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, pro-
drop, 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 Gen-
erative Diachronic Syntax Conference, held at the University of York in April
1990. The idea for that meeting was Adrian Battye's. Since then, the Con-
ference 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 re-
vision 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 mutu-
ally 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 pro-
vided 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, at-
tempting 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 intro-
duction; 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 de-
fence of this idea. Universal Grammar (UG) is taken to consist of an invari-
ant 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 prin-
ciples 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 lan-
guages 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 prin-
ciples 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 di-
rect object and verb is then a locus of parametric variation. The attested cross-
linguistic 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 ma-
terial on the right of the arrow in set notation so as to indicate that the rela-
tive 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 state-
ment given there is an instance of a more general schema for the internal struc-
ture 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 catego-
ries 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 involv-
ing heads of different categories are "harmonic" in the sense that heads con-
sistently 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 sim-
plified, is enough for our purposes here. We now have a notion of how cur-
rent 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 syntac-
tic 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 param-
eter change.
This approach has the immediate consequence that studying historically at-
tested (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 com-
binations 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 say-
ing that this language is both a verb-second language and a null-subject lan-
guage, 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 contem-
porary 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" to-
wards 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: phonol-
ogy, syntax, etc.? These are all traditional questions of diachronic linguis-
tics, 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 remain-
der 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 sys-
tem 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 acqui-
sition 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 asso-
ciated 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 sound-
meaning pairs" etc. that are inspired by work in analytic philosophy (see, for
example, Quine (1960); Montague (1974)). Also, most traditional work in his-
torical 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 E-
language is quite independent in principle of any properties that may be as-
cribed 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 per-
son who knows the language, acquired by the learner and used by the speaker/
hearer" (Chomsky (1986a:22)). Clearly, then, generative grammar is prima-
rily 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, pro-
ceeds on the basis of two things. On the one hand, there is Universal Gram-
mar "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 E-
language 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 Parents' E-language

Child's I-langua ge 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 be-
ing 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 I-
language representation for some E-language utterance U may differ from the
parents' I-language representation for U. Of course, the different representa-
tions must both fall within the confines defined by UG (unlike the case imag-
ined 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 cer-
tainly 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 lan-
guage acquisition, this partial overlap may in effect be imperceptibly differ-
ent 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 discus-

sion 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 analy-
sis and parameter setting on the part of the acquirer, or through some inher-
ent 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 Rob-
erts (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 "appropri-
ate 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 de-
velopment 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 acquisi-
tion 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 Prin-
ciple 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 lan-
guages and need positive evidence to set the null-subject parameter positively.
Something like the Subset Principle is clearly necessary, since negative evi-
dence 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 gram-
mars, 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 suffi-
ciently 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 prop-
erties of the acquirer and the process of acquisition. Of course, given the enor-
mous complexity of grammatical systems, it is frequently the case that other
things are not precisely equal. Nevertheless, the result is interesting, and in-
dicates 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 pro-
vide genuinely new insights into a range of both novel and traditional ques-
tions.
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 posi-
tive. 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 pa-
rameter 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 combi-
nations of parameter settings. However, at the level of individual construc-
tions 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 sim-
plify 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 lin-
guistics 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 sys-
tem of non-lexical categories carrying essentially "morphological" informa-
tion—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 gener-
ated 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 inde-
pendent 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 con-
tains 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 quantifi-
ers 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 quanti-
fiers 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 Struc-
ture 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 ini-
tial appearances, English also has this V-to-I movement rule, but its scope is
much more restricted than its French counterpart. In English, only auxilia-
ries, i.e. essentially aspectual have and be, can move to I; this is the have/be-
raising 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 9-
criterion, 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 rela-
tively "rich" agreement system. A number of the papers in this volume dis-
cuss 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 place-
ment of main-verb infinitives in relation to adverbs is more complex. Infini-
tives 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 domi-
nates 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 sys-
tem 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 com-
plex 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 ac-
cording 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 follow-
ing 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 Agr-
S—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 conse-
quence 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 nega-
tive 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 nega-
tion in English and French. This was NegP, the projection of a negative mor-
pheme. 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 negative-
polarity 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 infini-
tive 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 Pol-
lock, 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 con-
structions, i.e. when the verb moves over the subject (taking the usual posi-
tion 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 phe-
nomenon, found in German and other Germanic languages (but not contem-
porary 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 ele-
ment. 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 pro-
posed 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 analy-
sis, 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 po-
sition in verb-second languages as in related non-verb-second languages (al-
though 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 (forth-
coming)).
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 dif-
ferent 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. How-
ever, no further condition specifies the nature of that constituent. Roberts
suggests that the condition in (23) is related to the Extended Protection Prin-
ciple 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 sub-
jects 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 im-
plies 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 Gram-
mar 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 lan-
guage she is exposed to. A language will therefore change over time as
acquirers of that language come to set parameters differently (see the discus-
sion 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 param-
eter 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-Phenom-
enon in the History of Yiddish," also deals with the diachronic development
of verb second. Santorini shows that early Yiddish exhibits the same asym-
metry between root and embedded contexts as is to be found today in Ger-
man, 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 Nominative-
Case 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 "sym-
metric" 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-Asymmet-
ric Verb-Second Languages; the Case of Middle French" looks at the verb-
second 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-inter-
nal 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 ame-
nable to an analysis which has recourse to illocutionary elements in the sen-
tence.
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-sec-
ond configuration was lost from this language (with the exception of a cer-
tain residue). She bases her account on the proposals made in Roberts (1992),
which in turn derives from Koopmann and Sportiche (1991), that various para-
metric 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 inte-
grated into a typology of verb second in Medieval Romance.
Paul Kiparsky's paper "Indo-European Origins of Germanic Syntax" at-
tempts 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 lan-
guage, Proto-Germanic, and other branches of Indo-European represented by
Vedic Sanskrit, Hittite, and Homeric Greek. Evidence based on data concern-
ing verb movement, clitic movement, coordination and relative-clause forma-
tion 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 Medi-
eval 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 develop-
ment 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 con-
sequently available in Old and Middle French.
Vance's argument is that the availability of free inversion led to the weak-
ening 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 differ-
ence 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 unambigu-
ous indication of a "true" verb-second structure with the verb in C, while in-
version 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 evi-
dence 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 observa-
tion that scholars agree that the loss of verb second takes place around the
14th and 15th century in these languages, but different reasons, usually lan-
guage-specific ones, are put forward for this loss.
The leading idea in this paper is that the loss of verb second is to be corre-
lated 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 ini-
tially) are considered to have an I which is morphologically rich enough to
license referential null subjects, Old English does not possess such morpho-
logical richness in I and, therefore, only expletive null subjects can be identi-
fied. 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 mor-
phologically 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 morpho-
logical 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 (wh-
phrases) and those that appear with hybrid properties (allowing preverbal sub-
jects 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 intro-
duced 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 compo-
nents 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 (exple-
tive 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-sub-
ject system to that identified in Vigneulles' French by Hirchbiihler in his pa-
per 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 re-
cast 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 gram-
mar which makes use of functional projections and movement to C. The analy-
sis 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 verb-
second. Data drawn from Old French, Medieval Northern Italian dialects,
Medieval Southern Italian dialects, and Old and Modern (European) Portu-
guese 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 con-
figuration 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 pro-
posed, 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 me-
dial but the loss of the possibility of cliticisation to this node is to be attrib-
uted 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 lin-
guistics 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 Represen-
tations." 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 Cham-
pion [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. Ox-
ford: 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 Scandina-
vian 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 Trans-
formational 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." Linguis-
tic 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 Pro-
ceedings 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 Gram-
matical 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 lan-
guage 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 tantaliz-
ing 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 sug-
gests 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)CP[Spec
(1) c-tlVj+H IP[NP I'[VP[... ei ej] ] ] ]
Spec of CP may be filled by any phrasal category through what is often re-
ferred to misleadingly as a "topicalization" operation; this yields to the fa-
miliar 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 in-
flected 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 interest-
ing 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
'he is visiting The Hague'
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.
c. CP[Peteri [bezoekj+t]k IP[ei VP[den Haag ej] ek] ]
'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. CP [pourquoi[alli+ez]j IP [vous ej vp[ei a Paris] ] ]
why go you to Paris
'why are you going to Paris?'
 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 move-
ment 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" verb-
second languages; we shall return to this distinction later.
A more basic issue is: What exactly does the verb-second parameter con-
sist 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 plau-
sible 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 can-
not "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 indi-
rect 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 spe-
cific device which will block its derivation. This cannot be ruled out in prin-
ciple 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, nar-
rowly prescribed circumstances.
Nonetheless, one finds several proposals in the extensive verb-second lit-
erature 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 postu-
late 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-verb-
second 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 pro-
posals 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 struc-
tures 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 ex-
posed 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 analy-
sis. 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 underly-
ing 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 of-
fered 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 funda-
mental 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 lan-
guages 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 re-
quirement 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 com-
pares them with the "full verb-second" phenomenon which determines the
order of constituents in all main (and some embedded) clauses in Dutch, Ger-
man 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 prepo-
sitional 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 ac-
counts 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 dif-
ference 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 opera-
tor or trace in an A-bar chain and a [+I] head licenses a subject in the speci-
fier 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 for-
mation, was adopted in Lightfoot (1991, ch.3) and can be stated in terms of
constraints on phrase structure. Suppose that Dutch children learn that sen-
tences 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 par-
ticular 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 state-
ment 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 sec-
ond 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 verb-
second 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 trig-
ger 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 ac-
count was that the X-bar schemata of UG require that verbs occur alongside
their complements at D-structure; therefore Dutch children have ample evi-
dence 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). Fur-
thermore, 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 acquisi-
tional 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 con-
stituents; all verbal elements (including verbal complexes) occur
in first/second and final position with a preference for final po-
sition.
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 pro-
duced, 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 de-
spite 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 struc-
tures: 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 de-
gree-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 lan-
guages, 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 Scan-
dinavian 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 in-
vert (12a) or copying instead of substitution (12b) are not typical childhood
forms, whereas the equivalent forms are standard for English-speaking chil-
dren (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 ques-
tions: 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 at-

tain. This makes Weinberg's (1990) account interesting because she argues
that subject-auxiliary inversion involves marked and highly data-driven op-
erations, 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 ac-
quired 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 phe-
nomena have the virtue of making the presence of the finite verb in C condi-
tional 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 prepo-
sitional 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 trig-
ger 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 experi-
ence, they would also make the same predictions about the loss of full verb-
second 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 trig-
ger a different grammar, and, whatever the nature of the parameter, one would
expect the loss of verb-second to take place rapidly and catastrophically, re-
flecting 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 verb-
second 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 inver-
sion (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 sup-
porting 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 con-
structions 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 obliga-
tory, instances of verbs preceding subjects dropped out quickly.
On the first prediction, that the loss of verb second should have been pre-
ceded 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 be-
devilled 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 pre-
sumed 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) re-
ports 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 verb-
second mode, then even the later sections of the Chronicle remain well above
that threshhold. Recall, however, that Old English showed verb second obliga-
torily 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 dif-
ference 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 violat-
ing the Empty Category Principle (its trace would not be properly governed).
The major evidence offered by Emonds and others for this parametric distinc-
tion 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 auxil-
iary 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), signifi-
cantly 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 must have watched television on
Wednesdays .
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 de-
cline reached a certain critical point, there would be a parametric change re-
flecting 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 gov-
erned, 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, Ger-
man, 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 suffi-
ciently 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 inter-
rogatives 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 at-
tainability 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 lin-
guistic data taking place between 1400 and 1600, which might have the ef-
fect 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 in-
flectional 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 V-
to-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 ex-
tensively 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 agree-
ment cannot be a sufficient condition for absence of V-to-I despite sugges-
tions 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. Fur-
thermore, 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 in-
flection 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 En-
glish. 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 ("af-
fix 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 morpho-
logical 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) ap-
plies only where necessary, i.e. where there would otherwise be a stranded
affix. We know that "do support" was exercised increasingly during the rel-
evant 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 par-
ticular, 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 ear-
lier, which suggest that Dutch and German children acquire the I-to-C opera-
tion relatively early and without the kinds of errors made by English-speaking
children. The difference might arise through Rizzi's full vs residual distinc-
tion: 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 ad-
dressed 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. que manges-tu?
b. tu manges la poire?
c. tu manges quoi?
d. quand est-ce qu'il vient?
A common error, noted by Clark (1985) is failure to invert with subject pro-
nouns (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 pro-
noun (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 con-
firms 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 pro-
nominal, 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 English-
speaking 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 experi-
enced 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 rela-
tively 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 I-
to-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 op-
eration were triggered by precisely the positive data that the operation ac-
counted 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 ex-
perience which in turn entailed the new parameter setting. It may be true that
UG consists of principles and parameters, but an account of language acqui-
sition 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 prelimi-
nary 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). Sen-
tences 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] ] ]. Sen-
tences 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 con-
structions, 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 verb-
second 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 verb-
second 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 ver-
sion 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 subse-
quently 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 disap-
peared 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 Lexi-
cal 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 pre-
sented 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 18:613-
632.
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 Lan-
guages: A Description and a Tentative Explanation." In Fisiak, ed., 107-
135.
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., 49-
75.
Haider, H. and M. Prinzhorn, eds. (1986) Verb Second Phenomena in Ger-
manic 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 Syn-
tax. 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 News-
letter 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." Lan-
guage 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 Re-
view 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 Syn-
tactic 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 Subject-
Auxiliary 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 inver-
sion. The earliest and best-known generative analyses of the verb-second phe-
nomenon (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 re-
gard 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 ex-
hibit 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 be-
cause 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 sec-
ond during the medieval period but lost it (except for residual traces) in the
course of their history, Yiddish developed in the opposite direction, extend-
ing the domain of verb second to include subordinate clauses. This paper in-
vestigates 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 pre-

sented 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 as-
sumptions 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 lan-
guage move into a clause-medial Infl and that SpecIP is available for non-
subjects to move into. Further, I extend the analysis of verb movement in
asymmetrical verb-second languages put forward by Rizzi (1990b), accord-
ing 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-sec-
ond language as asymmetric or symmetrical depends (among other things) on
whether the highest [+I category in a clause is Comp or Infl. These assump-
tions form the basis for describing the transition of Yiddish from an asym-
metric 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 struc-
ture 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 Infl-
medial 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 Yid-
dish went from being Comp to being Infl. Evidence for this analysis comes
from the loss of empty categories and the concomitant rise of lexical exple-
tives and non-subjects in SpecIP. I conclude the paper by addressing the rela-
tion 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 asym-
metry illustrated in (1) and (2). The clause-initial constituent in a verb-sec-
ond 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-sec-
ond 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 symmetri-
cal 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 con-
tent of functional categories, I would like to propose an answer to this ques-
tion 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 fea-
ture 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 move-
ment 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 dis-
cussion 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 verb-
second language is open to parametric variation. In asymmetrical verb-sec-
ond languages like German, the highest [+1] category is Comp, whereas in
non-verb-second languages like English or in symmetrical verb-second lan-
guages 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 trigger-
ing verb movement, the highest [+1] category of a (finite) clause assigns nomi-
native 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 domi-
nates 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 neu-
tral 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 condi-
tion 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 Ital-
ian, 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 under-
lying 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] cat-
egory 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] cat-
egory is Infl, as is the case in modern Yiddish (and at least for some speak-
ers, 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 em-
bedded 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 condi-
tion 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 exhib-
ited synchronic variation between Infl-final and Infl-medial phrase structure.11
In this paper, I will therefore only briefly recapitulate the evidence that Infl-
final and Infl-medial phrase structure are both attested in early Yiddish, be-
fore 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

Highest [+I] Non-subjects

Type of Agr Category Languages in SpecIP?

Neutral Comp Dutch, Modern Mainland No

Scandinavian

Neutral Infl English, French, Italian No

Nominative Comp German, Early Yiddish No

Nominative Infl Modern Yiddish, Icelandic Yes

60 THE DIACHRONY OF VERB SECOND

2.1 Infl-Final Phrase Structure

Many early Yiddish subordinate clauses are not consistent with verb sec-
ond 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 superfi-
cial word orders in which the inflected verb is after second but before abso-
lute clause-final position. The treatment of such word orders as reflecting
Infl-final phrase structure is consistent not only with current generative as-
sumptions, but also with the traditional practice in German philology of tak-
ing 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 Yid-
dish, since the same generalizations obtain concerning the distribution of the
diagnostic elements as in the modern language. Thus, in addition to Infl-fi-
nal 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 Infl-
medial, since a verb projection analysis of them is inconsistent with language-
internal 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 Infl-
final 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 cor-
pus 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 nar-
rative 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 subordi-
nate 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 subordi-
nate 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 un-
ambiguously Infl-final subordinate clauses) and m (the number of unambigu-
ously 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 subor-
dinate clauses never exceeding 5%.16 After about 1700, Infl-medial subordi-
nate 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 lit-
erary and vernacular usage, I take this development to reflect a rise in Infl-
medial 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 ver-
nacular 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 struc-
turally 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 Fig-
ures 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. Ac-
cording 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 Writ-
ten 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 subordi-
nate clauses that are inconsistent with verb second because too many constitu-
ents 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 Infl-
medial phrase structure (so that the inflected verb can occur in second posi-
tion), 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 sub-
ject 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 con-
trast 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 his-
tory 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 vert gifindll eyn hurg fun di hrugi gzirh hn"l
which 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 Yid-
dish and the complete unacceptability of their counterparts in modern Yid-
dish 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 head-
governor 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 gap-
Vf 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 transi-
tion 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 lexi-
cal expletive is in complementary distribution with empty expletives: as shown
by the contrast between (19) and (20), it can appear in SpecIP but not VP-
internally.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 nominative-
case 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 nec-
essary consequence of the emergence of Infl-medial phrase structure. In or-
der 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 Infl-
medial verb-second languages without verb movement to Infl.like the mod-
ern 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 lan-
guages involves verb movement to a hybrid Comp. By contrast, in an Infl-
medial 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 Yid-
dish 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 chil-
dren 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 nomi-
native 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 Yid-
dish—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 exple-
tives 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 exhib-
ited some Infl-medial phrase structure, but did not in general exhibit embed-
ded 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 embed-
ded 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 Infl-

medial 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-recur-
sion 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 pre-
cisely the pattern that we would expect if Comp were hybrid for some speak-
ers of Icelandic and Faroese.23
While there is evidence, then, for the possibility of a parametric constella-
tion 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 that-
clauses with an empty expletive in first position. At least for the Icelandic
case, these frequencies are comparable to the frequencies with which perfor-
mance errors occur in Modern English (Anthony Kroch, p.c.) and Early Yid-
dish (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 in-
flected verb)
2. unambiguously Infl-medial (instance of stranding or apparent verb pro-
jection 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 verb-
second constraint is the sum of (3), (5) and (6), and the total number of sub-
ordinate 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

(1) (2) (3) (4) (5) (6)

Court testimony, 1440-1489 80 1 (1%) 6 KO) 0 0
Total: 1440-1489 80 1 (1%) 6 1 (0) 0 0
Court testimony, 1540-1589 20 0 (0%) 10 0(0) 0 0
Total: 1540-1589 20 0 (0%) 10 0(0) 0 0
Court testimony, 1590-1639 1 0 (0%) 3 0(0) 0 0
Letters, 1619 19 0 (0%) 3 1(0) 0 0
Total: 1590-1639 20 0 (0%) 6 1(0) 0 0
Court testimony, 1640-1689 67 4 (6%) 39 1(0) 0 0
Total: 1640-1689 67 4 (6%) 39 1(0) 0 0
West Yiddish Literary Texts
(1) (2) (3) (4) (5) (6)
Mints, third quarter of 1400s 3 1 (25%) 3 0(0) 0 0
Total: 1440-1489 3 1 (25%) 3 0(0) 0 0
Bovo, 1507 73 3 (4%) 21 4(0) 0 0
Goetz, 1518 15 0 (0%) 4 0(0) 0 0
Anshel, ca. 1534 32 1 (3%) 6 0(0) 0 0
Total: 1490-1539 120 4 (3%) 31 4(0) 0 0
Preface, Shir ha-shirim, 1579 35 7 (17%) 33 3(0) 0 0
Shir ha-shirim, 1579 77 4 (5%) 54 1(0) 0 0
Officials, 1588 10 0 (0%) 7 0(0) 0 0
Preface, Megilat Ester, 1589 9 3 (25%) 12 0(0) 0 0
Megilat Ester, 1589 68 1 (1%) 33 1(0) 0 0
Total: 1540-1589 199 15 (7%) 139 5(0) 0 0
Kine, 1648 22 1 (4%) 14 6(1) 0 0
Messiah, 1666 32 1 (3%) 19 0(0) 0 0
Witzenhausen, 1677 4 0 (0%) 4 0(0) 0 0
Total: 1640-1689 58 2 (3%) 37 6(1) 0 0
 TWO TYPES OF VERB SECOND IN THE HISTORY OF YIDDISH 71

(1) (2) (3) (4) (5) (6)

Purim-shpil, 1697 66 13 (16%) 70 4(0) 0 3
Total: 1690-1739 66 13 (16%) 70 4(0) 0 3
Zeeb, 1740 4 2 (33%) 6 1(0) 0 0
Moses, ca. 1750 22 9 (29%) 44 0(0) 0 0
Total: 1740-1789 26 11 (30%) 50 1(0) 0 0
East Yiddish Vernacular Texts
(1) (2) (3) (4) (5) (6)
Court testimony, 1540-1589 6 2 (25%) 8 1(0) 0 0
Total, 1540-1589 6 2 (25%) 8 1(0) 0 0
Court testimony, 1590-1639 25 11 (31%) 64 3d) 0 0
Total, 1590-1639 25 11 (31%) 64 3(1) 0 0
Court testimony, 1640-1689 10 8 (44%) 46 1 (1) 1 2
Total, 1640-1689 10 8 (44%) 46 1(1) 1 2
East Yiddish Literary Texts
(1) (2) (3) (4) (5) (6)
Meineket Rivkah, ca. 1550 1 1 (50%) 1 0(0) 0 0
Total: 1540-1589 1 1 (50%) 1 0(0) 0 0
Sam Hayyim, 1590 35 1 (3%) 15 0(0) 0 0
Preface, Lev Tov, 1620 32 3 (9%) 14 3(0) 0 0
Lev Tov, 1620 74 3 (4%) 42 4(0) 0 0
Preface, Ha-magid, 1623-1627 25 3 (11%) 11 3(0) 1 0
Hagen Abraham, 1624 32 1 (3%) 13 3(1) 2 0
Total: 1590-1639 198 11 (5%) 95 13(1) 3 0
Vaad, 1671 10 1 (9%) 9 0(0) 0 0
Ashkenaz un Polak, ca. 1675 38 5 (12%) 18 2(1) 2 0
Total: 1640-1689 48 6 (11%) 27 2(1) 2 0
Vilna, 1692 3 16 (84%) 31 0(0) 0 5
Sarah, first half of 1700s 0 5(100%) 16 0(0) 0 0
Ellush, 1704 1 8 (89%) 15 0(0) 0 1
Duties, 1716 3 8 (73%) 13 1(1) 0 0
Poznan, 1717 1 4 (80%) 9 0(0) 0 0
Total: 1690-1739 8 41 (84%) 84 1(1) 0 6
Nakhman, ca. 1800 0 31(100%) 50 0(0) 5 3
Naphthali, 1803 3 32 (91%) 61 2(0) 0 5
Geography, 1818 7 20 (74%) 52 0(0) 4 9
Ukraine, 1834 1 34 (97%) 93 2(0) 1 9
El Male Rakhamim, 1834 0 11(100%) 31 1(0) 2 2
Total: 1790-1839 11 128 (92%) 287 5(0) 12 28
Judah, 1848 0 27(100%) 64 0(0) 0 22
Grine Felder, ca. 1910 0 42(100%) 91 0(0) 4 8
Royte Pomerantsen, 1947 0 52(100%) 116 0(0) 4 16
Total: 1840-present 0 121(100%) 271 0(0) 8 46
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. Need-
less 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 head-
initial. 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 ar-
eas in contact with Slavic, and East Yiddish was originally spoken on Slavic-
speaking 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 mid-
1700s. 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 verb-
second 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 sug-
gested 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 pur-
poses of this paper, the precise position in which subjects originate is irrel-
evant 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 opera-
tor [+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: ac-
cording 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-com-
mand 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 sec-
ond in a language like Italian. The unacceptability of embedded verb second
in Italian is a serious difficulty for their approach and supports the case-mark-
ing 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 un-
known, 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 much-
studied phenomenon in West Germanic. Verb raising permutes the order of
auxiliary verbs and the heads of their infinitival complements. In verb projec-
tion 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 excep-
tional texts is comparable to East Yiddish vernacular usage of the same pe-
riod. 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 plau-
sible explanation for the unusually high incidence of Infl-medial phrase struc-
ture 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 excep-
tional 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 con-
stituent occupies clause-initial position, resulting in word orders that are su-
perficially 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. Pro-
ceedings 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 Lexi-
cal Deletive Rules." In W. Abraham, ed. On the Formal Syntax of the
Westgermania, 47-131. Amsterdam: John Benjamins. Draft version cir-
culated in 1977.
Besten, H. den and J.A. Edmondson (1983) "The Verbal Complex in Conti-
nental 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 Phenom-
ena in Germanic Languages. Publications in Language Sciences 21:111-
135. 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 Pro-
cesses: 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, 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.
Deprez, V. (1989) On the Typology of Syntactic Positions and the Nature of
Chains: Move a to the Specifier of Functional Projections. PhD Dis-
sertation, 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 Univer-
sity 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 His-
tory 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 8:128-
172.
GeilfuB, J. (1991) "Jiddisch als SOV-Sprache." In Verb- und Verbphrasen-
syntax, 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 Ger-
manic 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 Linguis-
tic 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 Scandina-
vian 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 Scan-
dinavian Syntax 48:1-43.
Kemenade, A. van (1987) Syntactic Case and Morphological Case in the His-
tory of English. Dordrecht: Foris.
Kitagawa, Y. (1986) Subject in Japanese and English. PhD Dissertation, Uni-
versity 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 Gen-
eralized COMP Analysis." In H.A. Sigur5sson, ed. Papers from the 12th
Scandinavian Conference of Linguistics, 182-193. Reykjavik: Linguis-
tic 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. Pro-
ceedings 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 24:71-
91. 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 Pennsyl-
vania.
Platzack, C. (1986) "The Position of the Finite Verb in Swedish." In H. Haider
and M. Prinzhorn, eds. Verb Second Phenomena in Germanic Lan-
guages. 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. Cam-
bridge: Cambridge University Press.
Sigur5sson, H.A. (1990) "Verb-Initial Declaratives and Verb Raising in Ice-
landic." 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, Pub-
lications 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. Cam-
bridge: 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 con-
straint 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 trig-
gering 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 inter-
esting 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 theo-
retical assumptions concerning the function of some illocutionary morphemes
involved in verb second.
The discussion will proceed as follows: SECTION 2 presents the data illus-
trating 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 Yid-

dish 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 Nomi-
native-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 char-
acteristics 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, (...) Berinus, I, p. 240
Then looked (he) down at the sea
'Then he looked down at the sea'
Postverbal full NP
(2) Longtemps fu ly roys Ellnas en la montaigne (...)
Melusine, p. 14.
For a long time was King Elinas 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 de-

rived 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 ex-
amples (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 cen-
turies. 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
Texts XPV Nom. Subj. V Pron. Subj. V Subtotal % Total

Berinus 189 40 107 336 91 371

Melusine 211 82 126 419 93 452

Policie 148 47 71 266 83 319

QJM 113 57 175 345 85 404

CNNA 152 64 130 346 87 397

Memoires 214 90 73 377 91 416

TOTAL 1027 380 682 2089 89 2359

 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 gener-
ated 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 verb-
second 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 gram-
mar 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 lan-
guages 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 deriva-
tion 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 char-

acteristics, this will motivate an alternative solution, especially if these argu-
ments 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. Some of the verb-second languages obey the so-called root/embed-
ded 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)).
B. Some verb-second languages are strictly constrained as to the possi-
bility of dropping the subject: this is the case with German, a semi-
pro-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 subordi-
nate 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 embed-
ded clause as is the case in the following sentences where vayn 'wine', shabes
bay nakht 'Saturday night' and hayntike tsaytn nowadays' trigger the verb-
second effect.
(8) a. Ir zolt visn zayn, mayne libe kinderlekh, az vayn
You (pi) should know be my dear children that wine
ken men makhn fun troybn oykn.
can one make from 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 distribu-
tion 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 respon-
sible 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 observa-
tions as those given for Yiddish, Rognvaldsson and Thrainsson (1990) (here-
after 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 follow-
ing sections, let us adopt R&T and Diesing's analysis without further elabo-
ration. 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 pro-
vided 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, be-
ing 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 [IP ce soit t; ]]
Queste, 112,1
because enemies believes he well that there is
'because he believes that there are enemies'
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 rel-
evance 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 struc-
ture 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 in-
finitival 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 que-
clause. The French subjunctive mood expressed by the verb in embedded
clauses is strictly dependent upon the higher predicate. Obviously, there ex-
ists a strong dependence relation between both clauses such that the embed-
ded 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 embed-
ded 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 re-
sult of a single verb movement to Intl.
Let us summarize the discussion so far. Extraction processes and depen-
dency 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 ne-
gation 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 Policie, p. 17
But now is not such rule respected
'But now such a rule is not respected'
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 posi-
tion of the following negative element. But the same derivation would obvi-
ously yield the wrong result for MidF.
The order of the negation in (18) is explained straightforwardly if we as-
sume 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 articu-
lated structure of inflectional projection proposed in Pollock (1989). Accord-
ing 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 Con-
straint.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 require-
ment 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 desa 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 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 be-
tween 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 show-
ing 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 struc-
tural relation. Bearing this in mind, let us explain how the subject in TP, pro
or an overt pronominal, can trigger agreement while being assigned Nomina-
tive Case. Our analysis is based on ideas developed in Rizzi (1990) follow-
ing 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 assign-
ment to the VP-internal subject follows from the Principle of Full Inter-
pretation 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 in-
troduces a dichotomy between Case and agreement and does not account for
the strict locality requirement necessary if one maintains that "rich" agree-
ment 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 gov-
ernment 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 un-
der 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 exam-
ined so far. It was shown that subordinate clauses tend to pattern with paral-
lel 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 characteris-
tics 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 em-
bedded clauses, which we call illocutionary factors, are predicted by our analy-
sis. 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 ac-
count; 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 asymme-
try 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 fig-
ures we find in Table 4.2. This table shows that there is much less verb-sec-
ond 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

Texts Null Subject Postverbal Subject Subtotal Total

# # # #

Root Clauses 645 382 1027 2385

Embedded Clauses 106 26 132 2819

TOTAL 751 408 1159 5204

 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 B6rinus, 1,4
Because truly it have (Is SUBJ) deserved
'Because I have truly deserved it'.
Sentential adverbials like si in (25) must be generated in some Spec posi-
tion 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 func-
tion like adverbials as far as verb-second is concerned, but whose categorial
status is not clearly established:
(26) a. Or avez vous ouy du roy Elinas etde 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 n' estoit nul qui 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 verb-
second constructions do not necessarily constitute a homogeneous class; in the
following discussion, we will distinguish the si-class, which is clearly an ad-
verbial 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 sum-
marize 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 sug-
gestion inspired by Banfield (1982).8 We shall pursue this idea that there is
something distinct about root narrative clauses, but without recourse to V-to-
C movement.
In his discussion of emphatic contexts with imperative do in English, Pol-
lock (1989) suggests, as an alternative to NegP, that in such cases, do occu-
pies 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 Laka (1991)
to nobody I will ask forgiveness
 THE LOCUS OF VERB MOVEMENT 97

Spec.ZP is also available in emphatic fronting contexts, i.e. in contexts con-

taining 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 op-
erator like Assertion which identifies si in [Spec.AgrP] as a legitimate asser-
tive 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 Melusine, p. 6
And the other two to answer
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 verb-
second 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 simi-
lar 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 Podsies, p. 3
'Would you please leave him in peace'.
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 posi-
tion.11 The preverbal position is filled by an operator in the Spec which is
licensed by discourse conditions. We argue that in such superficially verb-
first 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 sen-
tence 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 M61usine, p. 29
So it is really (the) truth
b. mais avant vous 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.
 THE LOCUS OF VERB MOVEMENT 99

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

Texts "et" "si" Other Discourse Total Discourse Total XPVs

XPs XPs
Berinus 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

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 pro-
posed 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
Texts "et" "si" Other Discourse Total Discourse Total XPVs
XPs XPs
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
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 verb-
second 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 re-
flex 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 inter-
pretation 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 sen-
tences like (34) as violations of the verb-second constraint, a V-to-Agr ap-
proach has nothing special to say about these examples.
In fact, a large proportion of the cases involving discursive elements dis-
cussed in SECTION 4.1 constitute verb-second violations for any theory. Nev-
ertheless, 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 V-
to-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 ex-
ample, certes 'certainly', and premierement 'firstly' are realized in the
[Spec.ZP] position. A V-to-C analysis would have to assume that they are ad-
joined 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 ad-
verbial 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 avail-
able because the pronoun is on the verb. In her analysis, such apparent verb-
second 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 Nominals Pronominals
V2 V>2 % Total V2 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

Total 380 122 24 502 682 148 18 830

Figure 4.5 following, shows that violations with both NP subjects and with
pronominal ones represent a stable phenomenon that is attested over two cen-
turies. 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 viola-
tions, 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
and also they know perfectly laugh between
themselves
'and so they perfecly know to laugh about it between themselves'
b. briefment il court par la maison QJM, p. 52
quickly he runs through the house
(38) a. Et auxi toutes les aultres sont en la buee QJM, p. 53
'And also all the other ones are in the laundry'
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 verb-
second is possible both in root and in embedded clauses. Our proposal en-
tails 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 mani-
festations of the verb-second constraint in Modern French must take into ac-
count 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 analy-
sis 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 in-
volve 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 non-
asymmetric 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. 410-
89-1409 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 exple-
tive 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 Nomi-
native Case is checked.
6. We borrow from Vance (1989:88-89) the following summary of these de-
bates: "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 ad-
verb, acting as the intial constituent in what is really an CVS structure. Oth-
ers (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 neces-
sary 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 predica-
tion, the one called "a second level predication" by Marchello-Nizia: as it
has been demonstrated by this author, si expresses just such a relation be-
tween 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]. How-
ever, 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 solu-
tion would have the advantage of giving a unified analysis of et in coordina-
tion 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-sec-
ond 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, Berinus, I, p. 15
'Surely, she was telling the truth'
(ii) Premierement je vueil respondre... Berinus, I, p. 90
'First I want to answer...'
(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 reis erranment 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) 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 mul-
tiple 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. Pro-
ceedings 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 Chang-
ing. 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 3:209-
274.
Belletti, A. (1989) Agreement and Case in Past Participial Clauses in Ital-
ian. Ms. Universite de Geneve.
Bennis, H. and L. Haegeman (1983) "On the Status of Agreement and Rela-
tive 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 Lexi-
cal 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 Publish-
ers.
Chomsky, N. (1991) "Some Notes on the Economy of Derivations and Rep-
resentations." In R. Friedin, ed. Principles and Parameters in Compara-
tive 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 Dis-
sertation, 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." Re-
vue 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, Func-
tional 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. Uni-
versity 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 Inter-
national Workshop on Japanese Syntax, 103-143. Stanford: CSLI.
Laka, I. (1991) "Negative Fronting in Romance: Movement to E." Paper pre-
sented 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 psycho-
systematique 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 Prin-
ciple." Ms. MIT.
 THE LOCUS OF VERB MOVEMENT 109

Pollock, J.-Y. (1989) "Verb Movement, UG and the Structure of IP." Linguis-
tic Inquiry 20:365-424.
Rizzi, L. (1990) Relativized Minimality. Cambridge, Mass.: MIT Press.
Rizzi, L. and I. Roberts (1989) "Complex Inversion in French." Probus 1:1-
30.
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. Syn-
tax 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 Se-
mantics." 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 Analy-
sis 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 sen-
tence [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 verb-
second structure. Rizzi (1991) calls "residual verb-second" the Wh-construc-
tions which show subject-auxiliary inversion in English, and verb-subject clitic
inversion in French. In those languages there is no generalization of the verb-
second order to main declarative clauses. On the other hand, "full verb-sec-
ond" identifies the phenomenon that determines the order of the constituents
in all matrix clauses and in some embedded ones in languages such as Ger-
man, Dutch and Scandinavian. The distinction between the two types of phe-
nomenon 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 verb-
second 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 111

b. Que manges-tu?
What do you eat?
c. Whom did Mary see?
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 in-
version 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. A qui tu as parle?
b. *Que tu manges?
c. *Who Mary saw t?
d. Who t saw Mary?
e. *Who did t see Mary?
f. Para quando os diretores querem esses relat6rios?
g. 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 sen-
tence (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 com-
mon in contemporary spoken BP.3 English, on the other hand, presents a pe-
culiar 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) stud-
ies 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 Wh-
interrogatives 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 occur-
rence of SV order (especially in 1882 and 1918) is observed, but a clear pref-
erence 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 com-
pare 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 pro-
posal for analysing the loss of verb-second in a diachronic perspective, fol-
lowing 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 show-
ing verb-final basic word order (SOV) such as German, for instance, the in-
flected 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 oth-
ers) 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 con-
structions 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 sub-
ject 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 sur-
face 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 ad-
verbial 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) identi-
fies 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 verb-
second 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 verb-
second 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 vir-
tually 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 con-
nect 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 construc-
tions 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 foron- se 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 pres-
ence 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 consid-
ered 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 fre-
quent 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 verb-
second contexts.
A strict correlation between null subject and context of inversion verb-
second 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 assign-
ment: 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 lan-
guage where Nominative is assigned only under government, pro may be li-
censed 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 licens-
ing 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 compari-
son 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 assign-
ment, i.e., in a context either under agreement or under government by Agr°.
However, when pro is licensed under agreement, rich inflectional verbal mor-
phology 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 undergo-
ing erosion. However, the 14th-century OP Agr-system contains person/num-
ber 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 agree-
ment 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 occu-
pied 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 em-
bedded 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-sec-
ond, 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 posi-
tion Spec/C (in a structure involving CP-recursion, under the analysis pro-
posed 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 symme-
try 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-cen-
tury 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 ex-
amples 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 Icelan-
dic, Yiddish, Medieval English and in OF). This rule fronts some VP con-
stituent (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 sen-
tences in Icelandic and OF. Considering Padua's (1960) remark that relative
structures as those exemplified in (15) (with verb-final order and omitted sub-
ject) are very common in OP, we may conclude that Stylistic Fronting con-
structions 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 struc-
ture 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 construc-
tions 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" sub-
ject 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, con-
ditionals and in imperatives. Hirschbiihler and Junker identify verb-first con-
structions 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 cors deporter
(Le Charroi de Nimes 1.26)
Hold (they) bird 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 struc-
tures. 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 sentence-
initial 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 con-
struction:
(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 op-
erator in Spec/C', these structures can be analysed as involving a Spec/C' po-
sition 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 struc-
ture 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 preposi-
tional 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 co-
occurs with CVS word order in matrix and embedded clauses. Verb-
subcategorized 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 cen-
tury, required the CVS order. In this construction, C is the equivalent of ver-
bal 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 con-
stituent subcategorized by the verb. Certain adverbial elements did not trig-
ger 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 struc-
ture 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 verb-
second 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. How-
ever, 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 repre-
senting 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 VP-
complement. 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 pos-
sibility 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 sys-
tems 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 verb-
first 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 him-
self 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 tech-
nical 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 consid-
ered.
In completive structures in DSG, the complement clitic pronoun system-
atically occurs adjacent to C°. Different types of elements can intervene be-
tween 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 com-
plementizer 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 hav-
ing 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 char-
acterized 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 pre-
positional 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 char-

acterized 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 verb-
second 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 struc-
tures). Stylistic Fronting clauses are found in Wh-complements (in AgrP-
recursion structures). Taking the pro licensing condition into consideration we
propose that in OP pro was licensed in pre/postverbal position, i.e., in con-
texts either of government under Agr° or under Spec/head agreement (agree-
ment 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 di-
rective and discursive constructions. We also observed that pro can be pre-
verbal 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 sug-
gested 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 verb-
second string, and finally that, even in direct interrogative constructions, verb-
second 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 rela-
tion to the element which triggered the change.
The Principles and Parameters model considers linguistic change as a modi-
fication 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 eliminat-
ing some of the properties related to the abandoned value, and by confining
some of them as relics on special stylistic registers or as crystallized construc-
tion 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 lan-
guages at former stages, but which is now confined to a few marked construc-
tions. 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 gram-
mar (grammar 1 = Gl) are totally discontinuous:
(33) Primary data 1 Grammar 1
Primary data 2 Grammar 2
Because the surface structures may be compatible with more than one gram-
mar 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 struc-
ture) as a basic SVO one. Having also lost the initial accent, typical of verb-
second 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 condi-
tion 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, chil-
dren 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 re-
analysis 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 comple-
ment. 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 assign-
ment. He assumes that the choice between one or the other option (govern-
ment-only or agreement-only) or of both options (government and agreement)
for Nominative assignment shown in the configurations in (34) is a paramet-
ric one. He exemplifies the application of this system of Nominative assign-
ment 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 inver-
sion 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 hy-
pothesis.
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, un-
der 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 to-
gether 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 as-
signs 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 ex-
plained in these terms. We saw in Duarte's (1989) work that direct Wh-ques-
tions 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. Con-
structions 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 Wh-
question 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 con-
structions 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 gov-
ernment; 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 verb-
second 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 as-
signed 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 infor-
mally 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 agree-
ment, and to analyse the XPVS evidence as marked structures involving Nomi-
native 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-assign-
ment 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 configura-

tions. We have put forward a preliminary analysis of OP Nominative-assign-
ment 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 fea-
ture [+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 verb-
second completives in verb-second Germanic languages and in OF. Verb-sec-
ond is also possible in causal and conditional clauses, as in examples (16b)
and (16a), respectively. Stylistic Fronting is possible in relative clauses (ex-
amples in (15)).
In terms of current proposals, the inflected verb cannot move to an embed-
ded [+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 ques-
tions, 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 agree-
ment. 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?'
(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.
4. 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)).
5. Mattos e Silva's work (1971) focuses on the analysis of a Portuguese docu-
ment, 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.
6. 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).
7. 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 in-
flected 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 lan-
guage 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 Rob-
erts' 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 19:1-
34.
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 Lexi-
cal Deletive Rules." In Abraham, ed., 47-131.
den Besten, H. and C. Moed van Walraven (1986) "The Syntax of Verb in Yid-
dish." 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 Pro-
cesses: 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 Represen-
tations." 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., 49-
75.
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 Rel-
evance 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. Pro-
ceedings 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: Cam-
bridge 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 Seman-
tics 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." Linguis-
tic 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 21:1-
30.
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 (1538-
13). Comissao do IV Centenario da Cidade de Sao Paulo.
Taraldsen, K.T. (1986) "On Verb Second and the Functional Content of Syn-
tactic 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." Lin-
guistic Inquiry 15:103-129.
Travis, L. (1984) Parameters and Effects of Word Order Variation. PhD Dis-
sertation, 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 wh-
phrases, 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 left-
peripheral 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 con-
stituents 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 ar-
guments and modifiers (Kayne (1982); Taraldsen (1986); and Holmberg
(1986)). This restructuring in turn led to several major syntactic characteris-
tics of Germanic, including the rise of V-to-C° movement, triggered by wh-
phrases and other focused elements in Spec-C. "Residual verb second" (Rizzi
(1990)) is then the original core of the verb-second system. The modern Ger-
manic 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 Ger-
manic 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 expla-
nation 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 fol-
lows. 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 lexi-
cal complementizer, blocking V-to-C° movement.5 Subordinate clauses lack-
ing 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° move-
ment, 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 speci-
fiers. 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. Verb-
second 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 Ice-
landic: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 lan-
guages, and its apparent optionality in Old English is attributable to the para-
metrically 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 Top-
ics must bind a correlative resumptive pronoun in the clause and that wh-
phrases and other elements triggering Subject-Aux Inversion move into
Spec-C:
(8) a. Beans I like. (focused NP in Spec-C)
b. Beans, I like them. (Topic adjoined to CP)
c. *Beans who needs? (focusing blocked by wh-phrase in
Spec-C)
144 THE DIACHRONY OF VERB SECOND

d. Beans, who needs them? (Topic adjoined to CP)

e. *Beans not once did I eat (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. *This shelf, what should we put on?
b. This shelf, what should we put on it?
c. On this shelf, what should we put?
d. This room, never once did John sleep in.
e. This room, never once did John sleep in it.
f. 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 previ-
ously 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 Swed-
ish, 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, Spec-
C 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 Spec-
C 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 differ-
ence 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 as-
sumptions. 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 re-
side in Spec-C, sentences containing them must be CPs, therefore must con-
tain 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 anal-
ogy of Dutch and German (e.g. Kemenade (1987)), these facts are mysteri-
ous. 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 ini-
tial 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 land-
ing 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) accord-
ing to which the placement of clitics arises from the interaction of indepen-
dent 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 sat-
isfaction of the phonological requirement, the clitic is moved by "prosodic in-
version" 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 illus-
trated 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 S-
initial, 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 require-
ment 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 Spec-
C and that V-to-C° is obligatory.
Our analysis is again confirmed by the fact that the focused element in Spec-
C 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 dif-
ferent 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 win-
ter 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 verb-
final 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 ex-
plained. 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 some-
times taken to mean that those conjuncts are subordinate clauses in some
sense—as if, mysteriously, "even co-ordinating conjunctions are subordinat-
ing," 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 syn-
tactic and semantic respect they are like main clauses and unlike subordinate
clauses. For example, Old English allows certain long-distance wh-dependen-
cies 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 subordinate-
clause 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 Spec-
C and C° may be omitted in non-initial conjuncts.
This approach gains support if we examine the conditions under which con-
joined 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° move-
ment if and only if they begin with a lexically unfilled complementizer posi-
tion. Since the second conjuncts in (22) have overt wh-elements, they
necessarily have a C° head, and therefore show obligatory verb-second posi-
tion.
Now we also have the explanation for why the other Germanic verb-sec-
ond 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 op-
tional. 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 gen-
eral, 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 Ger-
manic 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 sug-
gest 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 follow-
ing:24
Preposition stranding is obligatory in bare-complementizer relative clauses,
but does not occur in pronominal relative clauses (Mitchell (1965, §2231-
2248)). This is especially significant in view of the status of preposition strand-
ing 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 pro-
pose 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 con-
trolled 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 confu-
sion 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 his-
torically 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-fi-
nal word order abound on every page of Old English. The reason is that V-to-
C° 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 sen-
tence-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 com-
peting 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. Tak-
ing 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. S O Vi M
b. S Vi Vf O
c. SOVfV
d. S Vf O V
e. 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 deri-
vations are obvious. The underlying order is S O Vinf Vf. Of the two main-
clause 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 subor-
dinate 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 his-
torically 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 par-
ticular 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 ar-
gument 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 constitu-
ency 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 re-
spectively 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 sys-
tem, are syntactically affiliated with S", and pronominal clitics being argu-
ments, 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 con-
stituent 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 155

b. 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 More-
over, Hale argues that Homeric Greek clitics are positioned by the same prin-
ciples, 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 posi-
tions 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 re-
maining 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 argu-
ment 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 argu-
ment 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 cor-
relative. 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-Euro-
pean 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 gen-
eral 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 Brah-
manaspati 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 indepen-
dently 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 Ger-
man 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 cun-
ning 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 embed-
ded 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 (gener-
ally 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 trans-
parently 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 Indo-
European, the most popular being cliticization and verb-topicalization. Theo-
ries of the second type view verb second as arising entirely within Germanic
in consequence of other changes that took place within that family. An ex-
ample would be the proposal of Vennemann (1975) that verb second is func-
tionally 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 V-
to-C° to be substantially inherited from Indo-European, but with V-to-C° it-
self 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 tradi-
tional 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 allit-
erative conventions of Germanic epic poetry, which was certainly flourishing
around the time verb-second word order gained ground in the Germanic lan-
guages, 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 cat-
egories 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 con-
struction in all the early Germanic languages, and one manifestly cognate with
verb second from a structural point of view, and very likely from an histori-
cal point of view as well (see the examples in (5) and the discussion in SEC-
TION 2.4).
According to the variant of the cliticization hypothesis proposed by Hock
(1982), verb second develops as a generalization of a process cliticizing Aux-
iliaries. Auxiliaries are not lexical categories and so it is a priori not impos-
sible 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 sec-
ond from cliticization is that the two processes have quite distinct properties,
a fact unfortunately .disguised by the use of non-committal "S VO" terminol-
ogy. Clearly the verb-second process that arises from the putative generaliza-
tion 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 mani-
festations of a principle which he formulates as follows: "locate the formal
reflection of a linguistic unit's relational properties by reference to a promi-
nent position in that unit." In Germanic, the reference point is the first con-
stituent, 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 verb-
second 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 al-
ready 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 re-
sult 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 Indo-
European 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 re-
flexes of the introduction of complementizers which itself is just one of sev-
eral 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 indeclin-
able subordinating complementizer, are pan-Germanic and most likely Proto-
Germanic. V-to-C° movement, on the other hand, belongs in a group of
innovations that are absent from Gothic (and probably from the earliest Scan-
dinavian 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 be-
hind 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 discus-
sants 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 move-
ment, with full discussion of the recent literature, see Viker (1990) and Rob-
erts (1993). Lenerz (1984, 1985), and Kemenade (1987), show that the V-to-
C° 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 uni-

versal principle that shorter derivations are preferred over longer derivations
(Chomsky (1991)). See Vikner (1990) for enlightening discussion of the is-
sue 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 func-
tion, being typical of "lively narrative," often linking a sentence to the previ-
ous 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 combina-
tion thereof. Therefore, examples like (15b) below cannot be cited as evi-
dence 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 surviv-
als 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 posi-
tion 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 ori-
ented 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 men-
tioned "whether" type) is explained on the assumption that they have an obliga-
tory 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 counter-
part 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 un-
realized 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 af-
ter a class of non-factive verbs. Wechsler (1990) argues that this is a conse-
quence of the special illocutionary status of these complements as "asser-
tions". These clauses seem to represent the thought or words of the implicit or
explicit agent of the containing verb. In Icelandic, however, embedded pad-
clauses are reported to undergo topicalization freely.
16. This statement anticipates the treatment of relative clauses discussed in SEC-
TION 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 evi-
dence. 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 subordi-
nate 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 Scandi-
navian languages, might very well have had "root" phenomena in subordi-
nate clauses, which would have constituted another source of apparent ex-
ceptions 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 condi-
tions 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 which-
ever 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.) how-
ever, 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 prop-
erties of main clauses; in particular, the relation between relative and correla-
tive is governed essentially by the same principles that govern anaphoric re-
lations between main clauses.
33. Hock appeals to the Penthouse Principle (Ross (1973)), but this is only legiti-
mate 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 character-
istic 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 idi-
oms, 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 un-
dergo 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 Occa-
sional 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 Uni-
versity Press.
Andersson, L.-G. (1975) "Form and Function of Subordinate Clauses." Mono-
graphs 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 Lexi-
cal 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 Represen-
tation." 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 Disserta-
tion, 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 tradi-
tion." 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-Euro-
pean." 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 Scandina-
vian 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 Transfor-

mations." 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 His-
tory of English. Dordrecht: Foris.
Kiss, K. E. (1987) Configurationality in Hungarian. Dordrecht: Reidel.
Koster, J. (1975) "Dutch as an SOV Language." Linguistic Analysis 1:111-
136.
Lehmann, C. (1980) "Der indogermanische A w i-/k w o-Relativsatz im
typologischen Vergleich." In P. Ramat, et al., eds. Linguistic Reconstruc-
tion 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 Seman-
tics. Dordrecht: Reidel.
McCloskey, J. (1990) "Resumptive Pronouns, A-Binding and Levels of Rep-
resentation 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 Pennsyl-
vania.
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, 375-
385. 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 Disser-
tation, 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 Syn-
tactic 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 Dis-
sertation, 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 Ger-
manic 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 sev-
eral Old and Middle French texts written between the 13th and the 15th cen-
turies 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 domi-
nate the grammar in Middle French: the coexistence of "Germanic" verb-
second inversions and Romance "free inversion" (SECTION 2), and the
dissociation of the initial constituent from the inverted clause, also respon-
sible for the rise of verb-third orders (SECTION 3). I will claim that the varia-
tions 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 reset-
ting 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 gram-

173
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 con-
cerns 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 pro-
portion of pronominal inversions, which we claim are generated only by V-
to-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 (ital-
ics) 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 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'.
(3) [CP Longuement [c resgarda [IP Perceval [I' t [ VP t l'ome...]]]
(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 move-
ment 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 175

(4) (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, how-
ever, 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 analo-
gous 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 restric-
tions on what can intervene between the finite verb and the subject in such
constructions, these restrictions are similar to those observed for Italian in-
version 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 unaf-
fected 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 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
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 sepa-
rated from the finite verb by any other VP material. If OF had obligatory V-
to-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 depen-
dent 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 pos-
sible sources of diachronic change. In so doing, we extend and unify sugges-
tions 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 in-
versions of either (5a-d) or (6). Vance, largely following Kayne (1983), ac-
counts 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 hypo-
thetical pronominal versions of (5a) and (6) are then (5a') and (6'), respec-
tively. 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 poten-
tially 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 Nomi-
native Case by special mechanisms. In such cases, then, there is no true "in-
version" 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 be-
tween the position of unaccusative and passive subjects and other subjects, a
contrast that is of potential importance but is more difficult to formulate un-
der 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 struc-
ture; 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 in-
creases 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 fol-
low 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
Total CVSp CVSn CVXSn
Queste (1225) 419 212 (.51) 138 (.33) 69 (.16)
pp. 1-41,71-115
Joinville (1306) 74 24 (.32) 43 (.58) 7 (.10)
pp. 104-125
Saintre (1456) 44 15 (.34) 23 (.52) 6 (.14)
pp. 239-294
Commynes (1491) 93 11(.12) 50 (.54) 32 (.34)
pp. 58-97

Although the total numbers of inversions in each text are not directly com-
parable, 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 in-
versions 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 be-
ginning of the 14th century and remains there. Recall, however, that the cat-
egory 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-pro-
nominal 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 po-
sition 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 sim-
ply 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# CVXSn Total VP-final
CVSnO
Q 212 (.51) 27 (.06) 111 (.26) 69 (.16) 180 (.43)
J 24 (.32) 6 (.08) 37 (.50) 7 (.10) 44 (.60)
S 15 (.34) 1 (.02) 22 (.50) 6 (.14) 28 (.64)
C 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 inver-
sions. 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 in-
creases 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 as-
sumed, 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-pro-
nominal inversions). Although more work remains to be done, it seems un-
likely 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 CVSn CVXSn
Queste 42 of 208, 20% 67 of 137, 49% 76 of 121, 63%6
pp. 1-41, 71-115
Joinville 2 of 24, 8% 24 of 43, 56% 3 of 7, 43%
pp. 104-125
Saintre 5 of 15, 33% 18 of 26, 69% 5 of 6, 83%
pp. 239-294
Commynes 5 of 11, 45% 39 of 50, 78% 25 of 32, 78%
pp. 58-97
 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 struc-
tures 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 struc-
ture 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 embed-
ded 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 good-
ness 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 af-
fairs is what we see reflected in Tables 7.1 and 7.2, where ambiguous non-
pronominal 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-pro-
nominal 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 de-
scribed here have in common the property that they have identical surface
structures in both matrix and embedded clauses in OF. The only clear evi-
dence 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 ques-
tion, 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 de-
tail 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 domi-
nate 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 non-
subject constituent. We first see evidence, once again, of synchronic differ-
ences 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 posi-
tion 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. Percent-
ages 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 constitu-
ents 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 re-
served 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
CVSp CVS CVXS
Adverb 84 (.42) 101 (.73) 47 (.68)
PP (adjunct) 66 (.33) 28 (.20) 17 (.25)
Dir.Obj. 23 (.12) 6 (.04) 4 (.06)
Clause 19 (.10) 0 0
Inf 3 0 1
Past Participle 1
>1 Constituent 3
Adjective 3
Misc Subcategorized 1
199 139 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 obliga-
torily 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 "mean-
ingless" 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 ex-
amples can be seen to make use of these adverbs approximately half as fre-
quently 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 con-
structions are in fact topicalization constructions. When si triggers the inver-
sion 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 17 (.20) 27 (.27) 16 (.34)

lors 7 (.08) 24 (.24) 6 (.13)

Total Adv 84 101 47

%si+lors 24/84=.29 51/101=.50 22/47=.47
 ON THE DECLINE OF VERB MOVEMENT TO COMP 185

the ambiguity that Roberts describes, then, is an ambiguity in the interpreta-

tion of the initial constituent of CVSn clauses. The presence of the "mean-
ingless" 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 con-
stituents in 13th-century prose are adverbs and adverbial PP's; the first con-
stituent of the new CSV clause carries over this tendency toward non-
subcategorized material. The competition between inversion and non-inver-
sion 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
CV(X)S CSV % Inv
Queste 219 7 97%
1225
Joinville 74 30 71%
1306
Froissart 48 38 56%
c. 1375
Quinze Joies 44 41 52%
1420
Jehan de Saintrd10 39 58 40%
1456
Commynes 91 89 51%
1491
186 VERB SECOND AND THE NULL-SUBJECT PARAMETER

Table 7.6 above shows the ratio of inversion to non-inversion after an ini-
tial 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 im-
portant 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 accord-
ing 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 sub-
set 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 pres-
ence of an initial constituent only the latter is possible.

Table 7.7
Inversion vs. Non-inversion of Pronominal
and Non-pronominal Subjects After XP
CVSp CSpV % inv CVSn CSnV %Inv
Queste 97 3 97% 122 4 98%
1225
Joinville 24 17 59% 50 13 79%
1306
Froissart 15 26 37% 33 12 73%
c. 1375
Quinze Joies 17 28 38% 27 13 68%
1420
Jehan de Saintre 9 28 24% 30 30 50%
1456
Commynes 11 60 15% 80 29 73%
1491
 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 under-
standing of diachronic change; when added to similar evidence from syntac-
tic 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 influ-
enced, 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 verb-
third 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) ar-
gues against this account on the basis of evidence that preverbal subject pro-
nouns 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 sub-
jects 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 Rob-
erts (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 vari-
ably 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, how-
ever, 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 constitu-
ents to Spec CP. In the Queste, entire clauses may trigger pronominal inver-
sion:
(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 cen-
tury 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 essen-
tial 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 in-
version 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 posi-
tion in MidF, for example in initial position in embedded clauses:
(23) et grant partie du surplus tel que pro voyez
(Saintre, 246,24)
and great part of-the surplus such as pro see-2pl.
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 im-
plies that an IP structure is at issue. Finally, such clauses have exact counter-
parts 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 sub-
urbs 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 il revenoit beaucoup de gens, qui avoient
nevertheless there returned many of people who had
este caches es bois (Commynes 65,10)
been hidden in-the woods
The expletive pronoun il is clearly not required, since its absence would sim-
ply result in structures of the type in (24) and (25). The fact that it is spo-
radically 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 struc-
tures 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 inver-
sion 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 pro-
nominal 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 reanaly-
sis (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 differ-
ent. 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 constitu-
ent. IP inversions, on the other hand, generally have VP-final subjects, strictly
exclude pronouns, and may occur in embedded contexts. Since they are es-
sentially "free inversions" and not verb-second structures, they may in fact
be verb-initial in MidF. It is the coexistence of these two underlyingly differ-
ent 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 gen-
eral and should be applicable to other cases than the history of French. Spe-
cifically, 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 vol-
ume 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 allow-
ing inversions in both main and embedded contexts. This matter is an empiri-
cal 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 recog-
nized 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 lan-
guages (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 imple-
ment Rizzi's suggestion for Old French would require a more detailed articu-
lation 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 con-
stituent.
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 con-
structions, and the postverbal subjects of unaccusatives and passives are usu-
ally 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 pas-
sive "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 ordi-
nary subjects.
On the other hand, this example may reflect the sort of movement of non-
finite 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. Al-
though 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 philo-
logical 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 syntac-
tic 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. Propo-
nents of the cliticization account interpret this fact as evidence for their posi-
tion, 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 syntac-
tic 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 fac-
tors are held to be ultimately responsible for the loss of verb second. Further-
more, 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 avail-
able 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 VP-
final 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. Proceed-
ings ofNELS 18, 1-21. Amherst, Mass.: GLSA, University of Massa-
chusetts.
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, Uni-
versity of Pennsylvania.
Besten, H. den (1983) "On the Interaction of Root Transformations and Lexi-
cal 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 Ital-
ian." Harvard Working Papers in Linguistics 91-127.
Chomsky, N. (1991) "Some Notes on Economy of Derivations and Represen-
tations." 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 Lan-
guage 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 His-
tory 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: Configu-
ration, 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, 67-
80. 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, Func-
tional 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 Literatuur-
wetenschap.
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 person-
nels 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 Re-

ports 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 Dia-
lects. 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 Se-
mantics 24:41-69. San Diego: Academic Press.
Sprouse, R.A. and B. Vance (1993) "Null Subjects in Surselvan." Paper Pre-
sented 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 Dis-
sertation, 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 Projec-
tions. PhD Dissertation, University of Washington.
Zwanenburg, W. (1978) "L'ordre des mots en francais medieval." In R. Mar-
tin, 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 un-
ambiguously 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. Con-
sider 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 pre-
sentation 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 lan-
guages 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 ac-
counts for the need to move the tensed verb to C° in main clauses of verb-
second 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 lead-
ing 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 expla-
nation. 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 rep-
resented. 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 predica-
tion 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 (par-
ticiples 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 fea-
ture [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 host-
ing 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 subject-
verb 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 lan-
guages 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 as-
sume 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 li-
censed 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 de-
fined. 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 gover-
nor 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 po-
sitions. 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) (in-
formation 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, how-
ever, 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 ac-
cording to (2b) (a specifier position is head governed by its head). If alterna-
tively [+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 assign-
ment 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, inde-
pendent 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 205

(7) 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 gov-
erned 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 be-
low, English and French have developed differently with respect to Agr, some-
thing 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 re-
turn 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 En-
glish 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 En-
glish 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, try-
ing 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 fre-
quency 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 em-
bedded 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 suffi-
cient 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 am-
biguity 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 indi-
cated the presence of verb second, and a bulk of sentences which were struc-
turally 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 el-
ements: hence, the loss of verb second in French is described as the intro-
duction 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 emer-
gence of blockades (mainly the negation not) which prevented the finite fea-
tures 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 projec-
tions, S-identification is conjugation (i.e., tense and mood). In the same
manner as for Case, conjugation can be assigned either inherently or structur-
ally (via C°, governing a finite Vmax; it is to be noticed that there is no I-pro-
jection 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-sec-
ond 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 be-
tween indicative and subjunctive. Finally, there are certain theoretical prob-
lems 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 hypoth-
esis 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 differ-
ently 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 prob-
ably sentences with a word order ambiguous between a verb-second interpre-
tation and a basic SVO-order interpretation. The structural ambiguity of such
sentences is illustrated in (8), where (8a) shows the verb-second interpreta-
tion 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 am-
biguous 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 out-
lined 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 ambigu-
ous; 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 utter-
ances 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 (di-
rect questions, wh-questions and cases with topicalization) unambiguously in-
dicated 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 consti-
tuted a further ambiguity. Together this led to a situation where true verb-
second indicators were few in number. In spoken language, and especially in
the language spoken to children, the number of sentences unambiguously in-
dicating 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 sen-
tences 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, ac-
cording to van Kemenade (1987:177)), the potential ambiguities found in SEC-
TION 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 pa-
rameter 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 ordi-
nary 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 consider-
ation the regular verb second found in Chaucer, she concludes that "the be-
havior 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 mis-
take. 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 ques-
tion 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-sec-
ond 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 ut-
terances 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 sen-
tence 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 'cer-
tainly'. Vance assumes that the cases with introductory clauses are only ap-
parent 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 intro-
ductory adverbials are considered to be real exceptions. This is a doubtful
conclusion, however, since similar cases are found in a true verb-second lan-
guage 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 situ-
ated 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 re-
stricted 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 comple-
ment 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-syn-
tactic 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, not-
withstanding 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 pa-
rameter. As I have argued, language learners of both English and French re-
acted 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 sta-
tus 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 indepen-
dent 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 corre-
lation 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 dif-
ferences 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 discus-
sion 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 V-
to-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 mod-
ern 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 hypoth-
esis, I must show that the weakening of Agr and the emergence of both aux-
iliary 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-lower-
ing, since there is no other way for tense to be realized within VP. As no-
ticed 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 ex-
ample 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 oc-
curs 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 fea-
ture: 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 differ-
ent 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 iden-
tified. 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-fea-
ture, 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. Mod-
ern German seems to lack (p-features in C° altogether: the only possible iden-
tification 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 ex-
amples like (21b,c) are not well formed, since C° without any (p-features can-
not 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 quasi-
argumental 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 con-
trast illustrated by (21c) and (22) indicates that C° may be marked for [+num-
ber] 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 embed-
ded 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 sub-
jects 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 subordi-
nate clauses. Consider the following example, taken from Vance (1989:3):
(24) Puet bien estre que n'en avez point (Jehan de Saintre)
may (3sg) well be that neg-en have neg
'It may well be that you have none'.
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, En-
glish 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. How-
ever, 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 ex-
planation: at the time of the loss of verb second, both languages had a pre-
dominant 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 situa-
tion is a necessary prerequisite for a syntactic change to take place.
According to Holmberg and Platzack, who provide the theoretical frame-
work 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 valu-
able 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 Ger-
man, 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 (Swedish)
that U. not bought the-book
c. a5 Jon keypti ekki bokina (Icelandic)
that J6n keypti ekki bokina
(ii) a. Karl kaufte das Buch nicht. (German)
K. bought the book not
b. Ulf kopte inte boken. (Swedish)
U. bought not the-book
c. Jon keypti ekki bokina. (Icelandic)
J. bought not the-book
The apparent lack of a root-subordinate distinction in languages like Icelan-
dic 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 contri-
butions 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 (con-
sider e.g. Platzack and Holmberg (1989)), this description should apply to
these languages as well. Consider also Travis (1984), who claims that sub-
ject-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 stan-
dard V-to-Comp analysis which was first suggested by den Besten in an un-
published 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 non-
nominal Agr. Such languages should have no V-to-I in spite of having sub-
ject-verb agreement. There are some MSc. dialects of this kind, e.g. the dia-
lect of Hallingdalen in Norway (see Trosterud (1989)).
7. Rizzi uses this formulation in order to specify the three subcases of anteced-
ent 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 ex-
ception. 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 ap-
proaches.
15. The example is found in van Kemenade (1987:200).
16. The example is found in Vance (1989:200); notice that Vance does not con-
sider 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 ex-
amples 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 Mod-
ern 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 em-
barrassing 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 struc-
ture 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 ex-
amples 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 em-
bedded 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 investiga-
tions of the exact contexts inwhich null subjects are possible in MF texts.

References
Abraham, W. and E. Reuland, eds. (1991) Issues in Germanic Syntax. Ber-
lin/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 Lexi-
cal 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 Posi-
tion." 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 Con-
trastive Perspective." Paper presented at the Seventh International Con-
ference 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 Scandina-
vian 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 His-
tory 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 Mor-
phological 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." Linguis-
tic 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 1:1-
30.
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 Dis-
sertation, 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 Norwe-
gian 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. 1-
111. 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. Mar-
tin, 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 pro-
drop in French and English. We shall identify two types of verb second lan-
guages, 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 assump-
tions concerning the licensing of functional properties, allows a very articu-
late 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 back-
ground 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 pro-
jections 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 asymme-
try 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 move-
ment 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 lan-
guages, 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 em-
bedded 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 lan-
guages, 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 verb-
second.3 We will come back to this in the next section.
Non-verb-second languages (Modern English and Modern French) are char-
acterized by movement of Vf to I (or vice versa4), and no further movement
to C, except under restricted circumstances such as in Modern English inter-
rogative 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 lan-
guages then is that in verb-second languages Comp is always lexically real-
ized, 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 num-
ber of crucial head properties that in non-verb-second language are character-
istic 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, interact-
ing 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 ab-
sence 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 ei-
ther syntactically (i.e. structurally) or morphologically. In this article, we con-
centrate 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 con-
dition (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-fea-
tures represents the case of languages such as Modern Italian where agree-
ment 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 fol-
lows 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 singu-
lar form. I is then not fully licensed, as there is no structural or morphologi-
cal 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 Mod-

ern 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 exple-
tive pro-drop is customary. Full pro-drop appears in languages such as Mod-
ern 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 Ital-
ian, 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 non-
thematic subject, as in impersonal sentences such as those in (9) above. Exple-
tive 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 condi-
tions 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 desig-
nated 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 cru-
cially with the dominant head for C/I orientation in terms of our theory. In C-
oriented 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 lan-
guage. 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 Ital-
ian 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-fea-
tures 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 pro-
nominal C, may contain (p-features: [±person, ±number, ±gender]. We follow
Rizzi (1982:42) in assuming that it is the specification [+person] that is nec-
essary to license referential pro-drop. If a pronominal I does not have the
[+person] specification, only expletive pro-drop is possible. In I-oriented lan-
guages, 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 lan-
guages:
(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 lan-
guages for which this is relevant, are Modern Dutch and earlier stages of En-
glish, as exemplified above. We will discuss the latter in more detail below.
The possibilities for pro-drop discussed are summarized in (15), with men-
tion of some of the core instantiations of the various options.11
(15) V2 Comp Infl
MoDu/OE/ME + +DH -phi expl.pro-drop
Cont. Scand. + -DH - Phi no pro-drop
Ital/Span - - +DH+phi full pro-drop
MoFr/MoE - - -DH-phi no pro-drop
Icelandic + -DH +DH-phi 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 mix-
ture 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 pio-
neering 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 lan-
guage change. Weerman's alternative, which we do not wish to discuss fur-
ther 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 phe-
nomena 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 hypoth-
esize 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 fol-
lowing sections provides crucial support for distinguishing at least two func-
tional 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 exple-
tive pro-drop that is not limited to root contexts, whereas OE has only exple-
tive 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 char-
acteristics 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 num-
ber of observations that were earlier made by traditional grammarians. Adams
analyzes OF as a verb-second language with referential pro-drop, but inter-
estingly, 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 ref-
erential 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 verb-
second 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. Move-
ment of V/I to C is derived by condition (3) above, that states that the domi-
nant 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 li-
cense 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 pro-
drop, 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 verb-
second language where C can act as a formal licenser for pro and I has (p-
features, 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 non-
root 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 ac-
count 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 con-
stituent 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 charac-
ter 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 9-
features. 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 ele-
ment 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 con-
texts 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 sec-
ond and referential pro-drop. By lexicalizing C, verb second creates the con-
text 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 counter-
examples 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 embed-

ded IP in OF, subsequent work by Adams (1988a, 1988b), Vance (1989),
Dupuis (1988), Hirschbuhler (1990, this volume) has shown that this assump-
tion cannot be maintained. According to Dupuis null subjects in a position
immediately preceding the finite verb (a non-verb-second context) were pos-
sible in specific genres even in OF (23). Others (Vance 1988) relate this pos-
sibility 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 pro-
drop 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 analy-
sis, as we propose that lexical C is the formal licenser for pro, but the con-
tent-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 for-
mulated 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 opera-
tions 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 pronoun-
Vf could be interpreted in two ways: as having an XP fronted to CP and clitic-
Vf 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 struc-
tures 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 dis-
cussed above, there was an increase in these structures in the MidF period,
cf. Lemieux and Dupuis (this volume). Hirschbuhler and Junker (1988) re-
port 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 feroit- ilz (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 interpre-
tation; as an IP interpretation for the verb-second phenomenon became avail-
able, 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. Dur-
ing 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 interpreta-
tion, triggered a shift to IV2. Verb-second structures in MidF are then ana-
lyzed 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-gener-
ated position. Evidence for this VP-internal position is given by Lemieux and
Dupuis (this volume) who cite examples in which a VP-adjoined adverb in-
tervenes between Vf and the postverbal subject:
242 VERB SECOND AND THE NULL-SUBJECT PARAMETER

(30) Et la commenca fort le trait d' arbalestres et de

and then started violently the firing of crossbows and of
canons (Melusine, p. 89)
canons
'and then 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 embed-
ded 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 gov-
ernment 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 lan-
guage 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-ori-
ented language can assign Nominative Case under government or under Spec-
head 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 Case-
marking 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 sug-
gested 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 un-
der Spec-head agreement could never be the only option, as postverbal sub-
jects 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 iden-
tified 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 his-
tory 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, sev-
eral 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 asym-
metry 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 imme-
diate 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 ex-
amples 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 referen-
tial pro-drop does not occur in OE. Given our theory, we analyze the OE situ-
ation 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 inher-

ently licensed, hence it must be licensed under structural conditions by speci-
fier-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-projec-
tion.
We will now consider how default licensing is instantiated in OE. The ex-
amples involving expletive pro-drop (37) are all examples where I is not fully
licensed: there is no morphological licensing (no inherent p-features) or syn-
tactic 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 there-
fore 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 Nomi-
native subject became obligatory in the early 15th century.
From the middle of the 14th century, verb second declines. As to the pri-
mary 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 fol-
lowed 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 system-
atic discrepancy between the position of the nominal subject and the pronomi-
nal 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 nomi-
nal 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); non-
inversion 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 pro-
nominal 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 li-
censed 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 cen-
tury, s jects come to appear more and more frequently to the left of the fi-
nite 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 caus-
ally 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 causa-
tion involved here. At the stage in the history of English where these devel-
opments 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 dif-
ference 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 referen-
tial subject could never be dropped, English, when it lost its CV2 characteris-
tic, 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 pro-
jections.

Notes
1. This paper was presented at the First Diachronic Generative Syntax Confer-
ence 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 particu-
lar 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 histori-
cal 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 lexi-
cal; 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 Case-
marking 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 cre-
ates 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 Ger-
manic stress pattern, that gradually changed to the Modern French stress pat-
tern, 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 prop-
erty 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 chain-
related to the base-generated position in Spec.VP. This is unproblematic for
Nominative Case and formal licensing; these two properties can be transmit-
ted 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 spec-
head 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 after-
wards its formal licensing properties. The precise relationship with Nomina-
tive 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 analy-
sis 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 conclu-
sively 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, V-
fronting is not completely obligatory; and second, the position preceding the
finite V in root clauses is sometimes occupied by two constituents. This, how-
ever, 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 rep-
resent an important literary standard, see Warner (1982). Allen (1990) criti-
cizes 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 observa-
tions 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 Linguis-
tics 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 Rela-
tive 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 Lexi-
cal 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 Gov-
ernment 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 Ger-
manic 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. Lon-
don: 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. Linguis-
tics in the Netherlands, 69-78. Dordrecht: Foris.
Hulk, A. and A. van Kemenade (1993) "Subjects, Nominative Case, Agree-
ment and Functional Heads." Lingua 89:181-215.
Kemenade, A. van (1987) Syntactic Case and Morphological Case in the His-
tory 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 Pa-
rameter." 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: Edi-
tions 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: Aca-
demic 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 em-
bedded 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 dis-
cussed 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 ex-
amined here are necessarily relevant for other Middle French texts. Some of
the facts are unexpected given previous literature. It may therefore be tempt-
ing 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 rel-
evant 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-sec-
ond sentences, which are analyzed as CPs, and SVO clauses which, follow-
ing 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 posi-
tion 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 ini-
tial 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 em-
bedded 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 sub-
jects 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 consider-
ation: 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 dis-
cussed 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 em-
bedded 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. 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.
2. 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.
3. Licensing subordinators: lexical wA-phrases (excluding those in-
stances 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 for-
mally 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 pre-
vails.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. Fol-
lowing 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 mor-
phologically uniform, with a slot for agreement affixes available for each per-
son), 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 agree-
ment with Infl. Thus we distinguish between:
1. Strong Agr. 2nd. p.pl. (i.e. person 5) allows verb first with null sub-
jects 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 Spec-
Head 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.
2. Healthy Agr. 3rd singular (expletive, i.e. person 31, and non-exple-
tive, i.e. person 3) and 3rd p. pi. (person 6). These are able to iden-
tify 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 261

3. 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 adja-
cent 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 subor-
dinate clauses, which are discussed in SECTION 4. There are "XP (S)VO" sub-
ordinate clauses just as there are main clauses of that type. When the subject
is null, they resemble verb-second constructions, but there are clear indica-
tions 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-licens-
ing 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 per-
sons 1 and 2.

1.2.4 Stylistic Inversion vs Null Subjects

Recently, it has been suggested that the so-called Stylistic Inversion con-
struction 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 (com-
parative)
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 wh-
gap, while impersonal null subjects are not, it would seem that the expletive
of Stylistic Inversion constructions, if there is one, should be treated differ-
ently 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 for-
mal 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 act-
ing as an formal licenser, and the relation of Agr to SpecIP, required for con-
tent 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 imper-
sonal 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 be-
ing 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 examina-
tion 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 em-
bedded 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 correspond-
ing 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 su-
perlative clauses, and indirect questions. Part B contains temporal and con-
ditional 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 wh-
clauses (indirect questions, relatives, comparatives), except for those intro-
duced 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 pre-
ceding 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 12 2 3 31 7 ce 4 5 6 Total %
Main Clause 127 12 927 415 28 28 53 300 1890 78.61
Subord. Clause 7 6 124 207 4 7 127 31 514 21.38
Total 134 18 1051 622 32 35 180 331 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 preverbal Spr immediately the subordinate clause intro-
ducer (cf. Table 10.3),
72 preverbal Spr separated from the subordinate clause in-
troducer by some phrase (cf. Table 10.14), (to which nine
examples of on might be added), and
25 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 coordi-
nations 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 propor-
tion 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 embed-
ded 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 con-
structions (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 construc-
tion 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 1 2 ce 3 31 4 5 6

purpose 7 5 1 4
cause 16 6 6
subject/object 114 22 22
misc. 13 5 5
concessive 7 3 3
result 47 2 1 1
optative 2 0 0
comp. que 6 2 2
R. Rel.: que 28 19 2 17
I.Q.: wh-que 0 0
Total A 240 64 1 3 60

B. Spec. Cases Total 0 Sub 0 V 1 2 ce 3 31 4 5 6

quant 'when' 13 5 1 4
Temporal phrase 4 3 1 2
conditional si 63 34 4 1 0 29
Total B 80 42 1 1 4 3 4 29

C. Wh # que Total 0 Sub 0 V 1 2 ce 3 31 4 5 6

I.Q.: si 6 3 2 1
I.Q.: wh # que 5 3 2 1
NP Appos. Rel. 42 36 11 22 1 2
CP Appos. Rel. 26 11 5 4 2
R.Rel. # que 7 3 2 1
comme 'as' 108 17 2 15 0
Total C 194 73 20 31 1 16 5
Total A, B, C 514 179 1 1 4 24 35 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 30 2 1 0 20 1 0 1 5
cause 143 14 4 3 79 10 10 3 20
subject/object 1027 86 32 108 554 103 16 38 90
misc. 62 1 3 2 40 4 1 2 9
concessive 57 6 2 3 34 5 0 4 3
result 379 11 9 10 246 42 3 3 55
optative 10 2 0 1 5 0 1 0 1
comparative /
superlative: que 106 6 4 1 67 1 3 8 10
Rest. Rel. : que 506 72 12 3 283 29 14 21 72
I.Q.: wh-que 129 12 2 24 37 27 5 7 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

quant 'when' 267 8 8 24 166 9 4 1 47
Temporal phrase 115 10 2 2 60 8 4 2 27
conditional si 221 26 19 8 97 17 4 33 17
Total B 603 44 29 34 323 34 12 36 91

C. Wh # que Sub Spr V 1 2 ce 3 31 4 5 6

I.Q.: si 66 2 2 2 42 6 0 3 9
I.Q.: wh # que 122 17 2 11 67 0 1 7 17
NP Appos. Rel. 87 15 I 0 55 4 3 4 5
CP Appos. Rel. 99 12 0 4 63 8 0 0 12
Rest.Rel. # que 183 38 1 2 98 11 14 5 14
comme 'as' 275 38 7 0 78 39 1 90 22
Total C 832 122 13 19 403 68 19 109 79
Total A, B, C 3884 378 111 208 2091 330 84 232 450

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, fac-
tors 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 signifi-
cant 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 non-
wh-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 ac-
tual one, because of the filled embedded C or SpecCP. Competition with an-
other 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 com-
parison between Spo and Spr for each person in Table 10.4 below. The per-
centages 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 repre-
sent 31.15% for person 5! This difference cannot be imputed to a formal dif-
ference 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 Sub Sp(o)V 1 2 ce 3 31 4 5 6

Total (T2a) Spo 42 1 1 4 3 4 0 29 0
Total (T 2b) Spr 603 44 29 34 323 34 12 36 91
% Spo 6.9' 6 2.222 3.33i 10.552 0.9:2 10.52 0 44.6il 0

Categ;ory C Sub Sp((3)V 1 2 ce 3 31 4 5 6

Total (T 2a) Spo 73 0 0 0 20 31 1 16 5
Total (T 2b) Sp 832 122 13 19 403 68 19 109 79
% Spo 8.06 0 0 0 4.72 31.31 5 12.8 5.95

Total B+C Sub Sp(o)Vl 2 ce 3 31 4 5 6

Spo 115 1 1 4 23 35 1 45 5
Spr 1435 166 42 53 726 102 31 145 170
% 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 percent-
age (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 examina-
tion 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 ex-
amples 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 clas-
sify 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 al-
ways 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... 091018

...when it was the moment to pay tax to the prince
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 cat-
egory 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, ... 011008
But before that [he] refl-left, ...
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, how-

ever, 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,... 096027
And if [it] weren't for that,...
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,... 041067
...if it was not for fear of damaging my bed,...
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 embed-
ded 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 per-
sons. The facts are clear for person 1, as there are 122 examples with a pro-
nominal 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. Ei-
ther the complementizer licenses SpecIP, in which case identification by per-
son 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 per-
son 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 1 2 ce 3 31 4 56 6 Total
Spo 0 0 0 20 31 1 16 5 73
Spr 122 13 19 403 68 19 109 79 832
% Spo 0 0 0 4.72 31.31 5 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 im-
personal, 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 particu-
lar, we will see that there is no clear indication that null subjects are regu-
larly 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 ei-
ther 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 discov-
ered, 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 previ-
ous 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 Sub Sp V 1 2 ce 3 31 4 5 6
Q.I.: si Spr 66 2 2 2 42 6 0 3 9
Q.I.: si Spo 3 0 0 0 2 0 0 0 1
% Spo 4.34 4.54 — 10

Q.I.: wh Spr 122 17 2 11 67 0 1 7 17

Q.I.: wh Spo 3 0 0 0 2 1 0 0 0
% Spo 2.4 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... 041056
I don't know if [he] suspected what was going on...
c. ...Dieu sail si 0 6 le raillerent... 052056
...God knows if [they] scoffed at him...
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. 031029
... [they] asked him why [he] had gone away
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 prob-
lematic 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 proper-
ties, 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 Sub Sp V 1 2 ce 3 31 4 5 6
I.Q.: wh-que 129 12 2 24 37 27 5 7 15
Indirect questions are the clearest case where que introducing an embed-
ded 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 wh-
que 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 indi-
rect 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. 094055
...I will tell you what it is necessary for you to do
c. ...et veit bien quelle heure qu'il estoit.10 091111
...and [he] saw very well what time it was
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 nomi-
nal antecedent, appositive relatives with a CP antecedent, and restrictive rela-
tives. 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 Total 1 2 3 31 4 5 6 Total
Relatives Spo 50 0 0 16 28 1 1 4 22
Relatives Spr 363 65 2 216 23 17 9 31 341
% Spo 11.89 0 0 6.89 54.90 5.55 10 11.42 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 5 6
R.A. Spo 36 0 0 0 11 22 1 0 2
R.A. Spr 87 15 1 0 55 4 3 4 5
%Spo 29.26 0 0 0 16.66 84.61 25 0 28.57
Table 10.8 clearly shows that relatives not introduced by que allow null sub-
jects, especially in the case of person 31, and to a lesser extent 3 and 6. Per-
sons 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 ante-
cedent 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 ex-
amples 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 sub-
jects, 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 divi-
sion of NP appositive relatives in two classes. The first one, with 13 Spo ex-
amples, 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 13 0 0 0 3 10 0 0 0
R.A. 1 Spr 6 0 0 0 4 1 0 0 1
%Spo 68.42 — — — 42.85 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 im-
personal 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 rela-
tives 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 23 0 0 0 8 12 1 0 2
R.A,.2 Spr 81 15 1 0 51 3 3 4 4
%Spo 22.11 0 0 0 13.55 80 25 0 33.33
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 phenom-
enon 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 ap-
positive 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 apposi-
tive 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 se-
cretly 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 279

f. ...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 differ-
ent places in the towns...
Rega ing the hypothesis (suggested by the first class of appositive rela-
tives) 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 sub-
jects to verb-second constructions in main clauses: the relative here can of-
ten 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 ap-
proach 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 sub-
jects 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 sub-
jects 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 1 2 ce 3 31 4 5 6
R. Ph. Spo 11 0 0 5 4 2
R. Ph. Spr 99 12 4 63 8 12
% Spo 9.48 0 0 7.35 33.33 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 imper-
sonal 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 ex-
amples 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 gram-
matical, 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 ap-
positive relatives, the fronted wh-phrase was taken as a topicalized constitu-
ent 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 re-
search.

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 com-
monly 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 rela-
tives not introduced by que, i.e. the remaining null subjects are limited to the
impersonal.
Table 10.13
comme
Category C Sub Sp V 1 2 ce 3 31 4 5 6
comme Spr 275 38 7 0 78 39 1 90 22
comme Spo 17 0 0 0 0 2 0 15 0
% of Spo 5.82 0 0 — 0 4.87 0 14.28 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 sub-
ject 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 verb-
second 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 distin-
guishing the cases with a fronted complement from those with the exple-
tive il.

3.4 Summary on Spo in Embedded Verb First

We have looked in minute detail at the different types and subtypes of con-
structions, 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-construc-
tions, 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 ques-
tions, in the remaining restrictive relatives, as well as in the examples intro-
duced 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 li-
censing 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] lo-
cated 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 per-
centages are not very high in the case of person 3, it reaches 6.89% in rela-
tive 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 explana-
tion. 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 verb-
first construction, is possible in wh-clauses introduced by que while the im-
personal verb-first construction is excluded from this context, indicates that
there is a null subject (pro) that needs content identification in standard im-
personal 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 construc-
tion. 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 li-
censed 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 construc-
tion 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... 002045
...but one thing I ask you...
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 2 3 31 ce 4 5 6 on
purpose 1 1
cause 5 1 2 1 1
complement 27 2 1 12 3 2 3 4
misc 1 1
concessive
result 19 13 1 1 1 1 2
optative
comp.-superl. 1 1
Rest.ReL: que 2 2
Total A 56 6 2 28 4 1 4 5 6

B: Spec. Cases Sub XP Sp V 1 2 3 31 ce 4 5 6 on

quant 'when' 3 3
temporal phrase
conditional si 2 1 1
Total B 5 3 1 1

C: Wh # que Sub XP Sp V 1 2 3 31 ce 4 5 6 on
Q.I.: si I 1
Q.I.: wh # que 0
NP Appos. Rel. 6 1 4 1
CP Appos. Rel. 5 4 1
R.Rel. # que 6 2 3 1
comme 'as' 2 1 1
Total C 20 4 11 1 1 3
Total A, B, C 81 10 2 42 4 1 4 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 Total Sub X V 0 1 2 3 31 4 5 6
purpose 7 2 1 1
cause 16 10 1 5 4
complement 114 92 1 36 43 1 4 7
misc. 13 8 1 4 2 1
concessive 7 4 2 1 1
result 47 45 1 28 8 1 7
optative 2 2 1 1
comp.-superl 6 4 2 2
Rest.Rel.: que 28 9 1 1 6 1
Total A 240 176 4 4 83 61 1 5 18

B: Spec. Cases Total Sub X V 0 1 2 3 31 4 5 6

quant 'when' 13 8 1 7
temporal phrase 4 1 1
conditional si 63 29 4 15 8 2
Total B 80 38 0 1 4 23 0 8 2

C: Wh excl. que Total Sub X V 0 1 2 3 31 4 5 6

Q.I.: si 6 3 1 1 1
Q.I.: wh # que 5 2 1 1
NP Appos. Rel. 42 6 4 2
CP Appos. Rel. 26 15 2 4 7 2
R.Rel. # que 7 4 2 1 1
comme 'as' 108
comme dit est 54 54
comme XP V 37 2 23 1 9 2
Total C 194 121 2 0 14 88 2 9 6
Total A, B, C 514 335 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. 047061
I don't know if since then they enjoyed them.
c. ...comme cy apres vous oyres. 020000
...as thereafter you will hear.
(23) a. ...puis que aultrement 03 n'en pouvoit faire... 016042
...as differently [he] could not do...
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 065111
I don't know if since [they] talked about it.
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 be-
loved
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. comple-
ment 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 comple-
ment clauses, as in (24):12
(24) a. ...il affermoit...que se n'estoit il point. 005025
...he was swearing...that that this was he not
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 pronomi-
nal 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 029023
...of the measure that so [you] forgot
d. va a tous les deable, que jamais plus 01 ne te voie. 066046
go to the devils, that never more [I] see you
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. 039083

...when me [you] can not touch
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 sim-
ply 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 identi-
fication 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 wh-
clauses.
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 rep-
resented 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 li-
censing 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 te-
dious this may be at times.

Notes
* This research has been made possible thanks to grants 410-87-0332 and 410-
89-1131 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 prohi-
bition 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 de-
fends the idea that empty expletives must be (i) licensed by being head-gov-
erned 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 wh-
clauses 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). Consider-
ing embedded clauses, an empty SpecIP could not be licensed by a [+F] fea-
ture 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 embed-
ded clauses the finiteness feature would be located in C0 only in those wh-
clauses not introduced by que (as I consider that the interrogative que is a wh-
word 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 well-
established 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 environ-
ments 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." Re-
vue 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 Pro-
drop 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 devel-
oped different systems in which subject clitics have specialized as the pro-
drop 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 wide-
spread 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 com-

peting head that replaces some of Agreement's syntactic functions. The analysis
of this phenomenon can thus help us to define the mechanisms that are in-
volved in the syntactic mapping of the relation between a subject and its predi-
cate.

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 mor-
phological 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 pro-
jections, this analysis of subject clitics can be rewritten as in (1) where sub-
ject 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 mor-
phemes 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 Venetian
Your mother not comes
(3) Elle ne vient pas French
She not comes not
(4) No la vien Venetian
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 sub-
ject 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 coor-
dination:
(5) Nane lese el giornale e fuma un toscan Venetian
John reads the newspaper and smokes a cigar
(6) II lit le journal et fume un cigare French
He reads the newspaper and smokes a cigar
(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 coor-
dination, 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 un-
der 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 Trentino
The Mary she eats
(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 obliga-
tory even when the subject DP is a Quantifier phrase, which cannot be left
dislocated:3
(10) Tut *(l)'e capita de not Trentino
Everything it is happened by night
(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, be-
cause they have subject clitics, but their structure is similar to Standard Ital-
ian 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 re-
spect 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 base-
generation 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 pre-
cisely 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 theta-
role 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 sub-
ject 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... Veneto
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.
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 sub-
ject clitic can (indeed it must).
The explanation for the contrasts in (12), (13), (14) and (15) is that, as men-
tioned above, argumental subject clitics leave a trace in the basic subject po-
sition 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 vari-
ety 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 sub-
ject 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 previ-
ous 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 Ital-
ian 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. 2. 3. 1pl. 2pl. 3pl. expl.
a/e te/ti m. el a/e a/e m. i 1
f. la f. le
As Vanelli (1987) notes, these subject clitics do not present any of the fea-
tures 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 theta-
role. 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 ques-
tions 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 ungrammati-
cal. The only negative proof that can be given is the absence of such a se-
quence 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 coun-
terparts 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 possibil-
ity 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 sepa-
rately 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 sub-
ject 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 el-
ement 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 theta-
role to its subject and cases of expletive subject clitics with postverbal sub-
jects. 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 subjunc-
tive 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 plu-
ral 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, be-
cause it contains the selectional features assigned by the matrix verb. Never-
theless, 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 sub-
ject. 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 pro-
drop theory only if it is "visible" in some sense to be defined.
On the other hand, the distribution of argumental null subjects of first per-
son singular and plural and second person plural does not seem to be depen-
dent 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 li-
censed both in a main and in an embedded context. On the other hand, exple-
tives 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 fol-
lowing schema:
(29) MAIN CL EMBEDDED CL. EMBEDDED CL
-wh /subjunct. +wh /+subjunct.
expletive + + +
pro -theta
l.person + + +
sing.pro
l.person + + +
plur.pro
2.person + + +
plur.pro
 THE DIACHRONIC DEVELOPMENT OF SUBJECT CLITICS 305

2.person - - +
sing.pro
3.person - - +
sing.pro
3.person - - +
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 "ex-
tended pro-drop." The second group of null subjects which can only be li-
censed 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 reminis-
cent 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) MAIN CL. EMBEDDED CL. EMBEDDED CL.
-wh/-subjunct. +wh/+subjunct.
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
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 plu-
ral and second person plural null subjects are admitted both in main and em-
bedded 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 be-
haves 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 exam-
ine how the system can be formalized within the context of the theory of pro-
drop 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 pa-
rameter 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 sec-
ond 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 sub-
jects 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 +per-
son and +number feature.
A system like that described in (31) and (32) generates the split between ex-
tended 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 vari-
ety 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 ex-
presses 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 pro-
nouns 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 infor-
mation about the subject.
As in RVe subject clitics are not always distinguished for person and num-
ber, so RVe has no other choice than to maintain the person and number fea-
tures on Agreement, reinterpreting subject clitics as heads for the persons that
are not fully specified by verbal morphology. French on the other hand, hav-
ing 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 fea-
tures 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. SEC-
TION 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 pre-
sented 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 Accusa-
tive, because the verb is an ergative one. It cannot be the Partitive Case pos-
tulated 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: Spec-
head 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 Rob-
erts 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. Fol-
lowing 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 as-
signed 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 morpho-
logical 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: Nomi-
native 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 num-
ber 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 sub-
ject 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 num-
ber with the verb. Hence also in RVe the AgrP projection is active for Nomi-
native 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 re-
ceives Partitive Case, because it is a definite DP. It cannot receive Nomina-
tive 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 can-
not be in the same chain and share the same Case, while an overt and a silent
element can. Considering the Nominative Case-assignment possibilities ex-
pressed in (34), we can make the hypothesis that Case is assigned to the post-
verbal 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 any-
thing 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 exple-
tive 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 theta-
role, which is assigned inside the VP.
The only situation in which the two Nominatives can be assigned thus cor-
responds 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 ab-
sorb 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 Ital-
ian 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 ele-
ments 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 struc-
ture 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 gram-
mar: 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 sub-
ject 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 vari-
ety 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 knowl-
edge 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 sub-
jects 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 end-
ings of Tense and agreement is quite often observable in this group of lan-
guages. 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 con-
servative 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 sub-
jects 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 posi-
tion to the head of AgrP where the inflected verb is placed after having incor-
porated 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 posi-
tions (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 posi-
tion.
As (44) illustrates, subject clitics move to the head of AgrP. We can imag-
ine 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 influ-
ence 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 sys-
tem. 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 phoneti-
cally 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 lan-
guage, in which a subject pronoun must always be present. The subject pro-
noun 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? French
When will come John?
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 construct-
ing 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. 2. 3. l.plur. 2.plur 3.plur
i ti el/la i i i/le
When a subject DP is not realized, a subject clitic is obligatory:
(51) a. Sa *(ti) me dive la paca Rovignese
If you to-me give a hit
b. * (A) ta par
It to-you seems
The subject clitic is not obligatory when there is a phonetically realized sub-
ject DP:
(52) a. Se Paron Giacomo gira furbo... Rovignese
If Mr. Giacomo was clever
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 can-
not 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 per-
sons 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 Dia-
lects is connected to, at least, three components of the grammar: the pro-drop
parameter, the Case-assignment conditions, and the visibility of empty Agree-
ment heads. It is possible to summarize the entire discussion about the de-
velopment 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 ap-
pear 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 morphol-
ogy following Renzi and Vanelli's (1983) generalization that the subject fea-
tures 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 syn-
tactic 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 sub-
ject 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 sub-
ject 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 verb-
second 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 morpho-
logically 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 phenom-
enon 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 pre-
sented 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 in-
stance, 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-assign-
ment configuration only once: if the subject gets Nominative from T, it can-
not 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 Chang-
ing. 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 19:1-
34.
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, 115-
225. 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 Param-
eter, 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 Agree-
ment." 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 Repre-
sentations." In R. Friedin, ed. Principles and Parameters in Compara-
tive Grammar, 417-454. Cambridge, Mass.: MIT Press.
Chomsky, N. (1992) "A Minimalist Program for Linguistic Theory." MIT Oc-
casional 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: Configu-
ration, Parameters, and Empty Categories. Cambridge: Cambridge Uni-
versity Press.
Giorgi A. and A. Pianesi (1991) "Toward a Syntax of Temporal Representa-
tion." 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 Climb-
ing." In L. Tasmowski and A. Zribi Hertz, eds. De la musique a la
linguistique. Hommage a Nicolas Ruwet. Communications and Cogni-
tion, 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 1:1-
30.
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 posi-
tions. Subsequently, I will show that the same interpretation is valid for other
Medieval Romance languages, which differently from Old French and North-
ern 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 fol-
lowing 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 posi-
tion; 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 sen-

tence-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 con-
texts where it cannot follow the verb, both in sentence-initial and sentence-
internal 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 re-
garding 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 consid-
ering are verb-second languages, and that the relevant variation observed is
the effect of a parametric difference, namely of the activation of a Top posi-
tion 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 fol-
lows:

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 projec-
tions; 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 sub-
ject 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 regard-
ing 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 centu-
ries, 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 phenom-
ena 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-sec-
ond 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. the verb moves into the position of the head of CP;
2. a constituent moves into the position of SpecC;
3. no other constituent moves to the front of the sentence, or, in the de-
scriptive 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 con-
sidered 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 non-
argumental 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 light-
eninged'.
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 di-
rect object in first position always has a resumptive pronoun. In other Mod-
ern 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 operator-
like 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 be-
fore 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 quali-
fications: if compared with NIDs, or Old Spanish and Old Portuguese, it ap-
pears 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 sen-
tences 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 ad-
jacent 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 re-
turn 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 verb-
second, 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 occa-
sions for the verb to be in first position, but this does not imply that the con-
straint 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 posi-
tion: 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 331

(9) 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, struc-
turally 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 pro-
hibits 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 constitu-
ent 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 de-
scribed 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 pos-
sible preverbal constituent. In this sense I continue to consider them to be
verb-second languages, though the verb is not constrained to surface in sec-
ond position. The behaviour of these languages, being more richly varied,
gives us the possibility to check some predictions in a wider range of con-
texts.
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 asymme-
try 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 gath-
ers 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, un-
like 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 assump-
tion, 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 speci-
fied 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 pro-
noun:
(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 Medi-
eval Italian texts with regard to clitic pronouns. In contrast with the French
type, the Italian type, while prohibiting clitics from beginning a sentence, al-
lows clitics to follow a verb in sentence-internal position. In Mussafia's view,
the sequence "verb-clitic" is never ungrammatical, only the sequence "clitic-
verb" 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 sup-
port 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 posi-
tion 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 consis-
tently 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 depen-
dent 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 fol-
lowing 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 pro-
nouns 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 for-
you-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 constitu-
ent of the main clause (or a SpecCP occupier)—in which case enclisis is im-
possible—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 al-
ways 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 hypoth-
esis. Since we have ways to characterize the SpecC with respect to Top posi-
tion, 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 coordi-
nate 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 re-
spect 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 coor-
dinate sentences depends on the level where coordination occurs: if the at-
tachment 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 move-
ment 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-ques-
tions. 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 pro-
noun 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 complemen-
tary 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 Hang-
ing Topic is only chi facesse contra, while la prima volta is a separate con-
stituent, 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 pe-
riphery of the sentence come from the third group of Medieval Romance lan-
guages 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 vari-
ant 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 avail-
ability 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 ob-
served 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 con-
firmed by the available data.
But including pro-drop Medieval Romance in the discussion, and in par-
ticular 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 im-
portant, 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 obliga-
tory).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 re-
stated 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 semi-
nar in Venice in 1989 and at the First Generative Diachronic Syntax Confer-
ence 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 observa-
tions 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 be-
tween 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 pic-
ture, 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 al-
ways to be enclitics, as in Modern Italian (excepted some cases, in fact), but
again the position of complement clitics relative to gerund and infinitive de-
pended 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 pos-
sibility 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 ap-
pears, 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, how-
ever, 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 comple-
mentary distribution of enclisis and lexical or abstract elements in CP (wh-
elements 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-ele-
ment. 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 coor-
dinate 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 docu-
ments. 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 sug-
gesting 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 Span-
ish: 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. Mar-
tin, 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 pat-
terns 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 clitic-
head relation and the w/t-construction.

1. Two Types of Verb Third

OHG syntax was characterized by the verb-second constraint (as in Mod-
ern German). This fact is clearly noted in the well-known traditional descrip-
tions, 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 oc-
cupy, 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 excep-
tions 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 pa-
per although we do not discount the possibility of a genuine syntactic expla-
nation 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 verb-
third 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 con-
struction 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 fol-
lowing properties hold for OHG:
a) OHG prose was characterized by the verb-second constraint
(subject-verb inversion in the main declarative clause);
b) the finite verb may shift to third position given the presence of a pro-
nominal 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) this construction is well attested in Old English prose (on the con-
trary, in OHG the construction under consideration is attested only
in the translation of Isidor and partially in the Monsee-Wiener
Fragmenten);
b) while in OHG the pronoun which intervenes between the fronted
element and the finite verb generally corresponds to the subject pro-
noun, 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 impor-
tant 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 summa-
rized by the following schema (cf. Kemenade (1987:139)):
(6) Main Clause:
a. XP - pronoun+Vfnt . . . . . .
b.(WH)

w
ne - Vfnt+pronoun - / * 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 de-
clarative 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 349

(7) 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
Comp Subj PP Vfnt NP NP(Dat) ObObj
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 comple-
mentizer - 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 verb-
second 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 sub-
ject 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) the lexical complementizer in the subordinate clause;
ii) 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 in-
teresting 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 sub-
ordinate clause:
(11) a. Comp Vfnt (verb-last)
b. Comp vfnt XP fnt
xp
(Extraposition of XP)
c. Comp .......V Vfnt (verb-last)
d. Comp... ....Vfnt V (VR)
e . Comp NP(Subj) V fnt NP(Obj) V (VPR)
Concerning (11), the following observations should be made:
i) While both OE and OHG clearly have OV order inside VP, the posi-
tion 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 oc-
cupy 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, Auxiliary-
Past 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 dialecti-
cal 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 sub-
ject 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 or-
der patterns can be found it would thus be possible to explain the two differ-
ences 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 in-
side IP (I°-VP versus VP-I°). More explicitly, our hypothesis is the follow-
ing: the syntax of both OHG and Modern German is characterized by the
verb-second phenomenon; the difference between these two grammars con-
sists 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 - +

West Flemish + + + +
Zurich Swiss German + + + +

German + + _ +
Dutch + + - +
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 XP-
pronoun-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 un-
der discussion, which does not seem to hold the other way round in
Williram (OHG) and more obviously in two Modern Germanic lan-
guages like West Flemish and Zurich Swiss German. In these lan-
guages, 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 re-
sult of V movement to a medial I0.8 Here in fact the correlation be-
tween the two word order patterns under consideration (verb-third
cases in the main declarative clause and bracket structure in the sub-
ordinate clause) seems to hold in both directions. The precise na-
ture 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) the relative order pronoun-Vfnt
b) the positions of the Vfnt in the subordinate clause.
In this section it will be tentatively assumed that every analysis which cap-
tures OE data should also capture OHG data and vice versa. Concerning the
word order pattern XP pronoun Vfnt . . . . . , two analyses have been put for-
ward 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 Ger-
man 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 S-
initial 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) 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).
ii) 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 pro-
posed 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 as-
sumption 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) if the topic position (= [Spec, ConflP]) is occupied by an operator
(wh-elements, pa and ne) the specifier and the head of ConflP ap-
pear to behave as one constituent within which cliticization is im-
possible;
ii) 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) ver-
sus (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) in verb-second Germanic languages the morpho-syntactic character-
ization of C° resembles, in many respects, the characterization of I°
in Romance languages;
b) this fact makes the postulation of an "extra," independent IP super-
fluous.
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) 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 differ-
ent syntactic processes);
ii) 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 359

iii) 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 head-
medial 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) be-
low:
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-me-
dial (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 ty-
pological 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 morpho-
syntactic 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 his-
tory of German the loss of cliticization corresponds to the strength-
ening 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 differ-
ent 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 dis-
tinct 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) Concerning the history of English, the postulation of an intermedi-
ate 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 con-
cerned, 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
II) 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 viola-
tion 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 sug-
gested to me by Ans van Kemenade and Tony Kroch (p.c.)), then
we would still need an explanation for why I° can never repre-
sent 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 fur-
ther 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 speci-
fier 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 inter-
related assumptions:
i) the wh-element, the clitic of negation (ne) and pa are operators;
ii) 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 opera-
tors, 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 conjunc-
tion. 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 wh-
element 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 wh-
elements 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 particu-
lar 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 lan-
guage crucially relies on the wh-criterion.
In fact, if we assume (following Rizzi (1990a:378-379)) that the feature
[+wh] may be generated:
i) in C° in the subordinate clause (through selection by the matrix verb),
ii) 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 verb-
second 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 at-
tributed 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-crite-
rion.
At this point it is important to note that:
a) 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));
b) 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) How does cliticization modify the status of the head?
II) Which principle underlies the constraint proposed in (41)?
Concerning II), note that cliticization has been generally analyzed as a pro-
cess 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 preposi-
tion 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) cliticization phenomena in Romance languages versus cliticization
phenomena in Germanic languages;
ii) cliticization of the subject pronoun versus cliticization of object pro-
nouns.
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 prelimi-
nary 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 syn-
tax is characterized by a process of extraposition, then (1c) could be ana-
lyzed, 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 convinc-
ing 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 pres-
ence 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): Enti sohuuer soquuidit le uuort uuidar mannes sune
Coor. REL. Vfnt O SP
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
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]) inter-
vene 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 char-
acterized 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, how-
ever, 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 charac-
terized by an SIOV stage finds an interesting parallelism within the typologi-
cal 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 as-
sumed 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 dis-
cussion 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 66:146-
152.
Behaghel, O. (1923-32) Deutsche Syntax. Eine geschichtliche Darstellung.
Heidelberg: Carl Winter.
Besten, H. den (1983) "On the Interaction of Root Transformations and Lexi-
cal 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 necessar-
ily) 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 17:417-
466.
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, Func-
tional 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 His-
tory 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 Dia-
lects." 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, 375-
385. 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 Analy-
sis 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 Linguis-
tics 26.2
Travis, L. (1984) Parameters and Effects of Word Order Variation. PhD Dis-
sertation, MIT.
Weerman, F. (1989) The V2 Conspiracy: A Syncronic and Diachronic Analy-
sis of Verbal Positions in Germanic Languages. Dordrecht: Foris.