Levels of Syntactic Representation [Reprint 2016 ed.] 9783110874167, 9783110130898

Table of contents :
Preface
Table of Contents
1. A Note on Government and Binding
2. Clitics and Binding in Spanish
3. French predication and linguistic theory
4. Logical Operators, Complete Constituents, and Extraction Transformations
5. Unambiguous Paths
6. Configurational Grammar
7. Empty quantifiers, LF-movement, and the ECP in French
8. On Case and Impersonal Constructions
9. Remarks on government, thematic structure and the distribution of empty categories
10. Aspects of multiple wh-movement in Polish and Czech

Citation preview

Levels of Syntactic Representation

Studies in Generative Grammar The goal of Studies in Generative Grammar is to publish those texts that are representative of recent advances in the theory of formal grammar. Too many studies do not reach the public they deserve because of the depth and detail that make them unsuitable for publications in article form. We hope that the present series will make these studies available to a wider audience than has been hitherto possible.

Jan Köster Henk van Riemsdijk

Robert May/Jan Koster(Eds)

Levels of Syntactic Representation

Ψ 1981 FORIS PUBLICATIONS Dordrecht - Holland/Cinnaminson · U.S.A.

Published by: Foris Publications Holland P.O. Box 509 3300 AM Dordrecht, The Netherlands Sole distributor for the U.SA. and Canada: Foris Publications U.S A. P.O. Box C-50 Cinnaminson N.J. 08077 U.S .A.

ISBN 90 70176 32 7 (Bound) ISBN 90 70176 30 0 (Paper) © 1981 Foris Publications - Dordrecht No part of this book may be translated or reproduced in any form, by print, photoprint, or any other means, without written permission from the publisher. Printed in the Netherlands by Intercontinental Graphics, H.I. Ambacht.

Preface

Few generative linguists will deny that 1979 was a memorable year for their field. In April, Noam Chomsky presented his lectures on government and binding at the Scuola Normale Superiore in Pisa (Studies in Generative Grammar 9). These lectures were not only the culmination point of the seminal work of the 1970s, they also marked the beginning of a new era: Government-Binding Theory. The new GB framework led to immediate reactions, some skeptical, others enthusiastic, but almost everywhere in the field there was a sense of excitement and a strongly felt need for discussion. It is in this climate that we decided to organize a colloquium. Preparations were made in Nijmegen, the Netherlands, where we were doing our research at the time, thanks to grants from the Max-Planck-Institut fur Psycholinguistik and the Netherlands Organization for the Advancement of Pure Research (Z.W.O.). The colloquium was held in Paris in December 1979. The articles in the volume are all based on presentations given during the colloquium. Against the stimulating background of Chomsky's Pisa lectures, the nature of syntactic representations is discussed on the basis of languages as diverse as Czech, Dutch, English, French, Norwegian, Polish, and Spanish. Because of its variety and scope, this volume is not only a unique document of a memorable period, it also gives a rich overview of current research in generative grammar. Needless to say, we did not organize the colloquium all by ourselves. We had generous support and cooperation from the Max-Planck staff in Nijmegen and from our friends and colleagues in Paris. In particular, we would like to thank Pierre Jacob, Richard Kayne, and Jean-Claude Milner for helping us to find such nice settings for our meetings in the Ecole Polytechnique, the Université de Paris à Vincennes, and the Ecole Normale Supérieure. Last but not least we would like to thank all the participants for making it an unforgettable colloquium. Robert May

Barnard College, Columbia University Jan Köster Department of Language and Literature, Tilburg University

Table of Contents

Preface

V

1. A note on government and binding Hans Bennis

1

2. Clitics and binding in Spanish Reineke Bok-Bennema

9

3. French predication and linguistic theory Daniel Couquaux

33

4. Logical operators, complete constituents, and extraction transformations J. Guéron

65

5. Unambiguous paths Richard S. Kayne

143

6. Configurational grammar Jan Köster

185

7. Empty quantifiers, LF-movement, and the ECP in French Hans-Georg Obenauer

207

8. On case and impersonal constructions J.-Y. Poüock

219

9. Remarks on government, thematic structure and the distribution of empty categories Knut Tarald Taraldsen 253 10. Aspects of multiple wh-movement in Polish and Czech Jindñch Toman

293

Chapter 1 Hans Bennis University of Amsterdam

A Note on Government and Binding

The purpose of the workshop on 'Levels of Syntactic Representation' (Paris, 7-9 December 1979) was to discuss research in progress. The talk I have presented consisted essentially of two parts. One was about the quantitative construction from a comparative point of view; the other was a proposal for a change in the Government-Binding theory, as was presented by Chomsky in Pisa (1979). In this article I will concentrate on the second part, which has the more speculative character of research in progress. The comparative study of the quantitative construction, which study was carried out in cooperation with Aafke Hulk, will be forthcoming elsewhere. Here I want to propose a few changes in the Government-Binding system and discuss the consequences. The main point of discussion in this paper is Proper Government and its definition. In Chomsky's (1979) proposal Proper Government is necessary in the formulation of the Empty Category Principle (ECP), which conditions the appearance of empty nodes: (1)

[Nr e \ must be properly governed

In general empty categories are traces of wh-movement and NP-movement (but not PRO, which in contrast to empty categories, contains pronominal features). The apparent asymmetry between wh-traces in subject and object positions follows from the definition of Proper Government, in which Government and Coindexing are combined : (2)

Proper Government α properly governs β iff 1. a governs β and 2. a) α = [±N, ±V] or b) α is coindexed with β

(3)

Government α governs β iff α minimally c-commands β

2 (4)

Hans Bennis Minimal c-command α minimally c-commands β iff a c-commands β and there is no y such that α c-commands y and y c-commands β and not y c-commands α and: 1. a (and y) =_[±N, ±V] 2. there is no S or NP bracket between α and β

The definition of Proper Government (2) should allow the trace of whmovement from subject position to COMP to be properly governed so as not to violate the Empty Category Principle. However interpreting definitions (2), (3), and (4) strictly, this trace is not properly governed. Although the trace in subject position is coindexed with the wh-phrase in COMP as a result of movement, following definitions (3) and (4) the wh-phrase in COMP does not govern the trace: since neither the wh-phrase in COMP, which is an NP, nor the COMP node itself is a lexical category, as is required by condition 1. in (4). Consequently one of the conditions in the definition of Proper Government is not fulfilled. How can this problem be solved. We could change the definition of Proper Government in such a way that government is only relevant if a = [±N, ±V] and not if condition 2.2b applies. However that would lead to other problems, since then whenever a phrase is coindexed with some phrase it is properly governed. In particular, PRO, in cases of control, would be properly governed. Alternatively, suppose that the definition of Proper Government in (2) is retained, but that the definition of Government (in fact the definition of minimal c-command (4)) is altered so that COMP (or something in COMP; I will not discuss that topic now, but I assume for the moment that COMP is a governor and can inherit an index of an wh-phrase which it contains) governs the subject. The proposal then is to add COMP to the first condition of the definition of minimal c-command (4). Now the whtrace in subject position is properly governed because it is governed by and coindexed with COMP. There are a number of consequences to this redefinition. For instance, it follows that subjects are always governed either by COMP, j n tensed sentences and most infinitives, or by the matrix verb, in case of S-deletion in Raising- and ACI-constructions. The implication is that PRO is governed too, contrary to the idea that PRO occurs only in ungoverned positions. At first glance this might seem an unfortunate consequence. But as can be seen from the definition of Proper Government PRO is not properly governed, since PRO is not coindexed with its governor (COMP) nor is PRO governed by something of the category [±N, +V]. So the difference between empty categories and PRO can be expressed by Proper Government: empty categories have to be properly governed (ECP) but not PRO.

A Note on Government and Binding

3

An advantage of the extension of governors to COMP is that there is no need to provide special status for nominative case. In the GovernmentBinding theory Case assignment takes place on the basis of government with the exception of nominative case, which is assigned structurally to anything outside VP in tensed sentences. In my proposal nominative case can be assigned if the subject is governed by a COMP which contains the feature [+Tense]. So Case is always determined by the governor. This proposal requires the postulation of a feature [±Tense] in COMP, for which I think there is some motivation. In the first place the form of the lexical complementizer agrees with the finiteness of the verb. Secondly, in Dutch root sentences the finite verb is moved into COMP, as is convincingly argued by Den Besten. Given the optionality of the expansion of the base rules, it is possible to have [+Tense] in COMP and PRO in subject position. Assuming that case assignment depends on government, the result will be a Case-marked PRO (nominative). I think it is possible to consider this Casemarked PRO as the PRO in tensed sentences in languages like Italian and Spanish, where the subject of a tensed sentence need not be lexically realized. Because of the language specific possibility of allowing this sort of PRO-subject, we have to introduce a parametrized condition, permitting the occurrence of P R O [ + c a s e ] . This would make the introduction of a category Agreement no longer necessary, which was introduced especially to capture these facts. In conclusion, I think that the addition of COMP to the first condition of minimal c-command makes it clear that even small changes can have rather wide-spread consequences. In the next section I want to propose another change in the definition of minimal c-command, made possible by the introduction of COMP as a governor. Considering COMP as a governor, it is possible to drop the second condition in the definition of minimal c-command: "there is no S or NP bracket between a and β". This means that government goes on until it reaches a domain of another governor. Furthermore a governor itself is governed by another minimally c-commanding governor, the main consequence of which is that wh-traces can be considered to be (non-lexical) anaphors instead of having the status of names. Now, anaphors are characterized by the condition that they have to be bound in every governing category. If we take S and NP to be the governing categories, it is clear that a wh-trace in NP-position can be considered as an anaphor if it is bound by a wh-phrase in COMP. Dropping the second restriction on minimal c-command, it becomes possible to consider wh-traces in COMP (intermediate traces) as anaphors too, since now COMP, which is a governor itself, is governed by a minimally c-commanding governor (V in complement clauses and Ν in relative clauses). As a result, the Minimal Governing Category of a governed complementizer is not the S by which it

4

Hans Bennis

is immediately dominated, but the S or NP of the higher clause. Now the relation between a wh-trace in COMP and a wh-phrase in a higher COMP can be considered an anaphoric relation, because the wh-trace is bound by the wh-phrase in its minimal governing category. Successive cyclic movement follows and at least in these cases there is no need to postulate Subjacency. The Complex NP Constraint follows too, since NP is a governing category; if we should move a wh-phrase from COMP to COMP out of an NP, the trace can no longer be considered as an anaphor because it is not bound in its minimal governing category (NP). Another consequence which I will only mention, but not pursue here, is that now every category is governed except the highest COMP. This exceptional status of the COMP in root sentences is attractive, as it could well serve as the basis of an explanation of root phenomena, like the position of finite verbs in root sentences in Dutch and German. A third consequence is that the fact that verbs in a matrix clause can select a complementizer in the embedded clause can be easily explained. To do this we have to connect government and subcategorization. Suppose that we consider subcategorization as a special case of government. This implies that only governors ([±N, ±V] and COMP) can subcategorize and that the domain of subcategorization is the domain of government. Consequentially verbs can be subcategorized for objects, but also for the complementizer of a complement clause, because the verb governs the complementizer. Since COMP is a governor, it follows that COMP is (can be) subcategorized for subject and verb, so we might say that the subjectpredicate structure of a sentence is primarily determined by subcategorization. Before discussing a few problematic cases, I will summarize the proposals and their consequences. The Empty Category Principle (1) has been replaced by the Anaphor Principle below (8), with the definitions of Proper Government (2) and Government (3) remaining unchanged. The definition of minimal ccommand (4) has been altered to : (5)

a minimally c-commands β iff α c-commands β and there is no y such that a c-commands y and y c-commands β and not y c-commands α and: a (and y) = [±N, ±V] or COMP

There are three kinds of NP's: 1) anaphors

- NP's that are lexically identified as anaphors: reflexives and reciprocals

A Note on Government and Binding

5

- empty NP's: wh-traces, NP-traces and base generated empty NP's 2) pronominals - pronouns: pronominale with a phonetic matrix - PRO: pronominals without a phonetic matrix 3) other, lexical NP's The Binding Theory can be formulated as: A: If α is an anaphor it is bound in every governing category B: If α is a pronominal it is free in its minimal governing category C: Other NPs are free in every governing category As can be seen Case plays no role in the Binding Theory. For that reason Case-assignment could even take place on the phonological side of the grammar. Nevertheless I will assume Case is always assigned by the governor of the NP either in the base, in which case we have inherent Casemarking, (which goes along with subcategorization) or at the level of S-structure, where Case is structurally assigned. Moreover, it must be stipulated that sometimes no Case or [-Case] is assigned. For example we could argue that in passive structures no Case is assigned to the object of the governing passive participle, to avoid Case-conflict if the object is preposed into subject position. Similarly a COMP which is specified as [-Tense] doesn't assign Case to the subject to avoid nominative subjects in infinitival constructions. Finally there has to be a Case-filter on the phonological side of the grammar which excludes non-case-marked lexical (with a phonological matrix) NP's. This Case-filter is the same as in the Government-Bin ding theory: (6)

* [NP_ case phonetic matrix]

Figure 1 indicates which NP's can be distinguished at the level of S-structure : (7)

1 2 3 4
V (CÌ)... b. CI -*• ([+reflexive] )

([-reflexive] )

(where . . . contains the position for V complements like direct and indirect objects) For reasons of exposition I will restrict myself to dialect b. in the remainder of this article. I have said that [—person] clitics can be treated as NPs in the base and now I wish to suggest that they should be treated as such; indeed nonhuman pronominal direct object NPs must be clitics. The question (14a), for example, cannot be answered by means of a strong pronoun which refers to la mesa (14b), if one uses a pronoun it must be a clitic (14c): (14) a. Viste la mesa? Did you see the table? b. *Si, vi ella Yes, I saw she c. Si la vi Yes, I saw it TTiese pronominal NPs acquire clitic status (i.e. a status equal to that of [+person] clitics) at a certain point in the derivation. For details, see section 5.

16

Bok-Bennema

3. M-BINDING AND A-BINDING The question I wish to address at this point is: what kind of relationship exists between a [+person] clitic and the direct or indirect object to which it refers? The answer implies some revisions of the GovernmentBinding system presented by Chomsky in Pisa (Chomsky 1979). The changes I will propose are rather speculative and are developed specifically for the problems presented in this article. As always, they raise a great many questions in other areas of the overall theory. As can be observed from the examples in (15), strong object pronouns can be absent in Spanish : (15) a. María le vio (a él) María saw him b. María me dio el libro (a mi) Maria gave me the book c. María se vio (a sí misma) en el espejo María saw herself in the mirror What seems to be the relevant parameter here is that objects can be absent because of the presence of the clitic. I assume that the "absent" objects are null objects coindexed with, and hence bound by, the clitics. The indices are assigned by a general rule,along the lines proposed by Chomsky (1979). Below I will address the question whether the null objects should be considered as PRO or as e, i.e. whether they have features or not. The Government-Binding theory allows for two kinds of binding: "argument" binding by an NP and binding by an "appropriate operator". A [+person] clitic is not an NP (not an argument), nor does it seem plausible to consider it as a Logical Form operator. Another problem is that the null objects are in case positions, and the only null NPs with case that are allowed by the theory are variables. To accommodate clitics in the system I want to introduce a new binding concept: "morphosyntactic binding" (M-binding). M-binding is defined as follows: (16)

M-binding: a M-binds β, iff α is coindexed with(3, and α c-commands j3, and ais a morphosyntactic category and β is an argument

M-binding differs from A (argument) binding in defining a relationship between a morphosyntactic category (e.g. a clitic) and an NP, whereas Α-binding holds for two NPs. I reproduce the definition of Α-binding in (17): (17)

Α-binding: a Α-binds β, iff ais coindexed with β, and ac-commands β, and α and β are both arguments.

Clitics and Binding in Spanish

17

Languages differ in morphosyntactic categories, and in the possibilities for these categories to serve as binders. M-binders can be: clitics, Agreement in "Subject PROdrop" languages like Italian and Spanish, the category [+r] in Dutch (see below), and wh-phrases (and their traces) in COMP. In (18) I reproduce the binding theory as presented by Chomsky (1979): (18)

A. If NP is an anaphor or has no phonetic matrix (1)it is a variable, or (2) it has to be Α-bound in every governing category B. If NP is case marked and nonpronominal (1) it is an anaphor (2) it has to be Α-free in every governing category C. If NP is pronominal (a pronoun or PRO) it has to be Α-free in every minimal governing category

In order to incorporate M-binding, a revision of this theory is necessary. I propose that A. be reformulated as follows: (19)

A.l. If NP has no phonetic matrix and is case marked it has to be M-bound in every governing category A.2. If NP has no phonetic matrix and no case or is an anaphor it has to be Α-bound in every governing category

B. and C. remain as in (18) This reformulation has various consequences. For one thing: case marked PROs are now permitted, if they are M-bound in their minimal governing category (MGC). I will return to this below. Another consequence is that variables are no longer treated as a separate category. They have case and no phonetic matrix, so by A.l. they have to be M-bound in every governing category. I consider S as a governing category for M-binding (see below), thus wh-extraction out of a (minimal) S is only possible if it leaves a trace in the COMP of that S, which will M-bind the ultimate trace (unless there is another M-binder). This has a number of advantages, but also poses certain problems. Succesive cyclicity is partially explained.7 May's QR (May 1977), which also leaves an empty NP with case, can be accomodated in the system if one assumes that its application is succesive cyclic as well. Furthermore subjacency no longer needs to be stipulated to block wh-extractions from NPs and from whislands: in both cases extraction would leave a case marked trace, which would not be M-bound in its MGC (the NP and the wh-island respectively).8 Another consequence of the requirement that the ultimate trace of wh-

18

Bok-Bennema

movement be M-bound in its MGC is that there have to be doubly filled COMPs (containing the lexical complementizer and the trace of wh) in logical form in languages like English, even when the wh-phrase comes from an object position, as for example in (20): (20)

Who do you think that he saw?

This means that the ungrammatically of (21) cannot be explained by means of an analysis requiring a binding wh-trace in COMP for subjects; but not for objects (as Chomsky (1979) does, using an ingenuous explanation involving the doubly filled COMP filter (see Chomsky and Lasnik 1977, and Pesetsky 1979) and a rule of free deletion in COMP, which applies at S-structure). 9 (21 )

*Who do you think that saw him

There is a difference between Α-binding and M-binding with respect to the categories that count as governing categories. As wh-phrases in COMP are crucially involved in M-binding, in the manner indicated above, one has to assume that S is a governing category for M-binding, whereas for Α-binding S may be the relevant category, see Chomsky (1979). Another difference concerns PPs. Thus, van Riemsdijk (1978) argue? that PPs function as islands with respect to extraction, unless they have a morphosyntactic "escape hatch". In Dutch this escape hatch is the [+r] position, which is the first position in the PP. In (22) the case-marked object of the preposition naast can be empty because it is M-bound by er (a [+r] word) within the PP (see 22b). (22)a. b.

Ik zit ernaast I sit next to it Ik zit [pp er¡ naast e¡]

er can subsequently move higher, but that is not relevant here. What is relevant is that [+r] words are the only word that can leave the PP, via the [+r] position, leaving as a trace an M-binder in this position. Thus, (23) is not grammatical, because wie is not a [+r] word: (23)

*Wie zit je naast? Who are you sitting next to?

Obviously traces in PPs have to be M-bound within the PP, which means that PPs are governing categories with respect to M-binding. In (24) the reciprocal elkaar is Α-bound, but not in the PP:

Clitics and Binding in Spanish (24)

19

Wijj zitten naast elkaar¡ We sit next to each other

Given the grammaticality of (24) and the fact that elkaar has to be Abound in its MGC (by A.2.) one has to conclude that PP is not a relevant governing category for A-binding. 10 I leave aside a number of other questions concerning differences between the system presented here and the Pisan Government-Binding theory (some are mentioned in f o o t n o t e 11), and return to [+person] clitics in Spanish. 11 The null objects bound by these clitic fall under binding condition A . l . : they have to be M-bound in their MGC (which they are, by the clitic). Notice that A . l . implies that there are both case-marked (M-bound) PROs and case-marked (M-bound) e. An instance of the latter are the traces of wh-movement. The question to ask now is: should null objects, bound by clitics, be considered as PRO or as e in Logical Form? Recall that Chomsky (1979) inserts them as PRO's, which loose their features before Logical Form (Chomsky's system does not allow for case-markes PROs). I wish to argue, however, that they still are PROs at the moment the binding conditions apply. The behavior of the null object (φ) in (25) is exactly like that of pronominal NPs: (25) a. *Juan¡ le vio φ\ Juan saw him b. Juan¡ dijo que María le vio φ¡ Juan said that María saw him c. Juan¡ dijo que María le vio (33) a.

Paco desiste de verte Paco gives up (the plan) to see you b. *Paco te desiste de ver

22

Bok-Bennema c.

Paco quiere encontrarte Paco wants to meet you d. Paco te quiere encontrar e. Paco quiere que yo te encontre Paco wants that I meet you f. *Paco te quiere que yo te encontre

It seems rather ad-hoc to stipulate that certain verbs constitute a bridge for Clitic Climbing when their complement is infinitival. Furthermore, the verbs that permit Climbing are exceptional in another way, as we shall see below, so the stipulation of bridge conditions probably would imply a missed generalization anyway. The hypothesis that Clitic Climbing is a structure preserving movement that moves the whole cluster of clitics would explain the clustering facts, but not the restriction of Climbing to a certain kind of construction. On the basis of these observations I conclude that Spanish clitics do not undergo structure preserving movement. As clitic positions are generated freely, it is necessary to exclude this kind of movement in some way. One could stipulate, for example, that clitics do not belong to the class a of the "move a " rule. A more principled assumption is that structure preserving clitic movement is excluded by a restriction on the relationship between two morphosyntactic categories (a clitic and its trace in this particular case). I have as yet no theory about this kind of relationship, and leave the question open. It is often assumed that verbs that allow Clitic Climbing trigger a rule of Verb Raising, which adjoins the embedded verb to the matrix verb and at the same time changes the complex structure of the sentence into a simplex one. This rule would apply optionally. Clitics can "climb" when Verb Raising takes place, but not when it does not. A clear exposition of the Verb Raising rule (for Italian) can be found in Rizzi (1978). One of the arguments that shows that Verb Raising applies is based on Gapping phenomena: normally a complement verb cannot be ommitted together with a matrix verb, but with Clitic Climbing matrix verbs both verbs can undergo Gapping together: (34) a.

Paquito quiere tomar limonada y María quiere tomar cerveza Paquito wants to drink lemonade and María wants to drink beer b. Paquito quiere tomar limonada y María cerveza c. Yo desisto de comprar la casa y María desiste de comprar el coche I give up (the plan) to buy the house and Maria gives up (the plan) to buy the car d. *Yo desisto de comprar la casa y María el coche

23

Clitics and Binding in Spanish

The verb querer in (34a and b) permits Gitic Gimbing, it also permits its complement verb to undergo Gapping; desistir does not permit Gitic Climbing, Gapping of its complement verb, as in (34d), results in ungrammatically. 14 The assumption that Verb Raising involves a "Restructuring" or "Gause Union" process makes it possible to integrate Clitic Gimbing in the framework I developed in the preceding section. Restructuring would erase the S boundary of the complement. It also erases its S and VP boundary, as well as the PRO subject, but that is not relevant at the moment. Recall that (31) was ruled out because the PRO object was not bound in its MGC, which was the internal S. If this S is not present anymore and the sentence becomes something like (35) the MGC for the PRO object becomes the higher S, in which the PRO is bound by te : (31)

Paco tej q u i e r e [S p R O ver PROi]

(35)

[s Paco te¡ quiere + ver PROi]

There are, however, several reasons for rejecting a restructuring analysis. As Restructuring is optional (i.e. it does not apply when the clitics stay in the complement 15 ) the Logical Form structure of (28a) (without Restructuring) would be entirely different from that of (28b) (where Restructuring would have applied): (28) a. Paco quiere verte b. Paco te quiere ver

(a complex structure) (a simplex structure for the Restructuring hypothesis)

The assumption that these two sentences differ so much in structure seems to me at least implausible. Furthermore one would expect that a reflexive or reciprocal object in the complement sentence could be coreferent with the matrix subject after Restructuring: it would be correctly Α-bound in its MGC, the matrix S. This is not the case, as one can see from the ungrammatically of (36b): (36) a.

Los candidatosi votaron el uno por el otro The candidates voted for each other b. *Los candidatosi les ordenaron votar el uno por el otroi The candidates ordered them to vote for each other

0ordenar is a Clitic Gimbing verb) In the last place it is highly undesirable to allow the theory of grammar to

24

Bok-Bennema

posses rules that turn complex sentences into simplex ones, i.e. rules that erase S, S and VP boundaries, as well as PRO subjects. Though I accept the hypothesis that Verb Raising takes place in Clitic Climbing structures, I reject, on the basis of the arguments I just mentioned, that the same holds true for Restructuring. Neither base-generation in the matrix sentence, nor structure preserving clitic movement, nor Restructuring provide an explanation for Clitic Climbing phenomena. In the next section I will elaborate the hypothesis that Clitic Climbing is essentially a consequence of certain local processes, which apply in simplex as well as in complex sentences. These processes belong to the "lefthand" part of the grammar, i.e. to the part that is not visible in Logical Form. This means that in the Logical Form structures of both (28a) and (28b) the clitic te belongs to the complement sentence and correctly M-binds the PRO object of ver. I assume for various reasons that Verb Raising applies before S-structure, and that it leaves a trace.16 The S-structures of (28a) and (28b) differ because of the fact that in (28a) there has been no Verb Raising, and in (28b) there has. The S-structure of (28a) is (37a) and that of (28b) is (37b): (37) a. Paco quiere [s PRO ver te PRO] b. Paco quiere + νβη [s PRO ei te PRO] (where e¡ is the trace of the verb ver) 5. LOCAL CLITIC MOVEMENT RULES In the preceding section I discussed the problems arising from taking Clitic Climbing as belonging to the transformational part of the grammar. In this section I would like to consider an approach which treats Clitic Climbing as the result of local rules. As local rules are generally language specific, their position in the T-model of Core Grammar has been somewhat neglected. I think that one can argue that they apply after S-structure, in the "lefthand" (phonological) part of the grammar, at least in a number of cases. I will not elaborate this general idea here, but I would like to show that in the case of Spanish clitics this approach to local rules may lead to explanations that are otherwise rather difficult to attain. In section 2, I proposed the base rule (13a), which contains a clitic node (Q) as a righthand sister of V: (13) a.

V^V(Cl).

Clitics and Binding in Spanish

25

The order provided by this rule is correct when the sequence V CI appears in a tenseless clause; when the clause is tensed, however, the order is inverted: (38) a.

Después de presentársenos Juana . . . (tenseless) After Juana having presented herself to us . . . b. Juana se nos presentò (tensed) Juana presented herself to us

This inversion can be accommodated by means of a local rule of Verb Clitic Node Inversion, which is triggered by the presence of the feature [+Tense] in Aux: (39)

Verb Clitic Node Inversion: [+Tense] V 0 - > [+Tense] Cl V

Notice now that not only the clitics generated under the clitic node participate in this inversion, but also the clitics that are generated as NPs in the base (the [—person] ones): (40) a.

Después de dártefc» Paco . . . After Paco having given it to you . . . b. Paco te /odio Paco gave it to you

(tenseless) (tensed)

To account for this, one has to assume that the [—person] clitics (like lo in (40)) belong to the clitic node at the point of application of Verb Clitic Node Inversion. Suppose then that there is a rule that adjoins [—person] clitics to the clitic node. This rule has to apply before Verb Clitic Node Inversion. As direct object NPs immediately follow the clitic node and [—person] clitics are always the last in a sequence of clitics, this rule does not change the order of the string. It changes the structure, however. 17 (41)

lo(s), la(s) Adjunction

α

-»• C1+

' lo ' los la > las.

A problem arises when there is a clitical direct object NP, but no other clitics, i.e. no clitic node. In this case also lo(s), la(s) have to appear before the verb in finite clauses :

26

Bok-Bennema

(42) a. Después de verla . . . After having seen i t . . . b. Juana la vio Juana saw it

(tenseless) (tensed)

Let us suppose that this result is attained by a rule of Verb lo(s), la(s) Inversion that inverts the order Verb [-person] clitic in tensed sentences (this rule can easily be collapsed with (39)). An alternative would be to stipulate that a [—person] clitical NP is adjoined to V when there are no other clitics intervening, and that it becomes the clitic node itself, which then undergoes Verb Clitic Node Inversion if the clause is tensed. Other solutions are possible, but at the moment I see no principled way to decide among them. Notice, by the way, that I am not speaking about real (morphological) cliticization rules that turn the verb and its clitic(s) into one word, I assume that these apply later on. The rules described above account for the movement of clitics in simplex sentences, but also for Clitic Climbing in sentences containing verbal complexes, created by Verb Raising. Consider, for example, structure (43): Paco [+Tense] [y quiere + ver,] [s PRO e¡ te PRO]

(43)

Where the clitic node of the complement sentence contains te. To this structure the rule of Verb Clitic Node Inversion applies. This rule, applying at the left-side of the grammar, does not take into account the PRO subject, nor the trace of the raised verb, since neither PRO nor e¡ seem to have phonological relevance. It analyzes the complex verb quiere + ver as V. The resulting surface structure, after application of Verb Clitic Node Inversiones: (28) b. Paco te quiere ver Notice, however, what happens when the matrix verb has clitics of its own, as in (44): (44)

Juana le permitió mirarse Juana allows him to look at himself

If Verb Raising has applied, its S-structure is (45a). The elements that are relevant to Verb Clitic Node Inversion are those in (45b): (45) a. Juana [+Tense] permitió + mirar, le [s PRO e¡ se PRO] PRO18 b.

[+Tense] [y permitió + mirar] [elle] [cï se]

Clitics and Binding in Spanish

27

Here the clitic node which contains se cannot be analyzed as the third factor of rule (39) (Verb Clitic Node Inversion) because the first clitic node intervenes. Rule (39) applies (with the first clitic node as its third factor), and the resulting structure is (46): (46)

[+Tense]

[Q le] [Y permitió + mirar] [Q se]

As the verb and [+Tense] are no longer adjacent, there is no possibility for a second application of (39), because its structural description is not met. If the complement sentence of a Verb Raising structure contains a [-person] clitic, this clitic is either adjoined to the first clitic node to its left, or, if there is no such node, it "climbs" by Verb lo(s), la(s) Inversion. There are four different cases: (1) neither the matrix sentence nor the complement sentence contain other clitics (i.e. neither of them posses a clitic node 19 ) (2) the complement sentence contains a clitic node, but the matrix sentence does not (3) both the matrix and the complement sentence contain a clitic node (4) only the matrix sentence contains a clitic node These four cases are illustrated in (47): (47) a.

Juana quiere verlo Juana wants to see it b. Juana quiere dártelo Juana wants to give it to you c. Juana se permitió comprármelo Juana allowed herself to buy it for me d. Juana me permitió comprarlo Juana allowed me to buy it

In (47a) Verb lo(s), la(s) Inversion applies (providing that Verb Raising has taken place). The result is (48): (48)

Juana lo quiere ver

In (47b) Verb /o(s), la(s) Inversion cannot apply, because of the intervening clitic node (te), lo can be adjoined to this node (by (41)), however, and the whole clitic node will then "climb" by Verb Clitic Node Inversion, giving (49): (49)

Juana te lo quiere dar

28

Bok-Bennema

Notice that this explains the clustering facts that I described in the preceding section: in (47b) neither to, nor te can "climb" alone: climbing of lo alone is prevented by the presence of te and climbing of te (alone) by- the fact that the adjunction rule for [—person] clitics (rule (41)) precedes the rule of Verb Clitic Node Inversion. In (47c) to will be adjoined to the clitic node of the complement (by (41)), which contains me. As there is another clitic node in the matrix sentence, the clitic node of the complement cannot climb (i.e. we have exactly the same situation as in (44)); thus (50) is not grammatical: (50)

* Juana se me lo permitió comprar

(47d) is an interesting case. In spite of the fact that the matrix verb contains a clitic, or rather a clitic node, of its own, "climbing" of lo gives the grammatical (51): (51)

Juana me lo permitió comprar

At S-structure we have the following configuration: (52)

Juana [+Tense] permitió + mirar; me [s PRO e¡ [Np lo] ] PRO

As the clitic node that contains me is the first clitic node to the left of to, rule (41) will adjoin to to it (recall that the phonologically irrelevant PRO and e\ are not taken into account by the kind of rules discussed here). Afterwards the rule of Verb Clitic Node Inversion places the clitic node before the verb. Thus, though [+person] clitics cannot "climb" to a verb which has clitics of its own, [—person] clitics can. This is a result of the fact that [—person] clitics undergo a special rule (rule (41)), which transforms them from NPs to clitics, by adjoining them to the first clitic node to their left. The difference in status in deep structure between [+person] and [-person] clitics was motivated in section 2. This difference is maintained until rule (41) (or the rule of Verb to(s), la(s) Inversion) applies. Although there are still some problems concerning this approach, it has a number of advantages: it explains "clustering", it does not need any ad-hoc conditions concerning the impossibility of Clitic Climbing to a verb which has clitics of its own, it treats movement of clitics in simplex and complex sentences as unitary phenomena and it does not interfere with the binding requirements between clitics and object NPs.20

Clitics and Binding in Spanish

29

6. CONCLUDING REMARKS As is always the case with ongoing research, the present article leaves open a great number of questions. These concern certain details of the analysis as well as the more general hypotheses that relate to the organization of grammar. One of these hypotheses is that the grammar contains the notion of M-binding at Logical Form. I have presented no more than a first approximation to a theory which incorporates this type of binding. Further elaboration is necessary, not in the least for the wh-movement cases. I have tried to show that M-binding is useful for the description of the relationship between (Spanish) clitics and NP objects, but only if Clitic Climbing is considered to be the result of rules that operate in that part of the grammar that is invisible to Logical Form. The possibility that local rules — such as those that are involved in the movement of clitics — operate in this "lefthand" part of the grammar is also a point of further research.

NOTES 1. That clitics seem to form one word with the verb in (la), but not in (lb) is the result of an orthographic convention. 2. A late rule (the spurius se rule) converts le(s) to se when it is followed by fo(s) la(s). See e.g. Groos (1978). 3. Groos uses different features: [+reflexive], [-reflexive] and [afeminine]. I have chosen [+person] and [-person] both for reasons of exposition, and to exclude the undesirable a. Below I will show that the features characterize certain base positions, which only clitics can occupy. One has to exclude their being occupied by nonclitics containing the same features, of course. This means that the base positions wül also contain features that exclusively characterize clitics, say ^ p r o nominal] , [-stress]. Ihave left these out throughout. One should know, furthermore, that there are still some minor restrictions on co-occurence not explained by the analysis presented here. I refer to Groos' paper for details. 4. Nonhuman indirect objects are possible, see e.g. (i). I assume that these are treated as human by Spanish speakers: normally indirect objects are human, other recipients (institutions and the like) are interpreted in some way as "humanized". (i)

Le di el libro a la biblioteca I gave the book to the library

5. For example by Strozer (1976), by Rivas (1977) and by Groos (1978). 6. The a also precedes the "humanized" indirect objects, referred to in footnote 4. Nonspecific human objects are not preceded by a. As clitics always refer to specific objects, the nonspecific ones are not relevant for the present article. 7. It is explained as far as the first step of wh-movement is concerned. Subsequent (COMP-to-COMP) movement is permitted to cross several S and S boundaries, to exclude this, some condition on binding of traces in COMP is required. 8. Rizzi (1977) shows that Italian permits wh-island violations (both from

30

Bok-Bennema

subject and from object position), i.e. in Italian a wh-phrase can move over a filled COMP into the next higher COMP. It seems that in this case it is not the first S in which the (ultimate) trace of wh-movement has to be bound, but the second one. One cannot stipulate the MGC for M-binding in general is the second S in Italian, because for M-binding of null NPs by Agreement and clitics it surely is not. As a special stipulation for wh-traces is rather ad-hoc the question clearly merits further investigation. 9. To account for the ungrammatically of (21) revision of the ECP (Chomsky 1979) is necessary. Notice that languages which permit lexically doubly filled COMPs do not permit nominative subjects to be extracted by wh-movement. Thus, in the Dutch dialect which accepts this kind of COMPs, (i) is grammatical, but (ii) is not: (i) (ii)

Hi) vroeg me wie of ik kende He asked me who that I knew Hij vroeg me wie of kwam He asked me who that came

At GLOW 1980 Hans Bennis presented a paper ("Coindexing and Complementizertrace-phenomena") in which the differences in grammaticality between (i) and (ii), as well as those between (20) and (21) - the latter two with a trace in COMP were explained. 10. Sometimes a PP can function as MGC for Α-binding. This is probably the case in (i): (i)

He saw a snake near him

In Dutch all locative PPs can optionally serve as MGCs for Α-binding. See Vat (1980). 11. In the system presented here case-marked traces of NP movement are permitted by A.l. of (19), if they are M-bound in the MGC (e.g. by Agreement). This is not so in the Pisan system, which does not allow for case-marked traces of NP movement at all. However, in most cases condition Β (2) (of (18)) will rule out the resulting configuration: if the trace is c-commanded by the moved NP it is A-bound in a governing category, which is not permitted; NP movement to a position that is c-commanded by the trace is also excluded (the NP will be Α-bound in a governing category, by its trace). Certain cases of NP movement from case-marked positions will (also) result in ungrammatically because of case clash. It remains to be investigated if there are other instances of NP movement that leave a case-marked, Mbound, trace, which should be ruled out, but at present are not. Notice that movement to a non-argument position is permitted: this could explain the free Stylistic Inversion phenomena of Italian and Spanish - where Agreement binds the trace of the "inverted" NP-, provided that the target of Stylistic Inversion is a non-argument position. Another consequence of the new system, as it stands, is that it predicts that caseless NPs (e and PRO) can be M-bound. This prediction is wrong, however, (i), for example, does not have any interpretation: (i)

Juan¡ se¡ fue visto e¡ Juan (se) was seen

The solution for this problem of course has to do with case. One could stipulate that all M-binders fall under the case filter and inherit their case from the category they (ultimately) bind (as wh-phrases in COMP do in the Pisa system) which would explain

Clitics and Binding in Spanish

31

that they cannot bind a caseless position). Another possibility is that M-binders like Agreement and clitics have case of their own (nominative and objective, respectively) and that they have to agree in case with the NP they bind. 12. Notice that a caseless PRO can now be anaphoric (Le. [+reflexive]) if it is Α-bound in its MGC. This means that the subject of the complement of raising verbs and the NP bound by the subject of a passive can be an anaphoric PRO, as well as an e. Introduction of the θ-criterion (Chomsky 1980), however, will rule out sentences with these PRO, because no θ-role will be available for the antecedent (PRO has to have a 0-role, just like all other NPs with features). 13. By an entirely different line of argumentation Luigi Rizzi also arrives at the conclusion that null subjects bound by Agreement have to be PROs (Rizzi: "whMovement, Negation and the 'PRO-drop* Parameter", paper held at GLOW 1980). 14. It is not clear whether de in (34c) and (34d) should be treated as a complementizer, or as a preposition. In any case there are verbs that select these kind of prepositional elements and permit Gapping of the complement verb as one can see from (i) and (ii) (the preposition also disappears): (i) (ii)

Yo voy a comprar una casa y Paco va a comprar un coche I am going to buy a house and Paco is going to buy a car Yo voy a comprar una casa y Paco un coche

This means that the ungrammatically of (34d) cannot be due to the presence of de in (34c). 15. Rizzi (1978) claims that Restructuring is a necessary condition for Clitic Climbing. He is not sure that it cannot apply when clitics stay in the complement, but his analysis requires at least that it applies optionally in the latter case, and that is the relevant point here. 16. The main reason to consider Verb Raising as a rule which belongs to the transformational part of the grammar is that it interferes with other rules (see Rizzi 1980), e.g. the rule of Gapping. The trace is necessary for the interpretation of the complex structure. 17. Chomsky (1973) suggests that rules that "never change the terminal string of a phrase marker but only its structure" be constrained to the component of grammar which relates syntax to phonology. It is precisely to this component that rule (41) belongs. 18. The third PRO is the indirect object of the matrix sentence, which controls the embedded PRO and is M-bound by le in LF. In Spanish direct object complement sentences do not necessarily extrapose. 19. The case in which the matrix clause would contain a [-person] clitic (and no other clitics) never occurs in the relevant Verb Raising constructions. 20. There seem to be Spanish speakers who allow Climbing in infinitival clauses (cf. (i) and (ii)). This constitutes a problem for the present analysis, in which Climbing is triggered by the feature [+Tense]. (i) (ii)

Juana va a poder dártelo Juana is going to be able to give it to you Juana va a podértelo dar

32

Bok-Bennema

REFERENCES Chomsky, Ν. (1973) "Conditions on Transformations," in Anderson, S. and P. Kiparsky, eds., A Festschrift for Morris Halle, Holt, Rinehart and Winston, New York. Chomsky, N. (1979) "The Pisa Lectures," unpublished typescript. Chomsky, N. (1980) "The Representation of Form and Function" to appear in The Linguistic Review, 1.1. Chomsky, N. and H. Lasnik (1977) "Filters and Control," Linguistic Inquiry 8, 425-504. Contreras, H. (1979) "Clause Reduction, the Saturation Constraint, and Clitic Promotion in Spanish," Linguistic Analysis 5, 161-182. Emonds, J. (1976) A Transformational Approach to English Syntax, Academic Press, New York. Groos, A. (1978) "Towards an Inflectional Theory of Clitics," ms., University of Amsterdam. May, R. (1977) "The Grammar of Quantification," Doctoral dissertation, MIT, Cambridge, Massachusetts. Pesetsky, D. (1979) "Complementizer-trace Phenomena and the Nominative Island Constraint, CUNYforum (proceedings of the NELS IX). Riemsdijk, H. van (1978) A Case Study in Syntactic Markedness, Foris Publications, Dordrecht. Rivas, A. (1977) "A Theory of Clitics," Doctoral dissertation, MIT, Cambridge, Massachusetts. Rizzi, L. (1977) "Violations of the wh-Island Constraint in Italian and the Subjacency Condition, ms., Pisa. Rizzi, L. (1978) "A Restructuring Rule in Italian Syntax," in Keyser, S.J., ed., Recent Transformational Studies in European Languages, Linguistic Inquiry Monograph Series 3, MIT Press, Cambridge, Massachusetts. Strozer, J. (1976) Clitics in Spanish, Ann Arbor, Michigan. Vat, J. (1980) "Zieh en Zichzelf," in Gerritse, M. and S. Daalder, eds., Linguistics in the Netherlands II, North Holland, Amsterdam.

Chapter 3 Daniel Couquaux Faculté des Sciences de Marseille-Luminy Groupe Intelligence Artificielle

French predication and linguistic theory* 0. In a more extended work (Couquaux, 1981), I have argued that a basic verb-subject (VS) order is more frequent, in French and English, than is commonly assumed. There, I consider several syntactic and semantic phenomena of English and French grammar; I would like, however, to focus here on the so-called French predicative sentences, of which (1) is an example. (1)

Pierre est gentil "Pierre is nice"

I am going to claim that this type of sentence derives from a deep structure roughly of the form (2), and not (3), contrary to current assumptions. (2)

[foipe] [yp être NP AP] ]

(3)

[NP [yp être AP] ]

I shall call this hypothesis the ETRE-EN-TETE ("to be in front") theory (henceforth EET). In Couquaux (1981), I argue that EET is supported by the semantic properties of adjectival predication, but I shall focus in this paper on some syntactic properties of predicative sentences. I shall try to show that EET accounts for the rather strange, and still unexplained, distribution of the clitic pronouns en and se in this type of structure. This study is relevant for linguistic theory in four respects. First, it shows that French has an NP raising transformation. Consequently, we must definitely reject, as far as French is concerned, the lexicalist (Brame, 1976, Bresnan, 1978) or interpretive (Köster, 1978) accounts of raising phenomena. Secondly, it confirms that movement transformation leave "traces", as advocated by EST for many years now and, thirdly, that traces cannot be properly bound by their "antecedent" unless the antecedent c-commands the trace. Thus, our data support clearly the theory of grammar outlined by Chomsky (1976, 1977, 1980). However, we shall see that NP movement and WH movement cannot be coalesced into a single "Move a " transformation, as proposed by Chomsky (1980).

34

Daniel Couquaux

1. THE EN-AVANT TRANSFORMATION French object clitics (i.e. non subject clitics) are generally associated with a post-verbal position. In the following (a) examples, the underlined complements illustrate the position to which the clitics of the (b) examples are clearly connected: (4) a. b.

(5) a. b.

(6) a. b.

(7) a.

b.

Odile connaît tes amis "Odile knows your friends" Odile les connaît Odile them knows "Odile knows them" Pierre a écrit à sa mère "Pierre wrote to his (or her) mother" Pierre lui a écrit Pierre to him (or her) wrote "Pierre wrote to him (or her)" Luc pense à son travail "Luc thinks about his work" Lucj> pense Luc about it thinks "Luc thinks about it" Ils ont parlé de ton livre They spoke of your book "They spoke about your book" Ils en ont parlé They of it spoke "They spoke about it"

One can imagine two ways to express the obvious association of an object clitic with a post-verbal position. The first is having a clitic placement transformation, as in Kayne (1975a) 1 . (8)

CLITIC-PLACEMENT (CL-PL) X - NP - V - Y - Pro - Ζ 1 2 3 4 5 6 1- 2 - 5 + 3-6 (Pro denotes any strong pronoun)

The second would consist in inserting the clitics directly into their surface position, and having a clitic interpretation rule, more or less as in Rivas (1977). The choice between these two accounts is irrelevant for the present study. I shall assume the CL-PL analysis because of its greater familiarity.

French predication and linguistic theory

35

In any case, it is clear that a particular property distinguishes en from the other object clitics. Thus, it seems that en can be extracted from a subject NP and, hence, be associated with a pre-verbal position. (9) a. b. a. b. a. b.

La préface de ce livre est trop flatteuse "The preface of this book is too flattering" La préface en est trop flatteuse "The preface of it is too flattering" La porte du garage était ouverte "The door of the garage was open" La porte en était ouverte "The door of it was open" La couverture de ce livre sera rouge "The cover of this book will be red" La couverture en sera rouge "The cover of it will be red"

It is apparent that CL-PL does not predict the possibility of having an object clitic associated with a pre-verbal position. It might then be supposed that en is not a clitic at all; that is, en might be a constituent of the subject NP in (9b). These examples would have a surface structure more or less as in (10). (10)

[ [ N P N P - e r t ] [yp être AP] ]

Ruwet (1972, p. 54) has shown that this position cannot be maintained. He remarks that en follows the negative particle ne in negative sentences, which is, of course, impossible for other NP complements such as deNP. (11 ) a. La porte du garage «'était pas ouverte b. *La porte ne du garage était ouverte "The door of the garage was not open" (12)a. *La porte en n'était pas ouverte b. La porte η 'en était pas ouverte The door not of it was open "The door of it was not open" Moreover, even when extracted from the subject, en must always occur to the right of other object clitics, which is, of course, impossible for de MP abdominal complements: (13) a. La porte du garage lui a été ouverte b. *La porte lui du garage a été ouverte

36

Daniel Couquaux

"The door of the garage has been opened to him (or her)" (14) a. ""La porte en lui a été ouverte b. La porte lui en a été ouverte The door to him (or her) of it has been opened "The door of it has been opened to him (or her)" Thus, even when extracted from the subject NP, en belongs to the VP, like the other object clitics. However, it cannot be cliticized by CL-PL, which moves pronouns from right to left over V (backward cliticization). Consequently, Ruwet (1972, ch. 2) proposed a special transformation, which he called EN-AVANT ("en forward"), in order to cliticize en from left to right: (15)

EN-AVANT W-tNpX-en] 2 3 1

-Y-V-Z 4 5 6 -* 1 - 2 - 4 - 3 + 5 - 6

2. THE SINGULARITY OF EN-AVANT I have just suggested that forward cliticization singles out en. It could be objected, however, that subject clitics undergo the same type of movement rule. Such is the opinion of Kayne (1972), who argues that French has a SUBJECT-CLITIC-ADJUNCTION (SUB-CL-ADJ) rule: (16)

SUB-CL-ADJ - [NP y - C L 1 - V - Z 2 3 4 5 1

χ

-* 1 - 2 - 3 + 4 - 5

According to Kayne's hypothesis, the surface structure of (17a), for instance, would be (17b) (irrelevant details omitted). (17)a. b-

Elle aime les épinards [[nP¡ e ] [vp [yellei aime] les épinards] ] "She likes spinach"

If SUB-CL-ADJ existed, EN-AVANT would no longer appear exceptional, since French would have a class of forward cliticization rules. However, this rule becomes unnecessary as soon as we do not adhere to Kayne's structural definition of clitics. As a matter of fact, this definition does not seem to hold: there is much evidence that subject clitics do not belong to the VP in surface structure. Note, first, that they precede the negative particle:

French predication and linguistic theory

37

(18) a. Elle n'aime pas les épinards b. *N'elle aime pas les épinards "She does not like spinach" Secondly, all the syntactic rules of French I am aware of treat subject clitics exactly as full NP subjects. For example, the direct interrogative particle que, whatever it is, must be adjacent to the verb. If it cannot be, we must use the periphrase qu'est ce que instead: (19)a.

Piene fait quoi? "Pierre is doing what?" b. Que fait Pierre? c. *Que Pierre fait? d. Qu'est-ce que Piene fait? "What is Pierre doing?"

In (19a), the interrogative pronoun quoi "what" has not been fronted. In (19b) the subject Pierre has been moved to the right, and the sentence is grammatical with que. In (19c), the subject has not been postposed, and the sentence is ungrammatical (on subject inversion, see Kayne, 1972, Kayne and Pollock, 1978). Exactly the same paradigm is observed if we replace the full NP Pierre by a subject clitic, tu "you" (sing.), for instance. (20) a. b. c. d.

Tu fais quoi? Que fais-tu? *Que tu fais? Qu'est-ce que tu fais? "What are you doing?"

Thus, subject clitics pattern like full NPs in their interaction with the syntax of que. Crucially, they also pattern differently from the object clitics, since que may be separated from the verb by one or more object clitic(s): (21 ) a. b. c. d.

Pierre lui a dit quoi? Que lui a dit Pierre? *Que Pierre lui a dit? Qu'est-ce que Pierre lui a dit? "What did Pierre say to him (or her)?" (22) a. Pierre lui en a dit quoi? b. Que lui en a dit Pierre? c. *Que Pierre lui en a dit?

38

Daniel Couquaux

d. Qu'est-ce que Pierre lue en a dit? "What did Pierre say to him (or her) about it?" (23) a. Il vous y dit quoi? b. Que vous y a-t-il dit? c. * Qu'il vous y a dit? d. Qu'est-ce qu'il vous y a dit? "What did he say to you there?" In (21b), que is separated from the verb by the dative clitic lui, in (22b), by the cluster of clitics lui en, and in (23b) by the cluster vous y. Therefore, these sentences, being grammatical, contrast with (20c). Now, the morphological status of interrogative que is not quite clear. Obenauer (1976) claims that it is nothing but complementizer que (the same claim was made, for relative que, by Kayne (1975b)). Goldsmith (1978) and Hirschbühler (1978) argue that it is the reduced form of quoi. In any case, both accounts lead to the conclusion that subject clitics are not syntactic clitics. Let us suppose that Obenauer is right. Then, we must resort to the following filter: (24)

*que, unless que is in the context - V

Now, this filter would have to follow SUB-CL-ADJ, for general reasons (Chomsky and Lasnik, 1977). Therefore, if this transformation existed, (20c) would be grammatical. But it is not. Alternatively, if que is the clitic form of quoi, as argued by Goldsmith (1978), the quoi -*• que rule would have to apply in the phonological component, i.e. after SUB-CLADJ. Again, if this transformation existed, (20c) would be grammatical. We conclude from the above discussion that the clitic properties of subject pronouns cannot be explained by a forward cliticization rule. Therefore, it seems that EN-AVANT is left as the only left-to-right cliticization transformation. 3. THE DISTRIBUTION OF THE TWO EN'S As we saw above, cf. (7b), en may undergo the "normal" backward cliticization. Consequently, it is necessary to choose between the two following hypotheses: (25)

(26)

French has two en clitics: backward en, which undergoes CL-PL like the other object clitics, and forward en, which undergoes EN-AVANT French has only one en. This clitic, however, has one additional

French predication and linguistic theory

39

property, in comparison to the other object clitics - it accepts some sort of forward cliticization To demonstrate that hypothesis (25) is true, it must be shown that forward (or progressive) en has (at least) one property that backward (or regressive) en does not have. If this cannot be shown, then it must be concluded that progressive en is just a subcase of regressive en ; that is, that (26) is true. It is, then, interesting to observe that backward en shows all the properties of forward en. (i) Progressive en is always extracted from an NP: but regressive en can be also: (27) (28)

La préface en est trop flatteuse "The preface of it is too flattering" Max en a lu la préface "Max read the preface of it"

(ii) Both progressive and regressive en may correspond to the last de NP phrase of an unlimited series (Gross, 1968): (29)a. b. (30)a. b.

L'extrémité du pied de cette table est usée L'extrémité du pied en est usée "The extreme end of the foot of this table / of it is worn" Luc a cassé l'extrémité du pied de cette table Luc en a cassé l'extrémité du pied "Luc broke the extreme end of the foot of this table / of it"

(iii) Both en's are reported to refer more easily to objects than to humans: (3 l)a. Le père de ces enfants est mort la semaine dernière b. ?Le père en est mort la semaine dernière "The father of these children / of them died last week" (32)a. J'ai rencontré le père de ces enfants la semaine dernière b. ?J'en ai rencontré le père la semaine dernière "I met the father of these children / of them last week" (iv) Both en's belong to the literary style (Kayne, 1975a) (v) Progressive and regressive en are ordered in the same place in a string of object clitics, immediately before the verb (cf. (14b)). It appears, therefore, that the properties of forward en are just a (proper) subset of the properties of regressive en. We conclude from this

40

Daniel Couquaux

that French has only one en, with two subcases, forward en and backward en. However, the distribution just pointed out raises a problem for ENAVANT: we see that every output of EN-AVANT could just as well be generated by CL-PL. In other terms, EN-AVANT is redundant with respect to CL-PL.

4. TWO GAPS IN THE DISTRIBUTION OF FORWARD EN Apparently, Kayne (1975a, §2.19) has perceived the reduncancy of EN-AVANT. He proposes a rule, EN-EXTRAPOSITION (EN-EXT), such that en would be extracted from the subject and inserted at the end of the VP or, perhaps, the sentence. CL-PL would then apply to the extraposed en, in the usual way. This suggestion has the merit of explaining why forward en does not have any property of its own. The proposal has, however, two kinds of defects. On the one hand, there is no direct argument in favor of EN-EXT. In particular, the same kind of rule does not apply to de NP phrases in the contexts allowing progressive en (remember that, in the general case, en is equivalent to de NP). Thus, the examples in (33) are ungrammatical, contrary to (9b). (33)

*La préface est trop flatteuse de ce livre "The preface is too flattering of this book" *La porte était ouverte du garage "The door was open of the garage" *La couverture sera rouge de ce livre "The cover will be red of this book"

On the other hand, as is well known, forward en is more restricted than backward en. EN-EXT does not seem to explain these extra restrictions any better than EN-AVANT. Actually, it seems to me that neither rule explains them at all. The restrictions I have in mind concern contexts in which one would expect forward en to occur, but where it does not. These contexts define what I shall call the "negative properties" of forward en. Once again, there is an inclusion relation between the properties of the two en's, but this time the relation is inverted: the negative properties of regressive en form a proper subset of the negative properties of progressive en. This configuration of the data raises a serious problem for EN-EXT as well as EN-AVANT. As we just saw, there is a unique en in French: why, then, is one instance of en more constrained than the other? It is clear that forward en has all the negative properties of backward en. I give here three sets of data showing this.

French predication and linguistic theory

41

(i) Certain de NP or de Ν phrases have no corresponding backward en (Gross, 1968, pp. 28, 32): (34) a.

Luc a étudié les molécules de ce genre, de ce type, de cette sorte, de cette espèce, etc. b. *Luc en a étudié les molécules "Luc studied the molecules of this kind, of this type, of this sort, etc. I of it" (3 5) a. Odile portait un manteau de fourrure b. "Odile en portait un manteau Odile was wearing a coat of fur / of it "Odile was wearing a fur coat"

As expected, on the presumption that there is only one en, no forward en can correspond to these kinds of phrases: (36) a. Les molécules de ce type, etc., sont difficiles à étudier b. *Les molécules en sont difficiles à étudier "The molecules of this type, etc., are difficult to study" (37)a. Un manteau de fourrure serait le bienvenu par ce froid b. ""Un manteau en serait le bienvenu par ce froid "A fur coat would be welcome with this cold" (ii) Colette Colmerauer (personal communication) has noticed that regressive en cannot correspond to the de NP complément of lequel "which": (38)a. Luc préfère lequel de tes amis'! b. *Luc en préfère lequel?" "Luc prefers which of your friends?" In the same fashion, progressive en cannot correspond to the de NP complement of lequel in subject position: (39)a. Lequel de tes amis est le plus intelligent? b. ""Lequel en est le plus intelligent? "Which of your friends of them is the most clever?" (iii) In contemporary French, backward en is generally incompatible with the clitic "there". As predicted, the same is true of forward en : (40)a. On y a lu le premier chapitre de ton livre b. "On y en a lu le premier chapitre "Someone read the first chapter of your book / of it there"

42

Daniel Couquaux

(41) a. La sortie de ton livre y a été annoncée b. *La sortie y en & été annoncée "The appearance of your book / of it was announced there" The fact that all the restrictions on the occurrence of regressive en also constrain progressive en confirms the uniqueness of en. However, as I have said, and as has been noted by several people (Gross, 1968, Ruwet, 1972, Kayne, 1975a, Milner, 1978), EN-AVANT (or EN-EXT) would have to be more severely constrained than CL-PL. In the first place, it seems that a great number of verbs do not usually accept forward en : (42) a. La couverture de ce livre jaunit au soleil b. ?La couverture en jaunit au soleil "The cover of this book / of it turns yellow in the sun" a. L'éclatement de ce pneu provoquerait un grave accident b. ""L'éclatement en provoquerait un grave accident "The bursting of this tyre / of it would provoke a severe accident" Backward en is entirely natural with the same verbs: (43) a. Le soleil jaunit la couverture de ce livre b. Le soleil en jaunit la couverture "The sun yellows the cover of this book / of it" a. Un choc provoquerait l'éclatement de ce pneu b. Un choc en provoquerait l'éclatement "A shock would provoke the bursting of this tyre / of it" Still more strikingly, forward en cannot co-occur with a complement clause, tensed or infinitival: (44) a. L'auteur de ce livre espère devenir célèbre b. * L'auteur en espère devenir célèbre "The author of this book / of it hopes to become famous" a. Le résultat de ces expériences a démontré que l'hypothèse de Luc était fausse b. *Le résultat en a démontré que l'hypothèse de Luc était fausse "The result of these experiments / of them demonstrated that Luc's hypothesis was false" Backward en is, of course, compatible with the verbs of (44), as soon as they accept a direct NP complement:

French predication and linguistic theory (45) a. b.

43

Max a démontré l'inutilité de ces expériences Max en a démontré l'inutilité "Max demonstrated the uselessness of these experiments / of them"

These observations lead to the conclusion that the occurrence of forward en does not depend on particular verbs, but on certain structure types. This conclusion is verified upon examination of a broader class of data. Ruwet (1972) and Milner (1978) note that most forward en's are found in NP — être — AP sentences. However, the correct generalization seems to be that forward en can occur in all predicative sentences. Predicative sentences are those of the form NP - V - AP, with V = être "to be", sembler "to seem", paraître "to look (AP)", demeurer, rester "to remain", se montrer, se révéler, se trouver, s'avérer "to show up as (AP)", and the idiomatic phrase avoir l'air "to look (AP)" 2 . (46) a. b.

a. b. a. b.

La conclusion de ses travaux s'est révélée fausse" La conclusion s'en est révélée fausse "The conclusion of his (or her) works / of them turned out to be wrong" La démonstration de ce théorème semble impossible La démonstration en semble impossible "The demonstration of this theorem / of it seems impossible" L'auteur de ce livre restera inconnu L'auteur en restera inconnu "The author of this book / of it will remain unknown"

The second well-known restriction (Kayne, 1975a, Milner, 1978) on progressive en is that it cannot co-occur with subject quantifiers (or quantified NPs), or with subject adjectives and numerals, even in predicative sentences. (47) a. Beaucoup de filles sont laides b. * Beaucoup en sont laides Many of girls / of them are ugly "Many girls are ugly" a. _ Deux de ces caisses sont vides b. *Deux en sont vides "Two of these boxes / of them are empty" a. Des gros poissons ont été péchés à cet endroit b. *Des gros en ont été péchés à cet endroit "Big fish were caught in this place"

44

Daniel Couquaux

It cannot be claimed that there is no source for en in this kind of examples, given the uniqueness of en, since backward en can be construed as a complement of object quantifiers or quantified NPs, adjectives and numerals: (48)a. b.

a. b. a. b.

Luc connaît beaucoup de filles Luc en connaît beaucoup Luc knows many of girls / of them "Luc knows many girls" Odile a ouvert deux de ces bottes Odile en a ouvert deux "Odile opened two of these boxes" Max a péché des gros poissons à cet endroit Max en a péché des gros à cet endroit "Max caught big fish in this place"

In what follows, I will not endeavour to explain these constraints in themselves, as this approach implies that they are, in some sense, irregular, and hence that forward en is a regular phenomenon, not in need of special explanation. Rather, I shall adopt the opposite point of departure, in asking the following question: why is forward en sometimes possible? What I will show is that the answer to this question leads to an account of the restrictions on progressive en which is simple, coherent, and consistent with current thinking in linguistic theory.

5. ETRE-EN-TETE I have attempted to justify above the following propositions: (i) (ii)

en is the only pronoun allowing forward cliticization; forward en is the same clitic as backward en.

However, we do not see why en would differ from the other French pronouns in cliticizing to the right. Then, the existence of forward en must be considered problematic in itself. In other analyses, this problem was skirted by postulating a special rule, such as EN-AVANT or ENEXT (the very name of these rules indicates that they apply idiosyncratically to en). What I shall propose here is that these rules become unnecessary once it is seen that there is no reason to assume that en moves progressively. It is not difficult to achieve this latter result. To this point, it has been presupposed that the deep structure of predicative sentences is identical

French predication and linguistic theory

45

to their surface structure in all relevant respects, that is, more or less as in (49). (49)

[NP [ v p NP AP] ]

Suppose alternatively that, contrary to usual assumptions, predicative sentences are derived from deep structures of the form (50): [[Npe] [yp V NP AP] ] 3

(50)

This hypothesis, which I have called ETRE-EN-TETE ("to be in front"), accounts straightforwardly for the possibility of forward en in predicative sentences. To see this, consider (51) as an example. In order to get the so-called forward en we need only apply CL-PL to en in (52a) before applying NP movement to the NP la porte (the question of the trace of en will be considered in the next section). (51) (5 2) a.

La porte en est ouverte [s [NP e] [VP [V est] [NP la porte - en] [AP ouverte] ] ] CL-PL ->

b

·

Is [NP e ] [VP t v

·

la

en

-

est

l [NP l a Ρ 0 « 6 ) [AP ouverte] ] ]

"Move NP" c

[s [NPJ porte] [VP tv

en

~ est] [NP¡ e] [AP ouverte] ] ]

Naturally, we explain, a contrario, the non-occurrence of forward en in (42b) and (44b) by postulating a basic subject-verb (SV) order in these structures. Accordingly, CL-PL cannot apply to en in these cases.

6. NP MOVEMENT, CL-PL AND LOGICAL FORM Reconsider now the second gap in the distribution of progressive en (as compared to regressive en). The relevant examples are repeated here for convenience: (53)

*Beaucoup en sont laides *Deux en étaient vides *Des gros en ont été péchés

I shall argue that the ungrammaticality of (53) follows from the c-command requirement, plus two independently justified hypotheses. The c-command constraint states that:

46 (54)

Daniel Couquaux In LF, a trace must be c-commanded by its antecedent

C-command will be defined as in Reinhart (1976). (55)

A node X c-commands a node Y if X does not contain Y, and the first branching node dominating X dominates Y too, or if the first branching node αϊ dominating X is itself immediately dominated by a branching node a 2 which dominates Y and is of the same category as αϊ

Our first additional hypothesis is that French has a restructuring rule, which we shall call SCISSION, that detaches en from an NP: (56)

[np NP-[pp en] ]

[np NP] [ P P en]

SCISSION converts the subtree (57a), for example, into the subtree (57b). Note that we diverge from Chomsky (1977, p. 114) in assuming that such restructuring rules do not create a trace. (57)

VP I NP

a. NP

Λ

b.

VP NP

Τ en

PP

nÎ

Λ

en

The second hypothesis is that SCISSION does not apply when the head of the maximal NP in (56) is a quantifier, numeral or adjective. As a matter of fact, this constraint on (56) need not be stated at all if quantifiers, numerals and adjectives are not NPs. Given this, consider (58), which is, according to EET, the surface structure of the first example (53) (irrelevant details omitted):

[np· e] is the trace of the surface subject, [pp. e] is the trace of en. SCISSION being inapplicable in the presence of Q, the trace of en has been raised to subject position. Therefore, en does not c-command its trace in

French predication and linguistic theory

47

LF, and (58), i.e. (53), is ruled out by the c-command constraint. Now, how are well-formed sentences, such as (9b), repeated here, derived? (59)

La préface en est trop flatteuse La porte en était ouverte La couverture en sera rouge

According to EET, this type of sentence has a deep structure of the form (60a), which is converted into (60b) by SCISSION: (60)a.

[ s [ N P e] [yp [y est] [np [np la préface] - [ P P en] ] [yy> flatteuse] ] ]

b

·

ts [NP e] [VP t v [^p flatteuse] ] ]

est

l [NP [NP l a préface] ] [ P P en]

Successively applied on (60b), CL-PL and NP Movement generate (61), which is not excluded by (54), since here en c-commands its trace (according to the or... clause of the definition (55) of c-command). (61)

la préface

enj

est

e

e

flatteuse

So far, the distribution of forward en supports the condition (54), which is a well-known theorem of EST. However, this theoretical conclusion depends on the truth of our two additional hypotheses. Therefore, it would be good to show that these hypotheses are independently confirmed. Such facts exist in French; moreover, they confirm our explanation of the contrast between (53) and (59). Milner (1978, p. 76) notes the ungrammaticality of (62), to be compared with (63). (62)

*Ces usines, dont tu vois deux, produisent des saucisses "These factories, of which you see two, produce sausage" *Ces thèses, dont Max a lu beaucoup, traitent de thermodynamique "These theses, of which Max has read many, deal with thermodynamics"

48 (63)

Daniel Couquaux Ces usines, dont tu vois les cheminées, produisent des saucisses "These factories, of which you see the chimneys, produce sausage" Ces thèses, dont Max connaît bien les auteurs, traitent de thermodynamique "These theses, of which Max knows the authors well, deal with thermodynamics"

What is striking with these data is that they are perfectly parallel to the facts with forward en. We observe that dont, just like forward en, may be extracted from an NP with an NP head, but not from an NP the head of which is a quantifier or a numeral (adjectival heads are excluded redundantly, by an indépendant reason). But, as suggested by Chomsky (1977) for a similar phenomenon in English, the contrast between (62) and (63) can be explained by the Subjacency Condition (SJC), if we extend the domain of SCISSION to dont. This extension looks very natural, since dont is equivalent to de NP. (64)

SJC No rule can move a constituent across more than one cyclic node boundary. (I take cyclic categories to be S and NP in French).

Now, consider (65), which is the relevant part of the surface structure of the first example in (62). (65)

[s [COMP [PP¡ ^ t ] I t s t u [VP v o i s [NP[Q d e u x l - [ppj el ] • ·

According to our second hypothesis, SCISSION cannot split the NP [deux - dont] into [NP[Q deux] ] [pp dont]. Consequently, when moved into COMP, dont has to cross the boundaries of two cyclic categories, viz. NP and S. Hence the ungrammaticality of (65), and (62). Now, according to our hypotheses, the relevant part of the surface structure of the first example in (63) would be as in (66). (66)

[S [COMP [PP¡ d o n t ] I [ S t u [VP v o i s InpInP l e s cheminées] ] [PP, e]

Here, by hypothesis, SCISSION can detach dont from the NP head les chiminées. Therefore, the WH-movement of dont does not violate the SJC, since it crosses only the boundary of the cyclic category S. Thus, the dont facts confirm the hypotheses concerning SCISSION and its functioning. Additionally, they also support the ETRE-EN-TETE hypothesis in a rather striking way.

French predication and linguistic theory

49

In (62) and (63), dont was extracted from an object NP. This is an important point, given that dont may be extracted from a subject NP, even if the head of this NP is a numeral or a quantifier. Thus, the sentences (67) are grammatical, in contrast to (62). (67)

La direction va fermer ces usines, dont deux produisent des saucisses "The management will close these factories, of which two produce sausage" Max a examiné toutes ces thèses, dont beaucoup ont coûté de la sueur et des larmes à leurs auteurs "Max has examined all these theses, of which many cost their authors sweat and tears"

The grammatically of (67) leads to the conclusion either that SCISSION can apply to dont in subject position, even in the presence of a numeral or a quantifier, or that the SJC does not operate in configurations of the form examplified by (68): (68)

[COMP [pPj

1 Is [NP d e u x / beaucoup [pp. dont] ] . . .

The first suggestion amounts to saying that the constraint proscribing SCISSION from NPs with quantifier/numeral heads is restricted to VP. The second might mean that the SJC "does not count" one of two boundaries of cyclic categories when they are contiguous, as in (68). The first of these possibilities appears to be true, though I do not understand the phenomenon (it might be suggested that [j^p quantifier/numeral] [pp dont] is base generated in subject position, but why?). The reason I think it is true is that there are numerous counter-examples to the second proposal. Thus, we wish to explain the ungrammatically of the (b) examples below by SJC (cf. Chomsky, 1977, p. 114): (69) a. b

Ta passion pour Monique scandalise mon beau-frère "Your passion for Monique scandalizes my brother-in-law

ta · * f s [COMP [pp¡1 P'otu r qui] ] [s [NP [NP passion] [ P PLJe] ] 1_ [yp scandalise-t-elle mon beau-frère]]]?

a.

"For who does your passion, etc.?" Notre collaboration avec les Russes a échoué piteusement "Our collaboration with the Russians failed piteously"

50

Daniel Couquawc b. * [g" [cOMP [pPj a v e c qui] ] [s [nP ÍNP n o t r e collaboration]

a. b

[yp a-t-elle échoué piteusement]]]? "With who did our collaboration, etc.?" Ses enquêtes sur le Bourouchaski ont bouleversé la théorie phonologique

· * [S [COMP [PPj s u r q u e l l e langue] ] [ s [ N P [Npses enquêtes] [yp ont-elles bouleversé la théorie phonologique] ] ]? "His investigations on Buruchaski upset phonological theory' "On what language did his investigations, etc.?"

It seems, then, that the application of SCISSION to dont is blocked by a numeral/quantifier head only in the VP. Presumably, the same is true of en, since it behaves exactly like dont in all other pertinent respects. If this generalisation holds, the ungrammaticality of (53) can be attributed to the non application of SCISSION to en under one condition only: en has to be a constituent of the VP at the point of application of SCISSION. If the surface subject of the sentences (53) and (59) originated in subject position, there would be simply no reason for the former to be worse than the latter. Finally, the fact that the c-command requirement is relevant for en provides us with a straightforward argument in favor of the EET theory. Because of this requirement, EN-AVANT and EN-EXT simply cannot exist. EN-AV ANT would lower en into the VP, so that it would not c-command its trace in LF. As for EN-EXT, if it extraposed en to the end of the VP, it would violate (54) exactly as EN-AV ANT would do. Now, if it extraposed en to the end of the sentence, the next application of CL-PL would yield an illicit configuration. Clearly, only EET can account for the grammatical occurrences of the so-called forward en.

7. NP STRUCTURE Chomsky (1980, p. 3) assumes that "transformational rules are restricted to the single rule: Move a, where α is a category". I am going to argue now that "Move NP" and "Move WH" are not just subcases of a single transformation, and that they must be kept distinct. Given the c-command requirement and the limitation of SCISSION to object NPs not containing a quantifier or a numeral, the (70b) sentences should be ungrammatical; but they are well-formed.

French predication and linguistic theory (70) a. b. a. b.

51

Combien de filles connais-tu? Combien en connais-tu? "How many " o f ' girls / "of them" do you know?" Combien de gâteaux Pierre a-t-il mangés Combien Pierre en a-t-il mangés "How many " o f " cakes / "of them" did Pierre eat?"

Combien "how many, how much" is obviously a quantifier. Accordingly, SCISSION cannot split the underlying object phrase [^p [q combien] [pp en] ] into [j>jp [q combien] ] [pp en]. Therefore, the surface structure of, for example, the first (70b) sentence must be (71) (overlooking irrelevant subject clitic "inversion"): (71)

It is clear that (71), that is, (70b), should be ill-formed, because en does not c-command its trace. Because of the grammaticality of (70b), one might hold that combien is not a quantifier after all; but this counter-proposal would clearly be false, because of the ungrammaticality of (72b): (72) a. Combien de filles sont-elles laides? b. *Combien en sont-elles laides? "How many " o f " girls / "of them" are ugly?" a. Combien de boites étaient-elles vides? b. *Combien en étaient-elles vides? "How many " o f " boxes / "of them" were empty?" It appears that forward en cannot be construed as the complement of a subject NP headed by combien. (72b) is parallel to (53) and (62); thus, combien patterns like beaucoup and deux, not like ordinary NPs such as la porte or les cheminées. In other words, combien is indeed a quantifier. Accordingly, the surface structure (71) appears to be correct. Given (54), then, the grammaticality of (70b) is surprising. To explain it, one might claim that, for some unknown reason, SCISSION can apply to [jsjp [q combien] - [pp en\], even in object

52

Daniel Couquaux

position. Thus, the surface structure of (the first example in) (70b) could be (73) rather than (71). (73)

S

This move is incompatible with the EET hypothesis, because of (72b).

Following EET, combien de filles, combien de boîtes and combien - en originate in object position. Therefore, if SCISSION could apply to combien — en in the VP, the sentences (72b) would have to be grammatical since they would be similar to (70b) in the relevant respects. Both kinds of sentences would have surface structures of the form (74). (74)

combien We have assumed, so far, the "Move a " framework. What appears is that it is incompatible with EET, because it does not allow expressing the obvious difference between (70b) and (72b). The difference is that the combien phrase of (70b) is moved directly into COMP, whereas it is moved first into subject position in the (72b) case. Clearly, the c-command constraint applies to the result of NP Movement, but not to the output of WH Movement. Van Riemsdijk (1980) has recently proposed an alternative to the "Move a " system, which he attributes to Williams and himself. He claims that the well-known differences between NP Movement and WH Movement are better explained if the two rules are kept distinct. He argues that NP Movement generates a systematic linguistic level, which he calls "NPstructure". Most importantly, the binding conditions would be checked at this level. As for WH Movement, it applies after NP-structure, and generates the systematic level Surface Structure.

French predication and linguistic theory (75)

53

The Τ model (Chomsky and Lasnik, 1977, Chomsky, 1980)

The linear model (Van Riemsdijk, 1980)

DEEP STRUCTURE

DEEP STRUCTURE

Φ "Move a "

Ψ 'Move NP"

\/ SURFACE STRUCTURE

\J/ NP STRUCTURE

PHONOLOGY Rules of construal, Quantifier Interpretation

\k LOGICAL FORM

Ψ Quantifier Interpretation \/ LOGICAL FORM \l/ "Move WH"

\/

SURFACE STRUCTURE The Van Riemsdijk-Williams model explains directly the contrast between (70b) and (72b). To see this, consider (72b) first. The deep structure of the first of these examples is (76a). It is converted by CL-PL and NP Movement into (76b), which is its NP structure. (76)a. b

·

[s [COMPe] [s[NPe] [vp[V*ont] [Np[Qcombien]-[ppe«] ] [AP laides]]]] tS [COMPe] [s [NPj [Q combien] - [PPj e] ] [yp [v e n ] - sont] [NPI E] [AP LAIDESL 111

As far as en is concerned, (76b) is identical to (58), and is ruled out for the same reason: en does c-command its trace. Combien-tizce is ultimately moved into COMP, but this is irrelevant to the en problem. Let us turn now to (70b). The deep structure of the first of these examples is (77a). While CL—PL can apply to this structure, NP Movement, of course, cannot. Therefore, (77b) is the NP structure of this sentence.

54

Daniel Couquaux

(77) a. b-

[S [COMP E] [s [NF tu] [VP [V connais] [NP [Q combien] [FPen]]]]] [S [COMPe] [S ÌNPtu] [VP [ve«j -connais] [Np[Qcombien] [PPj e] ] ] ] ]

Since en c-commands its trace at NP structure, (77b) satisfies the c-command requirement. Comò ¿en-trace is ultimately raised into COMP, but this does not render the sentence ungrammatical, because WH Movement applies later than NP structure. To sum up, the combien-en data seem to support some of the main features of the Van Riemsdijk—Williams model, viz. the separation of NP Movement from WH Movement, the postulation of a systematic NP structure level, and the claim that the binding conditions are checked at this level. 4 ' 5

8. A GAP IN THE DISTRIBUTION OF THE CLITIC SE It is well-known that the French clitic se must refer to the subject of its clause, as in (78). (78)a.

Jeanj sj'est acheté une moto "Jean bought a bike to himself' b· [NPj Pierre et Odile] Si'écrivent de longues lettres "Pierre and Odile write long lettere to each other"

However, Kayne (1975a) has noted that se cannot always be construed as an anaphor to the subject NP. Let us examine the circumstance in some detail. The following sentences are ungrammatical. (79) a. *Pierrej sj'est cher "Pierre is dear to himself' *Dans ces circonstances, Luc¡ s¡'est toujours égal "Under those circumstances, Luc is always "equal to himself', i.e. able to face the situation" b- * [NPi Pierre et Odile] se¡ sont semblables "Pierre and Odile are similar to each other" * [NPj Les angles A et B] se¡ sont superposables "The A and Β angles are superposable to each other" * [NPj Les droites A et Β] se, sont parallèles "The straight lines A and Β are parallel to each other" These examples are not improved if reflexive/reciprocal se is "doubled"

French predication and linguistic theory

55

by a pronominal phrase à lui-même, à elle-même, etc., or l'unie) à l'autre, les un(e)s aux autres. The examples in (80) are just as bad as those in (79). (80) a. *Pierre¡ sj'est cher à lui-mêmej *Lucj si'est toujours égal à lui-mêmej b. *[NP¡ Pierre et Odile] se¡ sont semblables l'un à l'autre} * [NPj Les droites A et B] se¡ sont parallèles l'une à l'autrei Compare with the examples in (81), which are grammatical, and synonymous to those in (78). (81 ) a. Jeanj sj'est acheté une moto à lui-méme¡ b · [NPj Pierre et Odile ] Si'écrivent de longues lettres l'un à l'autrej The ungrammatically of (79)-(80) does not result from any "semantic" incongruity; the sentences below are grammatical, with the meaning (79) and (80) would have if they were possible. (82)a.

Pierre est cher à lui-même "Pierre is dear to himself' Dans ces circonstances, Luc est toujours égal à lui-même "Under those circumstances, Luc is always "equal to himself" " b. Pierre et Odile sont semblables (l'un à l'autre) ."Pierre and Odile are similar (to each other)" Les angles A et B sont superposables (l'un à l'autre) "The angles A and B are superposable (to each other)" Les droites A et B sont parallèles (l'une à l'autre) "The straight lines A and B are parallel (to each other)"

Furthermore, it cannot be said that être is incompatible, for some yet undiscovered reason, with dative clitics which are underlying complements of adjectival predicates, since the sentences in (83) are perfectly regular. (83)

Odile lui est chère "Odile is dear to him (or her)" Cela leur est égal "This is equal (i.e. indifferent) to them" L'angle A lui est superposable (à l'angle Β) "The angle A is superposable to it (to the angle B)" La droite A leur est parallèle (aux droites B, C, D) "The straight line A is parallel to them (to the lines B, C, D)"

Etre is not even incompatible with the class of clitics which are ambiguously

56

Daniel Couquawc

accusative and dative, viz. me, te, nous, vous, in addition to se (nous and vous may also be nominative clitics). Thus, the sentences below are excellent. (84)

Odile m'est chère "Odile is dear to me" Cela nous est égal "This is equal (indifferent) to us"

The pertinent observation is that se cannot occur in NP—être—AP sentences, that is, in predicative sentences. That the gap in the distribution of se is not restricted to être sentences, but concerns predicative sentences in general, is demonstrated by the ungrammatically of (85). 6 Pierre; se¡ semble cher (à lui-même,) "Pierre seems dear to himself" *Malgré son âge, Luq sj'est resté égal (à lui-même,) "Despite his age, Luc has remained "equal to himself"" b· * [NPj Pierre et Odile] se¡ deviennent de plus en plus semblables (l'un à l'autre,) en vieillissant "Pierre and Odile become more and more similar to each other as they grow older" * [NPj Les angles A et Β] se¡ paraissent superposables (l'un à l'autrej), mais en réalité ils diffèrent de I o "The angles A and Β seem superposable to each other, but in reality they differ by one degree"

(85)a.

As predicted, the same sentences are well-formed without se: (86)a. b·

Pierre, semble cher à lui-même. Malgré son âge, Luq est resté égal à lui-méme¡ [NPj Pierre et Odile] deviennent de plus en plus semblables (l'un à l'autre,) en vieillissant [NP¡ Les angles A et B] paraissent superposables (l'un à l'autrej), mais en réalité ils diffèrent de 1°

Tp sum up, there is a gap in the distribution of se which, importantly, presents itself precisely in the structural configuration in which we have previously observed an excess in the distribution of en, viz. the predicative structure. Thus, in this environment, en and se exhibit inverse properties; se is impossible where it should occur, and en is possible where it should not occur. In an explanatory theory of French grammar, this configuration of data calls obviously for a unitary account of the two properties in question.

French predication and linguistic theory

57

We have seen that forward en can be eliminated by assuming the ETRE—EN—TETE theory. Presumably, this theory will also account for the missing se's. As is well-known, the antecedent of se must be the subject of its clause; contrary to himself, each other and l'un l'autre, se cannot have a direct object antecedent. Thus, (87a) can mean only "[NP¡ Pierre and Odile] introduced Max and Catherine to each other;", whereas (87b) can also have the reading "Pierre and Odile introduced [NP¡ Max and Catherine] to each otherj". (87) a. b.

Pierre et Odile se sont présenté Max et Catherine Pierre et Odile ont présenté Max et Catherine l'un à l'autre.

In order to explain the difference between se and himself, etc., we assume a formal system extrapolated from Chomsky (1980). We assume that every NP receives a referential index in the base. Referential indices are positive integers (ranging from 1 to n), and attributed "from top to bottom". Anaphors are not indexed in the base. They may receive an anaphoric index by rule (88): (88)

Co-index an anaphor to every c-commanding NP

An anaphoric index is, therefore, a set of positive integers, and is noted < . . . i, j, . . . >. An anaphor may also receive a counter-anaphoric index, which is a set of integers, noted > . . . k, 1 . . . > 2 < sont présenté [ΝΡ2 Max et Catherine] [NPi Pierre et Odile] ont présenté [NP 2 Max et Catherine] l'un à l'autre < ι , 2 >

In (90a), the identical anaphoric and counter-anaphoric indices 2 get erased, so that se may be construed only as an anphor to the subject NP, to which it is co-indexed. In (90b), l'un l'autre is not counter-indexed to the direct object, as it belongs to the PP l'un à l'autre and does not c-command the direct object. Consequently, l'un l'autre keeps its anaphoric

58

Daniel Couquaux

index 2 as well as 1, and may be construed as an anaphor to the direct object as well as to the subject NPs. Let us now see how EET can account for the ungrammatically of (79), (80) and (85). Consider, for instance, the first sentence in (79). According to EET, (91) is its NP structure ; co-indexing by (88) and (89) yields (92). (91) (92)

[NPJ Pierre] se est [NP 1 e] cher [NPI Pierre] se < ι > > ι < est [NPJ e] cher

Pierre has been indexed 1 (for instance) in the base. When it is moved to subject position, its referential index is copied on the trace (or, alternatively, both the antecedent and the trace receive the same new referential index). Se is then co-indexed with the subject (anaphoric indexing) and the trace (counter-anaphoric indexing). Both indices erase, since they are identical. Consequently, se has no anaphoric index at the end. Suppose that all anaphors have to be bound, that is, anaphorically linked (by an anaphoric index) to some NP. Then, the sentences (79), (80) and (85) will be ruled out because se is not bound.

9. TOWARD AN EXPLANATORY THEORY OF PASSIVE The distinction between the so-called adjectival and verbal passives is mainly based on interpretive facts. Thus, (93) conveys a permanent or stative reading, whereas (94) exhibits a punctual or active one. (93) (94)

Le gâteau est mangé "The cake is eaten" Le gâteau est mangé par les enfants "The cake is eaten by the children"

It has been claimed (e.g. Wasow, 1977, Bresnan, 1978) that this interpretive difference results from a categorial ambiguity of passive participles. According to Wasow and Bresnan, they are adjectives is the adjectival passive, but verbs in the verbal passive. In other words, Bresnan and Wasow agree, as does everybody, that the adjectival passive is just an ordinary predicative structure. However, this implies that this sort of "passive" derives via NP movement, contrary to what they both assume, if the EET hypothesis is true. Though they agree on the status of the adjectival passive, Bresnan and Wasow diverge on the so-called verbal passive. Wasow claims that it implies NP Movement, whereas Bresnan argues that the same kind of rule(s) that is necessary to interpret the adjectival passive may be used also to interpret

French predication and linguistic theory

59

the verbal one. One of the consequences of EET is that Bresnan's position cannot be correct; the verbal passive rule cannot be a lexical interpretive rule, since even the adjectival passive rule is not. Given EET, then, one may ask what is the difference between the two kinds of passives. We already know that it cannot be a difference between base derived versus transformationally derived structures, but it might still be the case that passive participles are adjectives in one case and verbs in the other. As surveying Wasow's arguments for distinguishing the two kinds of passives would lead us too far afield at this point, let us focus on what seems to be the heart of the matter. Since the so-called adjectival passive is an ordinary predicative structure, the verb être which occurs in it must be the predicative être (I shall refer to this être as the copula). What about the être of the verbal passive? If the participle is a verb, then être cannot be the copula, since presumably the copula does not co-occur with verbs. It could then be either auxiliary être, or a third kind of être, a special passive être. Bresnan seems to adopt the latter solution; she writes: "In a more detailed analysis, we would take be to be a verb subcategorized for passives, as get is" (Bresnan, 1978, p. 21). What I will try to show is that the être of the verbal passive is nothing but the predicative être. I shall mention briefly two classes of data which demonstrate that the e/re+participle string of the verbal passive behaves exactly like the éfre+adjective string of predicative sentences, including the être+participle string of the adjectival passive. A priori, it would not be a particularly disturbing assumption to claim that the être of the verbal passive is a kind of auxiliary, since être can be an auxiliary in French. Thus, there are sentences such as (95), where the participle is clearly a verbal form: (95)

Pierre est venu (parti, monté, sorti, descendu, etc.) "Pierre is (i.e. has) come, gone, gone upstairs, gone out, gone downstairs, etc."

The temporal interpretation of such sentences, however, contrasts with that of predicative sentences (including adjectival passive sentences). (96)

Pierre est gentil "Pierre is nice" La porte est fermée "The door is closed"

The predicative sentences in (96) are construed with present temporal reference, consistent with the morphological tense of the copula. In the

60

Daniel Couquaux

active sentences in (95), however, être is also morphologically present (est, but not était or fut), but the sentences are construed with past temporal reference. This contrast is corroborated by a test of compatibility with time adverbs: (97)

(98)

Pierre est gentil La porte est fermée "Pierre is nice "The door is closed Pierre est venu

j en ce momenti {*hier j J in this moment' [ yesterday" *en ce moment) hier j

It is emphasized by the fact that être, though morphologically present, cannot be put in the past tense by using auxiliary avoir, in active sentences. (99) (100)

Pierre a été gentil hier La porte a été fermée hier *Pierre a été venu hier

Now, the verbal passive exhibits the same temporal properties as predicative sentences: the tense of être determines the temporal reference of the whole sentence, and être can be put in the past by using auxiliary avoir: (101)

fen ce moment) {•hier J Le gâteau a été mangé par les enfants hier "The cake has been eaten by the children yesterday" Le gâteau est mangé par les enfants

The second set of data is just as simple. In predicative sentences, an adjective may be "replaced" by the so-called propredicate le. The same is true of participles in adjectival passives. (102)

Pierre est gentil, et Luc /'est aussi "Pierre is nice, and Luc is "it" too" Pierre est gentil, mais Luc ne /'est pas "Pierre is nice, but Luc is not " i t " " La porte est fermée, et la fenêtre /'est aussi "The door is closed, and the window is "it" too" La porte est fermée, mais /'était-elle au moment du crime? "The door is closed,but was it "it" whenthecrime wascommitted?"

We cannot, however, pronominalize the participle of an active sentence:

French predication and linguistic theory

61

(103) *Pierre est venu, et Luc /'est aussi "Pierre "is" come, and Luc "is it" too" *Pierre est venu hier, mais il ne /'est pas avant-hier "Pierre "is" come yesterday, but he "is" not "it" the day before yesterday" The so-called verbal passive behaves again like predicative sentences, not like active ones. Thus, the examples below are perfectly well-formed: (104)

Ce gâteau a été mangé par les enfants, mais celui-ci /'a été par le chat "This cake has been eaten by the children, but this one has been "it" by the cat" Ce gâteau sera mangé par les enfants, mais celui-ci ne le sera pas "This cake will be eaten by the children, but this one will not be "it" "

In view of such facts, it would probably be wrong to assume that the ê/re+participle phrase of passive is categorially distinct from the être+ adjective phrase of the predicative structure. In other words, it seems to me that passive is just a subcase of this structure. This conclusion has several interesting consequences. First of all, the arguments that have been given in the literature in favor of a transformational derivation of the verbal passive turn out to be indirect arguments in favor of ETRE—EN—TETE. Conversely, EET provides a strong argument for the NP Movement derivation of passive sentences: if they are a kind of predicative sentences, and if predicative sentences are transformationally derived, then passive sentences are transformationally derived. To conclude, I would like to claim that EET leads toward a deeper understanding of passive, in bringing together two main aspects of this construction: its being built with être, and its undergoing NP Movement. NOTES *I am indebted to several people for useful advice on the work reported here; particularly, for written comments, to Richard S. Kayne, Robert May, Jean-Claude Milner, Yves-Charles Morin, Nicolas Ruwet and Douglas C. Walker. I am thankful to Robert May and Doug Walker for trying to improve the style of this paper. 1. Our analysis of en is fully compatible with a more compact formulation of CL-PL, such as (i) (Rouveret and Vergnaud, 1980), or (ii) (Couquaux, 1980): (i) (ii)

Cliticize X onto V Move clitic

62

Daniel Couquaux

We mention here Kayne's more developed version of the rule for expository convenience. 2. Of course, I am not claiming that forward en is to be found in the predicative structure only, since this claim would clearly be false. Forward en occurs also with "existential" verbs (i.e. verbs predicating the existence of some thing or being) and with movement verbs, such as arriver "to arrive", venir "to come". For further details on the distribution of forward en, see Morin (to appear), Couquaux' (1981). What I am claiming is only that forward en is fully regular in the predicative structure. 3. As far as the syntax of forward en is concerned, the order être AP NP would do just as well. There is evidence, however, that theéíre NP AP order is to be preferred. First, the NP AP order is found in the verbless predicative phrases embedded under epistemic verbs such as savoir "to know" or croire "to believe": (i) (ii)

Je savais Jean malade ?Je savais malade Jean I knew Jean ill I knew that Jean was ill"

Secondly, the NP Predicate order might lead to an explanation of a puzzling contrast discovered by Ruwet (1975). He remarks that the possessive pronoun son may be construed as co-referential to Christine in (iii), but not in (iv): (iii) (iv)

Son argent est le meilleur atout de Christine "Her money is Christine's best luck" *Son meilleur atout est l'argent de Christine "Her best luck is Christine's money"

The base structures of (iii) and (iv) will be (v) and (vi), respectively, if we assume the NP Predicate order: (v) (vi)

[NP e l e s t [NP s o n argent] [jjp le meilleur atout de C.] [fsjp e] est [jsjp l'argent de C.] [Mp son meilleur atout]

We observe that NP Movement creates a cross-over configuration in (iv), but not in (iii). I shall not try to explain why such a configuration amounts to ungrammatically, in this case. 4. Consequently, we have to replace "Logical Form" by "NP structure" everywhere in section 6. 5. Robert May and Jan Köster (personal communication) object that the NP structure approach would not account for the contrast between (i) and (ii): (i) (ii)

Which woman that John loves did he buy a present for? *He bought a present for a woman that John loves.

They claim that "if the c-command on anaphora is stated at NP structure and hence before WH Movement, then we would expect binding possibilities to be the same in (i) and (ii)". They conclude from this that "the c-command requirement applies to the output of WH Movement". This objection could be skirted in several ways. First, it is not at all clear that the c-command requirement is relevant for pronouns (i.e. for non anaphors). It has been proposed recently by Guéron (1979) that pronouns are not subject to the c-command requirement, but rather to rule (iii):

French predication and linguistic theory (iii)

63

A pronoun cannot precede its antecedent in LF

Moreover, the fact that the binding conditions on pronominal reference have to follow WH Movement would not ruin ghe NP structure approach, if we admit that they do not belong to sentence grammar, but rather to the grammar of discourse. A discourse constraint on pronominal reference would have to follow the sentence grammar rule of WH Movement. See Guéron (1979) for a discourse account of pronominal reference. 6. Se cannot be anaphoric to the subject in passive sentences and seem sentences either: (i) (ii)

* ÍNP¡ Pierre et Odile] sj'ont été présentés par Jacques "Pierre and Odile have been introduced to each other by Jacques" *Jeanj se\ semble avoir raison "Jean seems to himself to be right"

Compare (ii) with (iii): (iii)

Jeanj /u/j semble avoir raison "Jean seems to him to be right"

The pertinent generalisation is that se is incompatible with NP Movement. This incompatibility follows from the system sketched below.

REFERENCES Brame, M. (1976) Conjectures and Refutations in Syntax and Semantics, ElsevierNorth-Holland, Amsterdam, New York. Bresnan, J. (1978) "A Realistic Transformational Grammar", in Halle, M., J. Bresnan and G. Miller (1978) Linguistic Theory and Psychological Reality, M.I.T. Press, Cambridge, Mass. Chomsky, N. (1976) "Conditions on Rules of Grammai", Linguistic Analysis, 4. Chomsky, N. (1977) "On WH Movement", in Culicover, P. et al. (1977). Chomsky, Ν. (1980) "On Binding", Linguistic Inquiry, 11,1. Chomsky, N. (to appear) Lectures on Government and Binding, Foris Publications, Dordrecht. Chomsky, N. and H. Lasnik (1977) "Filters and Control", Linguistic Inquiry, 8 , 3 . Couquaux, D. (1981) La transformation MONTEE en français, thèse de doctorat d'Etat, Université Paris VII. Culicover, P., T. Wasow, and A. Akmajian (1977) Formal Syntax, Academic Press, New York, London. Goldsmith, J. (1978) "Que, c'est quoi? Que, c'est QUOI", Montreal Working Papers in Linguistics, 11. Gross, M. (1968) Grammaire transformationnelle du français: syntaxe du verbe, Librairie Larousse, Paris. Guéron, J. (1979) "Relations de coréférence dans la phrase et dans le discours" in Ronat, M., Grammaire de phrase et grammaire de discours, Langue Française, 44. Hirschbühler, P. ( 1978) The Semantics and Syntax of WH Questions, Ph.D., U Mass.Amherst.

64

Daniel Couquaux

Kayne, R.S. (1972) "Subject Inversion in French Interrogatives" in Saciuk, B. and J. Casagrande, Generative Studies in Romance Languages, Newbury House, Rowley, Mass. Kayne, R.S. (1975a) French Syntax: The Transformational Cycle, M.I.T. Press, Cambridge, Mass. Kayne, R.S. (1975b) "French Relative QUE", Recherches Linguistiques, 3. Kayne, R.S. and J-Y. Pollock (1978) "Stylistic Inversion, Successive Cyclicity, and Move NP in French", Linguistic Inquiry, 9 , 4 . Köster, J. (1978) Locality Principles in Syntax, Foris Publications, Dordrecht. Milner, J-C. (1978) De la syntaxe à l'interprétation: quantités, insultes, exclamations, Editions du Deuil, Paris. Morin, Y-C. (to appear) "Some Myths about Pronominal Clitics in French", Linguistic Analysis. Obenauer, H-G. (1976) Etudes de syntaxe interrogative du français: QUOI, COMBIEN, et le complémenteur, Max Niemeyer Verlag, Tübingen. Reinhart, T. (1976) The Syntactic Domain of Anaphora, Ph.D., M.I.T. Van Riemsdijk, H. (1980) NP Structure, and a Linear Model of Core Grammar, lecture delivered at the GLOW meeting, Nijmegen. Rivas, A. (1977) A Theory of Clitics, Ph.D. thesis, M.I.T. Rouveret, A. and J-R. Vergnaud (1980) "Specifying Reference to the Subject: French Causatives and Conditions on Representations", Linguistic Inquiry, 11,1. Ruwet, N. (1972) "La syntaxe du pronom en et la transformation de "montée du sujet"", in Ruwet, N., Théorie syntaxique et syntaxe du français, Editions du Seuil, Paris. Ruwet, N. (1975) "Les phrases copulatives en français", Recherches Linguistiques, 3. Wasow, T. (1977) "Transformations and the Lexicon", in Culicover, P. et al. (1977).

Chapter 4 J. Guéron Université de Paris VIII

Logical Operators, Complete Constituents, and Extraction Transformations 1 Recent work in transformational grammar has revealed the importance of Logical Form (LF), that level of grammar intermediate between syntax and semantics.2 LF shares a property of syntax in being a locus for movement rules. It differs from syntax in other respects: (i) it feeds the semantic component directly; (ii) it contains interpretive rules which transform syntactic structures into logical structures. On this level, constituents containing a Wh-morpheme are interpreted as logical quantifiers and empty categories are interpreted as variables bound by these quantifiers under certain conditions. This property implies that constraints on interpretations which are both structure-dependent and mention variables, such as the Government-Binding principles of Chomsky (1981), operate on the output of LF. Another such constraint, the Complete Constituent Constraint, will be proposed in this study, (iii) Unlike syntax, LF does not feed the phonological component. This property will be used to explain why certain movement operations which are unacceptable in the syntax are acceptable in LF. In Section 1 of this paper, some movement and interpretive rules of LF will be described. In Section 2 the Complete Constituent Constraint will be proposed in conjunction with some elements for a theory of logical operators. Section 3 shows the importance of the Complete Constituent Constraint for a theory of reference. Finally, in Section 4, the role of the Complete Constituent Constraint with respect to extraction transformations will be examined.

1. RULES OF LOGICAL FORM If we suppose that the same movement rule which applies in the syntax, namely MOVE CATEGORY, applies as well in LF, then we can explain the synonymy of (1) and (2) below in terms of a single logical output structure (3).

66

J. Guéron

(1) (2)

Not a single term paper did the teacher insist that we write The teacher insisted that we write not a single term paper 3

(3)

[gn+l^iNOTX]

[sn...*1...]]

We may similarly explain the synonymy between (4) and (5) in terms of the LF output (6) (4) (5)

John is so tall that he can't fit into the car So tall is John that he can't fit into the car

(6)

[ S n + 1 [XiSOX] [ S » . . . x i . . . ] ]

It then becomes possible to posit a universal schema for quantification interpretation, (7) below, applicable to the output of LF. (7)

X* [+Q]

[8...χ*...]

or

b. [g .. .χ*... ]

X1 [+Q]

That is, any structure in which X both incorporates a logical operatior (represented in (7) by the feature [+Q] ) and governs an S is a quantificational structure. 4 Following discussion in Baker (1970) and Bresnan (1972), we assume that (7a) is the quantificational structure for languages which have COMP to the left of S and (7b) is the structure for languages which have COMP to the right of S. We assume, moreover, that it does not matter whether the syntactic node which dominates both S and the quantifier is S (as in the case of syntactic movement to COMP) or S i + 1 (in the case of logical movement) so long as the quantifier governs S. The interpretation of particular realizations of (7) will be a function of the interpretation of the schema itself plus that of the quantifier it contains (e.g. NOT vs. Wh) and the lexical support of the quantifier (a paper vs. tall). Positing a universal schema for quantification allows us to associate with (3) and (6) not only English structures, in which syntactic NEC and SO-movements seem to provide a syntactic "model" for movement in LF, but also structures in languages in which the parallel syntactic movements are impossible. We will thus associate the French Ss (8) and (9) with the same interpretive schema as their English counterparts (2) and (5), that is, with (3) and (6) respectively. (8) (9)

Il ne veut que tu épouses personne (cf. Kayne 1979) (He wishes that you marry no one) Jean est si grand qu'il ne peut pas entrer dans la voiture (Jean is so tall that he can't get into the car)

Logical Operators, Complete Constituents, Extraction Transformations 67 I will consider the fact that NEG and SO do not move in the syntax of French as something to be explained independently of the theory of quantification.5 (An explanation will be attempted below) (10)a. *Personne ne veut-il que tu épouses b. *Si grand Jean est-il qu'il ne peut pas entrer dans la voiture An argument in favor of positing logical movement parallel to syntactic movement in UG is that the same "bridge constraints" which block the first also block the second. Thus wide-scope SO movement is not possible from an embedded factive clause either in syntax or in LF. (11)a.

So weird did Mary claim Bill is that we won't invite him to dinner (cf. Liberman 1974) b. *So weird did Mary forget that Bill is that we won't invite him to dinner (12) a. Mary claims that Bill is so weird that we won't invite him to dinner b. *Mary forgot that Bill is so weird that we won't invite him to dinner

If we consider the constraint operating in (lib) to be a constraint on movements, then we must assume that the SO constituent moves in (12b). If we consider ( l i b ) to blocked by an interpretive constraint on outputs, then we must give (1 lb) and (12b) the same output. Either way we must assume a movement rule to have applied in (12b), or else espouse the highly improbable view that ( l i b ) and (12b) are unacceptable for two different reasons.6 The hypothesis that the rule MOVE CATEGORY functions in LF as well as in syntax allows us to posit (7) as the universal structure for quantification configurations directly feeding the rules of semantic interpretation to explain the synonymy of syntactically dissimilar structures in terms of a single logical structure and to explain why the same constraints apply to syntactically dissimilar structures. We now turn to two major problems raised by this hypothesis, (i) If indeed MOVE CATEGORY applies indifferently in syntax and in LF, and if quantifier phrase movement applies in (13a) below to create the LF (13b), as proposed in Chomsky (1976) and May (1977), (13)a. b

John saw everyone everyonex [g John saw x]

then why is (14) any illegitimate syntactic structure?

68

J. Guéron

(14)

*Everyone John saw *Everyone did John see

(ii) It appears that quantifiers may combine in Logical Form to create complex quantifiers. Thus Higginbotham and May (1981) propose that the two WH-constituents of (15a) function, through a process of "absorption", as a single complex quantifier in the LF (15b). (15)a. b.

Who loves who? (WH, WH)X ,y [ s χ loves y]

Why, then, is there no syntactic structure (16)? (16)

[ s *2 Whoi [ s +i Whoj [ s

ei

loves ej] ] ]

If two quantifier phrases can merge into a single complex quantifier in LF, thereby satisfying the requirement that a quantifier govern its syntactic domain, then the grammar should associate (16) with the acceptable LF (15b). Inasmuch as these problems have given some linguists doubts as to the reality of LF, we should like to attempt to solve them. Our explanation will be crucially based on the fact that syntax feeds the phonology and LF does not. We propose that the illocutionary rules of the language, which determine whether an S is declarative, interrogative, etc., apply on the output of the phonological component. Following Safir (1980), I will suppose that the illocutionary rules essentially read what is in the COMP node, or possibly the string of material starting with COMP and going to the tense morpheme. A rough idea of the rules linking a syntactic output with an illocutionary type is given in (17) for the major illocutionary types of English.

(17) a.

b. c.

X [+WH] L-COMP

+tense [+tense] L-S

ϊφ ΓΝΡ Lc o m p LLS S COMP a [+tense] '-COMP

α +1 +tense

[+tense]

WH-INTERROGATIVE

DECLARATIVE

YES-NO INTERROGATIVE L

S

Logical Operators, Complete Constituents, Extraction Transformations

69

An S such as (18) will be associated with the interrogative type (17a) in the output of the phonology and with the quantificational structure (7) in Logical Form. (18)

Who did you see?

As shown in (19), adapted from Chomsky and Lasnik(1977),illocutionary rules and interpretive rules apply independently of each other. (19)

SYNTAX: - Phrase Structure Rules - "MOVE CATEGORY" C s-structures1) PHONOLOGY

(

phonetic \ structures,/ ILLOCUTIONARYl RULES

LF-1 : "MOVE CATEGORY" "COINDEX" IDENTIFICATION OF QUANTIFIERS AND VARIABLES OUTPUT CONSTRAINTS — Government-Binding - Complete Constituent Constraint LF-2: (SEMANTICS): "INTERPRET" Rules of Reference Rules of Quantification, etc.

\1/ DISCOURSE FILTERS AND PRAGMATIC FILTERS

SYSTEMS OF KNOWLEDGE AND BELIEF

complete semantic representations J

Illocutionary rules are language specific. In English, alongside (17a), which defines an illocutionary type of S with [+Wh] in COMP, there will be illocutionary assignments for Ss with the following morphemes in COMP: NEG, SO, ONLY. Suppose, however, the syntax generates (14') 7 O4')

[COMP [nP 1 Everyone] ] [ s John saw

e] ]

It happens that English has no rule which assigns an illocutionary type to Ss with every-X in COMP. Nor can such a S be considered a declarative since, after movement, it no longer satisfies the description in (17b). If every S must be associated with some kind of illocutionary type, then (14') will be ruled out in the illocutionary component as ill-formed. If, however, the quantifier phrase everyone does not move until LF,

70

J. Guéron

it will then be assigned a quantiflcational interpretation in the semantic component and be associated with a declarative illocutionary type in the output of the phonology, as desired. The constituents which can trigger an illocutionary rule, and therefore may appear in COMP, are language specific and arbitrary to some degree. Thus, although SO and NEG may not appear in COMP in French, peutêtre and aussi may. (20) a. b.

Peut-être est-il déjà parti. (Perhaps has he already left) Aussi vous ai-je écrit pour avoir des nouvelles plus récentes. (Therefore did I write to you to get more recent news)

The difference between obligatory and non-obligatory movement to COMP also depends on the contents of the illocutionary rule component. Whereas WH-constituents must move to COMP in English, they need not in French. (21)a. *You have how many books? b. Tu as combien de livres? This difference stems from the fact that there happens to be an illocutionary rule in French which associates a particular intonational contour with an interrogative S-type. 8 Thus the following Ss are interrogative if they have a rising intonational contour and declarative if they have a falling contour. (21 ) c. d.

Tu as vu Pierre. (You saw Pierre) Deux et deux font quatre. (Two and two are four)

Assuming that, for some reason, ^//-quantifiers must be associated with an interrogative-type S, the desired association can take place in English only if Mi-constituents move in the syntax, but will take place in French if either of two conditions apply: WH-movement takes place in the syntax or the S is associated with a rising contour in the phonological component. The obligatoriness of syntactic WH-movement in English is thus explained independently of the theory of quantification. Note that the association of an intonational contour with an illocutionary rule in French provides support for our hypothesis that illocutionary rules apply to the output of the phonological component. 1.2. Interpretive constraints. Since LF directly feeds the semantic component, it is plausible to situate all structure-dependent constraints on interpretation on the output of

Logical Operators, Complete Constituents, Extraction Transformations

71

LF-I. In the schema (19) it becomes obligatory to do so for any constraint which mentions variables. It may then turn out that independently necessary constraints on logical form outputs allow us do away with constraints on syntactic rules or outputs. Just such a situation pertains with respect to rival explanations for "cross-over phenomena". Postal (1971) proposed that the difference between (22) and (23) with respect to coreference of WH-word and pronoun be explained in terms of a syntactic constraint which prohibits movement of a constituent over a coreferential one. (22) (23)

Who betrayed the women he loved? Who did the woman he love betray?

Chomsky (1976) shows that the output constraint on logical form (24) explains not only the difference between (22) and (23), but also the impossibility of coreference in the examples of (25). (24) (25) (26)

A variable cannot be the antecedent for a pronoun to its left The woman he loved betrayed everyone The woman he loved betrayed JOHN

Counterexamples to the "cross-over constraint", such as (27), due to Postal himself, show, moreover, that an output constraint on LF is more empirically adequate than one on movements. (27)

Which of Mary's suitors does she like best?

The rest of this study will be devoted to the justification of another constraint on interpretation which, like (24), mentions the term variable, and which we intend to replace constraints on movements.

2. LOGICAL OPERATORS AND THE COMPLETE CONSTITUENT CONSTRAINT. 2.1. Logical operators (28)

Definition. A logical operator is any constituent whose categorical status is not determined by the features [±N] [±V].

We adopt the traditional view that lexical items belong to two classes, lexical formatives and grammatical formatives. The lexical formatives are those items whose categorial status is determined by the features [±N]

72

/. Guéron

[±V], which we henceforth refer to as lexical features. The grammatical formatives are all the others. We propose that grammatical formatives are assigned the feature [+Q] in the lexicon whereas lexical formatives are assigned the feature [—Q]. Interpretive rules of LF such as those which interpret a case-marked trace as a variable or an ungoverned zero pronominal as PRO will interpret any lexical category with the [+Q] feature as a logical operator. There are a number of distinguishing properties of lexical formatives as opposed to grammatical formatives. The first class is productive. Nouns, verbs, adjectives and their morphological derivatives may be added to the language (or dropped) without the grammar undergoing a change. However the loss or addition of a grammatical formative in itself constitutes a change of grammars. It will be shown below that the difference between two grammars may be described in terms of the loss or addition of the [+Q] feature alone. This will constitute an argument in favor of our association of operator status in LF with the feature [+Q]. Some operators are constituents, such as articles, complementizers, the negative particle NOT, and the adverbials ONLY, MERE, and VERY. There also exist abstract operators which are not phonologically realized, such as WH, NEG, IMPERATIVE, or EMPHATIC. And some operators have an intermediate status. Although they have a phonological realization, they must be affixed or cliticised onto a lexical item. Examples are tense, the affixes -ing and en, and clitics such as ne in French or er in Dutch. The degree of syntactic independence of a given logical operator is a parameter over which languages vary. Thus the Dutch clitic er is equivalent, in one of its uses, to the independent locative adverbial there in English. Those operators which are themselves constituents (the first class mentioned above) we will consider pure operators. We will consider as mixed operators those lexical constituents (nouns, verbs, adjectives) which acquire the operator feature [+Q] by absorption of a phonologically empty operator, such as WH-constituents or imperative verbs. Finally, there are constituents which, being assigned both lexical features and the operator feature [+Q] in the lexicon, behave ambiguously as either lexical items or operators. The extremely interesting behavior of two such items, auxiliary verbs and prepositions, will be examined below. (We will not discuss the semi-independent operators such as aspectual verbal affixes or clitics, for these involve complex problems pertaining to the definition of their scope in LF, and deserve separate treatment.) 2.2. Complete Constituents (29)

Definition. A complete constituent is an X n constituent in which X n - 1 is governed by a logical operator.

Logical Operators, Complete Constituents, Extraction Transformations

73

If in S, S is governed by a complementizer such as that, a logical operator by definition (28), then S is a complete constituent. If in NP, Ν is governed by an article such as a or the, then NP_is a complete constituent. Ingenerai, the phrase structure rules provide for X constituents in which X is governed by a pure operator of the above types. If, following the discussion in Bresnan (1972), we consider articles and complementizers as specifiers with parallel logical functions in X, we may consider the schema (30) as a specificational schema. (30)

^ α [+Q]

X

The structure (31), on the other hand, in which X is governed by a complex operator (cf. (7) above) is a quantificatìonal schema. (31)

X? XJ

X1

[+Q] Specificational schémas and quantificatìonal schémas appear to be the prototypical realisations of complete constituents as defined in (29). 9 Our claim is that they are subject to some of the same constraints in LF. 2.3. Ambiguous operators As mentioned above, certain constituents are assigned both lexical features and the operator feature [+Q] in the lexicon. Such constituents may be interpreted as either operators or lexical constituents, depending on their position within S. 2.3.1. Auxiliary verbs In English, the auxiliary verbs have, be, and do are ambiguous operators. They are assigned the lexical features [+V], [—N] and the operator feature [+Q] in the lexicon. The auxiliaries are like true verbs in absorbing tense and, at least for have and do, assigning case and a theta-role to a complement when construed as the head of VP. They have a number of properties characteristic of operators, however, as well. (i) They assign neither case nor theta-role when in (or construed as part of) the AUXILIARY position. (ii) They are not governors for the government-binding theory

74

J. Guéron

(Chomsky, (1981)). Thus they may structurally c-command a PRO-VP, as in VP DELETION configurations in English.10 (32) (33)

John is nice and Bill is PRO too John has been working and Bill has been PRO too

The fact that any "level" of VP may be deleted provided there is an auxiliary verb governing it is an automatic consequence of the governmentbinding theory if a "deleted" VP is PRO, and auxiliaries may be construed as logical operators rather than true verbs. We assume the relevant phrase structure rules to be (34): (34)

S

NP AUX VP

AUX

i Τ (DO) \ {MODALJ

VP-»V VP We will assume, following Emonds (1976), that the auxiliary verbs have and be are generated in VP and may then be raised to AUX position. The derivation of John has been working in (33) above is as in (35) 11 (35)

(i) AFFIX HOPPING IN VP (ii) VERB RAISING TO AUX

work +ing [+V] [—Q]

Logical Operators, Complete Constituents, Extraction Transformations 75 We may now describe the conditions under which an auxiliary verb has verbal properties and those under which it has operator properties. An auxiliary verb loses its operator properties when it functions as the head of VP, as in (36) (36)

John has a book

NP I John a book

We assume the head of X to be that X constituent which is^ dominated by the greatest number of uninterrupted X projections within X. Under this definition, work is the VP head in (35) and has in (36). Assuming that a constituent cannot be simultaneously construed as a lexical item and as an operator, then the fact that the head of VP must assign case and theta-role (if there are V complements) precludes its being taken for an operator in LF.12 Moreover, since the verbal head is always part of a deleted VP, the grammar presented here predicts the contrast between (37a) and (37b) below. (37) a. John has left and Mary has too b. ?* John has a book and Mary has too The operation of verb raising into AUX is parallelen important ways to movement of mixed operators into COMP. In both cases, a lexical constituent with the operator feature [+Q] moves into a position which can contain only operators and from which it governs X n . COMP and AUX may then be considered two parallel "operator positions" in S in English, one "completing" S in the sense of (29) above, the other "completing" V p 13

(ili) A third characteristic of auxiliary verbs when they function as logical operators is that they can combine with pure operators. Akmajian, Steele, and Wasow (1979) present the contrast in (38), involving contraction of the auxiliary have and the operator not, as an argument in favor of verb raising to AUX; (38) a. You have not been listening - You haven't been listening b. He must have not been listening — ""He must haven't been listening

76

J. Guéron

Following Emonds (1976), they propose that contraction between have and not takes place only when have is in AUX. (39)

Auxiliary Reduction is possible only if the element to be reduced is dominated by the syntactic node AUX. (ASW, p. 49)

The same reasoning supports the hypothesis that auxiliary have is also moved to AUX. Consider the contrast (40) (40) a.

You

been a beautiful baby

b.

You

two dollars

We propose that movement to AUX in the syntax feeds contraction on the phonological side of the grammar and operator absorption on the logical side. If we consider logical absorption of the operators have and not or must and have as parallel to the absorption of such operators in COMP as the pair rather than or more than (contiguous in base structure), the transformationally-derived who that of Middle English and other languages, or the complex WH-operator of the LF output (15) above, then we have another parallelism between the COMP and AUX nodes. Not only are these nodes landing sites for operator movement, but they are also loci for operator absorption. Our discussion of English auxiliary verbs as ambiguous operators has shown that the natural class of grammatical formatives is more adequately described in terms of the feature [+Q] than in terms of morpheme classes. Auxiliary verbs have both verbal features and the operator feature [+Q]. Their contribution to the interpretation of the S depends on their syntactic position. They are operators if dominated by AUX and true verbs if dominated by V maximal. Our use of the lexical feature [+Q] for logical interpretation recalls the use of phonological features of segments for phonetic interpretation, and the discovery in phonology that natural classes are more adequately described by means of their feature composition than in terms of phonemes.14 Some rules will now be introduced which apply to auxiliary verbs and which are equally important in other areas of the grammar. One of these is the rule of OPERATOR REDUCTION. (41)

α α l + θ Γ [-Q]

OPERATOR REDUCTION

The lexically assigned [+Q] feature may be freely reduced, in which case

Logical Operators, Complete Constituents, Extraction Transformations 77 the constituent which bears it will no longer be interpreted as an operator in LF. We assume that auxiliary verbs which are not raised to AUX position undergo this reduction process. Thus whereas in (33) both has and been are in AUX (and undergo operator absorption in LF), in (42), only has is in AUX. (42)

John has been working and Bill has too John [j\UX Τ have] [yp been working] and Bill [aUX Τ have]

+Q +V [yp PRO] too The verb be in the VP of the first S will undergo operator reduction to produce the schema (43). (43)

been working

(43) in turn undergoes an obligatory rule of UG which coindexes two true verbs when they are string and structure contiguous. (44)

VERB COINDEXING a1

a'

a1

a1

Thus two true verbs may form a complex verb, as shown in (45), just as two WH-quantifiers or two auxiliaries may form a complex operator in LF. VP

(45)

V

V

-Q The two parts of the complex verb in (45) pool their semantic resources, be providing aspectual information and working the rest of the semantically pertinent lexical features. 15 ' 16

78

J. Guéron

There is one other grammatical rule exemplified in an interesting way in the auxiliary system of English. (46)

DELETE α

We assume the deletion rule applies freely, subject to the principle of recoverability. And we take "recoverability" to mean that no material may be deleted if it is necessary either for semantic interpretation on the "right side" of the grammar or phonetic interpretation on the "left" side. On the other hand, if a constituent contains no material which is necessary for either type of interpretive rule, it must be deleted.17 These principles are at work in the phenomenon of do-deletion. We assume do to be generated in AUX and to have the feature [+V], since it can absorb tense. If it is true that the auxilary verbs have and be get the [+Q] feature from their aspectual affix, then do either is not [+Q] or else TENSE must be its [+Q] affix, transfering the [+Q] feature to the stem at the moment of lexical insertion. We propose that do has no semantic features at all beyond its categorial features [+V] [+Q]. Assuming that AUX has a position for a verbal operator, which may be a modal or do+T, plus any operator verb which raises to AUX, we get the following schema for the verb system of English: (47)

S

Do in schema (47) can be deleted at any point in the derivation, with or without raising of the other auxiliary verbs. Since it is semantically empty, do will not be needed for rules of semantic interpretation. And other [+V] constituents are available to carry tense, whether deletion occurs before or after AFFIX-HOPPING. Therefore the absence of do will not block the interpretive rules of the phonological component. On the other hand do is not deletable, as predicted by the principle of recovery of deletion, if it is needed to absorb tense. This occurs in the well-known cases in which do is the only auxiliary verb available and AFFIX HOPPING

Logical Operators, Complete Constituents, Extraction Transformations

79

into the VP is prevented by an intervening operator, such as NEG in schema (48a) or an intervening NP in the subject-auxiliary inversion configuration (48b).18 (48) a.

The other case in which do may not be deleted is when it combines with the EMPHATIC operator, as in (49) (49) a.

AUX EMPH

AUX do +T ra

If do had been deleted in the syntax, then there would have been no [+Q] element to carry the EMPHATIC operator (or the strong stress which, when attached to an auxiliary, is interpreted as emphasis), resulting in a violation of the principle of semantic recoverability.19 The principle of semantic recoverability has the consequence that only operators can delete, on the natural assumption that true lexical items are never semantically empty. Operator constituents themselves will only delete when they are either inherently semantically empty, as is do, or when they have lost whatever semantic content was attached to their [+Q] operator feature as a result of the rule of OPERATOR REDUCTION. An example of this latter process which will be discussed below is the deletion of the complementizer that in English embedded Ss. The principle of phonological recoverability, which prevents deletion of even a semantically empty operator if its loss results in an ill-formed phonological representation, accounts for the impossibility of deletion in (50b), contrasting with its acceptability in (50a). (50) a. We want John to leave b. *Our desire John to leave The deletion of the for complementizer in (50a) has no ill effects, since the verb want can assign case to the embedded subject. But, as pointed out

80

J- Guéron

in Kayne (1980), the inability of a nominal to assign case means that the embedded subject has no case in b. The principle of phonological recoverability is thus violated in (50b). 2.3.2. Prepositions Like auxiliary verbs, prepositions have among their features the categorial features [+V] [+Q] ,20 Like verbs, prepositions assign case and a theta-role to the NP they govern, in PP configurations. Thus to assigns objective (or oblique) case and theta role GOAL to its object in the PP to the house and by assigns case plus the theta role AGENT in by everyone. Note, however, that the theta-role and case assignment property of the Ρ is limited to a single NP object. Although there appear to be "two-place" prepositional "predicates", such as with in (51a), the unacceptability of such Ss when the verb is not available as a governor shows that the Ρ is affecting only a single NP. (5 l)a. I want John with Mary at the table tonight b. *It was desired John with Mary at the table When a preposition is in COMP, it functions as a pure operator. In the general case, it no longer may assign either case or theta role to the NP which it structurally governs. The disambiguation of the categorial status of Ρ in LF as a function of its syntactic position is examplified in French. The Ρ pour, for example, can assign both case and theta role to its NP object in PP but assigns neither when it appears in COMP. (52) a. J'ai fait cela pour Jean. (I did that for John) b. *Pour Jean quitter Marie m'étonnerait (For John to leave Mary would surprise me) c. J'ai fait cela pour te plaire (I did that for to please you) The hypothesis that Ρ is an ambiguous operator explains the examples of (52). Pour is construed as [+V] in PP, assigning case and a.theta-role, in (52a). In COMP, on the contrary, it is interpreted as [+Q] and may assign neither case nor theta-role. (52b) is thus ruled out by the case filter; but (52c) is acceptable, since the PRO subject is not lexically governed. The passage of Ρ constituents from lexical status within PP to operator status in COMP is analogous to the passage of auxiliary verbs from lexical status in VP to operator status in AUX. COMP and AUX being the two operator positions in English, an ambiguous operator is necessarily construed as an operator in these positions. This surprising analogy between Ρ and auxiliary verbs, which sheds a good deal of light, I believe, on the

Logical Operators, Complete Constituents, Extraction Transformations

81

interaction of the lexicon, the syntax, and logical form, is a consequence of the hypothesis that these constituents are assigned the operator feature [+Q] as well as the lexical feature [+V] in the lexicon. English does not exemplify the ideal case of disambiguation of constituents by structure. In English, Ρ in COMP loses its theta-role assignment property but keeps its case-assignment property. (This means, of course, that the two properties cannot be collapsed). (53)a. I did that for John b. For John to leave Mary would surprise me c. *I did that for to please you (53b) and (c) contrast with the analogous French examples (52b) and (c). In (53b) for assigns case to the NP it governs, saving the S from the effect of the case filter. (53c) is unacceptable for the same reason (53b) is acceptable: the PRO subject is governed by a constituent which still qualifies as a lexical verb. This violates the government binding principles.21'22 Like all operators, Ρ may undergo the rule of OPERATOR REDUCTION (41). It then becomes [+V] [—Q] and is indistinguishable from a verb. a' a1 When operator reduction applies, the string undergoes the

KK]

rule of VERB COINDEXATION (44) which applies whenever its structural description is met, producing a single complex verb.23 It is this process which explains the contrast shown in (54). (54) a. John will call up his friend b. * John will call in three minutes his friend We assume, following Emonds (1976), that particles are "intransitive" prepositions. They are assigned the features [+V] [+Q] in the lexicon but do not have features for case or theta-role assignment. OPERATOR REDUCTION AND VERB COINDEXING apply to (55a), the S-structure of (54a), to generate the LF (55b). (55)a.

b.

m

...call'

up his friend =>

. . . call' up' his friend

82

J. Guéron

Call up now behaves like a simple verb in assigning both case and theta-role to the NP structurally governed by the Ρ up.M If the same processes apply in (54b), however, an unacceptable S will result. Assume OPERATOR REDUCTION and VERB COINDEXING to have applied in (54b) along the lines shown in (56). (56) a.

b.

· · · call1 in

three minutes his friend

. . . call1 in 1 three minutes his friend hV" -Q.

Call in now acts like a simple verb, assigning theta-role and cast case to its NP object. Let us assume further that in the case at hand call and in assign the same theta role, i.e. GOAL, and the same case. The S is still unacceptable since both case and theta role will be assigned to the NP object of in and not to the final NP his friend. The S will then be ruled out by the case filter in the phonology and the theta-criterion in LF. We may thus account for the contrast between the S's of (54) while maintaining that OPERATOR REDUCTION applies freely in the grammar. Differences between otherwise-related languages may be described in terms of the logical behavior of ambiguous operators. Thus the fact that the Ρ for in English maintains more verbal properties than the analogous pour in French in COMP position explains why the two languages are similar with respect to PP interpretation but differ precisely when a Ρ functions as a complementizer. Similarly, the difference between the two languages with respect to P-stranding boils down to a difference in the interpretation of their feature matrix in an identical syntactic configuration. R. Kayne suggests that if a Ρ in French may be considered a non-governor, the contrast between (57a) and its French counterpart (57b) may be explained (class lectures, 1980 and Kayne, to appear). (57) a. *I voted for b. J'ai voté pour Kayne also suggests that if in French a Ρ is unable to coindex with a verb, and therefore cannot undergo the reanalysis which, according to Hornstein and Weinberg (to appear) accounts for P-stranding in English, we explain the contrast in (58):

Logical Operators, Complete Constituents, Extraction Transformations 83 (58) a. Who did you vote for? b *Qui as-tu voté pour? In the theory presented here, the contrasts shown in (57) and (58) are due to a single difference between French and English. In French, OPERATOR REDUCTION does not apply to Ρ within PP. Thus when Ρ is in PP, with its lexical features [+V] [+Q], it may be interpreted as either [+V], in which case it assigns case and theta-role to its object, or [+Q], in which case it assigns neither and may minimally c-command a PRO object without counting as a lexical governor. Its inability to undergo OPERATOR REDUCTION means that it cannot satisfy the structural description of VERB COINDEXING (44). Since (44) is a necessary part of the "reanalysis" process which allows P-stranding, (58b) is ruled out. In English, however, OPERATOR REDUCTION does apply to Ρ in PP.26 Thus two striking differences between English and French reduce to the apparently arbitrary fact that OPERATOR REDUCTION does not apply within PP in French.27 Another well-known difference between English and French is that French has no VP DELETION, as shown by the ungrammatically of (59b). (59) a. Mary is nice and John is too b. *Marie est gentille et Jean est aussi If a deleted VP is PRO, then this difference is accounted for, in the framework of the Government Binding theory, by the hypothesis that whereas is is interpreted as [+Q] in (59a), est in (59b) is construed as [+V]. It may be assumed that avoir and être are assigned the lexical features [+Q] and [+V], just like their English counterparts, and obligatorily undergo OPERATOR REDUCTION. If OPERATOR REDUCTION is obligatory in auxiliary verbs dominated by a projection of V, and if, as suggested by Kayne (class lectures, 1979), French has no AUX position (or no position in AUX for a verbal element), it is the absence of verb-raising into AUX which explains the difference between French and English with respect to VP DELETION.28 On the assumption that only true verbs may have clitics, the fact that clitics attach to auxiliary verbs is a sign that they are not operators. (60) a. b.

Je le vois (I see him) Je l'ai vu (I him have seen)

This assumption is in turn supported by the fact that when a verb absorbs an operator such as the IMPERATIVE operator, becoming a mixed operator itself, then cliticisization is no longer possible.

84

J. Guéron

(61 ) a.

Je le lui donne (I give it to him) [+V]

b.

Donne le lui (Give it to him) ~+V +Q J-IMPER

c.

*Le lui donne ~+V +Q _+IMPER

If the copular verb in a PSEUDO-CLEFT structure such as (62a) is interpreted as a logical operator, then the quasi-unacceptibility of (62b) in French is explainable by the hypothesis that être is always construed as a true verb in French. (62) a. What she bought is a hat b. ?*Ce qu'elle a acheté est un chapeau The semantics of focus is conveyed in French by means of CLEFT structures as in (63a) or DISLOCATION structures as in (63b). (63) a. b.

C'est un chapeau qu'elle a acheté Ce qu'elle a acheté, c'est un chapeau

Finally, recall that auxiliary verbs have no case or theta-assignment properties. Thus, if it is true that clitics represent the case and theta-role assignments of a verb, the clitic on avoir in (60b) is justified only if avoir and vu have undergone VERB COINDEXING (44) with consequent merging of their semantic properties. The structural description of this rule requires, however, that both verbs be [+V] [ - Q ] . The grammar of operators adumbrated here links such apparently unconnected phenomena as the absence of VP DELETION in French and Í+V1 the presence of clitics on auxiliary verbs to the ^ QJ features on auxiliaries. Furthermore, superficially important differences between English and French, such as the possibility of VP DELETION, are describable in terms of a single feature distinction. A third difference between languages describable in terms of operator theory involves a difference between Shakespearean English and Modern English. In Shakespearean English, demonstrative adverbial pronouns triggered subject-auxiliary inversion.29

Logical Operators, Complete Constituents, Extraction Transformations (64) a. b. c. d. e.

85

So do our minutes hasten to their end 30 (Sonnet 60, 2) Thus is his cheek the map of days outworn (S. 68,1) Then need I not to fear the worst of wrongs (S. 95, 5) So shall I live, supposing thou art true (S. 93,1) But now is black beauty's successive heir (S. 127,3)

These constituents do not trigger subject-auxiliary inversion in Modern English. The framework presented here allows a simple description of this difference: the [+Q] feature in the lexical entry of this class of adverbial has been changed to [—Q] in Modern English. Once again, a striking syntactic difference between two languages is describable in terms of the single grammatical feature [±Q]. A theory incorporating the notion logical operator, based on the feature [+Q], makes it possible to envisage calculating the differences between languages in terms of the distribution of [±Q] on the level of Logical Form. Such a calculus may help answer interesting questions concerning the filiation of languages and the rapidity of linguistic change. 2.4. Complete constituents and constraints on LF In Section 2.1. we defined the term logical operator. In Section 2.2. we defined the term complete constituent as an X n constituent in which X n ~ 1 is governed by a logical operator. The detailed analysis of ambiguous operators in 2.3. was meant to justify the inclusion of the notion logical operator in the grammar by a demonstration of its explanatory power. The rest of this study is devoted to a justification of the notion complete constituent. Two output constraints on LF are given below which crucially mention complete constituents. It will then be demonstrated that the interaction of these constraints explains a great deal of otherwise puzzling syntactic and semantic phenomena. (65)

Only complete constituents are input to interpretive rules.

(65) is a constraint on the mapping from LF to semantics. An X constituent in which X is not governed by a logical operator is uninterpretable. The examples of (66) are consequently ruled out in the semantic component. (66) a. b. c.

[nP [] [Ñ m a n l ] [§-0[ S John left]] 3 1 [pp [0] [np the store] ] 3 2

86 (67)

J. Guéron The Complete Constituent Constraint (CCC) A complete constituent X1 may not contain a variable free in X1.

In this study the term variable will refer to the traces of fronted quantifying expressions. Although such traces are phonologically empty, there is nothing in our grammar which requires that a variable be empty either in the phonology or in LF. On the contrary, discussion in Zaenen and Maling (1980) suggests that resumptive pronouns in Swedish may be construed as variables in LF. Although such pronouns "save" a structure from the effects of the ECP, which presumably applies in LF, as shown in (68), structures with resumptive pronouns still are subject to the Complete Constituent Constraint. This will be shown in Section 4. 3 3 (68)

/Ta/Ze'undrade alia om /za«'skulle komma i tid Kalle wondered everyone if he would come on time

Constraints (65) and (67) work in tandem to create an indirect link between interpretation and movement rules. If XJs the input to an interpretive rule, it may not contain a free variable. If X contains a free variable, it may not be subject to an interpretive rule. The first consequence will be illustrated in Section 3, the second in Section 4.

3. RULES OF REFERENCE AND THE COMPLETE CONSTITUENT CONSTRAINT. 3.1. (69)

I assume a class of semantic rules which includes (69a.—c.) Rules of reference a. NP "is a referring expression" b. VP "is a property statement" (predicate) c. S "is a proposition "

These rules should be understood as assigning a semantic type to the output Ss of LF, just as Illocutionary rules assign an Illocutionary type to the output Ss of the syntax. 34 Note that although rules of reference apply only to complete constituents, the rules of (69) apply but to a proper subset of complete constituents. (69c), for example, applies to S if the complementizer is that, but not if the complementizer is for or a WHconstituent. The notions presupposition and assertion will be understood in terms of (69c). If (69c) applies to S, then S is (part of) the presupposition of the S containing it. Otherwise S is (part of) the assertion of S. Note that

Logical Operators, Complete Constituents, Extraction Transformations

87

as (69c) does not apply to root S s, which have no complementizers, only embedded S's may be qualified as propositions. Root S s will qualify as assertions — as proposed in note 31. 3S In (70), the embedded S is a complete constituent which obligatorily undergoes rule (69c). It then becomes the presupposition of S and is incorporated as an unanalysable segment into the assertion. (70)

I know [§" that John loves Mary ] ]

[+Q]

presupposition of S1 assertion of S1

3.2. Let us assume that a rule of reference must apply whenever its structural description (completeness plus mention of a particular operator) is met. It follows, given the PS rules of the language and the Complete Constituent Constraint (67), that every X constituent is normally an island. This seems to be a fair description of the unmarked case in UG.36 _ If, however, the rule of OPERATOR REDUCTION (41) applies in X, this constituent will no longer satisfy the definition of complete constituent (29), and consequently will no longer be subject to the Complete Constituent Constraint. In other words, X is an extraction island in (71a) but not in (71b). (71)a.

g

α

[χ...]]

[+Q] b

·

α

[-Q]

[χ···]]

Now let us suppose that in some context β, X must be subject to a rule of reference It follows from constraint (65) that in such a context, OPERATOR REDUCTION, which normally appUes freely, may not apply, for the operator will be needed to make the X constituent "complete". Let us suppose further that (72) describes such a context. (72)

The antecedent of a pronoun obligatorily undergoes a rule of reference

It now follows that any X constituent which functions as an antedecent must be complete by definition (29) and therefore cannot contain a free variable by the CCC (67). The "Internal Structure Constraint" of G. Williams (1972) may be taken as support for (72).

88

J. Guéron

(73)

Internal Structure Constraint: A constituent may not serve as an antecedent if any node properly contained within it has been moved outside of it by a chopping transformation.

Constraint (73) is illustrated by Williams' examples (74)) (74) a. ""Who did the police ignore the girlfriend of because the Orpheus mob would protect herí b. *The helicopter which Jeff believes that Heinrich bought does not exist, but Sandra believes it anyway The "chopping rule" of WH-Movement applies to a proper part of NP in (74a) and to a proper part of S in (74b). Because these constituents function as antecedents, the S s are unacceptable by (73). VP antecedents are also subject to (73), whether the pronoun is visible as in (75), or a PRO-VP as in (76). (75) a. b. (76) a. b.

John will paint the car if you don't do it * That's the car that John will paint if you don't do it John likes Mary and Bill does _ too *Who does John like and who does Bill? (E. Williams, 1977)

Interestingly, constraint (73) applies on the output of QR, the rule which moves quantifying phrases in LF. The subject of (77) may be construed in two ways, represented roughly as (77a) and (77b). (77) a. b.

The election of no candidate seems certain (Liberman, 1974) [no candidatex] [j>jp The election of x] seems certain] [PRO elect no candidatex] [ [j^p x] seems certain]

In (77a), the quantifying phrase no candidate has scope over S, whereas in (77b), the phrase remains inside the subject. Only in (77b) may the syntactic subject be construed as the antecedent for a pronoun. (78)

The election of no candidate seems certain to me and it does to you too

Constraint (73) prevents the subject of (77a) from functioning as an antecedent. As for S antecedents, consider (79), in which anyone is both the syntactic subject of the embedded S and a wide-scope operator over a matrix S.

Logical Operators, Complete Constituents, Extraction Transformations 89 (79) a. b.

Sue will allow anyone to use her books. anyonex [§ Sue will allow [§ χ to use her books] ]

The embedded S of (79b) may not function as the antecedent of a pronoun. (80)

*Sue will allow anyone to use her books but Mary won't allow it36

The unacceptable examples above have been ruled out by means of two stipulations in the grammar proposed here. (81)

(i) referential constituents may not contain a free variable (by (65) plus (67)), (ii) the antecedents of pronouns are referential (72).

However it has not been shown that (81), in which the notion of reference is crucial, is more adequate than Williams' global constraint (73), which handles the same examples adequately, or the non-global Double-Binding Constraint of Köster (1978), presented in (82). (82)

The Double Binding Constraint37 No rule refers to Χ, Y in: ... X... [a ... Y... ] . . . Ζ . . . where Y is empty [e], and Ζ is linked to a.

Linked in (82) refers both to the relation between a moved constituent and its trace and that between an extraposed complement and its head, as described in Guéron (1980a). It clearly can jpply to the pronoun-antecedent examples given so far, where α is a X constituent, Ζ a pronoun linked to a, and Y is a trace bound by a moved constituent X. There are both similarities and differences in the empirical effects of the CCC and of the Double-Binding Constraint which we will not discuss here.38 For the cases of pronominalisation which interest us, (82) is formally superior to the CCC. For although the CCC has the initial advantage of being strictly local — [ a n . . . Y . . . ] is ruled out when Y is a free variable and a " - 1 is governed by a logical operator — the possibility of reducing operators has the consequence that this configuration is in fact ruled out only when operator reduction itself has unacceptable consequences, in particular when a n is obligatorily subject to a rule of reference. Thus, if both stipulations of (81) above are correct, then a rule such as

90

J. Guéron

(82), which prohibits antecedents from containing empty constituents, must be taken as superior to one which prohibits referential constituents from containing them and stipulates that antecedents are referential. It appears, however, that (81 i) (=(72)) is false. We will show below that the antecedent of a pronoun is not always referential. It will then follow that the CCC has different empirical consequences for cases of pronominalisation than Williams' Internal Structure Constraint (73) or the Double Binding Constraint (82). Consider the following contrast 39 (83) a. *Who did you meet the brother of and is he nice? b. Who did you tell the brother of that he was a fool? There is a crucial difference between (83a) and (83b) which can be adequately described only on the level of LF and which is plausibly responsible for the acceptability difference. He is a free pronoun in (83a) and a bound pronoun in (83b). This means that the referential status of he in b. depends on that of its antecedent, whereas that of a. is independent of that of its antecedent. As pointed out in Evans (1980) the referential value of a free pronoun depends not on that of its antecedent but on the interpretation of the entire linguistic context. The difference between free and bound pronouns may be easily shown. If a negative quantifier is substituted for the antecedent in each of the S's of (83), a. becomes absurd, but b. does not. (84)a. *I met the brother of none of those children and he was very nice b. I told the brother of none of those children that he was a fool (84a) makes contradictory claims. The first S asserts that there was no one whose brother I met, and therefore no brother in the discourse world, whereas the second part presupposes that there was a brother. In b., however, where he is coindexed with a variable in the scope of a negative quantifier and interpreted as a variable itself, it has the same referential value as its antecedent, that is, no referential value. (85)

None of those children χ1 [I told the brother of xl that he was a fool]

A bound pronoun agrees with its antecedent, whereas a free pronoun need not in cases in which the grammatical number of the antecedent is distinct from the number provided by complete semantic interpretation. 40

Logical Operators, Complete Constituents, Extraction Transformations

91

(86) a. The father of each pupil took off fhis hat Ì (χ their hatsj b. The father of each pupil came into the classroom. Then he took off his hat Ì they took off their hatsj Configurational constraints on bound pronouns are more restrictive than for free pronouns. If, for example he in (84b) above is indeed a bound pronoun, then by constraint (24), the S will be unacceptable if the pronoun precedes the variable it is coindexed with. This prediction is correct. (87)

* Although he is a fool, the brother of none of those children is handsome

If we assume that a bound pronoun is always linked to a non-referential antecedent, whereas a free pronoun, if it is coindexed at all, is linked to a referential antecedent, then the antecedent of he in (83a) is a referential expression while the antecedent of he in (83b) is not. Stipulation (8li) above now rules out (83a) but not (83b). In terms of the Double-Binding Constraint (82), the grammar presented here predicts that where a is the antecedent of a pronoun Ζ in schema (82), the Constraint will be violated in cases in which a precedes Z, but not in cases in which a follows Z.41 Such appears to be the case. (88)a. b. (89) a. b.

I asked the mother of that boy for her address Who did you ask the mother of for her address? I got her address from the mother of that boy *Who did you get her address from the mother of!

In other words the Double Binding Constraint may be violated if the antecedent of the pronoun is non-referential. Since stipulation (8Iii) has been shown to be incorrect, the CCC and the Double-Binding Constraint no longer are empirically equivalent. (8li) is correct as is. For (82) to be empirically adequate a condition must be added "if Ζ is a pronoun then a is referential". In such cases X and Ζ may be eliminated without diminishing the empirical content of (82), resulting in a constraint stating that α may not contain an empty category if it is referential. This, modulo the difference between empty categories and free variables, is precisely what (8li) above asserts. The examples of (90) make the same point. In each case, extraposition has occured from a constituent which has itself undergone a movement rule.

92

J. Guéron

(90) a.

[ftp! who [§3 e] ] do you like [^pi] [§j who comes from Boston] ? =a

b.

Υ

Ζ

X

J [ppi 'en [s j e] ] ai [NP trois [ppi e] ] dans ma poche [s j qui sont =a Υ Ζ Χ

bons] Ik ben [ppi er [g"j e] ] altijd [[ppi e] tegen [e] ] geweest [s j dat =a Y Ζ X zij ook komt] (van Riemsdijk, 1978) d. [So many more people [ppj e] [ppk e] [g"1] e] [s have I met NI»1 = α Y Y Y

c.

[ N pi e] ] [ppj from Boston] [ppk than from New York] [s

1

that

Ζ I've become a Red Sox fan myself] X (90a—d) are counterexamples to constraint (82) but not to (8li). For in each case α is not referential; it is either a quantifier, as (90a) and (d), or a logical operator as in (90b) and (c). It may be objected that examples such as (83a) or (89b) above do not really show that a referential NP may not contain a free variable. For if, as suggested in Chomsky (1977), an NP splits into two constituents NP + PP prior to extraction, then that NP will contain only the trace of an extracted PP, which is not a variable by our hypothesis. Our examples would merely show that an unbound pronoun may not refer to a "reanalysed" NP. The same demonstration may be made, however, on the output of LF, for cases in which no such split can have occured. (91) *Each boy's mother came and I admired her son (92) a. Each boy's mother came to get her son b. LF: Each boy x x's mother came to get her son (92b) violates (82) just as (83a) does. However, as the antecedent is not referential, it does not violate (8li). In this section it has been shown that the antecedent of a pronoun may contain a free variable, but only if it is construed as non-referential and the pronoun qualifies as a bound pronoun. It is^ therefore not antecedency per se which induces unacceptability of an X constituent with a missing argument, but referentiality. Antecedency is but one of a number of contextual factors which may force a referential reading, having the same status as other contextual factors with this effect, such as appositional phrases. (93a) and (94a) below are acceptable just so long as the NP antecedent may be construed as non-referential. The addition of an

Logical Operators, Complete Constituents, Extraction Transformations

93

appositive clause in the b. examples forces the referential interpretation and thus brings about the exclusion of the Ss by (83i). (93) a. b. (94) a. b.

Who did you tell the brother of that he was a fool? *Who did you tell John, the brother of that he was a fool? Everyone's mother complained about her son *Everyone's mother, Mrs. Jones, Mrs. Brown, and Mrs. White, complained about her son

An X constituent may be an antecedent and contain a free variable. But it may not both obligatorily undergo a rule of reference and contain a free variable. This is predicted by a grammar which links interpretation and extraction through the pivotal logical notion of completeness.

4. EXTRACTION ISLANDS 4.1. In this section the Complete Constituent Constraint will be applied to a number of cases in which the presence of a logical operator governing an X1 constituent is sufficient to create an extraction island. Cases have been chosen in which subjacency is irrelevant, in order to show that the CCC is needed independently of other putative constraints on rules or representations. 4.1.1. ONLY An X n constituent in the scope of ONLY is an extraction island. In (95)—(97) below, the relevant X" constituents are VP, S and NP respectively. (95) a. We didn't do much yesterday. We ONLY listened to music. b. *What kind of music did you ONLY listen toi c.

LF: . . . [ypi ONLY [ y p i - 1 listen to [np e] ] ]

[+Q] (96) a.

I ONLY know that the house is on 85th Street (but I don't know the number) b. *What street do you only know that the house is on ? (97) a. I ONLY chose three people b. *How many people did you only choose?42

Movement from inside a VP to S or S is ruled out by the CCC if VP is governed by a logical operator. Movement to the VP node itself is not, however (provided there is semantic compatibility between the moved

94

J. Guéron

constituent and the VP operator) for then the variable in VP is bound within VP itself. This accounts for the contrast between (98) and (99). (98) a. Je ne demande que trois volontaires (I ask for only three volunteers) b. Combien demandes-tu de volontaires? (How many volonteers do you ask for) c. *Combien ne demandes-tu que [e] de volontaires? (How many volonteers do you only ask for?) (99) a. J'en demande trois. (I of them ask for three) b. Je n'en demande que trois. (I of them ask for only three) Que is a polysémie grammatical formative in French (an "operator") whose interpretation depends on its context. Let us assume that que is generated as an operator inside NP in (98) and (99), and that in order to be interpreted as ONLY, the ^rue-constituent must be governed by ne in LF. 43 The LF corresponding to (98c) is (100) below and that corresponding to (99b) is (101). (100)

(LF of (99c)) * t s [OP* combien] tu [γρ

ne demandes [^p que [fsjp [zpi e] [+Q]

[+Q]

[pp de volontaires] ] ] ] (101)

(LF of (99b)) Je [ γ ρ

ne

[+Q]

[ppi en] demande [jsjp que [jsfp trois [ppi e] ] ] [+Q]

In (100) a VP in the scope of a ne (+que) operator contains a variable not bound within VP and is therefore ruled out by the Complete Constituent Constraint. In (101), however, the PP trace is either not a variable or if it is one (on the assumption that en, which binds the PP trace, is construed as a mixed operator) it is bound within VP. 44 ' 45 Further support for the CCC is provided by (102) (102)

Je ne demande qu'à connaître Jules (I ask only to get to know Jules)

which is ambiguous between a construal in which S is the scope of the operator ne+que, as shown in (103a), and one in which the embedded verb alone is its scope, as in (103b).

Logical Operators, Complete Constituents, Extraction Transformations

95

(103) a. Je ne demande que [g à connaître Jules] [+Q]

[+Q]

b. Je ne demande que [5 à connaître Jules] [+Q]

[+Q]

If WH-movement applies to the complement object, (104a) results, which, as expected, can only be associated with the LF (104b), corresponding to (103b). (104) a. Un homme queje ne demande qu'a connaître (A man that I ask only to get to know) b. Un homme [WH'-que] je ne demande qu' [+Q]

[+Q]

[g à PRO connaître [j^p x'] ] A structure in which Wh-movement occurs from an embedded S in the scope of the ne que VP operator is excluded by the CCC. An ambiguity comparable to that of (102) pertains in English as well. In (105) either the embedded S or the embedded verb may be taken as the scope of ONLY. (105)

Brutus only wanted to bury Caesar

Extraction of the embedded object eliminates one of the two possible scope construals, allowing only the one in which the embedded verb is taken as the scope of ONLY. (106)

The great man, whom Brutus only wanted to bury (and not to praise)

Whereas extraction from VP is possible only with verbal scope, other processes, such as pronominalisation, pick out sentential scope. (107) a. Je ne demande qu'à connaître Jules et Marie le demande aussi (I request only to get to know Jules and Marie requests it too) b. Brutus only wanted to bury Caesar and Cassius wanted it too Recall that the combination of extraction from VP and pronominalisation of S is not possible, as pointed out in Section 2. Thus, compare (108a) which contains a free variable, with (108b) which does not.

96

J. Guéron

(108) a.*Un homme que je ne demande qu'à connaître, mais tu ne le demandes pas (A man whom I ask only to get to know, but you don't ask it) b. Pierre, je ne demande qu'à le connaître, mais tu ne le demandes pas Given, then, that an ONLY operator in a matrix clause can take either the embedded S or the embedded verb as scope, an interesting range of rather subtle phenomena follow directly.46 4.1.2. EACH I assume each to be a mixed operator with the lexically-assigned features [+Q] [+N]. If MOVE CATEGORY is in fact MOVE X n , where X is defined in terms of the lexical features [±N], [±V], this explains why each can move in the syntax, unlike every, for example, which is a pure operator with no lexical features. EACH quantifies an NP when generated inside it in the base, as in (109), or when Chomsky-adjoined to it by Q-FLOAT, as in (110). (109) (110)

I sent each of the students an article I sent the students each an article

47

(structure of (110))',48

NP 3 of (110) is a complete constituent from which extraction is prohibited. Thus (110) above contrasts with (111), and (112b) with (113b). (111) *Which article did you send the students each? (112) a. I sent each of the students an article by Bill. b. Who did you send each of the students an article by? (113) a. I sent the students each an article by Bill. b.*Who did you send the students each an article by?

Logical Operators, Complete Constituents, Extraction Transformations 97 We have assumed that the CCC blocks extraction from a "complex" NP with a PP complement in (113b). If NP and PP are sister constituents in this structure, and if "reanalysis" of V and NP is a necessary preliminary step for extraction of the object of a P, then we can handle (113b) and similar cases, as mentioned above, by excluding a verbal string V'X1 which contains a logical operator. Once again, however, the argument made on the assumption that an article by Bill is a complex NP can be made on the basis of a QR output not involving "reanalysis". (114) below is acceptable with a wide scope interpretation of some teacher, but (115) is not. (114·) a. b. (115) a. b.

I sent each student some teacher's syllabus some teacherx I sent each student (x's syllabus) I sent the students each some teacher's syllabus some teacherx I sent the students each (x's syllabus)

(115b) violates the CCC. WH-movement from post-verbal position is similarly excluded when EACH is a VP operator, as in (116b), suggested by R. May.49 (116) a. Who did each of the men see? b.*Who did the men each see? c. structure of b. Who [g [np the men] [yp each [yp saw [fjp e] ] ] ]

4.1.3. NEG A constituent in the scope of the NEG operator may not contain a free variable. The following examples show the ill-effects of WH-movement from a VP governed by a NEG operator. (i) TOUGH-MOVEMENT (117) a. It is hard for Sally to see John these days b. John is hard for Sally to see these days (118) a. It is hard for Sally not to see John anymore b.*John is hard for Sally not to see anymore If the verb alone is the scope of NEG, then, as in (103b) above, movement is permissible, for the variable will not be part of the quantified constituent in LF. (119)

John is hard not to like

98

J. Guéron

But if there is a polarity item in the lower S, which must be part of the scope of NEG, extraction is not possible. (120) a. It is fun not to go to work until ten b.*Work is fun not to go to until ten (ii) COMPARATIVES (121)

*John saw more people than I didn't see so (iii)QR

(122) a. I asked everyone to do that problem b. LF: Everyonex I asked χ to do that problem (123) a. I didn't ask everyone to do that problem b.*Everyone x I [ γ ρ NEG [ y p asked χ to do that problem] ] (123b), which corresponds to the interpretation "For all χ, I refrained from asking χ to solve the problem" is unacceptable, for it violates the CCC. The only acceptable interpretations of (123a) are (124) and (125) (124) (125)

NOT-everyonex I asked χ to do the problem NOT (everyonex (I asked χ to do the problem))

In (124) NEG moves to the S node in LF, where it forms a complex operator with the quantifying phrase everyone, as in the syntactic output

(126) (126)

Not everyone did I ask to do the problem

In (125), the "contradictory" reading, NOT is an operator on a quantified S. Neither of these structures violates the CCC. In this section it has been shown that a constituent which normally allows WH-movement no longer allows it if it is governed by a logical operator such as ONLY, EACH, or NEG.S1 This phenomenon is explained by the hypothesis presented here that the logical operator "completes" the constituent, making it subject to the Complete Constituent Constraint.52 4.2. Operator Reduction As mentioned above, the phrase structure rules of a language provide each X constituent with a specifier which governs X^Consequently, since specifiers are logical operators by definition (28), X .constituents are normally

Logical Operators, Complete Constituents, Extraction Transformations

99

complete constituents by definition (29) and extraction islands by the Complete Constituent Constraint (67). However, there are ways to avoid the effects of the core principles. As seen in Section 3.1.3. above, a pure operator such as NEG may move in LF, thereby enlarging the domain of quantification and permitting extraction from the smaller domain it governed in the syntax (cf. (124) and (125)) Another way is by means of OPERATOR REDUCTION, (41) above, repeated here: (41)

OPERATOR REDUCTION: a

a

l+Q]

I—Q]

(41) has the effect shown in (127)

. [+Q]

X A

of [-Q]

X Δ

In (127b) X is not a complete constituent, since X is no longer governed by a logical operator. Extraction from X will now not be subject to the Complete Constituent Constraint. On the other hand, since only complete constituents are subject to interpretive rules, by (65), X is no longer interpretable and the S which contains it will be ruled out on the level of complete semantic interpretation. Therefore, although operator reduction applies in any context, it will occur successfully only in those contexts in which an incomplete constituent may be integrated into and interpreted as part of the next largest constituent which contains it. _ Recent studies suggest that the required context is one in which X is governed by a verb. 1 wil^ assume that integration is represented by coindexing. Normally each X constituent has a unique index. However an X may lose its own index in favor of that of a governing verb. The verb and X then become a single verbal string, as suggested in Hornstein and Weinberg (to appear) for governed PP. They pool their argument structure and semantic properties, as shown, in part, in (128), and the incompleteness of X is no longer an obstacle to interpretation.1S3 (128)

talk' + toi +0-AGENT +0-GOAL I -CASE _ _+ CASE-OBJECTIVE

talk1 to1 +0-AGENT +0-GOAL + CASE-OBJECTIVE

That extraction may occur only from an X constituent governed by a

100

J. Guéron

verb has already been suggested by G. Cinque, as noted in Kayne (1980), for S; by Chomsky (1977) for NP with PP complement; and by Hornstein and Weinberg (to appear) for PP. Typical extraction contexts are shown in (129). (129) a. Who do you think that John likes? (S) b. Who did you see a picture of? (NP+PP) c. Who did you talk to? (PP) Because coindexing depends on government by a verb, operator reduction, and consequently extraction, may not take place from an ungoverned X. This explains the unacceptability of extraction from extraposed constituents, dislocated constituents, adverbial complements, and sentential subjects, as shown in (130)—(133), respectively. (130) a. The fact came out that Jim had lied to Bill b.*Who did the fact come out that Jim had lied to? (131) a. He's nice, the father of that child b.*The child that he's nice, the father of (132) a. John went to that move because he likes Orson Welles b.*Who did John go to that movie because he likes? (133) a. That John likes Mary annoys me b.*Who does that John likes annoy you? (129a) is associated with the LF (134) in which V and S have been coindexed and operator reduction has taken place on that. (134)

[s Who [ s You think1 [g i that [ s John likes [e] ] ] ] ] [-Q]

When an operator is reduced, it loses that part of its semantic content which is linked to its operator function. A TH-operator (that, the, those, etc.) loses its specificational properties.54 A WH-operator loses its quantificational properties.551 If the semantic content the reduced operator maintains is redundant in context, then the operator itself may delete. 56 (135) a. I know (that) Mary was here b. I want (for) Mary to come 5 7 Since í/zaí-deletion is optional in English, the presence of that in COMP does not prove that operator reduction has not taken place. Its absence is a sure sign of operator reduction, however, and therefore of a government relation between the S which dominates it and a verb.

Logical Operators, Complete Constituents, Extraction Transformations 101 It follows that when S is ungoverned, and extraction is impossible, we also do not find that-deletion. (136)

extraposed complements a. *The fact came out Jim had lied to Bill b.*A man came by Mary had known at school (137) adverbial complements a. John went to work at an early age, that he might help his parents b.*John went to work at an early age, he might help his parents (138) gapping structures *John said he was tired and Mary _ she was raring to go (139) ?*John knows but Bill doesn't know the subway rates are going up next week Thatdeletion is thus a diagnostic for government.

4.3. Government by an ambiguous operator Coindexing X, and thus operator reduction and extraction, depends on the existence of a structural configuration in which X is governed by a verb. In this section it will be shown that structural government is a necessary but not sufficient condition for operator reduction. The governing verb may itself be an ambiguous operator, with the features [+V] [+Q], such as the copula be in English. In such cases one of two circumstances may obtain. (i) The ambiguous operator is analysed as a verb. It may then coindex with the X constituent it governs to form a verbal string V1 + X1. The resulting verbal string may itself be coindexed with another X constituent which may in turn undergo operator reduction and extraction. (140) shows this process. (140) a. Who is John glad that you like? b. LF: Who [ s John [yiis] [Apiglad] [§"» that you like [ e ] ] ] ? [-Q] (ii) The ambiguous operator is analysed as a logical operator. In this case it may not coindex with the X constituent it structurally governs, for 58

operators bind variables but do not coindex with lexical constituents. Consequently operator reduction and extraction may not take place in the embedded X constituent even though the structural configuration of government is satisfied. Such a situation exists in pseudo-cleft structures.

102

J. Guéron

(141) a. What happened was that Mary slapped John b.*John, who what happened was that Mary slapped In the grammar presented here, it is not necessary to assume that the structural relation between be and S in (141) is any different from that between be and AP in (140). Rather, the clear difference in the semantic interpretation of the two S's is taken as an indication that an interpretive difference is at stake. Be is interpreted as a verb in (140), but as an operator in (141), as shown in (142a), which corresponds to the semantic formula (142b). (142) a.

S NP

Δ

AUX

bi

VP

i

+

[ Q]

A

b. Whx (x happened) = S The semantic ambiguity of Ss such as (143) below, which have either a PREDICATIONAL or a PSEUDO-CLEFT interpretation, is discussed at length in Higgins (1973). (143)

What we saw was a picture of Mary

In our terms, this ambiguity is a consequence of the possibility of interpreting the copula as either a lexical verb or a logical operator. In the first case, the VP of (143) has the LF (144a), and in the second case, that of (144b).

(144) a. b.

[ V I was] [ N pi a picture] [pi of] [NP Mary]

PREDICATIONAL

[+V] [yi was ] [jsrpj a picture of Mary] [+Q]

PSEUDO-CLEFT

In Section 2 it was shown that when an X constituent is obligatorily input to a rule of reference, for instance, when it is the antecedent of an unbound pronoun, then_extraction, which implies operator reduction^ cannot take place from X. Here I will discuss the converse case. When X has undergone a "chopping rule", it is no longer input to rules of reference, for such rules apply only to complete constituents, by (65). Consider the contrast in (148)

Logical Operators, Complete Constituents, Extraction Transformations 103 (148)a. I believe that John is sick b. I believe it that John is sick Safir (1979) points out that (148a) and (148b) may not be used in the same discourse contexts. (149) A. Where's John today? B. a. I believe that John's sick b.*I believe it that John's sick (150) A. Mary says that John ate at your house yesterday and now he's sick B. a.*I believe that John's sick b. I believe it that John's sick, but it has nothing to do with my cooking The semantic difference between contexts (149) and (150) is that the embedded S that John is sick is asserted in (149) and presupposed in (150). Recall that "presupposed in S" means nothing more, in the framework presented here, than that a rule of reference applies to S independently of the interpretation of the S as a whole.61 A rule of this type applies to (148b), which is acceptable in context (150), but not (necessarily) to (148a), which is acceptable in context (149). 62 If a rule of reference does not apply to the embedded S of (148a), then, on that reading, the operator that must have undergone operator reduction. If so, then S is governed by the verb believe. On the other hand, since a rule of reference does apply to the embedded S of (148b), then that S is either structurally ungoverned or structurally governed but not coindexed with believe. It can be shown that S is not governed by believe in (148b). The ungrammatically of (151a) illustrates that it and S may not be part of a single constituent, while the ungrammatically of (151b), where believe takes an NP and an S complement, shows that they are not sister constituents both governed by the verb. (151)a. *It that John is sick is certain b.*I believe John that Mary is sick It thus follows that S is either extraposed or dislocated in (148b). On the proposals given here, the semantic difference between (148a) and (148b) is attributed to the interaction of structural and logical criteria. Moreover, given OPERATOR REDUCTION, other differences between the two structures are predicted to correlate with the semantic difference. For instance, extraction is predicted to be possible from the embedded S of (148a) but not from that of (148b).

104

J. Guéron

It now follows that if (143) undergoes extraction from VP, the output will have a predicationaí interpretation and not a pseudo-cleft interpretation, for the latter interpretation is ruled out by the CCC. This prediction is correct. (145)

What was what you saw a picture of?

The syntactic form of the subject is irrelevant for the extraction phenomenon examined here. 59 As Higgins points out, copulative sentences with simple subjects show the same ambiguity as sentences with free relative subjects. (146) below is adapted from Ruwet (1975)). (146)

Dona Ana's lover is Leporello's master

(146) is ambiguous between a PREDICATIONAL interpretation, as in (147a) and, in Higgins' terminology, a SPECIFICATIONAL interpretation, as in (147b). ( 147) a. NP has the property a^ b. The identity of NP = X. The subject of (147a) is referential and the VP denotes a property of the subject. The subject of (147b) is a formula which is satisfied by the focus of the operator be. Once again, extraction is compatible only with the predicationaí reading.60 (147) c. Which servant is Dona Ana's lover the master of? In this section it has been shown that structural considerations, while necessary, are not sufficient to determine the acceptability of extraction. Interpretive criteria are crucial, in particular the status in LF of the auxiliary verb governing an X constituent. The [+Q] feature on the copula verb, independently justified in Section 2 above as a way to filter VP DELETION outputs, is once again shown to have a crucial role for the interpretive rules of English. (152) a. How sick do you believe that John is? b.*How sick do you believe it that John is?63 Also predicted is that ί/ζαί-deletion is possible in the embedded clause of (148a) but not in that of (148b).

Logical Operators, Complete Constituents, Extraction Transformations 105 (153) a. I believe John's sick b.*I believe it John's sick The analysis presented here, then, specifies implicational links between three properties of (148a) — non-islandhood, ίΛαί-deletion, and assertive interpretation — and three properties of (148b) — islandhood, failure of í/¡af-deletion, and presuppositional interpretation —. The latter group of properties characterizes complete constituents; the former group, constituents which have undergone operator reduction. Although there are other proposals in the literature which explain extraction as an indirect consequence of government by a verb, I am aware of no proposal which even attempts to explain the other two phenomena, let alone predict that all three are correlated consequences of the structural configuration of government. 4.5. Bridge conditions It was shown inJJection 2 that if a rule of reference must apply to X, for example, when X is the antecedent of an unbound pronoun, then operator reduction and extraction may not apply. In this section, I will explore the idea that the bridge conditions discussed in Erteschik (1973) are simply those discourse factors which are compatible with the non-application of a rule of reference and the subsequent inclusion of the X constituent in the assertion of S. 4.5.1. S. Reinterpreting Erteschik's observations in terms of the framework presented here, extraction from an embedded S is blocked whenever the rule "S is a proposition" (69c) applies, for this rule can apply only to complete constituents. In (155b), which contrasts with (154b), extraction is blocked because the rule "S is a proposition" applies. ( 154) a. He didn't know that John loved Mary b. Mary he didn't know that John loved (155) a. He's so dumb he doesn't (even) know that two and two are four b.*Four he doesn't even know that two and two are (cf. How much are two and two?) (156b) is blocked for the same reason

106

J- Guéron

(156) a. It may seem to John that turtles can live for many years (but it's false) b.*How long may it seem to John that turtles can live? The contexts in which the rule "S is a proposition" applies obligatorily may be determined only on the level of complete semantic representation. In some cases, however, we may identify a triggering environment for the rule. The antecedency and appositive cases discussed in Section 2 constitute such environments. Another is "manner of speech" contexts, discussed in Zwicky (1971). (157) a. Who did John say that Mary had seen? b.*Who did John snort/giggle that Mary had seen? (158) a. John said that Mary was the culprit in French/slowly b.*Who did John say was the culprit in French/slowly the "manner of speech" adverbs in (158) force a propositional construal of S, thus blocking operator reduction and consequently, extraction. (157) shows that a "manner of speech" semantic feature incorporated in the main verb may serve the same function as a manner of speech adverbial on the level of complete semantic interpretation. Among the "bridge conditions" which determine when operator reduction, and thus extraction, may take place are such factors as the manner in which the lexico-semantic content of the operator restricts its functioning in LF. For example, some operators have a double logical function. They "complete" the X n constituent they govern. But they are also linked to a complement structure elsewhere in S and appear to be paired either in LF or on the level of semantic interpretation with the operator in that complement. An example is the operator pair more+than which always appear in the same structures. (159) a. John is taller than Harry is b.*John is tall than Harry is c.*John is taller Harry is We assume that the interpretation of comparative structures depends on the presence and interaction of both operators in LF. If so, it would follow that extraction is never possible from a ^«-clause, even when it is structurally governed by a verb. We suggest that it is the semantic contribution of the rAan-operator in comparative structures which accounts for an interesting contrast pointed out in Chomsky (1979). (160) a. The teacher thinks that his assistant told the class to study the lesson

Logical Operators, Complete Constituents, Extraction Transformations 107 b. The class that the teacher thinks that his assistant told [e] to study the lesson... (161) a. The lesson was harder than the teacher told the class it would be b.*The class that the lesson was harder than the teacher told [e] it would be Chomsky points out that (161b) is unacceptable even in languages which do not observe the "Wh-island" constraint, so that its ungrammatically can not be attributed to the presence of Wh-constituent in COMP. Under the analysis presented here, there is no syntactic difference between (161) above and (162), which allows extraction. (162) a. The teacher was eager that his assistant tell the class to study the lesson b. The class that the teacher was eager that his assistant tell [e] to study the lesson The possibility of extraction from the embedded S of (162b) shows that it is possible to coindex be with the AP it governs and to coindex this verbal string in turn with the embedded S, as in (163). (163)

[yi be] [Api eager] [¡ρ that S] [-Q]

By this process an S with a reduced operator is integrated into the next highest constituent. The only significant difference between (161b) and (162b) is that whereas operator reduction does no semantic damage in a that-clause, in a f/ια«-clause it destroys an operator which is needed for comparative interpretation. 64 4.5.2.

NP

There appear to be three possible interpretations for the NP object of (164) (164)

I saw a picture

(i) NP is a complete constituent in which Ν is governed by a logical operator (here the specifier a) . It undergoes rule (69a), "NP is a refer[+Q] ring expression". (ii) NP is interpreted as a quantifying expression.65 (iii) NP is an incomplete constituent subsequent to operator reduction.

108

J. Guéron

It must then be coindexed with a governing verb, in order to be interpretable. Such "reanalysis" structures are interpreted by rule (165). (165)

VERB1 NP1 is associated with the semantic formula a NP.

Rule (165) embodies the claim that existential quantification may occur independently of extraction, inside an S.66,67 4.5.3. V+NP+PP. The process of operator reduction in X and coindexing VERB and X is repeated when a reanalysed verbal string governs a second X complement, as shown in 166). (166) a. Who did you see a picture of? b. S-structure (where NP and PP are sisters either in the base or as a result of a transformational rule of extraposition.

OPERATOR REDUCTION and COINDEX picture c. intermediate LF [Npi a picture] [pp [p of ] [np [e] ] [-Q] [+Q] verbal string

OPERATOR REDUCTION and COINDEX

d. LF output [NP' a Picture] [pi of ] [flpe] [-Q] [-Q] verbal string

(166c) constitutes a violation of the CCC, but (166d) does not, since subsequent to the integration of Ρ into the verbal string, the PP node may be ignored and what remains is a complete constituent, namely, the NP trace. Erteschik points out that extraction from NP occurs only over a "bridge". (166) contrasts with the ill-formed (167).

Logical Operators, Complete Constituents, Extraction Transformations 109 (167) *Who did you destroy a picture of? Here, "bridge conditions" are construed as an (unbounded) set of contextual factors which determine the semantic coherence of an S in which X has been reanalysed with a verb. With respect to NP reanalysis, semantic coherence requires that the verbal string be interpretable in context as an expression of existential quantification. The difference between the acceptable (166) and the unacceptable (167) thus boils down to the observation that see a picture may easily be taken as an expression of existential quantification, but destroy a picture generally may not be. Suppose, however, someone is in the business of destroying pictures. Then surely one can say felicitously, (168)

Who did you destroy a picture of today, Bill?

And suppose some NP is so closely associated with destruction that the choice of this NP as a subject suffices to render the semantic content of the verb destroy redundant. (169) illustrates such a case. (169)

How many houses did the storm destroy the roof of? (L. Selkirk, p.c.)

Linguistic and extra-linguistic factors could be cited at length which, in a given context, justify or prevent the interpretation of a verbal string as an expression of existential quantification.68 The few examples cited here suffice to show that bridge is not a meaningful lexical concept. There is no class of bridge verbs as opposed to non-bridge verbs. Rule (165) interprets reanalysed V'+NP1 strings as existential expressions. When extraction applies, this interpretation is the only one available, since an NP with a reduced operator may no longer be interpreted by either (i) or (ii) of rule (164). Whether the existential interpretation is appropriate in a given context, and thus whether extraction is semantically tolerable or not, is ultimately determined in the pragmatic component.69 4.5.4. V+NP+S. 4.5.4.1. Complement clauses. Assuming that EXTRAPOSITION may apply in the syntax to transform a complex NP containing a complement clause into two sister constituents NP and S, V+NP+S cases are parallel to the V+NP+PP cases discussed in the previous section. Given a d-structure of the form (170),

110

J. Guéron

(170) a. They made the claim that the rain caused the accidents, b. . . . VP

nothing can be extracted from the embedded S because S is complete. Nor can the operator that be reduced, for S is not governed by a verb and reduction will therefore create an uninterpretable constituent. Extraposition of S may apply, however, once inside NP, creating (171), the structure typically associated with relative clauses, (171)

V [NPtNpltsIl

and once again, so that NP and S become sister constituents in VP; (172)

the [+Q]

the rain caused the accident

The string V+NP+S may now be treated just like a string V+NP+PP. The specifier the undergoes operator reduction and NP and V are coindexed. The new verbal string now governs S. The complementizer that may undergo operator reduction in turn, allowing extraction from S.

Logical Operators, Complete Constituents, Extraction Transformations 111 (173) a. The accident that they made the claim that the rain had caused b. LF NP NP the accident

S1 COMP κχ WHJ-that

/ NP

S1

M l [+Q] they 4

VP NP'

made DET I the

'S* Ñ I Ν

^^ claim

COMP S 2 that [-Q]

[e]NPj

After operator reduction and coindexing, neither the NP object nor the S 2 object are complete constituents. Although they are no longer independent input to rules of référencé, they may be interpreted as part of the verbal string. The relative clause is itself a complete constituent, since S 1 is governed by a complex Wh-that operator, but as the variable in S21 is [+Q] [+Q] bound by that operator, the CCC is not violated. The coindexing of V and NP, which permits operator reduction in the governed S, is also responsible for other peculiar properties of "make the claim" constructions observed in Emonds (1976). (i) "make the claim" structures permit i/zaf-deletion in the complement S. (174)

We made the claim the rain caused the accidents

70

In the grammar presented here, ffazi-deletion is a symptom of structural government of S by a lexical verb. Its occurrence in (174) is an argument in favor of interpreting a reanalysed verbal string as the equivalent, for government at least, of a lexical verb.71 (ii) The subject of the complement S may be extracted. (175)

The rain that John made the claim caused the accidents lasted for hours

If (175) is not to be ruled out by the EMPTY CATEGORY PRINCIPLE (Chomsky, to appear) which requires, in effect, that an empty category be governed by either its antecedent or a verb, then it is independently

112

J. Guéron

necessary to assume that the string made the claim counts as a verb for government. If the complement clause is extraposed from a subject NP, the effects observed by Emonds will not obtain, for even if the passive verb does allow reanalysis, the complement will be too "high" for a government relation between verb and complement to be established.72 (176) a. The claim was made that the rain caused the accidents b. ?*The accidents that the claim was made that the rain caused (extraction from S) c. ?*The claim was made the rain caused the accidents (thatdeletion) d. ?*The rain that the claim was made caused the accidents (subject-extraction) Whereas (176b)-(176d) are excluded for structural reasons - S is not governed by the verb - the peculiar properties of "make the claim" structures also depend on semantic considerations. Emonds notes that "discuss the claim" structures do not behave like their "make the claim" counterparts. (177) a. John discussed the claim that the rain caused the accidents b. *The accidents that John discussed the claim that the rain had caused, (extraction from S ) c. *John discussed the claim the rain caused the accidents (thatdeletion) d. *The rain that John discussed the claim caused the accidents lasted for hours (extraction of subject) The discussion of "bridge conditions" in the previous section accounts for the differences observed by Emonds. A verbal string is interpreted as an expression of existential quantification. Make the claim satisfies this requirement easily while discuss the claim does not. To discuss a claim is necessarily to do more than affirm the existence of a claim in the world of the discourse. One cannot discuss a claim unless it is assumed that the claim exists prior to the discussion of it; hence the existence of a claim is part of the presupposition of (177). If so, the NP must undergo a rule of reference independently of the interpretation of the S as a whole. But in order to undergo such a rule, it must be a complete constituent, by (65), and have retained its operator, by (29). Then operator reduction, reanalysis, and extraction may not apply. 4.5.4.2. Relative clauses. The above analysis of complement clauses within NP should in principle be applicable to relative clauses. The only

Logical Operators, Complete Constituents, Extraction Transformations 113 derivational difference between the two structures, on the analysis presented here, is that EXTRAPOSITION of S must apply twice in complement structures in order for S to be sister to the head NP, whereas in relative clauses it need apply but once. Therefore it is predicted that (178a) may be associated with the s-structure (178b). (178) a. John made the claim that you mentioned b.

VP V

NP

made DET I the

S Ν ι

COMP z ^

Ν

WH' -that

t + Q 1 claim Γ+Ql [+Q]

mentinoed

An important difference between the embedded COMP nodes in (173) and (178) is that the first contains a simplex operator that, while the [+Q] second contains a complex operator Wh'-that, formed in LF on the output [+Q] [+Q] of Wh-Movement. This difference can be utilized to proscribe operator reduction, and thus extraction, in relative clauses, via the following constraint. ( 179)

Complex operators may not undergo OPERATOR REDUCTION.

Recall that we have a complex operator when two simplex operators function as a unit in LF, as in the case of the double Wh-operator described in Higginbotham and May (1980). (179) allows the simplex "specificational" operator that to undergo operator reduction, but not the complex operator Wh'-that, which functions both as a specifier and as a quantifier within S. Where operator reduction cannot apply, the CCC forbids extraction. (180) is thus ungrammatical. (180) a. Those flowers, I know a man that sells b. LF: .. .know [^pi a man] [g Wh'-that [5 [Npi e] sells [^pj e] ] ] [+Q] [+Q] In (180) the relative clause is a complete constituent which contains a free variable. It therefore violates the CCC.

114

J. Guéron

Let us now imagine a language with relative clauses much like those of English except that in the place of that there would be a semantically empty constituent with the features [ - N ] [—V] [±Q]. Coindexation with a WH-operator in relative clause-COMP would produce either (181a) or (181b) on the level of LF. (181) a. WhMthat) [+Q]

[+Q]

b. Wh'-tthat) or (that') (cf. Pesetsky, (1978) [+Q]

[-Q]

[+Q]

Whereas (181a) is a complex operator and is unreducible by (179), (181b) would be interpreted as a simplex operator and would be subject to operator reduction. It follows that a (that)-relative would not necessarily be interpreted as a complete constituent in such a language, and that extraction from a relative clause would be possible. The -relative would behave much like a í/uzí-complement clause in English, allowing coindexing of its head with a governing verb, operator reduction in the relative COMP, and extraction from S. (180a), associated with the Lf (182), would be acceptable. 73 (182)

ÍNpj Those flowers] [j¡ I[yi know [^pk a man] [§ i [-Q] [s [NP*

E

L

SELLS

ENP> E ] ] ] ]

It appears that Swedish, Norwegian, and Danish, in which (180a) is acceptable, are just such languages. In these languages extraction is permitted from relative clauses introduced by the complementizer som. Note that som is not the same complementizer as the one which introduces complement Ss (at in Norwegian, att in Swedish). It is apparently not equivalent to the English that but rather, we propose to . Whereas at(t) is a specificational operator, som is semantically empty and can become an operator only by absorption of the semantic content of the Wh-operator in a relative clause. After absorption, the coindexed som1 is interpreted as a simplex operator, subject to optional reduction. If reduction applies, extraction from the relative clause is possible.74 The difference between English and Scandinavian relatives would then boil down to the fact that whereas a Wh-that operator is interpreted in English as a complex operator, a Wh-som operator is interpreted as simplex in the Scandinavian languages. A striking fact in favor of this proposal is the following: extraction from relative clauses is acceptable in the languages under discussion only

Logical Operators, Complete Constituents, Extraction Transformations 115 when the higher verb + NP string is interpretable as an expression of existential quantification. This term is in fact used explicitly in published discussions of such constructions. Erteschik (1973, p. 63) notes that extraction is possible from sow-relatives only when the higher verb is "an introducer parallel to the existential operator". Dubos (1980, p. 369) points out that in the Swedish version of (180), "I know a man" may not be interpreted as "I am acquainted with a man" but rather as "I know a man exists" and that in such relatives the higher verb "fonctionne comme un prédicat d'existence". Thus whereas (183a) is acceptable, (183b) and (183d), in which the verb kà'nna cannot be interpreted as an existential operator, are not. 75 (183) a. De blommorna känner jag en man som säljer (Those flowers know I a man that sells) b. *De blommorna känner jag mycket val en man som säljer (Those flowers know I very well a man that sells) c. *De blommorna har jag länge känt en man som säljer (Those flowers have I known for a long time a man that sells) The only acceptable semantic analysis for a Swedish relative clause with a missing argument is precisely the one we have associated with reanalysed V+NP strings in cases of extraction from NP with PP or S complement. The proposal is, then, that EXTRAPOSITION, OPERATOR REDUCTION, and REANALYSIS constitute a process which allows extraction, at a cost, from all types of complex NPs. Relative clauses resist this process more than NPs with PP or fftai-complements, because relativeclause operators, being complex, do not allow operator reduction. Reduction in the comp of a relative clause depends on the existence of a semantically empty S introducer, like som in Scandinavian languages, which is more of a phonological support for the Wh-operator than it is an equal partner in the logical process of operator absorption. 76 4.6. Subjacency Our analysis of extraction from complex NPs has profited from earlier studies. The interaction of EXTRAPOSITION and extraction has been documented by Chomsky (1977) for NP with PP complements, by Emonds (1976) for NP with rftaf-complements, and by Taraldsen (1979) for Norwegian relative clauses. These authors propose that EXTRAPOSITION circumvents the subjacency constraint which would otherwise exclude the extraction. That some kind of reanalysis process similar to that described by Hornstein and Weinberg (to appear) for dangling prepositions might apply to Norwegian relatives has been tentatively suggested

116

J. Guéron

in Kayne (1980). The view here is that the constraint which is circumvented by this process is not subjacency, but the Complete Constituent Constraint. There are, I feel, a number of advantages to this analysis. (i) The difference between English and languages like Swedish or Norwegian is hard to understand if EXTRAPOSITION is an "escape hatch" from subjacency. For then extraction from extraposed relative clauses should be possible in English too. Taraldsen (1979) discusses the hypothesis, first suggested by Chomsky, that EXTRAPOSITION is a stylistic rule, on the "left" side of the grammar, in English, suggesting that EXTRAPOSITION is a core rule in Norwegian. Then EXTRAPOSITION would follow extraction in English, but precede it in Norwegian, and the facts would be explained within the context of a grammar containing the subjacency constraint. In Guéron (1980a), however, I argued on the basis of examples such as (184) below that EXTRAPOSITION affects logical scope relations and is therefore a syntactic rule in English. (184) a. *Books on any of those subjects haven't come out yet b. Books haven't come out yet on any of those subjects Moreover, Emonds' analysis of extraction from "make the claim" complement structures crucially depends on EXTRAPOSITION being a syntactic rule. It was pointed out above that Emonds - analysis of complement clauses is essentially identical to Taraldsen's analysis of Norwegian somrelatives. In both cases the syntactic rule of EXTRAPOSITION creates a favorable structural environment for extraction. If EXTRAPOSITION is indeed a core rule of UG, then something more than subjacency is needed to account for the difference between English and Norwegian. I have proposed a minimal difference: whereas both languages observe the universal constraint (179), in English a Wh-that complementizer is interpreted as a complex operator, whereas in the Scandinavian languages a Wh-som operator in COMP is interpreted as simplex. This difference is ultimately traced back to a lexical distinction between that and som. That is a specifying operator in S, while som is semantically empty, a hypothesis which finds independent support in the fact that that introduces complement clauses and som does not. It has furthermore been shown in this study that one parameter along which languages vary is precisely that of the feature composition of logical operators. Thus an auxiliary verb in English may be analyzed as either [+V] or [+Q], but an auxiliary verb in French and other romance languages always functions as a [+V] item. Similarly, a preposition may be analysed as [+V] following operator reduction in PP in English, but is always [+Q] in the same position in French. Operator theory thus explains

Logical Operators, Complete Constituents, Extraction Transformations 117 the absence of both VP DELETION and P-stranding in French. The analysis of extraction differences between English and Norwegian relatives as a function of a difference in the interpretation and feature content of that and som complementizers is thus neither ad hoc nor unnatural. It appears then that whereas the subjacency constraint cannot explain the difference between English and Norwegian relatives without help from operator theory, that theory can explain these differences in a natural way without subjacency. (ii) The CCC makes different empirical predictions with respect to relative clause extraction than does the subjacency constraint. The CCC predicts that if there were a relative clause construction in English with a zero operator in COMP, then extraction could occur, provided S were governed by a verb so that reanalysis could take place. Infinitival relatives provide an example of such a construction, as shown by the contrast between (185) and (186). (185) a. b. (186) a. b.

I'm looking for a man to fix the sink The sink that I'm looking for a man to fix A man to fix the sink is needed *The sink that a man to fix is needed

(185b) has the structure shown in (187). (187)

... V I looking

VP PP

The processes described above are applicable here: (i) EXTRAPOSITION of S from the NP object of for onto the VP node, (ii) OPERATOR REDUCTION of P, (iii) OPERATOR REDUCTION of the DET in NP, (iv) COINDEX V and P, (v) COINDEX v ¥ with NP. (vi) COINDEX verbal string with S. This process gives (188). (188)

· · · looking [pi for ] [^pi a man ] [ s i φ [ s PRO to fix [np e] ] ] [-Q] [-Q]

118

J.Guéron

The logical form of infinitival relatives is parallel to such structures as (189) and (190) in which a verb is also followed by two constituents in a predicative relation. As there is no complementizer to "complete" the second constituent, extraction can occur freely. (189)a. b. (190) a. b.

I saw a man fixing the sink The sink that I saw a man fixing I consider John intelligent How intelligent do you consider John?

An analysis which presupposes the subjacency constraint, and in which EXTRAPOSITION is a stylistic rule, predicts, falsely that an infinitival relative is an extraction island.77 A grammar containing the subjacency constraint can be maintained, however, if EXTRAPOSITION is assumed to be a core rule, and if, furthermore, infinitival relatives are not formed by Wh-Movement but have empty COMP nodes. For if the infinitival relative had the form (191) below, subjacency (plus the double-COMP filter) would exclude Wh-Movement of the object NP within the relative clause. (191)

[NP1 a man] [§ Wh1 [g [^pi e] to fix the sink] ]

If infinitival relatives have the form shown in (187) above, however, in which COMP is empty, then the NP object could move in two steps to the higher COMP node without violating subjacency, as shown in (192), to produce (185b). (192)

[g" Is I'm looking for [ N P a man] [§ 0 [s PRO to t

It

fix I

e

]] ™

It would appear then that an analysis assuming subjacency may have the same empirical consequences as one without subjacency but with the CCC only if it adopts many of the same assumptions, in particular, (i) that EXTRAPOSITION is a core syntactic rule; and (ii) that a constituent can be extracted from a relative clause only in the absence of a Whelement in the COMP of that clause. The difference between the two positions now turns on the question of the doubly-filled COMP filter (cf. Pesetsky 1978 for discussion). The analysis presented here attributes the failure of extraction from English relatives to the presence of a Wh-element in COMP (or a complex Wh-that operator) which functions as a quantifier in LF. On the other hand, if subjacency is assumed, the Whconstituent is but an indirect cause of this failure. The doubly-filled COMP filter prevents a constituent in a relative from "stopping o f f ' at the

Logical Operators, Complete Constituents, Extraction Transformations 119 relative COMP on its way to the matrix COMP if this lower COMP is filled, as it is, for example, in (191) above. And the subjacency constraint prevents the constituent from rising directly to the matrix COMP in one swoop. However, if EXTRAPOSITION may apply to make a relative clause sister to its head, as it does in "make the claim" constructions and in infinitival relatives, and if the doubly-filled COMP filter is an output constraint, then, given free deletion in COMP, it is not clear why a Wh-constituent in the relative may not stop off in the relative COMP as part of its cyclic route to the matrix COMP. In other words, it is no longer clear why (193) is less acceptable than (192). (193)

[flpi the sink] [5 that [