact that a subject clitic appears after the negative marker shows that subject clitics and subject DPs do not occupy the same position in the syntax: subject DPs in fact can only appear at the left and never at the right of the negative marker. Hence, we can conclude that SVe subject clitics are analogous to their modern counterparts. So, it seems that subject clitics have been reanalyzed, during the period between the 16th and the 17th century, as part of the inflectional head of AgrP. As already discussed in SECTION 1, this does not mean that subject clitics at this point of their evolution are not arguments in the sense that they do not absorb the subject theta-role. Even if they are heads, they can start out from a thematic position inside the VP and adjoin to the head of Agr blocking the insertion of another subject, because the thematic position is filled by the trace of the subject clitic. Object clitics in Romance have normally the distribution of argumental heads: when the object clitics are inserted, no object DP can be phonetically realized and no variable can occupy the object position.11 If we apply the tests already discussed in SECTION 1, we are compelled to admit that subject clitics of this period are bound to an argumental position. A subject clitic is not required when a subject DP is present, as in (40): (40) I toroere vale pi che no valse qui de Hisperite (Oda p. 442) SVe Your oak woods are more precious that not those of Hesperide. Subject QPs always appear without a subject clitic and there is no subject clitic when the subject is marked +wh and moved outside the sentence: (41) Agno pomaro fea pumi indore (Oda p. 441) SVe Every apple tree made golden apples (42) Agnun che bita dentro i tredese comun (Oda p. 443) Everyone that lives in the thirteen villages (43) Chi po far retirare el mare si ingordo? (Oda p. 443) Who can let retreat the sea (which is) so greedy. (40), (41), (42) and (43) show that subject clitics in SVe are parallel to object clitics: they absorb the subject theta-role and are incompatible with other subjects in argumental position.

314

CLITICS AND VERB SECOND

Therefore, the structure of a sentence with a subject clitic will be (1) (here repeated as (44)):12

THE DIACHRONIC DEVELOPMENT OF SUBJECT CLITICS

315

In (44) the subject clitic starts in the VP internal subject position as the trace t! under DP1 indicates, and moves up to Agr. It ends up in an adjoined position to the head of AgrP where the inflected verb is placed after having incorporated the affixes of Tense and Agreement. This kind of adjoined position is the same as that postulated for Modern NIDs (cf. (1)). On the other hand, (45) describes the situation that we found in RVe, in which subject clitics are still equivalent to maximal projections in the syntax and are clitics only at PR The difference between (44) and (45) can be interpreted as a modification of the subject clitic, which changes its categorical status. It is no longer analyzed as an XP that does not branch, as it does not have a Specifier and a Complement position, but as a simple head. As the structure preservation principle states that all XPs must move to an XP position and all X must move to head positions (cf. Chomsky (1986)), the subject clitic can no longer move to the SpecAgr position, which is an XP position, it can only move up to the head of this projection from the basic subject position inside the VP. Hence, the reanalysis of subject clitics as heads implies that they move to a head position. As (44) illustrates, subject clitics move to the head of AgrP. We can imagine different motivations that induce subject clitics to move just into this head: first of all no head containing a trace can host the subject clitic. T and V are both occupied by the trace of the verb which has moved to Agr. If the clitic adjoined to T or to V, it would induce a minimality effect between the trace and the inflected verb in Agr, yielding a structure like (46):

A configuration like (46) is excluded by Baker (1988). In fact the subject clitic would be a closer potential governor for the trace in T and it would prevent the correct relation between the verb in Agr and its trace in T. Hence the subject clitic must adjoin to a head which is not filled by a trace, but by a phonetically realized element, and only Agr is such a head. Second, the movement of the subject clitic to a left adjoined position to Agr recreates the same configuration at the X level that subject DPs have with Agr at the XP level. Adjunction of the subject clitic is structurally similar to a relation of Spec-head Agreement, but at a lower level.13 Third, if we consider Kayne's (1989) proposal that all syntactic clitics move to the head of AgrP in Romance, then also subject clitics, being syntactic clitics, will be attracted by this head. At this point we can ask if the reanalysis of subject clitics from purely phonological clitics as in RVe to syntactic clitics in SVe exerts some influence on other fields of the grammar. In particular we expect that the processes connected with the AgrP projection are influenced by this readjustment of the structure of Agr. Let us for instance, take into consideration the pro-drop system. We saw that the pro-drop system of RVe is fairly complex. Two heads

316

CLITICS AND VERB SECOND

are marked as possible pro licensers, namely C and Agr. But only in the case where they are filled by a particular feature are they visible for the pro-drop licensing condition. Looking at the data, it may seem strange to postulate a pro-drop system for SVe, because in this dialect, there seem to be no cases of null subjects at all. In fact, a subject clitic or a subject DP is always phonetically realized. (47) a. Quand'a me tacco a cantare (Oda p. 440) SVe When I me begin to sing b. Te si ti solo You are YOU alone

"

"

c. La mormolla de ti She murmurs of you

"

"

d. A sagion darme... We know to give

"

"

e. O golusi slecaizzi ch'a si Oh, greedy that you are

"

"

f. Quel ch'j dise What that they say

"

"

At a superficial glance, it seems that pro-drop has completely disappeared from the language. In fact, there is a subject clitic which is obligatory for all the persons of the verb, a phonetically realized subject DP, or a variable in the case of wh-movement of the subject. One may assume that the pro-drop character of RVe has been completely lost during this century and that SVe is a non-pro-drop language. Things do not appear to be so simple if we consider that subject clitics are no longer true subjects in SVe, but heads, as indicated by the tests in (38) and (39) and by structure (44). At this point three questions arise: (a)

If subject clitics are heads, what kind of element fills the SpecAgr position?

(b) Why are subject clitics obligatory, when there is no other phonetically realized subject DP? (c)

Why have pro-drop phenomena disappeared?

On the basis of the Extended Projection Principle, we must assume that SpecAgr is filled by some element, because the preverbal subject position cannot be left totally empty in any language. Hence, a null category must fill it: this category cannot be a variable, because it is not bound by any operator, it cannot be an DP-trace or a pro, because it is a Case-marked position. The only category that can occupy the SpecAgr position is a pro. This element, like all null categories, has to be licensed by a head which in RVe was Agr or C. We have seen that in SVe subject clitics are obligatory when there is no subject DP, but that they do not co-occur with subject DPs. If subject clitics

THE DIACHRONIC DEVELOPMENT OF SUBJECT CLITICS

317

appear only when a pro, and no subject DP occupies the SpecAgr position, we can make the hypotheses that the head that licenses the null subject is neither Agr nor C, but the subject clitic. The pro-drop conditions of SVe are expressed in (48): (48) pro is licensed by a clitic head in Agr through Spec-head agreement. The null subject is co-indexed with the subject clitic which licenses it through Spec-head Agreement. We can thus answer question (b): subject clitics are always obligatory when there is no phonetically realized subject DP because they license pro. If the subject clitic is omitted there is no head that can license pro and the sentence is ungrammatical. At this point the answer to the third question is quite simple. Pro-drop phenomena have not disappeared from the language at all. On the contrary, they are more widespread in the language than before. The change regards only the type of head that licenses the null subject. This head is neither C nor Agr as it was in RVe, but the subject clitic adjoined to Agr. The obligatory presence of a subject clitic simulates the requirement of a non-pro-drop language, in which a subject pronoun must always be present. The subject pronoun of SVe is nevertheless not a true subject DP, but a syntactic clitic in Agr. Sve is thus a pro-drop language like Standard Italian is, but it differs from Standard Italian because the head that licenses pro is not Agr itself, but a subject clitic adjoined to Agr. The structural configuration is the same in the two languages, namely Spec-head Agreement, but the head that licenses the null element is different. SVe has lost both strategies of pro-licensing that we found in RVe: neither C, nor Agr are possible pro-drop licensers. It has developed in the direction of a simpler system, in which only one head can license pro and only through a unique structural configuration. We see that the evolution of French and the Veneto variety are parallel. Modern French has also completely lost the possibility of pro-drop licensing through Spec-head agreement with the head Agr: no first or second plural person null subjects are admitted in Modern French as is the case in RFr. French has maintained pro-drop licensing from C through government, when C is marked +wh or +subjunctive, even if only for expletive subjects (cf. Kayne and Pollock (1978)): (49) a. Quand pro viendra Jean? When will come John?

French

b. J'aimerais que pro sorte Paul I wish that goes out Paul SVe has lost pro-drop licensing from both C and Agr, but it has developed a new system, in which another head has this function. Both SVe and French have developed into systems in which Agr is not a possible pro-drop licenser. This is the reason why they both have maintained subject clitics. As has often been noted in the literature, the languages that have developed subject clitics are precisely those that, in their medieval stage,

318

CLITICS AND VERB SECOND

could only license a pro through government by Agr, which had moved to C in accordance with the verb-second constraint. Agr was not able to license a pro through the configuration of Spec-head agreement. The similarity between French and SVe is to be found in the fact that in both languages Agr was not able to take up the function of pro licenser through Spec-head Agreement as was the case in Southern Italian Dialects and other Romance languages like Spanish. This weakness of Agr (which we assume to be syntactic and not only morphological) has brought about the development of an alternative system in SVe: a subject clitic licenses pro because Agr is not strong enough to do it in the relevant configuration of Spec-head Agreement. This system is still adopted by some conservative varieties in isolated areas. This fact is very important because it permits us to study the licensing conditions of a dead language such as SVe more deeply and to check our predictions by constructing ungrammatical sentences. One such variety is Rovignese spoken in Slovenia in the town of Rovigno. The subject clitic series of Rovignese is complete for all persons (cf. TekavCk; (1986)): (50)

1. i

2. ti

3. el/la

l.plur. i

2.plur i

3.plur i/le

When a subject DP is not realized, a subject clitic is obligatory: (51) a. Sa *(ti) me dive la paca If you to-me give a hit

Rovignese

b. * (A) ta par It to-you seems

The subject clitic is not obligatory when there is a phonetically realized subject DP: (52) a. Se Paron Giacomo gira furbo... If Mr. Giacomo was clever

Rovignese

b. La Francia gaviva tuchisto tira veia suldadi de 1'Istria The France had had to take away soldiers from Istria c. La feila spativa The girl waited Indeed, subject clitics and subject DPs in SpecAgr must be incompatible. In fact if we substitute the definite subject DP with a QP, which cannot be left dislocated and can only occupy the SpecAgr position, the subject clitic cannot appear: (53) a. Qualunque pol meti la man sul fogo Rovignese Everyone can put the hand on the fire b. * Qualunque el pol meti la man sul fogo Everyone he can put the hand on the fire

THE DIACHRONIC DEVELOPMENT OF SUBJECT CLITICS

319

This case is analogous to SVe: in SVe a subject QP always appears without a subject clitic, but we do not know if the structure QP+subject clitic is excluded or simply is not realized in the corpus of data that we take into consideration. If Rovignese has the same system that we outlined for SVe, we can check if subject QPs are really incompatible with subject clitics. (53b) shows that this is correct. The type of system displayed by Rovignese seems to be quite common in the Southern part of Veneto, where subject clitics are obligatory for all persons and only possible when no subject DP is realized. In SVe, as in Rovignese, subject clitics have developed into syntactic heads specialized for the licensing of a null subject, taking up the role that Agr and C had in RVe.

4. Conclusion The status and the distribution of subject clitics in Northern Italian Dialects is connected to, at least, three components of the grammar: the pro-drop parameter, the Case-assignment conditions, and the visibility of empty Agreement heads. It is possible to summarize the entire discussion about the development of subject clitics making a quite simple hypothesis regarding the relation between verbal morphology and subject clitics. In Standard French, subject clitics have remained true subjects, which appear in the SpecAgr position like other subject DPs. On the contrary, in all Northern Italian Dialects, subject clitics have been reinterpreted as a possible candidate for substituting agreement in various syntactic mechanisms. In all these cases the function of agreement, both intended as a syntactic position and as morphological specification, is to identify the subject of a predicate. Subject clitics, starting as true subjects, have slowly been reanalyzed as heads that interfere in the strict relation between the head and SpecAgr. In SVe, for instance, subject clitics mimic the relation of Spec-head Agreement that Agr has with the subject adjoining to the head and licensing a pro in the SpecAgr position. The subject clitic thus constitutes a new type of Agreement morphology following Renzi and Vanelli's (1983) generalization that the subject features must be encoded at least on one of the two elements, but can also be encoded on both.

Notes Thanks are due to A. Battye, A. Belletti, G. Cinque, T. Guasti, I. Roberts, L. Rizzi, A. Tomaselli, R. Zanuttini, and in particular P. Beninca and L. Vanelli for comments and discussion. All errors are naturally my own. 1

The term "agreement" is ambiguous, because it indicates both the inflectional morpheme and its structural position as head of AgrP. I will refer to the syntactic position of Agreement using the capital letter and to the morpheme as agreement in small letters.

320

CLITICS AND VERB SECOND

2. Some varieties allow both the order negation-subject clitic and the order subject clitic-negation: (i)

Tu un vieni Fiorentino You not come

(ii)

Un tu vieni

Note however, that the order in (ii) is never admitted when the subject is a full DP. Moreover, there are reasons to believe that in (i) the subject clitic is a head adjoined to the negative marker (see Poletto (1993a)). 3. Sentences with compound tenses have a different series of subject clitics from sentences with simple tenses. 4. The Veneto variety used for the examples is the dialect of Oderzo. 5. We use here the second person singular subject clitic, which behaves like the / clitic. 6. We will use examples from plays by Ruzante for the Paduan variety and from a letter collection by Calmo for the Venetian variety. There are only some minor morphological distinctions between the two. In (16) the first form is the Paduan, the second corresponds to Venetian. 7. Ca seems to be a specialized form for the comparative complementizer which is found only in Venetian texts. 8. Subject clitics are still arguments also in Modern Veneto. They have lost the status of XPs, and are heads just like object clitics, but they never co-occur with subject QPs or subject variables either in preverbal or in postverbal subject position. 9. The pro-drop system of Renaissance French and Veneto is different from the Medieval system. In their Medieval stage, these languages were verb-second. Pro-drop was licensed by the verb in C, hence possible only in matrix verbsecond clauses. In the Renaissance period, French and NIDs lost verb second, but the licensing of pro still comes from the C head. As the verb does not move any more into C, this must be marked with a particular feature in order to be visible. Agr can only take on the function of pro-licenser if it is morphologically strong. 10. The theory that we propose here cannot be applied to Modem NIDs as it is formulated. NID subject clitics are in fact heads, and it is not obvious that they need to be independently Case-marked. We will not discuss the phenomenon of Quirky Agreement (cf. Battye (1990)) in Modern NIDs here. 11. We are not considering here cases of clitic doubling, which are quite frequent in NIDs, but only with indirect object clitics. 12. From the diachronic point of view, it seems quite reasonable to admit that the change in the structure must happen by means of ambiguous strings of words (cf. Lightfoot (1979) and Roberts (1992)) that give rise to a possibility of "misunderstanding" the structure of the sentence. This is surely not the only reason for diachronic change, because there must be some parametric choices that "push" a language in a given direction. In any case, the structures presented in (44) and (45) present just the case of ambiguity that seems to be

THE DIACHRONIC DEVELOPMENT OF SUBJECT CLITICS

321

implied in the reanalysis of a structure. For instance a sentence like (i) can be interpreted as having the structure (44) or 5): (i)

El vien He comes

This ambiguity must have been the "bridge" which permitted the reanalysis from (45) to (44). 13. It is interesting to note that there seems to exist a relation of mutual exclusion between Nominative Case-assignment through government from Agr and Nominative Case-assignment through government from T. English, for instance, is a language that does not permit free inversion of the subject. Hence, following the parameter in (34) it does not select T as a possible Nominative Case-assigner. Nevertheless, in main interrogative sentences, Agr can assign Nominative to the subject DP in SpecAgr, as in (i): (i) What has John done? Romance languages, on the other hand, select T as possible Case-assigner, but do not permit Nominative Case assigned by Agr in a sentence like (i): (ii)

*Qui a Jean vu? Who has John seen?

So, we can observe that a language can exploit a Nominative Case-assignment configuration only once: if the subject gets Nominative from T, it cannot get it from Agr under the same type of configuration. This could be valid not only for government, but also for Spec-head agreement.

References Baker, M. (1988) Incorporation: A Theory of Grammatical Function Changing. The University of Chicago Press, Battye, A. (1990) "Quirky Agreement in Genovese." Ms. University of York. Belletti, A. (1988) "Unaccusatives as Case Assigners." Linguistic Inquiry 19:134. Belletti, A. (1990) Generalized Verb Movement: Aspects of Verb Syntax. Torino: Rosenberg and Sellier. Beninca, P. (1983a) "II clitico a nel dialetto padovano." Scritti linguistici in onore di Giovan Battista Pellegrini, 25-35. Pisa: Pacini. Beninca, P. (1983b) "Osservazioni sulla sintassi dei testi di Lio Mazor." In C. Angelet et al., eds. Langue, Dialecte, Literature. Etudes romanes a la memoire de Hugo Plomteux, 187-197. Leuven. Beninca, P. (1985a) "II mutamento sintattico: di un aspetto della sintassi ladina considerate di origine tedesca." Quaderni Patavini di Linguistica 5:3-15. Beninca, P. (1985b) "Uso dell'ausiliare e accordo verbale nei dialetti veneti e friulani." Rivista Italiana di Dialettologia 8:178-194.

322

CLITICS AND VERB SECOND

Beninca, P. (1986) "Punti di sintassi comparata dei dialetti italiani settentrionali." In G. Holkus, K. Ringger, eds. Raetia Antiqua et Moderna, 457-479. Tubingen. Beninca, P. (1988) "L'ordine degli elementi della frase e le costruzioni marcate." In L. Renzi, ed. Grande Grammatica di Consultazione, 115225. Bologna: II Mulino. Beninca, P. (1989) "Friaulisch." Lexicon der Romanistischen Linguistik, Vol 3, 563-585. Tubingen: Niemeyer Verlag, Beninca, P. (this volume) "Complement Clitics in Medieval Romance: the Tobler-Mussafia Law." Beninca, P. and G. Cinque (1993) "Su alcune differenze tra enclisi e proclisi." Omaggio a Gianfranco Folena, Padova Editoriale Programma. Beninca, P. and C. Poletto (1992) "II modello generative e la dialettologia: un'indagine sintattica." Rivista Italiana di Dialettologia 15:77-97. Beninca, P., L. Renzi and L. Vanelli (1985/6) "Tipologia dei pronomi soggetto nelle lingue romanze." Quademi Patavini di Linguistic 6:49-66. Beninca, P. and L. Vanelli (1982) "Aspetti sintattici del portogruarese tra veneto e friulano." In R. Sandron, ed. L'area portogruarese tra veneto e friulano, 39-52. Portogruaro. Beninca, P. and L. Vanelli (1984a) "Italiano, veneto, friulano: fenomeni sintattici a confronto." Rivista Italiana di Dialettologia 8:165-194. Beninca, P. and L. Vanelli (1984b) "Appunti di sintassi veneta." In M. Cortelazzo, ed. Guida ai dialetti veneti 4:7-38. Padova. Benucci, F. (1991) " Le particelle infinitivali come specificatori di CP." Ms. University of Padova. Bracco, C., L. Brandi and P. Cordin (1981) "Sulla posizione soggetto in italiano e in alcuni dialetti." In A. Franchi De Bellis and L.M. Savoia, eds. Sintassi e morfologia della lingua italiana d'uso. Teorie e applicazioni descrittive, 185-209. Roma. Brandi, L. and P. Cordin (1981) "Dialetti e italiano: un confronto sul parametro del soggetto nullo." Rivista di Grammatica Generativa 6:33-87. Brandi, L. and P. Cordin (1989) "Two Italian Dialects and the Null Subject Parameter." In O. Jaeggli and K. Safir, eds. The Null Subject Parameter, 111-142. Burzio, L. (1986) Italian Syntax: A GB Approach. Dordrecht: Reidel. Calabrese, A. (1980) "Sui pronomi atoni e tonici dell'italiano." Rivista di Grammatica Generativa 5:65-116. Calabrese, A. (1982) "Alcune ipotesi sulla struttura informazionale della frase in italiano e sul suo rapporto con la struttura fonologica." Rivista di Grammatical Generativa 7:3-78. Calmo, A. (1888) Le lettere. V. Rossi, ed. Torino: Loescher. Calmo, A. (1978) La Spagnolas. L. Lazzarini, ed. Milano. Cardinaletti, A. and I. Roberts (1992) "Levels of Representation of Agreement." In W. Chao and G. Horrocks, eds. Proceedings of the 1990 GLOW Colloquium. Cambridge: Cambridge University Press.

THE DIACHRONIC DEVELOPMENT OF SUBJECT CLITICS

323

Chomsky, N. (1981) Lectures on Government and Binding. Dordrecht: Foris. Chomsky, N. (1986a) Barriers. Cambridge, Mass.: The MIT Press, Chomsky, N. (1986b) Knowledge of Language. New York: Praeger. Chomsky, N. (1989) "Some Notes on the Economy of Derivation and Representations." In R. Friedin, ed. Principles and Parameters in Comparative Grammar, 417-454. Cambridge, Mass.: MIT Press. Chomsky, N. (1992) "A Minimalist Program for Linguistic Theory." MIT Occasional Papers in Linguistics, 1. Cinque. G. (1990) Types of A'-Dependencies. Cambridge, Mass.: The MIT Press. Giorgi, A. and G. Longobardi (1990) The Syntax of Noun Phrases: Configuration, Parameters, and Empty Categories. Cambridge: Cambridge University Press. Giorgi A. and A. Pianesi (1991) "Toward a Syntax of Temporal Representation." Probus 3:1-27. Giupponi, E. (1988) "Pro-drop Parameter und Restrukturierung im Trentino." MA Thesis, University of Vienna. Guasti, M.T. and C. Poletto (1991) "Subject Positions and the wh Criterion." Ms. University of Venice, University of Geneva. Kayne, R. (1975) French Syntax: The Transformational Cycle. Cambridge, Mass.: MIT Press. Kayne, R. (1984) Connectedness and Binary Branching. Dordrecht: Foris, Kayne, R. (1989) "Facets of Past Participle Agreement." In P. Beninca, ed. Dialect Variation and the Theory of Grammar, 85-103. Dordrecht. Kayne, R. (1991) "Romance Clitics, Verb Movement, and PRO." Linguistic Inquiry 22:647-686. Kayne, R. (1992) "Italian Negative Infinitival Imperatives and Clitic Climbing." In L. Tasmowski and A. Zribi Hertz, eds. De la musique a la linguistique. Hommage a Nicolas Ruwet. Communications and Cognition, Ghent. Kayne, R. and J.-Y. Pollock (1978) "Stylistic Inversion, Successive Cyclicity, and Move NP in French." Linguistic Inquiry 9:595-621. Koopman, H. and D. Sportiche (1991) "The Position of Subjects." Lingua 85:211-258. Lightfoot, D. (1979) Principles of Diachronic Syntax. Cambridge/New York: Cambridge University Press. Poletto, C. (1991) "Three Kinds of Subject Clitics and the Theory of Pro" EUROTYP Working Papers 8.1. Poletto, C. (1993a) "Subject Clitic/Verb Inversion in North Eastern Italian Dialects." In A. Belletti, ed. Syntactic Theory and the Dialects of Italy. Torino: Rosemberg and Sellier. Poletto, C. (1993b) "The Aspect Projection: an Analysis of the passe surcompose." In E. Fava, ed. Proceedings of the XVII Incontro di Grammatica Generativa, Torino: Rosenberg and Sellier.. Pollock, J.-Y. (1986) "Sur la syntaxe de en et le parametre du sujet nul." In M. Ronat and D. Couquaux, eds. La Grammaire modulaire, 211-246. Paris: Editions de Minuit.

324

CLITICS AND VERB SECOND

Pollock, J.-Y. (1989) "Verb Movement, UG, and the Structure of IP." Linguistic Inquiry 20:365-424. Renzi, L. (1992) "I pronomi soggetto di due varieta substandard: fiorentino e francais avanc6." Zeitschrift fur romanische Philologie 108:72-98. Renzi L. and L. Vanelli (1983) "I pronomi soggetto in alcune varieta romanze." Scritti linguistici in onore di G.B. Pellegrini, 121-145. Pisa: Pacini. Rizzi, L. (1982) Issues in Italian Syntax. Foris: Dordrecht. Rizzi, L. (1986a) "On the Status of Subject Clitics in Romance." In O. Jaeggli and C. Silva-Corvalan, eds. Studies in Romance Linguistics, 391-419. Dordrecht: Foris. Rizzi, L. (1986b) "Null Objects in Italian and the Theory of Pro." Linguistic Inquiry 17:501-557. Rizzi, L. (1990) Relativized Minimality. Cambridge, Mass.: MIT Press, Rizzi, L. (1991) "Residual Verb Second and the Wh Criterion." Technical Reports in Formal and Computational Linguistics, 2. University of Geneva. Rizzi, L. and I. Roberts (1989) "Complex Inversion in French." Probus 1:130. Roberts, I. (1992) Verbs and Diachronic Syntax. Dordrecht: Kluwer. Tekavcic, P. (1986) Grammatica storica dell'Italiano. Bologna: II Mulino. Tomaselli, A. (1989) La sintassi del verbofinito nelle lingue germaniche. PhD Dissertation, University of Pavia. Tuttle E. (1983) "L'Oda rusticale di Nicol6 Zotti." Scritti Linguistici in Onore di G.B. Pellegrini, 431-464. Pisa: Pacini. Vanelli, L. (1987) "I pronomi soggetto nei dialetti italiani settentrionali dal Medio Evo ad oggi." Medioevo Romanio XIII173-211. Vikner, S. (1990) Verb Movement and the Licensing of NP-Positions in the Germanic Languages. PhD Dissertation, University of Geneva. Zorzi, L. (1967) Ruzante: Teatro. Torino: Einaudi.

12 Complement Clitics in Medieval Romance: the Tobler-Mussafia Law* Paola Beninca State University of Milan

The aim of this paper is very restricted: I want to provide a generalization defining the precise context where complement clitics in Medieval Romance change their position relative to the inflected verb, becoming enclitics. The theoretical discussion will be limited to the interpretation of the structural contexts triggering the phenomenon, leaving aside for the moment any attempt to describe the process through which a proclitic pronoun becomes enclitic.1 To begin with, I shall describe the position of complement clitics relative to the inflected verb in Old French and Northern Italian dialects in the Middle Ages. The constraints that are observed appear to be closely related to the structure of the left portion of the sentence, i.e. with the Top and SpecC positions. Subsequently, I will show that the same interpretation is valid for other Medieval Romance languages, which differently from Old French and Northern Italian, are completely pro-drop. I will propose a reformulation of a "law" identified at the end of the 19th century by Adolf Tobler and Adolfo Mussafia for the Romance languages.2 The Tobler-Mussafia generalization can be informally summed up by the following formula (notice that only inflected verbs are taken into consideration):3 (1) a. * # clitic-Verb b. # (X) Verb-clitic (la) says that a clitic before a verb is ungrammatical in clause-initial position; that is to. say, if nothing appears before the verb in a clause, clitics must follow the verb. (Ib) says that a clitic following a verb is grammatical both in sentence-initial and in sentence-internal position, i.e. the sequence verb-clitic is grammatical in any context, and obligatory if the verb opens the sentence. (Ib) reflects Mussafia's formulation, based upon data from Old Italian 325

326

CLITICS AND VERB SECOND

varieties; he observed (Mussafia 1886) that enclisis was obligatory in sentence-initial position, but was apparently possible everywhere. The purpose of the description I am proposing is to specify the syntactic contexts where a clitic cannot precede the verb in main clauses, and the contexts where it cannot follow the verb, both in sentence-initial and sentenceinternal position, in main clauses and in dependent clauses. The evidence will indirectly bear against various proposals trying to trace back the constraint to prosodic or rhythmic conditions,4 or to conditions regarding the informational structure of the sentence and relations between given and new information.5 The position of complement clitics relative to the verb depends instead, in my view, on structural conditions. I will argue that the Medieval Romance Languages (MRLs) we are considering are verb-second languages, and that the relevant variation observed is the effect of a parametric difference, namely of the activation of a Top position above CP: languages that have no access to this position appear to be of a rigid verb-second type. The structure of the left periphery of the sentence I will deal with is a follows:

As is clear from the figure, I will not take into account possible splittings of the Infl node, and, as far as CP is concerned, I will only consider Top as a functional projection higher than CP. It appears from recent work on Romance languages that both IP and CP are presumably a cover term for a series of functional projections. For the time being, it does not seem possible to me to explore and test these assumptions on MRLs. Nor will I be concerned with the structure under IP, taking for granted that the subject and the main verb, at a certain point of the derivation, happen to be in the SpecI and I° position respectively; the main verb then moves to the head of CP in main clauses, and a constituent moves to the SpecC, producing the typical configuration of verb-second languages. Top is here considered an unanalyzed projection, the

COMPLEMENT CLITICS IN MEDIEVAL ROMANCE

327

position in which left dislocated and topicalized constituents are found in the Modern Romance languages. Top is probably a cover term as well, concealing more functional projections; it has specialized functions, in particular, it is not able (for principled reasons?) to have a constituent in its Spec which enters a Quantifier-variable relation with an argument position in the sentence. I will not pursue this subject further, and I will call this upper projection Top; I will assume also that, as in Modern Romance languages, it can be indefinitely iterated. Using unanalyzed IP I also mean to avoid more precise hypotheses regarding the exact way the verb and its inflection join each other and incorporate complement clitics: if a complex head results, in main clauses the whole complex head moves to C.6 The aim of the present paper is to show that this happens if and only if the Spec of C is filled. If the sequence clitic-verb is to be seen as an incorporation, in order to get the order verb-clitic we are led to admit the possibility of excorporating the verb, whatever the process of encliticization may be. In the present paper I will keep these descriptive and interpretive problems—which still have a controversial status—out of the scope of the phenomenon considered. I shall compare three types of MRLs, exemplified here by: a) French, limited to the 12th and 13th centuries; b) Northern Italian dialects (NIDs) of the 12th to the early 14th centuries, represented here by three varieties sharing the relevant features; c) Old Portuguese and Southern Italian, as instances of pro-drop MRLs. Old Portuguese will allow us to take into consideration a Modern Romance language, namely Modern Portuguese, which shares with its old variant the essential syntactic features with regard to the phenomena we are considering. This will give us the possibility to elicit and test the relevant cases in a living language.

2. Complement Clitics in Old French It is generally accepted that Old French has to be analyzed as a verb-second language; especially in the time period I am taking into consideration, Old French can be considered a strict verb-second language; this means that it appears to be subject to the complete set of rules and constraints that cause the verb to surface in second position in main clauses: 1. 2. 3.

the verb moves into the position of the head of CP; a constituent moves into the position of SpecC; no other constituent moves to the front of the sentence, or, in the descriptive perspective we are adopting, no Top position is available for any constituent to move into.

As illustrated in the following sentences, different types of constituents can occupy the SpecC position, apart from the subject:

328

CLITICS AND VERB SECOND

(3)

a. Nouveles vos sei dire del tornoiement de Karahe's (Artu 87, 11) 'News you-(I) can tell of the joust of Karahes' b. A ceste paroles respont la reine (59, 84) 'to these words responded the queen' c. Un peu apres eure de prime fu Mador venuz a cort (81, 6) 'Soon after the first hour was Mador come at court' d. Or voiz tu bien (72, 7) 'Now see you well' e. Si en est li rois moult a malese (79, 3) 'So of-it is the king very troubled' f. Non est ce la premiere foiz que vos 1'avez quis (Artu 24, 11) 'Not is this the first time that you him-have looked for' (= that you have looked for him).

From this comparison some superficial features emerge which may be considered as diagnostic of the underlying structure. The sentences in (4) show the so-called "asymmetry of pro-drop," i.e., the inflected verb in main clauses can have a pro as subject in SpecI position, while in dependent clauses the verb must have a phonetically realized subject (a pronoun or a Noun Phrase). This property has been connected to the fact that in main clauses the verb is normally in IP (see Beninca (1983-84); Vanelli, Renzi and Beninca (1985); Adams (1987)). (4c) is an example of the same fact with a verb having a nonargumental subject: (4) a. Si errerent-tant en tele maniere qu'il vindrent en la praerie de Wincestre (Artu 16, 66) 'so wandered (3.pl)- so much in such a way that they came in the prairie of Winchester' b. Atant en lessent -la parole ester et chevauchent a petites jornees tant que il vindrent a Kamaalot (Artu 31, 1) 'Then make -the speech stop and ride by easy stages so that they came to Camelot' c. Cel jour fist -moult lait tans, car il plut et espart (Berthe au grand pied, 76-77) That day was -very nasty weather, because it rained and lighteninged'. A second important feature is shown in (5): a direct object in first position is never accompanied by a resumptive pronoun. (5)

a. Itiex paroles disoit la reine a soi meisme (Artu 32, 30) 'Such words said the queen to herself

b. Mes Lancelot ne connut il mie, car trop estoit enbrons (Artu 11, 13) 'But Lancelot (obj.) not recognized he (subj.), because too much was sullen'

COMPLEMENT CLITICS IN MEDIEVAL ROMANCE

329

c. La traison li a contee que li vasals a aprestee (Eneas, 23-24) The treason him-has told, that the vassal has prepared'. This is an important difference with respect to Modern French (as well as Modern Romance languages in general, with two interesting exceptions, one of which, namely Portuguese, I will turn to later): in Modern French the direct object in first position always has a resumptive pronoun. In other Modern Romance varieties such as Italian or dialects of Italy (apparently, not in French), a proposed object can lack a resumptive pronoun: in this case, the intonation gives relevance to the preposed constituent, which is followed by an intonative break: it has a contrastive interpretation, related to its operatorlike status. Observing the diachronic development of ld French, it appears that this is true as long as the preposed direct objct was out of necessity immediately before the verb; that is, as long as Old French was strictly verb second. In fact, to characterize Old French as a strict verb-second language needs some qualifications: if compared with NIDs, or Old Spanish and Old Portuguese, it appears that, as I said before, French has not as free an access to an upper TopP projection as other MRLs, that is, it has not the same freedom to have sentences with a verb in third or fourth position, or, on the other hand, in first position. But compared to Modern German we would incline to say that Old French is "less strict" with regard to this requirement. So, we have a few cases where the verb is in third position. When more than one constituent is found before the inflected verb in main clauses, an interesting correlation is observed between the lack of a resumptive pronoun and the adjacency of the direct object and the verb. This can be easily observed in Northern Italian and Portuguese Medieval texts, as we will see; for French this is supported by systematical observations by, for example, Zwanenburg (1978). The pattern observed in Old French is, as in the following example, with the direct object always adjacent to the inflected verb, without a resumptive pronoun. (6)

a. Erec le roi en mercia (Erec 6499) 'Erec the king (obj) for-it-thanked' b. li roi lui et Enide an mainne (Erec 4207) 'The king him and Enide from-there-takes'.

Only in later texts can a preposed direct object be separated from the verb by another constituent, and it has in this case a resumptive pronoun, in French as in other Romance varieties. I assume, then, that the direct object was obligatorily preposed without a resumptive pronoun if the direct object landed in SpecC position. I will return to this point later when considering Italian Dialects and Portuguese, which will provide evidence in favour of this assumption. At this point, we can briefly outline the Tobler-Mussafia (T-M) Law in French. In the period considered, French is, as I said before, strictly verbsecond, in the sense that the inflected verb is actually, in most cases, preceded by one and only one constituent. Nevertheless, cases of verb first (more fre-

330

CLITICS AND VERB SECOND

quent in the older texts), though rare, are possible, mainly for reasons that I will not investigate here, with non-argumental subjects. The facts are roughly as follows. When a verb in sentence-initial position is accompanied by a clitic pronoun, the latter cannot take the usual position before the verb: either the pronoun takes the "full" form (which means that it is no longer a clitic), as in (7a), or it must follow the verb, becoming enclitic, as in (7b). (7)

a. Moi semble que.... (Floire et Blanchifleur, 1538) 'Me seems that... (= it seems to me) b. Sire, ai le ge bien fait (Perceval, 1471) 'have-it I well done' (= I have done it well).

In what follows, I will try to provide a description that connects the enclisis with a precise syntactic condition. It could be argued that in the period considered the T-M generalization is no longer completely valid in French, because it seems to be contradicted by a number of facts. Occasions for enclisis are uncommon, as I said, because the verb is rarely in first position. Any constituent can be moved into first position without any particular requirements; otherwise, filler-words, such as si 'so' and or 'now', then appear in first position. Thus there are fewer occasions for the verb to be in first position, but this does not imply that the constraint against clitics appearing in first position is itself no longer valid. We can say that it begins to be "obscured" by being avoided with various devices. Two further cases apparently contradicting the T-M generalization deserve more careful consideration. There are, in fact, precise contexts in which clitics are found in first position: one case is in yes-no questions, where, at the time we are considering, clitics are allowed to appear before the inflected verb in first position, as in (8a,b). This was not the case before the 12th century, when only sentences like (8c) were possible; this fact is often interpreted in the literature as an indication that, whatever the reason for the prohibition against clitics in first position, this prohibition was getting weaker in the 12th century. (8)

a. S"est il donques corrouciez a nos? (Artu 66, 29) 'Himself-is then vexed with us' b. ...me fetes vos droit de doner a la reine si lonc respit? (68, 30) '(to) me-give you the right...' c. ...haez le vos donques si durement? (59,7) 'Hate-him you then so harshly?'

Another case is found in coordinate sentences: a coordinate sentence can begin with a clitic or a clitic cluster, flying in the face of Tobler and Mussafia and in fact of any theory considering clitics as being elements that, due to their very semantics or prosody, cannot begin a sentence.

COMPLEMENT CLITICS IN MEDIEVAL ROMANCE

(9)

331

a. Au soir dist Lancelos a la dame..., et la mercia moult (Artu 56,2) 'In the evening said Lancelot to the dame..., and her thanked very much' b.

Celui soir envia Boorz le chevalier...au roi de Norgales, et li manda que... (58, 1) 'That night sent Boort (subj.) the knight (obj.)...to the king of Norgales and him-ordered...'

c.

...ele se parti de la fenestre ou ele estoit et s'en entra en sa chambre (58, 13) 'she herself-departed from the window where she was and her self-entered her room'.

I would argue that the last two facts are not to be considered as evidence of a weakening of the constraint against clitics appearing in first position. I will suggest later that they depend on a different way of interpreting, structurally and syntactically, both types of sentence. The structure is such that the triggering of the condition which causes the enclisis of the pronoun can be avoided without weakening the condition itself. It is worthwhile at this point to discuss briefly the "prosodic hypotheses." The T-M Law has been interpreted as a purely prosodic constraint which prohibits a sentence from beginning with unstressed constituents. This kind of interpretation has been recurrently advanced throughout the long history of the studies on this topic, from Tobler and Mussafia themselves, through W. Meyer-Liibke, to Adams. It is difficult to think of a phonological (prosodic) constraint that is able to read syntactic labels in order to distinguish clitic pronouns from other kinds of clitic (unstressed) constituents. Moreover, the constraint would have to distinguish between different preverbal constituents: as I will show later, clitics have to be enclitics in certain circumstances even if they are preceded in a sentence by a word (or constituent) to which they could cliticize (phonologically), but they do not have to be enclitics if the word or constituent is not in the required syntactic position. If a phonological constraint were at work in the cases we are considering, it would only be able to displace the stress: processes of this kind are described and analyzed, for example, in Kenstowicz (1990). As far as I know, it has never been reported, nor can be expected on theoretical grounds, that phonological constraints feed syntactic movements. I will return briefly to this argument later.

3. Complement Clitics in Medieval Northern Italian Dialects NIDs, which constitute the second Romance type considered in this paper, share some properties with Old French and differ from it in some interesting respects. In essence, these dialects are of a "broader" verb-second type in the sense that they admit more than one constituent before the verb. Neverthe-

332

CLITICS AND VERB SECOND

less, they share with French other features which indicate that we are still dealing with a language with V movement to C and restrictions on the possible preverbal constituent. In this sense I continue to consider them to be verb-second languages, though the verb is not constrained to surface in second position. The behaviour of these languages, being more richly varied, gives us the possibility to check some predictions in a wider range of contexts. I shall give examples drawn from three distinct languages of Medieval Italy: Piedmontese, Venetian and Florentine. With respect to the facts considered, these languages are to be considered interchangeable, sharing as they do the relevant features. To begin with, the following sentences show the asymmetry between main and dependent clauses with regard to pro-drop:7 Piedmontese: Sermones Subalpini (ed. Babilas, 1968. 12th-13th century) (10) a. Done-li terme per tal convent que, si el al terme non aves paia quest aver, qu-el serea pendu (sermo VII) '(he) gave-him due-date...that, if he at the due-date would not pay..., that he will be hanged' b.

Torne-sen, se ane a 1'autre so ami, si lie ai coita so desasi (VII) '(he) came back, so (he) went to the other friend, so (he) told him his disease'

c.

Quar eu no savea que tu fuses tal hom com tu eres (II) 'Because I didn't know that you were such a man as you are'.

Venetian of the Lagoon (Lio Mazor, ed. Levi, 1904. early 14th century) (11) a. Et en questa lo Saracho dis "Pouse, cunpare!", et leva lo rem et mena-me 50 per lo brago si ch'e/ me lo scavega (3t, 48) 'at that point the Saracho said "Stop, my fellow!," and raised the sweep and stroke-me on the arm so that he broke it' b.

el dis ch'el me pagarave quando el vorave (2t, 3) 'he said that he would pay me when he wanted'.

Old Florentine (ed. Schiaffini (1954) 13th-14th centuries) (12) a. La formica € piu savia di te e ongni altro animale, inper6 che ella raguna la state ond'ella vive di verno (p. 24: 89 v, 5) 'The ant is wiser than you and any other animal because she gathers together in the summer what she lives on in the winter' b. E cosi ne prov6 de piu cari ch'elli avea (p. 74: 90 r, 30) 'And so tested some of the dearest that he had' The emphasized pronouns in the examples above would be ungrammatical in Modern Italian, or they would produce disjoint reference effects. In other Modern Northern Italian dialects, on the contrary, ungrammaticality would stem from the absence of subject pronouns.

COMPLEMENT CLITICS IN MEDIEVAL ROMANCE

333

Apart from the asymmetry of pro-drop, the examples above show that, unlike Old French, Italian varieties do not appear to be verb-second languages: in main clauses we also find the verb in first, third, fourth, etc. position. Nevertheless, I argue that the verb is in the head of CP in main clauses, and therefore that NIDs and Old French share the property of verb movement to C in main clauses. I interpret the difference between Old French and NIDs with respect to the possibility of verb third etc. by supposing that SpecC in NIDs is not the only position available for preposing a constituent. A preposed constituent can be in Top as well, and, moreover, as for Modern Romance languages, we assume a Top position which can be rewritten. This is an even more natural assumption, if we consider Top, as suggested above, as a recursion of Comp with specialized (reduced) properties. In Italian dialects, then, a constituent which appears on the surface before the verb, is not necessarily in SpecC. On the basis of NIDs we can try to fix a number of properties that distinguish constituents located in SpecC from other preposed constituents, supposedly in Top. Some items are lexically specified as "fillers" of SpecC: (co)si is such an item, as it is in Old French. In main clauses it is strictly preverbal and causes obligatory subject inversion: (13) et cosi lo mis-e co (Lio Mazor: 17, r 17) 'and so him put-I down'. The same is true of preverbal objects, which always lack a resumptive pronoun: (14) a. 90 dis-el plusor fiade (Lio Mazor: 1t, 61) 'that said-he many times' b.

una fertra fei lo reis Salomon. (...) Las colones fei d'argent e 1'apoail fei d'or; li degrai per unt hom i montava covrf de purpra (Serm. Sub., V) 'a sedan-chair made the king Salomon (subj.)... The columns made (3.sg.) of silver and the support made of gold; the steps by which one there-climbed, covered (3.sg.) with purple'.

Again, a direct object adjacent to the verb has no resumptive pronoun. Since the Italian varieties are freer than Old French with regard to the accessibility of structural positions in the left portion of the sentence, we can observe that the items that we suppose to occupy SpecC cannot appear together. We can have more than one preverbal constituent, but we never find, for example, si and a preverbal object with no resumptive pronoun together. We find instead a subject followed by si or by an object. Let us say that the subject in this case is left dislocated in Top. (15) L'autre ami si est la moillier (Serm., VII) 'The other friend so is the wife' More evidence in favour of this assumption will be given below.

334

CLITICS AND VERB SECOND

Let us summarize now the generalization that Mussafia drew from Medieval Italian texts with regard to clitic pronouns. In contrast with the French type, the Italian type, while prohibiting clitics from beginning a sentence, allows clitics to follow a verb in sentence-internal position. In Mussafia's view, the sequence "verb-clitic" is never ungrammatical, only the sequence "cliticverb" is, and precisely in sentence-initial position. I am going to argue that this is not true. For the moment we will have to be content with negative evidence and consider the absence of relevant cases as evidence of ungrammticality; comparison with a living language will support the assumption. Compared to French, Italian generally shows more frequent occurrences of the sequence "verb-clitic." A coordinate sentence is the typical context for verb-clitic in Italian, but not in French (I do not take into consideration Old Provengal, which behaves like Italian in this respect): sentences (16) contrasted with sentences (17) illustrate this difference. (16) a. e si la lave e forbf e retorne"-la en sen loc (Serm., X) 'and so her-washed (3.sg.) and wiped and put-her again in its place' b. et he li tras la fosina de man et branchai-lo per li caveli (4r, 36) 'and I him-tore the harpoon from his hand and caught-him by the hair' c.

Allora quelli il si mise in casa (...) e cominciollo a confortare (Schiaffini (1954:76)) 'Then he him-put in his house and began-him to console*

(17) a. Mes Nouvele (...) ala (...) a la Joieuse Garde et I'i porta uns valets (Artu 105, 15) 'But News went to the Joieuse Garde and it-there-brought a valet (subj.)' b. Je m' en irai arriere a mon signour et li conterai (Artu 110, 57) 'I will go after my lord and him-tell'. We could conclude from this feature that the constraint against first position is stronger in Italian than in French. But if we look at yes-no questions, the conclusion is the opposite: in this context, Italian varieties very consistently have the clitics in first position: (18) a. Me voj tu dar la taverna? (Lio Mazor: 8t, 28) 'Me-want you give the tavern? (= will you give me the tavern?) b. Se vastarave lo pes... ? (Lio Mazor: 20t, 46) 'Self-would taint the fish?' (= will the fish go bad?) I said that in French the clitic in first position in yes-no questions becomes more frequent in relatively later texts, and this is interpreted as evidence of the weakening of the constraint against clitics in first position. From this point of view, the Italian pattern, with clitics in first position in yes-no questions

COMPLEMENT CLITICS IN MEDIEVAL ROMANCE

335

and clitics after the verb in coordinate sentences, is inconsistent, the former indicating a strengthening, the latter a weakening of the constraint. It seems, then, necessary to interpret these facts as only indirectly having to do with the constraint itself; they depend instead, I would suggest, on the different structural descriptions assigned to these types of sentence. What is interpreted as "first position" (i.e., empty SpecC) in Old Italian, is not "seen" the same way in Old French syntax, but the constraint remains unaffected. Another difference, first noticed by Mussafia himself, is the fact that, as I mentioned before, in Italian varieties the enclisis of the clitic, while being obligatory in first position, is widely found in sentence-internal position as well. Some work has been devoted defining the type of contexts prohibiting, or else allowing, enclisis.8 As Schiaffini (1954) already pointed out in his comments on the collection of 13th-century Florentine texts he edited, we never have enclisis in a dependent clause introduced by a complementizer or a wh-pronoun, even in the case where the verb is clause-initial. This syntactic type is illustrated by the following examples, which are good evidence against any proposal to consider enclisis as a process connected with the "given" character of the clitics in contrast with the verb. This property is expected not to change—and so to be relevant—in a complement clause: but this is not the case, since we never have enclisis in a clause introduced by a complementizer, in any Romance variety of the Middle Ages: (19) a. (...) che vi si po a riporre, (...) che vi si possa sedere (Schiaffini (1954:68)) 'that there-one-can put..., that there-one-can sit...' b. (...) queli che ve dis ste parole (Lio Mazor: Levi 1904, 1 r, 21) 'those that to-you-tell these words'. The fact that enclisis is never found in a sentence introduced by a complementizer (which is a designated occupier of the head of C), suggests that verb movement to C feeds the enclisis. We can add that the occupiers of SpecC that we isolated before (si, an in Venetian, the direct object) also block enclisis; we never have enclitic pronouns if the verb is preceded by these elements: in (20a) we have preposed direct objects (quello che tu vorrai, gli altri), in (20b) an, in (20c) si. (20) a. Quello che tu .vorrai mi renderai e gli altri ti terrai (Florentine: Novellino, IX) 'What you want to-me-will give (2.sq) back and the others foryou-will keep' (= you will give back to me what you want and will keep the others) b.

An lo dies-tu ben! (Lio Mazor: a3 t, 68) 'Indeed it-said you well' (= indeed, you said it!)

c.

si la lave...e retornela... (cfr. 16a) 'so her-washed and put-her'.

336

CLITICS AND VERB SECOND

The first position of a main clause preceded by an adverbial clause, on the other hand, can work like an absolute first position, i.e. enclisis is possible (but not obligatory, as (21c) shows): (21) a. Ed essendo poveramente in arnese, mise-si ad andare ad Alessandro (Novellino) 'And being in poor condition, set-himself to go to A' b. ...e quando il vide, raffigurollo (Schiaffini 77, 17) 'and when him-saw, recognized~him' c. la famiglia volendoli bene, l'insegnaro a campare (Sch.) 'the family loving-him, him-taught (3.pl.) to get by'. This means that an adverbial clause can either be generated as a constituent of the main clause (or a SpecCP occupier)—in which case enclisis is impossible—or as an extra-sentential complement—in which case enclisis is obligatory. Examples (21a,b) represent the latter case, while in (21c) the untensed clause is interpreted as generated in SpecC position. In NIDs we have some (very few) cases of a preposed object which is not immediately adjacent to the verb. We can observe that in these cases, we always have a resumptive pronoun agreeing with the preposed object and, if the intermediate constituent is a designated occupier of Spec, the clitic is proclitic: (22) Messer Pepo mando in certa parte, e messer Cante, perche' era grande suo amico, st'l mando a Mantova 'M.P. (obj) (he) sent in a certain part, and M.C. (obj.), since he was a great friend of his, so him sent in Mantua'. On the basis of the observations provided so far, it seems correct to conclude that, in order to have the enclisis of the pronoun to the verb in the appropriate cases, it is necessary for the verb to have moved into C. At this point, the enclisis of the clitic complements is sensitive to an empty SpecC.10 Let us leave the question open as to what the exact process is, and informally define enclisis as the result of an upward movement of the verb alone triggered by an empty SpecC. Italian varieties having more than one position above the head C—i.e. SpecC and Top, give us the opportunity to check this hypothesis. Since we have ways to characterize the SpecC with respect to Top position, we can see that an empty SpecC triggers enclisis, also when a constituent precedes the verb, if this constituent is not in SpecC but in Top. The formulation adopted allows us to describe in quite a simple way the different behaviour of Old French and Medieval Italian varieties in coordinate sentences and yes-no questions. We saw above that comparing the two linguistic types, we reach contrasting conclusions: with respect to coordinate sentences, it seems that French has a weaker constraint than Italian; with respect to questions, it seems that the constraint is weaker in Italian. I assume, then, that Italian and French grammar interpret the conditions of SpecC in these sentences differently. In yes-no questions it is possible to interpret the Spec of CP either as a node filled by an abstract Q operator, or as an empty node since it contains no lexical elements.

COMPLEMENT CLITICS IN MEDIEVAL ROMANCE

337

If the grammar considers the Spec to be empty, we have the situation found in Old French, where there is enclisis. If the Spec is considered not to be empty, since it contains an abstract operator, we have the situation found in Italian varieties, i.e. we always have proclisis: the Spec being filled, the clitics appear in sentence-initial position. In this interpretation, the different behaviour of Italian and French in coordinate sentences depends on the level where coordination occurs: if the attachment is at CP level, we will have an empty Spec, which will trigger enclisis,11 this is the situation of Italian varieties. If the coordination is at the level of C', we do not have an empty Spec in the second clause; this is the case of French.12 As I said above, it is possible to make some simple and straightforward predictions stemming from the generalization that when we have an empty SpecC in a clause, we have enclisis of clitic pronouns, while when we have a constituent that must be in the Spec of CP, we never expect to have enclisis. Furthermore, verb movement is to be considered as feeding the further movement of the verb. Then if verb movement to C is blocked, we predict that "verb-clitic" order does not occur. As expected, we never have "verb-clitic" in relative clauses or in wh-questions. In relative clauses we do not have an accessible head of CP for the verb to move into, in a wh-question there is no empty SpecC, as the wh-word is obligatorily there. I also assumed that a direct object in first position with no resumptive pronoun is obligatorily in SpecC. If the verb has other clitics corresponding to different arguments, these are always expected to be proclitics. This is also confirmed by the data and will be checked later in slightly different languages, Old Portuguese and Old Southern Italian, which are pro-drop languages. If, on the other hand, we could have a constituent in first position, which cannot be in the SpecC but has to be in Top, we could build up a complementary argument. In my theory, a preposed object with a resumptive pronoun has to be in Top, even if in the surface it appears immediately adjacent to the verb. If between a preposed object and the governing verb there is no intervening constituent and the object has a resumptive clitic, this clitic will always be enclitic. For some reason, a "heavy" constituent—in particular a NP containing a relative clause—is likely to be left dislocated (i.e. in Top), as shown in Vanelli (1986). So we have in the following sentence (from Schiaffini (1954:282, n.9) an appropriate example, with two preverbal constituents that are both in Top: (23) [A voi], [le mie poche parole ch'avete intese], olle dette con grande fede. 'To you, my few words that you have heard, (I) have-them said with great faith'. A Hanging Topic could also be a constituent that has to be in Top, as far as we know. If there is no other constituent between the Hanging Topic and the verb, I expect a clitic—if there is one—to always be enclitic, as is in fact the

338

CLITICS AND VERB SECOND

case. See the following example from Old Florentine: we can argue that the first constituent is a Hanging Topic from the fact that it does not have the preposition a required by the subc tegorizing verb (see Cinque (1983)). (24) [Quelgli il quale ndasse per Firenze...in die di lavorare], debbialgli essere soddisfatto... (Schiaffini (1954:54, 1.13) 'He who should go through Florence...in a working day, be-to him paid compensation...' In the following sentence (25) (from Schiaffini (1954: 46, 11.32-33)) we have apparently a counter example: the first constituent is again a Hanging Topic—it lacks the preposition a—and we would expect enclisis of gli: (25) Et chi facesse contra la prima volta, gli sia imposta penitenca, et la seconda sia cacciato. 'And he who should act against the first time, to-him be imposed a penance, and the second be pulled away'. But we can easily accommodate this sentence both for the theory and for the sense itself; as indicated in (25') with a different punctuation, the Hanging Topic is only chi facesse contra, while la prima volta is a separate constituent, which we can suppose to be in SpecC: (25') Et chi facesse contra, la prima volta gli sia imposta penitenca, et la seconda sia cacciato. 'And he who should act against, the first time to-him be imposed a penance, and the second be pulled away'. Notice that the emendation permits a more natural contrast between the— now isolated and symmetrical—constituents la prima volta...la seconda volta.

4, Pro-drop Medieval Romance A last set of data regarding the distribution of constituents in the left periphery of the sentence come from the third group of Medieval Romance languages that I am taking into account. This group includes the earlier stages of Portuguese, Spanish, Catalan, Galician, and Southern Italian Dialects (=SIDs). I will consider SIDs and Old Portuguese. In these languages there is no asymmetry between main clauses and dependent clauses with regard to pro-drop, in the sense that we always have pro-drop. Moreover, the verb is not in second position, but, as in NIDs, we have also verb-first, verb-third, verb-fourth main clauses. Nevertheless, I consider these languages as a variant of the same Medieval Romance type. They exhibit syntactic patterns which I have argued are connected with verb movement into C. The main feature is again the preposed object with no resumptive pronoun together with verb-clitic order. In as far as they are both consequences of verb movement into C, they characterize Old Portuguese and Southern Italian as verb-second languages of the broad type. This leads us to say that

COMPLEMENT CLITICS IN MEDIEVAL ROMANCE

339

a language which does not show second position of the verb in main clauses, nor the asymmetry of pro-drop, can nevertheless be a verb-second language. The broad type, then, apart from pro-drop asymmetry which is independently related to verb second, differs from the strict type only owing to the availability of the Top position. In a dependent clause introduced by a wh-pronoun or a complementizer we always have the order clitic-verb, as in the following Old Portuguese examples (from Huber (1933:160 ff.)): (26) a. As tuas doces palavras per que me prometias 'your sweet words with which you promised me' b. O rato respondeo que Ihe prazia 'The mouse answered that to-him-pleased'. The behaviour of a preposed direct object is exemplified in (27-29): (27) a. Sao ko kelle terre, per kelle fini que ki contene, trenta anni le possette parte Santi Benedicti (the oldest Italian text, 960, from Capua) 'I know that those lands, in those boundaries that here contains, thirty years them-owned the party of St. Benedict' (28) a. La salamandra audivi... (Jacopo da Lentini, Sicily) 'The salamander (I) heard...' b.

La mia gran pena e lo gravoso affanno..., madonna lo m'ha in gioia ritornato (Guido delle Colonne, Sicily) 'My great sorrow and grievous pain...my Lady it to-me has into joy turned'

(29) a. Tal service Ihe pode fazer hun homen pequenho (OPortug: Huber) 'Such service to-him can do a little man (subj.)' b. O trigo que eu como, guanco-0 per meu trabalho 'The wheat that I eat, (I) earn-if with my work'. Example (29a) has a preposed object with no resumptive pronoun and proclisis of the indirect object pronoun; (29b) has a preposed direct object with a resumptive pronoun: the clitic is enclitic. This pattern is consistently observed in Old Portuguese.13 I conclude that the preposed object is in SpecCP when it has no resumptive pronoun. Sentences like (29a,b) are evidence in favour of this hypothesis. If we have a direct object in SpecC, a resumptive pronoun is not expected. If instead we have a clitic corresponding to a preposed object, the object has to be in Top and not in SpecC. If SpecC is empty, this will trigger enclisis of the clitic (as in 29b). Another symmetrical prediction is the following: we will never have enclisis of a clitic if the preposed direct object has no resumptive pronoun. The object has to be in the Spec of C in this case and thus no fur-

340

CLITICS AND VERB SECOND

ther movement of the verb is possible. These predictions are completely confirmed by the available data. But including pro-drop Medieval Romance in the discussion, and in particular Old Portuguese, has a useful consequence: in fact, with respect to these patterns, Old Portuguese differs very little from Modern Portuguese. We can test our prediction then in a wider range of contexts, and, what is more important, we can elicit judgments about ungrammatical sentences. In Modern Portuguese (of Portugal) the subject is regularly in Top, as we detect from the fact that clitics normally follow the verb when there is a preverbal subject (in Old Portuguese, this was very common but not obligatory).14 (30) a. O Joao disse-nos b. *O Joao nos disse 'John told us'. But some subjects are never followed by a verb with enclitic pronouns. I would predict that, if my hypothesis is correct, the subjects that block enclisis are the subjects that must be in SpecC, as the following contrasts show: (31) a. ...quern me chamou? 'Who called me' b.

*...quern chemou-me?

(32) a. Ninguem nos viu /*viu-nos 'Nobody saw us' b. Todos se lembran /*lembran-se 'Everybody remembers'. Quantified subjects are considered to occupy SpecC obligatorily. Thus, the verb must be in the head of CP and enclisis of the clitics is impossible. The same happens with the negation: (33) a. Nao os comprendo b. *Nao comprendo-os '(I) them understand'. There is here a difference with medieval varieties, where the negation could either satisfy the verb-second requirement, i.e. occupy SpecC, or be proclitic of the verb (see Vanelli, Renzi and Beninca (1985)). With the support of the proposed tests the Tobler-Mussafia law can be restated as follows: In Medieval Romance (and in Modern Portuguese) complement clitics occur after an inflected verb if and only if the governing verb is in C and the Spec of CP is empty.

COMPLEMENT CLITICS IN MEDIEVAL ROMANCE

341

Notes * Previous versions of this paper have been presented at the Linguistics seminar in Venice in 1989 and at the First Generative Diachronic Syntax Conference in York, April 1990. I am greatly indebted to Giulio Lepschy and Paolo Salvi for discussing both the form and content of this paper. I also owe a debt of gratitude to Adrian Battye: he made precise observations on the analysis presented in first version of this paper; his organization of the York conference gave that meeting an especially pleasant atmosphere. To a deeply missed friend and colleague, I dedicate these pages. 1. There is evidence that the relation between a verb and an enclitic pronoun in Romance involves a process in some sense going "deeper" into morphology than is the case with proclitics. For some evidence of this close relation between an enclitic pronoun and a verb, see Beninca and Cinque (1993). Mussafia began his note of 1886 with these introductory lines: "I pronomi personali oblique atoni mi, ti, si, ecc., e le particelle pronominali atone ci, vi, ne o precedono il verbo di forma finita (proclisi) o gli tengono dietro, formando con esso una parola sola (enclisi)." ("The oblique atonic personal pronouns me, ti, si, etc., and the atonic pronominal particles mi, vi, ne, either precede the finite verb [proclisis] or follow it forming a single word with it [enclisis]"— editors.) It seems to me that we still have no means to account for intuitions of this kind. 2. See Tobler (1875), Mussafia (1886). 3. In infinitival and gerundive clauses probably other factors complicate the picture, but it seems that the main lines of the description I will provide hold in the same way. This means that clitics were not constrained always to be proclitics with these verb moods, as in Modern French for example, nor always to be enclitics, as in Modern Italian (excepted some cases, in fact), but again the position of complement clitics relative to gerund and infinitive depended on a broader syntactic context. A similar, though more restricted, possibility is still found in Modern Neapolitan and in other Italian dialects. 4. For example, Mussafia (1886), Meyer-Lubke (1897), Foulet (1924), Adams (1987), Cardinaletti and Roberts (1991). Marcantonio (1976,1980), Antinucci and Marcantonio (1980) proposed an analysis connecting enclisis to syntax, into a generative semantics framework. 5. See Benacchio and Renzi (1987), Renzi (1992). 6. In Old Spanish, as in Old Portuguese and Galego-Portuguese, a different possibility is met, with complement clitics attached to the complementizer or a Wh-element in dependent clauses, and separated from the verb by the subject and/or a restricted set of adverbs: the following example is from Ramsden (1963:137): (i) quando la el rey dixo 'when-it the king said'. See also Salvi (1990, 1991).

342

CLITICS AND VERB SECOND

I do not deal with this particular aspect of cliticization. It seems that, at a certain point of the derivation, verb and clitics are not in the same complex head, but in distinct functional projections, and only the higher one—where complement clitics are—moves to C in dependent clauses. 7. For reasons still unclear, probably related to a stronger verb agreement, the asymmetry is less evident in Florentine: but see Renzi (1983) for discussion and data. 8. See Ulleland (1960), Schiaffmi (1954), Antinucci and Marcantonio (1980), Marcantonio (1980). 9. Cf. also Mussafia (1886, ch.4), and the numerous examples given in Melander (1929) for different purposes: for all MRLs, where a complementizer appears, proclisis is always observed. 10. The whole process reminds us of the enclisis that is found in Modern Italian untensed and imperative clauses (see Kayne (1990)), where we have again an empty Spec (a Specifier of Agreement) and the verb is obligatorily followed by the clitics, which become enclitics. I will not deal with untensed clauses in Medieval Romance (see note (3)), nor consider the possible resemblances between the topic discussed here and the position of complement clitics in untensed clauses in Modern Romance languages. It should be stressed, however, that the process involved with untensed verbs and imperatives, possibly an upward movement, seems to differ basically from any movement to C. This observation is relevant to the analysis proposed by Lema and Rivero (1990): in the case of untensed verbs, we never find any evidence of complementary distribution of enclisis and lexical or abstract elements in CP (whelements or quantifiers). If an upward movement of the V head has to be postulated, it is a matter of movement into a position lower than C. 11. Modern French still deals differently with a Spec of CP position occupied by an operator and a Spec of CP filled by a lexicalized or an abstract wh-element. The case is Stylistic Inversion (see Kayne (1972); Kayne and Pollock (1978)). The Spec of CP in yes-no questions contains an operator that, while triggering verb movement to C, is not able to license an argumental pro in SpecIP. 12. This formulation does not contrast with the restrictive interpretation of coordinate structure proposed in Beninca and Cinque (1993), where it is suggested that no empty argument positions are generated in a coordinate structure. Crucially, the present case does not concern an empty argument position; furthermore, the choice proposed here is between different levels inside a unique projection. 13. Thanks to (Gian) Paolo Salvi for supplying data from Medieval Iberian documents. Data from Medieval Sardinian are consistent, too, as appears, for example, from the detailed results of Melander (1929). 14. The sentence is perfect in Brazilian Portuguese. For some observations suggesting that this variant differs from European Portuguese, presumably no longer being pro-drop nor verb second, see Adams (1988:11-12).

COMPLEMENT CLITICS IN MEDIEVAL ROMANCE 343

References Adams, M. (1987) Old French, Null Subjects and Verb Second Phenomena. PhD Dissertation, UCLA. Adams, M. (1988) "Parametric Change: Empty Subjects in Old French." In D. Birdsong and J.P. Montreuil, eds. Advances in Romance Linguistics, 1-16. Dordrecht: Foris. Antinucci, F. and A. Marcantonio (1980) "I meccanismi del mutamento diacronico: il cambiamento d'ordine dei clitici in italiano." Rivista di Grammatica Generativa 5:3-50. Babilas, W., ed. (1968) Untersuchungen zu den Sermones Sabalpini. Miinchen: Hiiber. Benacchio, R. and L. Renzi (1987) Clitici slavi e ramanzi. Padova: Clesp. Beninca, P. (1983-84) "Un'ipotesi sulla sintassi delle lingue romanze medievali." Quaderni Patamini di Linguistica 4:3-19. Beninca, P. and G. Cinque (1993) "Su alcune differenze fra enclisi e proclisi." In Omaggio a Gianfranco Folena. Padova: Programma. Cardinaletti, A., and I. Roberts (1991) "Clause Structure and X-second." In W. Chao and G. Horrocks, eds. Levels, Principles and Processes: the Structure of Grammatical Representation. Berlin: Foris/de Gruyter. Cinque, G. (1983) "Topic Constructions in Some European Language and Connectedness." In I. Ehlich and H. van Riemskijk, eds. Connectedness in Sentence, Discourse and Text. Tilburg. Foulet, L. (1924) "L, accent tonique et 1'ordre des mots." Romania 50:54-93. Huber, J. (1933) Altportugiesisches Elementarbuch. Heidelberg. Kayne, R. (1972) "Subject Inversion in French Interrogatives." In J. Casagrande and B. Saciuk, eds. Generative Studies in the Romance Languages, 70-126. Rowley, Mass.: Newbury House. Kayne, R. (1990) "Romance Clitics and PRO." Ms. CUNY (Published as Kayne, R. "Romance Clitics, Verb Movement and PRO." Linguistic Inquiry (1991) 22:647-686.) Kayne, R. and J.-Y. Pollock (1978) "Stylistic Inversion, Successive Cyclicity and Move NP in French." Linguistic Inquiry 9:595-621. Lema, J. and M.L. Rivero (1990) "Types of Verbal Movement in Old Spanish: Modal, Futures and Perfects." Ms. University of Ottawa. Levi, U. (1904) / Monumenti del dialetto di Lio Mazor. Venezia: Visentini. Marcantonio, A. (1976) "Un aspetto dell'ordine delle parole nell'italiano del Due-Trecento." Rivista di Grammatica Generativa 1,2:57-77. Marcantonio, A. (1980) "Alcune considerazioni sulla legge Tobler-Mussafia." In P. Gerrettoni, eds. Problemi di analisi linguistica. Perugia. Melander, J. (1929) "L'origine de 1'italien me ne, me lo, te la, etc." Studia Neophilologica 2:169-203. Meyer-Lubke, W. (1897) "Zur Stellung der tonlosen Objekts-pronomina." Zeitschrift fur Romanishe Philologie 21:313-334. Mussafia, A. (1886) "Una particolaritd sintattica della lingua italiana dei primi secoli." In Miscellanea di filologia e linguistica, dedicata alla memoria

344

CLITICS AND VERB SECOND

di Napoleone Caix e Ugo Angelo Canello, 255-261; 474-475. [Now reprinted in A. Daniele and L. Renzi. eds. (1983) Adolfo Mussafia, Scritti di Filologia e Linguistica, 291-301. Padova: Antenore.] Ramsden, H. (1963) Weak Pronoun Position in the Early Romance Texts. Manchester University Press. Renzi, L. (1983) "Fiorentino e italiano." In F. Albano Leoni, ed. Italia linguistica. Bologna: II Mulino. Renzi, L. (1989) "I pronomi soggetto in due varieta substandard: fiorentino e franc.ais avanceV Zeitschrift fur romanische Philologie 108:72-98. Salvi, G. (1990) "La sopravvivenza della legge di Wackernagel nei dialetti occidentali della penisola Iberica." Medioevo Romanzo 15:177-210. Salvi, G. (1991) "Difesa e illustrazione della legge di Wackernagel applicata alle lingue romanze antiche." In Miscellanea G.B. Pellegrini. Padova: Unipress. Schiaffini, A. (1954) Testi florentini del Dugento e dei primi del Trecento. Firenze: Sansoni. Tobler, A. (1875) Review of J. Le Coultre, De I'ordre des mots dans Chretien

de Troyes. Reprinted in Vermischte Beitrdge zur Franzosische

Grammatik, V. Leipzig 1912, 400. Ulleland, M. (1960) "Alcune osservazioni sulla legge Tobler-Mussafia." Studia Neophilologica 32:53-79. Vanelli, L. (1986) "Strutture tematiche in italiano antico." In H. Stammerjohann, ed. Tema-Rema in Italiano, 249-273. Tubingen: Narr. Vanelli, L., L. Renzi and P. Beninca (1985) "Typologie des pronoms sujets dans les langues romanes." In Actes du XVIIe Congres Inernational de Linguistiques et Philologie Romanes 3:164-176. Zwanenburg, W. (1978) "L'ordre des mots en fran?ais medieval." In R. Martin, ed. Etudes de syntaxe du Moyen Age francais, 153-171. Paris: Klincksieck.

13 Cases of Verb Third in Old High German* Alessandra Tomaselli University of Pavia

This paper is organized into five sections. In the first section, I introduce and discuss the particular construction which this investigation will concentrate on: the cases of verb-third order in matrix clauses in Old High German (OHG henceforth). In SECTION 2 OHG data will be compared with Old English (OE) data (cf. Kemenade (1987)). The third section is devoted to the presentation and discussion of two alternative analyses of these verb-third word order patterns already proposed in the literature: that in Lenerz (1985), and that in Kemenade (1987). Finally, in the last two sections, I will propose a partially new analysis, which will lead us to discuss two important issues: the clitichead relation and the w/t-construction.

1. Two Types of Verb Third OHG syntax was characterized by the verb-second constraint (as in Modern German). This fact is clearly noted in the well-known traditional descriptions, e.g. Behaghel (1923-32) and Erdmann (1985). It has also been clearly confirmed by recent studies developed within the theoretical framework of generative grammar (cf. Lenerz (1984), (1985); Tomaselli (1989); Weerman (1989)). However, OHG allows one kind of order which deviates from the general verb-second pattern, and which is not found in modern verb-second Germanic languages: in the main declarative clause the finite verb could occupy, apart from the second position, a "later" or "delayed" position known as Spaterstellung in the traditional descriptions. Two important points should be clarified concerning the notion of "delayed" position. First, the cases of "delayed" position in the main declarative clause are a minority compared to "regular" verb second. Second, from a syntactic point of view, the term "delayed" position refers to different word order 345

346

CLITICS AND VERB SECOND

patterns. Under this label one could in fact subsume all the cases in which the finite verb occupies a position between the second and the final position of the sentence: verb-third, verb-fourth, etc.. Inside this "basket" of exceptions to verb second our attention will concentrate on what certainly seems more interesting with respect to the general problem concerning the "degree of realization" of the constraint in OHG syntax: the cases of verb third. First of all it should be stressed that even an apparently simple label like "verb-third" does not unambiguously refer to a unique word order pattern. It is well known that OHG prose consists of translations from Latin. This fact helps us to differentiate between two different types of verb third. Let us first consider the following examples: (1)

Isidors Schrlft contra ludaeos

(8th/9th century)

a. dhaz ir chichundida Obj Subj Vfnt that he showed 'He showed that' Braune and Ebbinghaus (1979:19,135) b. erino portun ih firchnussu Obj Subj V fnt iron doors I shatter 'I shatter iron doors'. Braune and Ebbinghaus (1979:16,25); cf. also Lenerz (1985:106) c. Dhes martyrunga endi dodh uuir fmdemes mit urchundin NP Subj Vfnt PP His martyrdom and death we demonstrate with evidence dhes heilegin chiscribes NP(Genitive)

of the Holy Writings 'We demonstrate his martyrdom and his death with evidence from the Holy Writings'. Lippert (1974:52) (2)

Tatian (9th century) Inti ubil man fon ubilemo tresouue bringit ubilu Coor Subj PP Vfnt Obj and (a)badman from a bad treasure brings ill Latin: et malus homo de malo thesauro profert mala Braune and Ebbinhaus (1979:24,15)

(2), which is taken from Tatian (East Frankish, 9th century), represents a case of verb third which is clearly due to Latin influence. First, the finite verb (bringit) follows two "full" constituents: the subject NP and a PP. Second, the order of elements corresponds perfectly to the one in the Latin sentence.1 This type of verb-third construction will simply not be considered in this paper although we do not discount the possibility of a genuine syntactic explanation for this order, in place of Latin influence.

CASES OF VERB THIRD IN OLD HIGH GERMAN

347

On the other hand, examples (la-c), taken from the OHG translation of De fide catholica contra Judaeos by Isidorus Hispalensis, exemplify a different type of construction which could be considered peculiar to Germanic syntax: the verb-second constraint is violated by the presence of the subject pronoun which intervenes between the fronted constituent and the finite verb.2 It is clear that examples (la) and (Ib) could be considered cases of verbthird as well as cases of verb last (and, in fact, a verb-last analysis has been proposed by Lenerz (1985)). However, I prefer to consider these examples as cases of verb third given that they clearly exemplify the same kind of construction together with example (1c), where the third position of the finite verb does not correspond to the final position of the sentence.3 What has been said till now is well known to people familiar with Germanic philology. There is, in fact, general agreement among Germanists that the following properties hold for OHG: a) b)

OHG prose was characterized by the verb-second constraint (subject-verb inversion in the main declarative clause); the finite verb may shift to third position given the presence of a pronominal element in second position (cf., among others Lippert (1974:15)).

In what follows I will concentrate on exactly this kind of construction.

2. Cases of Verb Third in Old English The situation previously sketched for OHG corresponds to that found in OE, as described by Kemenade (1987). Indeed, it comes out clearly from her work that Old English was characterized by a verb-second constraint, and that the verb-second constraint was systematically violated by the occurrence of a pronominal element in second position. This is shown by the following examples, all taken from Kemenade (1987:110): (3)

/Efter his gebede he ahof baet child up PP Subj Vfnt Obj Participle after his prayer he lifted the child up

(4) pas ping we habbab be him gewritene Obj Subj Vfnt PP V these things we have about him written 'these things we have written about him' (5)

Forbon we sceolan mid ealle mod & maegene to Gode gecyrran Adv Subj Vjnt PP PP V therefore we shall with all mind & power to God turn Therefore we shall turn to God with all our mind and power'.

This construction is much more general in Old English than in OHG in two respects:

348

CLITICS AND VERB SECOND

a)

b)

this construction is well attested in Old English prose (on the contrary, in OHG the construction under consideration is attested only in the translation of Isidor and partially in the Monsee-Wiener Fragmenten); while in OHG the pronoun which intervenes between the fronted element and the finite verb generally corresponds to the subject pronoun, in OE it can also be the object of the verb or the object of a preposition.

Note that this difference between OE and OHG seems, in fact, to confirm the traditional hypothesis that OHG was a more consistently verb-second language than Old English (cf. Fourquet (1938)). Despite these differences, Old English and OHG syntax show two important similarities: both allow the relative order pronoun-Vfnt (cf. SECTION 2.1); both have the same range of positions of the finite verb in the subordinate clause (cf. SECTION 2.2).

2.1 The Relative Order Pronoun-Finite Verb The order pronoun-finite verb is subject to the same constraints both in OE and OHG. In other words, the distributional facts captured by Ans van Kemenade for Old English also hold for OHG data. Her results are summarized by the following schema (cf. Kemenade (1987:139)): (6) Main Clause: a. XP - pronoun+Vfnt . . . . . . b.(WH) ne - Vfnt+pronoun -

w

/ * WH/ne/ha - pronoun+Vfnt

Subordinate Clause: c. Comp.+pronoun - . . . . . / * pronoun+Comp.

.....

Note that, as far as the main clause is concerned, the pronoun precedes the finite verb only when the fronted element corresponds to a non-negative declarative constituent (XP) (pattern (6a)— cf. (la-c), (3), (4) and (5)). On the contrary, when the first constituent of the main clause corresponds either to a wh-constituent or to the negative clitic or to the adverb ba 'then' the pronoun follows the finite verb in third position. This word order pattern (cf. (6b)) will be analyzed later (cf. SECTION 5) where the syntactic nature of the adverb ba will also be discussed. In the subordinate clause (cf. (6c)), the pronoun immediately follows the lexical complementizer: Old English (examples from Kemenade (1987:59)):

CASES OF VERB THIRD IN OLD HIGH GERMAN

(7)

349

a. baet he mehte his feorh generian that he could his property save 'so that he could save his property' b. baet hi mihton swa bealdlice Godes geleafan bodian that they could so boldly God's faith preach 'that they could preach God's faith so boldly'.

Old High German (Isidors Schrift contra ludaeos): (8) a. dhazs dhu firstandes heilac chiruni Comp Subj Vfnt Obj 'that you understand the Holy Secret'. b. dhazs ih fora sinemu anthlutte hneige imu dheodun ObObj Comp Subj PP Vfnt NP NP(Dat) that I in front of his face bow to him people 'that I make people bow to him before his face'. It is important to note that the strict adjacency requirement, lexical complementizer - subject pronoun, characterizes not only OHG syntax in general but also all modern verb-second languages. What uniquely characterizes Isidors Schrift contra ludaeos with respect to both other OHG texts and modern verbsecond languages is the possible violation of the verb-second constraint given the occurrence of a pronominal subject to the left of the Vfnt (i.e., the main clause word order pattern XP pronoun Vfnt (...)).4 In modern verb-second languages, as well as in other OHG texts, the subject pronoun immediately follows the syntactic position C° independently of the lexical item which fills it. In other words, the subject pronoun occupie a position immediately to the right of: i) ii)

the lexical complementizer in the subordinate clause; the finite verb in the main clause.

The following examples are taken, respectively, from Williram (a prose text of the llth century) and from Muspilli (a poem of the late 8th century): Williram (cf. Braune and Ebbinghaus 1979:75,77,78): (9) a. daz er da ezze ddz uuocher sines eigenen ebeze Comp Subj Adv Vfnt NP(obj) that he then eats the harvest of his fruits b. daz ih nieuudnne necume in conventicula haereticorum Comp Subj Adv Neg+Vfnt PP that I never not+come in the heretical congregations c. thicco gehiezzer mir sine cuomst per prophetas Adv V fnt +Subj NP(Dat) NP (obj) PP Often announce+he to me his coming by prophets Muspilli (cf. Braune and Ebbinghaus 1979:87):

350

CLITICS AND VERB SECOND

(10) a. denner mit den miaton marrit daz rehta Comp+Subj PP Vfnt NP(obj) when+he against payment prevents the right b. daz hortih Obj Vfnt+Subj that heard I As for the relevance of the adjacency requirement C°-subject pronoun in modern verb-second languages I refer here directly to Tomaselli (1989, 1990a).

2.2 The Position of the Finite Verb in the Subordinate Clause A second important similarity between Old English and OHG concerns the position of the finite verb in the subordinate clause. What is particularly interesting for our purposes is the fact that both OE and OHG are characterized by the phenomena of Verb Raising (VR) and Verb Projection Raising (VPR)— cf. Haegeman and Riemsdijk (1986) for a first analysis of these phenomena in West Flemish and Zurich Swiss German. The following scheme provides a first summary of the most common word order patterns attested in the subordinate clause: (11) a. Comp Vfnt xp b. Comp vfnt XP fnt c. Comp .......V Vfnt d. Comp... ....Vfnt V e . Comp NP(Subj) V fnt NP(Obj) V

(verb-last) (Extraposition of XP) (verb-last) (VR) (VPR)

Concerning (11), the following observations should be made: i)

While both OE and OHG clearly have OV order inside VP, the position of the Vfnt in the subordinate clause is certainly freer than, for example, in modern German. ii) In sentences with a simple tense (patterns (lla.b)), the verb may occupy the final position but it could also be followed by one (or more) constituents. This fact is generally attributed to the possibility of extraposition. iii) In sentences with a complex verbal form (e.g.: Modal-V, AuxiliaryPast Participle) we find essentially three different word order patterns: 1) the verbal complex occupies the final position. In this case the relative order Past Participle - V fnt is clearly subject to dialectical variations. We can find both the order Past Participle - Vfnt as in modern standard German and the order Vfnt - Past Participle (a phenomenon attested, for example, in Standard Dutch and which is usually referred to as Verb Raising). 2) the subordinate clause could be characterized by the so-called bracket structure (cf. (lle)); in other words, the object NP occurs between the Vfnt and the non-finite part of the verbal complex giv-

CASES OF VERB THIRD IN OLD HIGH GERMAN

351

ing rise to a construction otherwise typical of the main clause (Verbale Klammerbildung).5 What is really important to note here is that any instance of bracket structure in the subordinate clause is crucially different from the main clause Verbale Klammerbildung with respect to the possible positions of the subject NP. In the main clause the subject NP could occur either sentence initially to the left of the Vfm or to the right of the Vfnt (subject-verb inversion: XP Vfnt NP(subj) ...)• In the subordinate clause, on the contrary, the subject NP generally occurs to the right of the complementizer before the Vfnt. Some examples of the word order patterns just discussed are given below in SECTIONS 2.2.1 and 2.2.2. 2.2.7 Old High German (Isidors Schrift contra Idaeos) The position of the V fnt in OHG can be exemplified by the following data: a. Verb-last (12) dhem izs firgheban uuard REL Subj V Vfnt '(To) whom it forgiven was'. b. Extraposition (13) dhazs uuerodheoda druhtin sendida mih zi dhir Comp Subj Vfnt Obj PP 'that the Lord of the army sent me to you'. (14) dhazs dhu firstandes heilac chiruni Comp Subj Vfnt Obj 'that you understand the Holy Secret'. (15) dhazs ih fora sinemu anthlutte hneige imu dheodun Comp Subj PP Vfnt NP(Dat) Obj that I in front of his face bow to him people 'that I make people bow to him before his face'. (16) dhazs dher selbo gheist ist got Comp Subj Vfnt NP 'that the spirit himself is God'. c. Verb Raising (17) dher fona uuerodheoda druhtine uuard chisendit REL PP Vfnt V 'who from the Lord of the army was sent'

352

CLITICS AND VERB SECOND

d. Verb Raising + Extraposition (18) dhazs dhiz ist chiquhedan in unseres druhtines nemin Comp Subj Vfntfnt V PP 'that this is said in the name of our Lord' e. Verb Projection Raising (19) dhazs dhar ist Christ chizeichnit Comp Adv Vfnt Subj V 'that there is Christ meant' (examples from Braune and Ebbinhaus (1979); cf. Tomaselli (1989:101-103)). 2.2.2 Old English All the following examples are taken from Kemenade (1987:40, 55, 59): a. Verb-last (20) paet ic pas boc of Ledenum gereorde to Engliscre spraece awende Comp Subj Obj PP PP Vfnt 'That I translate this book from the Latin language to the English tongue' (21) Paet hie gemong him mid sibbe sittan mosten Comp Subj PP PP V Vfnt 'that they must settle in peace among themselves'. b. Extraposition (22) aefter disum gelamp (then it happened) paet micel manncwealm becom ofer Paere Romaniscan leode Comp Subj Vfnt PP 'that a great plague came over the Roman people'. c. Verb Raising (23) paet he Saul ne dorste ofslean Comp Subj Obj Neg+Vfnt V 'that he didn't dare to murder Saul'. d. Verb Projection Raising (24)

Paet he mehte his feorh generian Comp Subj Vfnt Obj V 'that he could save his property'

(25) baet hi mihton swa bealdlice Godes geleafan bodian Comp Subj Vfnt Adv Obj V 'that they could preach God's faith so boldly'.

CASES OF VERB THIRD IN OLD HIGH GERMAN

353

2.2.3 First Conclusion The most important issue is whether the XP-pronoun-Vfnt (...) order in the main clause can be connected to the position of the Vfnt in the subordinate clause in particular with the word order pattern (lle), i.e. Comp NPSubj Vfnt ... V. In other words if a systematic correlation between these two word order patterns can be found it would thus be possible to explain the two differences concerning the syntax of the Vfnt in OHG with respect to Modern German (namely the "apparent" exception to verb second in the main clause and the occurrence of the verbal bracket structure in the subordinate clause) by the resetting of one single parameter: the head-complement order parameter inside IP (I°-VP versus VP-I°). More explicitly, our hypothesis is the following: the syntax of both OHG and Modern German is characterized by the verb-second phenomenon; the difference between these two grammars consists in the placement of the syntactic position of the head of IP, which was medial in OHG while it is clearly final in Modern German. A quick overview of some well-known OHG texts6 and of some extensively studied Modern West Germanic languages provides us with the following picture: (26) An overview: Table 13.1 [VpNP V] V-2 XP pronoun+Vfnt ... Comp NP Vfnt ...V VR Old English

+

+

+

+

+

OHG Isidor Muspilli Williram Memento Mori

+ + + +

+ + + •f

+

+ (+) + -

+ + +

+ Zurich Swiss German +

+ +

+ +

+ +

German Dutch

+ +

_ -

+ +

West Flemish

+ +

This table shows that there is a correlation between the position of the Vfnt in the main clause and the position of the Vfnt in the embedded clase. The Comp NPsubj Vfnt NPobj V order is found in the subordinate clause whenever the XPpronoun-Vfnt order occurs in the main clause. Prima facie, the hypothesis that these two word order patterns should both be related to the medial position of I° seems to be confirmed. Before exploring this hypothesis (cf. SECTION 4), some points should be stressed concerning (26): 1)

Diachronically, the order XP-pronoun-Vfnt (...) is lost before VPR in the subordinate clause.

354

CLITICS AND VERB SECOND

2)

The word order pattern (lle) (i.e. Comp NP(Subj) Vfnt NP(Obj) V) could refer to two different syntactic constructions (cf. note 5). This fact obviously plays an important role in establishing the correlation under discussion, which does not seem to hold the other way round in Williram (OHG) and more obviously in two Modern Germanic languages like West Flemish and Zurich Swiss German. In these languages, it is clear that: i) the possibility of having instances of bracket structure in the subordinate clause does not correlate with verb-third cases in the main clause order pattern Comp NP(Subj) Vfnt NP(Obj) V must be attributed to VPR (see, among others, Haegeman and Riemsdijk 1986); iii) the presence of VPR is strictly dependent on the presence of "simple" VR (den Besten (1986)).7 On the other hand, in OHG and OE, at least some instances of bracket structure in the subordinate clause could be analyzed, in principle, as the result of V movement to a medial I0.8 Here in fact the correlation between the two word order patterns under consideration (verb-third cases in the main declarative clause and bracket structure in the subordinate clause) seems to hold in both directions. The precise nature of this correlation will be explored in SECTION 4.

3. Are Verb Third Cases Exceptions to Verb Second? In the preceeding section it has been clearly shown that both OE and OHG syntax present two important related similarities: a) b)

the relative order pronoun-Vfnt the positions of the Vfnt in the subordinate clause.

In this section it will be tentatively assumed that every analysis which captures OE data should also capture OHG data and vice versa. Concerning the word order pattern XP pronoun Vfnt . . . . . , two analyses have been put forward in the literature: for OHG on the one hand (cf. Lenerz (1985)) and for OE on the other (cf. Kemenade (1987)). In what follows these two analyses will be briefly presented and discussed.

3.1 Lenerz's Analysis Before we present the analysis proposed by Lenerz (1985), it is convenient to repeat here the relevant examples of matrix verb-third order (cf. (la-c)): (27) a. dhaz ir chichundida Obj Subj Vfnt that he showed 'He showed that'.

CASES OF VERB THIRD IN OLD HIGH GERMAN

355

b. erino portun ih firchnussu Obj Subj Vfnt iron doors I shatter 'I shatter iron doors'. c. Dhes martyrunga endi dodh uuir findemes mit urchundin NP Subj Vfnt PP His martyrdom and death we demonstrate with evidence dhes heilegin chiscribes NP(Genitive)

of the Holy Writings 'We demonstrate his martyrdom and his death with evidence from the Holy Writings'. Lenerz presents an analysis of matrix verb-third order which is based on his hypothesis about the historical development of the early stages of the German language. This can be summarized by the following quotation (cf. Lenerz (1985:126)): Early Germanic had (52a) [= (28a)] as base structure with (52b) [= (28b)] as a stylistic reordering, Xmax being any constituent, i.e. also the finite verb:

The preposed verb in (52b) [= (28b)] may then have been reanalysed as Sinitial Infl in base structure. Since there were also structures with Comp in OHG, S-initial Infl became identified with Comp (Confl). In other words, according to Lenerz (who bases his analysis of verb second on Platzack (1983)), the rise of verb second in OHG is crucially dependent on two different processes of reanalysis:

356

CLITICS AND VERB SECOND

(cf. Lenerz (1985:122,126), where W-position = Wackernagel position). Given the historical evolution illustrated in (29), there are two possible analyses of examples (27a) and (27b): i)

ii)

Examples (27a) and (27b) could be analyzed as "relics" of the first OHG stage (i.e.verb-last cases (cf. (29a)), which corresponds, in fact, to the Early Germanic stage (cf. (28a)). This analysis gives rise to one main objection: since the (subject) pronoun precedes the finite verb even in non-verb-last main clauses (cf. (27c)), such examples should be given a completely different explanation (but see note 3). Examples (27a), (27b) and (27c) could all be analyzed as instances of the third OHG stage (cf. 29c). The subject pronoun occupies the W-position (the position reserved for "light" elements), immediately preceding the Vfnt in Infl. There is again one main objection to this analysis. We would falsely predict that in the subordinate clause the following orders are both possible in OHG:

(30) a. OKComp - Subject pronoun - Vfnt b. * Comp - 0 - Vfm- NP(Subj) - . . . . .9 Apart from the objections just discussed, it is clear that only the analysis proposed in ii) is compatible with OE data. On the other hand, the analysis in i) predicts a difference between OE and OHG as far as the word order pattern XP-pronoun-Vfnt is concerned. Note, in fact, that in OE the verb-third cases never correspond to verb-last cases. This result is completely undesirable given that we want to provide the same analysis for both OE and OHG.

3.2 Kemenade's Analysis Turning now to the analysis provided by Kemenade (1987) for OE data, her main claim (contrary to Lenerz's proposal for OHG) consists in the assumption that examples like (3), (4) and (5) are instances of verb second and as such should be analyzed in terms of V° movement to C. Recall that Kemenade's analysis attempts to capture the following distributional facts concerning the relative order pronoun-Vfnt (cf. (6)): (31)Main Clause: a. XP - pronoun+Vfnt . . . . . .

CASES OF VERB THIRD IN OLD HIGH GERMAN

357

b.

- Vfnt+pronoun-....../* WH/ne/pa-pronoun+Vfnt

Subordinate Clause: c. Comp+pronoun -

/ * pronoun+Comp

Kemenade assumes that pronominal elements in OE could have the status of "syntactic" clitics. In these terms, the word order pattern XP-pronoun-Vfnt(...) (cf. (31a)) is attributed to a process of cliticization on the left of C°. More precisely, the S-structure representation of the word order pattern (3la) is as in (32) (where C°= Confl):

(adapted from Kemenade (1987:129,131,133)). Note that in order to explain the fact that in the word order patterns (31b) and (31c) the clitic position is on the right of the Confl-projection rather than on the left of Confl, Kemenade (1987:139-140) must assume that: i)

ii)

if the topic position (= [Spec, ConflP]) is occupied by an operator (wh-elements, pa and ne) the specifier and the head of ConflP appear to behave as one constituent within which cliticization is impossible; there is a crucial distinction between verb second as lexicalization of Confl0 and that as lexicalization of Confl. More precisely, that corresponds to the proper base-generated lexicalization of Confl0, whereas verb second must be viewed as a default lexicalizer.

A first immediate advantage of the analysis proposed by Kemenade consists in the fact that this is free from the objections that we raised against Lenerz's (cf. SECTION 3.1). Nevertheless two important theoretical objections do arise: First of all we must assume that, depending on circumstances (cf. (3la) versus (31b,c)), the clitic either adjoins on the left or on the right of the same structural head (Confl0);10 Second, Kemenade's analysis (like Lenerz's (1985))

358

CLITICS AND VERB SECOND

is crucially based on the assumption that the structure of the sentence of a verb-second language lacks an independent Infl-projection (henceforth IP). Note that this hypothesis, in its original formulation (cf. Platzack (1983)), was based on two distinct assumptions: a)

b)

in verb-second Germanic languages the morpho-syntactic characterization of C° resembles, in many respects, the characterization of I° in Romance languages; this fact makes the postulation of an "extra," independent IP superfluous.

Without compromising in any way the validity of the assumption in a), in the following section it will be argued, contrary to b), that the assumption of an (independent) IP is justified both by theoretical and empirical reasons (cf., amongst others, Platzack (1986); den Besten (1986); Tomaselli (1989, 1990a,b)).

4. Assuming an Independent Infl Projection: Advantages and Problems In what follows we will propose a substantial modification of Kemenade's analysis. We will assume that the structure of the clause in a verb-second language is characterized (as in non-verb-second languages) by two distinct functional projections, i.e. IP and CP; and that contrary to Modern Standard German, IP was a head medial projection both in OE and OHG. If we accept these assumptions, then the following analysis can be provided for the relative order pronoun-Vfnt both in OE and OHG: i) ii)

it is possible to assume that the (subject) pronoun cliticizes either on the left of I° or to the right of C° (possibly involving two different syntactic processes); the order XP-clitic+Vfnt-(...) in the main declarative clause is derived from: a) cliticization to the left of I; b) I to C movement.

This is illustrated in (33):

CASES OF VERB THIRD IN OLD HIGH GERMAN

iii)

359

cliticization to the left of the lexical complementizer is automatically excluded. Given that: a) the base-generation of the lexical complementizer in C° obviously prevents head-to-head movement (den Besten (1983); b) direct cliticization to the left of C° should be independently excluded;

there would be no way to derive the following unattested word order pattern: (34) clitic+Paet/dhazs-

(cf. (29c)).

The analysis just proposed gives a clear account of the word order patterns (31a) and (31c) but it is clearly not sufficent to explain (31b). Before turning to this important issue (cf. SECTION 5), something more must be said about structure (33).

4.1 IP as a Head-Medial Projection: Advantages At this point it is important to note that while the hypothesis of a headmedial IP is far from being unproblematic (cf. SECTION 4.2), it does present at least two immediate advantages which are briefly discussed in A) and B) below: A)

The possibility of correlating the process of cliticization to I° (which combined with verb second, i.e.: I° to C° movement, gives rise to the word order pattern: XP-clitic+Vfnt-(...)) with IP being head-medial (which combined with head-to-head movement gives rise to the word order pattern Comp-NP(Subj)-Vfnt-NP(obj)-V in the subordinate clause) is independently confirmed by the following well-known typological constraint:

(35) a. ok[IpNPi[r[rcli+Vfnt]VP] b. *[ipNPi[I'VP[Icli+Vfnt]]

In other words, the possibility of cliticizing to a final I° seems to be unattested cross-linguistically. Note, in fact, that cliticization to I should be excluded in an SOVI language both at the Phonetic Form (the movement would not be string-vacuous) and at S-Structure (movements to the right are usually limited to "heavy" elements); B)

It is possible to correlate the loss of cliticization to two different factors both concerning the category I°: For English, the loss of a process of cliticization to I° correlates with a change in the morphosyntactic characterization of I° (a process which is generally referred to as "deflection"; cf. Kemenade (1987) and Weerman (1989)); For German, the loss of cliticization to I° correlates with a change of the head-parameter within IP:

(36) NP(Subj) I° VP

> NP(Subj) VP I°

Note that this difference is perfectly compatible with the fact that the correlation between the loss of cliticization and the loss of verb sec-

360

CLITICS AND VERB SECOND

ond only holds for the history of English (cf. Kemenade (1987) and Hulk and Kemenade (this volume)). Both phenomena could in fact be reasonably traced back to the general process of deflexion which this language underwent. On the other hand, however, in the history of German the loss of cliticization corresponds to the strengthening of verb second. This can be explained if we assume that the loss of cliticization, in this case, is not to be linked to a change of the morpho-syntactic characterization of I° but rather to the different position occupied by I° at D-structure (cf. (36)). After this excursus through some of the most interesting consequences of the hypothesis that both OE and (what may be more problematical) OHG were characterized by a head-medial IP, let us briefly consider some problems which this hypothesis involves.

4.2 IP as a Head-Medial Projection: Problems Note that two different important assumptions underlie (33): IP is a distinct maximal projection with respect to CP; IP is analyzed as a head-medial projection. This second hypothesis poses, in fact, more than one problem: I)

II)

Concerning the history of English, the postulation of an intermediate SIOV stage naturally fits with the general hypothesis that the historical development of this language was characterized by a gradual change from an original SOVI to the actual SIVO type.11 On the contrary, as far as the history of the German language is concerned, the idea that OHG was characterized by a head-medial IP would imply that, at a certain point in its historical development, German changed from SIOV to SOVI.12 Since the Vfnt in both OE and OHG, could occupy the final position in the subordinate clause (cf. SECTION 2.2, point (9)), we must assume that: a) in the main clause V° obligatorily passes through I° on its way to C° in order to pick up possibly the clitic and to avoid a violation of the Head Movement Constraint; b) in the subordinate clause V° to I° movement should be stated in optional terms; c) we should draw a potential distinction between "full" verbs on one side and modal and auxiliary verbs on the other. In fact, if V° to I° movement, as far as the subordinate clause is concerned, was more precisely limited to modal and auxiliary verbs (as suggested to me by Ans van Kemenade and Tony Kroch (p.c.)), then we would still need an explanation for why I° can never represent a final landing site for "full" verbs.13

The possibility of giving an adequate and detailed answer to the problems just sketched goes beyond the purposes of this article and is therefore left to further research.

CASES OF VERB THIRD IN OLD HIGH GERMAN

361

5. Why Does the Presence of a Clitic on Cornp0 Interfere with the Specifier-Head Relation? A problem which has remained open until now is the following, given the word order pattern (31b), here repeated as (37): (37 V fnt+pronoun -

/ * WH/ne/pa - pronoun+Vfm

Why can the (clitic) pronoun not intervene between the element in the specifier of CP ([Spec, CP]) and the Vfnt in C°? The answer provided by Kemenade (1987) consists, as we have already noted above (cf. SECTION 4), of two interrelated assumptions: i) ii)

the wh-element, the clitic of negation (ne) and pa are operators; the relation between an operator in [Spec, CP] and the Vfnt in C° is such that nothing can interrupt it.14

The following objection can be raised regarding the first of these assumptions: while we can certainly assume that both wh-elements and negation are operators, the syntactic status of the adverb pa is uncertain. Stockwell (1977), in fact, assumes (on the basis of unpublished work by W. Rybarkiewicz; cf. Stockwell (1977:311, n.2)) that transitional adverbs like pa, ponne, paer, are to be considered conjunction elements together with the coordinating conjunction. Following this hypothesis, sentences introduced by pa should be more appropriately analyzed as verb-first sentences, like yes/no questions (cf. also Fourquet (1938)). Regarding the assumption in (ii), the idea that the relation between the whelement and the Vfnt is "special" in a certain way is immediately captured by the wh-criterion recently proposed by Rizzi (1990a). Following, in essence, May (1985) and updating his proposal in terms of Chomsky's (1986) theory of clausal projections, Rizzi assumes that the occurrence and position of whelements is determined by the following principles (cf. Rizzi (1990a:378): (38) Wh-Criterion: Principle A: Each [+wh] X° must be in a Specifier-Head relation with a wh-phrase Principle B: Each wh-phrase must be in a Specifier-Head relation with a [+wh] X° One of the main purposes which underlies the formulation of the wh-criterion consists in the possibility of accounting for what Rizzi calls "residual verb second." This label refers to the constructions which involve the mechanics of verb second (i.e. V° to I° to C°) in non-verb-second languages, in particular subject-aux inversion (SAI) in English and subject-clitic inversion in French (SCI). These two syntactic phenomena are exemplified in (39) and (40) below:

362

CLITICS AND VERB SECOND

(39) SAI a. Who i didj [Mary tj see ti ]? b. *WhOi [Mary I° saw ti ]?

(40) SCI a. Que manges-tu? b. *Que tu manges? Note that, while the adjacency requirement between the wh-word and the Vfnt is accounted for in terms of V° (to I°) to C° movement,15 on the other hand, the explanation for why the Vfnt must move to C° in a non-verb-second language crucially relies on the wh-criterion. In fact, if we assume (following Rizzi (1990a:378-379)) that the feature [+wh] may be generated: i) ii)

in C° in the subordinate clause (through selection by the matrix verb), in I° in the main clause,16

then it follows that I° (i.e. the Vfnt) must move to C in the main wh-clanse in order to satisfy principle B of the wh-criterion. The hypothesis that the movement of the Vtnt to C° in the main wh-clause is forced by the wh-criterion receives independent evidence from the diachronic perspective. If we consider the historical evolution of verb second, in fact, we do not find any violation of the verb-second constraint in wh-constructions. First of all it is well known that in OHG one possible exception to the verbsecond constraint in the main declarative clause consisted in the (strongly) limited occurrence of verb-last constructions.17 It conies as no surprise that no verb-last construction is attested in the main interrogative clause in OHG. Secondly, it is clear that the distributional facts presented by Kemenade (1987) with respect to the relative order pronoun-Vfnt in OE (here extended to OHG data) go exactly in this direction. The fact that the verb-second constraint seems to be "stronger" in the wh-construction should be, in fact, simply attributed to the wh-criterion, independently from whatever explanation one could provide for "full verb second." Turning now to our original problem, if what has been said so far proves to be reasonable, then in order to explain the word order pattern in (37) we have simply to assume the following: (41) The complex head which is derived from a process of cliticization (= [p clitic [r Vfnt]]) is unable to satisfy principle B of the wh-criterion. At this point it is important to note that: a) b)

The wh-criterion together with (41) cover not only the OE (and OHG) data under discussion (cf. (37)) but can be immediately extended to the phenomenon of subject-clitic inversion in French (cf. (40)); An important distinction must be drawn between subject clitics and object clitics. In fact, while the generalization captured in (35) does

CASES OF VERB THIRD IN OLD HIGH GERMAN

363

not imply any asymmetry between subject clitics and object clitics (as it seems to be the case in OE), in OHG and, more obviously, in modern French, the same generalization holds only for subject clitics. Compare (40) with the following two French examples: (42) Qui 1'a mangee? (Who it-has eaten = Who ate it?) (43) Quand 1'as-tu mangee? (When it-have-you eaten = When did you eat it?) As we can see, while the subject clitic must occur to the right of the finite verb (cf. tu in (43) and (40)), the object clitic (cf. /' in (42) and (43)) regularly occurs to the left of it.18 c)

the constraint (41) crucially implies that the process of (subject-) cliticization modifies the status of the head on which it applies.

From this last assumption at least two important related issues arise: I) II)

How does cliticization modify the status of the head? Which principle underlies the constraint proposed in (41)?

Concerning II), note that cliticization has been generally analyzed as a process of adjunction of a head (the clitic itself) to another head (I° or, more controversially, C°).19 Given this assumption, the interference caused by the clitic could be reduced to the notion of intervention as stated, for example, in Rizzi (1990b). In fact, the clitic adjoined to C° (either directly as proposed by Kemenade (1987), or indirectly through I° to C° movement as proposed here) interferes in the relation between the wh-element in [Spec, CP] and the Vfnt in C° in a way which is similar to the interference caused by a preposition intervening between a verb and its complements in a Case-assignment relation. This situation is illustrated in the following schema:

Turning now to I), one interesting solution would rely on a typology of possible cliticization processes (a goal pursued in recent work by Cecilia Poletto (1990a,b)). As Poletto (p.c.) pointed out to me, there are reasons to assume that the process of cliticization applies in at least two different ways:

364

CLITICS AND VERB SECOND

a)

cliticization through adjunction:

Given this distinction, it is clear that only cliticization through adjunction (cf. (45)) would be pertinent as far as the constraint (41) is concerned. On the contrary, the complex head resulting after cliticization to an X internal level (cf. (46)), should not cause any violation of principle B of the wh-criterion. Note that this distinction provides us with a principled explanation for the different (morpho-)syntactic behaviour between: i) ii)

cliticization phenomena in Romance languages versus cliticization phenomena in Germanic languages; cliticization of the subject pronoun versus cliticization of object pronouns.

A deeper investigation of the consequences of such speculative assumptions goes further beyond the purposes of the present paper and is therefore left to future work.

Notes *

This paper is a revised and expanded version of a talk presented at the First Generative Diachronic Syntax Conference in York (April 1990). A preliminary version of this work was submitted to the University of Geneva for the attainment of the Certificat de Specialisation en Linguistique (Theorie de la syntaxe et syntaxe comparative), academic year 1989/90, and appeared in GAGL 33 (1991). In addition to instructive discussion from the participants at the conference in York (particularly from Ans van Kemenade, Tony Kroch, Cecilia Poletto, Ian Roberts, Beatrice Santorini, Sten Vikner and Fred Weerman), comments and other help came from the following people: Werner Abraham, Harald Clahsen, Denis Delfitto, Giorgio Graffi, Maria Teresa Guasti, Lidia Lonzi, Andrea Moro, Luigi Rizzi and Raffaella Zanuttini. Of course, responsibility for errors is just mine.

CASES OF VERB THIRD IN OLD HIGH GERMAN

365

This work was partially financed by the research project 40% "Variazione e conservazione delle strutture sintattiche," headed by Prof. Riccardo Ambrosisi (Universita di Pisa) and Prof. Giorgio Graffi (local coordinator, Universita di Pavia). 1. For the relevance of the influence of Latin syntax in Tatian cf. the detailed study by Lippert (1974). 2. Note that in (la-c) the fronted constituent always corresponds to the object NP. This is due to chance; other XPs can appear here. 3. As Beatrice Santorini (p.c.) pointed out to me, (lc) is not the clearest example one could provide in order to differentiate this particular kind of verb-third construction from verb last, which was also possible in OHG. Since one has to assume, independently from the facts under consideration, that OHG syntax is characterized by a process of extraposition, then (1c) could be analyzed, in principle, as one of the possible variants which could be derived exactly from a verb-last construction through extraposition. A sentence with a complex verbal form (auxiliary/modal - past participle/infinitive) or a par ticle verb (e.g. liebhaben) would certainly provide more direct and convincing evidence in order to draw a precise distinction between (la-c) and the verb-last construction. 4. The so-called Monsee-Wiener Fragmenten present an ambiguous situation with respect to the position of the subject pronoun in the main declarative clause. If, on one hand, the subject pronoun generally follows the Vfnt as in Williram, on the other hand, some cases of verb third (XP-pronoun-Vfnt) are also attested. 5. As Ans van Kemenade (p.c.) pointed out to me, the word order pattern (9b), here repeated for the sake of simplicity: a) CompNP (subj) V fnt NP (0bj) V could ambiguously refer to two rather different syntactic constructions, which should be more precisely characterized by two distinct word order patterns: b) CompNP(Subj)XPVfntNP(0bj)V

c) Comp NP(Subj) *(XP) Vfnt NP(0bj) V Note that only the word order pattern in b), where a maximal projection XP intervenes between the NP(subj) and the Vfnt, can be unambiguously analyzed as an instance of VPR. On the other hand, the word order pattern in c) could be much more adequately analyzed as the result of V° movement to I° given the hypothesis of a deep structure word order of the type: d) [ I p NP ( S u b j ) [I°][ V P NP ( 0 b j ) V°]] where IP is a head-medial maximal projection. It is precisely this idea that we are going to explore in SECTION 4. 6. For a detailed investigation of the syntax of verbs in OHG, cf. Tomaseli (1989, ch. 2). 7. Muspilli shows a quite interesting situation. In fact we find an instance of VPR (just one example amongst 65 subordinate clauses) in a dialect which

366

CLITICS AND VERB SECOND

does not have simple VR. This is particularly interesting as far as the nature of VPR in OHG is concerned (cf. note 8). On the other hand, in Modern West Germanic languages the presence of VPR is strictly dependent on the presence of "simple" VR (cf. den Besten (1986)). 8. This hypothesis receives independent evidence from recent research by Ans van Kemenade on Old English (cf. Kemenade (1990)). 9. This ungrammatical word order pattern is, in fact, attested in at least one OHG text. The following example, taken from Muspilli, is the only example of complementizer-Vfnt adjacency usually cited in the historical grammars of German (cf. Erdmann (1985); Jolivet and Mosse" (1972)): a)

daz sculi der antichristo mit Eliase pagan CompVfnt NP(subj) PP V that must the antichrist with Elias fight Note that in the relative clause the possible adjacency of the relative pronoun and the Vfnt has a clearly different syntactic value. The following examples are taken from the Monsee-Wiener Fragmenten (M) and from Tatian (T): b)

(M):

der framtregit fona siemo horte niuuui ioh firni REL. Vfnt PP NP(0bj)

b')

(T): thie tha frambringit fon snemo treseuue nivvu inti altiu (Latin: qui profert de thesauro suo nova et vetera) who derives from his treasure the new and the old

c)

(M):

c')

(T): Inti souuer soquidit uuort uuidar then mannes sun (Latin: Et quicumque dixerit verbum contra filium hominis) and whoever says a word against the son of the man

Enti sohuuer soquuidit le uuort uuidar mannes sune Coor. REL. Vfnt O SP

Leaving aside the influence of the Latin syntax, it is clear that the adjacency of relative pronoun-Vfnt could not be stated in structural terms. Assuming that the relative pronoun occupies the specifier of CP, both the head of CP (C°) and the trace of the relative pronoun in subject position ([Spec.IP]) intervene between the relative pronoun and the finite verb. 10. Cf. the review by Allen (1990), where this objection against Kemenade (1987) is also raised. 11. According to Steele et al. (1981:285 ff.), the evolution of English went through the following stages: a. S O V Aux (Old English) b. S Aux O V (Early Modern English) c. S Aux V O (Modern English) Note that while, on one hand, there is a certain agreement an the relative chronology of these three stages (cf., amongst others, Roberts (1985) and, for independent evidence coming from a different field of research, Schwartz

CASES OF VERB THIRD IN OLD HIGH GERMAN

367

and Tomaselli (1988)), on the other hand, what is at stake here concerns the absolute chronology. In fact it is crucial for the purposes of our analysis to assume that OE was already characterized by a head-medial IP (cf. b). 12. Unless one wants to assume with Travis (1984) that Modern German is characterized by a head-medial IP as well. The fact that in recent works it has been convincingly shown that the system proposed by Travis is not an adequate description for Modern German syntax (cf. den Besten (1986); Tomaselli (1989); Schwartz and Vikner (1989)) does not compromise, however, the validity of her analysis for an older stage of the language (for related ideas concerning the historical development of French, cf. Roberts (1992)). Note, by the way, that the hypothesis that the history of German was characterized by an SIOV stage finds an interesting parallelism within the typological framework (cf. Lehmann (1971)). 13. The fact that I cannot represent the final landing site for verb movement (but just an intermediate step in V to C° movement) must be independently assumed for Modern Scandinavian languages (with the exception of Icelandic) (cf., amongst others, Platzack (1986)). 14. More precisely, Kemenade (1987:139-140) assumes the following: "if an operator with an index (wh-elements, Pa and ne) moves to Comp (= Spec, CP[AT]), it transmits this index to the head Infl (= C°, in a system where both CP and IP are distinct maximal projections [AT]) as in (44) (cf. a below [AT]): a. [infip Comp Oi Infl] > [InflP Comp Oi Infli] with respect to cliticization, (44) has the following effect: Comp and Infl behave as one constituent, so that cliticization is on the Infl projection rather than on Infl0." Apart from obvious terminological differences, Kemenade's basic idea is the following: when the specifier of CP is occupied by an operator, the specifier and the head of CP behave as one constituent preventing anything from intervening between them. 15. For a first formulation of SAI and SCI in terms of movement of the Vfnt to Comp, cf. den Besten (1983). 16. For an exhaustive and detailed explanation, cf. Rizzi (1990a). 17. This construction could be exemplified by the following word order pattern: XP YP (Z) Vfnt (Y) where both XP and YP are maximal projections. For a first analysis of this construction in OHG, cf. Lenerz (1985). 18. The situation represented by some Northern Italian Dialects (like Basso Polesano (cf. Poletto (1990a)) and Trentino (cf. Brandi and Cordin (1981) and Rizzi (1987)) seems to provide immediate support to this claim. A discussion of these data goes beyond the purposes of this paper. 19. Cf. Roberts (1991) and the literature cited there. 20. As for the postulation of X internal levels, cf. Selkirk (1982) and Roberts (1991).

368

CLITICS AND VERB SECOND

References Allen, C. (1990) "Review of A. van Kemenade (1987)." Language 66:146152. Behaghel, O. (1923-32) Deutsche Syntax. Eine geschichtliche Darstellung. Heidelberg: Carl Winter. Besten, H. den (1983) "On the Interaction of Root Transformations and Lexical Deletive Rules." In W. Abraham, ed. On the Formal Syntax of Westgermania, 47-131. Amsterdam: John Benjamins. Besten, H. den (1986) "Decidability in the Syntax of Verbs of (not necessarily) West-Germanic Languages." Groninger Arbeiten zur Germanistischen Linguistik 28:232-256. Brandi, L. and P. Cordin (1981) "Dialetti e italiano: un confronto sul parametro del soggetto nullo." Rivista di Grammatica Generativa 6:33-87. Braune, W and E.A. Ebbinghaus (1979) Althochdeutsches Lesebuch. Tubingen: Niemeyer. Chomsky, N. (1986) Barriers. Cambridge: MIT Press. Erdmann, O. (1985) Grundziige der deutschen Syntax. Ed. by W.A. Koch. Hildesheim: Olms. Fourquet, J. (1938) L'ordre des elements de la phrase en germanique ancien. Paris: Les Belles Lettres. Haegeman, L. (1991) Generative Syntax: Theory and Description. A Case Study from West Flemish. Oxford: Blackwell. Haegeman, L. and H. van Riemsdijk (1986) "Verb Projection Raising, Scope and the Typology of Rules Affecting Verbs." Linguistic Inquiry 17:417466. Hopper, P. (1975) The Syntax of the Simple Sentence in Proto-Germanic. The Hague/Paris: Mouton. Hulk, A. and A. van Kemenade (this volume) "Verb Second, Pro-drop, Functional Projections and Language Change." Jolivet, A. and F. Mosse (1972) Manuel de I'allemand du Moyen Age. Vol. 1. Paris: Aubier Montaigne. Kemenade, A. van (1987) Syntactic Case and Morphological Case in the History of English. Dordrecht: Foris. Kemenade, A. van (1990) "The Loss of Subject-Verb Inversion in Late Middle English." Paper given at Manchester University, April 1990. Lehmann, W.P. (1971) "On the Rise of SOV Patterns in New High German." In K.G. Schweisthal, ed. Grammatik, Kybernetik, Kommunikation. Festschrift fur A. Hoppe, 19-24. Bonn, Ferd: Dummlers Verlag. Lenerz, J. (1984) Syntaktischer Wandel und Grammatiktheorie. Eine Untersuchung an Beispielen aus der Sprachgeschichte des Deutschen. Tubingen: Niemeyer. Lenerz, J. (1985) "Diachronic Syntax: Verb Position and COMP in German." In J. Toman, ed. Studies in German Grammar, 103-132. Dordrecht: Foris. Lippert, J. (1974) Beitrage zu Technik und Syntax althochdeutscher Ubersertzungen. Miinchen: Wilhelm Fink.

CASES OF VERB THIRD IN OLD HIGH GERMAN

369

May, R. (1985) Logical Form. Cambridge, Mass.: MIT Press. Platzack, C. (1983) "Germanic Word Order and the COMP/INFL Parameter." Working Papers in Scandinavian Syntax, 2. University of Trondheim. Platzack, C. (1986) "COMP, INFL and Germanic Word Order." In L. Hellan and K. Koch Christensen, eds. Topics in Scandinavian Syntax, 185-234. Dordrecht: Reidel. Poletto, C. (1990a) "Subject Clitic/Verb Inversion in North East Italian Dialects." Paper presented at the Universtiy of Geneva (Seminaire Interdepartemental de Recherche en Linguistique). Poletto, C. (1990b) "Three Kinds of Subject Clitics in Basso Polesano and the Theory of pro " EUROTYP Working Papers 8.1. Reis, H. (1901) "Uber althochdeutsche Wortfolge." Zeitschrift fur deutsche Philologie 33. Rizzi, L. (1987) "Three Issues in Romance Dialectology." Paper presented at the 10th GLOW, Universita di Venezia. Rizzi, L. (1990a) "Speculations on Verb Second." In J. Mascaro and M. Nespor, eds. Grammar in Progress: GLOW Essays for Henk van Riemsdijk, 375385. Dordrecht: Foris. Rizzi, L. (1990b) Relativized Minimality. Cambridge, Mass.: MIT Press. Roberts, I. (1985) "Agreement Parameters and the Development of English Modal Auxiliaries." Natural Language and Linguistic Theory 3:21-58. Roberts, I. (1991) "Excorporation and Minimality." Linguistic Inquiry 22.1. Roberts, I. (1992) Verbs and Diachronic Syntax. Dordrecht: Kluwer. Schwartz, B.D. and A. Tomaselli (1988) "Some Implications from an Analysis of German Word Order." Ms. University of Geneva and Pavia. To appear in W. Abraham, E. Reuland and W. Kosmeijer, eds. Issues in Germanic Syntax (Proceedings of the Fifth Workshop on Comparative Germanic Syntax). Berlin: Mouton de Gruyter. Schwartz, B.D. and S. Vikner (1989) "All V-2 Clauses are CPs." Working Papers of Scandinavian Syntax 43:27-49. Selkirk, E. (1982) The Syntax of Words. Cambridge, Mass.: MIT Press. Steele, et al. (1981) An Encyclopedia of AUX. A Study of Cross Linguistic Equivalence. Cambridge, Mass.: MIT Press. Stockwell, R.P. (1977) "Motivations for Exbraciation in Old English." In C. Li, ed. Mechanisms of Syntactic Change, 291-314. Austin: Universtiy of Texas Press. Tomaselli, A. (1989) La sintassi del verbofinito nelle lingue germaniche. PhD Dissertation, University of Pavia. Tomaselli, A. (1990a) "COMP as a Licensing Head: an Argument Based on Cliticization." In J. Mascard and M. Nespor, eds. Grammar in Progress: GLOW Essays for Henk van Riemsdijk, 433-445. Dordrecht: Foris. Tomaselli, A. (1990b) "Review of F. Weerman (1989)." Journal of Linguistics 26.2 Travis, L. (1984) Parameters and Effects of Word Order Variation. PhD Dissertation, MIT. Weerman, F. (1989) The V2 Conspiracy: A Syncronic and Diachronic Analysis of Verbal Positions in Germanic Languages. Dordrecht: Foris.