Syntax and Semantics. Volume 2 0128613025, 9780128613023

Table of contents :
SYNTAX and SEMANTICS VOLUME 2
CONTENTS
List of Contributors
Preface
Contents of Volume 1
Explanation in Syntax
The Semantics and Syntax of Number and Numbers
The Semantics and Syntax of Numbers and Number in Natural Generative Grammar
The Syntax and Semantics of Numbers and Number in a Montague Grammar
Summary and Conclusion
References
Opacity and Reference in Language: An Inquiry into the Role of Modalities
Introduction
Opaque Environments
References
Focus and Relativization: The Case of Kihung' an
Introduction
Tone-Focus in Kihung'an
Cleft and Pseudo-Cleft in Kihung'an
The Derivation of Focusing Constructions
The Complex-Sentence and Neg-Raising Argument
The Anaphoric Deletion Argument
Wh-Questions
Discussion
Appendix 1: Object Pronoun Infixation in Kihung'an
Appendix 2: Is Igbo an Exception to the "Relative Clause and Predicate" Analysis?
References
On J. Emonds's Analysis of Extraposition
Introduction
Subject Replacement as a Root Transformation
That-Clauses and Infinitive Phrases as Noun Phrases
A Rule of Extraposition
The Status of the Rule of Extraposition
Summary
Appendix
References
Language Lateralization and Grammars
Get
The Problem
Get
's Got
Synonymy and Compositional Theories of Meaning
References
Sentence and Predicate Modifiers in English
The Thomason-Stalnaker Analysis
Lakoff' s Explanations
Problems in Lakoff's Explanation
Explaining Semantic Facts
References
Deriving S from S + Is
Be To
Not That Clauses
Pseudo-Cleft Sentences
Presuppositions and Negation
Interrogatives
Factive Complements
Sentences with Modals
That Is for You to Say
Is Deletion
Where Else to Look
References
Nonsources of Unpassives
Index

Citation preview

SYNTAX and SEMANTICS VOLUME 2

SYNTAX and SEMANTICS VOLUME 2 Edited by

JOHN P. KIMBALL University of California, Santa Cruz

SOAS WITHDRAWN LIBRARY

Jp SEMINAR PRESS

New York and London

A Subsidiary of Harcourt Brace Jovanovich, Publishers

COPYRIGHT© 1973, BY SEMINAR PRESS, INC. ALL RIGHTS RESERVED. NO PART OF THIS PUBLICATION MAY BE REPRODUCED OR TRANSMITTED IN ANY FORM OR BY ANY MEANS, ELECTRONIC OR MECHANICAL, INCLUDING PHOTOCOPY, RECORDING, OR ANY INFORMATION STORAGE AND RETRIEVAL SYSTEM, WITHOUT PERMISSION IN WRITING FROM THE PUBLISHER.

SEMINAR PRESS, INC.

111 Fifth Avenue, New York, New York 10003

United Kingdom Edition published by SEMINAR PRESS LIMITED 24/28 Oval Road, London NW!

LIBRARY OF CONGRESS CATALOG CARD NUMBER:

PRINTED IN THE UNITED STATES OF AMERICA

72-9423

CONTENTS

List of Contributors Preface Contents of Volume 1

Explanation in Syntax

ix xi xiii

1

THEO VENNEMANN GENANNT NIERFELD

The Semantics and Syntax of Number and Numbers

51

RENA TE BARTSCH The Semantics and Syntax of Numbers and Number in Natural Generative Grammar The Syntax and Semantics of Numbers and Number in a Montague Grammar Summary and Conclusion References

78 91 93

Opacity and Reference in Language: An Inquiry into the Role of Modalities

95

51

TALMY GIVON Introduction Opaque Environments References

Focus and Relativization: The Case of Kihung' an

95

96 122

123

ALEXIS T AKIZALA Introduction Tone-Focus in Kihung'an

123 126 V

vi

Contents

Cleft and Pseudo-Cleft in Kihung'an The Derivation of Focusing Constructions The Complex-Sentence and Neg-Raising Argument The Anaphoric Deletion Argument Wh-Questions Discussion Appendix 1: Object Pronoun Infixation in Kihung'an Appendix 2: Is Igbo an Exception to the "Relative Clause and Predicate" Analysis? References

On

J. Emonds's Analysis of Extraposition

130 132 135 138 140 143 145 146 148

149

F. R. HIGGINS Introduction Subject Replacement as a Root Transformation That-Clauses and Infinitive Phrases as Noun Phrases A Rule of Extraposition The Status of the Rule of Extraposition Summary Appendix References

Language Lateralization and Grammars

149 153 165 173 177

183 183 194

197

DIANA VAN LANCKER

Get

205 JOHN KIMBALL

205 206

The Problem Get 's Got Synonymy and Compositional Theories of Meaning References

213 215

Sentence and Predicate Modifiers in English

217

211

FRANK W. RENY The Thomason-Stalnaker Analysis Lakoff' s Explanations Problems in Lakoff's Explanation Explaining Semantic Facts References

217 223 227

238 244

vii

Contents

Deriving S from S + Is }AMES

247

R. HURFORD

Be To Not That Clauses Pseudo-Cleft Sentences Presuppositions and Negation Interrogatives Factive Complements Sentences with Modals That Is for You to Say Is Deletion Where Else to Look References

Nonsources of Unpassives

248 250 261 272 275 280 283 285 288 297 298

301

DOROTHY SIEGEL

Index

318

LIST OF CONTRIBUTORS

Numbers in parentheses indicate the pages on which the authors' contributions begin.

(51), Freie Universitiit Berlin and University of California, Los Angeles, Los Angeles, California

RENATE BARTSCH

(95), Department of Linguistics, University of California, Los Angeles, Los Angeles, California

TALMY GIVON

w. HENY (217), Department of Linguistics, University of Massachusetts, Amherst, Massachusetts

FRANK

(149), Department of Linguistics, Massachusetts Institute of Technology, Cambridge, Massachusetts

F. R. HIGGINS

JAMES R. HURFORD

(247), University of Lancaster, Lancaster, England

(205), College V, University of California, Santa Cruz, Santa Cruz, California

JOHN KIMBALL

(301), Department of Linguistics, Massachusetts Institute of Technology, Cambridge, Massachusetts

DOROTHY SIEGEL

(123), Department of Linguistics, University of California, San Diego, La Jolla, California

ALEXIS TAKIZALA

(197), Department of Linguistics, University of California, Los Angeles, Los Angeles, California

DIANA VAN LANCKER

(1), Department of Linguistics, University of California, Los Angeles, Los Angeles, California

THEO VENNEMANN GENANNT NIERFELD

ix

PREFACE

The Syntax and Semantics series, of which this is the second volume, is devoted to rapid publication of the most recent research in contemporary linguistics. The series developed originally as a publication source for papers read at the Summer Linguistics Conference at the University of California, Santa Cruz, in conjunction with the Summer Linguistics Program. Since its beginning, the conception of the series has been broadened. It is now intended that the series should be a yearly forum for papers on linguistics drawn from national and international sources. And indeed, of the papers found in the present volume, less than half were presented at the Summer Conference. The scope and focus of this volume is significantly different than that of a collection of symposia papers. The selection contains longer, more substantial contributions to theoretical linguistics. It is hoped that a volume with such an emphasis will be of more lasting value. The pair of papers by Vennemann and Bartsch represent the product of researches in a field the authors call natural grammar. The attempt is to provide an explanation for facts which are felt to be left unexplained by classic transformational grammar. This attempt involves use of logical notations for underlying semantic representations, plus various principles, which are not as precisely worked out as the standard theory of transformations, to relate these underlying representations to surface structure. In Vennemann' s paper one finds a general sketch of this system of syntax, while the details for a particular area are worked out in Bartsch' s paper. The success of natural grammar, and the question of where it differs from transformational grammar, if it does, remain to be evaluated. The papers by Givan and Takizala illustrate the benefit to be derived from the study of non-Inda-European languages for the study of current theoretical problems in linguistics. In Givan' s paper it is xi

xii

Preface

claimed that the difference between opaque and nonopaque contexts is demarked by overt morphology on surface structure, and in Takizala' s work we find evidence that there is underlying syntactic difference which are responsible for differences in surface structure focus, represented in English by differences of stress center. This bears on issues raised recently in linguistic theory concerning the necessity of surface structure rules of semantic interpretation for such phenomena as differences of stress. In Higgins's article we find a massive quantity of evidence arrayed on the question of the adequacy of a proposed reanalysis of the Extraposition by Emonds. The question involved is not only that of the operation of this particular rule, but also the general scheme of syntax presented by Emonds in which a particular analysis of Extraposition plays a crucial role. The lateralization article by Van Lancker is the only contribution in the area of psycholinguistics, but the point made is an important one for the theory of grammar. Briefly, there is evidence that the brain processes linguistic data in different ways depending on whether the input is productive in the sense of being generated by internalized rules or is stored as a whole. The article on get attempts to show that all occurrences of this word may be derived from roughly the same syntactic source. Besides the intrinsic interest of this fact, the question bears on the test for meaning by substitution of synonyms. The article by Heny deals with a debate between Lakoff and Thomason and Stalnacker that is to this date unpublished. The reader without access to the underground circulation of these papers should be able to reconstruct the essentials of the debate from Heny' s introduction. Hurford attempts to show that every sentence is embedded in the predicate be in deep structure. If it should turn out eventually that his conclusion is not warranted, the arguments are substantial, and the data presented will need to be accounted for by any proposed solution to the problem. The final paper by Siegel is a quite excellent treatment of sentences such as Alaska was uninhabited by white men for many years. It is shown that such sentences both cannot come by means of the standard Passive transformation, yet are passives; and that the uncannot come from sentential negation, yet is a negative. Contributions to the Syntax and Semantics series are welcome and should be submitted to the Editor. Thanks to Mary Ann Hilmes and Elaine Tringali for typing portions of the manuscript.

CONTENTS OF VOLUME 1

Possible and Must LAURI KARTTUNEN

The Modality of Conditionals -A Discussion of "Possible and Must" JOHN

P.

KIMBALL

Forward Implications, Backward Presuppositions, and the Time Axis of Verbs TALMY GIVON

Temporally Restrictive Adjectives DAVID DOWTY

Cyclic and Linear Grammars JOHN

P.

KIMBALL

On the Cycle in Syntax JOHN GRINDER

Discussion GEORGE LAKOFF

Action and Result: Two Aspects of Predication in English MICHAEL

B.

KAC

Three Reasons for Not Deriving 'Kill' from 'Cause to Die' in Japanese MASAYOSHI SHIBATANI xiii

xiv

Contents of Volume I

Kac and Shibatani on the Grammar of Killing JAMES D. MCCAWLEY Reply to Mccawley MICHAEL

B. KAc

Doubl-ing JOHN ROBERT Ross

Where Do Relative Clauses Come From? JUDITH AISSEN

On the Nonexistence of Mirror Image Rules in Syntax JORGE HANKAMER

The VP-Constituent of SVO Languages ARTHUR SCHWARTZ

Lahu N ominalization, Relativization, and Genitivization JAMES A. MATISOFF

Navaho Object Markers and the Great Chain of Being NANCY FRISHBERG

The Crossover Constraint and Ozark English SUZETTE HADEN ELGIN

Author Index-Subject Index

EXPLANATION IN SYNTAX THEO VENNEMANN CENANNT NIERFELD University of California, Los Angeles

Much descriptive work has been done in syntax over the last fifteen years, but little insight has been gained into the nature of syntax. No explanations have been proposed for why syntax is the way it is, because none have been sought. (0)

(a) (b) (c) (d) (e) (f)

(g) (h) (i)

(j)

(k)

Why do grammars contain syntactic rules? Why do syr>.tactic rules work the way they do? Why do languages have relative adjectives? Why do languages construct adverbials as operators on predicates? Why do languages construct negation as operators on predicates and arguments? Why do languages have modal verbs? Why do languages construct quantifiers as operators on arguments? Why do languages have complex verbs such as causatives? Why do languages have surface syntactic categories? Why do some languages have grammatically functional word order while others do not? Why do languages with grammatically functional word order have the kinds of word order that they do?

These are the kinds of questions that one would expect students of the theory of grammar to ask and to seek answers to. But these ques-

1

2

Theo Vennemann genannt Nierfeld

tions are not asked, and answers cannot be found in the linguistic literature. I will show in this paper that such questions can be asked meaningfully and can be answered. I will show, in fact, that the answers follow from a small number of universal principles which in tum can be easily explained.' Let me begin by asking one further question: Why have students of the theory of grammar not asked questions like these? The answer is that the function of grammars was never fully understood by linguists. To be sure, most linguists are agreed about the general definition that a grammar is a device which associates the meaning and sound of natural language discourse or, in more formal and more adequate terminology, semantic representations with phonetic representations. Most linguists agree furthermore that such a device is a system of rules which convert semantic representations into phonetic representations and conversely. But since no one until very recently had any clear notion of what semantic representations are like (though a good deal is known about the nature of phonetic representation), very little was known about the kind of work that rules in this system have to carry out, and even less was, therefore, known about the principles they obey. In particular, pretransformational grammar was at best interested in the meaning and changes of meaning of words, without any discernible interest in the semantics of syntax; and transformational grammar itself, although interested in principle in developing a theory of semantics, has never even come near its goal of describing a "semantic component," whether of an "interpretive" or a "generative" character. 2 If we accept the view that the grammar of a language is a device which associates semantic representations with phonetic representations, we must realize that nothing of value can be said about the nature of such devices until we establish first a theory of semantic representation. All we knew until very recently were a number of 1 This paper has grown directly from the work on the relation between semantics and syntax that I have carried out over the last six months with Renate Bartsch of the Freie Universifat Berlin and that is reported in our forthcoming book Semantic structures (Bartsch & Vennemann, 1972). I would not have been able to write this paper were it not for this continual and continuing collaboration. I am grateful to Dr. Bartsch for permission to use ideas and materials that originated in our joint work.

2 It is shown in Part I of Semantic structures that both Chomsky's model of transformational grammar (as described for example in Chomsky, 1971) and "generative semantics" (as described for example in Lakoff, 1972) are purely syntactic systems without a semantics and are thus not theories of natural language grammar. The same is shown of the model described in Chafe (1970).

Explanation in Syntax

3

properties of phonetic representation and of surface syntactic structure. We d:d not know anything about the rules which convert such surface structures into meaning structures because we did not know what meaning structures are like, and we did not know anything about the rules which convert meaning structures into surface structures for the same reason. While the obvious goal of linguists was thus to study the nature of semantic representation, linguists in fact embarked on a study of grammar itself as though grammars were not devices linking meaning and sound, but devices relating the surface structures of some sentences to the surface structures of others. This was done in the hope that insight into the nature of grammar and, consequently, into the nature of semantic representation, would develop in the process. Underlying this endeavor was a belief (which I cannot document but which I have found expressed in numerous discussions with and among syntacticians) that syntax is somehow more "concrete" than semantics and therefore more accessible to empirical investigation. 3 This belief is entirely false. Syntactic rules (those rules of a grammar which are not phonological rules) are not at all directly open to empirical investigation. What is concrete about language, in the sense that it is directly accessible to empirical investigation, are phonetic representations and semantic representations. We know what acertain discourse sounds like, and we know what it means (assuming that we know the language in which the discourse is conducted). Everything "in between" is abstract, in the sense that it can be known only by way of inference, by systematically comparing the meanings and sounds belonging together in discourses. To be a little more specific, phonetic structure can be investigated experimentally. Also, statements about surface syntactic structure can be tested empirically by means of various kinds of substitution and movement tests. Finally, statements about semantic representation can be tested in various ways, as I will illustrate directly. What cannot be tested are statements about syntactic rules in the absence of knowledge about the structures which these rules are supposed to operate on. For example, the belief that grammars contain movement rules which transport lexical material from one place in a syntactic structure to another 3 As a matter of fact, since the relations obtaining among surface structures of sentences were found to be fairly complex, rather abstract structures were posited as "underlying" them, structures which were not themselves amenable to direct empirical vindication but could be justified (never incontrovertibly) only by tenuous inferential "arguments." Strange talk about "mentalism in linguistics" developed as a consequence (Katz, 1964).

4

Theo Vennemann genannt Nierfeld

place in the structure is entirely unfounded because without knowledge about sentence representation other than surface representation, it is meaningless to contend that those other structures are the kind of structures in relation to which a notion of shoving lexical items about is at all possible; in fact it is meaningless to contend that those structures contain lexical items. The following example may illustrate my contention that statements about semantic representations can be tested empirically. Assume some hypothetical linguist proposes the view that semantic representations are generated recursively in the following way. 4 There is an initial symbol S. S may be rewritten as Predn (Arg 1 , Arg2 , • • • , Argn), for any natural number n. Predn must be rewritten as any one of an inventory of atomic n-place predicates. Argi, Arg2 , • • • , Argn must be rewritten as any one of an inventory of constants and variables, or as the symbol S, in accordance with the requirements of the chosen atomic predicate. There are no other ways of generating semantic representations. It is easy to show empirically that this is not a device that generates the semantic representations of natural languages. Let us ask the hypothetical linguist to give the semantic representation of the sentence (1).

(1)

Sam sliced the salami carefully.

Since to slice carefully is not an atomic predicate, a case of sentence embedding must occur in the generation of the semantic representation of (1). A possible representation then is (la), considering that (la)

s

pr r r ~

careful (in)

x ~

Pre!,

Arg,

Arg,

slice

X

y

5

Explanation in Syntax

the meaning of (1) can also be rendered as Sam was careful in slicing

the salami. Now let us ask the hypothetical linguist to represent the sentence (2) or, equivalently, (2'). (2)

Carefully Sam sliced the salami.

(2')

Sam carefully sliced the salami.

It seems to me that the meaning of these sentences is quite distinct from that of (1), even though some native speakers have to overcome some initial difficulty before they realize this. Consider the sentence Sam carefully sliced the salami carefully. (I owe this sentence to Grover Hudson, UCLA.) Also note that sentence (1) can be paraphrased as "Sam applied care to the action process of his slicing the salami." Sentence (2) [(2')] can be paraphrased as "The act of Sam's slicing the salami grew out of carefulness on Sam's side." In (1) the slicing process is characterized as careful; careful is used as a predicate adverbial, viz. a manner adverbial. In (2) the act of Sam's slicing the salami is characterized as careful; careful is used as a sentence adverbial. In (1) there is no implication that the act itself grew out of carefulness; it may in fact have been careless of Sam to slice the salami carefully. In (2) there is no implication that Sam applied any care to the slicing of the salami; he may have sliced the salami sloppily, while the act of slicing it at all grew out of carefulness. This difference between a manner adverbial and a sentence adverbial is perhaps more clearly recognizable in the following pair of sentences. (3)

The engine was humming mysteriously.

(4)

Mysteriously the engine was humming.

Also, the following sentences are clearly different in meaning.

(5)

It was raining annoyingly.

(6)

Annoyingly it was raining.

If he represents (1) as in (la), how will the hypothetical linguist represent (2)? If (3) is represented as in (3a) [or (3a')],

4 Some readers may notice a certain similarity between the recursion process described here and the generation of "logical forms" in generative semantics (cf. Lakoff, 1972). This similarity is accidental and, in any event, irrelevant to the argument.

6 (3a)

Theo Vennemann genannt Nierfeld

S

pr r i' ~

mysterious

x

pr·

~

Arg,

hum

(3a')

pr

s

X

r

~

mysterious

S

r

~

pr· hum

x

how is (4) going to be represented? And if (5) is represented as in (5a), then how is (6) going to be represented? Note that it is not possible to account for sentence pairs such as (1)-(2), (3)-(4), (5)-(6) by means of a lexical-categorial classification of all adverbs as either predicate adverbs or sentence adverbs. For while it is true that some adverbs are only predicate adverbs (e.g., well) and others only sentence adverbs (e.g., often), most adverbs are not in the lexicon at all but are generated from adjectives morphologically; and whether they can serve only as predicate adverbs (e.g., beautifully), or only as sentence adverbs (e.g., fortunately), or as both (e.g., carefully, mysteriously,

7

Explanation in Syntax

(5a)

s

~

Pr,

1•

Annoying

S

I

Preda

I

rain annoyingly, strangely), depends on the semantic properties of the underlying adjective. Furthermore, not only adverbs can function as predicate or sentence adverbials but many other constructions as well, e.g., adverbial clauses and prepositional phrases, such as in the sentences (7)-(8) and (9)-(10). (7)

The engine was humming in a surprising way.

(8)

To everyone's surprise, the engine was humming.

(9)

The engine was humming as mine never would.

(10)

Although everyone would have rather expected it to explode, the engine was humming.

There can be no question of marking all these constructions as either predicate adverbials or sentence adverbials. The difference between predicate and sentence adverbials is purely a matter of sentence semantics and can only be accounted for at the level of sentencesemantic representation. 5 Since the model of semantic representation hypothetically developed above is incapable of systematically differentiating between predicate adverbials and sentence adverbials, it fails on empirical grounds to qualify as a model of natural language semantics. There exists a more general test for semantic representations. There must be a level of representation for natural language discourse at which valid discourses can be distinguished explicitly from invalid ones. The minimal assumption is that this level is the level of semantic representations. 5

For a full account of the semantics of various types of adverbials cf. Bartsch (1972a).

8

Theo Vennemann genannt Nierfeld

As a first example, consider Fillmore's (1971) observation concerning the use of even with comparatives. Looking at another use of "even," consider the two sentences "John is taller than Bill" and "John is even taller than Bill." In the first sentence, John and Bill may both be giants or midgets-only their relative height is in question. In the second sentence, the presence of "even" entails that Bill is tall. I would say that "John is even taller than Bill" entails the two sentences, "John is taller than Bill" and "Bill is tall." Recall that "Bill is tall," now, is to be interpreted as "Bill is taller than average" [pp. 538-539].

I believe that Fillmore's observation is correct, as an observation. The question is how can we account for such observations in a semantic theory? For surely listing such observations is not doing semantics yet but merely a collecting of data for the purpose of creating an impression of what kinds of phenomena such a theory has to account for. Before we write down the semantics of Fillmore's sentence, let us see if Fillmore's observations, in addition to being correct, are also exhaustive. Do both of the entailed sentences have the same status in relation to the studied sentence? Let us negate the sentence. What are we denying when we deny that John is even taller than Bill? In my opinion, the sentence It is not the case that John is even taller than Bill denies that John is taller than Bill rather than that Bill is tall. If I were to deny the latter entailed sentence, I would have to say You are using "even" improperly because I don't consider Bill as tall. In other words, that Bill is tall is not denied when Fillmore's sentence is being denied. This is, then, a test which shows that only John is taller than Bill is asserted when John is even taller than Bill is said but Bill is tall is presupposed. Of the two entailed sentences, then, one is an assertion and the other a presupposition. We certainly would want a semantic theory to take account of this fact. The semantic theory of natural generative grammar does take account of this fact, in the following way. The sentence John is taller than Bill receives a semantic representation which also underlies John's height exceeds Bill's height. For reasons given in Bartsch and Vennemann, I represent 'John's height' and 'Bill's height' as f¥(x 1 ) andf¥(x2 ), with x 1 for 'John' and x2 for 'Bill'. The sentence Bill is tall receives a semantic representation which also underlies Bill is taller than average. For reasons given in Bartsch and Vennemann, I represent the average as Nr.v, i.e., the average height value of the set of all objects within which Bill is compared, e.g., the set of all American adult males, or all boys of Bill's age, or all giraffes, depending on the context. Now we can represent Fillmore's sentence in such a way that the semantic representations of the two entailed sentences

9

Explanation in Syntax

simply occur as conjuncts in the semantic representation of the entailing sentence. Furthermore, the presupposed conjunct can now be marked explicitly by enclosing it in angle brackets ( ). Since we are about to use all this formal apparatus, let us show a little more than we started out to show. Let us show that from John is taller than Bill one may not infer John is tall, but that from John is even taller than Bill one may infer John is tall (not merely Bill is tall, as Fillmore does). The invalidity of the first inference as opposed to the validity of the second becomes explicit in the semantic representations of natural generative grammar, cf. (11) and (12). (11)

John is taller than Bill=:/=? John is tall f¥(x1) > f¥(x2) -=f::=? f¥(x1) > Nr,Y

Indeed: a > b =:/=? a > c (12)

John is even taller than Bill~ John is tall f¥(x1) > f¥(x2) & (f¥(x2) > Nr,Y>

Indeed: a > b & b > c

~

~

f¥(x1) > Nr,Y

a > c, the law of transitivity for >.

As a second example, consider the following dialogues. (13)

A: B:

Roderich is not married. That is not true. He is a bachelor.

(14)

C: D:

Roderich is married. That is not true. He is a bachelor.

If it is correct to assume that the level of semantic representation is the same level as that on which valid and invalid inferences are distinguished in an explicit fashion, then the semantic representations of these dialogues must make it explicit that (13) is an invalid discourse but (14) is a valid one. That is, the semantic representations of these discourses must reveal formally that B in (13) is involved in a self-contradiction while D in (14) is not. How can this be done? In natural generative grammar a distinction is made between sentence semantics and word semantics. Given a sentence, its sentence semantics is represented as a logical form which contains all the information provided by the syntactic structure of the sentence, including the surface-syntactic categorial information of its lexical items. Its word semantics is represented in meaning postulates associated with lexical items in the lexicon. These meaning postulates are again logical forms. More specifically, they are strict implicative open sentences, i.e., implicative open sentences which are necessarily true for all terms substituted for the variables. These open sentences may be called upon when a lexical item appears in a dis-

10

Theo Vennemann genannt Nierfeld

course sentence. The terms occurring in the sentence semantic representation of the discourse sentence are substituted for the variables of the open sentences which are thereby turned into propositions. The propositions representing the sentence semantics and the propositions representing the word semantics together constitute the semantic representation of sentences in a discourse. The semantic representation of a sentence in a discourse is thus itself a logical form, viz. a conjunction of propositions some of which represent its sentence semantics and the others, its word semantics. Note that it is usually unnecessary to call upon all of the meaning postulates of all the lexical items of a discourse to show that it is invalid. But it is important that all of them may be called upon for this purpose. Let us apply this model of semantic representation to the discourse (13). Let x1 stand for 'Roderich,' M for the predicate 'married,' Ba for the predicate 'bachelor,' and Mn for the predicate 'man.' Applying these abbreviations and the rules of English syntax to the sentence of A, we arrive at a representation (not [M])(x 1 ). The lexicalization rule for not relates this syntactic structure to the logical form - (M(x 1 )) [or: - M(x 1 )], which may be read "it is not the case that Roderich is married.'' This logical form represents the sentence semantics of the sentence of A. As we will see, no meaning postulates for M need to be called upon to show that dialogue (13) is invalid. So far then what has been said in dialogue (13) can be represented as in (15a). (15a)

(A)

Next, let us study what B says. First, he says that what A says is not true. If we abbreviate what A says as p, then B is asserting -p. Since p is (15a), B starts out by asserting - (- M(x 1 )) [or: - - M(x 1 )]. (15b)

- - M(x 1 )

(15a),(B)

Then B goes on to say Ba(x 1 ). (15c) Since B obviously intends both of his statements to be true, the content of his part of the dialogue is fully represented in (15d). (15d)

- - M(x1 ) & Ba(x 1 )

(15b,c),(B)

In order to show that (15d) is self-contradictory, we have to call upon one of the meaning postulates associated with the word bachelor in the lexicon of English: If someone is a bachelor, it is necessarily the case that he is not married. (15e)

Ba(x) -< - M(x)

11

Explanation in Syntax

Substituting x 1 for the variable x in (15e), we get the proposition (15f). (15f)

Ba(x1 ) -< - M(x 1 )

Since B asserts (15d), he asserts in particular (15c). (15c), (15f), and the inference rule (15g) together allow us to infer (15h). (15g)

Modus ponens for the strict implication.

(15h)

(15d,c,f,g)

Again, since B asserts (15d), he asserts in particular (15b). From (15b) we can infer (15i). 6 (15i) For by accepting A's statement at all (rather than saying, e.g.: Well, how can he be married, he is only thirteen years old or . . . He is a dog, and dogs don't marry), he assumes that Roderich is the kind of object of which one can meaningfully say that it is married; this follows once again from B's stating Ba(x 1 ), because another meaning postulate of the lexical item bachelor allows us to infer that Roderich is a man, Mn(x 1 ), 7 and men are among the kind of objects of which 6 I make use of the laws of logic without mentioning them explicitly in each case. Note that it is not necessary to derive (15i) in order to receive a contradiction, because a contradiction follows also from (15b) and (15h) directly. I derive (15i) in order to illustrate the difference between a presuppositional and an assertive meaning postulate, cf. fn. 7. 7 This meaning postulate is presuppositional rather than assertive because Mn(x,) is not negated in the sentence Roderich is not a bachelor, nor is it questioned in the sentence Is Roderich a bachelor? (Cf. Kiparsky & Kiparsky, 1970, p. 156.) Thus, much of the meaning of bachelor is expressed in the following partial lexical matrix;

b::ec;}lr ] [ Ba(x) -< (Mn(x)) Ba(x) -< - M(x) (More exactly speaking, the assertive meaning postulate of bachelor is that Ba(x) strictly implies "x has never been married," i.e., (with M'(x,t) for "x is married at time t," Ba'(x,t) for "x is a bachelor at time t," and t 0 for the moment with reference to which the word is used in a sentence-in the example, the moment of uttering the sentence): Ba'(x,t0 ) -< ('ilt)(t ,;;; t 0 :::> - M'(x,t). The simpler postulate given follows from this and suffices for the example.) Similarly, married has the presuppositional meaning postulate M(x) -< (Mn(x) V Woman(x)) associated with it. Since it is presuppositional, A does not deny Mn(x 1 ) V Woman(x,) by saying - M(x 1 ). Again, by denying -M(x,) B does not deny Mn(xi) V Woman(x 1 ) but presupposes it. Since Roderich is a name unsuitable for women, Women(x,) is excluded and Mn(x,) is thus entertained by B as a presupposition.

12

Theo Vennemann genannt Nierfeld

it is meaningful to say that they are married. If B then denies the truth of the statement that Roderich is not married, he asserts that Roderich is married. Now, since both (15i) and (15h) are logically implied by (15d), so is (15j). M(x 1 ) & - M(x1)

(15j)

And since B asserts (15d), he is, by logical implication, asserting (15j), which is a contradiction. We can only conclude that B either does not know the meaning of the word bachelor, viz. that part of its meaning which is expressed in (15e), or that he "was not thinking" when he answered A, or that he had not understood what A had said. In any event, his answer is inappropriate; the discourse (13) is invalid. By contrast, the dialogue (14) is perfectly valid. C says (16a). (16a)

M(x 1 )

(C)

D says (16b). (16b) There is no meaning postulate for either married or bachelor which makes it possible to derive a contradiction from (16b). D properly denies what A says. (D may, of course, be wrong, but that is not what is at issue. What is at issue is that D's statement is neither self-contradictory nor meaningless.) In fact D denies (16a) twice, because by stating Ba(x 1 ) he is asserting once again -M(x 1 ), by implication. He is also presupposing Mn(x 1 ) and perhaps some other properties of Roderich, depending on what exactly the word bachelor means, but all of these implicata are compatible with - M(x1 ). Natural generative grammar makes it possible to make explicit the difference between valid and invalid discourses. This lends empirical support to its theory of semantics. All aspects of the meaning of a word cannot be expressed by means of meaning postulates of the kind exemplified above. Consider the sentences (17) and (18).

John believes that Mary wears padded bras, but I don't.

(17) (18)

0

John knows that Mary wears padded bras, but I don't.

(17) is grammatical but (18) is ungrammatical if know is taken to be used in its normal sense. The difference between sentences with believe (hope, wish, fear, etc.) and know (realize, be aware of, etc.) is that the former do not commit the speaker to the truth of the em-

13

Explanation in Syntax

bedded sentences while the latter do. The former verbs are in the category of nonfactive verbs while the latter are in the category of factive verbs. The use of a factive verb requires that the speaker considers the proposition expressed by the embedded sentence as true. Considering this proposition as true is a condition on using the factive verb. Use conditions are presuppositional, but not in the same sense that certain implicata of meaning postulates are presuppositional. They are presuppositional in relation to the belief the speaker has about the truth value of the proposition. These use-conditional presuppositions must be expressed directly in the semantic representation of sentences containing factive predicators, i.e., they occur in the lexicalization rules for factive predicators rather than in their meaning postulates. Let h be the lexical representation of a factive predicator. Let ( indicate the presuppositional character of a proposition. Let • designate the sense operator, i.e., an operator which eliminates the illocutionary force of a proposition, reducing it to its propositional content. Let H stand for those aspects of the propositional attitude expressed by h which go beyond the speaker's considering the proposition expressed by the embedded nominalized sentence as true. Let a be a variable over propositions. Then (19) represents the lexicalization schema for factive predicators. (19)

(a) & H(x, • a)

~

h(x, • a)

It is now easy to show that the speaker of sentence (18) is involved in a self-contradiction. Let p represent the proposition expressed by Mary wears padded bras. Let x 1 stand for John, x0 for the speaker of sentence (18). Let k stand for the lexical item know. Now application of the rules of English syntax to ( 18) yields a representation k(xi, ·p) & -k(x0 , ·p), from which application of (19) reconstitutes the sentence semantic representation (20). [Remember that presuppositions remain valid under negation, which means that - ( (a) & b) is the same as (a) & -b.] (20) The contradiction now results not inside the system of semantic representation but outside it, in the next higher logical type. This can be understood most easily if we pretend we are overhearing a discourse in which sentence (18) occurs and reformulate (20) in our own words. Let us first reformulate ( p). From our point of view, ( p) means "x 0 considers p to be true." Using T for the predicate "consider to

14

Theo Vennemann genannt Nierfeld

be true," we learn that r(x0, p). Anyone who knows that some individual considers a proposition to be true knows necessarily that that individual believes he knows that what the proposition expresses is the case. Let us write K for believe to know on our level of formulation. Then what I said in the next-to-last sentence is expressed in (21), where furthermore ~ is used for the strict implication on this level, and a for any proposition. (21) From the third conjunct of (20) we learn that x 0 does not believe that he knows that what the proposition p expresses is the case, i.e., (with = for 'not') we formulate the information we receive as = K(x0, • p ). Substituting x 0 for x and p for a in (21), and applying the inference law of contraposition for the strict implication to the resulting form and = K(x0, • p ), we receive = r(x0, p ). Since we formulated the first conjunct of (20) as r(x0, p), we derive that r(x0, p) and =r(x0, p) can both be learned from what x0 said: r(x0, p) A =r(x0, p). That is, our reformulation of what x0 said reveals that we have "learned" a contradiction. We conclude that x0 either does not know that know is a £active predicator or was not thinking when he said (18). In any event, any discourse in which (18) occurs is invalid. Note that the same contradiction arises each time a speaker uses negation with a £active predicator in the first person present indicative. It is, therefore, ungrammatical to say "I do not h that p," where h is an expression for a factive predicator and p an expression for a proposition p: (22)

a.

0

I do not know that Mary wears padded bras.

b.

0

I do not realize that the moon is not shining.

c.

0

d.

A:

1 am not aware of Mary's illness (or: Mary's being ill).

B:

?Do you know that Mary wears padded bras? No.

0

B's no is ungrammatical because it has the same semantic representation as (22a) which is ungrammatical. The proper answer for B is No, I didn't (until you told me just now). In fact, I find (22d,A) questionable for the very reason that one of the two de facto possible answers is necessarily ungrammatical. The proper question for A to ask (if it is at all proper to ask such a question!) is Did you know that

Mary wears padded bras (before I told you just now)? The comparison of sentences (17) and (18) thus shows that a system of semantic representation must not only allow us to represent

Explanation in Syntax

15

the meaning of discourses as logical forms, it must furthermore allow us to represent meaning on different logical levels (in different logical types). This follows from the postulate that semantic representations must make it possible to make explicit the difference between valid and invalid discourses (valid and invalid inferences). I hope to have illustrated sufficiently the ways in which semantic representations can be tested empirically. In particular, I hope to have made plausible the position, advocated in natural generative grammar, that semantic representations are logical forms. 8 Starting from the assumption that semantic representations are logical forms, I will now address myself to Questions (a)-(k) asked at the beginning of this paper. I show that the answers to the greater portion of those questions, (c)-(h), follow from a single principle, the PRINCIPLE OF ARGUMENT RAISING first proposed in Bartsch and Vennemann. This principle says that "good" syntax raises arguments from deeper levels of embedding to the first level, with F(x, . . . ) as a goal for surface syntactic representation. Before we speculate on the reason for the existence of this principle, let us look at some examples illustrating the way it functions. In the process of illustrating the principle, we will answer questions (c)-(h). I give in each case an exemplifying sentence, a logical representation of its sentence semantics on the left-hand side of the lexicalization rule for the surface syntactic category exemplified in the sentence ("-"indicates lexicalization rules), and, to illustrate the 8 The contentioo that semantic representations are logical forms must be understood quite literally. There is no need to reinterpret the term logical form so as to designate objects of a special "natural logic" accommodated to the special requirements of natural languages. The only seemingly special things are presuppositions. But the special character of presuppositions disappears when we interpret them as axioms for discourses. These axioms are established by shared knowledge, circumstantial evidence, and propositions introduced into the discourse by one speaker that are not immediately challenged by other speakers. Usually, when a speaker uses a factive predicator, the proposition expressed by the embedded nominalized sentence is already among the axioms of the discourse. The use of a factive predicator over a new proposition is a sneaky way of introducing new axioms. Unless this axiom is attacked immediately (i.e., the 0ld discourse is ended and a new discourse is initiated), it is accepted by all participants in the discourse. For example, if A asks B, "Why don't you realize that your theory of grammar is wrong?" any reason that B may come up with is offered under the assumption that B's theory of grammar is in fact wrong. If B does not like the failure of his theory to become axiomatic, he must end the discourse immediately, starting a new one by denying A's presupposition, e.g., by saying, "What do you mean why don't I realize? My theory of-grammar is right. But why don't you admit that your theory can't express the fact that indefinite articles come from verbs?" Linguists know that factive predicators are the best rhetorical tool for the prevention of dialogues.

16

Theo Vennemann genannt Nierfeld

argument ra1smg graphically, a tree representation of the lexicalization rule, with the raised arguments circled. (23)

[cf. Question (c)] John is taller than Bill. ftf(x1) > ftf(x2) ~ {d{Comp} }(x1, X2)

RELATIVE ADJECTIVES. a. COMPARATIVES:

p ~

{d{Comp~

b.

(23b)

John is tall.

ftf(x1) > Nn,Y

~

{d{Pos} }(x1)

p

ffi

>

(24)

POSITIVES:

fl\, c

p

NDY

~

{d{Po~

ADVERBIALS (cf. Bartsch, 1972a). [cf. Question (d)] a. PREDICATE-LIMITING ADVERBIALS: (German) Hans

ist

gesundheitlich in Ordnung 'John is okay healthwise.' 9 9 This type of adverbial is very common in German, while English makes sparing use of it. (But note examples of parallelism, e.g., Hans ist emotional gestort 'John is disturbed emotionally.') German syntax is, thus, "better" than English syntax in this regard. We will see constructions later on which are "good" in English but truly "bad" in German.

17

Explanation in Syntax

okay

(JG (x1))

F

/

\ John

I

health of

{Aumc {F}} (xi)

,,.--..,___

\okay) \ John

I

(healthwise

p

(24a)

FAc

JG/\

b.

I

{A100~

MANNER ADVERBIALS:

(fF

G

slow

\

John drives slow.

(x1))

~

driving of

p

~

~

John

{Amanc {F}} (x1)

I

(slow

\

drive)

\ John

p

(24b)

G/\c

p ~{AmanG~

JF~

c.

G

SENTENCE ADVERBIALS:

(x0

I \

appears

,

speaker

A(F(x1)))

John is apparently ill. 10

~

\~~ 'that' ill John

{Ase {F}} (x1)

\ \

~ John

(apparently ill)

(sense operator)

10 Note that English has a competing construction for propositional attitudes like appear: John appears to be ill. This construction is likewise based on the principle of argument raising. Cf. (26) below.

18

Theo Vennemann genannt Nierfeld

~

(25)

NEGATION.

a.

[cf. Question (e)]

John is not foolish (or John is no fool), cf. It is not the case that John is foolish (or a fool)Y

NEGATION AS AN OPERATOR ON PREDICATES:

~ F(x 1 )

~

{no(t){F} }(x1 )

(25a)

p {no(t~

b.

NEGATION AS AN OPERATOR ON ARGUMENTS:

male hates. 12 ~(3x)(F(x) & G(x))

~

No fe-

G({no{FN}})

11 The choice between not and no depends in part on whether the predicate is verbal or nominal (but note John is not a fool). I use not foolish and no fool only to exemplify the constructions analyzed; I do not assert that they are synonymous.

12 This ·semigrammatical sentence is given to show with minimal verbiage the structure of the sentences analyzed. A fully grammatical example is No logician would ever say such a thing. Note that the variable of the semantic representation does not appear in the lexicalized structure. Instead, the lexicalized structure has a nominalization of one of the predicates, with the second predicate predicated over this nominal. The elimination of variables is one of the functions of "good" syntax; cf. also (27), page 20. Note that without some explicit characterization of topic and comment in semantic representations, -(3x)(F(x) & G(x)) could also be lexicalized as F({no{GN}}), as in No hater (or no one that hates) is female. I suspect that the proper way of characterizing the topic is by means of conjoined presuppositional existential propositions; in the example, ((3x)F(x)) & -(3x)(F(x) & G(x)), as opposed to ((3x)G(x)) & -(3x)(F(x) & G(x))-unless the intonation peak is on the noun following no, in which case the topic-comment relation is reversed. This may be clearer for sentences containing no unicorn as a topic. I cannot pursue this matter further in this paper.

19

Explanation in Syntax

(26)

John may be ill (cf. It is possible that John [cf. Question (f)]

MODAL VERBS:

is ill).

◊

F(x 1 )

,--.__,,,

{F {may} }(x1 )

(26a)

p

~F{rn~

Note especially John can't be ill (cf. It is not the case that it is possible that John is ill): (266)

~

p

{ F {can't}}

The description of the lexicalization of modality given in (26) is greatly simplified so as to not burden the reader with too much for-

20

Theo Vennemann genannt Nierfeld

mal apparatus. All these graphs are intended to do is illustrate the difference in embedding complexity of modal structures at the level of semantic representation on the one hand and at the level of surface syntactic representation on the other. See Bartsch and Vennemann for a more detailed analysis. (27)

QUANTIFIERS AS OPERATORS ON ARGUMENTS. 13

Question (g)]

a.

UNIVERSAL QUANTIFICATION:

[cf.

. Every female is under-

paid. (Vx)(F(x) ::J G(x))

~ Px :::>

G({every{FN}})

p

(27a)

'r/x

~

p

~ (F (x), G (x)) b.

~ G

EXISTENTIAL QUANTIFICATION:

~ {every {FN}}

Some man is over-

paid. (3x)(F(x) & G(x))

~

G({some{FN}})

13 The corresponding sentences with plural quantifiers are derived from parallel semantic structures with set variables rather than individual variables. Quantifiers occur also as operators on predicates. Here too the principle of argument raising is at work. The details are very complex. Therefore, I will only sketch the Iexicalization rules. Consider the sentences (i) and (ii).

(i) The Cadillacs in the yard are all the Cadillacs we have.

(ii) The Cadillacs in the yard are some of the Cadillacs we have.

Let Fp1(X) stand for X are Cadillacs and X are in the yard, and Gp1(X) for X are Cadillacs and we have X. The tests for presuppositions show that in both (i) and (ii) it is presupposed that All the Cadillacs in the yard are Cadillacs that we have, ((\;/X)(Fp1(X) :i Gp1(X))). What is asserted is in (i) that We do not have any other Cadillacs, (\;/X)(Gp1(X) :i Fp1(X)), and in (ii) that We do have other Cadillacs, (3X)(Gp1(X) & -Fp1(X)). Here Gp1(X) and Fp1(X) are embedded as in (27). On the surface, the variables are gone (as always in "good" syntax), and nominalizations GN and FN occur in the plural ('s) and with the definite article. The surface representation has then the general predicateargument structure which is always the goal of the argument raising principle: the complex predicates [all[the['s[GN]]]] in (i) and [some[the['s[GN]]]] in (ii) are predicated of the complex argument [the['s[FN]]]. (For a discussion of plurals and their relation to determiners, cf. Bartsch, 1972b.)

21

Explanation in Syntax

P

(276)

3x

~

p

~

Px

G

~ (F ( x) , G ( x))

~ {some {FN}}

& (28)

[cf. Question (h)]

COMPLEX PREDICATES AND COMPLEX VERBS, EXEMPLIFIED BY CAUSATIVE CONSTRUCTIONS:

(a) Roderich causes Mary to fly and (b) Sanskrit Rodhariijii Marzrri patayati. With F for "fly," G for "cause," and Reichenbach' s (1947) notation for event nominalizations, (w)[p] 0 (v) meaning "the event v which is described by the proposition p," we have the following lexicalization rule for sentence (a): G(x 1 , (w)[F(x 2 )] 0 (v))

~

{F{G} }(xi, x2 )

In order to show the argument-raising nature of this complex predicate formation rule graphically, I use a functor ff: for event nominalizations:

p

.--....___,,,.. {F{G}}

The nature of the causative-verb formation rule used for the causative verb in sentence (b) can now best be understood if we view its diachronic origin. Consider an SOV language with the lexical verb G meaning "cause." [S: Subject; 0: Object or verb complement; V: Verb.] With F for the main verb, this language will have the sentence pattern

s I

V

I

G

Now phonological change will turn G into an enclitic and eventually a suffix on F:x 1x2 (FG). For example, in Pre-Indo-European there was a verb meaning "cause" which in Proto-Indo-European itself is reflected as a causative suffix 0 -eyo-, Sanskrit -aya-; cf. Mar'i patati

22

Theo Vennemann genannt Nierfeld

'Mary flies' to sentence (b), which has the meaning of the English sentence (a) plus a semantic component which we find also in the difference between English kill and cause to die, etc., viz. "physical involvement of the agent (where applicable)." I note in passing that it seems unlikely that a causative affix originates before the main verb, i.e., as a prefix. This could happen only in an SVO language. But there the configuration corresponding to the graph above would be the following:

s

I

V

I

G

There is no way for G and F to be merged into a complex verb by phonological change, because of the intervening x2 • If this view is correct, I have just predicted that English will have to make do without productive causative verb formation, and other complex verb formation processes, for millennia to come. Synchronically, complex verb formation processes are lexical rules and thus outside of the domain of sentence semantics, i.e., the semantics of syntax as opposed to word semantics. However, they are important in the present context because they demonstrate the psychic reality of the argumentraising principle diachronically because they create structures of the simple F(x,y) form. The preceding analyses show that the lexicalization schemata for various surface syntactic (lexical) categories, even though they lexicalize vastly different semantic structures, all serve the same purpose: They change complex semantic structures with deeply embedded arguments into shallow surface syntactic structures of the form F(x, . . . ). They do this by replacing sentence operators and predicates over nominalized sentences with operators sometimes on arguments which arise from nominalizations of inner predicates, but usually on inner predicates. These substitute operators are members of surface syntactic categories. The lexicalization process works in both directions because the complex semantic hierarchy can be retrieved in the shallow surface syntactic structure precisely because of the categorial information contained in the constituents. The lexicalization rules discussed are all enactments of the argument-raising principle. I am not aware of any lexicalization rules which violate this principle (although I know of several others which further illustrate it, e.g., tense, aspect). The argument-raising principle is thus a principle by which syntactic systems are constructed in the evolu-

Explanation in Syntax

23

tion of languages. It must have some intrinsic conceptual value to the linguistic mind of human beings. If we accept the principle as valid, it explains, conversely, why grammars differentiate among various syntactic categories rather than having, for example, undifferentiated constituents and an undifferentiated set of lexical items. This answers Question (i) in general and Questions (c)-(h) in particular. There are, of course, many more such categories, each with fine subcategorizations. No language utilizes all subcategories or even all main categories. But, significantly, all languages use a large portion of them. No language exists that has no syntactic categories. In fact, we would not call a communication system lacking syntactic categories a natural language. I believe that the argument-raising principle can itself be explained. One reason for its existence is that the structures resulting from its application are more compact than verbalizations not based on it would be. For example, while It is not the case that it is possible that John is ill reflects the semantic hierarchy of John can't be ill more directly than the latter sentence, the latter is preferred because it is more compact. There is a prize for brevity in syntax. Note that the former sentence, though modeling its own semantics with great accuracy, is not at all easier to understand than the condensed rendering. In fact, it is harder to understand precisely because of its hierarchical (embedding) structure and its length. John can't be ill is easy to understand because it is short and not hierarchical (nonembedding)-we do not need to be spoon-fed the hierarchies underlying sentences because we can reconstruct them ourselves from their surface syntax. There is a second reason. If we say, for example, John's driving is slow, we are modeling the sentence semantics of that other sentence, John drives slow, rather well. But not only is the former sentence longer, it is also contrary to the most likely intention of the speaker. While the speaker most likely wants to tell the listener(s) something about John, his sentence says something about John's driving, i.e., some action process in which John is involved. If a speaker says, John's height exceeds the average, he speaks about John's height, i.e., some abstract value associated with John, where in all likelihood his intention is to tell the listener(s) something about John, the person. In short, the argument-raising principle allows us to speak directly about objects that interest us, rather than about processes, acts, events, or circumstances they are involved in, or properties or values associated with them. In the case of some lexicalization rules, there is a third reason for

24

Theo Vennemann genannt Nierfeld

argument raising. Quantification involves bound variables. The syntax of natural languages loathes variables. In "good" syntax, variables are replaced by nominalizations of predicates over variables bound by the quantifiers. These nominalizations at the same time attract the quantifier, cf. (27). Here argument raising must be understood as the construction of a first-degree argument from a nominalized, second-degree embedded predicate plus a quantifier. The target structure, however, is the same as in all other cases: F(x). This is why I subsume quantifier lexicalization under the argument-raising principle. It should perhaps be emphasized that the argument-raising principle, as a principle governing the construction of grammars, could not possibly be discovered without a clear concept of the nature of semantic representation. The precision reached in natural generative grammar concerning our understanding of semantics has thus led to first important insights into the nature of syntax which could not otherwise be gained. Natural generative grammar has thus begun to answer Question (b), "Why do syntactic rules work the way they do?" in a nonimpressionistic way. Next I tum to Question (j), "Why do some languages have grammatically functional word order while others do not?" It is a part of traditional linguistic knowledge that grammatical relations can be expressed either by morphology with no reliance on word order (as in Latin and Sanskrit, cf. Staal, 1967 for Sanskrit) or by word order (as in Chinese) or by both (as in German and English, German relying more on morphology than English which relies almost entirely on word order). A drastic way of showing this is by comparing English Roderich seduced Erdmuthe-which means what it does mean only with the linear order in which the words occur in it-to Latin Rodericus Erdmutham seduxit, which means what it does mean with any one of the six possible word orders. A slightly subtler example is the distinction between manner adverbs and sentence adverbs in German and English. German does not mark de-adjectival manner adverbs morphologically but usually does mark de-adjectival sentence adverbs, most commonly by adding -er+ -weise. English, by contrast, does not distinguish de-adjectival manner adverbs and sentence adverbs morphologically but only by means of word order: Seltsam sang Maria "America the Beautiful" (or Maria sang "America the Beautiful" seltsam) 'Mary sang "America the Beautiful" strangely' versus Seltsamerweise sang Maria "America the Beautiful" 'Strangely Mary sang "America the Beautiful."' (See Bartsch, 1972a for further discussion.)

Explanation in Syntax

25

The failure of transformational grammar to appreciate and accommodate this traditional and simple insight that morphology and word order are alternative devices for expressing semantic relations on the surface is aptly critized in Bever and Langendoen (1972). Transformational grammar had to ignore it because one basic assumption of this model of grammar is incompatible with it: the assumption that deep structures (both of Chomsky's model and of generative semantics) are linearly ordered. Assume that deep structures are ordered. An earlier stage of English (Proto-Indo-European and, to a lesser extent, Old English) marked grammatical relations morphologically rather than by word order. (Word order was "free," i.e., not used to express basic grammatical functions, but to express stylistic or discourse functions such as topicalization.) Old English thus had a rich set of morphological rules, plus a set of "scrambling rules" (Ross, 1970) to undo the linear order of the deep structures. Modem English has a very rigid word order and very little morphology. Historically speaking, it has lost most of its morphological rules and almost all of its scrambling rules. Since the complexity of a transformational grammar is measured by the number of symbols needed to describe it, a transformational grammarian must conclude that English has become simpler, because fewer rules (hence: fewer symbols) are needed to describe it (viz. no rules to express the basic grammatical functions where Old English had the morphological rules; one could argue that the function of the stylistic scrambling rules must still be carried by some rules of the language -intonation, extraposition, etc.). The conclusion that English has become simpler cannot, of course, be accepted. Languages, if anything, are evolving toward greater rather than lesser complexity; at least they have in the past two million years, and there is no reason to believe the development has recently become retrograde. Languages develop cyclically from "morphology with few grammatically functional word order rules" to "word order with few morphological rules" and back again, with sound change being the causal factor throughout: first sound change grinds off the morphology and thus forces the grammar to respond by substituting word order rules in order to counter the threat of ambiguity. Next sound change degrades the positionally fixed independent function words of the language into a new morphology, which makes the word order rules redundant and leads to their loss. And so on indefinitely. Since the conclusion that English has become simpler is unacceptable, so must be the assumption that underlying representations of sentences are linearly ordered (cf. Vennemann, 1971 for this argument against ordered underlying structures of sentences).

26

Theo Vennemann genannt Nierfeld

Natural generative grammar, therefore, assumes its deepest representations of sentences not to be linearly ordered. While the answer to Question (j) is thus straightforward and traditional, Question (k), "Why do languages with grammatically functional word order have the kinds of word order that they do?" is much more complex and has, to my knowledge, never been answered (or asked). I will answer this question for a number of constructions. In the case of some constructions, the answer simply follows from the fact that if the construction were different, ambiguity would result. It is, of course, a well-known universal principle of syntax that languages do not tolerate systematic syntactic ambiguity. Where it arises, the construction is shunned, or changed; the ambiguity disappears. The PRINCIPLE OF AMBIGUITY AVOIDANCE, or the DISAMBIGUATION PRINCIPLE, by itself answers a number of questions concerning word order. Since quite a few word order universals are known, and the principle has likewise been known, it is amazing that no one ever seems to have attempted to explain the word order universals by the disambiguation principle. We will see later that all phenomena of word order cannot be explained by a simple application of the undifferentiated disambiguation principle. Some word order phenomena can only be explained by a special principle, the PRINCIPLE OF NATURAL SERIALIZATION, proposed by R. Bartsch. This principle turns out to be, in the final analysis, a highly specialized subprinciple of the disambiguation principle. In order to understand it, however, one needs an explicit system of semantic representation. Since the natural serialization principle thus once again lends empirical support to the hypothesis that the semantics of natural language is adequately represented in logical syntax, I will treat it later as a separate principle. A large number of word order universals are collected in Greenberg, 1966. Most of these universals are stated in the form of implications: If a language has such and such, it also has such and such. They are all presented as mere observations. In no case did the author attempt an explanation. 14 No attempt is made to separate the universals according to their raisons d'etre. I will, in the remaining portion 14 It seems that Greenberg made no attempt to understand the ways in which the universals relate to each other. For example, universal 22 says, "If in comparisons of superiority the only order or one of the alternative orders is standard-marker-adjective, then the language is postpositional. With overwhelmingly more than chance frequency, if the only order is adjective-marker-standard, the language is prepositional." This statement is entirely superfluous. Object-verb languages tend to have standard-adjective order and postpositions; verb-object languages tend to have adjective-standard order and prepositions. Universal 22 simply follows from this.

Explanation in Syntax

27

of this paper, explain those of the universals which I consider as systematic, rather than as accidental to the limited corpus of languages considered by Greenberg for his study; and I will limit my discussion to word order phenomena. 15 With S for subject, 0 for object (or verb complement in the wider sense), and V for the verb (or verbs), Greenberg observes that of the six possible arrangements (SVO, SOV, VSO, VOS, OSV, and OVS) only three occur as the only or dominant pattern of declarative clauses, viz. those in which S precedes 0: VSO, SVO, and SOV (universal 1). This is readily explained. In unaffected speech, the subject represents the topic of a sentence, and the object is part of the comment (or what 15 Several of Greenberg's universals follow from the principles that quite generally (a) languages do not have morphology for a tertiary category unless they have morphology for the corresponding secondary category, etc.; (b) languages do not subcategorize a secondary category unless they also subcategorize the corresponding primary category; and (c) languages do not mark secondary categories by zero. (There are, of course, exceptions arising by historical accident, e.g., the Russian gender marking in the past but not the present of verbs, contrary to (b).) Principle (a) accounts for universal 34, "No language has a trial number unless it has a dual. No language has a dual unless it has a plural." Principle (b) accounts for the universals 37, "A language never has more gender categories in nonsingular numbers than in the singular"; 44, "If a language has gender distinctions in the first person, it always has gender distinctions in the second or third person or in both"; 45, "If there are any gender distinctions in the plural of the pronoun, there are some gender distinctions in the singular also"; and probably 43, "If a language has gender categories in the noun, it has gender categories in the pronoun." Principle (c) accounts for the universals 35, "There is no language in which the plural does not have some nonzero allomorphs, whereas there are languages in which the singular is expressed only by zero. The dual and the trial are almost never expressed only by zero"; and 38, "Where there is a case system, the only case which ever has only zero allomorphs is the one which includes among its meanings that of the subject of the intransitive verb." There are other types of universals in Greenberg's list which are likewise outside the scope of the present paper. For example, the only clearly nonimplicational universal in the list, 42, "All languages have pronominal categories involving at least three persons and two numbers," seems to state simply a minimum of differentiation necessary for a language to be a useful communicative tool for human beings. Other universals express the fact that certain categories are more important conceptually than others and are therefore more often marked morphologically, e.g., 30, "If the verb has categories of person-number or if it has categories of gender, it always has tense-mode categories"; 32, "Whenever the verb agrees with a nominal subject or nominal object in gender, it also agrees in number"; and 36, "If a language has the category of gender, it always has the category of number." It seems inappropriate to me to state these differences of importance in the form of implications because implicational formulations create the impression that a causal connection exists between the two observed phenomena. The use of an implicational formulation should be avoided here.because for most of Greenberg's universals, such a causal connection does exist, and the two types of universals should be distinguished overtly.

28

Theo Vennemann genannt Nierfeld

is predicated of the subject). The topic is what the speaker is going to make a predication about; he will naturally introduce it first and will then proceed to make his comment (predication). This explains why S precedes 0. What does an interpretation of the universally preferred SO order in terms of topic and comment entail for the position of the verb? In the languages whose intonation patterns I have certain intuitions about, the intonation peak of an unaffectedly intoned sentence containing S, 0, and V is on 0. This is true of English which is SVO. It is also true of German independently of whether the sentence pattern is SVO or SOV. And it is true of Proto-Inda-European, which was SOV (Lehmann, 1970). To be sure, my intuitions about Proto-Inda-European intonation are weak on most details, but this one statement can be made with confidence because we know by the absence of tone from the finite verb that this constituent was treated phonologically as enclitic to the preceding nominal, i.e., as lacking sentence stress (e.g., Meillet, 1937, p. 367). Even though one cannot say that the verb is topical in the kind of sentence we are discussing, it is not comment in the same sense as the object either, or else we would expect it to be stressed. The verb here defines the special relation in which the object is a comment to the subject. Where then would one expect the verb to stand in relation to S and O? My own expectation would be: outside the SO pattern. Since the topic-comment relation as revealed by sentence intonation holds between S and 0, with V as the element which characterizes the particular nature of this relationship, I would expect S and O to be juxtaposed in the sentence, with V either preceding or following the SO complex. If pressed further to express my expectation concerning a possible preference for one of the two possibilities, I would say that SOV would be favored over VSO, because (a) only the former normally permits the speaker to begin the sentence with the topic, and (b) only the SV pattern guarantees a constant topic-comment serialization both in SOV sentences, where O is comment, and in SV sentences, where V is comment. VSO, by contrast, has the property of having topic-comment, in this order, in VSO sentences but comment-topic in VS sentences. I would say, then, that SOV is the most natural serialization of S, 0, and V. (The complete argument on which this suggestion is based will be given in Vennemann, in preparation.) When we look at Greenberg's classification of the 30 languages he considered in terms of a preference for the verb position, the expec-

Explanation in Syntax

29

tations sketched above are in part borne out, but in part contravened. Of the 30 languages, 6 have VSO, 11 have SOV, and 13 have SVO (Greenberg 1966, p. 77, Table 1). From a ratio of 11: 6, or 1.83: 1, it appears thus that SOV is indeed favored over VSO, as expected. As Greenberg (1966, p. 77) says, "Type I [VSO] is a definite minority." However, there is no equal confirmation of the other expectation that the preferred verb position is outside the SO complex: The sample ratio of VSO and SOV languages to SVO languages is 17: 14, or 1.21: 1. Is it, then, impossible to support the assertion that the most natural position for the verb is outside the SO complex? It seems to me that in order to assert this with absolute confidence, one would have to show that in each case of a different preference, i.e., SVO order, there is a particular reason for the language to have this order rather than one of the otherwise predicted ones. This would be an arduous task which no one could hope to complete. I believe, nevertheless, that evidence for the contention exists. It is of an indirect nature, and it is given by Greenberg even though he did not have this assertion in mind. As I discuss this evidence, I will for a while disregard the less common VSO type and concentrate on the difference between SOV and SVO languages. Universal 41 says, "If in a language the verb follows both the nominal subject and the nominal object as the dominant order, the language almost always has a case system." Greenberg says nothing about SVO languages. Let us assume that while SOV languages typically have an effective case system, SVO languages typically have none, or only very ineffective ones (like German). One function of case systems is to distinguish S and 0. SOV is then the typical pattern of languages in which S and O are differentiated from each other by an independent syntactic device, and SVO is the typical pattern of languages in which S and O are not differentiated from each other by an independent syntactic device. I propose, as a tentative conclusion, that in SVO languages it is the placement of the verb between S and O which is the syntactic device separating S and 0. If this conclusion is correct, then it is correct to say that SOV represents the sentence pattern with the natural verb position because in this type the verb has only the semantic function of specifying the relation of the comment O to the topic S, while in

30

Theo Vennemann genannt Nierfeld

SVO languages the verb has the additional syntactic function of separating Sand 0. 16 There is, fortunately, further support for this conclusion. In Old English, the sentence pattern is, roughly, SOV for dependent clauses and SVO for independent clauses. Proto-Indo-European had SOV, with many free word order possibilities because of its rich inflections (Lehmann, 1970), and contemporary English has SVO in both clause types. English thus changed SOV to SVO first in main clauses, and only subsequently in dependent clauses. Contemporary German represents the same syntactic type as Old English. Let us boldly conclude that when languages change from SOV to SVO, they do this first in independent clauses and only subsequently in dependent clauses. Why should this be so? The topic-comment relation plays an important role only in main clauses, not in dependent clauses. A typical way of changing O into the topic and S into the comment is to exchange the position of S and 0. If this is done in an SOV language without case marking, the effect is not one of topic-comment exchange but one of subject-object confusion: John the prime minister killed, if changed to The prime minister John killed, assumes not a different topic-comment structure but a different meaning altogether. If, however, the normal sentence pattern is SVO, then the topic-comment exchange can be signaled by OSV, or VOS: John killed the prime minister versus The prime minister John killed or Killed the prime minister Jbhn. While this consideration is admittedly speculative, the following is as good as Greenberg's universal 25: "If the pronominal object follows the verb, so does the nominal object." That the complementary statement is not true is shown by French and Spanish which have (nonemphatic) pronominal objects before the verb but nominal objects after it. Why should this be so? 16 One may think at first that S and O could be distinguished merely by their relative order, SO. The difficulty which arises for the listener in a putative SOV language without case marking becomes apparent if complex S and O are assumed, e.g., noun phrases with adjectival, adverbial, and clausal modifiers. Without some help from the syntax, it will be hard to tell where S ends and O begins. The need to mark the end of nominal constituents is underlined by Greenberg's universal 40: "When the adjective follows the noun, the adjective expresses all the inflectional categories of the noun. In such cases the noun may lack overt expression of one or all of these categories." Note also the remarks on topicalization of O below.

Explanation in Syntax

31

To answer this question, I first have to propose a new Greenberg universal, number 46: "If a language marks cases in the noun, it marks cases in the pronoun." My argument is as good as this universal. The universal implies that when an SOV language loses case markers through phonological change, the cases are first lost in nominals in general and in nominal objects in particular. Ambiguities arise earlier in structures with nominal objects than in structures with pronominal objects. Our interpretation of verb shift from SOV to SVO as a disambiguating move predicts that the shift occurs earlier in structures with nominal objects than in those with pronominal objects. The prediction is correct if Greenberg's universal 25 (as well as my universal 46) is correct. This supports my contention that the shift from SOV to SVO is an emergency measure to separate S and 0. Next, I would like to attempt a few explanations for the properties accompanying VSO order. In order to make my statements intelligible, I must explain a number of concomitants of SOV and SVO order. Universal 4 says, "With overwhelmingly greater than chance frequency, languages with normal SOV order are postpositional." How can this be explained? Adpositions (i.e., prepositions and postpositions) derive historically from either verbs or nouns, cf. recent English prepositions, e.g., concerning and because of (.. {[Operator[Operand]] in OV languages {o perator {o peran d}} ---,,, . [ [Operand]Operator] m VO languages

The mirror image behavior of the types with regard to serialization is illustrated below. It should be noted that the subject-predicate relationship is not an operator-operand relationship and thus is not governed by the above principle. The preferred serialization of the subject-predicate relationship is subject-before-predicate: {{Predicate}(Subject)} = } [(Subject){Predicate}]

42

Theo Vennemann genannt Nierfeld

It is shown in great detail that clauses of the kind exemplified here are constructed semantically by first predicating the content of the verb over the subject (I disregard tense and aspect in this discussion), then predicating the first adverbial over the result of a proper nominalization of the innermost proposition, then predicating the second adverbial over the result of a proper nominalization of the expanded proposition, etc. The order in which the adverbials are predicated is crucial to the meaning of the sentence. Using ff for a nominalization functor (i = 0, l, 2, . . . to distinguish different nominalizations, I= F 0 , F 1 , F 2 , • • • for different predicates), and x1 for 'John', the semantic representations of sentences like the above can be given schematically as in (32). (32) When such a structure is lexicalized "well," i.e., in such a way that the argument raising principle is satisfied, F 1 , • . . , Fn will all become operators on the innermost predicate, F 0 • In our examples, F 0 is lexicalized as a verb, V, and F 1 , • • • , Fn are lexicalized as adverbials, A 1 , • • • , An. The effect is that while x1 is deeply embedded in (32), it is an argument at the highest level in the surface representation, with V, A 1 , • • • , An forming a complex predicate F predicated of it: F(x 1 ). Bartsch's principle of natural constituent structure is satisfied only if the semantic hierarchy among the operators Fn,f~·:.\, ... ,ff;• is reflected in the surface by corresponding operator-operand relations and closeness of constituents in the string, cf. (33). (33) Bartsch' s principle of natural serialization is only satisfied if the serialization of V, A 1 , • • • , An is unidirectional. This can be done either from right to left or from left to right. The right-to-left order tends to be employed by SOV languages, including-in this regardGerman which retains many word order properties of the Proto-IndoEuropean SOV order; cf. Greenberg's universal 7:

"If in a language with dominant SOV order there is no alternative basic order, or only OSV as the alternative, then all adverbial modifiers of the verb likewise precede the verb. (This is the "rigid" subtype of III [SOV].)" The left-to-right order tends to be employed by SVO and VSO languages, including English.

43

Explanation in Syntax

(34)

a.

b.

[An[An-1[· " ·[A2[A1[Y]]]- " ·]]] [ [ [ · " "[ [ [V]A1JA2J- .. JAn-1JAnJ

in OV languages in VO languages

Note that the natural constituent structure and natural serialization principles also manifest themselves in the fact that only the highest adverbial can be isolated from the adverbial-verb nexus, cf. (3la,b) to (30a,b) and, in particular, (35) which is ungrammatical if the intended meaning is as in (30) and (31). (35)

a.

b.

To school, John went during the recess because of the exam. 0 To school, John went because of the exam during the recess. 0

The corresponding German sentences would likewise be ungrammatical on the intended readings. The same principles apply to many other structures as well. Consider the clauses in (36). (36)

a.

b.

(dass) Hans es gebracht haben muss. (German) (that) John must have brought it.

These clauses exemplify natural constituent structure and natural serialization in the verbal complex: (37)

{V {A

ux

{M d}}} =? { [V[Aux [Mod]] J in OV languages O [[[Mod]Aux]V] in VO languages

See Greenberg's universal 16: "In languages with dominant order VSO, an inflected auxiliary always precedes the main verb. In languages with dominant order SOV, an inflected auxiliary always follows the main verb." 23 23 Perhaps I should point out that I am not suggesting that Mod(als), Aux(iliaries), and (main) V(erbs) have to occur in the hierarchy exemplified above. {V{Mod{Aux}}} is possible and will be serialized in the natural way in English (John has wanted to bring it), but somewhat less naturally in German ((dass) Hans es hat bringen wollen). Note that two tense auxiliaries make a sentence ungrammatical: 0 {

{V{Aux} }{ {Mod{Aux}}} },

while several Modals are common, e.g. {V{Mod{Aux{Mod}}}} in He may have wanted to come. The reason is transparent: If the outer modal modifies the assertive force of the sentence (i.e., indicates an epistemic relation between the speaker and his proposition) it must be in the present tense; the inner modal may in this case carry a tense auxiliary. If the outer modal does not modify the assertive force of the sentence (i.e., does not relate the speaker to his proposition), but is a deontic operator, it may carry tense, but in this case no further tense operators may occur in its scope.

44

Theo Vennemann genannt Nierfeld

The natural serialization principle also sheds light on the traditional problem of the relation between passivization and the relative order of quantifiers (R. Bartsch, personal communication). The examples in (38) are self-explanatory. (38)

a.

(3x)(Vy)[Woman(x) & (Man(y) => Love(x,y))] "GooD" SURFACE: Some woman loves every man. "BAD" SURFACE: Every man is loved by some woman.

b.

(Vy)(3x)[Man(y) => (Woman(x) & Love(x,y))] "GooD" SURF ACE: Every man is loved by some woman. "BAD" SURFACE: Some woman loves every man.

(The corresponding sentences with plural quantifiers usually used in discussions are derived from parallel semantic structures with set variables rather than individual variables. The relation between some and a and between all and every is analyzed in Bartsch, 1972b.) These examples also show with great clarity that the natural serialization principle is a special application of the disambiguation principle. Bartsch' s natural serialization principle also explains several of Greenberg's universals. Consider universal 20: "When any or all of the items -demonstrative, numeral, and descriptive adjective-precede the noun, they are always found in that order. If they follow, the order is either the same or its exact opposite." Take the example these five happy girls. 24 The noun (N) girl has as its extension the class of all girls. The adjectival operator (A) happy on the noun girl cuts the subset of happy girls out of the set of all girls; the extension of {A{N}} is {x:A(x)} n {y:N(y)}. The plural operator (Plu) applied to a (simple or complex) noun (N) forms the set of all subsets of the set {x: N(x)} which is the extension of the noun; the extension of {Plu{N}} is thus {L :L C {x :N(x)} }. A numeral operator (Num) applied to a (plural) noun cuts from this set the subset of all sets whose cardinal number is expressed by the numeral; the extension of {Num{Plu{N}}} is thus (with Pl as an operator which 24 My analysis of this noun phrase is inspired by Bartsch's (1972b) treatment of number and numbers. There it is also shown that the syntactic rules for noun phrases with determiners, number, plurals, and descriptive adjectives once again operate in conformity with the argument-raising principle.

45

Explanation in Syntax

maps a set on its cardinal number): {L' :ftf(L') = Num}

n

{L :L C {x :N(x)} },

which is equal to {L :ftf(L) = Num & L C {x :N(x)} }. In our example, where instead of a simple noun N we have the complex noun {A{N} }, we get {L:ftf(L)=5 &LC {x:A(x)} n {y:N(y)}}, which is the extension of five happy girls. The determiner (D) these singles out one of the sets in this set by adding a specification F, e.g., by pointing at one set of five happy girls; the extension of {D{Num{Plu{A{N}}}} }, here these five happy girls, is thus (tL)[fM(L) = 5 & L c {x :A(x)} n {y :N(y)} & F(L)]. The natural serialization principle predicts that {D{Num{Plu{A{N}}}}} will be serialized either as [D[Num[Plu[A[N]]]]] or as [[ [ [ [N]A]Plu]Num]D].

The order [[[N]][D[Num[Plu[A]]]]] is unusual. I explain it as appositional in origin: girls, these five happy (ones). Where it is the only construction of these constituents, I would expect to be dealing with a young SVO language developed from an SOV language. English may start doing such things soon as it continues to develop toward the pure SVO type which has, of course, [[ [[ [N]A]Plu]Num]D]. Note incidentally that English does not treat the constituent Plu according to the above formulas but marks it at the end of the noun (or in the noun), and on some determiners. This is the residue of a system in which number was marked as a harmony on noun phrases (agreement). Other Greenberg universals are explained with even greater ease by Bartsch's principles. Number 28 says: "If both the derivation and inflection follow the root, or they both precede the root, the derivation is always between the root and the inflection."

Indeed, since derivation is performed in the lexicon where it builds stems, and inflection comes from the syntax and is added to lexical items, whether primitive or derived, we would not expect any other serialization than that described by Greenberg. Number 39 says: "Where morphemes of both number and case are present and both follow or both precede the noun base, the expression

46

Theo Vennemann genannt Nierfeld

of number almost always comes between the noun base and the expression of case." Indeed, number is an operator on nouns and is inherent in noun phrases. Case comes from the role a noun phrase plays in relation to the verb. It is easy to state other such universals, without even looking into individual languages, e.g., universal 47:

If number and gender marking either both precede or both follow the noun base, gender marking will stand closer to the noun base than number marking. Bartsch' s principles of natural constituent structure and natural serialization have predictive power for language change. They explain many of the word order changes that follow the basic verb shift which is at the heart of all syntactic type changes. This is very transparent in contemporary German. German has a very bad syntactic system due to its incomplete shift from SOV to SVO. The finite verb has shifted in main clauses, but not in dependent clauses. The order of all other constituents is still that of an SOV language: adverbials are serialized from right to left (in English, from left to right); verbs serialize from right to left (in English, from left to right); direct and indirect objects are serialized from right to left (in English, from left to right in john gave the book to Mary, with john gave Mary the book still available from the old SOV days). All these concomitants of the SOV pattern are perfect for dependent clauses, but for main clauses they are all "bad." Since the verb position of the main clause pattern is basic (it is "unmarked": most frequent, learned earliest by children, etc.), Bartsch's principles predict that German will change all its old serialization rules until all constituents are again serialized naturally. In short, German must undergo similar syntactic development in the future as English has in the last 1500 years. As pointed out in Vennemann, 1971, transformational grammar makes the wrong prediction when it assumes for German the deep structure order of an SOV language, plus a transformation moving the finite verb to the second position in main clauses. This is the simplest analysis, but it characterizes German as a syntactically simple language, which it is not. And it makes the incorrect prediction that German may revert to a pure SOV language by losing that one movement transformation. Natural generative grammar assumes German to be an SVO language with very many complicated rules arranging all constituents in an unnatural order. The theory predicts that German will develop into

Explanation in Syntax

47

a pure SVO language by replacing its unnatural serialization rules with natural ones. Bartsch' s principles also predict that discontinuous constituents are disfavored. This explains Greenberg's universal 26:

"If a language has discontinuous affixes, it always has either prefixing or suffixing or both." 25 I believe the above remarks are the nucleus of an answer to my Question (k), "Why do languages with grammatically functional word order have the kinds of word order that they do?" The answer follows from three simple and immediately plausible principles of syntax: the disambiguation principle, the natural constituent structure principle, and the natural serialization principle. In answering this question, we were led to develop a theory of syntactic type change and to explain most of Greenberg's famous universals in the process. The answer to Question (k) also completes my answer to Question (b), "Why do syntactic rules work the way they do?" A syntactic rule is either a lexicalization rule or a serialization rule (or an intonation rule, about which I cannot say anything in this paper, primarily because I do not know anything about intonation). If it is a lexicalization rule, it will tend to abide by the argument-raising principle which assures that its effect at the surface is the creation of the universally preferred short and non-embedding F(x 1 , • • • ) structure; or, if it is a morphological rule (a special kind of lexicalization rule), it serves to symbolize semantic function-argument relationships on the syntactic surface. If the rule is a serialization rule, it will tend to abide by the disambiguation principle, with its two subprinciples, the natural constituent structure and natural serialization principles, in order to symbolize the semantic function-argument relations on the syntactic surface with a minimum of ambiguity. This then leads us to answer the initial question, (0, a), "Why do grammars contain syntactic rules?" Grammars convert semantic representations into phonetic representations and conversely. Semantic representations are logical forms and are as such hierarchical, unordered, and nonphysical. Phonetic representations are idealized representations of programs which enable speakers to pronounce sentences. As such they must be interpretable physically and must 25 Again I find the statement of this universal as an implication objectionable (cf. fn. 15). There is no causal relation between the existence of discontinuous and continuous affixes, but only one of frequency of representation in languages.

48

Theo Vennemann genannt Nierfeld

model the linear unidirectional flow of speech in time. It follows from this that each grammar needs at least these two kinds of rules: lexicalization rules which introduce elements that can be converted into physical signals, and serialization rules which arrange these elements in linear order interpretable as temporal order. The problem of a grammar is that it relates structures which are hierarchical (semantic structures) to structures which are nonhierarchical (phonetic representations). A grammar must therefore provide devices which make the function-argument relationships of the hierarchical structure retrievable from the nonhierarchical linear surface string. These devices are threefold: syntactic categories which can stand in function-argument relationships to each other in surface syntactic representations; closeness of constituents in the surface syntactic representation; and morphology and unidirectional serialization of function-argument constituents of any given type. The principles of syntax discussed in this paper guarantee the effective implementation of these devices. Human beings are born with the ability to develop in time a system of cognitive capacities whose linguistically relevant aspects are adequately modeled in the syntax of a truth-functionally interpretable logic. In short, the ability to form semantic representations is innate. The ability to analyze speech sound signals phonetically and to expect them (or the phonetic representations idealized from them) to be associated with semantic representations is likewise innate. Since there is an unlimited number of semantic representations and phonetic representations, the association cannot be established by memorization but must follow rules. Human beings are therefore endowed with the ability to discover these rules of association, i.e., the rules of grammar. They must therefore be born with the principles by which grammar rules are constructed. The principles of syntax discussed in this paper are chief among the innate principles by which syntactic rues are discovered by language learners and constantly created anew (cf. Bartsch & Vennemann, 1972). Explaining syntax is therefore nothing but discovering and applying these principles, by systematically comparing semantic representations and phonetic representations (or surface syntactic representations by way of shortcut) that belong together in discourse. This is what I have done in this paper. I have solved many problems of syntax which had not only been unsolved but had not even been seen to be problems in the past. I trust that by developing further the theory of grammar which has made these solutions possible, we will be led to numerous further and deeper explanations in syntax.

Explanation in Syntax

49

Note Added in Proof: After proofreading this paper, I feel that a few remarks and references should be added in the interest of clarity. (1) The terms "operator" and "operand" are taken from Bartsch (1972a) and are used in accordance with Bartsch and Vennemann (1972), without explicit definition in this paper. Simon Dik (Amsterdam) has pointed out to me that the impression may arise that their definition is based on the word-order universals so that the whole procedure of explaining the universals would be circular. There is no such circularity in the theory. The definition of "operator" and "operand" in natural generative grammar is, semantically, that in a constituent structure AB the specified element is the operand and the specifying element the operator and, syntactically, that the element, which determines the syntactic category of AB is the operand, the other the operator. A more concise presentation of the theory of word-order change, based on this definition, may be found in my paper "Analogy in Generative Grammar: The Origin of Word Order," forthcoming in the proceedings of the 1972 International Congress of Linguists, Bologna. (2) That the verb shift from SOY to SVO when subject and object inflection weakens is a response to a resulting topicalization problem is apparent once again from the intermediate TVX (topic-verb-remainder-of-sentence) stage, as in Old English and in German; cf. Vennemann (in preparation), which will appear in the proceedings of the First International Congress on Historical Linguistics, September 1973, Edinburgh. (3) That the verb shift depends only on subject and object case marking and not on the other cases (note Finnish) and begins very early in the case weakening process (note Old English) is brought out more clearly in my paper "An Explanation of Drift," where it is also shown that the drift phenomena discussed in E. Sapir's Language (1921) and R. Lakoff' s "Another Look at Drift" (in Stockwell & Macaulay, 1972) are automatic consequences of my theory of word order change. (4) The study of the topic-comment relation (cf. footnote 12) and of intonation (sentence accent) in natural generative grammar is begun in my papers "Warum gibt es Syntax? Syntaktische Konstruktionsprinzipien in logischer und psychologischer Sicht," forthcoming in Zeitschrift fur Germanistische Linguistik, 1973, 1, No. 2, and "Topics, Sentence Accent, Ellipsis: A Proposal for Their Formal Treatment," forthcoming in the proceedings of the Conference on Formal Semantics of Natural Language, April 1973, Cambridge (England).

REFERENCES Bartsch, R. (1972a) Adverbialsemantik: Die Konstitution logisch-semantischer Repriisentationen von Adverbialkonstruktionen. Frankfurt am Main: Athenaum Verlag, 1972. Bartsch, R. (1972b) The semantics and syntax of number and numbers. This volume. Bartsch, R., and Vennemann, T. Semantic structures: A study in the relation between semantics and syntax. Frankfurt am Main: Athenaum Verlag, 1972. Bever, T. G., and Langendoen, D. T. The interaction of speech perception and grammatical structure in the evolution of language. In R. P. Stockwell & R. K. S. Macaulay (Eds.), Linguistic change and generative theory. Bloomington: Indiana University Press, 1972. Bierwisch, M. Grammatik des deutschen Verbs. (Studia Grammatica, 2.) Berlin: Akademie-Verlag, 1963. Bierwisch, M., & Heidolph, K. E. (Eds.) Progress in linguistics. The Hague: Mouton, 1970.

50

Theo Vennemann genannt Nierfeld

Chafe, W. L. Meaning and the structure of language. Chicago: Chicago University Press, 1970. Chomsky, N. Deep structure, surface structure, and semantic interpretation. In D. D. Steinberg & L. A. Jakobovits (Eds.), Semantics: An interdisciplinary reader in philosophy, linguistics and psychology. Cambridge, England: Cambridge University Press, 1971. Davidson, D., & Harman, G. (Eds.) Semantics of natural language. Dordrecht, Holland: D. Reidel, 1972. Fillmore, C. J. Entailment rules in a semantic theory. In J. F. Rosenberg & C. Travis (Eds.), Readings in the philosophy of language. Englewood Cliffs, N. J.: PrenticeHall, 1971. Giv6n, T. Historical syntax and synchronic morphology: An archaeologist's field trip. Papers from the Seventh Regional Meeting, Chicago Linguistic Society, 1971, 394415. Greenberg, J. H. Some universals of grammar with particular reference to the order of meaningful elements. In J. H. Greenberg (Ed.), Universals of language. (2nd ed.) Cambridge, Massachusetts: M.I.T. Press, 1966. Katz, J. J. Mentalism in linguistics. Language, 1964, 40, 124-37. Kiparsky, P., & Kiparsky, C. Fact. In M. Bierwisch, & K. E. Heidolph (Eds.), Progress in Linguistics. The Hague: Mouton, 1970. Lakoff, G. Linguistics and natural logic. In D. Davidson, & G. Harman (Eds.), Semantics of Natural Language. Dordrecht, Holland: D. Reidel, 1972. Lehmann, W. P. Proto-Germanic syntax. Unpublished manuscript, University of Texas, 1970. Lehmann, W. P. On the rise ofSOVpatterns in New High German. In K. G. Schweisthal (Ed.), Grammatik, Kybernetik, Kommunikation: Festschrift fur Alfred Hoppe. Bonn: Dtimmler, 1971. Lehmann, W. P. Converging theories in linguistics. Language, 1972, 48, 266-75. Meillet, A. Introduction a l'etude comparative des langues indo-europeennes. (8th ed.) University, Alabama: Alabama University Press, 1964. (Originally published in 1937.) Reichenbach, H. Elements of symbolic logic. New York: Free Press, 1966. (Originally published in 1947.) Rosenberg, J. F., & Travis, C. (Eds.), Readings in the philosophy of language. Englewood Cliffs, N. J.: Prentice-Hall, 1971. Ross, J. R. Gapping and the order of constituents. In M. Bierwisch, & K. E. Heidolph (Eds.), Progress in linguistics. The Hague: Mouton, 1970. Staal, J. F. Word order in Sanskrit and universal grammar. (Foundations of Language, Supplementary Series, 5.) Dordrecht, Holland: D. Reidel, 1967. Steinberg, D. D., & Jakobovits, L. A. (Eds.) Semantics: An interdisciplinary reader in philosophy, linguistics and psychology. Cambridge, England: Cambridge University Press, 1971. Stockwell, R. P., & Macaulay, R. K. S. (Eds.) Linguistic change and generative theory:

Essays from the UCLA conference on historical linguistics in the perspective of transformational theory, February 1969. Bloomington: Indiana University Press, 1972. Vennemann, T. (1971) Rule inversion. Lingua, 1972, 29, in press. Vennemann, T. Topics, subjects, and word order: From SOV to SVO via TVX, in preparation.

THE SEMANTICS AND SYNTAX OF NUMBER AND NUMBERS RENATE BARTSCH Freie Universitat Berlin and University of California, Los Ange/es

In this article I will discuss plural formation and the application of numerals. 1 I will first give the semantics of plural nouns and the application of numerals in extensional logic, extended by set theory, and relate the semantics to surface syntax including morphophonology. Then I will extend the treatment to plural terms (plural noun phrases). This program is carried out in the framework of the theory of natural generative grammar (NG) which is outlined in Vennemann, 1971 and, subsequently-for semantics and syntax-in Bartsch and Vennemann (1972). I will also carry out the same program starting from surface syntax which will be related to semantics. This will be done in the framework of a Montague grammar. Finally I will indicate why I think that the two approaches to the problem may be notational variants of each other. THE SEMANTICS AND SYNTAX OF NUMBERS AND NUMBER IN NATURAL GENERATIVE GRAMMAR

The Generation of Plural Nouns Before I describe the semantics of plural formation I want to emphasize the distinction between "noun" and "noun phrase" (or 1 I am grateful to Professor Theo Vennemann for various suggestions which led to an improvement in this article.

51

52

Renate Bartsch

"term"). We must be aware of this distinction not only in the singular, but also in the plural; and we must distinguish systematically between "plural noun" and "plural noun phrase" (or "plural term"). Singular nouns and plural nouns have no determiner, whereas a noun phrase (term) has a definite or indefinite determiner if it is not a pronoun. (Also, proper names generally have no determiner.) In English and German the indefinite article applied to plural nouns and mass nouns is morphologically zero (fj). THE SEMANTICS OF PLURAL NOUNS

In this section I will use the usual set theoretic notation with which I assume the reader to be familiar. I will use a prime (') to indicate that not the word a but its corresponding translation into logic, namely a' is meant. (This is done following Montague, 1970.) Thus, house' is the predicate corresponding to the noun house. It is to be interpreted as a set, namely the set of all houses, which is the extension of the sense (or concept or intension) Ahouse'. The symbol is the sense operator which, applied to an extension, yields the corresponding concept. The concept corresponding to an extension is a function by which, when one applies it to a realm of entities, one discriminates those entities to which the concept applies from those to which it does not apply. Thus it identifies the extension corresponding to it. A property is a sense (or concept or intension) of a predicate. Since a predicate F designates an extension, F(x) means x E F (x is an element of F). For F(x) we could also say "x has the property AF." I will use x, y, z for individual variables and X, Y, Z for variables over sets (set variables). X C Y means X is included in Y, i.e., for all elements of X it is also true that they are elements ofY: (Vx)(x E X - x E Y). X n Y means the intersection of X and Y, i.e., the set of all elements which are in X and in Y. X U Y means the union of X and Y, i.e., the set of all elements which are in X or in Y (or in both). ex means the power set of X, i.e., the set of all subsets of X. A is the predicate abstractor. It is an operator which, applied to a variable and a sentence open in that variable, yields a predicate. For example, AX house'(x) is the set of all individuals which are houses. Thus AX house'(x) = house'= {x: house'(x)}. The extension of the singular noun house in extensional logic is house', the set of all houses. For the extension of the plural noun houses in extensional logic, houses', I assume the set of all subsets of the extension house' that have more than one member. Thus, just as the singular noun house can be predicated of any house, i.e., of any member of its extension, so can the plural noun houses be prediA

The Semantics and Syntax of Number and Numbers

53

cated of any set of houses which has more than one member, i.e., it can be predicated of any member of its extension. Therefore the extension of a plural noun, for example houses, is in extensional logic:

(1)

houses' = >-..X(X C >-..x house'(x) & f';!(X) > 1).

For the predicate abstractor A we will also use an arc (r,) on top of the variable abstracted from. (l')

= X(X ("'\

C

("'\

x house'(x) &f';!(X) > 1).

f';! is the power functor, designating a function which maps a set on its cardinality ("power" or German Miichtigkeit). The cardinality of a set is the set of all sets which can be mapped onto each other by a 1-1 mapping, i.e., the set of all sets with the same cardinality. Instead of (l') we can write: (1")

("'\

= X(X c house'

& f';!(X) > 1).

Instead of the inclusion relation between sets we can use the element-of relation. Instead of saying "X is included in the set house'," we can then say "X is an element of the power set of house'." Let c be the operator forming the power set of a set (i.e., the set of all subsets of that set). We then can write: (l" ') or

("'\

= X(X E chouse' & f';!(X) > 1) = X(chouse'(X) &f';!(X) > 1). ("'\

I now want to convert the given semantic representation into its phonetic representation. I will use English for the purpose of illustration, but analogous procedures apply in other languages. THE SYNTAX OF PLURAL NOUNS

(a)

("'\

X(chouse' (X) & f';!(X) > 1).

To this semantic representation in extensional logic (including set theory) we apply a constituent building or categorization rule which transforms this representation into an operator-operand hierarchy which is a complex predicate. This complex predicate is treated in extensional logic as a unit, like a primitive predicate. But internally the complex predicate is structured as an operator-operand hierarchy in which an operation is performed on a concept. This inner structure of the complex predicate is built in accordance with the rules of intensional logic. The syntactic rule which converts the semantic representation into the representation with complex predicates is a constituent-building or categorization rule because it introduces syn-

54

Renate Bartsch

tactic categories together with the new logical structure. Furthermore, together with the formation of the operator-operand hierarchies and the introduction of syntactic categories-especially lexical categories -the constituents which are indicated by the bracketing of the operator-operand hierarchies are labeled with symbols of categories according to the rules of categorial grammar. We now apply constituent building rule (CRl) to (a) and derive (b). (CRI)

Let a' represent any predicate over individuals. (Nsg symbolizes the category of singular nouns, and Npl symbolizes the category of plural nouns.) Then X(ca'(X) & ftf(X)

> 1) ~ X((Pl( aJsg)NP1)(X))

(The constituents inside the parentheses are not linearly ordered.) (b)

X ((Pl( house~sg)Np 1)(X) ).

Since the categories of the derived constituents are determined by the rules of categorial grammar from the lexical categories and the operator-operand hierarchies, i.e., the constituent structure, we can dispense with labeling the parentheses of the derived constituents with category symbols. The labels are redundant information if we know the lexical categories and the operator-operand constituent structure. But I will label the derived constituents also, because I do not state the rules of categorial syntax here. 2 I will now employ a lexicalization rule which introduces the phonological representation for the concept in its syntactic category. By introducing the phonological representation the sense operator ~ is extinguished. We obtain (c) from (b). 3 2 Note that in Bartsch and Vennemann (1972) the constituent-building rules were viewed as a certain type of lexicalization rule because there was no distinction made between introducing the full lexical material (including the phonological representation) and the introduction of the lexical categories in which concepts can be used. It seems more appropriate to introduce syntactic, and especially lexical categories, separately from the phonological representation of the lexical items. Since C-rules introduce lexical categories especially, they may be taken as a special type of lexicalization rule. Cf. also fn. 3.

3 Vennemann (personal communication) has pointed out that if one does not believe that deletion of complete lexical material (including phonological properties) is a linguistic process-I believe that it is not-one must distinguish an intermediate step which introduces all grammatical properties of lexical items but not their phonological properties. The evidence Vennemann suggests for this is the relevance of grammatical (as opposed to semantic) gender in pronominalization. German example: Peter nahm den Becher (die Tasse, das Glas) und zerbrach ihn(sie, es) 'Peter took the mug (the cup, the glass) and broke it.'

The Semantics and Syntax of Number and Numbers

(LRl) (c)

(Pl( CX~sg)NPI) A

55

(Pl(/cx/ Nsg)Np1)'

X(Pl(/haus/Nsg)NPl)'(X)[=(Pl(/haus/Nsg)Npl)'].

I will also write (c'), in case I use an orthographic rather than a phonological representation of the lexical items. (c')

(Pl(houseNsg)NP1)'.

To this representation another type of lexicalization rule, namely a morphological rule applies. The morphological rule (MRI) introduces the plural marker. 4

(MRI) From (c) we derive (d) by (MRI). (d)

(Zp1(/haus/Nsg)Np1)'.

Then (d), which still is structured hierarchically, will be transformed into a linear sequence. This is done by serialization rule (SRI). 5

(SRI) (The brackets indicate that the constituents inside are linearly ordered.) Linearizing (d) according to (SRI) will yield (e). (e)

[(/haus/ Nsg)(/ z/ p1)Np1] '.

To (e) the phonological rule (PRI) applies to yield (f). 6 4 Certain irregular nouns (e.g., mouse) must be exempted from this rule by exception markers. 5 It is an empirical question as to whether lexicalization and serialization should be collapsed into single rules in the case of affixation. Evidence from linguistic change and speech errors seems to suggest that affixation is a unitary process (Vennemann, personal communication). In this case (MRI) and (SRI) would be collapsed into the morphological rule (ARI).

(ARI) 6 I owe this analysis of regular pluralization in English to Vennemann (lectures at UCLA, Fall 1971). Its virtue is that it treats a-epenthesis in plural and past formation as a single process. Vennemann points out, however, that it is only available if one permits extrinsic rule ordering in grammars. Since this device is not permitted in natural generative grammar, the proper analysis is with separate sets of rules, e.g., for plurals:

Pl..:_ {a;/[+sibilant] --} The output of this rule (and the past tense rule) is subject to an additional voice assimilation rule.

56

Renate Bartsch

0-

(PRl) (f)

8/[~~t~[bilant]--[~~t~[hilant]. +coronal +coronal /haus8z/NPI'•

Since the ls/between /au/ and /8/ gets voiced by a morphophonemic process, we finally have (f '). hauz8z

(f ')

(The underlining here indicates that this is the phonetic representation.) After the last morphophonemic process all nonphonetic information is removed. In the following sections I will pay no further attention to phonology and only derive the orthographic representation of plural nouns from structures like (c'). I now want to describe the semantics and syntax of adjective application and of numeral application especially. THE APPLICATION OF NUMERALS

Certain adjectives can be applied to plural nouns. The semantic representation of plural nouns given above shows that these adjectives are adjectives which express predicates over sets. Such adjectives are the plural relative adjectives many and few, and the numerals one, two, three, etc. One is not a plural adjective, but it is a predicate over certain sets, namely those with one element only. Examples: few houses'= X(chouse'(X) &ftf(X) > 1 &ft/(X) < Np,y),

with NP,Y as a norm or average of cardinality relative to a reference set Y of sets. 7 two houses'= X(chouse'(X) & ftf(X)

= 2).

The last formula, generalized, is also the semantic representation for the dual which is used in several languages. The dual is derived in a manner parallel to the plural, with rule (CR2) and a morphophonemic rule introducing the dual marker. (CR2)

~

~

~

X(ca'(X) &ftf(X) = 2) "-" X((Dual( a~sg)Nd1)(X).

The trial, which is employed by some languages, has the semantic representation [for three houses.] X(chouse' (X) & ftf (X) 7

= 3).

Cf. the analysis of relative adjectives in Bartsch and Vennemann (1972).

The Semantics and Syntax of Number and Numbers

57

Adjectives like red, big, built in 1900, etc. are not plural adjectives but are, rather, adjectives which are basically singular. They are expressions of properties of single objects. The application of a singular adjective to a basic or complex singular noun yields a new complex singular noun to which the plural operator can then be applied. That means, if singular adjectives occur with plural nouns, they are not applied to the plural noun but to the singular noun before the plural operator is applied. Note that in German the plural marker is applied to the adjective and to the noun, while in English it is applied to the noun only; and there are languages in which the plural marker for the complex noun is applied to the adjective only, or in which there is no plural marker if the adjective is a numeral (as in Ngizim, a Chad language). 8 Before explicating the grammar of numerical adjectives I briefly want to show the derivation of red houses. (2)

r,

house'(x) & red'(x)} &ftf(X) > 1).

X(X C {x:

By (CR3) we get (2'). 9 (CR3)

x'(a' (x) &

/3' (x))

~ x'(/3~sg( a~sg)Nsg)(x)),

with Asg symbolizing the category of singular adjectives. (2')

X(X C {x:

(red~sg( house~sg)Nsg)(x)} & ftf (X) > 1).

i.e., X(c(red~sg( house~sg)Nsg)(X) & ftf(X) > 1). By application of (CRl) we get (2"). (2") Lexicalization applied to (2") yields (2' ") 8 This information was given by Russell Schuh (UCLA) in personal communication. Cf. further Greenberg's universal 40 in Greenberg, 1966:

When the adjective follows the noun, the adjective expresses all the inflectional categories of the noun. In such cases the noun may lack overt expression of one or all of these categories. For example, in Basque, where the adjective follows the noun, the last member of the noun phrase contains overt expressions of the categories of case and number and it alone has them [p. 95]. 9 Which of the two concepts is lexicalized as a noun and which as an adjective depends in part on presuppositions, in part on the structure of lexicons.

58

Renate Bartsch

Finally we employ the serialization rule (SR2) and then (SRl). 10 (SR2) ([//3/ Al] ([/a/Nsg])Nsg) ~

[[//3/ Asg][/a/Nsg]Nsg] if /3 itself is serialized in such a way that its last lexical item is an adjective. [[/a/ Nsg] [/ /3/ Asg]Nsg] otherwise.

(Note that instead of the phonological representation I have given the orthographical representation.) Numerical operators can be applied to the result of the application of the plural operator on nouns. I represent this by indicating the hierarchy of constituent building in: (/3~p1( (Pl( a~sg)NPl)NPt), A

A

(Apl symbolizes the category of plural adjectives to which plural numerals, i.e., two, three, ... belong.) The application of numerical operators is described as an intersection function operating on sets of sets. In two houses', houses', i.e., the set of all subsets of house' (with more than one element) is intersected with two', i.e., the set of all sets with two elements.

(3)

two houses'= X(chouse'(X) & ff(X) > 1 & ff(X) = 2).

By rule (CRl) this amounts to (3'). (3')

r-.

X((Pl(house~sg)NPI)(X) & ff(X)

= 2).

This means: the extension of the predicate two houses' is the set of all sets of two houses, i.e., the set of all sets of houses which have the cardinality 2.

(4)

r,

one house'= X(chouse'(X) &ff(X) = 1).

The difference between the semantic representations of one man and a man is the reason for the different syntactic properties of the two phrases. Thus we can say Only one man entered (but not more) and The one man who entered ... , but not ~only a man entered 10 Note that my serialization rule for the category A (adjectivals) is formulated in such a way that the order AN is the special case, the order NA is the general case. This formulation - as generally the separation of special cases and general case in rules (Vennemann, lectures at UCLA, 1971-72)-predicts that the pattern AN will be lost in time in favor of a generalized NA pattern which is indeed the regular construction for SVO languages. This ongoing change of AN into NA in English is a consequence of the verb's changing its position from the final to the preobject position (Lehmann, 1970; Vennemann, 1972).

The Semantics and Syntax of Number and Numbers

59

(but not more) and "'The a man who entered. . Despite the difference in their semantic representations A man entered and One man entered can be shown to be equivalent, i.e., the semantic representation of the first sentence implies the second and vice versa. If One man entered is interpreted as meaning that only one man entered, then the second sentence implies the first, but not vice versa. 11 The following paragraphs are concerned with determining what kind of syntactic rule is needed to form complex predicates like two houses' from (3'). But first a few remarks about the semantics of numerically quantifying expressions are in order. The meanings of numerically quantifying expressions and quantifying relative adjectives (few and many) do not characterize individuals but sets. Numbers themselves are sets of sets. Among these sets of sets an ordering relation obtains. Numbers are also referred to by terms to which two-place measure comparative predicates are applicable, expressed by the comparative phrases equal, more, less. (Note that these two-place predicates are one type higher than numbers, i.e., they are sets of pairs of sets of sets.) The application of a two-place predicate to one term yields a one-place predicate, e.g., =2 (equal to 2), > 2 (more than two), < 2 (less than two). These oneplace predicates are predicated of ft!(X), where Xis a set. For example, (5)

The cardinality of the set X1 is 2.

(5')

ft!(X1) = 2.

Cardinal numbers are sets of sets of individuals, e.g., 2 is the set of all sets with two elements. This set of sets has a certain property defined by its role in the ordering relation holding between these sets of sets, which is explicated in the axiomatic system for natural numbers. ft! maps a set X on the equivalence class relative to cardinality of which X is a member. Thus we can also write X E 2 or 2(X). If we would read this as X is 2 the is of this expression cannot be understood as the is expressing identity. Rather it is to be interpreted as the is of predication. Of course, this way to read 2(X) is inappropriate, it has to be read rather as X has two elements or X is two in number. (5") The following tree diagrams illustrate (5') and (5"). 11 The reader may easily convince himself that the sentences One man entered and A man entered are equivalent:

One man entered: A man entered:

(3X)(cman'(X) & f';f(X) = 1 & (\;/x)(x E X----> entered'(x)) (3x)(man'(x) & entered'(x)).

60

Renate Bartsch

s

(5'a)

~ Arg Arg

Pred

I

I

I

C1

=

A

2

(5"a)

S

~Arg I I 2 X

Pred

1

Arg

Op

I

I

J1j!

X1

The denotation of the constant C 1 is the measure value in cardinality of the set denoted by the constant X 1 . After a brief excursus on the principle of argument raising, formulated by Vennemann as a principle of syntax by which semantic structures are related with their surface representations, I will show that structures like (5') are convertible to the same surface structures as structures like (5"). EXCURSUS ON THE PRINCIPLE OF ARGUMENT RAISING

In Bartsch and Vennemann (1972) and also in Bartsch (1972) it is shown that structures like (a) and subsequently (b) are lexicalized in structures like (c) or (d).

S

(a)

(b)

~

Pred

Arg

>

c'

I

I

Arg

1:11\

(CR4) a'(x;y')

I

-------

s

~

Arg

by (CR4)

Nn,Y

I

C

Ml\

fn ~

(an;(~y~h;)(x).

T occurring alone symbolizes the category of terms; and! is a metasymbol referring to the categories intransitive adjective (IA), intransitive verb (IV), and intransitive noun (IN), for which I use simply the symbols A, V, N. Tl is a metasymbol referring to the categories transitive adjective (TA), transitive verb (TV), and transitive noun (TN).

X

61

The Semantics and Syntax of Number and Numbers

Rule (CR4) says that a two-place predicate applied on one term yields an intransitive predicate. By a syntactic rule following Vennemann's principle of argument raising, 12 (b) is converted into (c), and, provided xis a member of the reference set Y, (c) is converted into (d). (c)

~Arg

Pred

~Arg

Pred

l

I

Comp d Nv, y

s

(d)

S

I

I

Pos d

X

X

D represents the dimension (e.g., tallness) in which the measure function JM is applied to x. The relative adjective corresponding to the dimension D is d. N D,Y represents the norm or average relative to the dimension D and the reference set Y. (For example, Ntallness',men' is the norm or average of tallness relative to men.) Pas dis the positive form and Camp d is the comparative form of d. I will now demonstrate how tall man is derived in accordance with Vennemann's principle. (6)

x is a tall man

(6a)

f¥((rel x) man'(x)) > Nr,men'

(6')

(tall~ct( >TA(Nr,men' T)A)A)((rel x)man'(x))

(6a)

S

(6')

~ Arg Arg

Pred

I

~Arg I I

Pred

I

A, c'

S

NT men'

tall (>(NT men')) '

x'

~

Op

Arg

Op

Arg

IT

x'

(rel x)

Sx

I

I

A

I

Op

Arg

(relx)

Sx

I

I

A

Pred

I

I

/"-.Arg

man'

Arg

I

x

Pred

I

man'

I

x

12 This rule is given in Bartsch and Vennemann (1972) and in Bartsch (1972). The principle is discussed in further applications in Bartsch and Vennemann (1972).

62

Renate Bartsch

(6a) is converted into (6') by (CR4) and the syntactic rule for relative adjectives and comparison stated in Bartsch (1972), which follows the principle of argument raising. The operator rel is used to form the phrase x such Sx. For "rel(x,Sx)'' I write "(rel x)S/'. 13 The operator is defined by the following definition in use.

Definition:

G(x) & F(x).

F((rel x)G(x)) -

Note that (rel x) is different from the iota- and eta-operator in that it does not carry any existential commitment. The operator rel is interpreted as an operation which narrows down the domain over which the variable x' ranges. In case a constant x 1 is to be determined by Sx, the eta-operator or the iota-operator may be used instead of rel. Both can also be applied after the application of rel. Thus we have

= (ix)(G(x) & F(x)) ('ryx)F((rel x)G(x) = (71x)(G(x) & F(x)) (tx)F((rel x)G(x))

It is also possible to narrow down the realm of (rel x)G(x) further by applying rel a second time: (rel y)(F(y) & y = (rel x)G(x)) = (rel x)(G(x) & F(x)). From (6') we derive (6") by (CR3), and from (6") we can get (6 111 ) by a lexicalization rule (LR). (6")

((tall~ct( >TA(Nr,men'T)A)A)( man~)N)(x)

(6"')

(tall~( man~)N)(x)

s

s

r-----__

Pred

I

((tall'(> (NT y))) (man') '

Arg

I

X

(LR)

~Arg

Pred

I

tall '(man')

I

X

Example (6") is lexicalized as man, taller than the norm of tallness relative to Y, or as man, tall relative to Y. Since in this case Y = men', the part NT,Y is not lexicalized and we obtain instead of man, tall relative to men simply (tall A( man~)N), serialized as tall man'. After this excursus we return to the derivation of two houses. 13 The two-place operator rel applied to its first argument x yields the one place operator (rel x) applied to Sx, i.e., the sentence S open in x.

63

The Semantics and Syntax of Number and Numbers THE SYNTACTIC DERIVATION OF NUMERALS AND PLURAL ADJECTIVES

Vennemann' s principle, recapitulated for the example tall man, applies equally for the example with plural nouns and numerals.

(7)

[=(3')]

Combining the two-place predicate"=" with its second argument by (CR4) and raising X' by the syntactic rule for relative adjectives and comparison following Vennemann's principle we arrive at (7'). 14 (7') (in number~dum ( (=TApI( 2Tp1) API)) Ap1)( (rel X)(Pl( houseNsg)Np1)(XhP1),

S

(7)

~ Arg Arg

Pred

I

I

p (= (2))

X'

~Arg

Arg

Op

X'

(rel X)

I

I

JM D

I

2

~

~Arg

Pred

I

I

C'

Op

s

(7')

A

Op

I

I

I

Sx

A

Arg

Pred

Arg

Sx

Pl(''house')

X

I

(rel X)

I

I

Pr~Arg

I

Pl("house')

I

X

In (7') p corresponds to P as, generally, a relative adjective d corresponds to the dimension D. 15 We lexicalize pas in number or in cardinality (German: in Miichtigkeit or anzahlmiissig). For example, in This group of boys and this group of girls are equal in number (or in German Diese Manner sind sieben an der Zahl 'These men are seven in number'), the phrase in number (p) is a predicate limiting adverbial (Adum), It carries the information of P inf';!, namely that the general measure function fM is applied under the aspect of cardinality. Only 14

Cf. footnote 12.

15

Cf. footnote 12.

64

Renate Bartsch

by carrying the predicate limiting adverbial in number' does the structure (7') make sense: X' is a variable ranging over sets of individuals with the property predicated in Sx, namely being an element of the set of all subsets of the extension of house'. The predicate equal to 2' cannot be predicated of a set of individuals because equal to 2' is not a predicate of a set of individuals but of the cardinality of such a set (ft! (X')), i.e., a set of sets of individuals. Thus we cannot say X' is equal to 2 or This set of houses is equal to 2 (without violating the hierarchy of types). But the predicate 2 in number' can be predicated directly of a set of individuals. Thus we have X' is two in number or These houses are two in number. The variable X' refers to any set described by the predicate X(Pl( house~sg)N))(X), which is lexicalized by the plural noun houses. By further argument raising according to the definition of the rel-operator and an extension of rule (CR3) to plurals we obtain (7"). (7")

((p Adum ( (=TApl( 2Tsg)Asg))Asg)( (Pl( house~sg)Np1))Npl)(X)

s

~Arg

Pred

I

(P( = (2))) (Pl (house'))

I

X

By fully lexicalizing and serializing this structure we would obtain houses, equalling two in number or houses, equal to two in number. But the adjective (PAct ( (=TAp1(2Tp1)API))Ap1) normally is not fully lexicalized. It reduces to (2API) by lexicalization rule (LR2). (LR2) By applying serialization rules (SRI) and (SR2) we get two houses. Note that I apply (CR3) and (SR2) to singular nouns as well as to plural nouns. These rules are not sensitive to the special marker pl on Npl or Apl. They thus could be formulated by using a metacategorial symbol referring to sg as well as to pl. Using v as such a symbol and using g to refer to individual variables as well as to set variables we can write (CR3') and (SR2') instead of (CR3) and (SR2). (CR3')

65

The Semantics and Syntax of Number and Numbers

(SR2') 16

([/

!

/3/ Av]([/a/NJ )Nv) ~ [[//3/Av][/a/NJ] if /3 itself is serialized such that its last lexical item is an adjective. [[/a/ Nv][/ /3/ Av]] otherwise.

Note that besides basic plural adjectives like numerals (> 1) there are also derived plural adjectives which are applicable to plural nouns. For example, red in the red two houses. The semantics for plural adjectives is the same as for plural nouns: (8)

X(Pl( /3Asg)Apt)(X)

= X(c/3' (X)

& ft{(X) > 1).

By the syntactic rules (CRl) and (CR3') X(Pl( redAsg)Apt)(~ (Pl(house~sg)Np1))Np1)(X) is equivalent with X(chouse'(X) &ft!(~)> 1 & cred'(X)), and by the rules of set theory this is equivalent to X(X C {x: house'(x) & red'(x)} &JM(X) > 1), which is, by the syntactic rules (CR3) and (CRl) equivalent to X(Pl( (redAsg( house~ 8g)Nsg))Npt)(X). Thus, for derived plural adjectives a with a' = Pl( /3' ), we have a'((Pl(y'))) = Pl((/3'(y'))), where y is a singular noun, and /3 is an adjective which is basically singular. Substituting Pl( /3') for a' we derive (9).17 (9)

((Pl( f3Asg)API)( (Pl( '}'~sg)Npl))NPI) = (Pl( (/3Asg( Y~sg)Nsg))NPI).

with (9) we have derived a DISTRIBUTION LAW FOR THE PLURAL OPERATOR.

The Generation of Plural Terms (Noun Phrases) THE SEMANTICS OF PLURAL TERMS (NOUN PHRASES) First I want to give the semantic structures of the quantifiers all and some, and show that they turn out to be equivalent with the semantic structures of every and a (at least one). Consider example (10). (10)

Some houses are red.

16 Note that numerals in some languages are not serialized like other adjectives. For example, in Spanish, numerals are preposed but descriptive adjectives are postposed to the noun.

17 This equality explains Vennemann's (1972) observation that languages may treat plurality on Adjective+ Noun constructions morphologically as a "harmony": cf. English red rock+ s to German rot+ e Stein+ e.

66

Renate Bartsch

The plural noun /;!puses was analysed in sentence semantics in (l") by the predicate X(X C house' & f';!(X) > 1). Thus we have for (10): There is at least one subset of the set of all houses whose cardinality is greater than 1 and whose elements are red. This is represented in (10'). (10')

(3X)(X C house' &f';!(X) > 1 & ('r;/x)(x EX -

red'(x))).

For this we can write (10"). (10")

(3X)(chouse'(X) & cred'(X) &f';!(X) > 1).

Sentence (11) has the semantic representation (11'). (11)

There is a red house. At least one house is red.

(11')

(3x)(house' (x) & red' (x)).

(11') would be equivalent to (10') in case (10') contained f';!(X) ~ 1 instead of f¾(X) > 1. We would have (10"') instead of (10') as thesemantic representation of (10) if (10) were considered to be true, even if it were demonstrated that there is only one red house.

(10"')

(3X)(X C house' &f';!(X) ~ 1 & ('r;/x)(x E X -

red'(x))),

i.e., One or more houses are red.

Application of the definition of inclusion C and of the usual inference procedures shows that (11') and (10'") are logically equivalent. There is a distinction to be observed between some houses [s /\ m hauz;n] and some houses [slJl hauz~z]. [s /\ m] is used contrastively as in Some houses are red and some are not. The meaning of this sentence is intended in Some houses are red. In the expression some houses, some is not used contrastively. (10') and (10'") represent the noncontrastive use of some. Since in negation and questioning of the sentence Some houses are red, the part some houses are not red is neither negated nor questioned, it is a presupposition. Thus we have the semantic representation (12') for the contrastive use. (12)

(12')

Some houses are red. (3X)(X C house' &f';!(X) > 1 & ('r;/x)(x EX red'(x))) & ((3X)(X C house' &f';!(X) > 1 & ('r;/x)(x EX - red'(x)))).

The angular brackets indicate a presupposition. Propositions in angular brackets belong to the premises of a discourse and may there-

67

The Semantics and Syntax of Number and Numbers

fore neither be questioned nor negated. If they are, the discourse ends and a new one is begun with new premises. Now consider a sentence with an all-quantifier.

All houses are red.

(13)

Its semantic representation is (13'). (13')

(VX)(X c house' &ftf(X) ~ 1 - (Vx)(x EX -

red'(x))).

[In natural language it belongs to the presuppositions of this sentence, used as a statement, that there are houses ( (3X)(X C house' &ftf(X) ~ l))]. Compare (13) and (13') with (14) and (14').

(14)

Every house is red.

(14')

(Vx)(house'(x) -

red'(x)).

The equivalence of (13') and (14') is easily seen by using the definition of the inclusion C and the transitivity of the conditional besides other inference procedures. Thus (13) and (14) are equivalent. Next I will discuss the plural indefinite and definite articles. In English the indefinite plural article is 0; thus, the plural noun is not distinguished morphologically from the indefinite plural noun phrase (term). For example, men in three men is the plural noun men (which expresses a predicate over sets) while men in Men are entering the women's gym is the indefinite plural noun phrase men with a zero determiner, 0 men. The same distinction, of course, must also be observed for complex plural nouns and noun phrases like three men in Three men are entering the gym where three men is a noun phrase (term) and in The three men are entering the gym, where it is a plural noun. The plural noun phrase men in Men are clever is generic, more or less equivalent with all men. The plural noun phrase men in Men are entering the gym is to be interpreted like some men. In the generic interpretation there is no difference in the semantics of sentences (15a) and (15b) in extensional logic, though there is a difference in the conditions under which they are used in natural language. (15a,b,c) are logically equivalent, if boy' -# ¢. (15)

(15a,b') (15c')

a. Boys are clever. b. All boys are clever. c. The boys are clever (but not the girls). (VX)(X c boy' &ftf(X) > 1 - (Vx)(x E x-clever'(x)). (tX)(X = boy') C clever'

or

cclever'((tX)X

= boy').

68

Renate Bartsch

For sentence (16) we have the semantic representation (16'). (16)

Boys are entering the women's gym.

(16')

(3X)(X C boy' & f1)f (X) > 1 & (\:Jx)(x E X -

G(x)),

with G = enter the women's gym. Thus, in nongeneric use, the indefinite plural article is treated like the quantifier some and, it is treated like the quantifier all in generic use. Some is not applicable to plural nouns with numerals: '>Some three men, ,:, some ten men. In the event that several is applicable to plural nouns with numerals, it means 'more than two times the set designated by the complex noun.' Thus, several hundred men means 'more than two sets of a hundred men.' The use of several in several dozen eggs follows the same pattern. The semantic representation for (17) is (17'). (17) (17')

Several hundred men are entering the arena. (3.x)(.x C {X: X C man' &ft;!(X) = 100} &ft;!(.x) > 2 enter the arena'(x)))). (\:Jx)(x E X & (\:JX)(X E .I -

The semantic treatment of the definite plural article faces the same problems as that of the definite singular article. Since, in this article, I do not want to go into pragmatics and text theory, I will assume Russell's analysis of definite description to be the framework in which the definite plural article is treated. Therefore, the revisions, necessary from the point of view of pragmatics and text theory, apply equally to the definite singular and the definite plural article. (18)

Men are entering the hall. The men are blond.

(19)

Men and women are entering the hall. The three men are blond.

The semantic representation of The men are blond is (18'). (18') or:

(tX)(X

c

man' & G(X))

c

blond'

cblond'((iX)(cman'(X) & G(X)).

The semantic representation for The three men are blond is (19'). (19')

cblond'((iX)(cman'(X) &ft;!(X) = 3 & G(X))).

Following Russell's theory of description, the iota-operator is defined by a definition in use which gives us (18") instead of (18'). (18")

(3X}(Cman'(X) & G(X) & (\:JY}(Cman'(Y) & G(Y) Y = X) & cblond'(X)).

69

The Semantics and Syntax of Number and Numbers

Since the definite description in a sentence has to be presupposed in order for the sentence to be used as a statement, this analysis is not quite adequate for natural language. (3X)(cman' (X) & G(X) & ('v'Y)(cman'(Y) & G(Y) Y = X) has to be presupposed, as it is in (18). Thus we can represent (18) more adquately by (18"'). (18'") (3X)(cman'(X) & G(X)) & (('v'Y)(cman'(Y) & G(Y) Y = X1 )) & cblond'(X 1 ).

In (18'") Men are entering the house is asserted. Then the uniqueness condition and the identification of these men with X1 is presupposed. Furthermore, They are blond is asserted with reference to X 1 • THE SYNTAX OF PLURAL NOUN PHRASES (TERMS)

I now want to show how plural articles and quantifiers are derived in surface syntactic structure from semantic representations by certain devices of syntax. I will not give an account of the differences between these, those, and the because this would require a developed text theory and pragmatics. I begin by deriving sentence (20) from its semantic representation (20'). Some men are clever.

(20) (20')

(3X)(X C man' &ft;!(X) > 1 & ('v'x)(x E X -

clever'(x))).

Note that by the definition of inclusion ( C) the last conjunct ('v'x)(x E X -

clever'(x))

can be replaced by X C clever' or cclever'(X). Thus we have (20"). (20")

(3X)(X C man' &ft;!(X) > 1 & X C clever').

By (CRl) we obtain (20'").

(20"') or:

(3X)((Pl( man~sg)NP 1)(X) & X C clever') (3X)((Pl(man~sg)NPI)(X) & cclever'(X)).

The first new step is the derivation of (20"") from (20"') by which the existential operator 3, which is an operator on a variable and a sentence, is transformed into another category, namely into a determiner E which is an operator on a noun yielding a term (noun phrase). This is done by rule (CR5), a constituent building rule. (CR5) If J denotes the quantifier 3 or 'v' and g denotes any variable, then (J()(~~(g) & o'(()) ,.,_.,, o'(f Det(~~h).

(CR5)

is applied to (20"'). The result is (20"").

70 (20"")

Renate Bartsch

(E 0 e/Pl( manNsg)Np1hp1) C clever' cclever' ((Eoe/Pl( manNsg)NPihp1),

or

Here, E is the determiner corresponding to the variable-sentence operator 3. The determiner E applied to a noun yields a noun phrase (term). This term has as its referent the union of the sets which were referred to by the quantifier 3 together with the open sentence in its scope. The step from (20'") to (20"") has the following correspondence in semantics: (20"') says that there are some subsets X (=l=~) of man' which are also included in clever'. Formula (20"") is interpreted as saying that the union of all these subsets is included in clever', i.e.,

some men are clever:

U {X:

X C man' &X C clever'} C clever'.

Of course, thus interpreted, the application of rule (CR5) does not add any new information but only states the same state of affairs differently, as illustrated below. (20"')

clever'

man'

clever' man'

t=::::3-----1-the set which is the union of the subsets with Xe man' & Xe clever': U {X: Xe man' & Xe clever'}

The determiner E applied to a plural noun, e.g., men, singles out a specific member from the extension of the plural noun men satisfying (20"'). This member is a set designated by some men'. The determiner E applied to a singular noun, e.g., man, singles out a member of the extension of the singular noun. This member is an individual, designated by a man'. E applied to plural nouns is lexicalized as some. E applied to singular nouns is lexicalized as a. Thus we have the following lexicalization rule: (LR3)

71

The Semantics and Syntax of Number and Numbers

A further constituent building rule is needed, which is only an extension of (CRl). (CR6)

X C /3' & ft! (X) > 1

I'"'-"'!

(Pl( f3vsg)VP1)(X),

or, in another form: c13'(X) &ftf(X) > 1 ,,.._

(Pl(f3vsg)vp1)(X).

In our case, /3v is (be(clever~)v). (Pl(be(a~sg)y 8g)vp1)(X) is lexicalized and serialized as [X[are[aApilvp1]8 ]. Applying (CR6) and lexicalizing as well as serializing we finally derive (20). Some men are clever. Next I will derive sentence (21). (21) (21')

All men are clever. (VX)(X C man' &ftf(X) > 1 -

(Vx)(x E X -

(21")

(VX)(cman'(X) &ftf(X) > 1 -

(21"')

(VX)((Pl( man~ 8g)NP 1)(X) -

clever'(x)))

cclever'(X)) cclever'(X)).

Now we apply (CR5) and get (21""). (21" ")

eel ever' (Aoe1(Pl( man~sg)Npthp1).

The step from (21"') to (21"") has the following correspondence in semantics: (21"') says that all subsets X with X C man' are included in clever'. Formula (21"") says that the union of all these subsets is included in clever': U {X: X C man'} C clever'. This is illustrated below. (21'")

man' subsets with X

2 (

!""•-----=-:~::

c

man'

which is the union of all subsets of man': all men'

72

Renate Bartsch

The determiner A applied to the plural noun men singles out a specific member, the set designated by all men, of the extension of the plural noun. Applied to the singular noun man A singles out every member of the extension of the singular noun. The operator A, applied to a plural noun, is lexicalized as all; and A, applied to a singular noun, is lexicalized as every: (LR4)

(Aoet(a~sghsg) -

(everyoet(a~sghsg)

(Aoela~p1hp1) -

(alloe1(a~pthp1).

By applying (CR6) and further lexicalization and serialization we get (21). We have now derived the surface structures of (20) and (21) from their semantic structures. The main step of the derivation was the application of the constituent-building rule (CR5) by which we dispensed with the variable-sentence-operators 'r;/ and 3 and constructed a constituent structure with terms (noun phrases) which are not variables. These terms are built by application of the corresponding determiners A and E as operators to nouns. Rule (CR5) has a semantic correlate, namely the singling out of a certain subset which is the union set of those sets which are included in man' and in clever'. These union sets are the referents of the constructed noun phrases (terms). That (20 1111 ) refers to the union set is shown in the following discourse: Some men are clever. All of those will be successful. Here, those refers to the union set of subsets of man' which are included in the set of clever individuals. Thus, we perform different conceptual operations by interpreting Some men are clever on the one hand, and Some man is clever or A man (at least one man) is clever on the other hand. The same is true for All men are clever and Every man is clever. These different surface forms represent different ways of capturing the same state of affairs. They are derived from different, though equivalent, semantic representations by syntactic rules conforming to Vennemann' s principle of argument raising. This principle states that constituent building takes place in such a way that surface sentences are preferably of the form a(/3) with a as a predicate and term f3 as an argument. Rule (CR5) is simply the implementation of the principle for quantifiers. Note also that variables are eliminated by this rule; variables do not occur in surface syntax. The following figures illustrate the effect of the principle graphically. (a) 'tfx

s

S

~

Sx

~ (F (x), G (x))

->

-------

G

~ (everynet(FNsg )Tsg)

The Semantics and Syntax of Number and Numbers

(c)

s

S

~Sx

3x

73

~ (F(x),

~

G

~ (a Det (FNsg) Tsg)

&

G(x))

(d) 3X

s

~

&

s Sx

~

--------

0

G~(Pl(FNsg)Nplhp1)

(°F(X), 0

G (X))

Finally, I want to mention briefly that the conjoined nouns men and women (e.g., in all men and women or in some men and women) and men or wo~en (e.g., in some men or women) have the semantic representations X(X n man' #- 0 & X n woman' #- (/)) and X(X n man' #- 0 V X n woman' ¥c (/)), respectively. The semantic representations of the noun phrases all men and women, some men or women, etc. are built by the application of quantifiers to the representations of the conjoined nouns. The conjoined terms (conjoined noun phrases) all men and all women, some men or some women, etc. have semantic structures different from those of the terms built from conjoined nouns and quantifiers. These phrases are treated in detail later on in this article, in the section on the syntax and semantics of number in a Montague grammar.

The "Group" Interpretation of Plural Predication Now I will describe the semantic distinction between plural predication as applied to several singular individuals and as applied to

74

Renate Bartsch

groups. Sentence (22) has the semantic representation (22') or, equivalently, (22" ). 18 (22) (22') (22")

Three men enter. (Three men are entering.) (3X)(X C man' &ffl;!(X) = 3 & ('tfx)(x E X (3X){Cman'(X) &ffl;t(X)

=3&

enter'(x))).

center'(X)).

(22), as well as (22') and (22"), leaves it open whether the three men enter, each independently of the other, or whether they enter together. The difference between the interpretation of (22) as (22') or like (23) can be shown in semantics. (23)

Three men enter together. (Three men are entering together.)

For the semantic representation of (23) I will use a notation which I introduced in Bartsch, 1970 for the analysis of manner adverbials and used subsequently (Bartsch, 197la,b) for the semantic analysis of several classes of adverbials and for the semantics of verb extraposition and the extraposition of certain classes of adverbials. Let us assume a realm of phenomena which are processes (including action processes) or states. Let the variable r range over this realm. Let I represent the predicate constant be involved in' (or: be engaged in' or be in'). This two-place predicate constant takes individuals as its first argument and states or processes as its second argument: x is involved in the state or process r. -V is an operator which maps F, a predicate over individuals, on the corresponding predicate F - V, a predicate ranging over states or processes. For example: -V:

drive'

------ ➔

driving process'

-V:

F

------➔

F-V_

The following equivalence is an axiom, or definition in use, for the predicate constant I and the operation symbol -V. (A)

F(x) ¢:::::::? (3r)(I(x,r) & F - V(r)).

Now we can formulate the semantics of (23), where it is stated that three men are involved in one single process of entering. (23')

(3X)(X C man' &ffl;!(X)

=3

& (3r)('tfx)(x EX I(x,r) & enter' - V(r))).

18 Note that the semantics of plural formation of verbs and adjectives is the same as for nouns, namely, essentially the formation of the power set of the extension of the corresponding singular verb or adjective. In examples (22) and (23) and elsewhere I have disregarded the continuous form in English.

75

The Semantics and Syntax of Number and Numbers

For (22') there is a formulation using the same notation, (22"'). (22"')

(3X)(X C man' &ftf(X)

=3&

(Vx)(x E X (3r)(I(x,r) & enter' - V(r)))),

or equivalently: (22"")

(3X)(X C man' &ft!(X) = 3 & (Vx)(3r)(x EX I(x,r) & enter' - V(r))).

Here, (22"') is shown to be equivalent to (22') by replacing (3r)(I(x,r) & enter' - V(r)) by enter'(x) in accordance with (A). (22"") and therefore (22"') follow from (23') by the inference rules of logic, but (23') does not follow from (22""). This makes explicit the fact that (23) implies (22), but (22) does not imply (23). The order (3r)(Vx)Sr,x of quantifiers in (23') is reversed in (22""), where we have (Vx)(3r)Sr,x• In (23') there is a single process in which all members of X are involved; in (22"") for every member of X there is a process in which he is involved. (22"") leaves it open as to whether it is the same individual process for all members of X or not. This is just what the implication (3r)(Vx)Sr,x ===? (Vx)(3r)Sr,x expresses, which is the essential step in the inference from (23') to (22""). Similar to (22) and (23) are (24) and (25). (24)

John and Mary go to the movies.

(25)

John and Mary go to the movies together.

(24')

(3X)(X

= {John, Mary}

& (Vx)(x E X -

H(x))),

with H representing to go to the movies'. For (24') there is a more elaborate form, namely (24"). (24")

(3X)(X = {John, Mary} & (Vx)(3r)(x E X I(x,r) & H - V(r))).

(25')

(3X)(X

= {John, Mary}

& (3r)(Vx)(x E X I(x,r) & H - V(r))).

(24') and (24") are equivalent to (24"'). (24" I)

HUohn) & H(Mary).

But (25') is not equivalent to (24"'). (24") follows from (25'); hence (24') and (24"') as well follow from (25'). For the semantics of sentences with cooperational adverbials, like

76

Renate Bartsch

with Mary in john goes to the movies with Mary, see Bartsch, 19716. 19 There are predicates which are not applicable to singular terms but to plural terms only. One of them is the verb to gather. (26)

a.

b. c.

d. (26) or:

Men gather. (Men are gathering.) The men gather. (The men are gathering. ) The committee gather. (The committee are gathering.) The committee members gather. (The committee members are gathering.) a'.

b'.

(3X)(cman'(X) & gather'(X)), gather' (('riX)°man' (X)) gather' ((iX)(cman' (X) & G(X))).

(G serves to determine the subset X uniquely.) c'. gather' ((iX)(committee' (X)) d'. gather'((iX)(X = committee member')). Some of the verbs that cannot be predicated of referents of singular terms include in the set of meaning postulates describing the meaning of these verbs, a meaning postulate which states that the individuals involved in a process r, described by the verb nominalization F - V, are involved in one single state or process, e.g., (27). (27)

(□

□ (VX)(gather'(X)

-

(3r)(Vx)(x E X come to'(x,y) & I(x,r) & stay at y' - V(r))).

means: It holds necessarily that . . . )

19 Note that (a) John and his parents are walking has a semantic representation different from that of (b) John is walking with his parents. The semantic representation of a cooperational adverbial like with his parents is given in Bartsch (1971b). It is-disregarding tense and aspect:

(b')

(3r)(I(John',r) & walk' - V(r) & COOP(]ohn', John's parents', r))).

COOP(x,y,r) is a predicate constant lexicalized as x cooperates in r with y. In contrast to (b') we have for (a): (a')

(3r)(I(John',r) & [(John's parents', r) & walk' - V(r))).

Good evidence for having different semantic structures for (a) and (b) is given in Edmonson and Lindau (1972) from Swedish reflexivization. Swedish has nonreflexive possessive pronouns (hans) and reflexive possessive pronouns (sin). In sentence (a) the possessive pronoun in his parents is hans in Swedish. In sentence (b) the possessive pronoun in his parents is sina, the reflexive form, in Swedish. From this Edmonson and Lindau, who work in the framework of transformational grammar, argue for different deep structures for (a) and (b). In natural generative grammar the same phenomenon can be used as an argument for different semantic structures: reflexives

77

The Semantics and Syntax of Number and Numbers

In this meaning postulate, it is left open whether the individuals come together to the place y or whether everybody comes by himself. But it is stated that there is a state or process r in which all individuals are involved, namely staying at y. The predicate gather' is one of the meanings of the verb come together. The other meaning is that of coming together rather than everyone coming by himself. No special meaning postulate is needed for this sense of come together since this phrase is built following a general rule for cooperational adverbials and especially for the adverb together. The application of this general rule yields: □ (v'X)(come

together'(X) -

(3r)(v'x)(x E X -

I(x,r) & come' - V(r))).

Here, the process of coming to a place is shared by all individuals of X, rather than the state of staying at a place. Another example of a verb applicable only to plural terms or mass terms is stromen 'pour' (or 'stream') in German, as in (28). (28)

Menschen stromen in die Arena. 'People are pouring into the arena.'

Sentence (29) is ungrammatical. (29)

Da stromt schon wieder einer. 'There is another person pouring in.'

But (29a) is also ungrammatical because stromen is not applicable to small groups but rather to masses of individuals.

1 2 3

occur in simple sentences of the form R(x,y,z, . . . ) for those arguments in positions other than 1 if these arguments are coreferential with the argument under 1, or if they contain in the term identifying these arguments a term coreferential with the argument under 1. In languages with reflexive possessives like Swedish, we thus have a 1

2

3

reflexive in the surface form of the simple sentence COOPUohn', John's parents', r). Therefore Swedish has a reflexive possessive pronoun in (b) but not in (a). In (a) John and John's parents are not terms functioning as argument 1 and argument non-1, respectively, in a sentence of the form R(x,y,z, . . . ). Rather they are argument 1 in two different simple sentences, namely in I(John', r) and in IUohn's parents', r). The fact that in sentences with conjoined noun phrases like (c), John and his girl friend walk with his parents the possessive pronoun in with his parents is not reflexive in Swedish is explained by the fact that the conjoined term John and his girl friend, which functions as a whole to determine the first argument in COOP( . . . . ) is not coreferential with the term John', which functions in the argument 2 in the term John's parents'. The analogous situation exists in Latin which has a difference between reflexive possessive pronouns (suus) and nonreflexive possessive pronouns (eius).

78 (29a)

Renate Bartsch

"Da stromen schon wieder zwei. 'There are two other persons pouring in.'

[Sentences (29) and (29a) could be used ironically.] Besides verbs, there are also adjectives which are applicable only as predicates of plural terms or as operators on plural nouns. Adjectives of this type are plural numerals, which have been treated above. Furthermore, there are nouns which are basically plural, like school in school of fish, flock in flock of geese, herd, cattle, people, group, committee, government, etc. Note, that nouns, adjectives, and verbs that are basically plural are of the same type as sets of sets, like the set of all subsets of a set which is the extension of a singular noun, adjective, or verb. THE SYNTAX AND SEMANTICS OF NUMBERS AND NUMBER IN A MONTAGUE GRAMMAR

In natural generative grammar (NG) the generation of sentences starts from the generation of semantic representations in extensional logic extended by set theory and model-theoretic interpretations in possible worlds. Intensional logic, which is interpreted by model theory is used in the generation of semantic representations only as far as operations on senses of sentences are concerned, e.g., in the representation of propositional attitudes. On the semantic representations, which are generated by the formation rules of extensional logic and the part of intensional logic mentioned above, rules of syntax are applied: constituent-building rules (C-rules) are applied to yield the conceptual level of constituent structure. This level of constituent structure is similar to Montague's level of intensional logic in having the same function-argument hierarchies. But, unlike the level of intensional logic in a Montague grammar, the function-argument hierarchies have syntactic category and subcategory symbols attached, which subdivide the types of intensional logic. 20 This constituent structure, still hierarchically organized, is mapped into linear sequences by serialization rules, after morphophonemic rules and other lexicalization rules have applied. Compared to NG, a Montague grammar works just the opposite way. In the following section I will pursue this opposite way, deriving the semantic representation from the surface syntactic structure. 2° For further discussion of the relation between Montague Grammar and the semantics and syntax of natural generative grammar cf. Bartsch and Vennemann, 1972.

The Semantics and Syntax of Number and Numbers

79

I now want to remind the reader of some terminology used in Montague (1970). A singular noun a is translated into a predicate a' in intensional logic, namely the set of all individual concvepts x to which it applies. The corresponding intension (sense) is, a', which is a function by which one is able to single out from a set o( individual concepts those elements which belong to the extension of a', namely a'] =a'. The corresponding extension a'. in extensional logic is the set of individuals u with a'..(u). (The subscript asterisk indicates that the predicate a' is used in extensional logic.) The variable x refers to individual concepts and the variable u refers to individuals. Thus we have for the singular noun house the sense or concept ~ house' and its extension in intensional logic house'. Its extension in extensional logic is house'., i.e., the set of all individuals that are houses: 11.u house'.(u) or u house'.(u), or simply house'• .21 P and Q are variables over properties of individual concepts, and P and Q are variables over properties of properties of individual concepts. In Montague's articles plural formation and the application of numerals are not treated. Yet there are a few unpublished notes about plurals in which he gives hints about the semantics of plural formation.22 He writes in a note dated Oct. 22, 1969: V

(

Perhaps 'committee' and plural common nouns designate sets, and plural verbs designate sets of sets.

the committee gather the horses eat grass,

horses eat grass.

And in a note dated March 23, 1970: I still think plural nouns designate sets (at least if we dismiss considerations of intensional verbs) and plural verbs designate sets of sets, but I now think that such common nouns as 'committee' (again if we ignore intensional verbs and the like) designate a set. Thus we can say 'the committee gather' and 21 Note that, in part I, I dealt with extensions in extensional logic and set theory only, and thus used a', e.g., house', for the set of individuals which is the extension of a, house, in extensional logic. In Montague's notation this is house'•. I used x to refer to individuals. In Montague's notation u is used instead, while x is used as a variable over individual concepts. In the range of problems treated in this article the distinction between individual concepts and individuals is of no importance. Thus we may as well forget about the star notation and the difference between x and u. I only use x as a variable over individual concepts in this part II to make my treatment of numbers and plural formation in a Montague grammar look like an extension of Montague's (1970) fragment of English, in which he treats quantification in ordinary English. 22 I am grateful to Michael Bennett for showing to me these unpublished notes of Montague's.

80

Renate Bartsch 'horses gather'. Of course, we can also say 'the committee is numerous', but this involves a higher-order 'is' and a higher-order adj 'numerous'; notice, we can say 'numerous lion'. Notice that 'seek lions' appears syn with 'seek a lion.' Perhaps: lions= P[ · P'u[u • lion]]

I am not sure what this formula fo_r lions is supposed to mean, but it could make sense as: the set of all properties of properties which have as extensions subsets of the set of all lions. Before stating the grammar of plural formation I want to emphasize the distinction between "noun" and "noun phrase" (or "term"). In the case of plurals as well as with singulars, we must be aware of this distinction, and distinguish between "plural noun" and "plural noun phrase" (or "plural term"). Parallel to a singular noun, a plural noun has no article (determiner) applied to it; whereas a noun phrase (term) has a definite or indefinite determiner, in the event that it is not a pronoun or a proper name. In English and German the indefinite article applied on plural nouns is morphologically zero (@).

The Grammar of Plural Nouns Rules for the formation of derived singular nouns, i.e., complex nouns consisting of adjectives and nouns, were given in Bennett (1971) and in Bartsch (1972). I thus presuppose the existence of basic and derived singular nouns (category: N) and start with plural formation, thus deriving phrases of the category Np 1 and giving their semantics. The category of plural nouns also has basic elements.

Np1 :

school (of fish), flock (of geese), herd (of elephants), people, government, committee, group, etc.

I will now present the syntactic rule for the derivation of plural nouns and then give the translation rule which translates the plural nouns into intensional logic.

Syntactic rule (S:Pl): If a E PN (i.e., if a is a singular noun), then f 20 (a) E PN (i.e., then f 20 (a) is a plural noun), with

suppletive plural form of a if marked for irregularity 23 .. ( ) = { [a;:)z] if morphophonemic cond. I 1 20 a [as] if morphophonemic cond. 2 [az] if morphophonemic cond. 3 23 Note that the regular cases can be treated together if the process is split up into a morphological rule, which yields /az/, and a phonological rule, as in section I of this article.

81

The Semantics and Syntax of Number and Numbers

fzo(man) = men fzo(house) = houses fzo(cat) = cats fzo(dog) = dogs

Examples:

Translation rule (T:Pl): If a E PN translates into a' (of type: predicate over individual concepts) in intensional logic, thenf20 (a) E PNpt translates into Pl( a') (of type: predicate over sets of individual concepts).

Example: a.

houses'= PI(house').

Now I will give a meaning postulate which explicates how the operator Pl functions.

(MP:l)

For all X it necessarily holds that PI(house')(X) X C house' &f')'!(X) > 1 X C {x: house'(x)} &f']'!(X) > 1.

where x is a variable ranging over individual concepts, X is a variable ranging over sets of individual concepts, and f')'!(X) is the cardinality of the set X. If 0 house' means the power set of the set house', i.e., the set of all subsets of the extension house', we can write (MP: 1) in a slightly different fashion: D(v'X)[Pl(house')(X)- 0 house'(X) &f']'!(X) > l]. Before dealing with the application of adjectives, especially numerals, to plural nouns I want to e~plain formula (a). The plural operator Pl is applied to the intension house'. The plural operation maps the sense house' on the power set of its extension house'. This operation is a combination of two functions: first the application of to the intension house' to yield the extension house' and then the application of the power set formation operator c. At the least, the empty set, which is a subset of every set, has to be excluded as an element from the power set. I am not sure whether the sets with only one element ought to be excluded, as done in (MP 1), since there are sentences with plurals which are already true if they are satisfied by one element, e.g., There are lions is true if there is only one lion. Without exclusion of sets of one element the plural formation in intensional logic is: A

V

A

c house' - {f)} = chouse' - {@} = X[chouse' (X) & X ¥ f)] n

PI( house')=

VA

= X[X

C house' & X ¥

0].

82

Renate Bartsch

Since the function of the plural operator is definable only with reference to the extension of the concept it operates on, we could as well define the plural operator as operating on the extension directly. Thus we have

Pl*(house') = chouse' - {0} = X[chouse'(X) & X #- 0], and Pl= Pl*

0

V.

For systematic reasons we might stick with the first proposal in which Pl operates on the intension. The general formulation of (MP 1) is: □ (VX)[Pl( a')(X) -

(MP I)

X c a' &f';!(X) > l] ca'(X) &ff(X) > l]

or less restrictively: □ (VX)[Pl( a')(X) -

(MP I')

X C a' & X =Jc. 0] ca'(X) & X #- 0].

Now the rules for the application of numerals are introduced. The category Num is a subcategory of the category Ap 1 of plural adjectives.

If a E

Syntactic rule (S:Num):

then F6 (a, /3) = a/3 E

PNum

C PApl

=

PNPilNPI

and /3 E

PNPh

PNPI·

Example: F6 (two, houses)= two houses If a E PNum C PA translates into a' (of the type: operator on (senses of) predicates over sets of individual concepts), and if {3 E PNPI translates into /3 (of the type: predicate over sets of individual concepts) in intensional logic, then F6 (a, /3) E (of the type: predicate over sets of indiPNPI translates into a'( /3 vidual concepts).

Translation rule (T:Num):

1

1

)

Example: two houses'= two'(houses') = two'(Pl(house')). There is a meaning postulate to explicate what kind of operation the application of a numeral operator on a predicate is: (MP:Num) □ (VX)[a'( {3')(X)- /3'(X) & ff(X) = a']. Instead of this we can also write: □ (VX)[a'(/3')(X) -f3'(X) & a'(X)].

To avoid having two different categories for nume!als (and plural adjectives generally), we assume a'(X) to mean a'( units')(X). This

83

The Semantics and Syntax of Number and Numbers

is analogous to having y'( entity')(x) instead of y'(x) for singular adjectives, e.g., red entity' (x) instead of red' (x). Thus the meaning postulate for numerical operators is correctly formulated in the following way:

(MP 2)

D(VX)[a'( /3')(X) -

/3'(X) & a'( units')(X)].

Without this maneuver there would be a category change, or rather a type change for a' involved in the application of the meaning postulate. Something like this is not provided for in Montague Grammar. 24 Note, however, that plural nouns like red houses are built by applying the adjective red to the singular noun house before the plural formation takes place. Thus red is a singular adjective. Yet another rule for plural noun formation besides (S:Pl) and (S:Num), namely a rule for the formation of conjoined plural nouns, will be given in this paper. But that will be done later when the rules for the formation of conjoined noun phrases are given, to which conjoined plural noun formation will be contrasted. Before dealing with the grammar of plural terms I will mention a notational variant for the formation of plural nouns, which will be used in the following section. In (MP I) we had the formula

PI( house' )(X) -

chouse' (X) & X ¥- (/J,

with X as a variable over sets of individual concepts. Note, however, that any such set can be described by a property P of individual concepts. Thus we can express plural noun formation in the following way:

(MP l") houses'= PI(house') = p[chouse'(P) & VP¥-

O].

That is, the extension of the plural noun houses in intensional logic, houses', is the set of properties of individual concepts the extensions of which are subsets of the extension of the singular noun house, house', and which are not the empty set. Thus we have generally for plural noun formation:

(MP l")

Pl(a') = p[ca'(P) & p ¥- (/J].

24 But it would not be a problem for Natural Generative Grammar, because there a category change is a syntactic device which permits compression of complex structures having fewer categories into more compact structures closer to the surface level where more categories are used (cf. Bartsch and Vennemann, 1972). Thus in Natural Generative Grammar we have the meaning postulate in its first two versions as a constituent-building rule, if it is read from right to left. If it is read from left to right, it works as an interpretive rule.

84

Renate Bartsch

I further want to make a short remark about plural verbs. There are verbs which are basically plural.

(IV)

gather, disperse, etc.

And there are plural verbs which are derived from singular verbs, like walk in Men walk. The semantics of plural verb formation is the same as the semantics of plural noun formation. I will briefly state the syntactic rule and the translation rule for plural verb formation. (S:IVp1):

If /3 E P 1v, then F 21 (/3) = /3 E P1v,,•

(T:IVp1)

If f3 E P 1v translates into /3 then F 21 (/3) translates into Pl(/3 (of the same type as the translation of plural nouns). 1

1

,

)

(MP 1) holds for plural verbs as well for plural nouns. The Grammar of Plural Terms (Noun Phrases)

In a Montague Grammar, sentences with singular or plural determiners in noun phrases are directly interpretable in intensional logic. Montague (1970) provided for the treatment of terms with the singular determiners a, the, and every. There, sentences like A man walks are interpreted in the following way: the property 'walk' belongs to the set of properties of individuals (exactly: individual concepts) that are compatible with being a man, i.e., that ~e carried by at least one individual concept of a man. This set, P(3x)[man'(x) & P{x}], is the set of all these properties. It is the translation of a man into intensional logic. Then A man walks is interpreted in intensional logic as n ' P(3x)[(man'(x) & P{x})]( walk'). This transforms by the rules given in Montague (1970) into:

(3x)[man'(x) & 'walk'{x}]. This transforms into

(3x)[man'(x) & walk'(x)], which is in extensional logic (following the rules given):

(3u)[man'.(u) & walk'.(u)]. In this way, terms like a man, the man, and every man can be handled like proper names, where we have e.g., John walks: n

'

'

PP{ j}( walk'), that becomes 'walk' (j}, then walk' (j), and finally walk' (j).

85

The Semantics and Syntax of Number and Numbers

I now want to extend Montague's method to quantification with plural terms. For this I need to introduce a new type of concept: g> and 9, are properties of extensions of properties of individual concepts.

Syntactic rule (S:Tp1): F;(a) = F;(a) = F~(a) = Examples:

If a E PN, then some a E PT, F;'(a) = a E PT,, all a E PT,,, F;'(a) = a E PT, the a E PT . " F; (houses) = some houses F;'(houses) = houses 1,

1,

1

Translation rule (T:Tp1): If a' translates a E PN into intensional logic, then F; (a) and F;' (a) translate into l(3P)[a' ( P) & .2,( P}], F~(a) and F~'(a) translate into 2('v'P)[a'(P) 9,(P}], and F~(a) translates into l(3P)[a'( P) & ('v'O)[a'( O) 0 = P & 9,( P} ]. The next rule is the syntactic rule by which a plural term and an intransitive plural verb are connected to yield a sentence. This rule is parallel to Montague's (1970) S4 for singular terms.

(S4:Pl)

If 8

PT,

E

1

C Pl/Iv,,

and a

E P 1v, 1,

then F~(8,a) = F 6 (8,a) =

8a E P1•

Example: F~(Some men, disperse)= Some men disperse. The corresponding translation rule is parallel to rule T4 for singular terms given in Montague (1970).

(T4:Pl)

If 8 E PT, C P 111 v,, and a E P 1v,, translate into S' and a' respectively, then F~(8, a) translates into S' ( a'). Example: Some men disperse . 1

.Q(3P)[men'(P) & 9,{ "p} ]( disperse') This transforms to

(3P)[men'(P) & ~disperse'(P}]. This transforms to

(3P)[men'(P) & disperse'(P)]. This is the same as

(3X)[men' (X) & disperse' (X)].

86

Renate Bartsch

In extensional logic we have correspondingly:

(3U)[men~ (U) & disperse~ (U)]. In the case of a plural verb derived from a singular verb, like walk in All men walk, a version of meaning postulate (MP 1) can be employed.

(MP I"') If /3 is the translation of {3 E P,v and a'= Pl( /3 is the translation of F 21 (/3) = a E P,v,,, then 1

1

)

□(3Q)(VP)[a'(P)--.+ (Vx)[P{x} - - Q{x}] & Q = A/3'].

I will now illustrate this meaning postulate by applying it to the translation of All men walk.

Example: All men walk. The tree structure below shows the syntactic derivation of this sentence. all men walk (S4 : Pl)

~walk (S: IVpl)

all men (S: Tpl)

I

men (S : Pl)

I

walk

I man The corresponding derivation in intensional logic is:

man:

man'

men:

Pl(man')

all men:

.P(VP)[Pl(A man')(v P) - - f),{ VP} J

all men walk: .P(VP)[Pl( man')( P) - - f2,( P} ]( (Pl( walk'))). Applying the same rules as in the last example we obtain: (VP)[Pl( man')(P) - - Pl(walk')(P)J.

Using (MP l'") yields: (VP)[Pl( man')(P) - - (Vx)rP(x) - - walk'(x)JJ.

Applying (MP l') yields: (VP)[VP C man' & VP#-" - - (Vx)[VP(x) - - walk'(x)J.

87

The Semantics and Syntax of Number and Numbers

This is the same as: (Vx)[X C man' & X ~ ~ -

('vx)[x E X -

walk'(x)].

By applying the definition of the inclusion C and the transitivity of the conditional this turns out to be equivalent to

walk'(x)].

('vx)[man'(x) -

That is, the semantic representation of All men walk is equivalent to that of Every man walks. Instead of applying (MP l"') above we can apply (MP l') and obtain

cwalk'(X)].

(VX)[cman'(X) -

This is the same as (VX)[X c man' -

X C walk'].

This formula can be read as Every subset of the set of men is a subset of the set of those entities that walk. Of course this is equivalent to

walk'(x)].

('vx)[man'(x) -

Example: Some men walk. Using (T:Pl), (T:Tp 1), (T:lVp1) and (T4:Pl) we obtain r,

A

V

V

A

A

2(3P)[Pl( man')( P) &2{ P} ]( (Pl( walk'))). By using the same rules as in the example above we obtain (3X)[X C man' & X ~ ~ & ('vx)[x E X -

walk'(x)]],

or using a different notation (3XWman'(X) & X ~

0 & cwalk'(X)].

It is easily shown that this is equivalent to (3x)[man'(x) & walk'(x)].

Thus Some men walk turns out to be equivalent to A man walks. (I omit the conversion into extensional logic.)

Example: seek unicorns, seek some unicorns Following Montague (1970), seek a translates into seek'(a') in intensional logic, by the rules given. Thus we have: seek'(j(3P)[Pl(unicorn')(P) & .E?(P}]). This would be equivalent to seek'(Q[3x)[unicorn'(x) & Q{x}]), the translation of seek a unicorn into intensional logic, only if the following assumption holds: Every property 2 of extensions VP consists in

88

Renate Bartsch

the fact that all individual concepts in the extension of P have the property Q. This assumption does not hold generally. It only holds for every property 2 which is identical with the sense of a derived plural verb a' wit~ the extension Pl( /3' ), or "Yith the sense of a derived plural noun a' with the extension Pl( /3 In these cases 2 applies to P if and only if every x in VP has the property Q which is identical with /3 The weaker assumption (v'Q)(32)[2{ VP} Q{x}] holds generally. Using this assumption we can (v'x)[P{x} show that seek a unicorn implies seek unicp!ns by tra~sforming the translation of seek a unicorn into seek' ( Q[(3P)[Pl( unicorn')( P) Q{x}]]]), and then applying the assumption. 25 & (v'x)[P{x} Finally I want to give the rules for conjoined noun phrases. In Montague (1970) we find only disjunctive noun phrases in the singular like Peter or Paul, the man or the woman. The rules for the derivation of plural alternative noun phrases like three men or four women and for the derivation of all kinds of conjunctive noun phrases like Peter and Paul, men and women, a man and two women, some men and all women etc. will now be stated. A

A

1

A

1

).

•

Syntactic rule (S:T~1)

f is a symbol referring to the category of sin-

gular terms (T) as well as to the category of plural terms (TP 1). If a E Pr, and /3 E Pr, then F 8 (a, /3) = a and /3 E PT . If a E Pr, and /3 E Pr, then F 9 (a, /3) = a or /3 E PT Translation rule (T:T~1): If a' translates PT , and /3 translates f3 E PT , then F 8 (a, /3) translates into i[a' (2) & /3' (2)] and F 9 (a, /3) translates into i[a'(2) V /3'(2)]. If a' translates a E PT and /3' translates f3 E PT, then F 8 (a, {3) translates into i(3P)[(a'(P) & /3'(P)) & 2{ VP}], and F 9 (a, /3) translates into i(3P)[(a' (P) V /3 (P)) & 2{ VP}]. If a' translates a E PT and /3' translates f3 E PT then F 8 (a, /3) translates into i(3P)[(a'(P) & 2(P}) & /3'(2)], and F 9 (a, /3) translates into 2(3P)[(a'(P) & 2(P}) V /3'(2)] (and correspondinglyfora E PT and /3 E PT)In case the variable 2 denotes a property designated by a derived plural noun or verb, the meaning postulate (MP l"') applies, by which we have 0

,.

1

1

0,,

0,

2(P} -

(v'x)[P{x} -

Q{x}],

where 2 stands for a property designated by a derived plural noun or verb a and Q stands for a property designated by the correspond25 I am in doubt as to whether the replacement of equivalent phrases under the scope of a sense operator leaves the sense of the whole expression unchanged: predicate abstraction fJ would apply vacuously after replacement according to the assumption. Instead a new predicate abstraction Ji has to apply.

89

The Semantics and Syntax of Number and Numbers

ing singular noun or verb {3 [i.e., a'= PI( /3')]. Thus the translation of F 8 (a, /3) for singular terms a and {3 transforms into ()(3P)[(a'(P) & /3'(P)) & (Vx)P{x} -

Q{x}].

From this it follows for a and {3 with singular terms a and {3: ()[a'(Q) & /3'(Q)].

Example: John and Mary walk. 26 This sentence translates into 2(3P)[(j*(P) & m*(P)) & j?(P}]((Pl( walk'))). This transforms to (3P)[(j*(P) & m*(P)) & (Pl( walk')){ VP}]. By (MP l"') we obtain (3P)[(j*(P) & m*(P)) & (Vx)[P{x} -

Awalk'{x}]].

Now,j*(P)P(j}, and m*(P)P( m}. Therefore, we obtain (3P)[(P(j} & P( m}) & (Vx)[P{x} -

A

walk' {x}]].

This implies

waJk'fj} & wal]umuana aa-somene ci-tabo

98

Talmy Givon

Thus, ici-tabo 'book' above may be either definite or indefinite, but it is obligatorily referential. In contrast, following the negative of the verb, one finds:

umuana taa-somene ici-tabo 'The child did not read the book.' (referential, definite) umuana taa-somene ci-tabo (cili cyonse) 'The child didn't read a/any book.' (nonreferential) One may also demonstrate that the scope of NEG for producing opacity is downward transitive, so that nouns embedded not directly under its scope also represent an opaque environment:

umukashi taa-tampa uku-lemba ici-tabo 'The woman didn't begin to write the book.' (referential) umukashi taa-tampa uku-lemba ci-tabo (cili cyonse) 'The woman didn't begin to write any book.' (nonreferential) Furthermore, the negative need not be overt ("syntactic"). Rather, the semantic property of negation is involved. Thus, opacity-in English as in ChiBemba, is produced under the scope of the verb 'doubt':

umukashi aa-twiishiike ukuti umuana aa-lembele ici-tabo 'The woman doubted that the child wrote the book.' (referential) umukashi aa-twiishiike ukuti umuana aa-lembele ci-tabo (cili cyonse) 'The woman doubted that the child wrote a/any book.' (nonreferential) Finally, note that not only the object position, but also the subject position in the sentence becomes opaque under the downward scope of NEG:

umukashi aa-twiishiike ukuti umu-ana aa-ishile 'The woman doubted that the child came.' (referential) umukashi aa-twiishiike ukuti mu-ana(uli onse) aa-ishile 'The woman doubted that a/any child came.' (nonrefer~ntial) Modality Verbs One of the most classical examples of opacity in English involves the verb want or would like, as in:

John wanted to marry a girl with blue eyes.

99

Opacity and Reference in Language

It seems that in English, indefinite nouns coming under the scope of such a verb may be interpreted either referentially or nonreferentially, with the first alluding to 'a particular girl known to the speaker,' while the second to 'such a girl.' Note that the verb marry itself is implicative, so that if not embedded under want, as above, it allows only a referential interpretation:

John married a girl with blue eyes. In Givon (1972), modality verbs are defined as Vfrbs which require a sentential complement whose subject is identical to that of the modality verb itself. Furthermore, these verbs are divided into three subgroups: implicative

neg-implicative

nonimplicative

begin finish continue manage succeed remember

fail forget neglect avoid

want, decide, plan, agree, try hope intend

An implicative verb is a verb implying the truth of its complement verb. Apparently a strong implication is involved, since the negation of the complement verb implies the negation of the implicative modal verb:

John managed to read the book ::J John read the book ~ (John read the book) ::J ~ (John managed to read the book) A neg-implicative verb implies the falsity of its complement verb. Apparently again strong implication must be involved, since the negation of the negated complement verb also implies the negation of the neg-implicative verb:

John avoided reading the book ::J John didn't read the book ~ (John didn't read the book) ::J ~ (John avoided reading the book) Finally, nonimplicative verbs are those which do not imply either the truth or falsity of their complement verb:

John read the book } John wanted to read the book 1) { john didn't read the book Implicative and neg-implicative verbs differ from nonimplicative verbs with respect to one interesting quality, which one may desig-

100

Talmy Givon

nate as future-gap. In all modal verbs, the act of the complement may not precede in time the act of the main verbs, so that the following are equally ungrammatical: l)yesterday John managed to read a book the day before. !)Yesterday John failed to read a book the day before. !)Yesterday John wanted to read a book the day before.

However, with nonimplicative modal verbs, a future-gap may be opened between the time of the modal verb and that of the complement verb:

Yesterday John wanted/decided/planned/hoped to read a book the following day. In contrast, this future-gap is not possible for either implicative or neg-implicative verbs:

!)Yesterday John managed to read the book the following day. !)Yesterday John failed to read the book the following day. In other words, implicative (and neg-implicative) verbs require simultaneity of time. And this is of course obvious, since the truth of the implicative (or neg-implicative) verb at the designated time impliesat the very same time-the truth (or falsity, respectively) of its complement verb. By definition, a future-gap may not be opened. Turning now to opacity, one would by now suspect that neg-implicative verbs create an opaque environment for nouns in sentences embedded under their scope, given the earlier discussion of semantically inherent NEG. And in fact, in both English and ChiBemba, one finds examples such as:

umuana aa-keene uku-soma ici-tabo 'The child avoided/failed to read the book.' (referential) umuana aa-keene uku-soma ci-tabo (cili cyonse) 'The child avoided/failed to read a/any book.' (nonreferential) As to nonimplicative (or future-implicative) modal verbs, they turn out to create an opaque context under their scope (and in fact the classical examples from philosophers often use want, would like, etc.), as in:

umuana aa-salile uku-soma ici-tabo 'The child decided to read the/a book.' (referential) umuana aa-salile uku-soma ci-tabo (cili cyonse) 'The child decided to read some book.' (any, nonreferential)

Opacity and Reference in Language

101

One may further show that the opacity-producing power of these verbs is downward transitive, like that of NEG, above. So that the noun in question need not be embedded directly under the futureimplicative verb, but may be embedded further below, as in:

umuana aa-salile uku-tampa uku-soma ici-tabo 'The child decided to begin to read a/the book.' (referential) umuana aa-salile uku-tampa uku-soma ci-tabo (cili cyonse) 'The child decided to begin to read some book.' (any, nonreferential) Later on I will show that the future modality by itself produces opacity in its scope. This may explain the opacity of future-implicative modal verbs - since their complement sentence is always in a future modality vis-a-vis the modal verb itself. That is, they may be described as obligatorily involving the structure:

Finally, implicative verbs, in contrast with the two subgroups discussed above, exhibit no opacity in their scope. For example:

umuana aa-tampa uku-soma ici-tabo 'The child began to read a/the book.' (referential) umuana aa-tampa uku-soma ci-tabo (cili cyonse) This is hardly accidental, since if commitment of the speaker to the modal implicative verb implies also his commitment to the truth of the complement sentence, then (if no other opacity-producing agent intervenes) the object of a (nondefinitional) proposition is assumed to have referential identity. Further below it will be shown that a similar three-way dichotomy holds with respect to coercive-manipulative verbs.

Future-Implicative Transitive Verbs Among verbs which require a direct-object nominal complement, a small group are nonimplicative. That is, they do not imply the refer-

102

Talmy Givon

entiality of their object. Contrast, for example,find, which is implicative, with look for, which is not: a. John has found a/the book. (referential) b. John was looking for the book. (referential) c. John was looking for a book. (ambiguous, either referential or nonreferential) In English the indefinite is used ambiguously in this context. In

Spanish the ambiguity may be resolved with the use of the subjunctive modality in a relative modifier, as in:

Juan esta buscando a una criada que es bonita. (referential) Juan esta buscando a una criada que sea bonita. (nonreferential) 'John is looking for a beautiful maid.' (ambiguous) In ChiBemba the possibility of ambiguity does not arise, since in

this opaque environment the VCV/CV contrast is operative:

umuana aa-fwaayile ici-tabo 'The child was looking for a/the book.' (referential) umuana aa-fwaayile ci-tabo (cili cyonse) 'The child was looking for a book.' (nonreferential) There are only very few verbs in this class, such as look for, want, expect. I believe a plausible argument could be made for treating their seemingly nominal objects as, at least semantically, sentential complements. First, notice the paraphrases which normally obtain:

look for a house = 'act with the intent of finding a house' want an apple = 'want to have/eat an apple' expect a guest = 'expect the arrival of a guest' Furthermore, it is easy to see that these verbs are intentional. That is, the semantically underlying sentential complement is in the [FUTURE] modality with respect to the main verb. This is rather clear from the infelicities in:

J've found a house and I'm still looking for it. I've eaten an apple and I want it. The guest has arrived and I am still expecting her. It thus seems to me that the most natural explanation for the opacity of objects of these verbs lies in construing them as verbs with sentential complements obligatorily requiring the [FUTURE] modality. The verb look for, for example, will then be analyzed semantically as:

103

Opacity and Reference in Language

--------VP________ I ~s~ [ : :h

l

intent

I

«::::::-::: s '---:::::--,_ [ ••• find • .. ]

[FUTURE]

The Equi-NP condition will characterize these verbs just as it characterizes other modal verbs, see above.

Coercive-Manipulative Verbs Much like modal verbs, verbs of this type may be divided into implicative, neg-implicative and nonimplicative (for some discussion see Karttunen, 1971): implicative

neg-implicative

nonimplicative

force make

prevent

permit/allow forbid order/tell persuade dissuade request/ask

As earlier, implicative verbs imply the truth of their complement sentence:

John forced Mary to read a book :J Mary read a book - (Mary read a book) :J - (John forced Mary to read a book) Neg-implicative verbs imply the falsity of their complement sentence:

John prevented Mary from reading a book :J Mary didn't read a book -(Mary didn't read a book) :J - (John prevented Mary from reading a book) Finally, nonimplicative verbs imply neither the truth nor falsity of their complement sentence:

Mary read a book } {M 't rea d a b oo k ary d'd i n

John told Mary to read a book

1J

Mary read a book }) - ({ Mary didn't read a book

1J -

(John told Mary to read a book)

104

Talmy Giv6n

Since this scheme parallels the one shown for modal verbs, one expects the same alignment vis-a-vis opacity as above. And indeed, the neg-implicative verb prevent induces opacity under its scope:

umukashi aa-lesheeshye umuana uku-soma ici-tabo 'The woman prevented the child from reading the book.' (referential) umukashi aa-lesheeshye umuana uku-soma ci-tabo (cili cyonse) 'The woman prevented the child from reading any book.' (nonreferential) As before, the opacity may be ascribed to the scope of NEG. Implicative-coercive verbs, such as force or make, turn out to produce no opacity in their scope:

umukashi aa-konkomeeshye umuana uku-soma ici-tabo 'The woman made the child read a/the book.' (referential) "'umukashi aa-konkomeeshye umuana uku-soma ci-tabo (cili cyonse) Finally, nonimplicative coercive verbs tum out to produce opacity in their scope, as in:

umukashi aa-sumiine umuana a-some ici-tabo 'The woman permitted the child to read a/the book.' (referential) umukashi aa-sumiine umuana a-some ci-tabo (cili cyonse) 'The woman permitted the child to read a book.' (nonreferential) As above, one may argue that nonimplicative verbs involve the [FUTURE] modality in their complement sentences, while implicative verbs do not. This is supported by data such as:

,.,yesterday she made him read a book today. Yesterday she told him to read a book today. Cognition Verbs and Opacity In Karttunen (1971) and Giv6n (1972) cognition verbs, i.e., verbs requiring a sentential complement but without an Equi-NP condition, are divided, according to their presuppositions, into three subgroups:

105

Opacity and Reference in Language factive

neg-factive

nonfactive

know realize remember forget

pretend

think believe hope doubt be sure be afraid

The full list is not given here since the status of several of the verbs is still in doubt. A person using a factive verb presupposes the truth of its complement. That is, the complement sentence's being true is a precondition to both the positive and negative use of the factive verb. Thus: 0 John knew that it was Tuesday though it wasn't. 0 John didn't know that it was Tuesday and it wasn't. Neg-factive verbs, on the other hand, presuppose the falsity of their complements, and this presumably accounts for the infelicities in: 0 John pretended that he was sick and indeed he was. 0 John didn't pretend that he was guilty though he was. Nonfactive cognition verbs carry no presupposition of this type one way or the other, a fact reflected in the acceptability of all four sentences below:

John thought that he was sick, and he was. John didn't think that he was sick, but he was. john thought that he was sick, but he wasn't. John didn't think that he was sick, and he wasn't. Quite apart from their presuppositional status, two verbs in the lists above are inherently negative -forget and doubt, roughly corresponding, semantically, to NEG-remember and NEG-be sure, respectively. Given earlier discussion concerning opacity under the scope of NEG, one expects to find the referential/non-referential contrast in complements of these verbs, which is indeed the case: umukashi aa-labile ukuti umuana aa-soma ici-tabo 'The woman forgot that the child had read a/the book.' (referential) umukashi aa-labile ukuti umuana aa-soma ci-tabo cili cyonse 'The woman forgot that the child had read any book.' (nonreferential) umukashi aa-twiishiike ukuti umuana aa-soma ici-tabo 'The woman doubted that the child had read the book.' (referential)

106

Talmy Givon

umukashi aa-twiishiike ukuti umuana aa-soma ci-tabo cili cyonse 'The woman doubted that the child had read any book.' (nonreferential) One must note, however, that the verb forget loses its £activity (i.e., it loses its presuppositions) when used in this frame. That is, if a nominal in its scope is interpreted nonreferentially, it could not be that the complement sentence in which that nominal is found is considered fact. Thus:

Mary forgot that the child had read any book. could not possibly imply that the speaker is committed to the truth of:

The child has read a book. This is shown by the fact that if the subordinator fact is used after forget, which is a device strenthening the presuppositional interpretation of forget, nonreferential nouns in the same scope become inadmissible:

Mary forgot the fact that the child had read the book. (referential, definite) Mary forgot the fact that the child had read a book. (referential only) "'Mary forgot the fact that the child had read any book. ("'nonreferential) Factive cognition verbs (except for forget) do not produce opacity in their scope (all other things being equal):

umukashi aa-ishiibe ukuti umuana aa-soma ici-tabo 'The woman knew that the child had read a/the book.' (referential) "'umukashi aa-ishiibe ukuti umuana aa-soma ci-tabo (cili cyonse) Indeed it would be surprising if the speaker presupposed the truth of a nongeneric statement without also presupposing the referentiality of the nominals about which the statement is made. A baffiing case is that of the verb pretend. In many environments it behaves like a neg-factive verb, though it has been claimed to not be "truly presuppositional" (for some discussion, see Neubauer, 1971). There are two verbs roughly corresponding to pretend in ChiBemba, -beepelekeshya 'pretend' and -beepa 'lie.' (They are derivationally related, but the relation is not straightforward semantically.) In English lie is not as presuppositional as pretend, which is suggested by the fact that while the speaker is committed to the falsity

Opacity and Reference in Language

107

of the complement sentence of lie, he is not committed to the falsity of the complement of NEG-lie:

Laura lied to Bernie that she was pregnant, and indeed she was. Laura didn't lie to Bernie that she was pregnant, she really was. Perhaps then lie has negative implication rather than negative presupposition. At any rate, so far as opacity is concerned, it seems that one gets an opaque environment under the scope of both pretend and lie in English: John pretended that he saw a book. a. 'a specific one' b. 'some book' 0

John lied (to x) that he saw a book. a. 'a specific one' b. 'some book' In ChiBemba I have been able to obtain the VCV/CV contrast only for lie but not for pretend (although perhaps the VCV form following pretend is ambiguously referential/nonreferential):

umukashi aa-beepelekeshya ukuti umuana aa-soma ici-tabo 'The woman pretended that the child had read a/the book.' (referential, nonreferential) 0 umukashi aa-beepelekeshya ukuti umuana aa-soma ci-tabo (cili cyonse) With lie, however, opacity is obtained:

umukashi aa-beepele ukuti umuana aa-soma ici-tabo 'The woman lied that the child had read the book.' (referential) umukashi aa-beepele ukuti umuana aa-soma ci-tabo cili cyonse 'The woman lied that the child had read a/some book.' (nonreferential) Let us tum now to nonfactive cognition verbs. In both English and ChiBemba they tum out to produce an opaque environment under their scope:

umukashi aa-tontonkeenye ukuti umuana aa-somene ici-tabo 'The woman thought that the child had read a book.' (referential) umukashi aa-tontonkeenye ukuti umuana aa-somene ci-tabo (cili cyonse) 'The woman thought that the child had read a/some book.' (nonreferential)

108

Talmy Givan

The same is also true for -cetekela 'believe' and -subila 'hope.' To sum up this section, we have seen that, in addition to [NEG] and [FUTURE], one more modality produces opacity under its scope, the modality of [NONFACTIVITY]. It will be further discussed below.

Simple Tenses and Opacity In this section I will deal briefly with the relation between tenses (time modalities) and opacity. The use of the past tense in the examples up to this point was not accidental. All else being equal, past and present-progressive tenses commit the speaker to the belief that the act did take place (that is, if the speaker is not insincere). These two modalities are thus factive, and the nominals associated with sentences in these modalities are therefore presupposed to be refertial. Thus:

umuana a-a-somele ici-tabo 'The child read a/the book.' (referential) (>umuana a-a-somele ci-tabo (cili cyonse) umuana a-lee-soma ici-tabo 'The child is reading a/the book.' (referential) (>umuana a-lee-soma ci-tabo (cili cyonse) In contrast, the future modality is clearly not factive and, in fact, its use (with sincerity) implies that the action in question has not yet occurred. As hinted in the discussion earlier, one should expect nominal contexts under the scope of this modality to be opaque, and indeed it seems to me that the object of the following English sentence may indeed be interpreted either referentially or nonreferentially:

John will see a doctor tomorrow. a. 'a specific doctor' (referential) b. 'some doctor' (nonreferential) It is thus reassuring to find the VCV/CV contrast of ChiBemba in this modality:

umuana a-ka-soma ici-tabo 'The child will read a/the book (tomorrow).' (referential) umuana a-ka-soma ci-tabo (cili cyonse) 'The child will read some book (tomorrow).' (nonreferential) The status of definitional statements, which involve the habitual (always) modality, will be discussed further below. Taking a referen-

109

Opacity and Reference in Language

tial subject (i.e., a subject that is presupposed to have a unique referent), a verb in the habitual mode does not necessarily require an object with unique reference. So that the sentence below is ambiguous, with a book interpretable either referentially or nonreferentially:

John always read a book. a. 'a specific book (that he had found long ago)' (referential) b. 'some book' (nonreferential) It is thus encouraging to see that in ChiBemba one could find the VCV/CV contrast in nominals in this environment, as in: lyonse m-mu-sanga, a-lee-soma ici-tabo 'Always (when) I meet him, he is reading a/the book.' (referential) lyonse m-mu-sanga, a-lee-soma ci-tabo cili cyonse 'Always (when) I meet him, he is reading some book.' (nonreferential) Furthermore, one could also show that the opacity-producing scope of HABITUAL is downward-transitive, as in:

lyonse ilyo m-mu-kumanya, a-tampa uku-soma ici-tabo 'Always when I meet him, he starts reading a/the book.' (referential) lyonse ilyo m-mu-kumanya, a-tampa uku-soma ci-tabo cili cyonse 'Always when I meet him, he starts reading some book.' (nonreferential) The same downward-transitivity may be shown for the tense:

[FUTURE]

umuana-a-ka-soma ici-tabo/ci-tabo cili cyonse 'The child will (tomorrow) read a/the book.' (referential/nonreferential) umuana a-ka-tampa uku-soma ici-tabo/ci-tabo cili cyonse 'The child will (tomorrow) begin to read a/the book.' (referential/ nonreferential) Some Preliminary Generalizations From the discussion so far, it seems that the opaque environments in language are due to the presence of four distinct modalities: NEGATION,

FUTURE,

HABITUAL,

and

NONFACTIVITY.

Furthermore, it seems that a hitherto nonopaque nominal context may be rendered opaque whenever the sentence (proposition) in which it is included falls under the scope of one of these four modali-

110

Talmy Givon

ties. These four modalities contrast sharply with those of [PAST] or [PRESENT-PROGRESSIVE], in which the action imbued in a proposition is assumed to either have taken place in the past, or else to be at the moment taking place. It seems to me that one obtains here a sharp division between two types of modality in language-factive and nonfactive. A factive modality is one by which the speaker commits himself to the (past or present) truth of a certain proposition-and therefore also commits himself to the referentiality of the participating nominals. A nonfactive modality, on the other hand, is one in which the speaker does not make such a commitment. Turning to our four nonfactive modalities, one can observe the following division: (a) In the use of the [FUTURE] and [NONFACTIVE] modalities, the speaker is not committed to the actual occurrence (factivity), in a specific time-place, of an event or an action. These are thus, both, uncertainty modalities; (b) In using the [NEG] modality, however, the speaker is committed to the nonoccurrence of a certain event/action. This is, thus, a negative certainty modality. Put in terms of their logical scope, the characterization:

[FUTURE]

and

[NONF ACTIVE]

have

NEG(certain(P))

while

[NEG]

has the characterization: certain(NEG(P))

The latter is of course a much stronger condition-and it implies the former, the weaker one: (x is certain that NEG-p) ::J (x is NEG-certain that p)

though not vice versa: (x is NEG-certain that p)

1>

(x is certain that NEG-p)

It seems, thus, that [FUTURE], [NONFACTIVE] and [NEG] all share the weaker condition, their common denominator, the uncertainty modality. One may thus conclude that when the speaker embeds a proposition under this modality, that is when he is not obligatorily committed to the certainty (time-space occurrence) of the embedded proposition, then he is also not obligatorily committed to the referentiality of the nominals involved in that proposition. In other words, the nominal context in question is opaque.

Opacity and Reference in Language

111

One out of the four opacity-inducing modalities stands in seemingly sharp contrast to what was summarized above, the [HABITUAL] modality. Rather than having [UNCERTAINTY] involved, habitual, generic or definition propositions are not, to begin with, made about specific, time-space bound actions or events. Therefore, they are not necessarily made about specific (referential) nominals. Rather, they are made about classes or genera. Further, their internal structure involves a conditional expression. Thus, the sentence All humans are mortal may be logically rendered as: (x)

Hx

:::J

Mx

(informally: for all x, if x is human, then x

is mortal) Now, as will be seen shortly, conditional expressions are opaque, and that opacity is most likely traceable to the [FUTURE] - and thus to the uncertainty modality. At this point suffice it to say that even if this reduction did not hold, definitional expressions do not commit the speaker to the time-space occurrence of a certain act or event, and therefore do not obligatorily commit him either to the referentiality of any of the nominals included in the expression. I shall return to this discussion later on.

Other Opaque Environments In the preceding sections I have attempted to show that opacity may be traced to two distinct modality sources - [NEG-CERTAIN] and [GENERIC]/[HABITUAL]. With respect to those (and for the moment confining the discussion to object nominals), we have seen that it is possible to assert that a nominal is always interpreted referentially, unless it falls under the scope-directly or indirectly-of an opacityproducing modality. In the following sections I will sketch briefly how several other opaque contexts may be shown to arise from being embedded, either overtly or implicitly, under the scope of one of the three opacity-producing modalities. Simple Conditionals In English the situation with conditionals closely parallels, with respect to surface marking of articles, that of negation:

If John will read the book, ... (referential, definite) If John will read a book, ... (ambiguous, though without extra description usually interpreted as nonreferential) If John will read any book, ... (nonreferential)

112

Talmy Givon

The same is seen in ChiBemba:

nga umuana a-a-soma ici-tabo, 'if the child will read the book, .. .' (referential) nga umuana a-a-soma ci-tabo (cili cyonse ), . . . 'if the child will read a/any book, .. .' (nonreferential) One may argue that the opacity here arises from the presence of [FUTURE], overtly in English and implicitly in ChiBemba. Note, however, that [FUTURE] need not be on the surface in English either:

If John saw the girl, then I'm a submarine. (referential) If John saw a/any girl, then I'm a submarine. (nonreferential)

a. b.

In (a), (b) above the past tense is overtly present. Nevertheless, I would like to claim, some [FUTURE] is implicit, so that (a), (b) above are correctly paraphrased by:

a'.

b'.

If it will turn out that John has seen the girl, then ... (referential) If it will turn out that John has seen a/any girl, then .. (nonreferential)

In other words, some [FUTURE] modality is always inherent in these conditionals, whether it does or does not appear on the morphological surface. In ChiBemba past tenses may be also used in simple conditionals, as above, and there again opacity is obtained:

nga umuana a-a-soma ici-tabo, . . . 'If (it turns out that) the child has read the book, .. .' (referential) nga umuana a-a-soma ci-tabo (cili cyonse ), . . . 'If (it turns out that) the child has read a/any book, .. .' (nonreferential) Opacity produced in if-clauses of simple conditionals may be thus traced to the scope of [FUTURE] or [NEG-CERTAIN]. Counter-Factual Conditionals The sentence embedded in the if-clause of these conditionals is presupposed, if the speaker uses the construction sincerely, to be false. This is evident from the infelicity of:

"'If John had come home on time-and he did-then

Opacity and Reference in Language

113

Thus, since [NEG] is presupposed as the modality of the if-clause sentence, one would expect opacity under its scope, which is indeed the case in English:

If John had read the book, ... (referential) If John had read a/any book, ... (nonreferential) The same seems to be true of ChiBemba:

a-a-soma ici-tabo, . . . 'Had he read the book, .. .' (referential)

a-a-soma ci-tabo cili cyonse, ... 'Had he read any book, .. .' (nonreferential) Yes-No Questions This is one of the most obvious opaque environments, allowing the use of any in English to mark the nonreferential noun. No factive presuppositions are attached to sentences embedded in this construction, since presumably the speaker has deferred his judgment, assuming that the proposition in question may be either true or false. Thus, a nominal within a proposition that is not held to be true could not be obligatorily interpreted referentially, and this applies to both subject and object nominals:

Has Has Has Has

the man read this book? (referential subject) any man read this book? (nonreferential subject) the man read the book? (referential object) the man read any book? (nonreferential object)

It is thus not surprising to find the VCV/CV contrast in ChiBemba in is context, as in:

bushe, umukashi aa-somene ici-tabo?

'Say, did the woman read the book?" (referential)

bushe, umukashi aa-somene ci-tabo cili cyonse? 'Say, did the woman read any book?' (nonreferential) Perhaps the most natural way of accounting for this opacity involves the observation, made earlier, that coercive-manipulative verbs such as ask, request produce opacity under their scope. Including the underlying performative verb, yes-no questions have, roughly, the following underlying structure:

114

Talmy Givon

Thus, either the [FUTURE] or [NONFACTIVE] modality must be involved here, given what we have seen about the scope of these modalities. Constituent (Identity) Questions

In general, these questions involve a presupposed sentence (proposition) in which the identity of one element (here we shall deal only with nominals, such as subject, object) is not fully known-and is queried. Thus, the referentiality of the constituent under query is also presupposed. Also involved, obviously, is the same performative structure(. [REQUEST] . . . [TELL] . . . ) seen above. For a query such as: Who killed a lion? the conceptual (underlying semantic) structure involved is roughly: a. b.

Someone killed a lion. (presupposed) (and) I ask you to tell me who was the one who killed it.

Given no opacity producing agent, one would expect that if a nondefinitional proposition is presupposed to be true, the nominals involved in it are therefore presupposed to have referentiality. However, is not the query under the scope of the opacity-producing [REQUEST] modality? Not if we assume, after Morgan (1969), that the presupposed materials associated with a construction are semantically conjoined before the construction. Part (a) of the underlying structure above is thus outside the scope of the [REQUEST] modality in part (b), and hence no opacity is expected. In this vein, notice that after NEG one may get opaque nominal contexts even if the nominal is modified by some relative clause, as in:

Mary didn't see the lion running after the car. (referential) Mary didn't see any lion running after the car. (nonreferential)

115

Opacity and Reference in Language

However, if the relative clause is loaded with enough material to convey that the proposition involved is presupposed to have occurred (i.e., to be true), nonreferential interpretation becomes decidedly odd:

Mary didn't see the lion that ran after our car and tore her dress. (referential)

"Mary didn't see any lion that ran after our car and tore her dress. (nonreferential) In other words, a presupposed relative clause falls outside the scope of NEG. Further support for this claim comes from examining the scope of factive-presuppositional verbs such as know or remember. Under normal circumstances, these verbs are considered presuppositional:

Mary remembered that the child had read a book => The child had read a book. Mary didn't remember that the child had read a book => The child had read a book. However, when a nonreferential NP is used in the complement sentence (i.e., in English, an NP with any), many speakers may still accept these verbs - but without any presuppositions:

Mary did not remember that the child had read any book. Mary did not remember that anyone was there. In these cases, it seems that a paraphrase with if rather than that subordination is quite appropriate:

Mary did not remember if the child had read any book. Mary did not remember if anyone was there. Finally, there is a way of strengthening the presuppositional power of factive verbs, by using the noun fact before the complement sentence. When this is done, the complement sentence may not be subordinated by if, nor would it accept nonreferential (any) nouns in it:

Mary Mary Mary "Mary "Mary

did did did did did

not not not not not

remember remember remember remember remember

that the child had read any book. if the child had read any book. the fact that the child had read a/the book. the fact if . .. the fact that the child had read any book.

In other words, when the presuppositional status of the embedded clause is established beyond doubt, it becomes immune to the downward-transitive scope of the [NEG] modality with respect to opacity.

116

Talmy Giv6n

Thus the generalization suggested by Morgan (1969) seems to hold again. It has been often observed (for some instances see Myers, 1971; Schachter, 1971; Takizala, 1972) that a relative-clause construction underlies focused and constituent-question constructions. In ChiBemba this is clear from the variant form of constituent questions, below, in which the relative-clause structure is overtly present-and from which the same [ACTION-FOCUS] tenses which are excluded from relative clauses are also excluded:

ni-naani uo Joni aa-mweene?

'Who is it that John saw?' (lit.: 'it's who that John saw?')

"ni-naani uo Joni aali-mweene? The same restriction holds for the normal question pattern, where the relative clause does not appear overtly:

Joni aa-mweene naani? 'Whom did John see?' (lit.: 'John saw whom?')

"Joni aali-mweene naani? It is thus not surprising to find that VCV prefixes are obligatory for nominals in constituent questions in ChiBemba (provided no other opacity-producing modality intervenes):

naani aa-mweene ici-tabo? 'Who saw a/the book' (referential) "naani aa-mweene ci-tabo (cili cyonse)? Nominal Predicates -Attribution and Reference One of the most obvious contexts in which one may find the referential/nonreferential contrast is in nominal predicates. This is so because nominal predicates may be used to perform two functions: they either are attributive and thus describe the properties of the subject, or else they are referential and identify the subject. In ChiBemba one thus finds:

Joni muu-puupu. 'John is a thief.' (attributive)

Joni ali muu-puupu. 'John is the thief.' (referential) In the referential expression above it is clearly presupposed that a certain thief exists and is well-known to the speaker (and hearer).

117

Opacity and Reference in Language

The expression is thus used to identify one referential noun (John) with another (the thief). In the attributive expression, on the other hand, no such presupposition exists. It is quite baffiing to find the same contrast used with predicate adjectives in ChiBemba, as in:

Joni muu-bi. 'John is bad.' (attributive)

Joni ali umu-bi. 'John is the bad one.' (referential) However, the use of adjectives as referential predicates is clearly a case of anaphora. Most noun modifiers in ChiBemba may be used as anaphoric pronouns when the head noun is deleted. And further, they assume the CV/VCV prefix of the head noun. This may be seen in:

nshia-mweene umu-ana mu-bi

===}

nshia-mweene mu-ana mu-bi

'I didn't see any bad child.'

nshia-mweene umu-bi 'I didn't see the bad one.' (referential)

'I didn't see the bad child.' ===}

nshia-mweene mu-bi 'I didn't see any bad one.' (nonreferential)

One is then justified, I think, in deriving the seemingly referential use of adjectives from referential nominal predicates in the same manner: Joni ali umu-ntu mu-bi ==? Joni ali umu-bi 'John is the bad person.' ==? 'John is the bad one.' The data concerning the opacity of predicate nominals must still be explained in some way compatible with the generalizations made above concerning opacity-producing modalities. One natural way of doing this is by treating the modality of these sentences as a [HABITUAL]. Thus, the sentences: a. John is a thief. b. john is the thief who stole the crown jewels. are typically without reference to any particular time. Rather, they express timeless properties of the subject, either attributive (a) or referential (b). Though they may both be put in the past tense, it seems rather clear that what is involved, semantically, is a habitual past:

c. John was a thief. d. John was the thief who stole the crown jewels.

118

Talmy Givan

Sentence (c) above describes what used to be John's inherent attributes at some period in the past. The seeming past tense of sentence (d), however, does not imply that the identity in question was true only in the past but not at present. Rather, both (b) and (d) make the same timeless assertion with respect to John's identity. The past tense in (d) may indicate remoteness of reference or the fact that John is no longer alive. Thus, note that while the conjunction of (c) with the negation of (a) above is quite natural, the conjunction of (d) with the negation of (b) is rather odd:

John was a thief, though he is not a thief now. (>John was the thief who stole the crown jewels, though he is not that thief now. Lastly, notice that the following, seemingly universal, restriction holds:

Referential subjects may take either referential or attributive nominal predicates, but nonreferential subjects may take only non referential (attributive) predication. This may be illustrated by the acceptability of both:

Men are thieves. These men are thieves. As against:

These men are the thieves who stole the crown jewels. (>Men are the thieves who stole the crown jewels. The restriction may thus be formulated in a more general way, to read:

A predicate may never be more referential than its subject. This definition may need further widening, in order to accommodate the following observation, due to Gruber (1967): (>

I saw the man, therefore I saw a man. I saw a man, therefore I saw the man.

(the man = 'a man') (f> a man = 'the man')

In the expressions above definiteness, rather than referentiality, is involved. On some scale of generality, the referential is less general than the attributive, while the definite is less general than the indefinite. Finally, it seems that a similar restriction may be involved in the following asymmetry:

Men are primates. (>Primates are men.

Opacity and Reference in Language

119

In these definitional expressions, neither nominal is used referentially. Nevertheless the subject cannot be more general than its predicate nominal. Our restriction may be thus redefined as:

A predicate may never be less general than its subject. Opacity of Subject Nominals Traditional descriptions have it that the subject in ChiBemba may only have a VCV prefix. This contention is not altogether accurate. Note, first, that if a sentence is embedded and then under the scope of an opacity-producing modality, the referential/nonreferential contrast may appear in the subject noun as well:

Joni taa-toontoonkeenye ukuti umu-ana aa-ishile 'John did not think that the child came.' (referential) Joni taa-toontoonkeenye ukuti mu-ana uli onse aa-ishile

'John did not think that any child came.' (nonreferential)

Joni aa-fwaaya ukuti umu-ana a-ye 'John wanted the child to leave.' (referential) Joni aa-fwaaya ukuti mu-ana uli onse a-ye

'John wanted some child (be it any) to leave.' (nonreferential) Of the two nonfactive tense modalities, [FUTURE] and [HABITUAL], the second allows the VCV/CV contrast at least for some nouns, as in:

umu-tondo muuti 'The (this) Mutondo is a tree.' (referential)

mu-tondo uli onse muuti

'A Mutondo is a tree.' (nonreferential)

imi-tondo miiti

'The (these) Mutondos are trees.' (referential)

mi-tondo miiti 'Mutondos are trees.' (nonreferential) A special restriction seems to apply to human nouns, whereby only VCV-prefixes are allowed-and a VCV-subject in the [HABITUAL] modality is opaque:

aba-aume baa-kosa { 'Men are strong.' (nonreferential) } 'The men are strong.' (referential) 0

ba-aume baa-kosa

120

Talmy Givon

The [FUTURE] modality does not allow VCV/CV contrasts in subject position, but that position remains opaque and subject nominals may receive a referential or nonreferential interpretation:

(>mu-aume (uli onsee) a-ka-isa umu-aume a-ka-isa } { 'the man will come.' (referential) 'some man will come.' (nonreferential) Reference and Modality in Language

From the discussion above an account of opacity and reference in language does indeed emerge, and that account stands in considerable contrast to earlier attempts to deal with this problem in linguistics. In Fillmore (1966), Baker (1966), Bach (1968), among others, referentiality was treated, essentially, as a marked feature optionally associated with all nominals in the grammar. However, this smdy suggests an altogether different account of the siruation: (a) Nonreferentially interpreted nominals are possible only if they fall within sentences which are under the scope of a nonfactive sentence modality. (b) Otherwise, all nominals are obligatorily interpreted as referential. Thus, referentiality (or existential quantification) in human language, unlike what one may see in formal logic notation, is not a problem confined to nominals (variables), but rather is predicted from properties of sentences (propositions) in the following manner:

If a sentence is presupposed to be true and characterizes an event that took place at some time-and-place point, all the nominals in that sentence are obligatorily interpreted as having referential identity. Human language, it thus seems, presupposes the referentiality of nominals unless those nominals are associated with sentences which are not presupposed to describe events that have occurred, in the speaker's belief. The data cited earlier, concerning the resistance to opacity shown by sentences which are presupposed, tend to further strengthen this account. It is thus the modality of sentences and the speaker's commitment to believing in the reality of the events they describe, which determines the speaker's ability to use and interpret nominals nonreferentially. Of the nonfactive modalities dealt with above, [NEG] is the stronger

121

Opacity and Reference in Language

one and contains, by implication, the weaker [NEG-CERTAIN] modality. The modality [FUTURE] is quite often associated with lexical verbs, and in those instances the verbs involved are intentional. That is, the complement sentence stands in a future time-relation relative to the time of the main verb itself. The modality associated with nonfactive cognition verbs, such as think, believe, hope or fear, is, similarly, an uncertainty modality. Future is an uncertainty modality because it has not yet occurred, and thus may or may not occur. Finally, the [HABITUAL]/[GENERIC] modality characterizes either propositions about genera and thus does not pertain to specific individuals at specific time-place points, or else it pertains to referential individuals (subjects) with respect to which a proposition in the habitual modality holds true. That is, again no claim is made about a time-place bound event or action. In both cases, further, universal quantification is involved. Since conditional expressions underlie the semantics of universal quantification, and since the [NEG-CERTAIN] modality underlies conditional expressions, a natural explanation may be yet found for the opacity produced in the scope of [HABITUAL]/[GENERIC]. If that explanation holds, one may trace all opacity in language to the scope of one sentential modality, that of UNCERTAINTY. The general semantic rule for language may thus be given as: NOMINAL--➔ ([REFERENTIAL])/NEG-CERTAIN[ .. - - --➔

.. ,]p

[REFERENTIAL]/elsewhere

From the discussion above two sentential modalities have emerged which play a crucial role in language, [NEG] and [CERTAIN]. The modality of negation in language may be combinatorially obtained by the ordering [CERTAIN-NEG], while the modality of uncertainty, relevant to the discussion of opacity in language, is obtained by the opposite ordering, [NEG-CERTAIN]. The factive modality in language is then that of [CERTAIN]. It is of interest to show that the primitive sentential modalities discussed above, when combined with logically-defined notions of time and, often with some notion of presupposition, seem to underlie the tense-aspect modalities in language. Given the 'privileged member' ts (time of speech) in the ordered set of T (time), and the defining ts+ as 'time after ts' (future) and ts- as 'time before t; (past), one may characterize some of the major tenses in language as: PAST: [CERTAIN]-P-tsPRESENT PROGRESSIVE: [CERTAIN]-P-ts, [CERTAIN]-P-tsPresupposed: [CERTAIN-NEG]-P-tsFUTURE: Asserted: [NEG-CERTAIN]-P-ts+ HABITUAL: [CERTAIN]-P-at all t

122

Talmy Givon

These characterizations are informal and crude, but they seem to suggest that the sentential modalities [NEG] and [CERTAIN], which control opacity and referentiality in language, are not ad-hoc entities but rather figure out, in some combinatorial fashion, in the semantic structure of other modalities in language.

REFERENCES Bach, E. Nouns and noun phrases. In E. Bach, & R. Harms (Eds.), Universals in linguistic theory. New York: Holt, Rinehart and Winston, 1968. Baker, L. Definiteness and indefiniteness in English, unpublished M.A. thesis, University of Illinois, 1966. Bokamba, G. Reference and definiteness in Dzamba, Studies in African Linguistics, 1971, 2.3, Fillmore, C. On the syntax of pre-verbs. Unpublished manuscript, 1966. Giv6n, T. Forward implications, backward presuppositions and the time axis of verbs. Language, 1972, in press. (a) Giv6n, T. Studies in ChiBemba and Bantu Grammar, Studies in African Linguistics, 1972, 3, Supplement 3, in press. (b) Gruber, J. Functions of the lexicon in formal descriptive grammars. Systems Development Corp., Santa Monica, California, 1967. Karttunen, L. Implicative verbs. Language, 1971, 47, 340-358. Morgan, J. On the treatment of presupposition in transformational grammar, Papers from the Fifth Regional Meeting, Chicago Linguistic Society, 1969, 167-77. Mould, M. The semantics of the initial vowel in Luganda, Studies in African Linguistics, 1972, 3, Supplement 5, in press. Myers, A. Interrogatives and emphatics in Kikuyu, Studies in African Linguistics, 1971, 2, Supplement 2. Neubauer, P. The 2 3 surface verb pretend. Unpublished manuscript, University of Michigan, 1971. Quine, W. van 0. Word and object, Cambridge, Massachusetts: M.I.T. Press, 1960. Sambeek, J. van (1955), A Bemba grammar. Schachter, P. Focus and relativization. Language, 1971, in press. Sharman, J. Nominal and pronominal prefixes in Bemba, African Language Studies, 1963, 4, Sims, G. W. (1959) An elementary grammar of ChiBemba. Takizala, A. (1972) Focus and relativization in Kihung'an. Studies in African Linguistics, 3.2, in press.

FOCUS AND RELATIVIZATION: THE CASE OF KIHUNG' AN 1 ALEXIS TAKIZALA University of California, San Diego

INTRODUCTION

In recent years two approaches to the analysis of stress-focus constructions in English have emerged. These two approaches coincide with the two approaches to deep structure in transformational linguistics. For example, consider the sentences: (1)

John bought a chair.

(2)

John bought a chair.

(3)

John bought a chair.

While sentence (1) is neutral, (2) and (3) are semantically more complex, and may be paraphrased by (4) and (5), respectively: (4)

The thing that John bought is a chair.

(5)

The person who bought a chair is John.

1 Kihung'an (pronounced [kihu1)an]) is a Bantu language spoken in southwestern Congo (Kinshasa). I will use the tonal markings / ·, ', and ' / to indicate HIGH, LOW and MID tone, respectively. I am greatly indebted to S. Y. Kuroda for his kindness and patience in going over earlier drafts of this paper. I have also benefited a great deal from comments from Talmy Giv6n. Neither of them may agree with the analysis presented below, for which-as well as for inadvertent errors -I am solely responsible.

123

124

Alexis Takizala

In other words, stress-focused sentences involve presupposed sentences which may be given in the form of relative clauses, as in (4), (5) above. For sentences (2) and (3), the corresponding presuppositions are, respectively: (6)

John bought something.

(7)

Someone bought a chair.

Given its presupposition, the stress-focused sentence supplies - in its predicate-the identity of the unidentified constituent, i.e., chair in (2), (4) and John in (3), (5). The first approach to the analysis of focus construction corresponds to generative semantics and may be found in Postal (1971) or Lakoff (1965). Both propose deriving stress-focused sentences from their corresponding cleft sentences, which exhibit the very same presuppositional structure. Taking this approach, (2) and (3) above will be derived from, respectively: (8)

It was a chair that John bought.

(9)

It was John who bought a chair.

Since the underlying relation between cleft and pseudo-cleft constrnctions has been noted (see, for example, Akmajian, 1970), a variant of this approach may also derive both stress-focus and cleft constructions from pseudo-clefts, such as (4) and (5) above. In English this approach has two things going for it. First, it accounts for the correct semantic interpretation in terms of presuppositions. Second, it accounts for the stress, since post-copular predicates in English are stressed under "normal" intonation, while subject and object nominals are not. A radically different approach, loosely corresponding to interpretative semantics, has been taken by Chomsky (1971). He defines the focus as "a phrase containing the intonation center" and the presupposition as "an expression derived by replacing the focus with a variable [p. 205]." He offers two objections to the "relative clause and predicate" approach of Postal and Lakoff. First, that it has no syntactic support, but only a semantic justification involving presuppositions. Second, that many cases do not yield to "relative clause and predicate" analysis without considerable artificiality. For example, for the sentence: (10)

It was an ex-convict with a red shirt that he was warned

to look out for.

Focus and Relativization

125

Chomsky points out that either of the expressions in (11) below may be put in stress-focus: 2 (11)

a. an ex-convict with a red shirt b. with a red shirt c. a red shirt d. shirt

Chomsky then concludes that focus must be a property of surface structure, thus purporting to justify the existence of surface interpretation semantic rules. Since the derivation of stress-focused constructions from cleft or pseudo-cleft sentences remains semantically attractive, it is of interest to see whether languages other than English offer more syntactic evidence to support it. Languages in which both semantic or syntactic parallelisms between focus constructions, cleft or pseudocleft and relative clauses are quite numerous. For recent discussion, see Robinson (1972) and Schachter (1971). 3 In this paper I will concern myself almost exclusively with Kihung'an, a language in which interesting syntactic features are shared by stress-focus constructions (here manifesting themselves as tone-focus), cleft/pseudo-clefts, and relative clauses. Furthermore, I will also show that several of the rules required for deriving focus constructions from pseudo-clefts in Kihung' an have independent justification in the grammar of this language. In addition, I will also show how this treatment may be generalized to Wh-questions in Kihung' an, which share the presuppositional structure of cleft and pseudo-cleft. One may of course argue that this analysis may apply only to Kihung' an and not to English, where comparable syntactic support for the "relative clause and predicate" analysis does not exist. It seems to me unlikely, however, that given the universal nature of the semantic structure involved, languages should be analyzed differently on purely syntactic grounds. It is also very likely that a language may have had more syntactic evidence to support this semantic parallelism in the past and then lost it, and it is an open question whether a language of this kind has thus changed in such a funda2 Given Chomsky's definition, one would expect the entire phrase (i.e., ex-convict with a red shirt) to be the focus in (10), regardless of what constituent bears the stress. Item (11), however, shows that the focus may vary with the "intonation center" or, in other words, it is identified with it.

3 See also Awobuluyi (1972) for Yoruba, Callender (1971) for Old Egyptian and Langacker (1971) for French. Langacker also argues that Wh-questions are structurally related to cleft sentences in French.

126

Alexis Takizala

mental way, from a stage in which such deep semantic facts were represented in the deep structure, to a stage in which they are determined by surface interpretation rules. The structure ofWh-questions in Kihung'an seems to be undergoing such a change right now. One may also argue that the "relative clause and predicate" analysis must be wrong, since it derives three surface constructions from one underlying representation. This type of an argument is best countered by asserting that when several surface constructions share some (or most) of their semantic structure, they do not necessarily share all of it. This is particularly true at the level of topic, which is the most likely level where stress-focus, cleft and pseudo-cleft differ from each other semantically (for further discussion see Shopen, 1972). TONE-FOCUS IN KIHUNG' AN

Sentences (Ila), (12) and (13) below are comparable to the English sentences (1), (2) and (3) above: (Ila)

K'ipes ka-swiim-in k'it zoon4 Kipese -buy-past chair yesterday 'Kipese bought a chair yesterday.'

(12)

K'tpes ka-swiim-in kit z66n6 Kipese -buy-past chair yesterday 'Kipese bought a chair yesterday.'

(13)

Kipes a-swiim-in k'it z66n6 Kipese -buy-past chair yesterday 'Kipese bought a chair yesterday.'

Sentence (12) is contrastive on the object, kit 'chair'; sentence (13) is contrastive on the subject, Kipese; sentence (Ila) is neutral. 5 Unlike English, however, Kihung' an provides enough evidence that contrastive sentences involve relative clauses and, therefore, that the syntactic similarity between them and their neutral counterparts is only superficial. Let us first examine the relative clause. 4 Except in certain environments, which need not be specified here, the final vowel is always deleted in Kihung'an. When it does appear, it is almost always identical with the preceding vowel, thus suggesting vowel harmony. 5 Sentence (Ila) could also have a contrastive reading, e.g. 'Kipese bought a chair yesterday,' but this is not the reading intended here. The a/ka prefixal alternation doesn't concern us here. Roughly speaking, it has to do with the subject being in focus in one case but not in the other. A systematic investigation of this will not be pursued here.

127

Focus and Relativization

Relative Clauses in Kihung' an Given sentence (Ila) above, the following represents an object relative clause derived from it: (14)

kit ki a-swiim-in Kipes zoon 6 chair pro -buy-past Kipese yesterday 'the chair that Kipese bought yesterday'

The object relative pronoun ki agrees in gender and number with the head noun kit 'chair.'7 Two properties of relative clauses are of particular interest here. The first involves negation. In unembedded sentences the negative particle lo is used, as in: (15)

lo-i-mween kiti neg I-saw chair 'I didn't see the chair.'

In relative clauses, on the other hand, the particle lo may not be used. Rather, the negative verb -khoon- 'fail' is used instead, as in: 8 6 As may be seen here, the subject of the embedded clause must be postposed in object relativization. For further discussion of this phenomenon, see Giv6n (1972b). Giv6n, using my data, also argues that the relative pronoun in (14) must be bound to the verbal word, but I see no conclusive evidence why this has to be the case, and have therefore left it free. Unlike other elements attached to the verbal word, the object relative pronoun may be deleted under certain conditions. On the other hand, it may still turn out that Giv6n's analysis is correct. 7 This is in line with the characteristic grammatical agreement in Bantu, see for example Giv6n (1972a), Guthrie (1948, 1953), van den Eynde (1969). The rules require that predicate verbs and adjectives agree with the gender-number of the subject noun (as expressed in its prefix, henceforth PA), and that modifiers and pronouns agree with the gender-number of the head noun. In Kihung'an, for example:

bit bi-min by-ool bi-beeng bi i-swiim-in zoon bi-bir-ir chairs PA-my PA-red pro I-buy-past yesterday PA-get-lost-past 'My two red chairs that I bought yesterday got lost.' 8 The arrows below indicate that the two elements involved may come in either order. Like several other Bantu languages, Kihung'an tolerates a considerable measure of freedom in word order (a notion which is yet to be properly characterized, and may indeed be inappropriate, given stylistic considerations such as discourse structure and topic assignment). For example, (i) and (ii) below correspond to (lla) and (12), respectively:

(i) kit Klpes ka-swiim-in zbon ' 'Kipese bought a chair yesterday.' (ii) kit Klpes ka-swiim-in z66n6 'Kipese bought a chair yesterday.'

128

Alexis Takizala

....____.,.

kit ki a-khoon-in Kir.,es ku-suum

(16)

chair pro PA-fail-past Kipese to-buy 'the chair that Kipese didn't buy' 0 kit ki lo a-swiim-in Kipes (17)

kit ki a-khoon-in ku-su~es kye (ki-e) ki-beeng chair pro PA-fail-past to-buy Kipese PA-is PA-red 'The chair that Kipese failed to/didn't buy is red.' 0 kit ki lo a-swiim-in Kipes kye ki-beeng

(18)

lo i-mween kit ki a-khoon in Kir.es ku-suum ....____.....,, not I-saw chair pro PA-fail-past Kipese to-buy 'I didn't see the chair that Kipese failed to/didn't buy.' 0 lo i-mween kit ki lo a-swiim-in Kipes

The second relevant property of relative clauses involves the infix object pronoun. In unembedded sentences in Kihung'an the object pronoun (OP) may be used as a definitizer [as in (19)] or as a pronoun [as in (20)]: (19)

Kipes ka-ki-swjim-in kit zoon Kipese PA-it-by-past chair yesterday 'Kipese boughf the chair yesterday.'

(20)

Kipes ka-ki-swiim-in zoon Kipese PA-it-buy-past yesterday 'Kipese bought it yesterday.'

In object relativization, on the other hand, the OP may not appear: 9 (21)

0

kit ki a-ki-swiim-in Kipes zoon chair pro PA-it-buy-past Kipese yesterday

It is only natural that this restriction does not apply to subject relativization, where one may get either (22) or (23): (22)

muut wu a-ki-swiim-in kit zoon person pro PA-it-buy-past chair yesterday 'the person who bought the chair yesterday'

9 This is unlike many other Bantu languages, such as Dzamba or Tswana, which require the presence of the object infix pronoun in object relativization. The situation in Swahili is a bit more complex, with the object pronoun required in some patterns but not in others, where its presence or absence may be used to signal a semantic distinction.

Focus and Relativization

129

muut wu a-ki-swiim-in zoon person pro PA-it-buy-past yesterday 'the person who bought it yesterday'

(23)

Negation and OP in Tone-Focus Constructions While tone-focus constructions in Kihung'an superficially resemble neutral sentences in their word order, they follow the relative (rather than unembedded) pattern in negation:

Kipes kd-kh66n-in ku-suum kit z66n6 Kipese PA-fail-past to-buy chair yesterday 'Kipese did not buy a chair yesterday.' (i.e.: 'What Kipese did not buy yesterday is a chair.')

(24)

And the use of the negative particle lo here is as ungrammatical as it was in relative clauses:

° Kipes 16 ka-swiim-in kit zo6n6

(25)

Similarly: (26)

Kipes a-kh66n-in ku-suum kit z66n6 Kipese PA-fail-past to-buy chair yesterday

'Kipese didn't buy a chair yesterday.' (i.e.: 'The person who did not buy a chair yesterday was Kipese.')

But not: (27)

°Kipes 16 a-swiim-in kit z66n6

Stress-focus constructions also follow the relative pattern in that they reject the OP in object relativization: (28)

° Kipes kd-ki-swiim-in kit z66n6 Kipese PA-it-buy-past chair yesterday

As expected from the discussion above, however, the OP is acceptable if the subject (rather than object) is in focus: (29)

Kipes a-ki-swiim-in (kit) z66n6 Kipese PA-it-buy-past (chair) yesterday 'Kipese bought it yesterday.' (i.e.: 'The person who bought it (the chair) yesterday was Kipese.')

130

Alexis Takizala

As seen above, neutral sentences accept both the lo negative and the OP. As we shall see below, cleft and pseudo-cleft sentences share the relative pattern with tone-focus sentences in both respects. CLEFT AND PSEUDO-CLEFT IN KIHUNG'AN

Much like tone-focus sentences, clefting in Kihung'an may be used to either focus on the object, as in: (30)

(kwe) kit Kipes ka-swiim-in zoono (it's) chair Kipese PA-buy-past yesterday 'It's a/the chair that Kipese bought yesterday.'

or on the subject, as in: (31)

(kwe) Kipes a-swiim-in kit zoono (it's) Kipese PA-buy-past chair yesterday 'It's Kipese who bought a/the chair yesterday.'

Similarly with pseudo-cleft sentences: 10 (32)

(kiim) ki a-swiim-in Kipes zoon (kwer' kye) kit (thing) pro PA-buy-past Kipese yesterday (is) chair 'What Kipese bought yesterday is a/the chair.'

(33)

(muut) wu a-swiim-in kit zoon (kwe/"ke) Kipes (person) pro PA-buy-past chair yesterday (is) Kipese 'The one who bought a/the chair yesterday is Kipese.'

So far as presuppositional structure is concerned, cleft and pseudocleft sentences are identical with tone-focus sentences. As far as negation is concerned, clefts and pseudo-clefts - much like tone-focus sentences in Kihung'an-take the negative -khoon-, characteristic of relative clauses, rather than the normal particle lo. Thus, for object focus: (34)

(kwe) kit Kipes ka-khoon-in ku-suum zoono (it's) chair Kipese PA-fail-past to-buy yesterday 'It's a/the chair that Kipese didn't buy yesterday.' 0 (kwe) kit Kipes lo ka-swiim-in zoo no

10 Note the ungrammaticality of the copula kye (ki-e) and ke (ka-e). The copulas in pseudo-clefts do not exhibit grammatical agreement with the subjects (kiim 'thing' and muut 'person,' respectively). The significance of this will be discussed later on.

Focus and Relativization

(35)

131

(kiim) ki a-khoon-in Kipes ku-suum zoon (kwe) kit (thing) pro PA-fail-past Kipese to-buy yesterday (is) chair 'What Kipese didn't buy yesterday is a/the chair.' "(kiim) ki lo a-swiim-in Kipes zoon (kwe) kit

Similarly, for subject focus: (36)

(kwe) Kipes (wu) a-khoon-in ku-suum kit zoono (it's) Kipese (pro) PA-fail-past to-buy chair yesterday 'It's Kipese who didn't buy a/the chair yesterday.' "(kwe) Kipes (wu) lo a-swiim-in kit zoono

(37)

(muut) wu-a-khoon-in ku-suum kit zoon (kwe) Kipes (person) pro PA-fail-past to-buy chair yesterday (is) Kipese 'The one who didn't buy a/the chair yesterday is Kipese.' "(muut) wu lo a-swiim-in kit zoon (kwe) Kipes

The same parallelism with tone-focus and relative clause is observed with respect to the OP. Object-focus clefts and pseudo-clefts cannot take it, as is shown in: (38)

"(kwe) kit Kipes ka-ki-swiim-in zoono 11 (it's) chair Kipese PA-it-buy-past yesterday 0 (kwe) kit ki ka-ki-swiim-in Kipes zoono 11 (it's) chair pro PA-it-buy-past Kipese yesterday

(39)

"(kiim) ki a-ki-swiim-in Kipes zoon (kwe) kit (thing) pro PA-it-buy-past Kipese yesterday (is) chair

Given the syntactic similarities between relative clauses, on one hand, and focus, cleft and pseudo-cleft sentences on the other, one must conclude that syntactically, too, a relative clause is involved in focus constructions. This is in addition to the semantic arguments already mentioned. Taken together, the evidence strongly suggests deriving all three constructions from a similar underlying source which closely resembles the structure of a pseudo-cleft. Thus, for sentences (12), (30) and (32), the following underlying structure is suggested (this is without forgetting that at some discourse/stylistic level, perhaps involving Topic, they may still differ):

11 The object relative pronoun ki is present only if the subject of the embedded sentence (i.e., subject of the object relative clause) is postposed. If the subject is not postposed, the object relative pronoun is deleted. For further discussion of this, see Givon (1972b).

----S

132 (40)

Alexis Takizala

0

Copula

~ S1 ~

NP

Adv

/VP""'

NP

NP

V

kiim thing

- - - -

/VP~ Predfcate

/NP~

Kipes

I -suum-il

buy -past

I

kiim thing

zoono -kal yest. be

NP

I

kiti chair

where the relative clause stands for the presupposition and the predicate, kiti 'chair,' for the (object) noun in focus. THE DERIVATION OF FOCUSING CONSTRUCTIONS

Given the underlying structure in (40) above, one must consider the derivation of tone-focus, clefts and pseudo-clefts from it. It turns out that most of the (transformational) operations necessary for this already have independent justification in Kihung'an. Those are illustrated below. Copula Deletion The copula, in the present tense, is optionally deleted in Kihung' an as in many other Bantu languages. Thus, for example: (41)

{kit kye ki-beeng} ki·t k't- beeng

'Th

d, h . . e c aIT 1s re .

Similarly: (42)

{ mwaan WU kke'd.kidi} mwaan wu t i

. an 1'd'10t. , 'Th'1s c h'ld 1 1s

In tone-focus, as seen earlier, no copula appears. In cleft and pseudocleft sentences it appears optionally. Relative Pronoun Deletion Object relative pronouns in Kihung'an may be optionally deleted, and if they are deleted the subject of the embedded clause is not postposed:

133

Focus and Relativization

(43 ) {k~t ki_ a-swiim-!~ K_ipes zoon ... ·} kit Kipes a-swum-in zoon .

'the chair (that) Kipese bought yesterday'

The same is true with.object cleft sentences: 't k' .. . K' } (44 ) {(kwe ) k~ i_ a-swum-i~. ires zoon (kwe) kit Kipes ka-swum-in zoono

'It's a/the chair (that) Ki ese b h P d , oug t yester ay.

A similar deletion may be shown in subject-focus cleft sentences: } .. . k' ) K , , wu a-swum-in it zoon (45 ) {(kwe ipes .. . k ·t , , (k we ) Kipes a-swum-in i zoono

'It's Kipese (who) b ht /th h . oug d a , e c au yester ay.

Tone-focus sentences of course show no relative pronoun. As we shall see later on, Wh-questions exhibit a similar optional deletion of a relative pronoun. One may further argue that both copula deletion and relativepronoun deletion are involved in the embedding of adjectives as modifiers in Kihung'an (and Bantu in general). 12 In Kihung'an the optionality of these rules is attested in the existence of all three variants in (46) below: (46)

kit ki-beeng 'the red chair' kit ki ki-beeng 'the red chair' (lit.: 'the chair that red') kit ki kye ki-beeng 'the chair that is red'

Head-NP Deletion

In Bantu languages the subject of a sentence may be anaphorically deleted, pending a suitable discourse environment, after grammatical agreement has applied: (47)

Kipes ka-swiim-in kit zoon 'Kipese bought a chair yesterday.' ka-swiim-in kit zoon '(He) bought a chair yesterday.'

A similar deletion also applies to object nouns, following (infix) object pronominalization: (48)

Kipes ka-ki-swiim-in kit zoon Kipes ka-ki-swiim-in zoon

12

For details, see Takizala (1972).

'Kipese bought the chair yesterday.' 'Kipese bought it yesterday.'

134

Alexis Takizala

Optional Movement and Deletion in Focusing Of considerable interest is the following stylistic movement rule, by which the tone-focused object noun is preposed. Thus, given the normal tone-focus sentence (49), one also finds its paraphrase (50): (49)

Kipes ka-swiim-in kit zoono 'Kipese bought a chair yesterday.'

(50)

kit Kipese ka-swiim-in zoono '(It's) a chair (that) Kipese bought yesterday.'

Thus, sentence (50) may represent an intermediate between tonefocus and cleft. Notice also the possible variants, all with the same object focus: (51)

(kwe) kit kiim ki a-swiim-in Kipes zoon 'It's a chair, the thing that Kipese bought yesterday.'

(52)

(kwe) kit ki a-swiim-in Kipes zoon 'It's a chair, what Kipese bought yesterday.'

One may thus consider (51) and (52) as earlier intermediates in the derivation of (50), and thus ultimately of the tone-focus sentence (49). As we shall see later on, a similar range of intermediates is also found in Wh-questions. Although sentence (50) may resemble, in its syntactic order, a topicalized structure, it clearly could not be, since 'chair' is there in focus, and a noun cannot be topic and in focus at the same time, since it is the case that predicates are never in topic position. From the discussion above it seems that most of the syntactic rules through which one may derive either cleft or tone-focus constructions in Kinhung'an from the (pseudo-cleft-like) underlying structure given in (40) exist in the language. Furthermore, the existence of syntactic intermediates such as (50)-(52) above strongly suggests that in essence this manner of derivation has not only synchronic validity but perhaps also some diachronic support. While many details of the required pruning and predicate-lowering conventions (for discussion see Ross, 1967) have yet to be elaborated, it seems to me that the following sequence can be tentatively proposed. Beginning with (40) above, one applies the relevant relativization rules, including subjectpostposing, to obtain a pseudo-cleft with the head-noun intact, as in (53): (53)

kiim ki a-swiim-in Kipes zoon kwe kit 'The thing that Kipese bought yesterday is a chair.'

Optional copula deletion yields:

135

Focus and Relativization

(54)

kiim ki a-swiim-in Kipes zoon kit 'The thing that Kipese bought yesterday (is) a chair.'

Optional head-noun deletion yields: (55)

ki a-swiim-in Kipes zoon kit 'What Kipese bought yesterday is a chair.'

Predicate-fronting rule may now yield a cleft-like construction: (56)

kit ki a-swiim-in Kipes zoon 'It's a chair, what Kipese bought yesterday.'

If this movement rule had applied without copula deletion, one may get the cleft: (57)

kwe kit ki a-swiim-in Kipes zoon 'It's a chair, what Kipese bought yesterday.'

If, in addition, relative-pronoun deletion had applied, one may get the cleft without subject postposing: (58)

kwe kit Kipes a-swiim-in zoono 'It's a chair that Kipese bought yesterday.'

For the derivation of tone-focus sentences, however, fewer rules are necessary. Beginning with (55), to which copula deletion and head-noun deletion have already applied, one also applies relativepronoun deletion, which results in the movement of the subject noun back to its original place, to yield: (59)

Kipes ka-swiim-in zoon kit

Which is, in turn, a mere stylistic variant of the tone-focus construction: (60)

Kipes ka-swiim-in kit zoon 'Kipese bought a chair yesterday.'

Thus, in terms of existing rules of the language, as well as in terms of the existence of surface forms resembling to quite an extent the putative intermediates in the derivation, deriving both cleft and tone-focus constructions in Kihung'a11 from pseudo-cleft-like underlying structures does not appear to be very costly. THE COMPLEX-SENTENCE AND NEG-RAISING ARGUMENT

Sentences with complex verb phrases tum out to present no further difficulty for the analysis suggested above. Thus, consider the fol-

136

Alexis Takizala

lowing examples of tone-focus, cleft, and pseudo-cleft sentences:

tone-focus: (61)

cleft:

kwe beet (ha) tu-mon-aak ni beet lo tu-biing it's we (who) PA-think-prog that we neg PA-win 'It's we who think that we will not win.'

(62)

(63)

beet tu-mon-aak ni beet lo tu-biing we PA-see-prog that we neg PA-win 'We think that we will not win.'

beet baat ha {t:-_mon-aak ni beet lo tu-biing 13 we people who P A-think-prog that we neg PA-win 'It's we (who are) the people who think that we will not

win.' (64)

beet ha tu-mon-aak ni beet lo tu-biing we who PA-think-prog that we neg PA-win 'It's we who think that we will not win.'

{

'We are (the ones) who think that we will not win.'

}

pseudo-cleft: (65 )

(baat) ha {t:-_mon-aak ni beet lo tu-biing kwe beet 13 (people) who PA-think-prog that we neg win be we '(The people) who think that we will not win are we.'

Sentences (61) through (65) above may all be represented by the

13 The prefix-agreement variation ba-/tu- exemplifies how a language may be caught in the middle of grammatical reanalysis. In this case, it is caught in the middle between an overt relative-clause analysis (of cleft and pseudo-cleft) and a tone-focus analysis. Thus, for the pseudo-cleft sentence (65) 'The people who think that we will not win are we,' we expect the agreement ha- to correspond with the head noun baat 'people.' But we already get the alternation with tu-, which agrees with beet 'we.' In the cleft sentence (63), where beet 'we' has been fronted but the head of the relative clause is still baat 'people,' we again get the alternation. In the clefts (62) and (64), however, where the underlying baat 'people' has been deleted, we get obligatorily tu-, agreeing with the fronted beet 'we,' and this represents already, in terms of the prefix agreement, the same situation as in the tone-focus sentence (61), where no superficial evidence of the relative pronoun (ha) has survived. The existence of intermediate forms of this type lends credence to the hypothesis of deriving focus constructions from (pseudo- )clefts, since it suggests that this process may have in fact occurred diachronically.

137

Focus and Relativization

underlying structure in (66) below: (66)

- - - - - - - - S0

----------

/NP~ NP

/VP~

~ S1

NP

Copula Predicate

JP

~VP

v / ~s

baat people

baat

I

-mon-aak

people think-prob

/[~

lo neg

beet

-biing -kal

we

win

beet

be

we

Now, the verb -mon- 'think' is a neg-raising verb. (The arguments in support of neg-raising in Kihung' an are the same as those for English and will not be repeated here.) That is, the meaning of a sentence remains the same whether the negative is in the complement of think or on think itself. In Kihung' an, when the negative is in the complement of think, as in (61)-(65) above, one obtains the normal lo negative particle. However, if the negative is raised to the higher clause, only -khoon- 'fail' may be used, regardless of whether a relative clause overtly appears on the surface. Thus, sentences (67)-(71) below correspond to (61)-(65) above, respectively.

beet tu-khoon-aak ku-mon ni beet tu-biing

(67)

we PA-fail-prog to-think that we PA-win 'We don't think that we'll win.' 0

beet lo tu-mon-aak ni beet tu-biing

kwe beet (ha) tu-khoon-aak ku-mon ni beet tu-biing

(68)

it's we (who) PA-fail-prog to-think that we PA-win 'It's we who don't think that we'll win.' 0

(69)

kwe beet (ha) lo tu-mon-aak ni beet tu-biing

b eet b aat b a { btu-khoon-aak ku-mon ni beet tu-biing awe people who PA-fail-prog to-think that we PA-win 'It's we (who are) the people who don't think that we'll

win.' b b b l {tu-mon-aak ni beet tu-biing 0 eet aat a o b a-

138

Alexis Takizala

(10)

beet ha tu-khoon-aak ku-mon ni beet tu-biing we who PA-fail-prog to-think that we PA-win 'It's we who don't think that we'll win.' 'We are (the ones) who don't think that we'll win.'

,:, beet ha lo tu-mon-aak ni beet tu-biing (7l)

(baat) ha {~:~khoon-aak ku-mon ni beet tu-biing kwe beet (people) who PA-fail-prog to-think that we PA-win be we '(The people) who don't think that we'll win are we.' ,:,(baat) ha lo {~:~khoon-aak ku-mon ni beet tu-biing kwe beet

Thus, if Neg-raising is considered a valid (stylistic) rule of Kihung' an, then at the stage when this rule operates, the structure of focus sentences must involve a relative clause, since the raised NEG appears in the form appropriate to relative clauses. THE ANAPHORIC DELETION ARGUMENT

I have mentioned above that under the appropriate discourse conditions, subject and object nouns in Kihung'an may be deleted, after agreement or pronominalization have applied. However, the same does not apply when the noun in question is in focus, either through tone-focus, cleft or pseudo-cleft. Thus, for subject nouns:

Kipes ka-swiim-in kit zoon ==? ka-swiim-in kit zoon 'Kipese bought a chair '(He) bought a chair yesterday.' yesterday.'

(72)

but not: (73)

Kipes a-swiim-in kit zoono ='/='=-r ,:,a-swiim-in kit zoono 'Kipese bought a chair yesterday.'

nor: (74)

kwe Kipes (wu) a-swiim-in kit =/? ,:,kwe (wu) a-swiim-in zoono kit zoono 'It's Kipese who bought a chair yesterday.'

nor: (75)

(muut) wu a-swiim-in kit zoon =/? ,:,(muut) wu a-swiim-in kwe Kipes kit zoon kwe '(The man) who bought a chair yesterday is Kipese.'

139

Focus and Relativization

Data of this kind are not readily explained under the Chomsky (1971) analysis. The "relative clause and predicate" analysis, on the other hand, explains them quite naturally. Anaphoric deletion applies only to a noun previously mentioned in discourse; or, in other words, to a topic noun. Nouns in focus, however, present the new information in the sentence, the one contained in the predicate. Thus, notice [as may also be seen from the pseudo-cleft (75) above] that predicate nouns in Bantu (as well as in any other language) may not be anaphorically deleted: (76)

muut wu Kipes =f-~ 'This man (is) Kipese.'

0

muut wu

At the stage where anaphoric deletion applies, then, the noun in focus must still be in predicate position-and thus undeletable. Let us now consider object nominals. It has already been mentioned before that (anaphoric) object pronominalization may not apply to focused constructions, just as it does not apply to relative clauses. Thus, for neutral sentences, one may get: (77) Kipes ka-swiim-in kit zoon ==~ Kipes ka-ki-swiim-in kit zoon 'Kipese bought a chair 'Kipese bought the chair yesterday.' yesterday' ==~ Kipes ka-ki-swiim-in zoon 'Kipese bought it yesterday.' but not for object focus sentences: (78)

Kipes ka-swiim-in kit zoon =f-~ ° Kipes ka-ki-swiim-in kit zoon 'Kipese bought a chair =f-~ ° Kipes ka-ki-swiim-in-zoon yesterday'

Both restrictions, i.e., on both pronominalization and deletion, may now be explained quite independently of the restriction on object pronominalization in relative clauses. The infix object pronoun in Kihung' an is used either as a definitizer or as anaphoric pronoun. Both processes are discourse phenomena, depending upon establishing the noun in question as Topic in previous discourse. However, as I have just argued above, a noun in focus cannot be a topic noun. Now, if at the time when anaphoric object pronominalization and deletion applies the noun in focus is still in a predicate position, a natural explanation emerges for these restrictions: a predicate noun cannot be anaphorically deleted, since it contains the new information. Nor can it be replaced by a pronoun, since semantically it has not been mentioned in previous discourse, and syntactically, the pronominalization process in question applies only to object nouns

140

Alexis Takizala

but not to predicates in Bantu. Thus, the restrictions on deletion and pronominalization of both subject and object nouns turn out to involve the very same explanation, one which is consonant with the "relative clause and predicate" analysis of focus constructions. 14 Wh-QUESTIONS

The semantic relation between Wh-questions, on the one hand, and focusing sentences and relative clauses on the other, has been noted before (for some discussion of this subject in Bantu see Myers, 1971 and Heny, 1971). Thus, it may be shown that a Wh-question involves the same presuppositional structure as stress-focus, cleft or pseudo-cleft, which may be likewise expressed by the presence of a relative clause. For example, compare sentences (79) (focus) and (80) (Wh) below: (79) Kipese bought a chair= Presupposed: 'Someone bought a chair.' Asserted: 'That person is Kipese.' (80) Who bought a chair?

= Presupposed: 'Someone bought a chair.' Queried:

'That person is who?' Thus, while the performative involved is different (declare for focus, request for Wh), the remainder of the semantic structure is the same. Much like focus constructions, Wh-questions in Kihung'an require negation with -khoon--the one typical of relative clauses-rather than with the normal (lo) neg-marker: (81)

khi Kipes ka-swiim-in? 'What did Kipese buy?' khi Kipes ka-khoon-in ku-suumu? 'What didn't Kipese buy?' ,:, khi Kipes lo ka-swiim-in?

(82)

na Kipes ka-mweene? 'Who did Kipese see?' na Kipes ka-khoon-in ku-mono? 'Who didn't Kipese see?' ,:,na Kipes lo ka-mweene?

Further, one may show that these WR-questions further resemble relative clauses in that relative pronouns may optionally appear in them, much like they may appear in the corresponding cleft sentences. Thus,· (83) below paraphrases (81), while one reading of the ambiguous (84) paraphrases (82): 14

For further discussion of object-infixation in Kihung'an, see Appendix 1, below.

141

Focus and Relativization

(83)

khi ki a-swiim-in Kipes?

(84)

. ~ {'Who (is it) that Kipese saw?'} na wu a-mween Kipes. 'Wh o ('1s 1't) w h o saw Ki pese.?'

'What (is it) that Kipese bought?'

The obligatory subject-postposing in (84) above creates the ambiguity. is Much like relative clauses and focus constructions, Wh-questions do not allow object-pronoun infixation (or, for that matter, the deletion of the interrogative pronoun itself, which corresponds to the noun in focus): (85)

(>na Kipes ka-mu-mweene? (> 'Who did Kipese see him?' (>na wu ka-mu-mweene Kipes? { (>'Who is it that Kipese saw him?'} (>'Who is it that saw him Kipese ?'

The reason for this restriction must be identical to the one discussed in the preceding section. The schematic underlying structure for Wh-questions may be thus represented by (86) below (in which the performative structure I request you has been omitted), which corresponds to the underlying structure for focus constructions given in (40), above: (86)

The difference between focus sentences and Wh-questions may be then traced down to their different performative function. Given an identical presupposition (expressed in the relative clause modifying the subject noun of the construction), focus constructions supply (state) the new information, while Wh-questions request it. Otherwise, the syntactic parallelism shown above tends to support the "relative clause and predicate" analysis of focus constructions. 15 And subject postposing, in tum, is motivated by the presence of the relative pronoun, in contrast with (81) and (82) where it is not present. For further discussion, see Giv6n (1972b).

142

Alexis Takizala

Some Possible Difficulties

An analysis proposed by Akmajian (1970) poses an apparent difficulty for the "relative clause and predicate" approach to which, I have claimed above, the Kihung'an data lend considerable support. Akmajian has offered a rather similar analysis, but one involving a headless relative clause. He rejects the full-relative-clause analysis because of difficulties raised by pseudo-cleft sentences such as (87) below, which are better paraphrased by sentences such as (88), in which a relative pronoun (locative pronoun) appears: (87)

(tThe place where John beats his wife is in the garden.

(88)

Where John beats his wife is in the garden.

This solution, it seems to me, suffers from a considerable drawback, in that it generalizes on the exceptional case while treating the general case as an exception: (89)

The {one } who got killed was John. person (tWho got killed was John. The person to whom I gave the book was John. (To) whom I gave the book was (to) John.

(t

In addition, note the instances, below, in which both variants are admissible: (90)

The thing that I saw was an armadillo. What I saw was an armadillo. How he did it was by working hard. The way he did it was by working hard.

From this data it seems clear that the significance of these restrictions is at best superficial, and should better be expressed at the level of the lexicon, where specific restrictions on lexicalization, of precisely the same type, ought to be expressed. That these restrictions are. superficial and highly lexical-specific is also apparent from the fact that while in the case of nonhuman direct objects [in (90) above] both the headed and headless relative clauses are accepted, the same is not true for human direct objects: (91)

The person (whom) I saw was John. (tWhom I saw was John.

143

Focus and Relativization

DISCUSSION

(a) The difficulty raised by Chomsky (1971) concerning the possibility of multiple-focusing in one sentence [see examples (10), (11) above] still remains. It seems to me, however, that much of the apparent difficulty may be handled by the "relative clause and predicate" analysis. 16 Multiple-focus sentences of the kind discussed by Chomsky (1968) may be found in Kihung'an, as in: (92) muut wu a-fuun-aak Anglais phiim 'The man who speaks English well.' (not poorly) (93) muut wu a-fuun-aak Anglais phiim

'The man who speaks English well.' (not French)

(94) muut wu a-fuun-aak Anglais phiim 'The man who speaks English well.' (not reads) Their corresponding negatives must take -khoon- rather than lo, thus again suggesting that a relative clause is involved:

muut wu a-khoon-aak ku-fuun Anglais phiim. 'The man who doesn't speak English well.' muut wu a-khoon-aak ku-fuun Anglais phiim 'The man who doesn't speak English well.'

(95)

16

Take for example (i) below, taken from Chomsky (1971, ex. 52i):

(i) No, he was warned to look for an automobile salesman.

As Chomsky points out, it would indeed be quite artificial to derive it from: (ii) "No, the thing that he was warned to look out for a salesman was an automobile. The artificiality of the paraphrase-though not its acceptability-decreases, however, with: (iii) ?No, the thing that he warned to look out for a salesman of was an automo-

bile.

The unacceptability of (iii), however, arises from a violation of Ross's Complex NP Constraint, rather than any deep semantic reason. It may be easily paraphrased by: (iv) Presupposed: He was warned to look for a salesman who sells something. Asserted: What that salesman sells is automobiles. Thus, I think that Chomsky has raised difficulties which pertain to more superficial constraints and to lexicalization, but not to the validity of the semantic claims made by the "relative clause and predicate" analysis.

144

Alexis Takizala

muut wu a-khoon-aak ku-fuun Anglais phiim 'The man who doesn't speak English well.' Finally, it seems that the semantic structure of these sentences may be characterized reasonably well by pseudo-cleft-like constructions, as in e.g.: (92')

The way he speaks English is good.

(93')

The language he speaks well is English.

(94')

The thing he does well with English is speak it.

The underlying presuppositions and assertions involved are thus: (92")

P: He speaks English in a certain way. A: That way is good.

(93")

P: He speaks some language well.

A: That language is English.

(94")

P: He does something well with English.

A: What he does is speak it well.

While the accuracy of paraphrastic formulae of this kind is not complete, I think they clearly suggest that, at least in principle, Chomsky's (1971) objections are not extremely damaging to the analysis at which they are directed. (b) I have shown that the data of Kihung'an strongly support the more abstract, "relative clause and predicate" analysis of focus sentences rejected by Chomsky (1968). Clear syntactic data link tonefocus, clefts, pseudo-clefts, and relative clauses together. For data of this type Chomsky's analysis can give no natural explanation. (c) The fact that similar syntactic support is also found for linking focusing constructions (and relative clauses) with Wh-questions strengthens the analysis. This is so because Wh-questions share much of the presuppositional-semantic structure of focusing constructions, in which the presupposed sentence is represented as a relative clause. Thus, the fact that the syntactic analysis proposed here meshes well with the semantic analysis in all three cases, should be considered as additional support for its viability. One may of course wish to argue that this type of an analysis holds for Kihung'an and the languages surveyed by Schachter (1971), Myers (1971), and Heny (1971), to mention only a few, but not for English. It seems to me, however, that adopting a highly language-specific solution such as Chomsky's

Focus and Relativization

145

(1971) for phenomena as seemingly universal-certainly at the semantic level and to quite an extent at the syntactic level as well- as focusing constructions tum out to be, does little to convey the underlying generalizations of language structure. (d) The plausibility of the analysis presented here is enhanced, it seems to me, by the real possibility that the derivation of clefts, stressfocus sentences and Wh-questions from pseudo-cleft like constructions has in fact taken place diachronically in Kihung' an (and very likely in other languages as well). The presence of several intermediate forms in Kihung' an, with or without the head of the relative clause, with or without the copula and with or without the relative pronoun, tends to lend credence to this diachronic hypothesis.

APPENDIX 1: OBJECT PRONOUN INFIXATION IN KIHUNG'AN

Sentences (1), (2) below are related to the topicalized sentences (3), (4), respectively:

(1)

?Kipes bu ka-yis-ir ayi kit Kipese asp. PA-come-past with chair 'Kipese came with a chair.'

(2)

Kipes bu ka-ki-swiim-in kiti Kipese asp. PA-it-buy-past chair 'Kipese bought the chair.'

(3)

kit Kipes bu ka-yis-ir akyo chair Kipese asp. PA-come-past with-it 'The chair, Kipese came with it.'

(4)

kit Kipes bu ka-ki-swiim-in chair Kipese asp. PA-it-buy-past 'The chair, Kipese bought it.'

While the topicalized sentence (4) is an infixed object pronoun, the topicalized sentence (3) as a preposition-bound object pronoun, and infixation in this case is ungrammatical, e.g.:

(5)

kit

Kipes bu ka-ki-yis-ir ayi chair Kipese asp. PA-it-come-past with

Now, notice that the restriction on the appearance of object pronouns in relative clauses applies only to infixed object pronouns, as in (6) below, but not on preposition-bound object pronouns:

146 (6)

Alexis Takizala

kit ki a-swiim-in Kipes zoon chair pro PA-buy-past Kipese yesterday 'The chair that Kipese bought yesterday.' "'kit ki a-ki-swiim-in Kip es zoon chair that PA-it-buy-past Kipese yesterday

In contrast, an object pronoun must appear if the relativized object is prepositional, as in: (7)

(8)

Kipes ka-di-ir ayi mbeel ===> mbeel yi a-di-ir ayo Kipes knife pro PA-eat-past with-it Kipese PA-eat-past with knife Kipese 'Kipese ate with a knife.' 'The knife that Kipese ate with' Kipes ka-yis-ir ayi mwaan

===>

Kipese PA-come-past with child 'Kipese came with a child.'

mwaan wu a-yis-ir ando Kipes child pro PA-come-past with-him Kipese 'The child that Kipese came with'

It thus seems that the restriction on anaphoric object pronouns in relativization in Kihung'an applies only to the infix pronouns, but not to other, preposition-bound pronouns. The appearance of anaphoric object pronouns in relativization is widely attested in Bantu. Swahili allows the infix OP in relativization. Both Dzamba and Tswana require them obligatorily. The "double pronominalization" shown in preposition-object relativization in Kihung' an is also attested in Bantu (see for discussion Giv6n, 1971, 1972b). What this suggests is that the restriction on object pronominalization in Kihung'an relativization is a superficial restriction, specifically applying to the infix OP but not to others. This is important because, as argued on page 138 the restrictions on the appearance of object pronouns in focusing constructions (also in Wh-questions, see page 141, above) are independently motivated by the semantics of the underlying construction involved. This restriction is therefore not related to relativization, as I have erroneously suggested above. APPENDIX 2: IS IGBO AN EXCEPTION TO THE "RELATIVE CLAUSE AND PREDICATE" ANALYSIS?

Robinson (1972) has argued that in Igbo (Nigeria) one should not postulate an underlying relative clause for focus sentences. There

147

Focus and Relativization

are no surface relative pronouns in Igbo, and a relative clause is characterized by a tonal change, as shown in: (1)

Vee 1}yere 1}Wo'ke akwa Uce gave man cloth 'Uce gave the .man cloth.'

(2)

akwa masiri Vee cloth please Uce 'The cloth pleased Uce.'

===>

akwa Vc'e vyere vwo'ke cloth Uce gave man 'The cloth that U ce gave the man.'

===> a'kwa 'masjrj Vee

cloth please Uce 'The cloth that pleased Uce.'

In focus construction, the same tonal change appears in object-focus as seen in the relative clause in (1) above: (3)

ij by akwa ka (J' ce 1Jyere nwb'ke it is cloth Uce gave man 'It's cloth that Uce gave the man.'

In contrast, no tonal change appears in subject-focus constructions, as seen from (4) below (compared with (2)):

(4)

g bi.t, akwa masiri Vee it is cloth please U ce 'It's cloth that pleased U ce.'

To begin with, it is significant that in the object-focus (3) the relative tone does appear. Part of its absence in (4), further, may be readily explained. Notice first that the relative tone of the subject relative clause in (2) appears in two places: on the head noun (cloth) and on the verb itself (please). In the object relative (1), on the other hand, the relative tone appears on the underlying subject of the embedded clause. Now, notice that cloth in (4) does not, under the "relative clause and predicate" analysis, arise from the relative clause (whose underlying form is "Something pleases Uce"). Rather, it is the transported predicate noun brought into focus. Our analysis would in fact predict this behavior. In contrast, the subject Uce in (3) does arise from inside the underlying relative clause ('U ce gave something to the man'). It is thus not an accident that it does indeed preserve the characteristic relative tone. The only fact which remains to be explained is the nonrelative tone on the verb in (4). This may necessitate positing some kind of analogical leveling, of the type discussed earlier above in connection with Wh-questions in Kihung' an. But the Igbo data are not necessarily as damaging to our analysis as one may have assumed at first glance.

148

Alexis Takizala

REFERENCES Akmajian, A. On deriving cleft sentences from pseudo-cleft. Linguistic Inquiry, 1970, 1, 149-168. Awobuluyi, 0. On the nature of relativization. Paper presented at the Third Conference on African Linguistics, Bloomington, 1972. Bokamba, G. Immediate dominance, topicalization and relativization. Paper presented at the Third Conference on African Linguistics, Bloomington, 1972. Borkin, A. Polarity items in questions, in Papers from the Seventh Regional Meeting, Chicago Linguistic Society, 1971, 53-62. Callender, J. B. Notes on constructions with in, Studies in African Linguistics, 1971, 2(1), 1-23. Chomsky, N. Deep structure, surface structure and semantic interpretation. In D. D. Steinberg, & L. A. Jakobovits (Eds.), Semantics: An interdisciplinary reader in philosophy, linguistics and psychology. Cambridge, England: Cambridge University Press, 1971. Giv6n, T. The magical number two, Bantu pronouns and the theory of pronominalization. Studies in African Linguistics, 1971, 1(3), 279-300. Giv6n, T. (1972a) Studies in ChiBemba and Bantu grammar. Studies in African Linguistics, Supplement 3, 1972. Giv6n, T. (1972b) Pronoun attraction and subject postposing in Bantu. In P. M. Parenteau, J. N. Levi, & G. C. Phares (Eds.), The Chicago which hunt: Papers from the relative clause festival. Chicago: Chicago Linguistic Society, 1972. Guthrie, M. The Classification of the Bantu languages. London: Oxford University Press, 1948. Guthrie, M. The Bantu languages of western equatorial Africa. London: Oxford University Press, 1953. Heny, F. Focus and interrogative in Bali. Paper presented at the Second Conference on African Linguistics, Los Angeles, 1971. Lakoff, G. On the nature of syntactic irregularity. The Computation Laboratory of Harvard University Mathematical Linguistics and Automatic Translation, Report No. NSF-16. Cambridge, Massachusetts, 1965. Langacker, R. W. French interrogatives revisited. Unpublished manuscript, 1971. Myers, A. Emphatics and questions in Kikuyu. Proceedings of the Second Conference on African Linguistics, Studies in African Linguistics, Supplement 2, 1971, 11-18. Postal, P. Cross-over phenomena. New York: Holt, Rinehart and Winston, 1971. Robinson, J. 0. Focus, presupposition and WH-questions in Igbo. Paper presented at the Third Conference on African Linguistics, Bloomington, 1972. Ross, J. R. A proposed rule of tree-pruning, In D. A. Reihe!, & S. A. Schane (Eds.), Modern Studies in English. Englewood Cliffs, N. J.: Prentice-Hall, 1967. Schachter, P. (1971) Focus and relativization, to appear in Language. Shopen T. Logical equivalence is not semantic equivalence. Papers from the Eighth Regional Meeting, Chicago Linguistic Society, Chicago: University of Chicago Press, 1972. Takizala, A. Random notes on Kinung'an. Unpublished manuscript, 1972. Van den Eynde, K. Elements de grammaire Yaka. Kinshasa: Universite Lovanium, 1969.

ON J. EMONDS'S ANALYSIS OF EXTRAPOSITION 1 F. R. HIGGINS Massachusetts Institute of Technology

INTRODUCTION

In his thesis, P. S. Rosenbaum set up a transformational rule which he called Extraposition (Rosenbaum 1967) in order to account for alternations of the following kind: (1)

a.

b. (2)

a.

b. (3)

That John will come is unlikely. It is unlikely that John will come.

That the buses are dirty is generally admitted. It is generally admitted that the buses are dirty. a.

b.

To learn French well isn't easy. It isn't easy to learn French well.

This rule has since then been discussed several times, in particular by Lakoff (1968, pp. 14-29) and Ross (1967, passim, especially pp. 178-182). Ross gives the following formulation (ibid., p. 179): 1 This work was supported by a grant from the National Institute of Mental Health (Number 5 POI MH 13390). I am grateful to J. Bresnan and R. Faraci for discussion prior to the writing of the original version of this paper, and to A. N. Chomsky, J. Kimball, D. Perlmutter, and J. R. Ross for their comments, corrections and encouragement. Each of them would probably wish to dissociate himself from at least some part of the paper even in its mildly revised form.

149

150 (4)

F. R. Higgins

Extraposition.

SD:

X - [it - S]NP - y

SC:

1 1

2 2

3 0

OPT.

4 ==;> 4+3

I assume that their arguments are familiar, and will not repeat them here. More recently, J. Emonds (1970) has, within the framework of a broader hypothesis, reexamined the material and come to different conclusions. In essence, he suggests the following: 2 · (i) In examples (1)-(3), the (b) sentences correspond more closely to the underlying forms, with the extraposed sentence in fact being generated inside the verb phrase by the base rules and the it inserted as a pronoun in the subject position (pp. 95-96; pp. 103-104).3 (ii) It can function as the antecedent of a sentence, and hence the noun phrase into which it has been inserted and a sentence can be marked as coreferential with one another (pp. 119-122). (iii) A rule of Subject replacement (p. 123)4 can, under certain circumstances, substitute a sentence for a noun phrase in subject position with which it is coreferential (pp. 104-110; pp. 122-125). (iv) A rule of It-deletion (p. 122) deletes any it which immediately precedes a sentence with which it is coreferential (p. 122). This analysis, in conjunction with the distinction between "root" transformations (Emonds, Chapter 1) and "structure-preserving" 2 My exposition does less than justice to Emonds's arguments, which can only be fully appreciated as a part of the total hypothesis and are very difficult to summarize. I have presented sufficient detail to show the crucial status of two claims: (i) the claim that that-clauses and infinitive phrases are never noun phrases; (ii) the claim that his subject replacement rule is a root transformation. The first of these claims is discussed more fully in an appendix to this paper. The reader must consult Emonds's own exposition for a justification of his analysis. Since the original version of this paper was written, Emonds has published an appendix to his thesis (Emonds, n.d.) in which he presents more evidence in support of his analysis of the facts of extraposition (Emonds, n.d., pp. 4-8). Although some of the material is interesting, his conclusions are, I believe, overridden by those which I present. I refrain from discussing the contents of this appendix for reasons of space.

3 Since Emonds' s dissertation is the primary reference for this chapter and will be discussed throughout, it is to be understood that when page references only appear in parentheses, these references are to Emonds (1970). 4 The term "intraposition" has apparently become current as a name for this process and the associated rule. See, for instance, Baker and Brame (1972, p. 64).

On J. Emonds's Analysis of Extraposition

151

transformations (Emonds, Chapter 2), allows Emonds to give what he considers a less ad hoc explanation of various facts than Rosenbaum was able to give. The chief of these facts are: (a) In Rosenbaum's terms, sentences containing a Poss-ing complementizer do not permit Extraposition to apply. Rosenbaum dealt with this by adding a condition to the extraposition rule which prevented it from applying where the sentence immediately dominated by constituent 3 in the structural description [see (4)] contained a Poss-ing complementizer. (b) In Rosenbaum's terms, Extraposition appears to be obligatory where the constituent 2-3 is the subject of an embedded sentence, or where any transformation such as Subject-auxiliary inversion moves such a subject. An attempt was made by Ross (1967, pp. 56-59) to account for this by means of a surface structure constraint, but a correct version of the constraint has so far eluded discovery. (For further remarks on this matter, see the discussion of Ross' s output condition, below.) Clearly, (a) and (b) are, with respect to the analysis Rosenbaum gives, mere quirks, and Emonds's attempt to correlate them with other facts is welcome. Fact (a) is, for him, a result of the fact that "gerunds" (the term used by him for sentences with a Poss-ing complementizer in the highest sentence) are always noun phrases, and sentences introduced by that and infinitive phrases never are. A major part of his effort is devoted to trying to demonstrate that no noun phrase node dominates thatclauses or infinitive phrases (see particularly Section III.I). Once this central point is established, the rest of the analysis follows. Gerunds do not undergo extraposition from subject position because no such extraposition rule exists. However, since that-clauses and infinitive phrases manifestly do appear in subject position [as, for instance, in the (a) sentences of (1)-(3) above], and since a noun phrase node always dominates the deep structure subject position, it is necessary to generate such constituents elsewhere and move them into subject position in such a way that they are at no stage noun phrases. Emonds achieves this by allowing the Subject replacement rule to destroy the basic structure of the sentence whose subject is replaced: the subject noun phrase node is destroyed, and the new subject is a sentence without being a noun phrase. Emonds gives the following as an example [p. 104: (31) and (3), respectively-I have added referential indices in accordance with (52), p. 120]:

152

F. R. Higgins

s

(5)

~

it

VP

M

NPi

/ T~i'

would

irritate

him for the house to be painted

s

(6) S

~

I

for the house to be painted

VP

I i~r him

would irritate

A transformation of the kind that relates (5) and (6), that is, one which destroys some fundamental aspect of the structure of the sentence (here, the noun phrase node in subject position), is what Emonds calls a "root" transformation. A major characteristic of root transformations is that they do not apply in embedded sentences; which explains the first arm of the disjunction in fact (b) above. Moreover, Emonds points out (p. 108) that there is "a general condition on all the fronting root transformations of English: only one of them can apply in a given S," and, therefore, Subject replacement cannot apply if any other fronting root transformation applies. 5 This explains the other half of fact (b). 5 There is a minor blemish here, as Emonds admits (p. 108), for he wishes to analyze the rule of Wh-fronting as a structure-preserving rule although it behaves like a root transformation with respect to this condition. I suspect that the condition should read: "at most one root transformation may apply to a given sentence." Nearly all the root transformations that Emonds lists cause subject-auxiliary inversion to take place, effected by a root transformation, and so does Wh-fronting except out of subject position. A double application of Subject-auxiliary inversion would be undesirable, and so Wh-fronting cannot be combined with any other root transformation that causes subject-auxiliary inversion, even though it is not a root transformation itself. Furthermore, it is possible that rules like Wh-fronting and Topicalization move the fronted item into the COMP position (on this notion, see Bresnan, 1970). If the COMP position can only be singly occupied, the impossibility of combining certain fronting operations is accounted for-even though Topicalization, for instance, does not cause subject-auxiliary inversion.

On J. Emonds's Analysis of Extraposition

153

Thus, Emonds succeeds in explaining two apparently anomalous facts about English by relating them to the fundamental distinction that he wishes to draw, that between root and structure-preserving transformational rules. The crucial points in his explanation, it would appear, are the claims that that-clauses and infinitive phrases are never noun phrases and that Subject replacement is a root transformation. The purpose of the present paper is to show that neither of these claims is justified; neither can be upheld without considerable and otherwise unmotivated complication of the grammar. I shall show this using Emonds's framework as far as possible, 6 and will deal with the two claims separately-artificially, since essentially the same evidence serves to refute either of them. SUBJECT REPLACEMENT AS A ROOT TRANSFORMATION

Five kinds of cases will be presented to show that Subject replacement must be able to apply in embedded sentences. In other words, it cannot be a root transformation, but must be a structure-preserving transformation [which in itself means that it must be possible for that-clauses and infinitive phrases to be noun phrases and removes the basis for Emonds' s explanation of fact (b )] . The basis for the argument is the same in each case: I show that there are derivations in which the extraposed sentence must have occupied the subject position in an embedded clause at some point in the derivation. No doubt

6 Part of the argument involves the use of pseudo-cleft sentences. However, neither of the widely discussed analyses of pseudo-cleft sentences is completely satisfactory. The extraction analysis (Chomsky, 1970a; Akmajian, 1970a, 1970b) cannot be reconciled with the notion "strict cycle" (Chomsky, 1971, passim) without resort to ad hoc devices, and (as J. Bresnan has pointed out to me) involves an unfortunate kind of relative clause structure that contravenes the otherwise general requirement that the relative clause should contain a constituent that is coreferential with the antecedent at the deep structure level. The deletion analysis (Ross, 1972) requires a deep structure of a strange kind and also cannot be reconciled with the strict cycle. Therefore the reader will find me fumbling considerably when arguing on the basis of pseudo-cleft sentences and the uncertainties in the analyses hinder any attempt to follow through the arguments to their logical conclusions-there are too many possible variables. Therefore, such arguments are left unfinished. As far as I know, nothing crucial hangs on the pseudo-cleft examples and all can be omitted without detriment to the overall argument.

154

F. R. Higgins

further constructions might be found which would prove the same point. Much more might be said about each of the constructions that I use as evidence, but that would not be pertinent to the present discussion. All the constructions are reasonably transparent in their behavior with respect to the phenomena that are of interest here.

Pseudo-Cleft Constructions Consider the following pairs of sentences: a.

(7)

b. a.

(8)

b. a.

(9)

b. (10)

a.

b.

What is most likely is that Susan said she would be late. 0 What it is most likely is that Susan said she would be late. What was foolish of them was not to tell us when they were arriving. 0 What it was foolish of them was not to tell us when they were arriving. What John resents is that you refuse even to discuss the matter. 0 What John resents it is that you refuse even to discuss the matter. What you may depend upon is that we won't abandon him. 0 What you may depend upon it is that we won't abandon him.

(7) and (8) suggest immediately that one cannot extract the complement sentence from the extraposed position-that is, the basegenerated position in Emonds' s terms - but only from the subject position. The distribution can be explained using Akmajian's analysis of pseudo-cleft sentences as follows: (a) The extraction rule as given by Akmajian (1970b, p. 30) is: [8

[X-A-Y]s be [a]] -

[ 8 [X- [+PRO, +WH] -Y]s be [A]]

That is, a constituent is extracted from the full sentence and placed in focus position, to the right of the copula, and a pro-form is left in its place. (b) This rule operates as shown in (11) to derive (7), as long as the complement sentence is in subject position at the relevant point:

On J. Emonds's Analysis of Extraposition

155

(11)

[+

~:o

s NP PRO-------------._ S

~

t--(_2_)---+----l-Nf

+WH

S

~ that Susan said she would be late

VP V----------NP

VP

~

is most likely

I

I

t.

is

(1)

An alternative possibility, which preserves Emonds' s analysis, involves putting the wh-marker onto the it-subject and extracting the sentence that is coreferential with it, roughly as shown in (12). (12)

------

s

NP

PRO------S NP ~ [ +PRO +WH

2)

I

NPi

it

V

VP

~V~P--...,,--------, AP S·

I~

is most likely

v~s

~

that Susan said she would be late

I

is

I

t.

(1)

This too predicts the distribution found in (7) and (8) correctly, but only at the expense of adding a special branch to Akmajian's rule to deal with such cases. Notice also that the it to which the wh-pro-form is added may be in the object position (as in (9) above) or in a prepositional phrase (as in (10) above), and therefore further branches of the rules will be needed for such cases. Perhaps one could accommodate the possibility of extraction (rom all of these positions by means of a rule such as: "A 'plain' sentence (i.e., one not dominated by a noun phrase node) can be extracted if there is an it which is coreferential to it and to the left of it. A suitable wh-pro-form is substituted for the it." A formal statement of this rule would require a nonBoolean condition in the structural description, of course. Furthermore, if one once extends Akmajian's analysis in this fashion, the way is open for the derivation shown in (13) since Emonds

156

F. R. Higgins

apparently puts no constraints on assignment of coreference to it and a sentence.

s

(13) NP

PRO---------S NPi-----VP

I+ ;: l ( ~ l~ r-0

WH

2)

1

it

v~ s

I is

I most

likely

L..======::=J:=

that Susan said (it

VP V~S

Si

I

~

that she would be late

is

I

.That Susan would be late John didn't think it was very likely.

(47)

Relative clause formation. a. John hasn't arrived yet, which is very odd. b. (>John hasn't arrived yet, which it is very odd.

Apparently, each of these rules will move a constituent out of the subject position but not out of the extraposed position. It is clear that the movement out of subject position can be characterized as the movement of a noun phrase node, in keeping with well-known properties of the three rules in question. However, if these rules move noun phrases, and if the sentence in extraposed position is dominated by a noun phrase node, as in (44), how can one account for the impossibility of moving that noun phrase out of the extraposed position? If one refuses to admit the obvious, one might think of rescuing (44) by adding some kind of constraint to the grammar that will block the ungrammatical sentences. For example, as Robert Faraci pointed out to me, a supporter of the configuration (44) might impose a surface filter of the following kind: (48)

In other words, in surface structure it must always precede a sentence with which it is coreferential. Apart from the mental discomfort engendered by such a tailor-made condition, one may adduce objectimis of the following kind to such a proposal: (i) The proscribed configuration is, in fact, the one which the normal behavior of anaphoric elements would lead one to expect. Anaphors usually follow their antecedents. (The same objection can, of course, be leveled against Emonds's proposal for coreferentiality markings.) (ii) The proposed condition does not suffice to block (45b), which exhibits the permitted configuration: iti ... that Susan said she

On J. Emonds's Analysis of Extraposition

169

would be latej, Nor will it block similar ungrammatical sentential relative clauses in what, which normally precede their antecedent clause: (49)

a.

b.

. . And, what is even odder, John has never apol-

ogized. " .. And, what it is even odder, John has never apologized.

Part of this objection could perhaps be countered by the fact that pseudo-cleft sentences in any case do not appear to obey Langacker' s (1969) constraints on pronominalization, in that backward pronominalization into them is blocked: (50)

a.

b.

"What surprised himi was that Johni had not been told. What surprised Johni was that hei had not been told.

(This was noted by Postal, 1971.) But the simplicity of the surface filter is beginning to vanish. (iii) One can split the antecedent sentence, as in (5lb,c): (51)

a.

Iti wasn't easy [to conceal that they had had dealings

with him]j, b. That they had had dealings with him it wasn't easy to conceal. c. They had had dealings with him, which it wasn't easy to conceal.

In (51b) the sentential object of conceal has been topicalized and now precedes the it which is supposed to be marked as coreferential to the whole infinitive phrase introduced by to conceal. In (51c) the sentential object of conceal has been relativized and, on the assumption that the relative pronoun which somehow retains the referential index of the deleted element, the it is preceded and followed by parts of the infinitive phrase with which it is supposed to be coreferential. Again, one could invent all kinds of devices to cope with such examples while retaining the condition (48), or one could make the condition more complex, but it would be preferable to avoid such moves. (iv) It may not be easy to state the domain of application of (48). Even within a sentence one can find configurations such as the following: (52)

Everyone was amazed [that he was leavingt apart from those who had heard about iti beforehand.

170

F. R. Higgins

Furthermore, Emonds wishes coreference relations to hold not only between it and that-clauses or infinitive phrases but also between it and certain other kinds of subordinate clauses (pp. 119, 123), and this position certainly has some intuitive appeal. Notice now the effect of fronting the protasis in his example (59): (53)

Iti would have been pointed out by John if [there had been any danger] 1• b. If [there had been any danger]i, iti would have been pointed out by John. 11 a.

Examples (52) and (53b) contravene condition (48) and yet are grammatical. Moreover, it is of course very common to find it used as a sentence anaphor in discourse, across sentence boundaries. Since the it follows its antecedent in such examples, the configuration also contravenes (48). 12 11 Emonds does not actually give any example like (53b) in his thesis, though he does give a couple in the appendix (Emonds n.d., p. 6). I suspect that my (53b) and his examples may be illegitimate. Can it in (53b) really refer to the whole sentence which precedes it, or can it in fact only refer to danger? Similarly, in Emonds's example:

(i)

If the dog did tricks, it amused my friends.

Is this sentence equally acceptable if the dog is replaced by a noun phrase referring to a human being? Compare: (ii)

If John did conjuring tricks, it amused my friends.

This example is improved if it is in a context where that can be substituted for it: (iii)

If John did conjuring tricks, that too amused my friends.

But by now it is not clear that one is really dealing with the same kind of construction, for the following is certainly ungrammatical: (iv)

"That too amused my friends if John did conjuring tricks.

Thus, it is possible that my example (53) should be ignored. Objection (iv) is correspondingly weakened. 12 I shall make no attempt to give examples, since the point is difficult to illustrate with short examples and not crucial. When sentence and anaphor are only a short distance apart it is much more common to find that (or this) rather than it. This probably has to do with the fact that it, in some ill-defined way, cannot be stressed. The same phenomenon is found with simple noun phrases, as in:

(i)

. . . John's father ... oh, everyone likes him.

(ii)

?? . . . John's dog ... oh, everyone likes it.

(iii)

. John's birthday party? ... oh, that's an occasion everyone likes.

(iv)

?? . . . John's birthday party? ... oh, it's an occasion everyone likes.

On J. Emonds's Analysis of Extraposition

171

These four objections suggest that (48) is hopelessly inadequate. All other similar alternatives seem to be. Therefore, it is unlikely that any ancillary restrictions will render adequate a structure like that in (44). There remains (43), repeated here:

s

(54)

NP--------------VP

t

v~

Descriptively, this seems to be adequate, and it explains the fact that sentences in the extraposed position do not behave like noun phrases even though they do when in subject position (as shown by their behavior with respect to the rules of Pseudo-cleft formation, Topicalization, and Relative clause formation). It is the simplest solution compatible with this fact. Nevertheless, it is a solution that Emonds's own principles perhaps do not permit. Emonds requires independent motivation for the nodes into which items are moved by structure-preserving rules. As far as I know, one only needs to generate the configuration in which a noun phrase node exhaustively dominates a sentence node to deal with precisely the class of sentential complements. Hence, the only evidence independent of that-clauses and infinitive phrases is provided by gerunds. If gerunds really are generated as sentences that are exhaustively dominated by a noun phrase node, then (54) is weakly but, presumably, adequately supported, and Emonds's (pp. 111-112) explanation of why gerunds do not extrapose 13 is 13 In fact, the claim is not completely uncontroversial. Even if one accepts the possibility that many apparent cases of extraposed gerunds are really to be explained as right-dislocated forms, there remain examples like:

(i)

It's no use telling me that.

Emonds suggests in a footnote that these are "to be classed with the non-NP ing complements to verbs of temporal aspect . . . [p. 111] ." But neither this suggestion, nor the other possibility-that they are progressive forms with a deleted infinitive-will explain examples like: (ii)

It's no fun owning an elephant.

(iii)

It's no fun weighing fifteen stone.

Here the verb normally has no progressive form, although it forms gerunds and participles freely. The gerund can appear in subject position, even in embedded clauses: (Continued on page 172)

172

F. R. Higgins

successful in all points but one: why can one not generate gerunds in the extraposed position in the first place? At this point I cannot offer any cogent or even coherent direct counterargument or counterproposal. I do not know enough about the syntax of gerunds. With some considerable diffidence I should like to suggest (as a research proposal) that gerund phrases do not in fact contain a sentence node (or a complementizer) at all, and, therefore, would not be generated in extraposed position in (54) in Emonds' s terms, or extraposed in Rosenbaum's terms. That is, they are, as the traditional term "verbal noun" suggests, nonsentential noun phrases with some such structure as: (55)

NP Det --------------N

I

I

NP

VP

~

I

John

shooting the hunters

If a structure such as (55) could be motivated for gerund phrases, there would be no possibility of adopting (54), since there would then be no evidence for the required substructure apart from thatclauses and infinitive phrases themselves.

(iv)

Owning an elephant is no fun.

(v) John said that because owning an elephant had turned out to be no fun, he

was going to sell his. As Jespersen (1946) points out in connection with such examples: "there is an alternative construction (chiefly Irish) with there [p. 90]," and he gives examples such as: (vi)

If they were all like her there'd be very little use our coming here.

The evidence is unclear, but, I would judge, points away from an explanation either in terms of extraposition or in terms of verbs of temporal aspect. Could one adopt a suggestion of M. Halie's, as reported by Emonds (Footnote to pp. 47-48), to the effect that sentences like: (vii)

It's Bill in the back seat.

(viii)

It's my friends outside.

might be produced by a structure-preserving rule of It-insertion, similar to the rule of There-insertion, when the subject noun is definite? The extraposed gerunds in (i)(iii) would be positioned by such a rule, and they would somehow count as indefinite in dialects that show there instead of it.

173

On J. Emonds's Analysis of Extraposition

A RULE OF EXTRAPOSITION

It is not necessary to abandon (54) entirely, however. The following retains all its virtues: (56)

This is, of course, Rosenbaum's rule of extraposition interpreted as a structure-preserving rule. Emonds himself gives a diagram similar to (56) and comments: "However, if one assumes that Rosenbaum's analysis of sentences and infinitives is correct, it is easy to see that his important rule of extraposition ... is structure-preserving [p. 92, fn.]."

The sole argument that I have offered against (54), and therefore, by a process of elimination, in favor of (56), has involved a quibble about the status of gerunds and is at present completely vacuous. Can any stronger argument in favor of (56) be offered? Emonds is forced to add a condition to the Subject replacement rule to prevent elements "from too far down the tree" from being moved into subject position, as in examples like (53a) above. He states the rule as follows (p. 123): (57)

[s NPi - X - Si - Y]

==> 3 -

2 - (/J - 4,

where no NP, PP, or S dominated by the S whose subject is replaced dominates 3. He comments: "The condition on (58) prevents S's which are 'too far down' in the tree from replacing the subject." Emonds's (59) [my (53a)] is given as an example where the rule of Subject replacement cannot apply. The if-clause is, for Emonds, a prepositional phrase, and therefore the sentence embedded in it cannot be extracted and put into subject position. Unfortunately the condition is probably not correct. I suspect that Emonds intended it, in part, to go proxy for a condition on the assignment of indices of coreference to it and sentences, since he himself states no such condition. At any rate, it is easy to see that some such condition is needed if his (58) [my (57)] is to be upheld. Consider the deep structure (58) (in which the passive auxiliaries are omitted).

174

F. R. Higgins

(58)

So NP~VP

v ~s

I PRO

1

/

think

NP~VP

---::r---._ S AP I I ~

V

it

2•

(to be) unwise

that you didn't try harder

The rules of Subject raising (I adopt a conservative stance here, since the point is of no importance- see Chomsky, 1971) and Passivization on the S0 -cycle lead to the structure (59). (59)

S0

NP;~VP

v~s it

I

is thought

i1 S AP v ---------:i::-VP

======--=:::::::.= that you didn't try harder 2 •

I

(to be)

I

unwise

1

Now one cannot apply the rule of Subject replacement to put S2 into the subject of S0 , since Emonds's condition blocks it: S2 is dominated by S1 , which is in tum dominated by S0 • However, the rule of Subject replacement must be able to apply to give: (60)

That you didn't try harder was thought (to be) unwise.

(See pages 115-117 for a discussion of the relationship of the rules of Passivization and Subject replacement. In Emonds' s framework, Passivization takes place first.) It is possible that one could invoke pruning to explain the grammaticality of (60) while preserving Emonds's constraint: the node S1 would be pruned and hence the constraint would no longer apply. Therefore consider the result of adding one further similar level of embedding to this example, and of utilizing the freedom that Emonds permits in the assignment of indices of coreference, as in the deep structure (61).

175

On J. Emonds's Analysis of Extraposition

(61)

So

NP~VP

v~s

I

I

i

~

think

PRO

NPi

I it

V

I

VP

~S I

(to be) unwise

2

~ to admit (it)

S3 i

~

that you didn't try harder

Note that S3 in (61) has been marked as coreferential with the it in S1 • Again apply the rules of Subject raising (to the subject of S1 ), Passivization, and Subject replacement on the S0 -cycle. One obtains (62). (62)

What I resent is it that Bill didn't tell me.

(Recall that even if one uses an analysis with an It-deletion rule, it must be prevented from applying after resent.) Since the rule of Extraposition is obligatory with this verb, one can apparently only avoid (67) and obtain (65) by ordering Extraposition after the deletion rule on the S0 cycle. This does not show, of course, that Extraposition is last-cyclic, since S0 may not be the highest node in the tree, but it does show that it may be incorrect to apply the rule of Extraposition in the lowest cycle in which its structural description is met. The simplest way of ensuring that this is so is to make it last-cyclic. This argument does not work for examples where the rule of Extraposition applies optionally (such as most examples with sentential subjects), for cyclic application of the rule would merely prevent the structure from meeting the structural description of the deletion rule (as was pointed out earlier, see p. 157).

F. R. Higgins

A similar argument can be provided using the extraction analysis in the form in which it is set up by Akmajian. The underlying tree is roughly as in (68). (68)

NP

P R O ~ s1

VP

~

I resent NP

I

V~NP

S2

~

that Bill didn't tell me

is

If the rule of Extraposition applies on the S1 cycle, and if S2 can then be extracted, one obtains: (69)

0

What I resent it is that Bill didn't tell me.

This can only be avoided in any obvious way by making the rule of Extraposition apply after the extraction rule. An analogous argument can be developed using topicalized sentences, where the situation is clearer. I repeat examples (23) and (25): (70)

a.

b. (71)

a. b.

That Susan would be late John didn't think was very likely. 0 That Susan would be late John didn't think it was very likely. That you refuse even to discuss the matter I most certainly do resent. 0 That you refuse even to discuss the matter I most certainly do resent it.

It is easy to account for (70) using the extraposition analysis: Extraposition is optional and could be cyclic. If it does not apply, the noun phrase that Susan would be late can be extracted from the subject position and topicalized in the usual way. If Extraposition has applied, that Susan would be late is a plain sentence and does not meet the structural description of the rule of Topicalization. However, (71), where the verb resent normally requires extraposition of the complement sentence, is a problem, for it appears that Extraposition

On

J. Emonds's Analysis of Extraposition

181

must follow Topicalization to obtain the correct result. But, since one can topicalize noun phrases out of extraposed sentential subjects, it would appear that Topicalization must follow Extraposition, unless one is willing to countenance violations of the sentential subject constraint that later on are redeemed by the application of the rule of Extraposition: (72)

a.

b.

f>That book that John will have read is unlikely. That book it is unlikely that John will have read.

(This argument was used by Ross (1967, p. 286) to establish that the rule of Wh-movement must follow the rule of Extraposition. In a theory in which global rules are utilized it obviously would not be a particularly strong argument.) Thus, an ordering paradox has arisen. The argument with respect to both pseudo-cleft and topicalized structures depends crucially on the fact that the rule of Extraposition is obligatory in the complement structures of verbs like resent, like, and depend upon. If it could be shown that these did not involve extraposition, sentences such as (67), (69), and (71) would be irrelevant to the issue. Since the assumption that extraposition is involved leads to a mild kind of ordering paradox, it is to be hoped that that assumption is incorrect. Clearly, these structures deserve further investigation. The other important question is whether the rule of Extraposition is structure-preserving. Emonds makes an unwarranted deductive leap when he states: " . . . it is easy to see that his [Rosenbaum' s] important rule of extraposition . . . is structure-preserving [p. 92]." The most that one can state with any confidence is that the rule may be structure-preserving. Owing to the structure of English and the lack of any detailed analysis of the verb phrase, there is no obvious way of testing Emonds' s claim in English at present. However, the situation is clearer in German, in which a rule of Extraposition operates that is probably identical to the rule found in English. A strict application of Emonds's hypothesis to German leads to the conclusion that the Extraposition rule both is and is not structurepreserving. The structure-preserving hypothesis requires that the underlying position of the finite verb in German is at the end of the clause, as Emonds points out (pp. 11-12). The rule which moves the finite verb to second position only applies in root sentences, and the structurepreserving hypothesis would not permit the movement of the verb in embedded clauses, because the base rules would have to generate otherwise unmotivated nodes to accommodate it.

182

F. R. Higgins

Given a verb-final analysis, embedded sentential complements in German will be generated to the left of the verb in structures of the following kind:

s

(73)

--------------VP

NP~V

I s

(74)

s

------------VP ~ S

V

However, this sentential complement is frequently, sometimes obligatorily, extraposed. It then follows any clause-final nonfinite parts of the verb phrase in root sentences, and the whole verb complex in embedded sentences. Thus, one finds examples such as: (75)

Er hat ein Haus zu bauen versucht. he has a house to build tried 'He has tried to build a house.' b. Er hat (es) versucht, ein Haus zu bauen.

(76)

a.

a.

b.

. .. dass er ein Haus zu bauen versucht hat. that he a house to build tried has ' . . . that he has tried to build a house.'

. .. dass er (es) versucht hat, ein Haus zu bauen.

The (b) example is in each case synonymous with the (a) example but differs from it in that the infinitive phrase has been extraposed to the right. This phenomenon is quite general, and occurs with zuinfinitive phrases, dass-clauses, and indirect questions. A pronoun such as es 'it' may appear to the left of the verb in object position when the clause is extraposed out of the object position, corresponding to it in English (for details, see, for instance, Bech 1955, 1957, passim; Bierwisch 1963, especially pp. 142-45). The extraposed clause in such examples is not moved into a node which can be argued for on any independent grounds that I am aware of. That would require the generation of structures of the shape:

On J. Emonds's Analysis of Extraposition

183

s

(77)

-------------VP NP~S

I s

s

(78)

--------------VP

s~s Hence, it is, prima facie, unlikely that the sentential extraposition rules of German can be fitted into the structure-preserving framework. The same could be true of English. Until this matter can be clarified, the structure-preserving nature of the rule of Extraposition must remain in doubt. SUMMARY

It has been shown that Emonds's reanalysis of the facts that Rosenbaum accounted for by means of an extraposition rule cannot be upheld when further constructions are considered. In particular, it has been shown, first, that, within the framework which Emonds has described, the subject replacement rule which he posits cannot be a root transformation but must be a structure-preserving transformation, and, second, that generalizations will be missed if that-clauses and infinitive phrases cannot be generated as noun phrases. A further set of facts can be accounted for most simply by the readoption of a leftto-right movement rule similar to Rosenbaum's extraposition rule. Finally, the question has been raised, as a spur to further research, whether one can take it for granted that the rule of Extraposition is either cyclic or structure-preserving. APPENDIX

Sentences as Noun Phrases No attempt has been made in the main body of this paper to answer Emonds's general claim that that-clauses and infinitive phrases never show noun phrase behavior; discussion was limited to subject position, where strong arguments could be offered that override the neces-

184

F. R. Higgins

sity for a detailed refutation of this claim. However, for the sake of completeness the more important of his arguments for the claim will be reviewed here. OBJECT COMPLEMENTS

Emonds offers no argument, but gives examples of the following kind (pp. 97-99):

(79)

a. b.

They told how to build a kite to the children. They told the children how to build a kite.

(80)

a.

The man taught that books were important to his sons. The man taught his sons that books were important.

b. (81)

a. b.

1 take to fix the lamp upon myself I take it upon myself to fix the lamp.

(82)

a. b.

They expect that you co-operate of you. They expect it of you that you co-operate.

He concludes: The underlined PP's in (18) are deep-structure sisters to V, and in every case an infinitive or sentence object complement must follow them. (I make no attempt to explain in principled fashion why an it sometimes appears in object position in (18) and sometimes does not.) Thus, there is no distributional evidence that such complements occur in object NP position in deep structure [p. 99].

Even if one disregards his cavalier treatment of the it, one can produce "distributional evidence" of the required kind. There are two main kinds of constructions with both an indirect object and an object complement sentence in English, those with a preposition marking the indirect object and those without such a preposition. They display characteristic differences in syntactic behavior. Consider the effect of passivization on such constructions (I omit agents in the passive sentences to give more natural-sounding examples): (83)

a. b. C.

d.

They suggested to us that it might be better to wait. we were suggested to that it might be better to wait. It was suggested to us that it might be better to wait. That it might be better to wait was suggested to us.

On

J. Emonds's Analysis of Extraposition

(84)

a.

b. C.

d. (85)

a.

b. C.

d. (86)

a.

b. c. d.

He divulged to the committee that there hadn't been time to prepare a report. 0 The committee was divulged to that there hadn't been time to prepare a report. It was divulged to the committee that there hadn't been time to prepare a report. That there hadn't been time to prepare a report was divulged to the committee. They warned Bill that John was out to fault him. Bill was warned that John was out to fault him. 0 It was warned Bill that John was out to fault him. 0 That John was out to fault him was warned Bill. They informed the police that the car was stolen. The police were informed that the car was stolen. 0 It was informed the police that the car was stolen. 0 That the car was stolen was informed the police.

The same pattern is found with indirect questions: (87)

a.

b. C.

d. (88)

a.

b. c. d.

He explained to them how to do it. They were explained to how to do it. It was explained to them how to do it. ?How to do it was explained to them.

0

They announced to the press where they were makingfor. 0 The press was announced to where they were makingfor. It was announced to the press where they were making for. Where they were making for was announced to the press.

(89)

a. He showed them how to do it. b. They were shown how to do it. c. ??It was shown them how to do it. d. ??How to do it was shown them.

(90)

a. b. C.

d.

They told the man where they were making for. The man was told where they were making for. 0 Jt was told the man where they were making for. 0 Where they were making for was told the man.

185

186

F. R. Higgins

These examples show exactly the same patterning as is found in sentences that have undergone the rule of Dative movement (for a recent discussion, see Jackendoff and Culicover, 1971). If one assumes that the underlying order in (83), (84), (87), and (88) is V S to NP, then the fact that the sentential object can become the subject in the passive sentence and the noun phrase in the to-phrase cannot is explained in the same way as: (91)

a.

b. c.

John gave the book to Mary. ""Mary was given the book to. The book was given to Mary.

That is, the rule of Passivization moves the noun phrase which is adjacent to the verb. Conversely, if the underlying order in (85), (86), (89), and (90) at the point where passivization takes place is V NP S, the fact that the noun phrase can become the subject in the passive sentence and the complement sentence is reluctant to can be explained in the same way as: (92)

a.

b. c.

John gave Mary the book. Mary was given the book. ?The book was given Mary.

The parallelism here is not exact, since (92c) is not as bad as might be expected- but it is (92c) that needs a special explanation, not the (c) and (d) forms of (85), (86), (89), and (90). The special fact about all these object complement constructions, compared with most verbs that exhibit dative movement, is that one does not find both alternants with a given verb, and hence, if one group is to be explained by means of a dative movement rule, that rule must be obligatory within that group of verbs. Thus, if the type exemplified by suggest and explain is basic, then the rule of Dative movement is obligatory for the type exemplified by warn and tell. (Incidentally, with nearly all the verbs that require a prepositional phrase as indirect object, that indirect object is optional; with nearly all the verbs that have a pure noun phrase as indirect object, that object is obligatory. The latter class is much smaller, and promise and show are obvious exceptions among those taking that-complements to the generalization just stated, for their indirect object is optional though not marked by a preposition when present.) Otherwise, the verbs behave as expected, given only the configurations V S to NP and VP NP S at the point where the passive rule applies. Note furthermore that the following, which have indirect questions as complements, are acceptable:

187

On J. Emonds's Analysis of Extraposition

(93)

a.

b.

He explained how to do it to them. They announced where they were making for to the press.

This all suggests that a surface constraint does block unacceptable sentences of the kind adduced by Emonds, but that sentences are indeed generated in the same position as noun phrase objects. Two extraposition rules are probably involved: one extraposes Sand leaves it; the other extraposes complex noun phrases and leaves no trace (probably the complex NP shift in Ross, 1967). PREPOSITIONAL PHRASES

Emonds wishes to show that prepositions do not govern thatclauses or infinitive phrases in the way that Rosenbaum (1967, passim; see also Burt, 1971, pp. 115-118) suggested: A clear-cut indication that infinitives and sentences are not in the same category as gerunds and noun phrases with head nouns is that the latter two but not the former two appear after the traditional class of prepositions (from, at, into, toward, by, with, on account of, because of, despite, etc.) [pp. 99-100].

He tries to show that the evidence from pseudo-cleft constructions for the presence of a preposition before that-clauses and infinitive phrases in underlying structure is not cogent. He argues that a sentence such as: (94)

What we are aware of is that she is poor.

could have roughly the following underlying structure: (95)

[NP PRO [s we are aware of wh-something]] is '----v------' X [8

we are aware [ 8 that she is poor]] X

The deletion rule removes the right-hand portion marked X by identity with the portion marked X in the left-hand portion, and then other rules give (94). There is no underlying preposition before the complement sentence in the right-hand clause. In the left-hand clause, of governs a noun phrase. It turns out to be very hard to argue directly against this analysis (though I do not see how it could be upheld if the extraction analysis of pseudo-cleft sentences is correct), in part because it is so vaguely formulated. However, one may doubt whether such a loose, self-

188

F. R. Higgins

defining identity requirement in the two clauses of (95) is viable. What prevents the generation of examples such as (96) and (97)? (96)

~what we are annoyed about is at her defeat.

(97)

~what we looked at was into how to do it.

The segment X cannot be self-defining, by which I mean that the rule cannot be stated thus: "Find identical portions of maximal length in the two clauses and delete the portion in the right-hand clause as long as what remains is a constituent." On this basis, (96) and (97) could be generated from: [PRO [ we are annoyed about wh-something]] is

(98)

X

[ we are annoyed at her defeat] X

[PRO [ we looked at wh-something]] was

(99)

'---y--J

X

[ we looked into how to do it] X

In other words, the prepositions must be checked for identity in some way, even though one of them is not deleted. The checking procedure (which may, of course, be part of the correctly stated deletion rule) can only be made uniform if prepositions also appear before that-clauses and infinitive phrases. Further problems of subcategorization may arise when a given verb can take more than one kind of complement structure. For instance, there is a class of verbs that appear both with a complement sentence (often factive) preceded by a preposition and also with a plain that-clause. (In the latter case, such verbs often report the speech act which the sentence embedded below them is used to perform.) Examples are object, insist, and hint, as in: (100)

a. John objected that he had never seen them before. b. Mary insisted that she had recently talked to Bill. c. They hinted that twenty dollars was not enough.

If one follows Emonds's analysis, one might expect to be able to derive: (101)

a.

??What John objected to was that he had never seen them before.

On J. Emonds's Analysis of Extraposition

b. c.

189

??What Mary insisted on was that she had recently talked to Bill. ??What they hinted at was that twenty dollars was not enough.

It is not clear to me what the status of these sentences is; all of them seem to be relatively unacceptable. What is clear is that they are not pseudo-cleft versions of the examples of (100). [In fact, "quotative" sentences like those in (100) do not appear to permit pseudo-clefting or topicalization of the complement sentences.] The only obvious way of blocking (101) is to say that the complement sentence in (100) is not preceded by a preposition, and therefore fails the checking procedure mentioned above. Notice now that if one follows Ross's formulation of the deletion rule (see page 156), then, as long as one interprets the noun phrase mentioned in the rule as covering only pure noun phrases and not prepositional phrases as well, the checking procedure is indeed part of the rule, and, if sentences can be noun phrases, then (97) and readings of (101) that paraphrase (100) will be blocked automatically. Prepositions may appear on the surface in at least one other construction when sentential noun phrases are moved, namely in passive sentences where the deep object is a sentential noun phrase. Emonds himself provides the following example (p. 117): (102)

That Mary stay the night was insisted on by Susan.

While most examples of this kind are relatively unacceptable, they are clearly better with the preposition than without it. They are not susceptible of the same explanation as Emonds attempted for the pseudo-cleft examples. In short, while it may not be correct to posit underlying prepositions before that-clauses and infinitive phrases as freely as Rosenbaum does, it will probably add to the complexity of the grammar if one bars them altogether in this context. POSSESSIVE FORMS

Emonds here claims that at some point in the transformational derivation a rule must insert 's in the context [NP [NP X _ ] Y] . . . But according to this rule, gerunds should be noun phrases and sentences and infinitives should not be, since the initial (subject) NP in gerunds but not in infinitives and sentences can be followed by an 's suffix [pp. 110-111].

190

F. R. Higgins

Even if the rule is correct, the argument has little cogency, for one would not expect the 'S-insertion rule to insert 'son the subject of a lower sentence. At most it would, if correct, suggest that gerunds contain no sentence node and it still fails to account for various differences between the possessive form in gerunds and in noun phrases with a noun head, as, for instance, the fact that the possessive form is often optional in gerunds. Note that the same argument proves that relative clauses cannot have the structure (103). (103)

No insertion of 's occurs in relative structures. If any good evidence can be found in favor of this popular analysis, then Emonds's argument cannot be correct. In any case, since I have already shown that sentential subjects are noun phrases, and since they do not behave any differently from any other sentential forms with respect to Emonds' s 'S-insertion rule, it is apparent that the argument carries no weight. Emends presents a further argument about possessive forms (pp. 102-103), but I do not understand it and cannot comment on it. CONJUNCTION

The argument runs: If sentences and infinitives are not NP' s, they should not conjoin freely with NP's. It was pointed out by Gleitman (1965) that this is the case. On the other hand, gerunds and NP's with head nouns can be conjoined [p. 111].

The force of this argument is not clear without much more investigation. If gerunds, infinitive phrases and that-clauses are different in semantic function, it is not surprising that they cannot be conjoined. From the fact that infinitive phrases and that-clauses also cannot in general be conjoined, does one conclude that they are not both sentences? If that-clauses are sentences, and gerunds are noun phrases, what then are infinitive phrases? Furthermore, one must choose the gerunds and noun phrases with noun heads carefully to ensure smooth conjunction. It is easy enough to find sylleptic examples: (104)

a.

b. c.

l'>You mustn't think of your wife or talking. I acknowledged my failings and having been at the meeting. l'>He regretted his lost youth and having invited me. I'>

On

J. Emonds's Analysis of Extraposition

191

From the fact that Miss Bolo went home in a flood of tears and a Sedan chair is sylleptic, does one conclude that a flood of tears and a Sedan chair cannot both be noun phrases? Categorial identity may be a necessary condition for conjunction, but it is rarely a sufficient one. CLEFT AND PSEUDO-CLEFT CONSTRUCTIONS

Emonds (pp. 112-114, 127-132) wishes to maintain "that the focus position of the pseudo-cleft construction is not a diagnostic context for the category NP," and he tries to show that Smay also be found in the focus position of the pseudo-cleft construction. However, it is only an S of a special kind that can appear there: one that is associated by the coreference rule with a noun phrase in the way described earlier. In a sense, this position is virtually impregnable, since there is no direct way of testing it. I hope to have demonstrated earlier that it cannot be maintained for sentences in subject position. Presumably, one can therefore discard the hypothesis for all sentences and claim that it is only sentences that are also noun phrases that can appear in the focus position in pseudo-cleft sentences. Many other explanations can be devised to account for the fact that, say, relative clauses and consecutive clauses cannot appear in the focus position in pseudo-cleft sentences, and discussion of the matter is futile at this point, in the absence of a detailed description of the syntax of pseudo-cleft sentences. PASSIVE CONSTRUCTIONS

Emonds (pp. 115-118) points out, quite correctly, that application of the passive transformation gives unacceptable results where the deep subject is a that-clause or infinitive phrase, and concludes: This means that the agent postposing rule requires an ad hoc condition in Rosenbaum's framework. This condition simply states that sentences and infinitives do not act like NP's as far as agent postposing is concerned, thus reflecting directly my contention that these constructions are never NP's [p. 117].

There is an alternative way of looking at the matter. If one moved the sentential subject into the by-phrase, the preposition by would be deleted by the preposition deletion rule, leaving nothing to mark the structure as the agent. Once one moves the sentence into the by-phrase one is bound to end up with an unacceptable sentence whatever one does. It is not clear how to state this within the present theory, but it is as adequate as Emonds' s approach and avoids the problem that he attempts to deal with in the footnote to pages

192

F. R. Higgins

117-118, in which, as far as I understand him, he is forced to set up an arbitrary condition to bar the generation of sentences such as: (105)

a.

b.

(>Embarrassment would be caused by it for John to arrive. (>The necessity of discipline is shown by it that the children are always late.

(Recall that for him, the it is in subject position in the deep structure and the subject clause in extraposed position.)

Ross's Output Condition Thus, the facts that Emonds adduces in support of his claim that

that-clauses and infinitive phrases are never noun phrases can, on the whole, be explained without that assumption, and sometimes more cogently without it. One is still left with the problem of accounting for the unacceptability of the various constructions which led Ross (1967) to set up the output condition: "Grammatical sentences containing an internal NP which exhaustively dominates S are unacceptable [p. 57] ." Ross defines an internal noun phrase as follows: " . . . an NP is internal to a sentence if it is both preceded and followed by non-null parts of that sentence [p. 105, Footnote 8]." Emonds objects that this does not account for the fact that free relative clauses, which do have the structure specified in Ross' s (3.27), are not at all unacceptable in internal position. If one revises the constraint so that it rejects internal noun phrases that do not have a noun as head, the restriction can perhaps be understood as not excluding internal free relative clauses, since they presumably do have a noun head, even if it is only [PRO]. It also may permit, correctly, internal indirect questions, as in: (106)

a. John told us that what it was you had said to annoy

them was unknown. b. They explained what it was they were working on to the students. It is possible that such indirect questions also have a noun head of some kind. This is perhaps indicated by the fact that indirect questions in what and who can take restrictive relative clauses in that, as in: (107)

a. John was asking me what (it was) you said that

annoyed them.

On J. Emonds's Analysis of Extraposition

b.

193

I don't remember who I was talking to that told me that.

In contrast to this, that-clauses can only be antecedents as a whole, as in sentential relative constructions, and infinitive phrases do not appear to take relative clauses at all. This cannot be considered a convincing argument for the presence of a head noun, for it would depend on the assumption that the class of clauses that can take restrictive relative clauses and the class of clauses with noun heads are identical. This assumption would be incorrect, since there are exceptions: relatively empty nouns with complement sentences, like fact, and question, are extremely reluctant to take relative clauses in that, as are gerunds. Perhaps one might claim that the presence of a noun head was a necessary condition for the presence of a relative clause in that, but I am unable to show this at present. The postulated head might help to explain the relationship discussed by Baker (n.d.) between indirect questions and "concealed" questions, as in Baker's examples (6.3) and (6.7) (pp. 83-84): (108)

a.

b.

Susan found out the place where the meeting was to be held. Susan found out where the meeting was to be held.

If there is an underlying noun head in indirect question noun phrases, the fact that that head can be filled by a genuine noun head becomes a little less surprising. It is even less easy to argue for the presence of a noun head in indirect questions in when, why, how and so on, but such a position can perhaps be upheld and is worth investigating. At any rate, there is clearly no evidence for the presence of a head noun in the noun phrase dominating that-clauses and infinitive phrases. A gerund phrase, if my vague suggestion is correct, does contain a head noun and contains no sentence node. Ross's condition (3.27) in fact needs no modification if construed correctly; for there can be no head noun in a noun phrase if it "exhaustively dominates" a sentence. It does not seem unreasonable to assume that the lack of a clearly nominal element in a noun phrase might cause difficulties for perceptual strategies, and therefore the suggestion that the unacceptability of the relevant sentences is due to a performance factor is not totally devoid of explanatory force. One stubborn problem remains: the characterization of "internal" in Ross's constraint. Ross has to define it as he does to prevent the constraint from blocking sentences like (109):

194

F. R. Higgins

John said that he had come.

(109)

where there is no material to the right of the that-clause. But now one is left with no way of blocking (110a) and (Illa): (110)

a.

0

a. b.

0

b.

How easy is to please John? How easy is it to please John? How likely is that John will come? How likely is it that John will come?

The ungrammaticality of these examples is predicted by Emonds's theory, since he imposes the constraint, roughly, that at most one root transformation may apply in a given sentence. In the (a) forms, two root transformations have applied, on his analysis: the rules of Subject-auxiliary inversion and Subject replacement. Although this explanation is now untenable, I suspect that the ungrammaticality of the (a) examples does indeed have to do with the fact that a root transformation has applied to constituents that are not inherently characterized (by the possession of a noun head) as noun phrases, and with the fact that they are moved by it to a position that is not inherently a noun phrase position. Somehow, they lose their identity as noun phrases in this process. It is not possible to generalize Ross's constraint to cover such cases, but if the hindrance of a perceptual strategy related to the identification of noun phrases lies behind that constraint, it may lie behind the explanation of (110a) and (Illa) as well-albeit, only if one acknowledges the essential correctness of the distinction between root and structure-preserving transformations. It would probably be fruitful to investigate the problem further along these lines.

REFERENCES Akmajian, A. (1970a) On deriving cleft sentences from pseudo-cleft sentences. Linguistic Inquiry, 1970, 1, 149-168. Akmajian, A. (1970b) Aspects of the grammar offocus in English. Unpublished doctoral dissertation, M.l.T., 1970. Baker, C. L. Indirect questions in English. Unpublished manuscript, n.d. Baker, C. L., & Brame, M. K. 'Global rules': A rejoinder. Language, 1972, 48, 51-75. Bech, G. Studien uber das deutsche Verbum infinitum. Det kongelige danske videnskabemes selskab, Historisk-filologiske Meddelelser. Bind 35, No. 2, Copenhagen: Ejnar Munksgaard, 1955. Bech, G. Studien uber das deutsche Verbum infinitum. Det kongelige danske videnskabemes selskab. Historisk-filologiske Meddelelser, Bind 36, No. 6, Copenhagen: Ejnar Munksgaard, 1957.

On J. Emonds's Analysis of Extraposition

195

Bierwisch, M. Grammatik des deutschen Verbs. Studia Grammatica II, Berlin: Akademie-Verlag, 1963. Bresnan, J. On complementizers: Toward a syntactic theory of complement types. Foundations of Language, 1970, 6, 297-321. Bresnan, J. On 'A non-source for comparatives.' Linguistic Inquiry, 1971, 2, 117-124. Burt, M. From deep to surface structure. New York: Harper and Row, 1971. Chomsky, A. N. (1970a) Remarks on nominalization. In R. A. Jacobs, & P. S. Rosenbaum (Eds.), Readings in English transformational grammar. Waltham: Ginn and Co., 1970. Chomsky, A. N. (1970b) Some empirical issues in the theory of transformational grammar. Unpublished manuscript, M.I.T., 1970. Chomsky, A. N. Conditions on Transformations. Bloomington: Indiana University Linguistics Club, 1971. Curme, G. 0. A grammar of the English language. Vol. 3. Syntax. Boston: D. C. Heath, 1931. Doherty, P., & Schwartz, A. The syntax of the compared adjective in English. Language, 1967,43,903-936. Emonds, J. Root and Structure-Preserving Transformations. Unpublished doctoral dissertation, M.I.T., 1970. Emonds, J. Appendix. Root and structure-preserving transformations. Bloomington: Indiana University Linguistics Club, n.d. Gilbert, K. F. Which-sentential relatives and as-clauses. Unpublished manuscript, M.1.T., 1967. Hornby, A. S. A guide to patterns and usage in English. London: Oxford University Press, 1954. Jackendoff, R. Some rules of semantic interpretation in English. Unpublished doctoral dissertation, M.I.T., 1969. Jackendoff, R., & Culicover, P. A reconsideration of dative movements. Foundations of Language, 7, 397-412. Jespersen, 0. A modern English grammar on historical principles. Part 3. London: G. Allen and Unwin Ltd., 1928. Jespersen, 0. A modern English grammar on historical principles. Part 5. London: G. Allen and Unwin Ltd., 1946. Lakoff, G. Deep and Surface Grammar. Bloomington: Indiana University Linguistics Club, 1968. Langacker, R. On pronominalization and the chain of command. In D. A. Reihe!, & S. A. Schane (Eds.), Modern studies in English. Englewood Cliffs, N. J.: PrenticeHall, 1969. Perlmutter, D. Deep and surface structure constraints in syntax. New York: Holt, Rinehart and Winston, 1971. Postal, P. Cross-over phenomena. New York: Holt, Rinehart and Winston, 1971. Poutsma, H. A Grammar of Late Modern English, Part 2, Section 2. Groningen: Noordhoff, 1926. Rosenbaum, P. The grammar of English predicate complement constructions. Cambridge, Massachusetts: M.1.T. Press, 1967. Ross, J. R. Constraints on variables in syntax. Unpublished doctoral dissertation, M.1.T., 1967. Ross, J. R. Guess who? In Papers from the Fifth Regional Meeting, Chicago Linguistic Society, 1969, 252-286. Ross, J. R. Act. In D. Davidson, & G. Harman (Eds.), Semantics of natural language, Dordrecht, Holland: D. Reidel, 1972.

LANGUAGE LATERALIZATION AND GRAMMARS DIANA VAN LANCKER University of California, Los Angeles

The homogeneity of the ideal speaker-hearer's competence for language is currently in question. Observations in syntactic research, psycholinguistics and sociolinguistic studies suggest that there are nonuniformities in language competence (Gleitman and Gleitman, 1970; Greenberg, 1969; Labov, 1969; Quirk, 1966), both across speakers and within speakers. I would like to point out another kind of heterogeneity in language behavior. I will describe a dimension along which identifiable variations might be said to fall, present clinical and neurophysiological evidence for the dimension, and relate its properties to linguistic theory. The evidence to be presented suggests that in language behavior, we can differentiate creative or propositional use from automatic use of language. Propositional language behavior includes all newly created, original, novel sentences; automatic language encompasses conventional greetings, overused and overlearned expressions (such as Be careful and First things first), pause fillers such as you know and well, certain idioms, swearing and other emotional language, perhaps stereotyped questions and answers, commands, and so on. In fact, it is likely that instead of the disjunctive sets propositional and automatic, there is a dimension along which subsets of language use fall. Further research will determine such refinements. In this paper I will give evidence from clinical reports and studies for the existence of an automatic versus propositional dimension in language

197

198

Diana Van Lancker

behavior, and I will suggest possibilities for establishing this claim for normal speech. The last part of the paper considers features of language inhering in notions propositionality and automaticity and their implications for grammars. It is important to note that the automatic-propositional language dimension has a physiological basis; namely, lateralization in brain function. It has long been accepted that language is normally lateralized in the left half of the human brain. This, in actual fact, is only partly true. It is true that propositional language is represented and processed in the left hemisphere; however, more automatic types of language behavior are stored and processed in the right hemisphere as well. Evidence strongly suggests, as Hughlings Jackson (1878; 1879) maintained a century ago, that propositional language is lateralized to the left, but that automatic uses are bilaterally represented in the brain. The well-known functional asymmetry for language, then, is not true for all of language ability. Clearly a firmer grasp on these details of language processing will lead to insights about human language. I mentioned that Hughlings Jackson was the first aphasiologist to distinguish propositional from automatic language. Many researchers and clinicians in the hundred years since his time have provided confirmation of his theories. For example, the French aphasiologist, Alajouanine (1956) observed that for the expressive aphasic (the aphasic who shows difficulty in the production of language), "situations leading to emotional, expletive, interjectional, automatic [and] habitual language are always more effective [in producing speech] than those requiring volitional or propositional speech." Furthermore, he notes that in agrammatical aphasia (or deviation from syntactic constraints), speech "no longer under volitional control," such as "ready-made formulae of salutation, exclamations, etc." is grammatical. Alajouanine observes that "the agrammatism [of agrammatical aphasia] does not extend to automatic language." Descriptions of how these types of speech are selectively affected in disease can be found throughout the aphasia literature. Some writers, Goldstein (1948) and Head (1926), for example, distinguish automatic or "highly structured" speech from what they call "volitional" speech. It is clear that they are referring to essentially the same impairment. Similarly, Bay (1965) reports that "mainly propositional" language is impaired in expressive aphasia, leaving intact emotional reactions (swearing), stereotyped reactions (salutations and other conventional phrases), and automatisms (such as lists and serials). Lenneberg (1967) notes that "expressive disorders of Ian-

Language Lateralization and Grammars

199

guage are less severe in emotional discourse than in propositional discourse." And Wepman, Bock, Jones, and van Pelt (1956) described anomia, or object-naming difficulty, as the almost exclusive use of more general, more frequent words in the language- "overlearned" words and phrases which are "virtually automatic" in the aphasic' s speech. Aphasia commonly results from injury to the left hemisphere. Often only automatic language remains in the victim. Language is seldom affected when the right hemisphere is injured or entirely removed. From these facts, we can conclude that automatic language can reside in both hemispheres. Espir and Rose (1970), in their recent handbook on language pathology, provide a list of what they call "nonintellectual" speech, typically retained in aphasia patients with left hemisphere injury: These are emotional utterances and expostulations; yes, no, and other "first-learned primitives"; recurrent utterances; songs and poems; serial speech-such as the alphabet, and days of the week; and social gestures of speech, such as How do you do. These linguistic abilities, they say, may well be retained "by virtue of the contribution of the nondominant [or right] hemisphere." Other clinical observations provide more evidence that these types of language are somehow different, and somehow involve the right hemisphere. Only a few patients have undergone left hemispherectomy, or removal of the left language hemisphere for treatment of malignancy; these have been capable of some speech soon after surgery. One case, a 14-year-old, said "mother," "father," and "I don't know" after the 16th postoperative day (Hillier, 1954). The one wellreported adult case was able very soon to swear, and to utter expletives and short emotional phrases; after ten weeks he could repeat words and give short answers to questions; in the fifth postoperative month he began recalling and singing whole familiar songs (e.g., Home on the Range) and hymns (Smith, 1966). For the case reports involving injury or removal of the left, dominant hemisphere, we can assume that the right hemisphere is in part or fully involved in whatever language remains. Studies on epileptics with speech disturbances as part of their seizure patterns are more difficult to interpret, but some role of the right hemisphere, as well as of the left, in this form of automatic language seems certain. The evidence is that ictal speech automatisms (defined as unremembered but linguistically correct expressions) occur "slightly more often with seizures originating in the [right] than in the [left] hemisphere." On the other hand, dysphasic utterances (for which the epileptic patient is conscious but unable to express himself grammatically) are associ-

200

Diana Van Lancker

ated almost exclusively with left hemisphere seizures (Falconer, 1967; Serafetinides and Falconer, 1963). Thus, on the basis of clinical observations of aphasic, hemispherectomy, epileptic patients and from other sources not detailed here (including experiments on split-brain humans and sodium amytal tests for cerebral dominance), it seems reasonable to conclude that the left hemisphere is specialized for propositional language, while other modes of language can be processed and stored elsewhere in the brain. The subsets need to be identified and characterized in detail. Further study of clinical correlations in performance data may suggest how these modes from propositional to automatic are mutually related, and what their differences imply for linguistic theory. For example, palilalia, the repetition two to twenty times of a final word, phrase, or sentence, is said to occur in propositional speech and in reply to questions; it occurs sometimes in emotional speech, oaths and interjections; but not in automatic recital such as prayers, naming days of the week, counting, and reading aloud (Critchley, 1970). With a great deal more such data, it might indeed be possible to identify and describe other interesting variations in language processing. There is some evidence for the propositional-automatic dimension in normal language. Studies on pauses, or hesitation phenomena, indicate different processing for newly-created, propositional use of language as compared with automatic, overleamed, read, and repeated speech (Goldman-Eisler, 1968). One study using delayed auditory feedback suggested that normal speech is processed differently from two other modes, swearing and epileptic ictal speech (Chase, Cullen, Niedermeyer, Stark, and Blumer, 1967). Labov's (1970) finding that casual and emotional speech modes are distinguished from formal modes by selective linguistic variables is a model example of what future studies on normal linguistic behavior might achieve: the identification and formal characterization of modes and styles across spectra within normal language use. It is possible that speech-error data would tabulate differently for various such modes and styles. One might investigate suprasegmentals with systematic variability in mind. In experiments now underway in the UCLA phonetics laboratory, I am utilizing the phenomenon of language lateralization to test hypotheses about automatic language behavior. This is done through a listening task, called dichotic listening. The subject hears two different sounds, one at each ear, simultaneously, and is asked to identify one or both sounds. When the stimuli are linguistic, he generally

Language Lateralization and Grammars

201

gets more correct in his right ear, because those stimuli are processed in the left or language hemisphere. This is a reliable finding in psycholinguistic research. My present interest is to show that more automatic-type stimuli do not show the right ear effect of propositional language. For linguistic theory, we want to ask what these data from abnormal and normal performance imply about language behavior. What properties do nonpropositional modes of language have? Are those properties representative of general human language behavior? How are these observations to be incorporated into grammatical description? We must face the performance-competence issue here, and consider at what level to describe observations on automatic language. First, the propositionality-automaticity dimension pertains to language use. Any automatic expression can also occur propositionally. For example, the same phrase Good morning can occur in different modes: as the automatic greeting; in perhaps semi-propositional use as in She didn't say "Good morning" to me; or propositionally, in It's a good morning to play tennis. The reality of these various modes is reflected in puns, word games, and the shock effect of minor substitutions in over-learned phrases, such as A rolling stone gathers

momentum.

An interesting property of automatic expressions is that they seem to be stored whole; they do not have dissoluble structure. To use Weinreich's (1969) term, they are transformationally defective. To structurally analyze, say, an idiom or other frozen expression -to vary it in some way, is to use it propositionally. We might also note that automatic language is low in semantic content. Meaning, in whatever way considered (whether philosophic or feature analysis) plays a different, lesser role for automatic use of language. People often do not know meanings of key words in cliches they use, and do not care. Even more curiously, automatic language is used in close association with situation context. Some instances one might even call highly stimulus bound, such as saying Excuse me to a piece of furniture. This, of course, is an extreme example; but I believe that everyday, domestic language contains a good deal of degrees of automaticity when the notion of stimulus is broadly defined to include internal states. We might present these data about automaticity as a principled basis for excluding emotional speech, over-learned phrases, pause filler, and troublesome idioms from consideration in grammatical analysis. We could even strengthen our position by noting that this is language-like behavior; it is not lateralized, as is true language, but

202

Diana Van Lancker

is present also in the nonlanguage hemisphere of the brain. Further research directed at ascertaining in detail the language-like abilities of the right hemisphere, then, could help us subtract (from performance data) those peripheral subsets, leading to a finer, clearer definition of language. Alternatively, we might agree that the propositional-automatic dimension is meaningful at the level of ideal or systematic performance, and the proper subject matter for a performance model. In this view, characteristics of automaticity, such as nonnovel expression, indissoluble structure, low-meaning content, and high situation dependency, might be seen as performance properties, not grist for the competence mill. Automaticity could be represented as a type of quantificational rule in a contingency grammar, comparable to rules for registers or style-shifting (Houston, 1970). However, it will be less easy to select out and set aside automatic use of language, or relegate its features to a performance model, if it turns out that automaticity is inherent and pervasive in human language. And it seems reasonable to assume that instead of whole, unanalyzable formatives on the one hand, and fully analyzable structures on the other, language ability recognizes degrees of structural cohesion. This notion of degrees of automaticity reminds one of other observations and insights about language which have not yet found a comfortable place in grammatical description. In investigating automatic language, it might be interesting, for example, to reevaluate the literature on collocation, or transition probabilities across successive items. Another well-researched factor, frequency of occurrence, prominent in aphasia studies, psycholinguistics, and in perceptual studies, seems obviously related to automaticity. Furthermore, some types of automatic language are highly familiar; the meaning of the complex concept "familiarity" in language behavior might be worth investigating in the context of a propositional-automatic dimension. It is interesting to recall that in his transformational grammar solution for compound nouns and idioms, Weinreich (1969) proposed building in the notion "degrees of familiarity." Still further, it seems likely that grammaticality judgments might correlate in interesting ways with a propositional-automatic dimension; in fact, recent studies have begun to investigate how the factor "familiarity" interacts with grammaticality and meaningfulness in subjects' ratings of sentences (Danks and Glucksberg, 1970). Finally, all these aspects (e.g., automaticity, frequency of occurrence, and familiarity) must crucially intersect for language acquisition. In Bever and Langendoen's (1971) view, systematic behavioral fac-

Language Lateralization and Grammars

203

tors in perception and production determine language competence to some extent. Competence is generally said to underlie language behavior, and to fully specify human linguistic ability. To exclude characteristic, systematic behavioral properties in language from consideration in a model of competence is to underdescribe human language. I am suggesting here that if normal linguistic behavior is interspersed with noncreative and less creative as well as less and more novel linguistic expression, then an adequate model should reflect this kind of heterogeneity at a significant level of description. Such a decision depends on how to assess automaticity in language behavior, and how linguists define levels in their descriptive models.

REFERENCES Alajouanine, T. Verbal realization in aphasia. Brain, 1956, 79, 1-28. Bay, E. Principles of classification and their influence on our concepts of aphasia. In A. V. S. de Heuck, & M. O'Connor (Eds.), CIBA foundation symposium: Disorders of language. London; J. & A. Churchill, 1964. Bever, T., & Langendoen, D. T. The dynamic model of the evolution of language. Linguistic Inquiry, 1971, 2, 433-464. Chase, R. A., Cullen, K. J., Jr., Niedermeyer, E. F. L., Stark, R. E., & Blumer, D. P. Ictal speech automatisms and swearing. Journal of Nervous and Mental Disease, 1967, 144, 406-420. Critchley, M. Aphasiology and other aspects of language. London; Arnold, 1970. Danks, J., & Glucksbe1,.g, S. Psychological scaling of linguistic properties. Language and Speech, 1970, 13(2), 118-139. Espir, M., & Rose, F. C. The basic neurology of speech. Oxford: Blackwall, 1970. Falconer, J. A. Brain mechanisms suggested by neurophysiologic studies. In C. H. Millikan, & F. L. Darley (Eds.), Brain mechanisms underlying speech and language. New York; Grune and Stratton, 1967. Gleitman, L., & Gleitman, H. Phrase and paraphrase. New York; Norton, 1970. Goldman-Eisler, F. Psycholinguistics: Experiments in spontaneous speech. London and New York: Academic Press, 1968. Goldstein, K. Language and language disturbances. New York; Grune and Stratton, 1948. Greenberg, S. R. Experimental study of certain intonation contrasts in American English. UCLA Working Papers in Phonetics, 1969, 13. Head, H. Aphasia and kindred disorders of speech. Cambridge, England: Cambridge University Press, 1926. Hillier, W. E. Total left hemispherectomy for malignant glioma. Neurology, 1954, 4, 718-721. Houston, S. Competence and performance in child black English. Language Sciences, 1970, 12, 9-14. Jackson, J. H. On affections of speech from disease of the brain. Brain, 1887, I, 304330 Jackson, J. H. On affections of speech from disease of the brain. Brain, 1889, 2, 203222.

204

Diana Van Lancker

Labov, W. Contraction, deletion and inherent variability of the English copula. Language, 1969,45, 715-776. Labov, W. The study of language in its social context. Studium Generale, 1970, 23, 30-87. Lenneberg, E. Biological foundations of language. New York: Wiley and Sons, 1967. Quirk, R. Acceptability in language. Proceedings of the University of Newcastle upon Tyne Philosophical Society, 1966, I, 79-92. Serafetinides, E. and Falconer, J. Speech disturbances in temporal lobe seizures: A study in 100 epileptic patients submitted to anterior temporal lobectomy. Brain, 1963, 86, 463-473. Smith, A. Speech and other functions after left (dominant) hemisphere hemispherectomy. Journal of Neurology, Neurosurgery and Psychiatry, 1966, 29, 467-471. Weinreich, U. Problems in the analysis of idioms. In J. Puhvel (Ed.), Substance and structure of language. Los Angeles: University of California Press, 1969. Wepman, J. M., Bock, R. D., Jones, L. V., & van Pelt, D. Psycholinguistic study of aphasia: A revision of the concept of anomia. Journal of Speech and Hearing Disorders, 1956, 21, 468-477.

GET JOHN KIMBALL University of California, Santa Cruz

THE PROBLEM

The verb get seems to carry diverse meanings in its occurrences in various environments in surface structure. Consider the following uses of get: (1)

a.

b. C.

d.

get a book get down get fired get tired

e. f.

g.

get to go to the movies get paint on one's face get the car fixed

When we examine these locutions looking for paraphrases for get, it seems that we have a different verb in each case. Thus, (la) can mean to receive a book, or to obtain a book, while (lb) means to move to a certain position. (le) means to be fired, and (ld) to become tired. Thus, in the first four cases of get we have four different synonyms, receive, move, be, and become. Furthermore, the synonyms do not freely occur in all environments as does get; so we do not have 0 re-

ceive down, 0 receive fired, 0 receive fired, 0 move fired, 0 move tired, etc. It appears that we are forced to the conclusion that the different occurrences of get are homonyms. The logic of the argument is that words synonymous to get in different contexts are not themselves synonymous, and synonymity is a test for sameness of meaning. The result outlined above is not only intuitively repugnant, but

205

206

John Kimball

also it is demonstrably wrong. The conclusion was forced on us by the test used to determine meaning-namely synonymy; yet this test seems inimicable as an instrument to investigate meaning. If we can show that the different examples of get above in fact come from the same source syntactically, i.e., that they are not homonyms, then we will be able to examine in some detail where it is that the test for meaning by substitution of synonyms went wrong above. The first section below will be devoted to constructing a common underlying source for all occurrences of get. The following section will deal with a related lexical item, the 's got of Carol's got a book. In the final section I will return to the more general question of the test for meaning by synonymy, and its relation to compositional theories of meaning.

GET The first step in constructing an underlying source for get consists in noticing that it is possible to paraphrase get with come to have or come to be, with a few exceptions to be discussed below. (2)

a. b. c. d. e.

= get a book = get down = get fired = get tired get paint on one's face=

'come to have a book' 'come to be down' 'come to be fired' 'come to be tired' come to have paint on one's face

In these cases the paraphrase is straightforward. We can treat get as the lexicalization of come to have or come to be after an operation such as Predicate Raising (McCawley, 1968) has applied. Thus, a deep structure for Carol got a book would look like (3a), which would subsequently be transformed into (3b) by Predicate Raising, and finally into (3c) by lexicalization.

(3) a.

s

b.

~

I

I

~ come

NP

I Carol

VP

~

V

NP

have

book

I

~

NP

VP

NP

S

I

I Carol

VP

~

V

I

have + come

NP

I

book

207

Get

s

c.

~

NP

VP

I/'--_

Carol

V

NP

get

book

I

I

Notice that from an underlying structure like for Carol to be tired came (about), we will have after Predicate Raising Carol became tired, which is a viable surface structure. The synonymy of Carol became tired and Carol got tired is thus explained by the proposed analysis of get. It is now possible to provide an account of the origin of the get passive. Lakoff (1971) argues that the get passive differs from the be passive in underlying structure. The origin of John was fired by the boss would on her account appear as (4a), while that of John got fired by the boss would be (4b). (4)

a.

S

I

N~P

I~ V NP

I

John

~ be NP VP

I"---.

the boss

r------._

V

I fire

s

b.

~ NP VP

NP

I John

I

get

I

S

~

the boss fire John

In favor of the difference in underlying structures she argues that the speaker attitude toward get and be passives is different. So the difference between This program has been/ gotten prerecorded is claimed to lie in the speaker expressing that the prerecording was detrimental in the get but not the be case. Whatever the difference is in terms of speaker attitude, however, it is highly dialect variable, and does not seem to hold for all cases of get versus be passives. And even if the difference in attitude were firm, it could not be taken as an indication of difference in underlying structure. By way of showing that the get passive contains the real verb get in underlying structure, Lakoff attempts to establish that both gets have the same ambiguity between agentive and nonagentive readings. Thus, a sentence like (3c) above-Carol got a book-can mean

208

John Kimball

either she bought one (agentive) or she received one (nonagentive). As Lakoff points out, adding for her mother versus from her mother to (3c) brings out each meaning unambiguously. The factual claim made is that the same ambiguity holds for sentences which are get passives. The only evidence adduced for this claim is the somewhat contrived After a lot of pushing in commercials, the claim of Zotz, the miracle detergent, finally got believed. Concerning such a sentence it should be observed that although there is an "idea of effort" expressed, there is still no agent, and so the agentive-nonagentive ambiguity is lacking. For standard get passives, such as Joe got arrested, there is no ambiguity. There are agentive sentences of the form Joe got himself arrested which will be discussed below. However, as the factual claim concerning ambiguities in the get passive is incorrect, no support for Lakoff's hypothesis is provided. In fact, the lack of ambiguity where Lakoff thinks there should be one is evidence against her hypothesis. Likewise, Lakoff argues that the three-way ambiguity in a sentence like (5), which she terms a "passive get-sentence," is reducible to the two senses of the "real verb get." (5)

John got his dishes washed.

It is important to notice first that a sentence such as (5) is not a passive, and thus in particular not a get passive, although it does contain an embedded passivized sentence. Thus, the existence of sentences like (5) does not bear on Lakoff's proposal concerning the origin of get passives. However, it is relevant to see how the proposal made above concerning the origin of get handles a sentence like (5). The three senses of (5) are paraphrased in (6a-c). (6)

a. John finished his dishes. b. John had someone wash his dishes. c. Someone washed John's dishes on him.

(6a) and (6c) are both senses of the noninchoative reading of sentence (7). (7)

John had his dishes washed.

A sentence like (5) is derived in the manner outlined above from the embedding of (7) under the inchoative come. The causative reading of (5), partially paraphrased in (6b), presents

a somewhat more complex case. It would be semantically incorrect to represent this reading of (5) as the inchoative (of 6b), as shown in (8).

209

Get

(8)

S ~

NP

VP

I S

I

come

~ NP VP

IV~ NP NP I I I cause someone S

John

~

NP I someone

VP

.,......---___

V

I wash

NP

~

John's dishes

This would be to represent the causative sense of (5) as the inchoative of a causative, whereas semantically it is the causative of an inchoative. That is, the correct underlying representation would be (9)

s

-----------

NP

John I

VP

---------------NP

V

I cause

I

--------S

NP I S

VP

~

NP

I someone

I

come

VP

~

V

I

wash

NP

~ John's dishes

To see that this is the right semantic representation, observe what happens in the negation of (5), considering the causative reading. (10)

a. John didn't get his dishes washed. b. John didn't get Susan to wash his dishes.

It is an established principle that negation applies to the highest predicate. (Thus, She didn't eat quickly means It was not quickly

210

John Kimball

that she ate, and not O It was quickly that she didn't eat.) Observe

that (10a) means that John didn't cause his dishes to be washed. This is more clearly seen in the unambiguously causative (10b), which means that John failed to cause something, not that it didn't come about that John caused Susan to wash his dishes. A deep structure like (9) is also appropriate for sentences like Tom got himself arrested, which would come from something roughly like Tom caused it to come about that someone arrest Tom. The derivation involved has an added bit of complexity over that for the noncausative sentences. Let us consider the derivation of the causative sense of (5) from (9). After Passive has applied on the lower cycle, and two steps of Predicate Raising on succeeding cycles, the resulting tree is (11).

s

(11)

---------~=---I I

~

NP

I

John

VP

V

be+ come+ cause

NP

~ his dishes

VP

washed

The lexicalization process must allow be + come + cause as well as be + come to be realized as get. The zero causative morphology on

inchoatives has precedent elsewhere in the grammar, however. Compare the simple inchoative The metal hardened with the causative form George hardened the metal. Here again the causative of the inchoative carries no extra morphology. Of the initial list of get sentences (la)-(lg) presented at the outset of this paper, underlying structures for all have been proposed, except for (le): get to go to the movies. The meaning involved is come to have permission. In this respect, get in sentences like (le) is like the inchoative of the modal may of sentences like Tom may be excused in the have permission sense. Thus, we may represent the underlying structure of (le) as (12). There is an asymmetry in the modals with respect to lexicalization by get. Thus, it has been shown that may + come is lexicalized, but there is no corresponding lexicalization of must + come by get. That is, no get sentence corresponds to It came about that Tom had to

jump out of the hole. It is appropriate to conclude this section with the observation that although it has been shown that all occurrences of get are derived from inchoatives, most generally come to have or come to be but

211

Get

------s

(12)

NP

VP

I

I S

come

---------I /'---

NP

someonei

VP

V

I may

NP

I

S

,~

~

NP

VP

I~ V PP

someone j

go

to the movies

also cause to come to be and "'come to may, the converse fails. That is, not all instances of, say come to be are lexicalizable by get. The sentences below are a few such examples. (13)

a.

George came to be president. "'George got president.

b.

George came to be a fast runner. "'George got a fast runner.

c.

How did Carol come to be eating pizza? "'How did Carol get eating pizza?

I have no proposal concerning how one ought to catalog or treat such cases. Evidently Predicate Raising is blocked in such sentences, so the constraint perhaps lies on this operation rather than on lexicalization by get. 'S GOT

Fillmore (1970 La Jolla Syntax Conference) pointed out a difference between sentences like Pat has a baby and Pat had (bore) a baby. The first in which have means possession is paraphrasable by Pat's got a baby; the second does not paraphrase as the corresponding O Pat had got a baby. The difference, Fillmore hypothesized, lies in the distinction between active and stative have, where only the latter is paraphrasable with 's got. This difference is particularly

212

John Kimball

visible in pairs of sentences like (14a) and (14b), in which only the latter paraphrases with 's got. (14)

a.

b. (15)

a.

b.

The play has them (the actors) roll in the aisles. The play has them (the audience) rolling in the aisles. (>The play's got them roll in the aisles. The play's got them rolling in the aisles.

An interesting peculiarity of 's got is that it never shows up as a paraphrase for past tense have, even though that have may be stative. Thus, (16a) has no corresponding (166).

a. b.

(16)

Carol had a book. (>Carol had got a book.

Also, 's got may be used for the obligatory sense of have. (17)

a.

b.

Susan has to swallow three guppies. Susan's got to swallow three guppies.

The significant question concerning the got of 's got is whether or not it is related to the inchoative get discussed above other than phonologically. In what follows I shall try to show that there is a relation. It is important to notice that the past participial form of get, gotten, is turned into got by a low-level phonological rule in certain dialects. Thus, (18a) can become (186), (19a) can become (196), etc. (18)

a.

b. ( 19)

a.

b.

He has gotten fired. He has got fired.

They have gotten down from the chimney. They have got down from the chimney.

The got that comes from this source is not the got of what is called 's got here, even though contraction of a sentence like (186) can give the homonym, He's got fired. Having avoided the possible confusion, we can confine the subsequent discussion to the real 's got. The first fact to be noticed in finding a source for 's got concerns the historical development of the perfect aspect in English. This aspect originated from the present tense of transitive verbs: I have the letter written became syntactically I have written the letter. "As having the letter in a written state implies previous action, have written gradually acquired verbal force, serving as a verbal form, pointing to the past and bringing it into relations with the present [Curme, 1931, p. 358]." A special class of verbs resisted development of the perfect aspect, a class which has been termed the mutative in-

Get

213

transitives, including come, go, arise, assemble, fall, etc. So we had Christ is arisen instead of Christ has arisen, They are assembled, The tree is fallen, etc. The same class in contemporary German takes sein instead of haben. Notice that the mutative intransitives are all inchoatives. In past participial form they represent the completion of some process of change. Arisen is the completion of rising, fallen the completion of falling, etc. This completion is responsible for the semantic stativity of the past participles of these verbs, and from the same source I shall claim arises the stativity of 's got noted above. We have here also a semantic explanation for the fact that the mutative intransitives did not participate initially in the development of the present perfect in English. For to say He has worked leaves open the possibility of his continuing to work in a way that He worked or He had worked does not. Thus, The tree has fallen might initially suggest that the action is not yet completed, so the stative be was used. (A deeper explanation than the one outlined here is eventually required; the sketch above will suffice for present purposes, however.) I shall claim that the got of 's got is the idiomatic lexicalization of the complex structure perf + come to have. Support of an indirect sort comes from the fact that the Greek word for to possess consisted morphologically of the same structure. This got is stative in the same way that the past participial forms of the mutative intransitives are stative. One can postulate that the idiom developed historically from gotten by the phonological process of dropping -en mentioned above. Once idiomized, however, it became distinct from has gotten. In favor of this account of 's got as an idiom is the fact that it relates this form to the productive inchoative get, and that it accounts for its stativity. Further, we might expect that ambiguities in the word have will show up as corresponding ambiguities in 's got, as the latter contains the former. Thus, the obligatory have of Susan has to go does appear in 's got, as in Susan's got to go.

SYNONYMY AND COMPOSITIONAL THEORIES OF MEANING

The general construction of all compositional theories of meaning is the same. The theories begin by hypothesizing a set of primitives, which one may call semantic primes, associated with or perhaps identified with some semantic reading. Such units enter as terminals in complex phrase structure configurations. The semantic representation for the entire configuration is then a determinate function of the readings of the primes, where the function itself is given by the

214

John Kimball

phrase structure tree. Put simply, readings on higher nodes of the tree are given in terms of readings on constituent nodes by some rule which is determined by the particular configuration. In this general sense of what constitutes a compositional theory of meaning the theory known as generative semantics is also a compositional theory. This theory identifies the meaning of a sentence with its underlying representation. It is a compositional theory in the sense that the semantic reading assigned to a higher node is assigned in each case by the identity function -the reading is the set of nodes dominated. The reading assigned to the highest S node is identical to the tree dominated by that S node. So again in this theory, the reading of the whole is a determinate function of the readings of the parts, only the function involved is in each case the identity function. An assumption held in common by all compositional theories of meaning is that a word has a meaning independent of its occurrence in a sentence. The basis of this assumption seems at the outset to be unassailable. Thus, it seems as though we can factor out the meaning of a word, say cat, by considering what portion of semantic content all sentences which contain this word have in common. It follows further from this assumption that if one substitutes for a particular word in a sentence one of its synonyms, the resulting sentence will have the same meaning. The assumption basic to compositional theories of meaning was challenged first by Frege, who maintained that words have meanings only in the context of propositions. It is interesting to see what evidence can be brought to bear on the question based on the discussion of get above. In particular, I should like to suggest that the situation is more complex than envisioned in any compositional theory. The evidence noted at the outset was that words which were synonyms for get in some environments were not only not synonymous in others, but also were ungrammatical. This made it look as though get were not one word but many. One way out of the seeming perplexity is to say that get is not a word at all, in the sense of occurring as a deep structure element. This is in fact the case, as the discussion above was concerned with showing. However, this does not clear up the difficulty with the test for meaning by substitution of synonyms. It simply pushes the problem back one more step. For whatever synonym we have for come to be in come to be down, it will be different from that in come to be fired. So instead of inquiring after the meaning of get in get down and get fired, we are faced with the equivalent problem for come to be. The point is that these elements in some sense in addition to having meanings of their own, also have parts of their mean-

Get

215

ings derived from their occurrence in particular sentences. The idiomization of 's got illustrates this point, for in this case the 'derived' meaning of perf + come to have in certain contexts became frozen.

REFERENCES Curme, J. 0. Syntax. Boston: D. C. Heath, 1931. Lakoff, R. Passive resistance. Papers from the Seventh Regional Meeting, Chicago Linguistic Society, 1971, 149-162. McCawley, J. Lexical insertion in a transformational grammar without deep structure. Papers from the Fourth Regional Meeting, Chicago Linguistic Society, 1968, 71-80.

SENTENCE AND PREDICATE MODIFIERS IN ENGLISH FRANK W. HENY University of Massachusetts, Amherst

THE THOMASON-STALNAKER ANALYSIS

I want to take up a few of the more interesting points arising from the recent (and so-far unpublished) discussion between Thomason (1970), Stalnaker (1970a,b) and Lakoff (1970a) on aspects of the logical and syntactic structure of English adverbs. 1 Since these papers are not freely available-and in any case perhaps not very transparent to many linguists - I shall summarize the relevant parts, simplifying and elaborating somewhat, but I hope not distorting their content. I shall not, in general, distinguish between the contributions of Thomason and Stalnaker since their proposals interest us mainly as a background to Lakoff' s, and Thomason has indicated (personal communication) that he accepts Stalnaker's modifications. With a little pushing from Lakoff (1970a), Thomason-Stalnaker identified two major logical classes of adverbs in English: (1) sentence adverbs like necessarily, and (2) predicate adverbs like reluctantly and slowly. Sentence adverbs can be used to modify a whole 1 This is an enlarged version of a paper read at the LSA summer meeting, July, 1971. An earlier draft was read by Osten Dahl, F. R. Palmer, George Lakoff, Cliff Abbott, and Rich Thomason. The present paper has benefited considerably from their comments and I should like to thank them and publicly exonerate them from any blame for what follows.

217

218

Frank W. Heny

sentence or just the predicate of a sentence while predicate adverbs cannot be used to modify a full sentence. Various criteria were proposed to distinguish the two classes, but the only ones that will concern us much here have to do directly with the distinction between sentence and predicate modification. To understand this basic distinction let us look first at a sentence which is ambiguous according to whether the adverb necessarily modifies the whole sentence or just the predicate. (1)

The (U.S.) president is necessarily a citizen (of the United States).

(In future, we shall not repeat the material in parentheses but it is to be understood.) Under the first interpretation, (1) is roughly synonymous with (la) A certain person, who happens to be the president, is

necessarily a citizen. Clearly this is false under the normal interpretation of the words. Nixon could at any time renounce his citizenship and join the LR.A.; Johnson could by now be a citizen of Mexico. The other interpretation of (1) may be roughly paraphrased thus: (lb)

It is necessarily the case that whoever is president is at the same time also a citizen.

This is presumably true. These two senses of (1) can be associated with the following logical formulas: (2)

a. b.

NECESSARILY NECESSARILY

x [Citizen x [Citizen

(x)] (the president) (x)] (the president)

In these formulae, modified a little from Thomason and Stalnaker (1968), 2 the scope of the modal operator NECESSARILY is indicated 2 In Thomason and Stalnaker (1968) the formulas corresponding to (2a) and (2b) are actually:

(2)

a'.

y□ Q(y)(ixP(x))

b'.

□ yQ(y)(ixP(x))

The most significant change that I have made is in (2a). In the original, the modal operates on the basic predicate Q; in (2a) on the other hand NECESSARILY is set up as an operator on the result of abstraction to bring it into line with Thomason's treatment of predicate modifiers. In the earlier analysis of the ambiguity of (1), the variation in the relative scope of the modal and the abstraction operator was the crucial factor. The formulae (2a) and (2b) in contrast show the modal always operating on the result of abstraction and the ambiguity depends solely upon whether or not in

Sentence and Predicate Modifiers in English

219

by underlining. In order to represent to their satisfaction the two meanings of (1) (and some other phenomena of no relevance to the present discussion), they imported a device from higher order logic, the abstraction operator x, which has the scope shown by the brackets ([ ]) and functions in each case to produce from a predicate an abstract predicate which is applied to the relevant argument of the original predicate. Only two things need concern us here with respect to these formulas. First, the president in (2a) is not within the scope of NECESSARILY (i.e., it is not underlined) while in (26) it is. Secondly, in (2a) NECESSARILY is modifying only a predicate, while in (26) it applies to a full (i.e., closed) sentence consisting of a predicate and its argument. Now we notice that in general two sentences like (3)

a. The president is ill. b. Nixon is ill.

are always true and false under the same conditions if Nixon is the president-so that (3a) can never be true and (36) false, or vice versa. Nevertheless, while there is a true reading of (1), there is apparently no way of construing (l')

Nixon is necessarily a citizen.

so as to get a true sentence. 3 (We are, of course, assuming that Nixon addition the subject of the sentence (i.e., the president) lies within the scope of the operator. When the subject is inside its scope, NECESSARILY is acting as a sentence operator; but when it is not then, like RELUCTANTLY in formulae (8a), (8b), it is a predicate modifier. It seems to me-though I am not really certain-that what Thomason (1970) has now done is to assume sufficient revision of the syntax and semantics of the ordinary predicate calculus to enable him to carry out the project (rejected in favor of abstraction in Thomason and Stalnaker, 1968, p. 364) of distinguishing directly and unambiguously between what we might represent as [□ Q](,xP(x)) and □ [Q(,xP(x))]. He still uses the device of abstraction, but for a different purpose. 3 This claim is palpably false as it stands. To it might be added the words " . . . with the same interpretation as either of the previously described readings of (l)." Since Nixon is at this moment president and since while he is president a man must be a citizen, Nixon must at this moment (and throughout his presidency) be a citizen. This seems to be one possible interpretation of (l') and is true. Notice that sentence (1) may have a similar additional reading, where it refers to the present incumbent as the president and predicates of him that throughout his incumbency it must be the case that he is a citizen. These readings, which both seem to be derived from the de dicta reading of (1), call for further analysis, but they do not appear to be immediately relevant to the present discussion.

220

Frank W. Heny

as a man is entirely free to renounce his citizenship.) Hence, we cannot substitute referentially equivalent expressions in the subject of (1) and always preserve truth values. In contrast with the situation in (3a), (3b), the subject of (1) is in a not fully referential or opaque position in the sentence; one aspect of this is that substitution under identity is invalid. One mark of members of the class of sentence adverbs, observed and rightly used as a criterion by Stalnaker, is that they can induce opacity in subject position. Superficially, at least, we can already see why this might distinguish sentence from predicate adverbs: NECESSARILY contains the president in its scope only when it modifies a full sentence, as in (2b), and the ambiguity under discussion apparently arises only because NECESSARILY may modify either a predicate or a sentence. Since predicate adverbs cannot under any circumstances modify a full sentence, we would not expect this ambiguity to arise with them. It does not. As Thomason-Stalnaker observed, we find (4a) and (4b) always true together: a. The president reluctantly signed the treaty. b. Nixon reluctantly signed the treaty.

(4)

Substitution under identity is valid (other things being equal) in the subject of sentences containing a predicate modifier like reluctantly. On the other hand, reluctantly can induce opacity in object position. So (5a) and (5b) do not necessarily have the same truth conditions, even if we assume that China is the only country that can/

could guarantee world peace. 4 (5)

a.

b.

The president reluctantly signed a treaty with China. The president reluctantly signed a treaty with the only country that could guarantee world peace.

Let us suppose (counterfactually?) that the president would like nothing better than to bring about world peace, and would very much like to sign a treaty with whichever countries could contribute to 4 There is an interpretation of (5a) [similarly (6a)] which is probably related in some way to the interpretation of (l') discussed in the previous footnote and which would have the same sense (and hence truth conditions) as (5b) [respectively, (6b)]. If the speaker assumes that his hearer believes China = the only country that can guarantee world peace then, even if he knows Nixon does not, he is entitled to describe the situation corresponding directly to (5b) by means of (5a) [similarly for (6b) and (6a)]. Linguists are properly concerned with such pragmatic aspects of interpretation but it must not be allowed to interfere with the basic semantic analysis. We refer no more to such additionally possible derivative interpretations of examples like (5a) and (6a).

Sentence and Predicate Modifiers in English

221

bringing this about. There is then clearly a sense in which (5b) is false - far from being reluctant to sign with a country that could guarantee peace he would be delighted. At the time, however, it might be the case that no one (including the speaker of the sentence) knew (or believed) that China could guarantee world peace. Then there might be no false reading for (5a). Some speakers may find the difference in truth values more obvious in the following pair.

. {carefully . Jocasta. Oedipus . t t· ll } married in en iona y . {carefully . d h·is moth er. b . O e d ipus . t t. ll } marrie in en iona y

(6)

a.

A predicate modifier like reluctantly may also introduce quantifier scope ambiguities in the object position: (7)

a.

b.

The president signed all the treaties reluctantly. The president reluctantly signed all the treaties.

If we assume a set of treaties to be signed, say, with China, South Africa, and Ireland, (7a) has a reading in which it is true only if the president was reluctant to sign with China, reluctant to sign with South Africa, and reluctant to sign with Ireland. The most natural reading of (7b), on the other hand, would be true if he was quite willing to sign with South Africa and China, reluctant to sign with Ireland, yet knew China wouldn't sign unless Ireland did-hence his reluctance to sign of them. (Whether the two sentences can, for some speakers under certain conditions, have the reverse interpretation is irrelevant to the present discussion.) This ambiguity of scope in object position can be represented thus in the Thomason-Stalnaker system, if y ranges over the set of treaties. 5

all

(8)

a. b.

(Vy) RELUCTANTLY x [Sign (xy)] (Nixon) RELUCTANTLY x [(Vy) Sign (xy)] (Nixon)

In both cases, reluctantly modifies the predicate resulting from abstraction, i.e., to the material within the scope of x. But in (8a), corresponding to (7a), the universal quantifier is outside the scope of the adverb while in (Bb) it is inside. When we compare (8a) and (8b) with (2a) and (2b) 5 Cliff Abbott has quite rightly pointed out that it is a far from trivial problem to find a general system in which the semantics of e.g., reluctantly will operate appropriately in the two formulae (Sa) and (Sb) to give appropriate truth conditions for the two cases.

222 (2)

Frank W. Heny

a.

NECESSARILY

b.

NECESSARILY

x [Citizen x [Citizen

(x)] (the president) (x)] (the president)

we see that there is no formula for RELUCTANTLY corresponding to (2b): (8)

C.

"'RELUCTANTLY

x [('v'y)

Sign (xy)] (Nixon)

Predicate modifiers by definition cannot modify a closed sentence and the material underlined in (Sc) or (2b) is a closed sentence. So no formula like (8c) is defined for RELUCTANTLY. If, as we have seen, the ambiguity of (1) is ascribed to the fact that its subject (the president) falls either inside or outside the scope of necessarily, and if substitution under identity fails when the adverb contains the subject within its scope, then we do not expect to find comparable ambiguity and failure of substitution in the subject of sentences modified by reluctantly. And we do not. Examples (4a) and (4b) illustrate this fact. (They cannot, of course, be conclusive.) On the other hand, we found that a quantifier binding the object could apparently be either inside or outside the scope of reluctantly in the formulas (8a) and (8b). These two formulas appear to represent adequately the difference in meaning between (7a) and (7b). Moreover, (5a) and (5b) [or (6a) and (6b)] exhibit ambiguity and substitution failure in the object quite comparable to that found in the subject of (1). In (2a) and (2b), that ambiguity was represented as a difference in the scope of necessarily; but (8a) and (8b) represent quantifier ambiguity as a similar difference, this time in the scope of reluctantly. What is the relationship, then, in the Thomason-Stalnaker system, between the quantifier-scope ambiguity in object position in (7a) and (7b) and the two cases of ambiguity of a rather different kind found in (1) and (5) or (6)? We are really dealing with two questions (Thomason-Stalnaker answer neither explicitly). First, are there quantifier scope ambiguities as in (7) found in subject position with necessarily and if so how will they be represented; and, second, how are we to represent the ambiguity/substitution failure found in the object with reluctantly and other predicate modifiers? It isn't really very important for us to answer the first question. But I think there may be the right kind of scope ambiguity in the following-although the meaning of (9a) is better represented by a modal. (9)

a. b.

?All the runners will possibly win. Possibly all the runners will win.

223

Sentence and Predicate Modifiers in English

If there is such an ambiguity, I think we could represent it thus, without doing too much violence to the system: a. b.

(10)

(Vy) POSSIBLY POSSIBLY (Vy)

x [Win x [Win

(x)] (y) (x)] (y)

For the moment, at least, let us assume that this answers our first question. The second is more difficult. The rest of this paper is really just an attempt to answer it. Although Thomason-Stalnaker did not explicitly say so, I think they probably assumed-and Lakoff certainly did-that the ambiguity of (5b), for example, (5)

b.

The president reluctantly signed a treaty with the only country that could guarantee world peace.

could be represented by permitting the description the only country that could guarantee world peace to arise either within or outside the scope of RELUCTANTLY; substitution failure would result when RELUCTANTLY has wider scope. If we represent the description as P, we could render the two readings of (7b) as something like: (11)

a.

b.

x [Sign-treaty-with (xy)] (Nixon)] (iy(Py)) x [Sign-treaty-with (x, iy(Py))] (Nixon)

y[RELUCTANTLY RELUCTANTLY

In (Ila) the description is outside the adverb. That is the transparent reading. (llb) corresponds to the opaque reading, where the description is inside the adverb and substitution fails, just as it did in the opaque reading of (1). Thus we appear to have provided some account of all the cases in the Thomason-Stalnaker system, by providing formulae to represent the opaque and transparent readings in object position.

LAKOFF'S EXPLANATIONS

Lakoff observed however, that in addition to "psychological" adverbs such as reluctantly, there are adverbs like slowly or with a knife, which seem at first sight good candidates for predicate modifiers, yet do not induce opacity even in object position. In particular, substitution under identity is valid in object position with such adverbs. Thus, in contrast with (5a) and (5b), (12a) and (12b) are always true together under the previous assumption that China is the only country that can guarantee world peace.

224 (12)

Frank W. Heny

a.

b.

The president slowly signed the treaty with China. The president slowly signed the treaty with the only country that could guarantee world peace.

On the basis of this observation, which seems so far correct, Lakoff proposed to distinguish a third major logical class of English adverbs. For reasons that will be apparent directly, let us call the "psychological" adverbs like reluctantly "two-place" adverbs. The simplest modification to the Thomason-Stalnaker binary classification into sentence and predicate modifiers is the following which appears to be assumed by Lakoff (19706). 6 Sentence adverbs-opaque subjects: necessarily Two-place adverbs-opaque objects: reluctantly Predicate adverbs-no opacity: slowly, with a knife Having noted the contrast between slowly and reluctantly, Lakoff sets out to explain a number of semantic facts by making reference to syntactic results already established for English. He observes, quite correctly, that the logicians' formulas merely codify the facts and explain nothing about the observed behavior of slowly, reluctantly, or necessarily. In particular, he asks: I. Why do two-place adverbs never induce opacity in subject position? II. Why can two-place adverbs nevertheless induce opacity in object position? In other words, concentrating on the psychological two-place adverbs, he tries to explain precisely those characteristics which set them off 6 Lakoff (personal communication) has claimed that he does not accept the Thomason-Stalnaker analysis of slowly or with a knife as predicate modifiers. However nothing in Lakoff (19706) makes that clear, and, as I shall point out later, a two-place analysis of the psychological adverbs explains none of the opacity facts unless it is contrasted with (and closely related to) some other analysis in which slowly and with a knife differ in precisely the right way from reluctantly. To suggest, as Lakoff (personal communication) has now done that slowly and with a knife are just like reluctantly in being two-place predicates but that their "truth conditions, when understood, would account for the lack of substitution failure" is to concede the main point of my paper; namely, that nothing in the syntax ("logic" for Lakoff?) of these adverbs will explain, or can be expected to explain, their behavior in potentially opaque contexts. Notice, too, that just in making this point Lakoff has accepted the essential correctness of the interpretivist position in one important area of the grammar. It seems likely to me that in the near future arguments of the sort advanced in the present paper will increasingly tend to delimit areas where an explanation must be given in terms of rules of semantic interpretation, set up somewhat after the fashion first proposed in Heny (1970).

225

Sentence and Predicate Modifiers in English

from sentence adverbs on the one hand and "pure" predicate adverbs the other. Of course, it is implicit in his explanation that there are syntactic analyses available for sentence and predicate adverbs which permit the former to induce subject opacity and prevent the latter from inducing opacity in object position. But he only discusses the two-place adverbs. For them he proposes to adopt his well-known analysis of manner adverbials (Lakoff, 1970c, pp. 157-159, 165-167). Under that analysis, (4a) has an underlying syntactic structure approximating the following: (13)

/s~

a.

NP

~

VP....,_____

I . . . ___ s y I~

the president reluctant

NP

~

the president

VP

I

I~

~

sign the treaty Hence, reluctantly is derived from reluctant, which is a two-place predicate. It is, in fact, a transitive verb possessing a sentential object. Like try, reluctai:it requires identity between its own subject and the subject of the embedded sentence. Just as ~John tried for Henry to go is ungrammatical, so would a sentence be which had John in place of the president in one sentence of (13a). On these putative facts rests Lakoff' s explanation. It seems an interesting one. Consider, for example, the following:

All the presidents reluctantly signed the treaty.

(14)

First let us assume that the quantifier has wider scope than reluctant. (13)

b.

/s""-

/Q\

V

X

s""vr~ V I

NP I

reluctant S

vr~

\

V

NP

I~

sign the treaty

226

Frank W. Heny

Identity holds between the two instances of the bound variable (required independently within his theory) in the subject of reluctant and the lower sentence. The normal reading of (14) is correctly represented. But if the quantifier is within the scope of reluctant we have something like (13)

c0

/ \VP............

NP

I

.......

I I

x/z

NP I

V

reluct~S \

s----

/ \ CT) -----Vr~ Q

v

NP

1

X

V

I

NP

~ sign the treaty

The subject of the embedded sentence is bound, but not the subject of reluctant itself. Hence there is no formal identity between the relevant NPs; and Lakoff has explained by using his two-place analysis of such adverbs why they cannot induce scope ambiguity in subject position. At the same time, as (13d) shows, there is no effect on the required subject identity if we introduce a lower quantifier binding the object variable. (A higher one a fortiori has no effect on that identity.) ( 13)

d.

/ NP

s '------VP----....,__

Ij "~p ~ the president

reluctant/ s"--.

/Q\

V

y

/s~ NP

~ the president

VP-..........

I

V

I sign

"----

NP

I

y

Sentence and Predicate Modifiers in English

227

It seems reasonable to conclude with Lakoff that he has hit on very interesting independent confirmation for his old syntactic analysis and a nice explanation for what the logicians have been content to describe. PROBLEMS IN LAKOFF'S EXPLANATION

Before agreeing with Lakoff, however, let us look at his explanation in more detail. First of all, what are the supposedly independent arguments for the analysis of reluctantly as a transitive verb? The best known concern questions, negatives, and imperatives (Lakoff, 1970c, pp. 165167). These particular arguments, as Lakoff has pointed out (personal communication) were intended only to establish that the items in question are predicates which take as arguments the remainder of the sentence in which they are (superficially) embedded. The following are the crucial examples. In (15a) the speaker assumes that the president signed and asks about his reluctance. (15)

a.

Did the president sign the treaty reluctantly?

In (15b) it is again assumed that he signed-while the claim is made that he did not do so reluctantly. b.

(15)

The president didn't sign the treaty reluctantly.

If (15c) is grammatical (and it is with carefully) it enjoins reluctance (care) and again assumes that the signing will take place. (15)

c.

?Sign reluctantly (carefully).

Why such examples as (15a)-(15c) should be regarded as evidence for an analysis of reluctantly as a higher predicate need not concern us; the reader is referred to Lakoff's papers for details of his reasoning (see also Carden, 1968). The basic premise of the argument is simply that negation, questioning, etc., apply always to the highest predicate in the underlying structure-i.e., the one immediately below the relevant operator in the phrase structure. What is important here is that if (15a)-(15c) provide evidence for the analysis of reluctantly as a higher predicate, then to the same extent they show that many of the sentence and (of more relevance) predicate adverbs are to be analyzed similarly. Consider (16) and (17) (16)

a.

b.

Did the president sign the treaty slowly? The president didn't sign the treaty slowly.

228

(17)

Frank W. Heny

d.

c.

Sign the treaty slowly. I don't think the president signed the treaty slowly.

a.

Is the president necessarily a citizen?

b. The president isn't necessarily a citizen. c.

d.

I don't think the president is necessarily a citizen.

In each of these, the signing (or that the president is a citizen) is assumed, and the question, denial, or whatever refers to the adverb alone. Hence these arguments appear to suggest that contrary to the original arguments of Thomason-Stalnaker (apparently accepted by Lakoff) and their analysis represented in formulas like (2a) and (2b) even predicate modifiers must be higher predicates. They do not, of course, force us to abandon the Thomason-Stalnaker logical analysis unless (a) we believe that (14a)-(15c) force us to adopt an analysis of the two-place adverbs as higher predicates, and (b) this turns out, when more precisely defined, to be inconsistent with formulae like (2a), (8), etc. That issue must be regarded as at least potentially separable from the question of underlying syntactic structure. (See also Footnote 6.) But whatever the outcome, Lakoff's criteria have failed (so far) to provide us with independent syntactic grounds for distinguishing the two-place adverbs from the others. If all three classes of adverbs have to be analyzed alike as higher predicates, is it also the case that all three have to be analyzed as twoplace predicates? If so, that will go a long way toward destroying the force of Lakoff's explanation of opacity. But it is clear that there are adverbs for which the arguments for binary structure do not apply. Lakoff (1970c, pp. 157-159) has specific arguments for the two-place nature of a manner adverb like carefully or reluctantly, and these more specific arguments certainly do not apply to a sentence adverb like necessarily. There is, for a start, no paraphrase of (1) like: (1')

frJ} being a citizen.

a.

0

The president was necessary

b.

0

The president was necessary to be a citizen.

Hence, although some such form could presumably be posited as an underlying structure for (1), its existence cannot (as it can in the case of reluctantly) constitute part of an argument for setting necessarily up as a two-place predicate. Is there such an argument? Perhaps we should concentrate our attention on the arguments on pages 157-159, assuming that Lakoff intended us to do so. After all,

Sentence and Predicate Modifiers in English

229

as we have mentioned, the other arguments show at best only that necessarily, reluctantly, and slowly alike are (sometimes) higher predicates at some level. We hope therefore, on Lakoff's behalf, to find that the arguments specifically for a two-place analysis apply only to his class of twoplace adverbs. But unfortunately, they do not. Among the examples used in the original argument on page 159 are the pair; a. John was good at playing gin rummy. b. John played gin rummy well.

(18)

He could just as well have cited; (19)

a.

b.

The president was slow in signing the treaty. The president signed the treaty slowly.

The relationship between the members of these pairs is precisely the same in all relevant respects as the relationship between, for example; (20)

a.

b.

The tailor was careful in fitting me. The tailor fitted me carefully.

All the arguments for deriving (20b) from (20a) apply equally to (19) and (18). Yet, as we have seen, slowly never brings about substitution failure in object position and is, in fact, Lakoff's (1970b) own example of a plain predicate modifier. The reader can easily prove that well must be, too. Hence Lakoff' s specific arguments for a two-place analysis of carefully, reluctantly, etc. apply at least as well to slowly, well, and in fact a number of other pure predicate modifiers. Hence, by his own account, and contrary to fact, they should induce opacity (in particular, failure of substitution) in object position [see (13d)]. The fact that they do not do so means that we cannot accept Lakoff' s syntactically based explanation of the behavior of adverbs but must search for some other explanation. Let us return briefly to the arguments from questions, negation, etc. It is worth observing that not only do these apply to too many adverbs (i.e., adverbs in all three classes) but they actually fail to apply to sentences in which reluctantly occurs-when it stands outside a quantifier. (21)

a.

Did the president {calrefut llytl } sign all the treaties? re uc an y

230

Frank W. Heny

In (21a), contrary to expectation,7 it is clearly not the president's reluctance that is at issue. The speaker is asking some kind of question (precisely what is not clear) about the signing of the treaties. In the same way (2lb)-(2ld), in so far as they are grammatical, don't negate or otherwise focus on the adverbs. 8 7 That is, contrary to expectation under the supposition that questions, negation, etc., generally and in principle, apply to the highest predicate in the underlying syntactic structure-which is the basis of Lakoff's original argument. It is, however, perfectly easy to see that even when careful appears overtly as a higher predicate in the sentence

(i)

Was the president careful to sign all the treaties?

the question nevertheless does not apply to that predicate but to all the treaties or to some part of that constituent. As Chomsky (1968) and Anderson (1970) have pointed out, the scope of questions, negatives, and the like is apparently defined most naturally in terms of relatively superficial phenomena like stress and surface order. There is, of course, the further problem of determining the logical form of a question like (i). It is to beg the question to assume that logical form and underlying syntactic structure are necessarily the same. 8 A number of readers will have experienced discomfort every time two sentences like the following are compared;

(i)

a. John carefully ate all the bagels. b. John ate all the bagels carefully.

The first is perhaps most naturally paraphrased by something like (iia), the second by (iib): (ii)

a. John was careful to eat all the bagels. b. John ate all the bagels in a manner that was careful.

In (iib), careful describes the manner in which John ate the bagels while in (iia) there is no implication that John's manner was at all careful; the modifier signifies rather that John made sure that the event in question actually occurred. Thus, more than scope appears to be involved in distinguishing the two senses. I think that that is often the case. Such a meaning difference is not, I think, limited to obviously quantified sentences: (iii)

a. Slowly John turned to face Bill. b. John turned to face Bill slowly.

Moreover, adverbs placed before the verb rather than after the VP may tum out to have a sort of nonrestrictive force, being an interpolation or comment by the speaker. It may be this latter factor which makes (2lb)-(2ld) seem strange. But I can find no more than a difference in scope between the following-and they don't seem to constitute an isolated case, or to differ in principle from the others: (iv)

a. John slowly ate all the bagels. b. John ate all the bagels slowly.

and the situation described in the text for (21a) certainly seems to hold equally for the related (iv.c):

Sentence and Predicate Modifiers in English

(21)

b.

C.

d.

231

?The president didn't {calrefut llytl } sign all the re uc an y treaties. } . l . { ?Carefully Reluctantly sign al the treaties. ?I don't think the president {calrefut llytl } signed all re uc an y the treaties.

So even the basic arguments for setting up certain adverbs as higher predicates do not work in some cases for Lakoff' s two-place class in a certain general class of environments. We cannot, of course, conclude on the basis of such evidence that his analysis is wrong-but since the original arguments for analyzing adverbs as higher predicates and specifically two-place predicates apply to a large class of cases which his explanation demands they not apply to, and since we have now found a class of cases where the general arguments for adverbs as predicates do not apply even to his favored class, the basis for the explanation begins to look shaky. Worse is to come. Although, as we saw in (12a) and (12b) (12)

a.

b.

The president slowly signed the treaty with China. The president slowly signed the treaty with the only country that could guarantee world peace.

there is no failure of substitution under identity in object position after slowly, there can be scope ambiguities. Examples (22a) and (22b) are not always true under the same circumstances. 9 9

(iv)

Thomason (1970) noticed this fact but Lakoff apparently did not see its significance. c.

Did John slowly eat all the bagels?

The additional semantic or pragmatic factors sometimes introduced by placing an adverb early in a sentence require explanation. Perhaps it will turn out that they can best be explained by deriving carefully from careful in and careful to-though that will still call for some account of why the adjective careful differs in the two uses unless we give in and settle for two distinct adjectives and hence two adverbs. Until that is shown to be the case, I simply make the fewest possible assumptions counter to the overt form of the language and will search for a single function to interpret carefully in such a way that it interacts with the rest of the functions that map (ia) and (ib) onto truth values, to produce the observed differences in meaning (in some sense) in the two environments. Whether or not this can be done in a satisfactory manner is an interesting empirical question in linguistic semantics, part of the enquiry projected in the closing section of this paper.

232 (22)

Frank W. Heny

a.

b.

The president signed all the treaties slowly. The preside'(Jt slowly signed all the treaties.

The first can only be true if for each of the relevant treaties the · president signed it slowly. The second can be true even if certain of the treaties were signed quickly, but the signing of the whole set of treaties took a long time, i.e., was carried out slowly. Lakoff was assuming that substitution failure and quantifier scope ambiguities were essentially a single scope problem-which he calls "opacity." To do him justice, many logicians seem to have identified the two phenomena and it is not clear that Thomason-Stalnaker distinguish the two appropriately. At any rate Lakoff's explanation of the validity of substitution in the subject of a sentence containing reluctantly, and its failure in object position, is given only in terms of the impossibility of obtaining proper subject identity when the quantifier binding a variable in the subject of the lower sentence occurs inside the adverb [see (13c)], while a quantifier binding the object could occur either inside [as (13d) shows] or outside the adverb without interfering with identity. In other words, Lakoff attempted to explain failure of substitution by reducing it (in some manner which is never made explicit) to scope of quantification. But we have just seen that the two are at least partially independent. Slowly, which he regards as a pure predicate modifier does not produce failure of substitution but it does produce scope ambiguity. There must be at least one other factor involved in substitution failure and we cannot hope to explain it solely in terms of scope differences. The fact that substitution failure is not to be identified with scope ambiguities is of general importance. I have argued elsewhere (Heny, 1970, especially pp 70-122) that it is a mistake to identify the two in relation to verbs of propositional attitude, but it is worth establishing a further brief argument for separating these phenomena, using facts more immediately related to our present concerns. Assume that (23a) and (23b) and (23c) are true with respect to the same situation, under at least one normal interpretation of each. 10 10 Objections to the argument based on examples (23a)-(23c) have been suggested by a number of people including Osten Dahl, George Lakoff, Rich Thomason, and Cliff Abbott. Essentially (it seems to me) these amount either to the proposal that the sense of (23b) that concerns us [where all of (23a)-(23c) are true together], is obtained if we permit all in (23b) to have the group or collective interpretation, or alternatively by letting the description the wrong questions identify a set s outside the scope of carefully and have the quantifier all alone operate (within s) inside the scope of the adverb. I cannot see how the first proposal will help in any way, and I know of no way

Sentence and Predicate Modifiers in English

(23)

233

a. John carefully answered all the questions that he

thought he was supposed to do. b. John carefully answered all the wrong questions. c. John didn't answer all the questions (that he answered) carefully.

More specifically, John was careful to answer all of a certain set of questions. He believed that each of those he answered should be answered. In fact that set consisted of all the questions that he should not have done. Yet John was not deliberately failing the test by answering the wrong questions; he was intending to answer the right ones. Notice that by (23c), John was not careful in answering individual questions; only about the set that he intended to answer. There is a reading of (23b) which is false on this evaluation. Ignoring that, let us concentrate on the true reading. Carefully must haye wider scope than all since (23c) is true. The description the wrong questions must have narrower scope than all in standard predicate calculus. Yet that description must at the same time have wider scope than carefully, since (23a) is true. If (23a) were false then the other reading of (23b) might be true, where John actually wants to answer the wrong questions. Then the description could without any difficulty have narrower scope than carefully. But our assumptions make that reading false. We are dealing with a reading in which the description must not occur within the scope of the adverb. So the description the wrong questions both must and must not have narrower scope than carefully. That is clearly a contradiction. We cannot account for this reading of (23b) using scope differences involving the opacity inducing adverb carefully and any selection of quantifiers, etc. Now the reading in question is what has been called transparent; under such a reading substitution is at least marginally valid. [So (23a) is true, but see Kaplan, 1968; Quine, 1955]. of achieving the second aim within any available system. There may be one, but I greatly doubt it for the following reason. There is no direct relationship between carefully and the description of the set in question, as the wrong questions. What John was careful about was answering all of a certain set and in terms of his beliefs which are a prerequisite to the interpretation of carefully, that set was properly described as the right questions. Hence the only way of distinguishing the transparent from the opaque reading is by going beyond carefully itself directly to the ascription of responsibility for descriptions. The relevant sense of (23b) seems to arise because a speaker is not obliged to avoid the use of an expression like the wrong questions, if it is referentially appropriate, even when it occurs within an intensional operator and he knows that some other of the personae (such as the subject in this sentence) could not appropriately use it. For further discussion see Heny (1970, Chapter 3).

234

Frank W. Heny

The other reading, the false one, is opaque. It has been held that the contrast between opaque and transparent construal of a sentence could be represented by giving the operator inducing opacity wider imd narrower scope (respectively). If one term of the opposition cannot be unambiguously assigned wider scope then the opposition cannot be accounted for in terms of scope. We return briefly to the problem of representing this contrast in meaning later in the paper. For the moment it is only relevant that Lakoff now turns out to have been misguided in trying to explain failure of substitution solely in terms of scope differences-i.e., by reference to the putative identity constraints holding for a predicate like careful(ly) or reluctant(ly). But perhaps we are now in a position to salvage some part of his explanation. Recall that both slowly (a predicate adverb) and reluctantly (a two-place adverb) exhibit ambiguity of scope in the object [examples (7a), (7b), (12a), (12b)]. And the original arguments for setting up reluctantly as specifically a two-place predicate applied equally to slowly [examples (16)-(20) and the accompanying discussion are relevant]. We have now established that opacity (specifically, failure of substitution) must be due to some additional logical parameter not yet discussed. So if we set aside the problem of substitution (as we must), it looks as though Lakoff's original arguments, and the facts about relative scope just discussed, place slowly and reluctantly in a single class set off from the sentence modifiers like necessarily. Moreover, Lakoff's two-place analysis seems thus far to be correct in predicting the scope ambiguities for both slowly and reluctantly. So even if we cannot even use Lakoff 's analysis to account for how slowly and reluctantly differ, we might use it to account for what they have in common. But that will not do. There is a fourth major logical class of adverbs which we have not yet mentioned. They act as slowly was supposed to act. In other words they do not even exhibit quantifier scope ambiguity in object position. Some good candidates are:

(24)

well completely partially fumblingly a. John filled all the buckets expertly entirely adequately singly single-handedly

Sentence and Predicate Modifiers in English

235

4 well completely partially fumblingly b. John expertly filled all the buckets. entirely ?adequately 4 singly single-handedly

(25)

a. John tapped at all the doors softly. b. John softly tapped at all the doors.

Although it may be possible for some speakers to find scope ambiguities and even substitution failure in object position, I am confident that (with the possible exception of expertly) none can be found for my speech. There are, in any case, plenty of other examples of this class available. Notice that even when these adverbs occur before the verb as in (24b), (25b), they are still interpreted unambiguously with narrower scope than the quantifier (some, like well and singly, do not appear there). In both the (a) and (b) cases, it is the filling of individual buckets (or whatever) that is done completely, expertly, and so on. We seem to have in these adverbs a class which behaves as Lakoff thought slowly did: They never appear outside a quantifier. In one sense, all we have done is show that Lakoff's example, slowly, fits squarely into the same class as reluctantly, once we ignore failure of substitution, and find some new cases to add to his other example, with a knife-which certainly never seems to fall outside a quantifier. But to reestablish Lakoff' s class of plain predicate adverbs does nothing for his explanation of the special behavior of the two-place class. For all the old arguments - including those for a binary analysis -apply to at least some plain predicate adverbs. In fact, good/well was one of Lakoff's (1970c, p. 159) examples. Compare the following with the relevant cases of (24). (26)

a. John was good at filling buckets. b. John was expert at filling buckets. c. John was fumbling in (his) filling (of) the buckets.

d. John was single-handed {~hen} filling the buckets. True, there are some for which the arguments based on such forms are irrelevant:

236

Frank W. Heny

(27)

a.

b. c.

"'John was complete in filling the buckets. "'John was adequate in filling the buckets. ?John was soft in tapping at the doors.

But that really only makes the situation more complicated; it does nothing to advance Lakoff 's claims about the unique syntax of the two-place adverbs. Now that we have redrawn the lines of Lakoff's threefold classification and ignored substitution failure in object position we are still unable to isolate the slowly/reluctantly class syntactically from the well class. Consequently it seems very unlikely that his old two-place analysis will be of any use in explaining either (1) why neither of these classes induces substitution failure or quantifier ambiguity in subject position or (2) why only the first class (i.e. slowly/reluctantly) has scope ambiguity in the object. In fact, it turns out that if we explained both those sets of phenomena we should have explained too much. To add to the complexity of the situation - and detract from the force of Lakoff 's explanation -there are at least some nonsentence adverbials which under certain conditions appear to induce scope ambiguity in subject position. Consider the following pairs: 11 (28) (29)

(30)

a. Slowly, all the marbles dropped from the box. b. All the marbles dropped from the box slowly. b.

Carefully all the men held the rope. All the men carefully held the rope.

a. b.

Bravely, 500 soldiers faced the invaders. 500 soldiers bravely faced the invaders.

a.

The first pair seems to show the cleanest contrast. Thus, (28a) can be true even if some of the marbles actually fell quickly provided the whole set fell slowly-i.e., took a relatively long time to fall. But (28b) can only be true if each marble dropped slowly. The second pair seems to contrast in roughly the following way: the first sentence asserts that in all grasping the rope, the men (as a group) were careful-presumably it was important for them all to grasp it. The second sentence merely asserts that each of the men took care to grasp it. The third pair contrasts in a similar fashion to the second; (30a) asserts that it was brave of a (mere) band of 500 to resist, while (30b) asserts 11 Lakoff (personal communication) claims that these facts are accounted for in Lakoff (1971, p. 160), but I cannot see how anything is explained by describing certain quantifiers as "group" nonquantifiers -and nothing more. As I have shown (Heny 1970), it is possible and indeed desirable for many reasons to treat the group reading of quantifiers as no different in kind from any other reading of an English quantifier.

Sentence and Predicate Modifiers in English

237

of each of the 500 that he was brave to resist-numbers have nothing to do with it though. In addition, I think that (30a) could be true even if there were some soldiers of whom it couldn't be said that they were individually brave to resist- because they were drunk or misled or mad. Under such circumstances, (30b) would certainly be false. In the discussion of (28)-(30) I have made some rather fine distinctions. For my speech they certainly hold for some readings of the sentences; and I am reasonably certain that plenty of examples, some better for other dialects, could be turned up. What significance have such examples for the Thomason-Stalnaker and Lakoff accounts of English adverbs? They suggest that the basic distinction between sentence and predicate modifiers may have relatively little importance. The original distinction was based largely upon the fact that there was a possibility of substitution failure in subject position only with sentence modifiers. And that was attributed to the fact that sentence modifiers could modify just a predicate (an open sentence) whereas predicate modifiers could never modify a full, closed sentence. 12 We have now shown that, at times, they can. And the fact that they do not (as far as I can see) ever produce sub12 It is true that Thomason-Stalnaker mention other possible criteria. But only on those discussed here is there much agreement and certainly only they can be immediately related to the semantic distinction that is being made. One of their claims which looks superficially interesting is the claim that only when an adverb is a sentence modifier does it occur in the frame:

It is ADVERB true that S.

So we find (ia) but not (ib): (i)

a.

b.

0

It is necessarily true that S. It is carefully true that S.

I do not know what to make of such a frame as the following: (ii)

?It slowly {1b·s } true that all the marbles fell from the box. ecame

And notice that (iii) seems to be just as ambiguous as (1) in the text: (iii)

It is necessarily true that the president is a citizen.

If this syntactically based criterion were immediately relevant, then, assuming that the analysis of the ambiguity of (1) summarized in the text is more or less correct, (iii) should be interpreted only with necessarily acting as a sentence modifier. In the light of these facts, it seems quite possible that sentences like (i) are most comfortable when the adverb most commonly functions as a sentence modifier. But they do not force the interpretation to proceed along those lines. Hence, there is no necessary, immediate correspondence between syntactic and semantic structure, though it may be that there is some indirect relationship between the syntactic property of modifying true with a sentential subject, and the semantic property of bearing in general an interpretation along the lines of sentence modification.

238

Frank W. He.Jly

stitution failure in that position is no more significant than the fact that slowly does not do so in the object while reluctantly does. There is no doubt that necessarily is semantically-and perhaps syntactically-quite different from slowly, just as reluctantly is quite different from well. It would be very nice to (a) characterize these differences precisely and (b) explain them. However, we have found good reason to reject the Thomason-Stalnaker system and better reasons for throwing out Lakoff's explanation of some of its characteristics. The observation that slowly and other nonsentence adverbs can sometimes act as sentence adverbs actually complicates Lakoff's explanation only a little- but it is worth noting how it does so. Recall that for Lakoff a sentence adverb like necessarily is a one-place predicate having a sentential subject. An adverb like reluctantly (and hence by our arguments slowly) is supposed to have a sentential object and concrete subject. Consequently, when slowly changes from being a two-place to a sentential adverb it changes its syntactic characteristics radically. In the underlying structure of (28a), it is an intransitive verb with a sentential subject; in (28b) it is transitive with a concrete subject and sentential object. Moreover, necessarily in (2a) would also have to be a two-place adverb, despite (l'a) and (l'b), p. 228! EXPLAINING SEMANTIC FACTS

There seems very little left of Lakoff 's explanation. We have altered quite radically the set of facts that have to be explained, and have shown that his account could explain neither what he set out to account for nor what we have ended up with. In the light of such a disastrous outcome, it may be worth standing back a little and asking what kind of account is most likely to explain the relevant facts about the behavior of English adverbs. We examine three possibilities: syntactic constraints, logical form, and semantic functions. Syntactic Constraints It is actually very hard for me to see why Lakoff set out on his program at all. If the system had behaved itself properly this would have been interesting, but while his results might have confirmed and explained his original analysis of carefully, I don't see how they could possibly have explained the distribution of opacity which he set about accounting for. To relate the syntax of one unknown like careful to that of another like try is unlikely of itself to explain anything about their meaning.

239

Sentence and Predicate Modifiers in English

For one is immediately entitled to ask why both are subject to the equi-NP constraint. Intuitively one expects to find that if there is a relationship between the syntax and semantics it will go quite contrary to Lakoff's expectations: A sentence like 4 John tried for Mary to win is rejected because the very precise, unconscious rules that operate on English lexical items to build up appropriate truth values for all English sentences simply go wrong when we try to apply them to such a sentence. Presumably the same kind of anomaly results when we try to apply those rules to 4 John was careful in Mary's cutting the bread. So if carefully and try are subject to a similar syntactic constraint, that is presumably because there is something similar about the way they play a part in the rules for assigning truth values to English sentences. In a rather precise way, in other words, careful and try are semantically alike and this could (though it need not) account for any similarity that exists in their syntax. Now failure of substitution (which is what Lakoff set out to account for) has very obviously something to do with the rules for assigning truth values to English sentences. Almost by definition, it has to do with the need to assign values to certain parts of a sentence not merely relative to the real world but relative to other worlds as well. I suppose it is conceivable that independent syntactic constraints might interfere with such processes, but, other things being equal, that is a relatively unlikely kind of explanation of the distribution of opacity. Moreover, as Lakoff himself pointed out, the difference in meaning between slowly and reluctantly (and words like them) is that reluctantly makes crucial reference to the attitude of the subject towards the event described by the sentence, whereas slowly does not. It seems an incredible waste of effort to ignore this highly relevant fact and chase wildly after little understood syntactic constraints in an effort to explain the difference between the two classes. I cannot show that such an attempt will never work; it is simply most unlikely, in principle, to do so.

Logical Form What about the logical form of the adverbs? Can that explain why they behave as they do? At first glance it might appear to provide a partial explanation of some of the facts. Let us look again at the formulae for necessarily and reluctantly (the latter now seem to apply equally to slowly): (2)

a.

b.

NECESSARILY NECESSARILY

x [Citizen (x)] x [Citizen (x)]

(the president) (the president)

240 (8)

Frank W. Heny

a. b.

('dy) RELUCTANTLY x [Sign (xy)] (Nixon) [('dy) Sign (xy)] (Nixon) RELUCTANTLY

x

We need to add a further case, to handle the completely class. If we let y range over the set of buckets in question, then the sentence (31)

John completely filled all the buckets.

might be represented thus in order to complete the paradigm: (32)

('dy) COMPLETELY X [Fill (xy)] (John)

Now we noted in the beginning that nonsentence modifiers cannot generally modify a closed sentence. In other words, something like (31) is generally ill-formed-or as one might say, ungrammatical [cf. (2b)]. (33)

"'RELUCTANTLY

x [('dy) Sign (xy)]

(Nixon)

If we consider that there is a single, independently established level of logical form as part of the structure of natural language, a position that Lakoff (1970) has argued for, then it is clearly conceivable that elements operating on this level will be subject to various syntactic constraints. Such a constraint applying to the reluctantly class could easily prevent (33). To generate forms like (28)-(30) we should simply have to permit the relaxation of the constraint-a not unknown kind of weakening in grammatical structure. A very similar constraint would then have to operate on items of the completely class. They would have to be barred from structures of the following kind: (34)

"'COMPLETELY

x [('i/y)

Fill (xy)] (John)

Whether there are exceptions to this constraint as there were to (33) we do not yet know. But there appears to be a redundancy rule holding between the constraints, which a fortiori bars (35). (35)

"'COMPLETELY

x [('dy) Fill (xy)]

(John)

In other words, there appears to be a regular system of constraints of increasing power applied to the level of logical form. The existence of that system might be taken to explain the apparently idiosyncratic behavior of its members with respect to scope differences in subject and object position; it seems to do so at least as well as a great number of syntactic explanations that are framed in terms of similar constraints. As a matter of fact we can pursue this kind of explanation a little further. Notice that the essential difference between (2a) and (2b) is that in (2b) the argument to which the abstract predicate x ... is

Sentence and Predicate Modifiers in English

241

applied, namely the president, is inside the scope of necessarily while in (2a) it is not. In sentence (1), represented by (2a) and (2b), the president is the subject. In the formulas for reluctantly, of course, the subject never [under normal applications of the constraint implied by (33)] falls under the scope of the adverb. But why the subject? There is no a priori reason for always doing abstraction in such a way as to leave, of all the arguments of the original predicate, the subject as the argument of the new abstract predicate. Well, of course, we can easily stop that gap: There must be a global constraint (see Lakoff, 1969) placed upon the argument of the abstraction such that it must always become the (underlying) subject of the main verb. In fact I believe there is indeed a very close relationship between the subject-predicate form of many natural languages and the logical process of abstraction. But to search for a relationship of the sort we have been tentatively exploring is to put the syntactic cart before the semantic horse. Lakoff was right to regard the logicians' formulas as totally unexplanatory. In general logical formulas simply encode facts about interpretation. Logical form tells us nothing that cannot be derived from our knowledge of how to assign truth values to sentences - i.e., from our knowledge of how in general they relate to conditions in the world. A formal calculus of entailment is, after all, simply designed so as to give results which tally with the interpretation of the calculus relative to all possible situations (see Tarski, 1941), and whether or not there is a distinct level of logical form in natural language is still an open question. Lakoff (1971) has argued that there is. I happen to believe that there is not, and that none of the evidence which he has given stands up to scrutiny. It is impossible to go into that question in this paper, but there is some rather interesting data which suggests strongly that it would be quite wrong for the linguist to bother about looking for certain standard logical forms to assign to English adverbs as a stage in accounting for their behavior. If we take a single adverb and look carefully enough we quite often turn up cases like the following-which are clear for my speech: (36) (37)

a.

b. a.

b.

0

John furnished all the rooms of his house lavishly. John lavishly furnished all the rooms of his house.

°For the party, John filled all his bathtubs lavishly with beer. For the party, John lavishly filled all his bathtubs with beer.

242 (38)

Frank W. Heny

a.

b.

John furnished all the rooms of his house lavishly with period furniture. John lavishly furnished all the rooms of his house with period furniture.

In (36), lavishly appears to belong to the completely class; in (37) it appears to belong to another class which we have not yet touched on (and there are many such classes): the adverbs, like together and equally [and lavishly, in (37)], which generally refuse to take narrower scope than a quantifier-or more generally than some plurality operator. Finally, in (38) lavishly exhibits scope ambiguity, like slowly. Should we assign the adverb to three different classes? We ended up doing something like that for slowly, and the cases seem quite comparable. However, to adopt such a solution is obviously pointless from the linguist's point of view. It might make some sense as a temporary expedient for th_e logician, but even that I doubt. There is of course no question but that lavishly must enter the logician's formulas with wider and narrower scope on different occasions, and the linguist doing semantics must have some analogous way of capturing the different values of (38a) and (38b) relative to possible situations in the world. But that does not force us to conclude that there are three lavishlys subject to different sets of constraints which are part of the apparatus of language. There is just one word lavishly, and the proliferation of exceptions (which would, I believe, have to be made for a great number of other words, too) will eventually prove theoretically intolerable. But above all, as we show in the next section, we already have available a way of accounting, in part, at a much deeper level, for the facts observed in these examples. Semantic Functions What is interesting in the behavior of lavishly is the interaction that takes place between the meaning of that word and the meaning of the rest of the sentence in which it occurs, to determine the possibilities for narrower and wider scope. So, while furnishing all the rooms of a house cannot be lavish (it is perfectly normal), you can furnish individual rooms in a way that is lavish. You cannot, on the other hand, fill a bathtub with beer in a way that is lavish- but to fill all one's bathtubs with the stuff certainly can be. Merely to furnish all one's rooms, on the other hand, may be lavish if it is with period furniture; and each room may or may not then be furnished in a way that is lavish. That is not very profound or precise. It is only meant

Sentence and Predicate Modifiers in English

243

to indicate roughly how an account of the precise semantic properties of lavish 13 (i.e., the part it plays in leading to truth values for the sentences in which it occurs) must be expected to explain its behavior with respect to scope. Let us continue this same line of inquiry just a little further. If slowly is interpreted in such a way that it makes reference to some relationship between (and this is exceedingly tentative) standardized time series and intervals during which events actually took place (rather like an added dimension to a continuous tense marker) and if reluctantly is interpreted by a function which necessarily makes reference to an individual, then we can begin to see why slowly but not reluctantly can occasionally have scope ambiguities in the subject. Moreover those ambiguities will, correctly, be limited to cases where the members of the set of objects named by the subject noun phrase take part in a number of nonsimultaneous but sequential actions - as in our sentences about marbles dropping. At the same time we begin to see how the fact that reluctantly can induce substitution failure is to be attributed to the fact that it is interpreted in such a way that truth values are dependent in part on possible worlds (or states of affairs) which are purely relative to the intentions (etc.) of the subject of the sentence. In general, logicians like Reichenbach (1947), Parsons (1968), Montague (1969), Clark (1970), Davidson (1966), and now ThomasonStalnaker, have assumed that English adverbs can profitably be treated as primitives in their formulae (where they have bothered to treat them at all instead of simply allowing them to be swallowed up by predicates). What I have been trying to suggest is that if we are content to deal with modifiers at the level where they appear simply as primitive operators on predicates (as suggested by Clark, 1970; Montague, 1969; Parsons, 1968; Thomason-Stalnaker, 1968), 13 A difference in the meaning of lavishly comparable to that found with carefully remarked on in Footnote 8 obviously further distinguishes the pairs (34)-(35). We find essentially the same difference between:

(i)

a. John spread his bread lavishly with homemade butter. b. John lavishly spread his bread with homemade butter.

This difference interacts in an interesting way with the variation in scope which we are taking as the central phenomenon. Notice that there is no easy way of saying what the following is attempting to convey: (ii)

?As regards each of his bathtubs, John lavishly filled it with beer.

In particular, this is not a possible interpretation of (37b). I am not even sure that (ii) itself can carry the intended sense.

244

Frank W. Heny

we lose access to all their linguistically interesting and perhaps to many of their logically interesting properties. The alternative, I suppose, is to leap into the uncharted swamp that lies out there somewhere beyond tense logic. A frightening, if challenging prospect, perhaps. But on reflecting that the logicians have had about as much success with the logic of adverbs as linguists have had with their syntax, and taking into account our brtef ride (in this last section) over the edges of the swamp, I begin to wonder if it is really all that much worse out there. I strongly suspect it to be very fertile country- and at least we shall be traveling in the right direction. I have said nothing in detail in this paper about the functions that will be needed or how they will operate on English sentences. For a highly unsatisfactory account which does not touch on the problem of adverbs but is written from a linguist's point of view the reader is referred to Heny (1970). Parsons, Montague, Thomason-Stalnaker and many of the recent works by modal logicians might prove helpful to a linguist. But for a detailed account of the semantics of adverbs along the lines suggested in this section we will probably have to wait some time.

REFERENCES Anderson, S. How to get even. Language Research Foundation Report, September 1970. Carden, G. English quantifiers. National Science Foundation Report No. 20, Harvard Computational Laboratory, 1968. Clark, R. Concerning the Logic of Predicate Modifiers. Nous, 1970, 4, 311-335. Chomsky, N. Deep structure, surface structure and semantic interpretation. In D. D. Steinberg, & L. A. Jakobovits (Eds.), Semantics: An interdisciplinary reader in philosophy, linguistics and psychology. Cambridge, England: Cambridge University Press, 1971. Davidson, D. The logical form of action sentences. In N. Rescher (Ed.), The logic of decision and action. Pittsburgh: University of Pittsburgh, 1966. Heny, F. Semantic Operations on Base Structures. Unpublished doctoral dissertation, University of California, Los Angeles, 1970. Kaplan, D. Quantifying in. Synthese, 1968, 19, 178-214. Lakoff, G. On derivational constraints. Papers from the Fifth Regional Meeting, Chicago Linguistic Society, 1969, 117-139. Lakoff, G. (1970a). Adverbs and Modal Operators. Unpublished manuscript, University of Michigan, 1970. Lakoff, G. (1970b). Adverbs and Opacity: A Reply to Stalnaker. Unpublished manuscript, University of Michigan, 1970. Lakoff, G. (1970c). Irregularity in syntax. New York: Holt, Rinehart and Winston, 1970. Lakoff, G. Linguistics and natural logic. Synthese, 1971, 22, 151-271. Montague, R. English as a formal language. Proceedings of a Symposium: Language in Society and the Technical World, Milan, 1969.

Sentence and Predicate Modifiers in English

245

Parsons, T. A semantics for English. Unpublished manuscript, University of Chicago, 1968. Quine, W. V. Quantifiers and propositional attitudes. Ways of paradox and other essays. New York: Random House, 1966. Reichenbach, H. Elements of symbolic logic. New York: Macmillan, 1947. Stalnaker, R. (1970a) Notes on "A Semantic Theory of Adverbs." Unpublished manuscript, University of Illinois, 1970. Stalnaker, R,,. (1970b) Notes op. Adverbs in Response to Thomason and Lakoff. Unpublished manuscript, University of Illinois, 1970. Tarski, A. Introduction to logic and the methodology of science. London: Oxford University Press, 1941. Thomason, R. A Semantic Theory of Adverbs. Unpublished manuscript, Yale University, 1970. Thomason, R., & Stalnaker, R. Modality and reference. Nous, 1968, 2, 359-372.

DERIVING S FROM S + /5 1 JAMES R. HURFORD University of Lancaster, England

The purpose of this paper is to argue that the deep structures of many, and possibly all, English sentences contain embedding structures in which the verb of the matrix sentence is some form of the verb be, and the subject of this matrix sentence is an embedded sentence whose constituents are the deep structure reflexes of the surface simple sentence. In other words, I shall argue that in many cases a surface S is derived from (at least) a deep structure like (1).

s

(1)

NP~VP

A it

S

A

I

be

I shall not be concerned with details such as the justification (or lack of it) for the it, or the position of the Aux in (1). I shall argue only 1 The composition of this paper has benefited from conversations with Barbara Partee (University of California, Los Angeles), Ron Arbini (University of California, Davis), and Norman Fairclough (University of Lancaster) and from correspondence with Paul Schachter (University of California, Los Angeles), Geoffrey Leech (Lancaster), Stanley Peters (University of Texas), and John Kimball (University of California, Santa Cruz). Responsibility for these proposals is, however, entirely my own.

247

248

James R. Hurford

for the embedding as in (1) and the presence of some form of the verb be in the VP of the matrix sentence. The semantic content of the deep structure VP with be appears to be the same as that of be the case, be a fact, be true, be the truth, be so, etc. How the lexical component of a grammar is to cope with the phonological diversity of these synonymous predicates, including the bare be as in (1), I do not attempt to explain, but on the basis of their synonymy, I assume that evidence for the presence of one of them in deep structure reflects a general likelihood that any or all of them may also be found in deep structure. And in fact one of the arguments I shall present concerns the predicate be a fact, rather than a bare be, as in (1). Apart from the Noun fact, the negative element, modals, and a few adverbs mentioned in passing, I will not deal with cases where the matrix sentence might appear to have some constituents in addition to be and the embedded subject sentence. The significant claim of this paper is, then, that the deep structure of a sentence may represent explicitly that what it states is true. I emphasize that the specific formulation of this claim in terms of the notion "deep structure" obviously derives from certain general theoretical assumptions, which, though they are at present widely accepted for working purposes, may yet need revision. The data mentioned in this paper will, however, need explanation in any theory and I believe that any revised version of the theory of transformational grammar (under whatever name) will have to contain the equivalent, in its terms, of the claim I am making here. My arguments are drawn from eight sets of data: be to constructions, not that clauses, pseudo-cleft sentences, the general relationship between assertions and presuppositions, certain interrogatives, factive complements, certain sentences with modals, and sentences like that is for you to say. I will not discuss here the relationship of my proposal to Ross's (1970) performative analysis, though I believe that with a few necessary modifications to Ross' s theory, my proposals and his are compatible. BE TO

Consider the following sentences. (2)

a. We are soon to be in England. b. You are to leave immediately. c. Mary is to have breakfast in the kitchen. d. Snakes are to be found near the stream.

Deriving S from S + Is

(3)

249

a. We will soon be in England. b. You will leave immediately. c. Mary will have breakfast in the kitchen. e. Snakes will be found near the stream.

Each of the sentences of (2) is synonymous on at least one reading with one of those of (3). As Jespersen (1940, pp. 236-239) points out, the be to construction can have a variety of subtly different meanings, not all of them exactly paraphrasable by expressions with will, e.g., This is not to be endured, but from a survey of Jespersen's examples and my own, I conclude that in a significant majority of cases it is possible to paraphrase an expression with be to by a similar expression containing will. Both be to and will can be used to express simple expectation of the future, to impose one's will, to explain some plan, to describe some general possibility, to convey a certain fatalism about events, etc. This conclusion is corroborated by Huddleston (1971, pp. 313-314) in whose corpus nine out of a total of eleven instances of be to are used with the senses of Instruction or Potentiality. Huddleston' s examples of these senses of be to can all be paraphrased with will. Given this relationship between be to and will, it is worthwhile to try to derive sentences such as those of (2) and (3) from identical deep structure sources, thus providing the basis for an explanation of their synonymy. The sentences of (2) and (3) can in fact be derived from identical deep structures of the type proposed in this paper without postulating any otherwise unmotivated transformations. I propose deep structures such as (4). (4)

we will soon be in England Sentence (3a) is derived from this structure simply by deleting the VP be in the matrix sentence. I will argue in the penultimate section (p. 288) that this rule can be justified on independent grounds. The derivation of sentence (2a) from structure (4) proceeds as follows. Extraposition yields a structure like it be [we will soon be in England] 8 •

250

James R. Hurford

Applying It Replacement to this structure has the effect of deleting u;e from the embedded S and substituting we for it in the matrix. This yields something like l)We are will soon be in England. Deletion of the offending will in this environment is independently motivated by alternations such as I promise that I will be there versus I promise to be there; I have decided that I will go versus I have decided to go; I expect that I will be late versus I expect to be late. That is, I assume that the structure of this last sentence after Equi-NP Deletion is something like I ·expect will be late; deletion of will yields I expect be late. Similarly, in the example we are following, deletion of will gives l)We are soon be in England. Infinitivalization, i.e., insertion of to before a verb with no overt surface subject in the same S (see Kiparsky and Kiparsky, 1971), like the be here; yields finally the desired We are soon to be in England. The derivation of (2a), then, is exactly parallel to that of such sentences as We are likely to be in England, which I assume to be transformationally related (on one reading at least) to It is likely that we will be in England. The verb be shares with be likely, be certain, be sure the idiosyncratic property that permits It Replacement when will occurs in the extraposed subject complement. This property is not shared by seem, appear, happen. It Replacement cannot apply to It seems that we will be in England, for example. I)

I)

NOT THAT CLAUSES

English clauses beginning with that are for the most part restrictive relative clauses or surface structure complements of some kind. There is disagreement on the correct analysis of clauses introduced by the complementizer that. All proposals of which I am aware, however, analyze such clauses as deep structure embedded sentences and disagreement concerns only the type of constituent in which the sentences are immediately embedded (e.g., NP or VP) and the possibility of sister constituents [such as it or (Det) N]. All proposals insert the complementizer that by a transformation. There is, furthermore, general agreement on the semantic relationships between that clauses and other constituents of the sentences in which they occur. There is, however, a class of sentences containing that clauses where the usual clear semantic relationships between the that clause itself and other surface constituents is not manifested. These are sentences as in (5). (5)

a.

I'm mad at Jean for eloping with a saxophonist (not that I have anything against musicians).

251

Deriving S from S + Is

b. c.

d.

I admire Spiro's tenacity (not that I agree with what he says). Phil asked his boss for a raise (not that he needs one). It's an original idea (not that originality is a virtue in itself).

The that clauses in (5) do not function semantically as if they were complements of any of the other surface constituents, even allowing a little juggling with the not. Thus (5d), for example, is not synonymous with either of the sentences of (6). (6)

a.

b.

That originality is not a virtue in itself is an original idea. That originality is a virtue in itself is not an original idea.

In whatever way the parenthesized clauses in (5) are analyzed, they have an altogether unusual semantic relationship to the other elements in the surface structure of these sentences. One should clearly try to treat That Insertion as a unitary process if possible. The rule for inserting that into ordinary complements is something like (7).

(7)

NP I

NP I

s

/'-....

NP

VP

s

~

that

NP

VP

(Nothing crucial to my argument rests on the choice of exactly this formulation.) If that is to be introduced into the sentences of (5) by this rule, then their final clauses must be constituents of a higher S. What are the other constituents of this higher S? The higher S cannot be an abstract performative sentence because of the possibility of negatives inside the not that clause, as in (8). (8)

a. John is a bore (not that he isn't intelligent). b. John isn't attractive (not that I'm not fond of him). c. He really shouldn't have sent all those flowers (not that he can't afford it).

Performative sentences are by definition nonnegative, and, unless one allows more than one negation per sentence, the not of not that must belong to the higher S. The possibility of a negative inside a not that clause, as in (8), is at the same time further evidence that

252

James R. Hurford

there is more than a single S in the underlying structure of such expressions. Sentences containing performative verbs may of course be negated, but when negated they lose their performative force. Semantically one can see that the higher S to which not that clauses belong cannot be a negated performative such as I do not say that ... The not that clause in (5c), for example, is not synonymous with I do not say that Phil needs a raise. This latter sentence constitutes a deliberate avoidance of any assertion on the subject of Phil's need for a raise: it is noncommittal. The not that clause in (5c), however, definitely contains the assertion that Phil does not need a raise. This point can be made generally for all not that clauses. Now consider the sentences of (9). (9)

a.

b. c.

It's not that I'm squeamish, but shouldn't you wash your hands? It's not that Peter's a rat; he's just misunderstood. It's not that she's in trouble-no, nothing like that.

In these sentences, not that clauses are preceded by it is. In these not that clauses, just as in those of (5) and (8), what is asserted is the negation of what follows the that. The not that clause in (5a), for example, asserts that I have nothing against musicians; in (8a) it is asserted that John is not unintelligent; and in (9a) it is asserted that I am not squeamish. It seems plausible, then, that all these not that clauses are derived from the same source and furthermore that the verb in the higher S is is. That is, the deep structure of the first half of (9a) is something like (10).

s

(10)

~

Neg

NP I

S

~

VP

I

is

I am squeamish Given such a deep structure, It's not that I'm squeamish and the first parts of the other sentences of (9) can be derived by the familiar processes of Complementizer Placement, Negative Placement, and Extraposi tion. Deriving the not that clauses of (5) and (8) from the same type of

Deriving S from S + ls

253

deep structure source as the it's not that clauses of (9) captures a significant semantic generalization about the assertions that are made in these clauses. This treatment makes necessary the postulation of rules accounting for the absence of it is in examples such as those of (5) and (8). I argue below that the postulation of such rules is independently motivated. The distribution of not that clauses is significantly similar to that of not clauses and phrases generally. Neither not that clauses nor not phrases may occur as independent sentences, as the following examples show. (11)

a.

b. c.

(12)

(;Not that I'm squeamish. (;Not that he needs one. (;Not that I have anything against musicians. a.

b. c.

(;Not 'detergent'. (;Not my words. (;Not Jean's bottom.

Both not that clauses and not phrases must be followed by the conjunction but when they occur sentence initially. Thus we get: (13)

Not that I agree with what Spiro says, but I admire his tenacity. b. Not that Phil needs a raise, but he asked his boss for one.

(14)

a. Not Humphrey but Nixon is President. b. Not the candidate but the proposal had merit.

a.

while the following are ungrammatical: (15)

(;Not that I admire what Spiro says, (and) I admire his tenacity. (;Not that Phil needs a raise, (and) he asked his boss for one.

a.

b. (16)

a.

b.

(;Not Humphrey (and) Nixon is President. (;Not the candidate (and) the proposal had merit.

As examples (11)-(14) show, both not that clauses and not phrases must be coordinate with some other clause or phrase. In examples (13)-(14) the not that clauses and not phrases are the first members of their coordinations. The examples of (5) and (8) and those below show that both not that clauses and not phrases can also be the second member of a coordination.

254 (17)

James R. Hurford

I said "deterrent," not "detergent." These are the cock's words, not mine. c. We anticipated his breaking down, not his recuperating. d. I helped him to dig the tunnel, not to crack the safe. a.

b.

A difference between not that clauses and not phrases in general is that the conjunctions and and but may sometimes occur before not phrases but never before not that clauses. That is, while (18a) and (18b) are grammatical, (18c) and (18d) are not. (18)

a.

b. c. d.

I sai'd "deterrent ,, {but} and not "detergent. ,, (>Phil asked for a raise,

{!~~} not that he needs one.

I shall offer an explanation for this fact later in this section. It is clear, meanwhile, that this difference does not outweigh the significant similarities that I have noted between not that clauses and not phrases in general. We therefore seek a single unified account of all not phrases and clauses. I present below an account that is clearly adequate to capture the major significant generalizations that hold true of not phrases and I shall show that this account can be extended to not that clauses given the assumption that this paper sets out to justify, namely that the deep structure of all English simple sentences may be something like that shown in (1). The account of not phrases that I propose owes a lot to Gleitman (1965), although I have also used ideas of my own. The analysis I propose of not phrases is this: (19)

Given two coordinate deep structure sentences, say S1 and S2 , in either possible order, where S1 dominates the terminal string X Y, and S2 dominates either the terminal string Neg X Z or the terminal string Neg Z Y (X, Y, and Z being variables ranging over substrings); then grammatical sentences containing not phrases can be generated by carrying out the following transformational operations in the order given: a. Substitute the morpheme not for the abstract negative operator Neg and move it to a position immediately to the left of Z. b. Delete any string X or Y adjacent to the string not Z provided (i) it is identical to some other string X or Y remaining in the structure, and (ii) it is not a sentence-initial string. c. If S2 precedes Si, insert but between them.

255

Deriving S from S + ls

In this account, some restrictions must be placed on the type of constituent represented by Z; I will not touch on this detail here. I will illustrate the above analysis with several derivations. Take the coordinate structure (20).

s

(20)

s~s

~VP

NP

Nixon

VP

6

6

is President Neg Humphrey is President

Here Nixon is the string X, is President is the string Y, and Humphrey is the string Z. Application of (19a) results in Nixon is President not Humphrey is President. Now application of (19b) will result in deletion of either the first is President or the second one. Thus we can derive either Nixon not Humphrey is President or Nixon is President not Humphrey, both of which, given the appropriate intonation contour or punctuation, are grammatical. As a second example, consider (21).

s

(21)

s

s

VP

/\NP

V

Neg

these

16

~VP ~ NP V

NP

~

are my words these are the cock's words

In this example, these are is the string X, the cock's words is Y, and my words is Z. Application of (19a) yields these are not my words these are the cock's words. According to (19b), only the second occurrence of these are may be deleted, since the first is sentenceinitial. Applying (19b) we get these are not my words the cock's

256

James R. Hurford

words. Now (19c) applies, giving the grammatical these are not my words but the cock's words, reducible to the stylistically preferable these are not my words but the cock's. If (19b) permitted deletion of the first occurrence of these are, we would derive the ungrammatical not my words but these are the cock's. Rule (19c) inserts but in a particular environment, but there are other instances of but that are not accounted for by this rule. I claim that the but used in not X but Y constructions has rather different properties from other instances of but and hence may not be treated in exactly the same way as but generally. Note that the but in not X but Y constructions can frequently be followed by rather, as in these are not my words but rather the cock's. Rather may not be used after other, more usual, uses of but. Thus Frank is weak but rather David is strong and Bill didn't ask me but rather I didn't want to go anyway, for example, are both ungrammatical. Furthermore the but in not X but Y translates differently into some languages from the other uses of but. Thus in German Franz is weak but Otto is strong is rendered as Franz ist schwach aber Otto ist stark, whereas Not Humphrey but Nixon is President comes out as Nicht Humphrey sondern Nixon ist Priisident; Nicht Humphrey aber Nixon ist Priisident is ungrammatical. The treatment of not phrases in (19) captures all the facts about the semantic relationships between constituents of sentences in which they occur. In Nixon is President not Humphrey, for example, it is asserted that Humphrey is not President and this assertion is represented explicitly in the deep structure (20). Furthermore this treatment makes possible an explanation of the ambiguity of such phrases as: (22)

a. John frightens me not you. b. Garlic tastes like onions not paprika.

(22a) may be paraphrased either by John frightens me; he does not frighten you or by John frightens me; you do not frighten me. The structures underlying these sentences can be reduced to a structure underlying (22a) by rule (19). Similarly the nonambiguity of sentences such as those of (17) is accounted for by this treatment. Of the deep structures that could be related to (17a), for example, by the transformations I am proposing, only one is semantically well-formed. Thus I said "deterrent"; I did not say "detergent" is semantically acceptable, but I said "deterrent"; "detergent" did not say "deterrent" is not semantically acceptable.

Deriving S from S + Is

257

Gleitman (1965, p. 108) notes correctly that sentences such as (22) are ambiguous only in writing. In speech the two readings are associated with different prosodic patterns. A rule assigning prosodic prominence to the appropriate constituent can easily be incorporated into (19), but I will not go into this detail here. The derivation of not clauses by deleting one of two identical parts of a coordination makes it possible to account for the unacceptability of certain sentences in a general way. Consider the following: (23)

0

It {pleases} me that Nixon is President, not that Hum-

amuses phrey is.

If the deletion treatment of not phrases is accepted, then the unacceptability of (23) can be attributed to the same source as the unacceptability of (24). (24)

0

1t {pleases} me that Nixon is President; it does not

amuses

please} me that Humphrey is President. { amuse Similarly, in this analysis, the acceptability of (25) follows from the acceptability of (26). (25)

It {pleases} me that Nixon is President, not Humphrey. amuses

(26) It {pleases} me that Nixon is President and that Humamuses phrey is not President. Thus a large class of cases of ambiguity, nonambiguity, synonymy, and unacceptability can be accounted for by postulating the transformational processes of (19). I shall refer to the rules of (19) as "Not-clause Formation." As we have seen, not that clauses have the same distribution as not clauses in general. This generalization can be captured by deriving sentences such as those of (5) and (8) through Not-clause Formation from coordinate structures in which some string is repeated. I propose, then, that (5c), for example, is derived from something like (27). The derivation of Phil asked his boss for a raise (not that he needs one) from structure (27) proceeds roughly as follows. The comple-

258

James R. Hurford

s

(27)

s~s

~VP

NP

NP

s

s

I

I

~

Phil asked his boss for a raise

is

Neg

D

Phil needs a raise

is

mentizer that is inserted before Phil needs a raise; Neg is changed to not and attached to the left of that Phil needs a raise by (19a); the second VP is is deleted under identity with the first by (19b). After pronominalization the second clause has its appropriate surface form not that he needs one. It remains to be explained how the first clause becomes Phil asked his boss for a raise rather than It is that Phil asked his boss for a raise. I shall postpone discussion of this question until the next-to-last section of this paper, where I will argue that the postulation of a rule deleting it is that is well-motivated. I believe that this analysis of not that clauses is justified in its broad outline. Since I have argued for a parallel treatment of all clauses (or phrases) beginning with not, I must try to explain why the conjunctions and and but can be inserted into examples such as those of (17) without affecting the acceptability of the examples whereas these conjunctions never occur before not that clauses. That is, we must show why the examples of (18a) and (18b) are acceptable, while those of (18c) and (18d) are not. For a start, examples such as (18d), i.e., those with and, can be shown to be unacceptable on semantic grounds. It is in general true that whenever we have a string XY and not Z, as in (18d), then, if XZ is a sentence, XY and XZ are implied to be incompatible. If ZY is a sentence, XY and ZY are implied to be incompatible. Thus in (18b), for example, I said "detergent" is a sentence and is implied by (18b) to be incompatible with I said "deterrent." That is, (18b) implies that the speaker could have said either "deterrent" or "detergent" but could not possibly have said both. Similarly These are the cock's words and not mine implies that These are the cock's

Deriving S from S + Is

259

words and these are my words cannot both be true. The point is perhaps illustrated more clearly in sentences such as those of (28). (28)

a. b.

0

J was born in France and not in Italy. J was born in France and not in Paris.

J was born in France cannot possibly be true at the same time as J was born in Italy. They are incompatible sentences. Sentence (28a) is perfectly acceptable. However it is always possible for J was born in France to be true at the same time as J was born in Paris.

These two sentences - call them XY and XZ- are not incompatible, and sentence (28b), corresponding to XY and not Z, is not acceptable, because its implication is false. Now the semantics of not that clauses is such that they generally deny assertions that might in the context of the discourse be expected to be compatible with the assertion made in the preceding clause. Thus in our example, Phil asked his boss for a raise is quite compatible with Phil needs a raise. The purpose of the not that clause is not to deny the compatibility of the two assertions, but rather to deny the truth of the latter. Observe that a sentence like (29) would only be said by someone with eccentric beliefs about the American political system, specifically the belief that always voting Republican and being a Democrat are compatible. (29)

?J always vote Republican (not that I'm a Democrat).

Since the assertions in sentences with not that clauses, when stripped of their negative operators, are always compatible, and since sentences of the form XY and not Z imply that their component sentences, when stripped of their negative operators, are incompatible, as we have seen, there can be no semantically acceptable not that clause preceded by and. From these facts we can also construct a hypothesis explaining the nonoccurrence of but before not that clauses. We can formulate another statement like those above, concerning the acceptability of sentences with but not. If XY and XZ are semantically incompatible sentences, then a sentence of the form XY but not Z is unacceptable; and if XY and ZY are semantically incompatible sentences, then sentences of the form X but not ZY and XY but not Z are unacceptable. As evidence for this, consider the following. (30)

a. b. c.

J was born in France but not in Germany. Men but not women are the weaker sex. 0 Men are the weaker sex but not women. 0

0

260

James R. Hurford

Substitution of and for but in the sentences of (30) would make all the examples acceptable. Note that I was born in France but not in Paris is quite acceptable. Further, consider (31). (31)

a. These are the cock's words but not mine. b. John but not Harry is in the lobby. c. John is in the lobby but not Harry.

In (31a) it is implied that these words could, within the range of possibilities considered relevant to the discourse, be mine as well as the cock's, although in actual fact they are not mine. In (31b) and (31c) it is implied that both John and Harry could conceivably be in the lobby together, although in fact only John happens to be there. These implications would not hold if and were substituted for but in (31). Not that and but not have much in common. They are both signals by which a hearer can recognize that the second of two coordinate sentences, when stripped of its negative operator, is held by the speaker to be compatible with the first. To use both these signals, i.e., to say X but not that Y, would be redundant, as it would signal twice the fact that X and Y are held to be compatible sentences. I speculate that the unacceptability of examples such as (18c) may be due to this redundancy and a general tendency in language to avoid redundancy. It is, of course, not true that any sentence with the word sequence but not that or and not that is unacceptable. But and and are only excluded from in front of the particular type of not that clause that we are discussing, namely what might be called the parenthetic type, with no obvious overt matrix sentence, as in examples (5) and (8). If the analysis of not that clauses and not clauses and phrases in general is correct, then that constitutes one piece of evidence for the proposal that all sentences have a deep structure containing an S whose constituents include is ( or some form of the verb be) and an embedded complement sentence. This follows from the fact that any sentence can be coordinated with a not that clause. (32)

a. Flying planes can be dangerous (not that I mind). b. What time is it? (not that I really care). c. Talk to him once more (not that it will do any good). d. Not that I want to interrupt, but you're standing on my toe.

261

Deriving S from S + Is

e.

f.

Not that I'm nosey, but who were you with last night? Not that I deserve it, but give me one more chance.

Finally I should mention a possible objection to my analysis of not that clauses, brought to my attention by Paul Schachter. The conjunction of declaratives after interrogatives and imperatives is normally not permitted. Thus sentence (33) is ill-formed. (33)

"'What time is it?

{!~~} I really don't care.

Oddly, however, sentence (32b), which corresponds to (33), is wellformed. I have been careful in this study to distinguish between coordination and conjunction. I have called structures which lack any overt conjoining element, such as and or but, coordinations but not conjunctions. Conjoined structures (conjunctions) are a proper subset of the set of coordinated structures (coordinations). The restriction exemplified in (33) applies to conjunctions but not to coordinations generally, as shown by (34). (34)

What time is it?-I really don't care.

I know of no evidence to suggest that (34) is not a coordination. Note that my analysis derives sentences with not that clauses from deep structure coordinations, but not from conjunctions. This explanation is exactly parallel to that proposed by Ross (1967), and adopted by Thompson (1971), for the grammaticality of (declarative) appositive clauses with interrogatives, as in Is even Clarence, who is wearing mauve socks, a swinger? PSEUDO-CLEFT SENTENCES

Another argument for the proposal that simple surface sentences are derived from underlying sentences embedded in a matrix sentence which also contains some abstract form of the verb be is to be found in pseudo-cleft sentences. I quote first Chomsky's (1970) analysis of these constructions. (35) [=Chomsky's (45b)] what John read was a book about himself ... we might explain many of the properties of these sentences by deriving them from a base structure of roughly the form (46):

262

James R. Hurford (36) [= Chomsky's (46)]

s VP

NP

/\

~S

N

be

I~ VP Aux

Pred

II~

John Past V

I

a

it NP

NP

I~ PP book

read a

A

about

John

We might then derive (45b) in the following way: Familiar rules apply to the most deeply embedded S to give John past read a book about himself A new substitution transformation replaces the unspecified predicate !::,. of (46) by the object of the embedded sentence, a book about himself, leaving a 'PROform' in its place. This gives: it-John past read it-past be-a book about himself Relativization and other familiar rules, supplemented by a rule that replaces it that by what, give (45b) [p. 209].

Emonds (1970, pp. 128-131) suggests an analysis similar in all relevant respects to this. I will argue below that the Chomsky-Emo nds analysis of pseudo-cleft sentences is the most satisfactory one that has been proposed. Peters and Bach (1968) have argued against Chomsky's analysis. The essence of their argument is that Chomsky provides no account of how the constituent which is to be moved from the embedded sentence to the focus position is selected. Chomsky (1970) left this matter unresolved, but his discussion of focus in deft sentences in a later paper makes it clear how the problem should be dealt with. Peters and Bach are concerned with the issue of "whether transformations change meaning." It is obvious that the transformation al pseudo-cleftin g process suggested by Chomsky does change meaning. It allows, for instance, the two nonsynonymo us sentences of (37) to be derived from the same deep structure.

(37)

a. What struck the house was lightning. b. What lightning struck was the house.

Deriving S from S + ls

263

Note that the difference between (37a) and (37b) does not involve the truth conditions for either sentence. What is asserted in each case is that lightning struck the house. The basic semantic functions such as subject, object, and predicate are fulfilled by the same lexical items or strings of items in each case. The difference between (37a) and (37b) is a difference of focus. In (37a) the focus is lightning; in (37b) the focus is the house. The issue here, then, is whether transformations change focus, or better, whether they assign focus. Is focus determined in derived structure or in deep structure? Chomsky (1971) has given strong arguments that focus is determined in derived structure after the assignment of an intonation contour. A phrase containing the intonational nucleus is interpreted as the focus of the sentence. As Chomsky's arguments are available, I will not repeat them here. Lakoff (1971), who has attacked a detail of Chomsky's treatment of focus, also does not hold the view that focus is determined in deep structure. He writes: " ... it is clear that the phenomenon of focus does involve global derivational constraints of some sort involving derived structure [p. 262]." The relevance of this argument to the theme of this paper is that there can be no metatheoretical objection to Chomsky's (1970) analysis of pseudocleft sentences on the grounds that it permits derivation of sentences, such as those of (37), which differ in focus, from a common deep structure. Besides their semantic argument against Chomsky's analysis, Peters and Bach also raise a syntactic argument. They note the following sentences: (38)

[= Peters and Bach's (12)] i. I believe someone. ii. I believe John.

(39)

[= Peters and Bach's (13)] i. I believe something. ii. I believe that John is intelligent.

(40)

[= Peters and Bach's (17)] i. What I believe is that John is intelligent. ii. ~what I believe is John.

Peters and Bach write: The pseudo-cleft rule(s) must be restricted so as to select for pseudo-clefting only noun phrases filling appropriate positions in unclefted sentences, namely

264

James R. Hurford

those that could be filled by something. Now this is a decisive fact disconfirming all Extracting Analyses in which the selection of the pseudoclefted noun phrase is carried out by a transformation: for it is impossible to tell at the time when this rule applies whether the noun phrase selected could be replaced by something since it can have been moved arbitrarily far from the elements with which it participates in selectional restrictions. Thus, there is no way to tell from the surface structures of (18) whether the bold face noun phrases can be replaced by something. [= Peters and Bach's (18)] i. John is thought to have been amused by the joke. ii. John is likely to be surprised. iii. My brother is difficult to image being in love. iv. I expect Harry to be said by most people to resemble an elephant.

(41)

[p. 6]

It is quite possible that these facts noted by Peters and Bach do not constitute an argument against Extracting Analyses of pseudo-cleft sentences, such as Chomsky's and Emond's. Lakoff (1970, 1971) has given a number of examples of problems which appear to require that the theory of grammar permit "global constraints" on derivations, e.g., that there may be conditions on the applicability of some transformations which are statable only in terms of a structure existing at a stage in the derivation other than that at which that transformation applies. Grinder and Postal (1971) have also pointed out an example of a transformation which depends for its applicability not only on its structure index being matched at the stage in derivations where it applies, but also on some condition which must be expressed in terms of the deep structures with which the derivations originated. Grinder and Postal use Lakoff s term "global derivational constraint" to describe such a condition on the applicability of a transformation. Baker and Brame (1972) have criticized Lakoff s notion of global derivational constraints, but I believe that their criticisms are adequately refuted in Lakoff s reply (1972). Thus it appears that there is a solution to the difficulty in Chomsky's analysis noted by Peters and Bach, a solution which does not entail abandoning deep structures such as (36). And if we can show that Peters and Bach's proposed solution, which does involve abandoning deep structures such as (36), runs into grave difficulties, then an analysis with deep structures like (36) is more likely to be correct, and we shall, incidentally, have discovered more evidence for the existence of global derivational constraints. Peters and Bach use the example What John counted was the pigeons. They propose the following deep structure:

Deriving S from S + ls (43)

265

[= Peters and Bach's (25)]

s VP

Aux NP

NP

NP~S

it

/\ # s#

~ VP

NP~VP

NP Aux

/"--._

V

~

I

I

/\

V

~

I /\

the thing John Past count something Past be John Past count the pigeons The pseudo-cleft transformation must guarantee that the phrase markers contained in the deep structure represent two sentences that are exactly identical except for the pseudo-clefted noun phrase. We therefore state the following rule. (43)

[= Peters and Bach's (26)]

the thing [s that X Y]s Aux be it# [s X' NP Y']s #

1 1

2 3

'---v---- -

2 3

4 4

6 ~

7 8 7 ~

9 ~

We can order (26) after the relative clause transformation, which will convert (25) into (27). (44)

[= Peters and Bach's (27)]

s NP~VP NP~S

NP

NP~VP

\

1

it

/\ # s #

NP~VP V

/\

NP

I A

the thing that John Past count Past be John Past count the pigeons

266

James R. Hurford

Transformation (26) will then apply to produce structure (28). (45)

[= Peters and Bach's (28)]

s VP

NP NP~S

Aux

~VP j NP NP Aux V

I

NP

(\

the thing that John Past count Past be the pigeons

[ p. 9]

Later rules yield What john counted was the pigeons There are some serious objections to this proposal. One, which is made in UESP, is that "the base structure is not a plausible one for semantic interpretation (cf. 'the thing that John counted was (it) that John counted the pigeons') [ p. 837]." Another objection to Peters and Bach's proposal is this. The values assigned to the variables X and Y in their pseudo-clefting rule (43) are intended to be identical to those assigned to the variables X' and Y' respectively. But note that X and Y will not be identical to X' and Y' respectively if some optional transformation has applied to one of the embedded sentences and not to the other. It can be seen that this will lead to the generation of ungrammatical sentences, such as t.tWhat John counted was that the pigeons were counted by him, the derivation of unambiguous sentences, such as What bothers me is that she smokes, from two different deep structures, and the derivation of nonsynonymous sentences, such as the last mentioned and what bothers me is that it bothers me that she smokes from the same deep structure. (The reader can verify for himself that this is so; the detailed exposition will not be given here. The optional transformations relevant to these examples are Passivization and Extraposition.) Another defect in Peters and Bach's analysis involves Complementizer Placement. This rule must obviously not apply to the second embedded sentences in Peters and Bach's proposed deep structures (e.g., to insert a that after it in (42)), yet how can this be prevented without some ad-hoc restriction on the rule? A further class of sentences that presents a problem for Peters and Bach's proposal is exemplified in (46).

Deriving S from S + Is

(46)

267

a. What you heard something bump into was my head. b. What you heard bump into something was my head.

Peters and Bach's deep structures for these sentences would have as the subject NP of the matrix sentence the thing [ 8 you heard something bump into something] 8 • After Relative Clause Formation this NP will be either the thing that you heard something bump into or the thing that you heard bump into something. It is important to note that the choice of which particular reading of the Relative Clause Formation rule is applied must be made to depend, in Peters and Bach's analysis, on the sentence embedded in the predicate of the matrix sentence in deep structure. Thus if the sentence embedded in the predicate of the matrix in deep structure is you heard something bump into my head, then the Relative Clause Formation rule operating on the subject NP of the matrix sentence may only produce the thing you heard something bump into; it may not produce the thing that you heard bump into something. The reverse is true if the sentence embedded in the predicate of the matrix sentence in deep structure is you heard my head bump into something. Peters and Bach's analysis, then, forces one to formulate a somewhat complicated, unusual, and ad-hoc restriction on the applicability of the Relative Clause Formation rule. We see that Peters and Bach's proposal necessitates the stipulation of a number of ad-hoc restrictions or extensions to such transformations as Complementizer Placement and Relative Clause Formation, and to all optional transformations. 2 And the restriction that would have to be placed on Relative Clause Formation to account for sentences such as (46) would do exactly what Peters and Bach were trying to avoid in principle; it would state a condition on the applicability of a transformation in terms of "corresponding deep structure." Note that the Chomsky-Emonds analysis, with deep structures such as (36), runs into none of these difficulties. The structures underlying the problematic sentences we have discussed would be roughly as follows. 2 Stanley Peters has pointed out to me that most, though not all, of the syntactic problems referred to here can be overcome if one makes use of the convention proposed by Chomsky in Aspects which "filters out" those final derived phrase markers which contain sentence boundary symbols. There is, unfortunately, no space to argue in detail about the boundary symbol convention here, but it appears to me to be an ad-hoc and arbitrary device used to describe, rather than to explain, a variety of unrelated phenomena, and capturing no clearly significant generalizations. Boundary symbols have neither semantic significance nor phonetic correlates and sentence boundaries are in any case adequately indicated by labeled bracketing.

268

James R. Hurford

s

(47)

s

N

I

it

t-r VP

Aux

NP

be

That she smokes bothers me. That it bothers me that she smokes bothers me. You heard something bump into my head. You heard mq head bump into something.

~

The constituents to be pseudo-clefted are in bold italic. Peters and Bach consider another class of analyses beside Chomsky's extracting analysis (so called because it extracts the pseudoclefted constituent from a representation of the unclefted simple sentence). The defects which they note in this other class of analyses do indeed exist and are, as Peters and Bach remark, "extremely easy to discover [p. 7]" I will not, therefore, discuss them here. I am aware of only one other proposed analysis of pseudo-cleft sentences, which is alluded to briefly in UESP (p. 838) and wrongly attributed there to Kuroda (1968). Its actual origin is unknown to me. It is similar to Peters and Bach's proposal and suffers from some, though not all, of the defects I have noted above. The proposed deep structure for What John counted was the pigeons is (48). (48)

s

A

~VP

NP

I

TNS

A

BE

NP

ALAL

the

thing John counted the pigeons John counted something the pigeons

This proposal does not have the defects associated with the complementation in the predicate of the matrix sentence noted in connection with Peters and Bach's proposals. The rule of Complementizer Placement, for example, does not have to be restricted in any

Deriving S from S + Is

269

was as a result of this proposal. The actual rule of Pseudo-clefting in this proposal is not mentioned in UESP, but optional transformations are likely to be problematic as in Peters and Bach's proposal. And this proposal seems sure to encounter the same difficulties as Peters and Bach's analysis with sentences such as those of (46). There are a number of problems which remain to be solved in any analysis of pseudo-cleft sentences, for example, "that of deriving both the it of the cleft and the what of the pseudo-cleft from a common deep structure in a well-motivated way [UESP, 839]", that of characterizing correctly the class of elements that can be pseudoclefted, and that of the occurrence of certain reflexives in pseudo-cleft sentences. Despite these matters, I believe that I have shown the Chomsky-Emonds deep structure, namely something like (36) and (47), to be certainly the most likely to be correct underlying form for pseudo-cleft sentences. Actually the Chomsky-Emonds structure may be incorrect in particular details that are beyond the scope of this paper. The presence of the node Pred in the VP of the ChomskyEmonds structure is such a detail. My concern in this paper is to argue for the presence of some form of the verb be in the VP of a higher sentence and the question of whether or not there is also a Pred is irrelevant to this. 3 Let us proceed on the assumption that structures like (36) and (47) are indeed correct for pseudo-cleft sentences. It remains to be shown that all sentences must be derived from structures such as these. This might be argued in the following way. The semantic difference between Lighting struck the house and What struck the house was lightning is a difference of focus. The other components of the meanings of these sentences are the same. Focus is determined, as we have noted, elsewhere than in deep structure. Components of meaning other than focus (and perhaps topic) are determined in deep structure. Therefore the deep structure of Lightning struck the house is identical to that of What struck the house was lightning, and in general the deep structure of pseudocleft sentences is identical to that of the corresponding noncleft sentences. (I am, for concreteness, begging an important question here; for a qualification of this conclusion, see below, p. 290.) This means that the deep structure of every simple sentence which has a corresponding pseudo-cleft can be a complex structure with an embedded S and with some form of the verb be in the predicate of the matrix S. Thus the evidence from pseudo-cleft sentences can be said to support the proposal that declaratives, interrogatives, and probably also imperatives can have just such deep structures. Cf. (49). 3 It could be that the dummy Pred of Chomsky's analysis is the phonologically null form of the predicates true, a fact, the case, so, etc.

270

James R. Hurford

a. We want Watneys. b. What we want is Watneys. c. Do we want Watneys? d. ls what we want Watneys? e. Take courage! f. What to do is take courage. g. What to take is courage.

(49)

Some justification should be given for citing (49f) and (49g) as pseudo-cleft versions of the imperative (49e). There are clear semantic differences between these sentences. (49e) has obvious imperative force which is lacking in (49f) and (49g). (49f) and (49g) are interpreted as descriptions of a recommended course of action, rather than as commands. One might say, loosely, that the pseudo-clefting operation carried out on (49e) strips it of its performative force. But note that this seems in general to be the case when overt performative sentences are pseudo-clefted. (50)

a. I (hereby) promise to do better next time. b. What I (?hereby) promise is to do better next time. c. I (hereby) warn you that I will tolerate no nonsense. d. What I (?hereby) warn you is that I will tolerate no nonsense. e. I (hereby) request you to tell me what time it is. f. What I (?hereby) request is for you to tell me what time it is.

If one questions the validity of the hereby diagnostic, there can still be no doubt that there is a semantic difference between the pseudo-cleft sentences of (50) and the corresponding uncleft sentences. Ignoring the herebys in these sentences, (50a), (50c), and (50e) actually constitute a promise, a warning, and a request respectively, whereas (50b), (50d), and (50f) presuppose respectively a promise, a warning, and a request. A well known syntactic fact about imperatives is that the only reflexives that occur with them are second person. (51)

a. b. c.

d. e.

f.

g.

Pour yourself a drink. Pour himself a drink. 0 Pour themselves a drink. What to do is pour yourself a drink. 0 What to do is pour himself a drink. 0 What to do is pour themselves a drink. Dry yourself with this towel 0

Deriving S from S + ls

271

h. °'Dry himself with this towel. i. °'Dry themselves with this towel. j. What to do is dry yourself with this towel. k. °'What to do is dry himself with this towel. I. °'What to do is dry themselves with this towel. Geoffrey Leech has pointed out to me an exception to the general pattern illustrated in (51). Although one can say, at least in British English, What to do is dry oneself, the corresponding simple sentence °'Dry oneself is ill-formed. I surmise that this anomaly has to do with some peculiar property of the impersonal pronoun one. To my knowledge, no sufficiently detailed study of this pronoun has been made for us to be able to see just how it fits into the English pronoun system in general. Embarrassing as this gap in our knowledge is, the irregular behavior of one is not enough to cause us to dismiss the significant regularities exemplified in (51). Stative verbs and adjectives in pseudo-cleft sentences like (49g) seem peculiar in the same way as stative verbs and adjectives in imperatives, while active verbs and adjectives are acceptable in both constructions. (52)

a. °'Be tall. b. °'What to be is tall. c. °Know the answer. d. °'What to know is the answer. e. Be honest. f. What to be is honest. g. Eat an apple. h. What to eat is an apple.

Furthermore, both imperatives and pseudo-clefts such as (49f) may not occur with forms expressing past time. (53)

a. °'Give it to me last week. b. °'What to do is give it to me last week. c. Give it to me tomorrow. d. What to do is give it to me tomorrow.

These and other similarities make it very plausible that sentences such as (49f), and (49g) are in fact the pseudo-clefted versions of imperative sentences. Imperatives, then, fit in well with the generalization proposed in this paper, namely that the deep structure of all sentences can contain an embedded S in a matrix sentence which also contains some form of the verb be.

272

James R. Hurford

Insofar as cleft sentences are most probably derived from pseudoclefts (see Akmajian, 1970), the existence of cleft sentences also provides support for the general proposal of this paper. PRESUPPOSITIONS AND NEGATION

Another argument for the general proposal of this paper can be constructed out of some apparent inconsistencies in the relationships which are normally said to obtain between meaning and presupposition. A summary of what I believe linguists typically understand to be the relationship between meaning and presupposition is given in the following passage from Fillmore (1971). Sentences in natural language are used for asking questions, giving commands, making assertions, expressing feelings, etc .... We may identify the presuppositions of a sentence as those conditions which must be satisfied before the sentence can be used in any of the functions just mentioned. Thus the sentence identified as (54): (54)

Please open the door

can be used as a command only if the [hearer] is in a position to know what door has been mentioned and only if that door is not, at [the time of speaking], open. The test that the existence and specificity of a door and its being in a closed state make up the presuppositions of (54) rather than part of its meaning is that under negation the sentence is used to give quite different instructions, yet the presuppositional conditions are unaffected. (55) [= Fillmore's (54')]

Please don't open the door

[p. 380]

There are, then, (at least) two separate criteria for deciding what aspects of semantic interpretation are presuppositions. They are: (a) the conditions that must be satisfied in order for a sentence to be deemed appropriate, and (b) those aspects of semantic interpretation that remain constant under negation. In the example given by Fillmore, both these criteria lead to the same conclusions. The coincidence in many cases of criteria (a) and (b) comes as a satisfying confirmation that a distinction which is of usefulness and explanatory value can be identified in a clear way. I assume that the importance of the distinction between meaning and presupposition is not in doubt even if attempts to characterize it in a clear way are sometimes unsuccessful. But linguists must, of course, try to develop theories in terms of which completely reliable statements can be formulated relating to the distinction between meaning and presupposition. I argue below that criterion (b) (i.e., that the presuppositions of a sentence are those aspects of its interpretation which remain constant under negation), is an imperfect approximation to the truth, and that the truth can only be expressed if we accept deep structures for simple sentences such as are proposed in this paper.

273

Deriving S from S + Is

The problem is that "under negation" is not characterized clearly enough. Consider sentence (56), spoken with a falling intonation on pilot and a low level pitch on the rest of the sentence. (56)

The pilot wasn't killed.

This sentence, with the given intonation contour, might provide an answer to the question Who wasn't killed? and can be paraphrased by It was the pilot who wasn't killed. Now one of the presuppositions of sentence (56), i.e., one of the conditions for its appropriateness, can be expressed as (57). (57)

Some (particular but unspecified) person wasn't killed (on the occasion in question).

If we regard the negation of a simple negative sentence as the corresponding simple affirmative (with the same intonation contour), then the negation of (56) is (58). (58)

The pilot was killed.

But (57) is not one of the presuppositions of (58). We must be more precise about what we mean by "presuppositions being constant under negation." One obvious flaw in the above discussion of examples (56) and (58) is that it is couched rather vaguely in terms of "negative sentences" and the "corresponding affirmatives." For the sake of preciseness, we should prefer to talk of deep structures containing the element Neg and deep structures identical but for the absence of the element Neg. This is entirely appropriate, since (we assume) negation is an aspect of the interpretation of a sentence determined in its deep structure. For preciseness also, we should represent presuppositions as base phrase markers, using some appropriate dummy symbol to indicate particular but unspecified elements, as in (57), where necessary. In talking about deep structures we must remember that a given deep structure does not uniquely determine a set of presuppositions. As we have seen, sentences with different focuses and therefore different presuppositions may be derived from the same deep structure. Nevertheless, it is still possible to speak of presuppositions being appropriate or otherwise to given deep structures. For example, the presupposition The door is closed is inappropriate to the deep structure of Please close the door, while the presupposition The door is open is appropriate to it. Most possible presuppositions are, of course, neither appropriate nor inappropriate to a given deep structure. I shall say that such presuppositions are irrelevant. For example, a presupposition such as Lloyd George knew my father is irrelevant to

274

James R. Hurford

the deep structure of Please close the door. The presuppositions of a given (surface) sentence are a subset of those which are appropriate to its deep structure. If we postulate simple deep structures for both (56) and (58), with only one S in each, and differing only in that the deep structure of (56) contains the element Neg, whereas the deep structure of (58) does not, then our problem remains. The presupposition (57) is appropriate to the deep structure of (56), but is merely irrelevant to that of (58). That is, for me to regard (58) as appropriate, I need not necessarily have (57) in mind, since what is relevant to (58) is not who wasn't killed but rather who was. But let us now say that the deep structure of (56) is as in (59). (59)

Neg the pilot was killed

Remember that I have argued that pseudo-clefts are derived from deep structures such as this. (59), then, is not only the deep structure of (56), but also of The one who wasn't killed was the pilot and It was the pilot who wasn't killed. The presupposition (57) is appropriate to deep structure (59). Now the deep structure of the negation of (56) will be as in (60). (60)

s

~ NP

VP

s

is

Neg

I

I

~ Neg the pilot was killed

Sentences derived from (60) include The one who wasn't killed wasn't the pilot and It wasn't the pilot who wasn't killed. No acceptable simple surface sentences are derived from (60): the nearest perhaps is the very dubious 0 The pilot wasn't not killed. Note that the pre-

Deriving S from S + Is

275

supposition (57) is appropriate to (60). This argument can, of course, be repeated for many simple declarative sentences. INTERROGATIVES

There is more evidence from English for the type of deep structure that I am proposing, but I will delay presentation of it until after I have made some remarks about French. If one accepts the hypothesis that there is a universal base component, then evidence for these deep structures in any language is evidence for them in all languages. There is very clear and transparent evidence for these deep structures in French. Langacker (1965, 1971) has given a thoughtful and comprehensive account of the basic mechanisms of French interrogative formation. He derives questions such as (61) from sentences such as (62), which have sentential complements of c'est que.

a. Est-ce que cette femme est folle? b. Est-ce que son auto lui a coute cher?

(61) (62)

a. b.

C' est que cette femme est folle. C' est que son auto lui a coute cher.

The questions of (61) are formed by the same transformational processes of reduplication, pronominalization, and ellipsis as the synonymous questions (63).

a. Cette femme est-elle folle? b. Son auto lui a-t-il coute cher?

(63)

Since the questions of (61) and (63) are synonymous, they are derived from the same deep structure, namely that of the declaratives in (62) with the addition of whatever is needed to represent interrogation, i.e., something like (64).

s

(64)

VP

Q

ce

est

~ COMP I

s

~

cette femme est folle

276

James R. Hurford

Langacker shows such a structure to be independently justified by evidence from cleft sentences such as (65), which he derives by what Peters and Bach would call an extracting analysis, not dissimilar to Chomsky's treatment of pseudo-clefts. (65)

a. b.

C'est le medecin que tu vois. C' est cette femme qui est Jolie.

A French construction not mentioned by Langacker is the tag question, which might also be considered to provide evidence for deep structures such as those I am proposing in this paper. The argument depends, however, on the unproven notion that there are some universal principles governing tag-question formation, and so would not be persuasive without the supporting evidence given above. I conjecture that the following properties are likely to be found in tag questions generally. (66)

Negative tags follow affirmative declaratives and vice versa.

(67)

The tag repeats, either literally or in pro-form, certain elements, often the subject and the main verb, of the declarative sentence preceding it.

(68)

Tag questions are formed by similar rules to those that produce nontag questions, e.g., inversion, rising tone.

These statements hold true for English. (66) and (68) hold true for French, but what of (67)? (67) could only be true of French if we postulate that there is at some stage in the derivation of every French affirmative declarative an occurrence of ce est. This is so because tag questions in French are of the form given in (69). (69)

a.

b.

Cette femme est Jolie, n'est-ce pas? Tu vois le medecin, n'est-ce pas?

As with every topic in grammar, there remain problems connected with the analysis of the French constructions I have mentioned, including some quite fundamental metatheoretical ones. Langacker (1971) draws attention to what he calls "syntactic idioms" of which strings like est-ce que and qu'est-ce que are examples. N' est-ce-pas? is also a syntactic idiom. There is not yet any theory of how syntactic idioms should be described, but their existence is not in doubt. This fact makes the empirical significance of claims about deep structure less clear, but I believe that they can still, and indeed must still, be made. Langacker concludes his 1971 paper with the provocative remark, with which I agree, "If c' est que did not exist, linguists would have to invent it [p. 40]."

Deriving S from S + Is

277

Tag questions in some other languages seem to present evidence for an abstract higher sentence such as I am here proposing. It could be argued that the German tag nicht wahr (literally 'not true') repeats a higher predicate wahr 'true' which is present in the deep structure of every sentence. I.e., the deep structure of Sie sind ein Englander, nicht wahr 'You are an Englishman , not true' might be argued to contain something like [Sie sind ein Englander] ist wahr. In some dialects of English spoken in Wales, sentences like We are going to the pictures, isn't it are grammatica l, the constructio n here apparently paralleling one found in Welsh. The constructio n obviously resembles the French n' est-ce pas and, like it, can be taken to provide evidence for the higher sentence I am arguing for in this paper. For most interrogativ es English does not use the equivalent of French est-ce que. Thus for most people the examples in (70) are quite unacceptab le. (70)

"'Is it that he knows where to look? "'Is it that he eats beef? "'Is it that we are going?

a.

b. c.

But in a number of cases, as I will show, including those where a declarative contains a negative and an adverb such as just or simply, the correspond ing interrogativ es formed with is it that are quite acceptable and probably even somewhat more acceptable than the correspond ing interrogativ es formed without is it that. (71)

a.

b. c.

(72)

(73)

He just doesn't know where to look. It is that he just doesn't know where to look?

Does he just not know where to look?

He just won't eat beef Is it that he just won't eat beef? c. Will he just not eat beef?

a. b.

a. We simply can't go. b. Is it that we simply can't go? c. Can we simply not go?

Examples (71c), (72c), and (73c) here are more stilted, less natural than (716), (726), and (736) in my dialect. Questions formed without is it that seem even less acceptable in cases where a negative and an adverb such as just or simply are found in rather idiomatic expressions . (74)

a. b.

It is that they just can't help it? Can they just not help it?

278

James R. Hurford

(75)

It is that he simply doesn't give a damn? Does he simply not give a damn?

a.

b. (76)

a.

b.

Is it that you just couldn't care less? Could you just not care less?

Well-formed interrogatives with is it that, some of which are less stilted than their counterparts without is it that, are found in a wide variety of syntactic environments, as is illustrated below. (77)

a. b. c.

d. e. f.

g.

h. i.

j. k.

Is it that he sometimes doesn't let people watch him? Is it that scarcely anybody comes to church these days? Is it that you often don't need to help to park the car? Is it that many of your friends are Catholic? Is it that you have lost all your money? Is it that your husband is a drunkard and beats you? Is it that Fred is an alcoholic and we never knew it? Is it that Seymour sliced the salami with his penknife? Is it that her husband might know the answer because he's an expert? Is it that he takes a bath at ten o'clock every night? Is it that Johnson has found the solution we've all been looking for?

I. Is it that the Martians have landed? m.

Is it that his car has broken down again?

Some of these examples are ambiguous. They can be requests for information either about the factuality of some event or state of affairs or about whether some event or state of affairs, whose factuality is presupposed, is the reason or cause for some other situation being discussed. That is, in one sense they may be paraphrased by is it the case that ... and in the other sense they may be paraphrased by is it because ... Natural answers to (77k), for example, might be No, he merely thinks he has or No, that's not why we're celebrating. In the latter case the it of the question appears to have anaphoric or deictic reference. All the interrogatives with is it that given above can have the first interpretation. The ambiguity of the sentences of (77) is complicated by the fact that these interrogatives may apparently be used with both interpretations simultaneously. In other words, sentence (77j), for example, could in one of its uses be paraphrased as (78). (78)

Does he take a bath at ten o'clock every night and, if so, is that the reason for the situation we are discussing?

Deriving S from S

+ Is

279

Questions with is it that are, then, similar to the more familiar sort like (79). a. Would you like tea or coffee? b. Do you know where John went?

(79)

These sentences are also ambiguous. In one interpretation of (79a), for example, the speaker presupposes that the hearer wants tea or coffee and asks which of the two he wants. In the other interpretation, the speaker asks whether the hearer would like some beverage and names the two beverages available. Thus appropriate answers include Tea, please and Yes, please. Similarly, appropriate answers to (79b) include He went to the library and Yes. The simple yes answers to these questions are certainly uncooperative, but the fact that they are pedantic is evidence in favor of, rather than against, their linguistic appropriateness. But it is also possible to use the questions of (79) with both interpretations simultaneously, i.e., to mean what is expressed in (80). (80)

a. Would you like tea or coffee, and if so, which? b. Do you know where John went, and if so, where did he go?

Thus other possible responses to the questions of (79) are Yes please, tea and Yes, he went to the library. And similarly, a natural response to (77k), beside the two I have already mentioned, is Yes, but that's not why we' re celebrating. I have no idea what deep structures the double-barreled interpretations of these questions correspond to. The purpose of this digression is to attempt to clarify the sense in which one can isolate interpretations of interrogatives with is it that in which what is questioned is just the factuality of the embedded proposition. Further evidence for the isolability of these interpretations is provided by discourses as in (81), in which questions with is it that are acceptable, but questions with is it because are actually unacceptable. (81)

a.

CUSTOMER: WAITER: CUSTOMER:

b.

FRED: JOHN: FRED:

I'd like escargots, please. The escargots are not good tonight, sir. Is it { llcb that } you have something else ecause specific in mind?

Why do you dislike that girl so? Who said I disliked her? Or is it {l),b th at } you really like her a lot? ecause

280

James R. Hurford

The following quotation, found in Barker (1964), also illustrates the use of a question with is it that which clearly does not mean is it because. (82)

But if mathematicians permit the development of alternative geometries whose laws contradict those of Euclidean geometry, what has become of the notion of truth in mathematics? Can it be that these conflicting geometries are equally true? Or is it that mathematicians no longer even seek the truth about space? [pp. 37-38]

Interpretations paraphrasable by is it the case that ... are the ones that concern us here. The derivations involving the double-barreled and is it because . . . readings do not concern us. It is clear that questions with is it that, as in (71)-(77) can be derived by the familiar rules of Extraposition and Question Formation from deep structures such as I propose in this paper. I am unable to give a satisfactory account of why some questions with is it that, e.g., those in (70) are actually not well-formed, but this does -not weaken the claim that acceptable is it that questions, such as those of (71)-(77) are derived from the proposed deep structures. FACTIVE COMPLEMENTS

It is not plausible to relate the NP the fact that Harry was here to any hypothetical sentence (>Someone facted that Harry was here. In the case of the NP the idea that Harry was here such a step is quite plausible because of the existence of another NP someone's idea that Harry was here. This latter NP can be produced by a general process which possessivizes the subject of a nominalized sentence, e.g., Fred reported that Harry was here ? Fred's report that Harry was here. But we do not find NPs like (>Fred's fact that Harry was here. This is evidence that the underlying subject of fact is not a dummy element, of the kind represented here informally as someone. Consider now the sentences in (83).

(83)

a. b.

That Harry was here is a fact. It is a fact that Harry was here.

It is clear from these sentences that the underlying subject of (be a) fact must be the sentence Harry was here. Now the usual deep structure postulated for an NP such as the fact that Harry was here (e.g., Kiparsky and Kiparsky, 1971; Rosenbaum, 1967) is something like (84).

Deriving S from S + Is

281 NP

(84)

Det~S

,L

1L~

Harry was here

Such a structure is semantically inadequate, since it does not represent the fact that the sentence Harry was here is the subject of (be a) fact. To clarify this point with an analogy, consider (85) and (86) below, which are the deep and surface structures respectively of the NP the man that I saw, as frequently envisaged. (85)

NP NP~S

D

,l~~.. (86)

I saw the man NP

L

~

the man

that I saw If we ask which of these two structures is more adequate as a basis for semantic interpretation, the answer is quite clearly (85), because only in (85) is the proposition I saw the man explicitly represented. We can compare structure (84) to (87) below in the same way. (87) NP NP~S th~ct

NP~VP

~

~ Harry was here

/\

is

afact

282

James R. Hurford

In (87), but not in (84), the proposition that Harry was here is a fact is explicitly represented. (87) is clearly more adequate as a basis for semantic interpretation, i.e., as a deep structure. And if we adopt (87) as the deep structure of the fact that Harry was here we can achieve an economy in the base component, since we will not now need to generate any base phrase markers of the form Det N S. All structures of this form in English are derived from structures of other kinds, e.g., the man that I saw is derived from something like (85), the belief that Harry was here is derived from some sentence-like structure in which the dummy subject of belief/believe is represented explicitly, cf. Someone believes that Harry was here. (I do not commit myself here to either the lexicalist or the transformationalist position.) I claim, then, that NPs of the form the fact that S have the deep structure of relativized NPs, the relative clause being S is a fact. I am not sure how the transformational derivations of such NPs should proceed. There are several possibilities. One might simply delete the VP dominating is a fact in (87), for example, leaving, after the necessary tree-pruning and insertion of that, a plausible surface structure. If we assume that is a fact and the bare is discussed in previous sections have the same semantic content, then this deletion rule can be justified on grounds that are presented later (p. 288). Alternatively, one might apply the regular processes of relativization to structures like (87), treating them in a way parallel to the expressions after not in the examples below. (88)

a. Russell was not the genius (that) Einstein was. b. John is not the rogue (that) his brother is. c. London is not the hell-hole (that) New York is.

If we adopt this treatment, then the is in a structure like (87) still has to be deleted in order to produce the surface NP the fact that Harry was here. I shall argue in the next-to-last section that there is reason to postulate such a deletion rule. Note that the arguments of this section are not tied to any specific view of the deep structure of relative clauses. I have assumed a widely held view for expository reasons. If, for example, relative clauses turn out to be derived from deep structure conjunctions, as proposed by Thompson (1971), diagram (87) in particular will have to be rejected as a deep structure, but arguments parallel to those given in this section for the existence of S is a fact in deep structure could still be made. The distinction between relative clauses and complements introduced by that is sometimes illustrated by the impossibility of con-

Deriving S from S + Is

283

joining them, as in °The idea that struck me in the bath this morning and that Nixon is a woman. An objection to the analysis proposed here could be raised by pointing to the impossibility of expressions such as (89). (89)

0

The fact that you mentioned and that Harry was here.

I postulate that both the fact that you mentioned and the fact that Harry was here are derived from NPs containing relative clauses where the noun fact is the relativized element. Given this analysis, to what can we attribute the ill-formedness of (89), since in general it appears that relative clauses can be conjoined, e.g., the lady that I saw and that John raised his hat to? Note, however, that conjoined relative clauses as in (90) are not well-formed. (90)

a.

b.

The genius that Einstein was and that discovered relativity. 0 The genius that discovered relativity and that Einstein was. 0

The ungrammaticality of these expressions can be attributed to a constraint to the effect that a clause relativizing a predicate NP, e.g., genius, fact, cannot be conjoined with another relative clause. It is to this constraint that we can also attribute the ungrammaticality of (89). In concluding this section, I remind the reader that NP' s of the type the fact that S play a far more substantial role in the system of English complementation than might be apparent from their relatively low frequency in surface structure. Kiparsky and Kiparsky (1971) have argued fairly convincingly that a very wide range of complements are to be analyzed as the fact that S in deep structure even though there may be no actual occurrence of the fact that in surface structure. Thus structure (87) will, I claim, be found as a constituent of the deep structures of all of the following examples, and of many more. (91)

a. We all regret that Harry was here. b. We are aware that Harry was here. c. It's odd that Harry was here. d. It doesn't bother me that Harry was here.

SENTENCES WITH MODALS

Parallel to the interrogatives dealt with in a previous section there are declaratives with modals which must clearly be derived from

284

James R. Hurford

embedded structures in which the verb of the matrix S is be. Examples of such sentences are given in (92). (92)

a. b. c.

d. e.

It could be that John is there already.

It may be that we made a mistake.

It must be that you are mistaken.

It might be that the plane arrived late. It cannot be that I know your long-lost nephew.

These sentences appear to mean the same things respectively as those of (93) below. (93)

a. John could be there already. b. We may have made a mistake. c. You must be mistaken. d. The plane might have arrived late. e. I cannot know your long-lost nephew.

The point to be made in connection with these sentences is similar to that made in connection with the interrogatives discussed earlier. Whatever semantic differences there are between the sentences of (92) and (93), they are evidently quite subtle and it is possible that these differences should not be represented at the level of deep structure. In this case the deep structures of both types of sentence will be identical, i.e., something roughly like (94).

s

(94) NP

I

S

.-----r----_ Aux

VP

could

be

I

~

I

John be there already Syntactic arguments exist in favor of this treatment. In fact no new rules need be postulated in order to derive John could be there already from structure (94). It can be derived by the application of Extraposition, It Replacement, and Be Deletion. The stages in the derivation are roughly as follows. (95)

[John be there already] 8 could be It could be [John be there already]s c. John could be be there already d. John could be there already. a. b.

285

Deriving S from S + Is

This derivation is essentially the same as that proposed in the wellknown analysis of sentences like John seems happy, which is derived roughly as follows. (96)

a.

[John be happy] 8 seems

b. It seems [John be happy] 8 c. John seems be happy d. John seems happy. The only general difference between derivations with seem and those with could be are differences in the optionality or obligatoriness of the rules of Be Deletion and Infinitivalization. These conditions depend in any case on idiosyncratic properties of individual lexical items, as is clear from examples like John seems (to be) happy versus John happens to be happy, "'John happens happy, and John should ("'to) go versus John ought to go, "'John ought go. Notice that the alternation of past tense and perfect aspect met in the sentences of (92) and (93) is also met in the familiar examples with items such as seem. Thus just as It may be that we made a mistake alternates with We may have made a mistake, so It seems that we made a mistake alternates with We seem to have made a mistake. However this alternation is to be described, it is obviously desirable to provide a single account of it, treating modals as in (92) and (93) in the same way as verbs like seem, be certain, be likely, etc. Similarly, of course, it is desirable to provide a single account of the alternation in subject position of it and some other NP in sentences with modals and those with seem, be certain, be likely, etc. These are syntactic arguments which militate in favor of deriving both the simple sentences of (93) and the complex sentences of (92) from the same deep structures. And the deep structures that appear to be called for are as in (94), i.e., embedded structures with the verb be in the matrix sentence.

THAT IS FOR YOU TO SAY

There is another class of sentences that can be derived from underlying S be without postulating any new rules. These are sentences like (97) and (98). (97)

a.

It is for the governor to settle this matter.

b. It is for you to say that. c.

It is for the vice-chancellor to decide on promotions.

286 (98)

James R. Hurford

a. This matter is for the governor to settle. b. This is for you to say. c. Promotions are for the vice-chancellor to decide on.

These sentences all express the social appropriateness of some party fulfilling some particular role, e.g., of the governor being the party to settle this matter. There is thus a paraphrase relationship between the sentences of (97) and (98) and those of (99) on one reading of the latter. (99)

a. The governor should settle this matter. b. You should say that. c. The vice-chancellor should decide on promotions.

The paraphrase relationship between the sentences of (97)-(99) needs to be accounted for, and if we can derive them all from a common deep structure by rules that are independently motivated, we shall have succeeded in doing it. Now notice the paraphrase relationship between the following sets of sentences.

a. It is outrageous for you to say so. b. It's a gas for you to think of me as your uncle. C. I am anxious for you to go. d. I pray for a miracle to save him.

(100)

(101)

a. It is outrageous that you should say so. b. It is a gas that you should think of me as your uncle. c. I am anxious that you should go. d. I pray that a miracle should save him.

Common deep structures for the sentences of (100) and (101), are quite plausible since in both sets of sentences there is a surface complement of a common predicate, e.g., outrageous, a gas, etc., and these sentences are synonymous. I assume, then, a transformational relationship between some complements with for ... to and some with that ... should. Let us postulate that the derivations of sentences like (100) and (101) proceed in the following way. (102)

a. b.

Remote structure: [ you should say so] 8 is outrageous by Extraposition: it is outrageous [you should say so ]8

Now either the complementizer that is inserted and no further transformations apply, yielding sentences like those of (101), or else the subject NP of the complement is shifted into an oblique case (i.e., for is inserted), Infinitivalization (insertion of to) follows, and

Deriving S from S

+ Is

287

should is deleted, yielding finally sentences like those of (100). Insertion of for and that and Infinitivalization are fairly generally accepted transformational rules: the only relatively unfamiliar step here is the deletion of should. It seems that some such step is necessary in order to account for the paraphrase relationship between the sentences of (100) and (101). The deletion of should parallels the deletion of will discussed in the first section. The Kiparskys (1971) explicitly assume a rule deleting a future should in complements of what they call "emotive" predicates, e.g., outrageous, a gas, anxious, etc. Now it is clear that the sentences of (97) can be derived by the same process from a deep structure such as (103).

s

(103)

~VP

NP

I

I

~

the Governor should settle this matter

Applying Extraposition, For Insertion, Infinitivalization, and Should Deletion to this structure we get It is for the governor to settle this matter. The sentences of (98) can now be derived from here by the familiar rule of Tough Movement (raising-to-subject). That is, (98a), for example, is derived from (97a) in the same way as This matter is hard for the governor to settle is derived from It is hard for the

governor to settle this matter. Simple surface sentences like those of (99) can be derived from structures such as (103) just by deleting the VP is. I shall argue later (p. 288) that such a deletion is well motivated. It is also quite possible that the sentence This matter is for settling by the governor should be derived from structure (103). The derivation would proceed by Passivization of the embedded sentence, Extraposition, insertion of for, deletion of should, and It Replacement. This would yield a structure like this matter is [for be settled by the governor] s• This structure is converted finally into This matter is for settling by the governor by the same rule that is responsible for capturing the relationship between, for example, That boy needs to be walloped by a kid his own size and That boy needs walloping by a kid his own size. This analysis seems promising, but there are

288

James R. Hurford

problems, such as the detaching of this matter from the inserted preposition for by It-replacement and the ungrammaticality of an example like (;That is for saying by you. An objection that might be raised to the proposed analysis of sentences (97)-(99) is that be is clearly not an emotive verb; whereas according to the Kiparskys (1971), for is inserted into just those complements of predicates which "express the subjective value of a proposition rather than knowledge about it or its truth value," e.g., odd, tragic, hate. The Kiparskys' claim is incorrect, however. There are predicates which express a subjective judgment and which do not take a for in their complements, e.g., (; I want for you to do it, (; I like for you to do that. And there are predicates whose complements take for which do not express a subjective judgment, e.g., I'll arrange for a taxi to meet you, It is illegal for this paper to be Xeroxed without my permission, It is a felony for nonmilitary personnel to carry arms without a license. It seems that the insertion of for must be sensitive to idiosyncratic properties of individual predicates. Another objection that might be raised to this analysis is that the should postulated for the underlying structures of (97)-(99) is the should of social appropriateness, whereas the should postulated for (100) and (101) is the subjunctive should which seems to carry little meaning, if any, in these contexts. This objection has no weight, since the transformation deleting should is sensitive solely to its form, and not to its meaning. The difference in meanings between the various shoulds is presumably determined by the contexts in which they occur in deep structure. IS DELETION

In this section I will discuss the validity of postulating the deletion rule accounting for the absence of It is that in simple sentences. It appears that one it is that clause can be embedded in another, since pseudo-cleft sentences can appear inside not that clauses. (104)

a.

b.

Phil asked his boss for a raise (not that what he really needs is more money). I admire Spiro's tenacity (not that what I chiefly look for in a vice-president is tenacity).

If one it is that clause may be embedded in another, then it should logically be possible to find clefts and pseudo-clefts embedded in each other. And in fact we do find examples such as the following:

Deriving S from S

(105)

+ Is

289

a. It was lightning that was what struck the house. b. What it was that struck the house was lightning. c. Lightning was what it was that struck the house.

Other conceivable combinations of clefts and pseudo-clefts which are ungrammatical can be excluded in a principled way, but I will not elaborate on this here. It is quite possible that it is that clauses can be self-embedded indefinitely. What is one to make of the following sentences? · (106)

Phil asked his boss for a raise (not that what it is that he really needs is more money). b. It was lightning that was what it was that struck the house. a.

These examples do not strike me as particularly bad, though it is hard to think of contexts in which they would be completely appropriate. The conclusion that it is that clauses may be indefinitely selfembedded does not entail a prediction of infinite ambiguity for innocuous sentences. Consider the similar case of it is true that. (107)

It is true that prices slumped. It is true that it is true that prices slumped. It is true that it is true that it is true that prices

slumped.

etc. The aspect of semantic interpretation determined in deep structure for all these sentences must be the same (which is not to claim, of course, that these sentences are semantically identical). It is natural to apply the same conclusion to the deep structure aspect of the semantic interpretation of it is that clauses. It is that clauses contribute nothing to the deep structure aspect of semantic interpretation, but rather appear to provide a frame in which certain choices affecting the derived structure aspect of semantic interpretation can be registered. Ross (1970, p. 251, p. 271) mentions that some syntactic rules can apply to their own output, and if this is the case with the transformational rule that deletes the higher it is that, no syntactic problem at all arises from the ability of such clauses to be indefinitely self-embedded. Any theory that contains the trivial

290

James R. Hurford

calculus needed to account for the deep structure synonymy of the sentences of (107) can also account for the deep structure synonymy of multiply embedded it is that clauses. This calculus, or semantic rule, would presumably look something like (108). (108)

S2

~

VP

NP

I

s1

/\ (true)

is semantically equivalent to S,.

is

This rule is relevant only to the aspects of semantic interpretation determined in deep structure. But to postulate both a transformation, capable of applying to its own output, for deleting it is that clauses, and a trivial calculus accounting for the semantic nonsignificance of it is that clauses would be to do the same work twice in a grammar. In the case of simple sentences like prices slumped, why must one postulate a deep structure paraphrasable as it is that prices slumped, only to have both the semantic and the syntactic (ultimately phonetic) effects of postulating the higher it is that clause nullified by separate semantic and syntactic devices? It might appear that we should try to eliminate one or other of these devices, but I shall argue below that neither of them can in fact be eliminated and furthermore that the apparent redundancy here actually reflects an important generalization about languages and must therefore be retained. Consider the case of not that constructions discussed earlier (p. 250). In these constructions, as I have argued, the it is before not that is deleted under identity with an it is in a coordinate clause by the independently motivated rules of Not-clause Formation. This treatment implies the presence in the initial clause of an occurrence of it is, which must later be deleted. This seems to be evidence in favor of keeping the deletion transformation. On the other hand, there seems to be no way of doing away with the trivial calculus (108) accounting for the deep structure semantic nonsignificance of higher it is that clauses. If multiple self-embedding of it is that clauses is permitted, as seems necessary, then obviously some device is needed to explain the fact that English sentences are not infinitely ambiguous. In the arguments of some of the previous sections I begged the important question that I am now raising. I assumed earlier (p. 269) for example, that since the deep structure aspects of the semantic

Deriving S from S + Is

291

representations of pseudo-clefts are identical to the deep structure aspects of the semantic representations of the corresponding simple sentences, then the deep structures of both types of sentences must be identical. But given the semantic calculus (108) discussed here, this assumption is invalid. It could be argued that simple sentences have no higher it is that clause in their deep structures, while pseudoclefts, clefts, sentences with be to, the interrogatives cited earlier (p. 275), and the complex modal sentences given on p. 284 do have a higher it is that clause in their deep structures and that the equivalence in deep structure aspects of semantic representation is accounted for by the simple semantic calculus (108). This suggestion sounds plausible, but I repeat that there does seem to be a need for a deep structure it is that clause in simple sentences coordinated with not that constructions, which is deleted by a late transformation. We have reached an apparent impasse. We have postulated a deletion transformation and a simple semantic calculus, neither of which can be eliminated from the grammar, yet they essentially duplicate each other-they do the same work. Such a situation has rarely been encountered before in published linguistic research because hardly any grammars have included a semantic calculus of the type discussed here. I know at present of one other area of language where a semantic calculus must be postulated in order to account for the semantic equivalence of deep structures, but it is remarkable that in this case too a deletion transformation must be postulated which does the same work as part of the semantic calculus. The area of language to which I refer is the area of numeral systems. I can only sketch a few of the arguments here which justify the analysis of numeral systems I propose, but I believe the facts are fairly clear. The relevant arguments are given in detail in Hurford (in press). A number of arguments lead to the conclusion that something like (109) is the correct deep structure for the English expression six

thousand three hundred.

NUMBER

(109)

~

PHRASE

NUMBER

I\ I I

* I

6

3

NUMBER

M NUMBER

six thousand three 1000

M

I

hundred 100

292

James R. Hurford

A similar set of arguments leads to the conclusion that the correct deep structure for the Mixtec expression sia?u tuu sia?u siko, literally 15 400 15 20, is something like (ll0).

NUMBER

(ll0)

*

------------

NUMBER

PHRASE

I\

NUMBER

I

sia?u 15

'

M NUMBER

I

I

400

15

tuu sza?u

M

I

siko 20

(Mixtec data are from Merrifield, 1968; analysis is my own.) The English and Mixtec expressions given here both refer to the same number, 6300 in Arabic notation. Yet (disregarding, of course, the phonological forms of the terminal elements) the deep structures of these expressions are distinct. The problem of accounting for the semantic equivalence of the English and Mixtec expressions is very simply handled by postulating a set of interpretive semantic projection rules, relating the two distinct deep structures to a common semantic representation. The two rules necessary for interpreting (109) and (ll0) are given in (lll). (lll)

a.

b.

The value (reading) of any NUMBER is the sum of the values of its immediate constituents. The value (reading) of a PHRASE is the product of the values of its immediate constituents.

These projection rules and a few other similar ones are adequate for assigning semantic representations to the deep structures of numeral expressions in all languages, as far as I can tell. The deep structures involved can all be justified on internal and essentially syntactic grounds. Let us narrow our view to the constituent PHRASE. The semantic relationship between the constituents of a phrase, as we have seen, is one of multiplication. Now semantically (arithmetically) multiplying a number by 1 has no effect on it: the operation of multiplying by 1 is semantically vacuous. Yet frequently, in order to capture significant syntactic generalizations, one must postulate an element with the value 1 as an immediate constituent of a PHRASE. And it is

Deriving S from S + Is

293

frequently necessary to delete this element by a transformation. Consider an example from French. (112) is a list of some PHRASEs occurring in French. (112)

un cent cent deux cents un mille mille deux milles un million deux millions

'100' '100' '200' '1,000' '1,000' '2,000' '1,000,000' '2,000,000'

Note that (>million by itself is ungrammatical as a numeral. A clear generalization that can be made about the structure of these PHRASEs is that they have the form shown schematically below. (113)

un deux trois

neuf

r•t(,)

mille(s)

l

million(s)

This generalization can of course be captured in a phrase structure rule expanding PHRASE. A transformation must then be stated optionally deleting un before cent and mille. The optionality of un cannot be stated adequately by making it an optional element in a phrase structure rule expanding PHRASE, because that would generate the ungrammatical (>million. Un must be deleted by a transformation. Very similar facts to these can be found in a large number of languages. A transformation deleting an element with the value 1 from a PHRASE is very common in the grammars of numeral systems. But rules deleting any other element from a PHRASE are very uncommon. If there can be a transformation deleting un, why can't there be a transformation deleting deux, or trois, or neuf? It is clear that elements signifying 1 have some special status, in that they are much more deletable than other elements. Yet it is not the case that elements signifying 1 in numeral expressions can be deleted from just anywhere in a structure. They are deleted only when they are immediate constituents of PHRASEs. Just as it is intuitively inconceivable that cent could be used to mean 200, or 300, or 900, so it is

294

James R. Hurford

inconceivable that it could be used to mean 101, or that mille could be used to mean 1001. I know of no cases where an element signifying 1 is deleted when that element is semantically in a relationship of addition with a sister constituent. The statement needed to account for these facts is something like (114). (114)

Given a (part of a) deep structure D with the constituents X and Y, where the semantic rules of the grammar assign identical readings to both D and Y, there is likely to be a transformational rule deleting X.

This statement is imprecise, but something like it must exist in linguistic metatheory. Its existence makes a transformation deleting elements signifying 1 less costly in terms of an evaluation metric and minimizes the objection that this transformation and a part of the semantic component duplicate each other. Independent confirmation of principle (114) might be found in the Chinese numeral system. It is a peculiarity of Chinese to use the word ling 'zero' as a kind of place holder in certain numerals, though the presence of this word contributes nothing to the meaning of the whole numeral expression. Examples are: (115)

a. b.

ichian ling ell.shyrsyh (= 1,024) 11000 0 2 10 4 erh pai ling erh (= 202) 2 100 0 2

(These examples are respectively from Chao, 1968, p. 575; Brainerd and Peng, 1968, p. 78. In both cases I have used the romanization of the source.) Though I have not examined the Chinese numeral system in detail, it seems plausible that maximal generality of the syntactic description is attained if ling is regarded as corresponding to a PHRASE in deep structure. Thus the deep structure for (115b) is something like (116). (116)

NUMBER

----------

PHRASE

NUMBER

P~~BER

/\

NUMBER M NUMBER M

I

erh 2

I I

Pai ling 100

0

I

s hih 10 ·

erk 2

Deriving S from S + ls

295

Principle (114) predicts that the shih is likely to be deleted. Without this principle in the grammar shih in (116) would not appear to be mor.e likely to be deleted than pai. There are, of course, no instances of deletion of any element (except 1) which is in a multiplicational relationship with a number other than 0. It seems that the same sort of thing is going on with the deletion of it is that clauses as with the deletion of numeral elements. In the present connection, it is relevant to note another instance of the duplication of function by semantic rules and transformations in a grammar which makes use of semantic rules. In Emonds (1970) the deep structure proposed for both sentences of (117) is (118). (117)

a.

It would bother me for the house to be painted.

b. For the house to be painted would bother me.

s

(118)

it

would bother me for the house to be painted

To account for the fact that in both sentences for the house to be painted is semantically the subject of the sentence, Emonds proposes an interpretive rule which assigns subject function to a complement S when it is coreferential with an it in the subject position, as marked by the indices in (118). And to account for the fact that in (117b) the embedded sentence is actually the surface structure subject of the sentence, Emonds proposes a transformation of Subject Placement (which is the inverse of the well known Extraposition rule) moving the embedded S into subject position. I know of a further example from a numeral system where a transformation must be postulated which in a sense duplicates the function of what may be regarded as a semantic rule. In Biblical Welsh one finds expressions such as the following: (119)

chwe chan mil a thair mil (6 100 1000 + 3 1000) (= 603,000) gan mil ac wyth mil (100 1000 + 8 1000) (= 108,000)

296

James R. Hurford

I argue in detail in Hurford (in press) that expressions such as those of (ll9) must be derived transformationally from structures more like those of the expressions of (120). (120)

(= 603,000) (6 100 + 3 1000) chwe chant a thair mil (= 108,000) (100 + 8 1000) gant ac wyth mil

That is, there is a transformation having the effect shown roughly below. (121)

x

+y

1000:::} xlO00 + ylOOO

The rule shown roughly in (121) is clearly parallel to a particular case of the mathematical Law of Distribution, which may be expressed in its most general form as in (122). (122)

(x

+ y)z =

(xz)

+ (yz)

It can be shown that the mathematical Law of Distribution is necessary for the interpretation of numeral expressions in some languages and therefore belongs to the universal semantic calculus relevant to the interpretation of numerals. Here, then, we have another case in which a syntactic transformation duplicates the function of a semantic rule. Very few grammars propose semantic rules and so situations like this are infrequent in the literature, but it is possible that future research will reveal that it is common, perhaps even normal, for independently motivated semantic and syntactic rules to duplicate each other. If this is the case, then principle (ll4) can be extended to account for this more general phenomenon. For my present purposes, I believe I have shown that a grammar with both a transformation deleting a higher it is that clause and a semantic rule stating, in effect, that a higher it is that clause is semantically vacuous is not necessarily a bad grammar and indeed, as far as can be seen from research to date, is quite probably a good grammar, i.e., a grammar accurately reflecting the properties of human language. With this conclusion in mind, we return to the question of the correct deep structure for English sentences. If there is no trace in surface structure of a higher it is that clause, should we postulate such a clause in deep structure or not? I do not see how a general answer to this question can be given. In the case of many simple sentences standing in isolation we can, depending on intonation, account for exactly the same data if we postulate a higher it is that clause which is deleted transformationally as we can if we postulate

Deriving S from S + ls

297

a simple deep structure without the higher it is that clause. The question of whether such simple sentences have a higher it is that in their deep structures seems to be without empirical significance. The conclusion of this paper must be that such simple sentences may or may not have a higher it is that clause in their deep structures. But the more positive conclusion which is possible for the wide variety of rather more complex sentences we have discussed still stands.

WHERE ELSE TO LOOK

R. Lakoff (1969, p. 43) refers to unpublished suggestions by Postal, G. Lakoff, and Ross for deriving passives from deep structures such as I have proposed in this paper. Huddleston (1971, pp. 321-326) draws attention to some similarities between cleft sentences and sentences with existential there. It may well be that arguments for the deep structures proposed in this paper can be constructed from the data of passives and existential sentences. In conclusion, evidence may be found for an analysis essentially like mine in a number of other languages. I have come across the following studies in which conclusions like mine are adumbrated. For Mandarin, Ann Hashimoto (1969) has proposed deep structures for various emphatics and contrastives that closely resemble my deep structures and Makino (1968) has proposed similar structures for Japanese. A paper given by Amy Myers at the 1971 U.C.L.A. conference on African Linguistics suggests that there is evidence for the equivalent of a higher it is that clause in questions and emphatics in Kikuyu. James Matisoff, in a paper given at the Santa Cruz Summer Linguistics Conference, 1971, has drawn attention to a number of sentence types in Lahu, which I believe are susceptible to the kind of analysis I have proposed here. Finally, Ross (1970) cites an analysis of Arabic by Brame in which ?inna is said to be a complementizer "used only after the verb ?aquulu '(I) say' [p. 245]." The facts of Arabic as given by Ross beautifully support his performative analysis, but I have come across an account of the word Pinna which suggests that the facts might be even more interesting. Shehadi (1969) writes of the word, "It has the force of 'It is the case that .. .' [p. ll8]." Shehadi mentions ?inna as one of the Arabic equivalents of the English verb be and says that it can denote existence. Perhaps Arabic ?inna provides evidence both for Ross's analysis and for the equivalent of an it is that clause embedded in the performative.

298

James R. Hurford

REFERENCES Akmajian, A. On deriving cleft sentences from pseudo-cleft sentences Linguistic Inquiry, 1970, 1, 149-168. Baker, C. L., & Brame, M. K. 'Global rules': A rejoinder. Language, 1972 48, 51-75. Baker, S. The philosophy of mathematics. Englewood Cliffs, New Jersey: PrenticeHall, 1964. Brainerd, B., & Peng, F. C. C. A syntactic comparison of Chinese and Japanese numerical expressions. In H. B. Corstius (Ed.), Grammars for number names. Foundations of Language, Supplementary Series. Vol. 7. Dordrecht, Holland: Reidel, 1968. Chao, Y. R. A grammar of spoken Chinese. Berkeley: University of California Press, 1968. Chomsky, N. Remarks on nominalization. In R. A. Jacobs, & P. S. Rosenbaum (Eds.), Readings in English transformational grammar. Waltham, Massachusetts; Ginn and Company, 1970. Chomsky, N. Deep structure, surface structure, and semantic interpretation. In D. D. Steinberg, L. A. Jakobovits (Eds.), Semantics: An interdisciplinary reader in philosophy, linguistics and psychology. Cambridge, England: Cambridge University Press, 1971. Corstius, H. B. (Ed.) Grammars for number names. Foundations of Language, Supplementary Series. Vol. 7. Dordrecht-Holland: Reidel, 1968. Emonds, J. Root and structure-preserving transformations, Bloomington: Indiana University Linguistics Club, 1970. Fillmore, C. J. Types of lexical information. In D. D. Steinberg, & L. A. Jakobovits (Eds.), Semantics: An interdisciplinary reader in philosophy, linguistics and psychology. Cambridge, England: Cambridge University Press, 1971. Fillmore, C. J., & Langendoen, D. T. (Eds.) Studies in linguistic semantics. New York: Holt, Rinehart and Winston, 1971. Gleitman, L. R. Coordinating conjunctions in English. Language, 1965, 41, 260-293. Grinder J. T., & Postal, P. M. A global constraint on deletion. Linguistic Inquiry, 2, 1971, 110-112. Hashimoto, A. Y. The verb "to be" in modern Chinese. In J. W. M. Verhaar (Ed.), The verb "be" and its synonyms. Foundations of Language, Supplementary Series. Vol. 9. Dordrecht, Holland: Reidel, 1969. Huddleston, R. D. The sentence in written English, Cambridge, England: Cambridge University Press, 1971. Hurford, J. R. The linguistic theory of numerals. In press. Jacobs, R. A., & Rosenbaum, P. S. (Eds.) Readings in English transformational grammar. Waltham, Massachusetts: Ginn and Company, 1970. Jespersen, 0. A modern English grammar on historical principles. Part V, London: Allen and Unwin, 1940. Kiparsky, P., & Kiparsky, C. Fact. In D. D. Steinberg, & L. A. Jakobovits (Eds.), Semantics: An interdisciplinary reader in philosophy, linguistics and psychology. Cambridge, England: Cambridge University Press, 1971. Kuroda, S.-Y. English relativization and certain related problems, Language, 1968, 44, 244-266. Lakoff, G. Global rules. Language, 1970, 46, 627-639. Lakoff, G. On generative semantics, In D. D. Steinberg, & L. A. Jakobovits (Eds.), Semantics: An interdisciplinary reader in philosophy, linguistics and psychology. Cambridge, England: Cambridge University Press, 1971.

Deriving S from S + ls

299

Lakoff, G. The arbitrary basis of transformational grammar," Language, 1972, 48, 76-87. Lakoff, R. Abstract syntax and Latin complementation. Cambridge, Massachusetts: M.I.T. Press, 1969. Langacker, R. W. French interrogatives: A transformational description. Language, 1965, 41, 587-600. Langacker, R. W. French interrogatives revisited. Unpublished manuscript, University of California, San Diego, 1971. Makino, S. Japanese "be." In J. M. W. Verhaar (Ed.), The verb "be" and its synonyms. Foundations of Language, Supplementary Series. Vol. 3 Dordrecht, Holland: Reidel, 1968. Matisoff, J. Nominalization in Lahu. Paper presented at the Summer Linguistics Conference, University of California, Santa Cruz, 1971. Merrifield, W. Number names in four languages of Mexico. In H. B. Corstius (Ed.), Grammars for number names. Foundations of Language, Supplementary Series. Vol. 7. Dordrecht, Holland: Reidel, 1968. Myers, A. On similarities between interrogatives and emphatics in Kikuyu and English. Paper presented at the U.C.L.A. conference on African Linguistics, 1971. Peters, S., & Bach, E. Pseudo-cleft sentences. Unpublished manuscript, University of Texas, Austin, 1968. Reibel, D. & Schane, S. (Eds.) Modern studies in English. Englewood Cliffs, New Jersey: Prentice-Hall, 1969. Rosenbaum, P. S. Phrase structure principles of English complex sentence formation. Journal of Linguistics, 1967, 3, 103-118. Ross, J. R. Constraints on variables in syntax. Bloomington: Indiana University Linguistics Club, 1967. Ross, J. R. On declarative sentences. In R. A. Jacobs, & L. A. Rosenbaum (Eds.), Readings in English transformational grammar. Waltham, Massachusetts: Ginn and Company, 1970. Shehadi, F. Arabic and "to be." In J. M. W. Verhaar (Ed.), The verb "be" and its synonyms. Foundations of Language. Supplementary Series. Vol. 4. Dordrecht, Holland: Reidel, 1969. Steinberg, D. D., & Jakobovits, L. A. (Eds.) Semantics: An interdisciplinary reader in philosophy linguistics and psychology. Cambridge, England: Cambridge University Press, 1971. Thompson, S. A. The deep structure of relative clauses. In C. J. Fillmore, & D. T. Langendoen (Eds.), Studies in linguistic semantics. New York: Holt, Rinehart and Winston, 1971. UESP (= U.C.L.A. English Syntax Project) Integration of transformational theories on english syntax, U.C.L.A. Student Bookstore, Los Angeles, California. Verhaar, J. W. M. (Ed.) The verb "be" and its synonyms. Foundations of Language, Supplementary Series. Vols. 1,6,8,9. Dordrecht-Holland: Reidel, 1967-1969.

NONSOURCES OF UNPASSIVES 1 DOROTHY SIEGEL Massachusetts Institute of Technology

In this paper, I shall consider the source of sentences like the following: (1)

The president's blunder was unreported (in the press). 2 The brass knuckles were unpaid for. c. Antarctica is uninhabited (by man). 2 d. The garbage went uncollected.

a. b.

Throughout this paper, sentences like the above will be called unpassives. Unpassives are characterized by the presence of an auxiliary verb (either be or go), by the presence of a past participle

1 I am grateful to J. R. Ross (who, among other things, suggested that I call the sentences under consideration unpassives), David Perlmutter, G. H. Matthews, Edwin Williams, Mark Aronoff, Mary-Louise Kean, Bob Fiengo, and Roger Higgins for their helpful comments and criticisms. I would especially like to thank Howard Lasnik, who suggested this topic to me. All wrong proposals are, of course, my responsibility. 2 The by-phrase is in parentheses because there are many people who do not get unpassives with agent phrases. People who do get by-phrases in unpassives generally find that unpassives vary in acceptability according to the particular content of the byphrase. Thus, for me, (i) is better than (ii):

(i)

(ii)

Antarctica is uninhabited by man. "Antarctica is uninhabited by the guy next door who plays Olympic dominoes.

301

302

Dorothy Siegel

prefixed with un-, 3 and by the optional presence of an agent phrase. The most striking feature of unpassives, however, is the nonexistence of well-formed active sentences from which they could have been derived. That is, there are no grammatical sentences in English such as the following: (2)

a. b. c. d.

The press unreported the President's blunder. Someone unpaid for the brass knuckles. 0 Man uninhabits Antarctica. 0 Someone (went) uncollected the garbage. 0 0

Even though the sentences in (2) are ungrammatical, it is possible within the present theory to posit them as the underlying source of unpassives. Thus, the deep structure of a sentence like (le) could be:

NP

____-s~

j man

VP:::---___

v/

I

~NP

I

-----PP

uninhabit Antarctica

/ ~PASS

by

This source for unpassives is unacceptable, though, for several reasons. In the first place, this analysis entails positing the existence of many abstract verbs which never show up except as unpassives: 0 unacquit, 0 uninhabit, 0 unallude to. In other words, all unVs obligatorily select by PASS. This consequence, in itself, makes the 3 The un- under scrutiny in this paper is negative un-, not the privative un- which shows up in words like undress, unbutton, and unsaddle. Privative un- implies a reversal of the action specified in the verb. Aside from its semantic distinctness from negative un-, privative un- differs in another way from negative un-. Only privative un- shows up on verbs: to unbutton, "to uncollect. There are yet other differences. For example, privative un- can take progressive aspect, whereas negative un- cannot:

(iii)

The blouse was being unbuttoned. b. "The garbage was being uncollected.

a.

Also, private un-, unlike negative un-, can co-occur with subject selected adverbs like enthusiastically: (iv)

a. The blouse was unbuttoned enthusiastically. b. "The garbage was uncollected enthusiastically.

The existence of two un-s predicts that there should be ambiguities in un- words in which the action implied by the verb can be both reversed and negated; and in fact, such ambiguities exist: unbuttoned, unzipped.

Nonsources of Unpassives

303

analysis look suspicious, for I know of no other case in which some morphological property of a verb determines its selectional characteristics. Far more incriminating is the fact that this analysis complicates an otherwise perfectly regular generalization regarding the labeled bracketing of un- words. Explaining this fact entails a short digression. Negative un- shows up surfacely on lexical adjectives (unkind, unhappy), on nouns (untruth, unkindness), on adverbs (ungracefully, unendingly), and on paiticiples (uncollected, uninhabited). In every case, un- appears on a we,rd which contains an adjective somewhere in its derivation. For example, un- appears on nouns if and only if those nouns have been derived from adjectives. Thus, we correctly predict that we will find un- in none of the three following environments: (a) on underived nouns like '>unhonor and '>unlove; (b) on nouns which are derived from verbs: '>unintegration, '>unarrival, '>unresistance, 4 and (c) on de-adjectival nouns where unN is grammatical but unADJ is ungrammatical. Thus, the prediction is that where we do not find adjectives like '>uncold, we also will not find nouns like '>uncoldness. Similarly, since '>unhard is ungrammatical, we do not find '>unhardness. In fact, this prediction is borne out. In the simplest analysis, then, un- gets prefixed to adjectives only. Derivational suffixes which attach to adjectives may then be suffixed to [un-[]ADJ]ADJ• This means that the bracketing of untruth must be [[un[true]AoJlAoJ0]N. The bracketing of ungracefully must be [[un[graceful]AoJlAorly]Aov• Turning to the bracketing of un- with respect to participial adjectives, we see that the generalization governing the bracketing of un- predicts that the bracketing on participial adjectives must be [un-[[inhabit]v-ed]AoJ]ADJ· However, the analysis we are considering makes the claim that the bracketing is as follows: [[un-[inhabit]vJv-ed]ADJ• If the analysis under consideration were correct, then it would constitute the only case in which un- attaches to verbs. Furthermore, in just those cases where the analysis prefixes un- to verbs, the verbs must passivize and become participial adjectives. It seems, then, that a plausible solution to unpassive formation must incorporate some mechanism whereby un- gets attached to adjectives. This could be done in at least three ways. First, one could claim that unpassives are active sentences underlyingly and that unis the surface realization of the same NEG that underlies not. This NEG would get attached to the verb as un- after the S had been pas• The fact that we get unintegrated, unarrived, and unresistant shows that there is no restriction against un-s appearing on the stems in question.

304

Dorothy Siegel

sivized. Attending passivization would be the relabeling of Vas ADJ in surface structure. Second, one might argue that unpassives are not derived at all, i.e., that unparticiples are all deep adjectives and that unpassives are generated directly by phrase structure rules. Third, one might claim that unparticiples are the surface realization of without-being+ PARTICIPLE phrases. I will argue against all three of these proposals, concurrently outlining the essential features which a correct analysis of unpassives should incorporate. I am unable, however, to propose a satisfactory analysis-one which can account for the passive-like behavior and morphology of unpassives without resorting to the artifice of positing the existence of abstract verbs like (>uninhabit. There are myriads of reasons for rejecting the first alternative mentioned above-that un- is a surface realization of NEG. Most of these reasons follow directly from the fact that un-, unlike NEG, has as its surface scope only the word to which it is attached. In the first place, NEG creates ambiguities which do not arise with un-. Consider the following sentence. (3)

Harry is not well because of the weather.

Under one reading, well is negated and the sentence means that Harry is sick because of the weather. Under the second reading, NEG is understood as originating in the because-clause; Harry is well, but his good health cannot be attributed to the weather. Now notice that the corresponding sentence with un- is unambiguous. (4)

Harry is unwell because of the weather.

This sentence can only be taken to mean that Harry is sick. Thus, Neg and un- differ with respect to the ability to create ambiguities; there is no rule of Un- Transportation corresponding to NEG Transportation. If un- and not had identical sources, we would expect (4) to be ambiguous. The next class of examples shows that NEG may appear in environments where unpassives may not. The first case concerns verbs which take indirect objects. Verbs which require indirect objects, like give, may not co-occur with unpassives, although they may cooccur with NEG. 5 Thus, compare the following sentences: 5 David Perlmutter has pointed out to me the fact that it is only verbs which require indirect objects, not simply verbs which may take indirect objects, which cannot co-occur with unpassives. Thus, we get the book was unread but not "the book was

ungiven to the school.

305

Nonsources of Unpassives

(5)

a. The gift wasn't given to the school. b. 0 The gift was ungiven to the school.

A further difference between un- and NEG is that unpassives show up only in stative environments, whereas not is not so restricted. For example, not can co-occur with progressive aspect in passives, but un- cannot: (6)

a.

. . t h e press. N ixon wasn ,t {being} ff ri·d·icu led in

b. Nixon was { 0 b/ng} unridiculed in the press. Moreover, not + PARTICIPLE, unlike unPARTICLE, can co-occur with subject-selected adverbs like graciously, enthusiastically, and

reluctantly:

. ,s gaJJ ,.r-1". . d. (7) a. N ixon e wasn 't { enthusiastically} ff pu bl.icize

ff

0 . ,s gaJJ ,.r-1". . d. b . N ixon e was { enthusiastically} unpu bl.icize

In addition, when un- co-occurs with modals, like can, the only reading for the modal is the epistemic one. However, both the root and epistemic readings are possible with not: (8)

a. The garbage can be uncollected. b. The garbage can't be collected.

[I reading] [2 readings]

Furthermore, we observe that NEG can appear in environments which specify sentence negation, whereas un- fails to appear in these environments. This fact is a direct consequence of un-' s limited scope. I shall consider three such sentence negation environments: either-conjoining, negative appositive tag, and tag questions. In the first case, either-conjoining, we note that either, rather than also, appears in the rightmost conjunct of conjoined sentences only if the rightmost conjunct contains not or one of the negative adverbs like never, rarely, or barely. Thus, we get the following distribution of eithers and alsos: (9)

a.

Rightmost conjunct is positive:

. The trash { wasn't} picked up yesterday and the dog was was run over { o:::ier}· b.

Rightmost conjunct contains not:

306

Dorothy Siegel

The trash wasn't picked up yesterday and the garbage 0 also} wasn , t collected { either . c. Rightmost conjunct contains un-: The trash wasn't picked up yesterday and the garbage was uncollected { 0 :J:ier}· Likewise, we get so in sentences with un-, and neither in sentences with not, when either and also immediately follow the conjunction

and: (10)

a.

Lefunost conjunct is positive:

The garbage was collected and { 0 neis_oth er } was the trash. b. Lefunost conjunct contains un-: The garbage was uncollected and { 0 nets_oth er } was the trash. c. Lefunost conjunct contains not: The garbage wasn't collected and { nei~stho er } was the trash. In the next example, we note that not, but not un-, takes the negative appositive tag not even: (11)

a.

The garbage pails weren't dented, { ones.

0 b. The garbage pails were dented, {

,;t} even the

ones. c.

1:;} even the silver

0 The garbage pails were undented, {

silver

/t} even the silver

ones. Finally, sentences with not take positive tag questions: (12)

a.

The garbage wasn't collected, { 0 was wasn

't} it?

0 wa~t} it? b. The garbage was collected, { wasn

c.

0 The garbage was uncollected, { wa~t} it? wasn

An additional difference between un- and not is that when they ap-

Nonsources of Unpassives

307

pear immediately before lexical and participial adjectives which admit degrees of intensity, unAD] and not ADJ differ semantically. The notion "degree of intensity" needs clarification. Roughly, the adjectives which can be modified by very are those which admit degree-of-intensity interpretation, and these are the ones in which unADJ differs semantically from not ADJ. For example, words like hated, killed, collected, solved, identical, equal, and tied cannot be modified by very; and likewise, unhated seems to be semantically identical to not hated. On the other hand, words like satisfied, happy, and complicated can be modified by very. It is in words such as these that the difference in question shows up. Observe, for example, that not complicated admits interpretation of a greater range of complication than uncomplicated; something which is not complicated is necessarily uncomplicated, but the converse is not true. The difference can be sharpened by observing that the members of the pair not very complicated and very uncomplicated mean different things. If it were the case that un- could show up in a proper subset of the environments in which NEG could appear, then there might be motivation for deriving un- from a particular subset of NEG. However, this is impossible because there is an environment in which un- may appear, but NEG may not. Un-, but not not, can co-occur with the noninchoative go that appears in sentences like Sam went hungry, Susie went without sleep for three nights, and The refugees went homeless. Thus, we get (13a) but not (13b): (13)

a. The garbage went uncollected. b. (>The garbage didn't go collected.

We have just examined evidence for rejecting a hypothesis whereby unpassives are derived from underlying active sentences with un- as the surface realization of NEG. Next, we shall consider evidence which bears on another putative source of unpassives. It is conceivable that unpassives are not derived at all, i.e., that the rules of the base generate unpassives directly. In the section to follow, evidence will be presented which shows that this hypothesis is untenable. Concommitantly, it will be argued that unpassives must be derived from underlying active sentences, although not from an active sentence source that contains NEG. The first piece of evidence which helps vitiate the claim that unpassives are generated directly by the base involves verbs that take prepositional complements. With many verbs like tamper with, hint at, accede to, and pay for it is possible to passivize the object of the

308

Dorothy Siegel

preposition to get sentences like (14): (14)

Our wiretaps have been tampered with. b. Foul play was hinted at by the police. c. The demands were acceded to by the administration. d. The convention was paid for by big corporations. a.

Likewise, we find unpassives with the same verbs: (15)

Our wiretaps were untampered with. b. Foul play was unhinted at. c. The demands were unacceded to. d. The convention was unpaid for. a.

However, there are certain prepositions which do not allow their objects to passivize at all. For example, in the following sentences, the passivized ones are ungrammatical: (16)

a.

b. c.

d. e. f.

These duties devolve on the Vice President. The Vice President is devolved on by these duties. Our success hinges on his acceptance of the proposal. 0 His acceptance of the proposal is hinged on by our success. This result follows from your hypothesis. 0 Your hypothesis is followed from by this result.

0

In just those cases where it is impossible for objects of prepositions to passivize, it is also impossible for them to unpassivize. Thus, we get the following ungrammatical sentences: (17)

The Vice President is undevolved on (by these duties). b. 0 His acceptance of the proposal is unhinged on (by our success). c. 0 Your hypothesis is unfollowed from (by this result). a.

0

Now, if one claimed that unpassive participles were simply deep adjectives which were unrelated to their homophonous passive participles, one would be forced to claim that the facts in (16b,d,f), and (17) are unrelated. The starred sentences in (16) would be out because certain prepositional objects which are unpassivizable had been passivized. The starred sentences in (16), on the other hand, would be ungrammatical for some entirely different reason under the proposal we are considering. The impossibility of 0 undevolved on, 0 unhinged on, and 0 unfollowed from would be unrelated to the fact that the verbs devolve on, hinge on, and follow from cannot passivize; for under the proposal we are considering, the sentences in

Nonsources of Unpassives

309

("' 17) are underived. Therefore, the application of Passive in an impermissible environment is not a possible candidate for an explanation of their ungrammaticality. Clearly, the ungrammaticality of "'hinged on is related to the ungrammaticality of "'unhinged on. If the sentences in ("' 17) are derived from underlying active sentences their ungrammaticality can be accounted for by the same explanation that accounts for the ungrammaticality of sentences ("' 16,b,d,f). A second argument against the analysis we are considering comes from sentences like The animals were uncared for. Since care for is a transitive verb which can't take optional object deletion (cf. "'Sam cares for), animals must be the deep direct object of (un)cared for. If animals were not the deep object of for, we are hard put to explain why "'The animals were uncared for the bananas is ungrammatical. However, if the animals were uncared for has animals as the object of for somewhere in its derivation, then we can explain why we don't get "'The animals were uncared for the bananas. A third argument involves the agent phrases that may appear in unpassives. Most unpassives sound best without agent phrases, but there is a class of unpassive participles which require the presence of an agent phrase: (18)

a. The tests went uncomplicated by mishaps. b. "'The tests went uncomplicated.

(19)

Fortunately, his problems were uncompounded by the meddling of incompetents. b. "'Fortunately, his problems were uncompounded.

(20)

a. Sam was unstayed by minor obstacles. b. "'Sam was unstayed.

(21)

a. Sam's expectations were unfanned by any enthusiasm. b. "'Sam's expectations were unfanned.

a.

Since verbs-but not adjectives-are subcategorized for the manner adverbial by NP, it follows that the verbal roots which underlie the unparticiples in the (a) sentences in (18)-(21) above are responsible for the appearance of the by-phrases. Complicate and compound, for example, must appear in the deep structures of (18a) and (19a), respectively, in order to account for the selection of the agent phrase. Furthermore, it is sometimes the case that agent phrases in passives contain prepositions other than by: (22)

a. b.

The tragedy was reported in the press. The problem was discussed among his friends.

310

Dorothy Siegel

In just these cases where a regular passive may take an agent phrase headed by a preposition other than by, the unpassive may also. Thus we find: (23)

a. The tragedy was unreported in the press. b. The problem was undiscussed among his friends.

If unpassives did not have a passive source, then the appearance of prepositions other than by in certain unpassives in just those cases where regular passives may take agent phrases headed by prepositions other than by would be an accident. A fourth argument concerns the class of verbs which can be roots in unpassives. If unpassives were not directly related to verbs through the passive transformation, then we would expect to find X was un Y' d (by Z) where Y is a verb which cannot passivize. If there were a root in an unpassive such that the root could not passivize, then this would be strong evidence against deriving unpassives from actives. 6 Since only transitive verbs may passivize, the existence of an intransitive or otherwise nonpassivizable stem in an unpassive would constitute strong motivation for not deriving unpassives from underlying structures with transitive verbs. But in fact, there seem to be no cases of this sort. On the contrary, the verbs which appear in unpassives are a proper subset of those which can passivize. This fact does not vitiate the generalization that unpassive participles are derived from underlying verbs. The existence of roots that could appear in unpassives, but not in passives, would constitute damaging evidence. But the fact that there are verbs which can appear as passives but not as unpassives does the generalization no harm. In fact, unless we claim that unpassives are the products of derivations which include the passive transformation, we would be implicitly claiming that the distribution of unpassive roots was an accident; it might just as well have been the case that unpassive participles had as roots all and only those verbs that could undergo Psychmovement. Certainly to claim that unpassives are not derived is to deny the importance of the fact that their roots form a proper subset of passivizable verbs. Let us consider, now, the class of verbs which do not show up as unpassives. I have been able to discover seven categories of verbs 6 Apparent counterexamples such as unparalleled (by), unaccustomed to, unexampled (by), and unabashed (by) are not real unpassives, for they do not occur in all

environments in which true unpassive participles can occur. For example, all unpassives can co-occur with go. Thus, we get the garbage went uncollected but not "the

discovery went unparalleled.

Nonsources of Unpassives

311

which do not unpassivize. The first four categories include, predictably, verbs which do not passivize: intransitives (expire, gasp, relapse, prevail), verbs whose prepositions do not allow their objects to passivize (hinge on,falter in, rhyme with, devolve on), middle verbs (have, resemble, weigh, lack), and verbs like judge, feel, consider, guess, and find, which take sentential objects but do not passivize. 7 The next three categories include verbs which passivize but which do not unpassivize. The first of these includes transitive verbs in disX where X is a free verb and dis- is a negative prefix. So, for example, we do not get 0 undisqualified, 0 undisavowed, 0 undisowned, 0 undissatisfied, or 0 undispossessed, although we do find undismayed, undisputed, undissolved, undistracted, and undiscussed. This fact follows from a general morphological constraint in English that negative prefixes like un-, mis-, in-, ir-, and dis- cannot stack up. Thus, we find unmistaken and unmissed but not 0 unmismanaged or 0 unmistrusted. We get unirrigated and unirritated but not 0 unirresolved and 0 unirreclaimed. We find uninfluenced and uninhabited but not 0 uninexperienced. Likewise, words like 0 ununtouched, 0 ununarmed, and 0 ununstressed are ungrammatical. The second category of roots which can passivize but not unpassivize includes verbs which take two prepositional complements, such as make away with, do away with, put up with, and get away with. We never get unpassives of the form 0 undone away with. Contrasting ungrammatical unpassives having two prepositional complements like 0 undone away with with grammatical unpassives with prepositional complements like untampered with, unalluded to, and unprotested against, we note the following difference: in the ungrammatical cases, there is greater stress on the participle than there is on the preposition which follows. Now, suppose we make the generalization (24): (24)

In surface structure, there may not be a more highly stressed word in the VP than the word to which un- is attached.

Since un- must be attached to the participle, we get an ungrammatical output if a higher stressed word follows the unparticiple in the 7 J. R. Ross has noticed (class lectures, M.1.T., 1971) that although these verbs do not have good outputs as passives, if sentences containing them are passivized and then the subjects are extraposed, the output is grammatical:

(v)

a. Sam feels that Mary has no business in the cookie jar. b. 0 That Mary has no business in the cookie jar is felt by Sam. c. It is felt by Sam that Mary has no business in the cookie jar.

312

Dorothy Siegel

VP. I cannot prove that (24) is correct. In fact, I suspect that it is not. Nevertheless, it does account for yet another environment in which unpassives cannot occur, namely with verbs that are subcategorized for __ NP PP. This class of verbs includes not only idioms like take care of X, take advantage of X, and give credence to X but also give and bring, which take indirect objects. In neither case can we get good unpassive outputs: (25)

a. Susie will be C'un)taken good care of b. Good care will be ((>un)taken of Susie.

(26)

a. His kindness was ((>un)taken advantage of b. Advantage was ((>un)taken of his kindness.

(27)

The proposal was (0 un)given credence. b. Credence was ((>un)given to the proposal.

(28)

a. a.

b.

The chairman was ((>un)brought the minutes. The minutes were ((>un)brought to the chairman.

Moreover, (24) correctly predicts that verb phrases like defend X against Y and distinguish X from Y should be unable to underlie well-formed unpassives, and indeed they are: (29)

a. b.

Sam was (0 un)defended against perspiration odor. The lettuce was ((>un)distinguished from the cabbage.

The final argument against the claim that unpassives are underived concerns the Crossover Principle. If unpassives are underived, then it is difficult to explain why the Crossover Principle applies to them: (30)

. Sam ,s views were unc h all enge d b y { h.himself} . ts peers (>

. (31) S am ,s h atrcut was unri·d·tcu ledb y {(>himself} h. :+: • ts WtJe So far I have argued against three analyses which might be advanced as sources of unpassives. First, it was shown that unpassive participles could not come from underlying abstract verb sources like (>uninhabit and (>unpay for. Second, it was shown that a source in NEG inhabit was impossible because of the lack of parallelism between the occurrence of un- and the occurrence of not. Third, it was shown that if one claimed that unpassives were directly generated by the base, certain important generalizations could not be captured. J. R. Ross has made the interesting suggestion that unpassive par-

Nonsources of Unpassives

313

ticiples like unbeaten have as their source the phrase without being beaten. Thus, (32b) would come from (32a) under his proposal:

The Sox went without being beaten for four consecutive seasons. b. The Sox went unbeaten for four consecutive seasons. a.

(32)

This analysis has the advantage of accounting for the co-occurrence of noninchoative go with unpassive participles. Noninchoative go may co-occur with adjectives which imply deprivation (hungry, horny, thirsty, naked, barefoot), certain -less adjectives (carless, armless, topless, homeless), without-phrases (without sleep, without being beaten, without water, without anyone's knowing), and with unpassive participles (uncollected, unmourned, undiscovered). 8 If unPARTICIPLE comes from without being PARTICIPLE, we can automatically explain why unpassives co-occur with go. The problem with this analysis is that the distribution of unparticiples is not parallel to the distribution of without being phrases. For example, without being phrases can co-occur with negatively prefixed participles, as we see in (33)-(35): (33)

The racers went without being disqualified for the entire meet. b. The racers went undisqualified for the entire meet.

(34)

The funds went without being mismanaged until Max took over. b. The funds went unmismanaged until Max took over.

(35)

The soldiers went without being unarmed by the enemy for 6 years. b. The soldiers went ununarmed by the enemy for 6 years.

a.

a.

a.

In addition, maintaining this proposal would entail coming up with 8 David Perlmutter has pointed out to me that the complement of noninchoative go must be something the opposite of which is usually presupposed to be the normal case. An example will clarify this rather confusing generalization. Consider the sentence pairs:

(vi) (vii)

The elephant went trunkless. b. 0 I went trunkless.

a.

a.

b.

0

The Chinese went without nuclear weapons for years. I went without nuclear weapons for years.

The (b) sentences are strange because it would not normally be expected of me that I should have either a trunk or nuclear weapons.

314

Dorothy Siegel

ad-hoc ways to block the (b) sentences in the following sentence pairs: (36)

Our dreams were undisturbed by the raging storm. b. (>Our dreams were without being disturbed by the raging storm.

(37)

a. The children went for years without being amply cared for. b. (>The children went for years amply uncared for.

(38)

The proposal went without being given credence. b. (>The proposal went ungiven credence.

a.

a.

For these reasons, I would reject the proposal that unpassive participles be derived from without being phrases. I am unable at this point to propose a satisfactory analysis of unpassives, but I will attempt to show below that such an analysis must permit un- to be inserted into trees as an independent lexical item at the level of deep structure. The fact that the selectional restrictions on un- are distinct from the selectional restrictions on the verb to which it eventually becomes attached lends support to the claim that un- is underlyingly an independent lexical item. There are at least five selectional facts relating to un- that must be stated somehow in the grammar. The first of these is the fact that un- may select the noninchoative go, in the sense of remain, which appears in sentences like the tests went uncomplicated by mishaps. (I am not considering the go which appears in sentences like the bread went stale.) Notice that go may appear only in unpassives but not in passives: (39)

{ (>contested } . The verdict went uncontes t ed . collected } b. The trash went unco ll ec t ed . { (>named} . d . c. The ship went unname a.

{

(>

The selection of go must be positively specified by un- and not by the roots contest, collect, and name, as the examples in (39) show. A further selectional fact involves the verb get. Get may co-occur with regular participles but not with unparticiples: (40)

a.

b.

. acquitted } Sam got { unacqui'tt ed by the ;ury. ridiculed } Sam got { unri'd'icu led by the mob. (>

(>

315

Nonsources of Unpassives

c.

. inhabited } . h a b i't ed by Eskimos. Greenland got { .,, unin

Thus, we see that this selectional restriction must be stated separately for un-, which cannot take get, and for verbal roots like acquit, inhabit, and collect, which can take get. Another selectional fact involves remain to be. No unpassives can co-occur with remain to be, whereas many passives can: 9 (41)

. . remained t d }. to be { .,, contested The verdict uncontes e collected . b. The trash remained ll ec t ed }. to be { .,, unco . l'ists. . to b e { .,, examined . d } b y specza c. Th e tes t s remain unexamine a.

Thus, we see that a verbal root, like contest, may differ from un- in its ability to co-occur with remain to be. A further fact is that as yet may co-occur with unparticiples but not with regular passive participles: (42)

. uncompleted} The plans are as yet { .,, comp le t ed by the engineers. . . as yet {unemulsified} l :+: d by chemists. .,, b. The substance is emu si1 ie {uninhabited} b y E sk'imos. . . c. A n t arc t tea is as yet "'inhabited a.

Un- must be permitted to select the adverbial as yet. This fact can be explained by the fact that other semantically negative items may select as yet also: 9 It is unclear how the class of verbs which takes remain to be is to be characterized. The following incomplete lists show that co-occurrence with remain to be certainly cuts across distinctions made by stativity:

Occurs with remain to be

Does not occur with remain to be

Stative

Nonstative

Stative

Non stative

understood appreciated perceived

promised written divulged

entailed hated precluded implied suspected

whispered

316 (43)

Dorothy Siegel

a. He has as yet barely begun his study. b. He can't as yet hold his liquor. c. His plans for the future are as yet indefinite.

In addition, we note that un- must be able to select stative environments, even though the verb it eventually attaches to may be [-stative]. For example, some passives may co-occur with subject-selected adverbs like enthusiastically, but no unpassives may co-occur with these adverbs: (44)

The proposal was enthusiastically { ,:,ud:t:;:::ed }-

Furthermore, some passives can take progressive aspect, but no unpassives can: (45)

The proposal is being { ,:,u~!:;:::ed }·

The stativity of un- seems to overwhelm the nonstativity of any verb it is in construction with, at least with respect to selection. Summarizing, we see that an explanatory analysis of unpassives must include the following features: (1)

(2) (3)

Un- must be inserted into trees as a lexical item at the level of deep structure in order to account for its ability to select other lexical items in the tree. For reasons discussed earlier, uncannot be attached to the verb until it has become a participial adjective. It remains to be determined under what node un- is introduced. Unpassives must be derived from deep structures containing passivizable verbs, and Passive must apply in the derivation of unpassives. By some rule of Unattachment, un- must become attached to what was once the deep verb only after this verb has become a participial adjective. This will entail positing a Passive transformation in which Vis relabeled as ADJ. This preserves the generalization with respect to the bracketing of un- mentioned earlier.

REFERENCES Bolinger, D. Adjectives in English: Attribution and predication. Lingua, 1967, 18, 1-34. Chomsky, N. Aspects of the theory of syntax. Cambridge, Massachusetts: M.I.T. Press, 1965.

Nonsources of Unpassives

317

Jespersen, 0. A modern English grammar. Vol. 6. Morphology. London: Allen and Unwin, 1961. Marchand, H. Categories and types of present-day English word formation. University, Alabama: University of Alabama Press, The Oxford English Dictionary. Vol. II. T-U. Oxford: Clarendon Press, 1961. Wood, F. T. English prepositional idioms. New York: St. Martin's Press, 1967.

INDEX

A

Adjective(s), 16, 56, 57ff, 63-65, 303, 307ff, 313 lexical, 307 participial, 303, 307 plural, 63-65 relative, 16 Adpositions, 31, 32 Adverb(ial)s, 5ff, 16, 17, 18, 217, 218, 220, 224-228, 238-242 classes of, 5ff, 217, 218, 224 logical form of, 239-242 negative, 305 semantic function of, 242-244 sentential, 220 syntactic constraints on, 238, 239 two-place, 224-228 Agent phrases, 309 Akmajian, A., 124, 142, 154, 180, 272 Alajounine, T., 198 All, 65-233 Ambiguity advoidance, see Disambiguation Principle Anaphora, 117, 138-140 Anomia, 199 Aphasia, 198-202 agrammatical, 198 and automaticity, 202 causes of, 199 expressive, 198, 199 As yet, 315, 316 Argument raising, 60-62, 72 Argument Raising Principle, 15, 16, 22, 23, 24ff Automatic expressions, 201 Automatic language, 197, 199, 201, 202

318

Automatic/propositional dimension of language, 197-202 implications for linguistic theory, 200, 202 Automaticity, 203 Auxiliary deletion rule, 164 Auxiliary reduction rules, 162, 163 B Bach, E., 120, 262, 263 Backward pronominalization, 169 Baker, L., 120, 264 Bartsch, R., 8, 15, 24, 26, 40, 41, 48, 51 Bay, E., 198 Bech, G., 182 Be-deletion, 284, 285 Be-passive, 207, 208 Be to, 248, 250 Bever, T., 202 Bierwisch, M., 40, 182 Blumer, D. P., 200 Bock, R. D., 199 Bokamba, 96 Brahme, M. K., 264, 297 Brainerd, B., 294 But not, 259-261

C Carden, G., 227 Carefully, 233 Categorization rule, 53 Causative constructions, 21, 22, 209-211 Chao, Y. R., 294 Chase, R. A., 200 Chi Bemba, 95-102

319

Index Chomsky, N., 25, 124, 139, 143, 144, 161, 162,174,261,263 Clark, R., 247 Clause-mate constraint, 176 Cleft and pseudo-cleft, 130-132 Collocation, 202 Come, 208, 209 Comparative constructions, 162-165 Competence, 197, 203 Complementizer placement, 252, 266, 268 Complex NP shift, 187 Compositional theories of meaning, 206, 213-215 Conditional expressions, 111-113 Conjunction, 190, 191 Constituent building rule, see categorization rule Contingency grammar, 202 Copula deletion, 132 Critchley, M., 200 Crossover principle, 312, 313 Culicover, P., 186 Cullen, K. J., 200 Curme, G. 0., 160, 212

D Danks, J., 202 Dative movement, 186 Davidson, D., 243 Deletion rule, 156, 157 Dichotic listening, 200 Dis-, 311 Disambiguation principle, 33 Distribution, law of, 296 Dysphasic utterances, 200 Dzamba, 96, 97

E Edmonds, J., 150, 151, 152, 161, 165, 170, 171, 181, 189, 191 Epilepsy, 199, 200 Epileptic ictal speech, 200 Epistemic, 305 Equi-NP condition, 103, 239 Espir, 199 Extraction, analysis, 158, 268 Extraction rule, 154 Extraposition, 149,150,151, 173-183, 249,252,284,287

Extraposition, (Cont'd) and cyclicity, 177-183 in German, 181-183

F Factive complements, 280-283 Factive modality, 110 Falconer, J. A., 200 Faraci, R., 168 Fillmore, C., 8, 9, 120, 211, 272 Focus, 124, 132-135, 143-145 multiple, 143-145 For-insertion, 287 Frege, 214 Future-gap, 100, 101 Future modality, 108-111

G Generative Semantics, 214 Gerunds, 151, 171 Get, 205-214, 314,315 environments of, 205 inchoative, 212 paraphrases for, 202, 206 syntactic source of, 206 Get-passive, 207, 208 Gilbert, K. F., 160 Giv6n, T., 32, 95, 99, 104 Gleitrnan, 190,197,254,257 Global derivational constraint, 264 Glucksberg, S., 202 Go, 313 Goldman-Eisler, F., 200 Goldstein, K., 198 Grammers, 2, 3 function of, 2, 3 Greenberg, J., 26, 27, 28, 29, 197 Grinder, J., 264 Gruber, J., 118

H Habitual mode, 109-111 Hashimoto, Ann, 997 Have, 211-213 active, 211 ambiguities in, 213 stative, 211, 212 Head, H., 198 Head-NP deletion, 133

Index

320 Hemispherectomy, 199 Heny, F., 140, 144, 232, 244 Hesitation phenomena, 200 Hillier, W. E., 199 Houston, S., 202 Huddleston, R. D. 249 Hurford, J. R., 291

I Ictal speech automatisms, 199, 200 Igbo, 147, 148 Inchoatives, 208-213, 307ff source for get, 210, 211 Indefinite article, 96 Infinitivization, 285, 287 Interrogatives, 275-280 Is-deletion, 288-297 It-deletion, 150, 158, 179 It-replacement, 284, 287

J Jackendoff, R., 179, 186 Jackson, H., 198 Jesperson, 0., 160, 249 Jones, C. V., 199

K Kaplan, D., 233 Karttunen, L., 104 Kihung'an, 125-135, 138-141, 147 Kiparsky, L., 250, 280, 288 Kiparsky, P., 250, 280, 288 Kuroda, S. Y., 268

L Labov, 197, 200 Lakoff, G., 124,149,217,224,225,228, 235,240,241,264,297 Lakoff, R., 207, 208, 297 Langacker, R., 169, 275, 276 Langendoen, D. T., 25, 202 Language acquisition, 202 Language lateralization, 198-202 Leech, G., 271 Lehman, W. P., 28, 30, 32 Lennenberg, Eric, 198 Lexicalization, 206, 210, 213 Logic, 52, 53, 78ff, 91 extensional, 52ff

Logic, (Cont'd) intensional, 53, 78, 91 Luganda, 97

M Makino, S., 297 Matisoff, J ., 297 May, 210 May+ come, 210, 211 McCawley, J., 206 Meaning,9, 214,215 Meaning postulates, 9, 10 Meillet, A., 28 Merrifield, W., 292 Mixtec, 292 Modals, 283-285, 305 Montague grammar, 78-91 Montague, R., 52, 78, 84, 85, 243 Morgan, 114, 116 Mould, M., 97 Must+ come, 210, 211 Mutative intransitives, 212, 213 in contemporary German, 213 stativity of, 213 Myers, A., 116, 140, 144, 297 N

Natural constituent structure, principle of, 41 Natural generative grammar, 12, 26, 51-92 Natural serialization, 26, 40 NECESSARILY, 218-220 Neg., 98, 104, 303, 304, 307 in con tr 1st with un-, 304ff and opacity, 98 Negation, 97, 98, 109-111, 129, 130 Negative appositive tag, 306 Negative placement, 252 Neg-raising, 135-138 Neg-transportation, 304 Neubauer, 106 Niedermeyer, E. F. L., 200 Nonfactive modality, 110 Nonopaque environment, 96 Not-clause formation, 257, 290 Not that clauses, 250-261 Noun phrases, see also Plural conjoined, 85-91 and referentiality, 96

321

Index Nouns, see also Plural conjoined, 73 Number, 29-295 Numerals, 63-65 Numerically quantifying expressions, 59, 60 0 Opacity, 96-109, 222, 223, 232, 234 Opaque environments, 96, 97, 109 cause of, 109 Optional object deletion, 309 Output condition, 192-194 Palilalia, 200 Parallelism, 163, 164, 312 Parsons, T. A., 243 PARTICIPLE, 305 Passive, 174, 186, 191,192,287,302,304, 308-310, 316 by_, 302 Past tense, 98, 108 Pause fillers, 197 Pauses, see Hesitation phenomena Peng, F. C. C., 294 Perfect aspect, 212, 213 Performance-competence, 200-202 Performance data, 200, 202 Perlmutter, D., 163 Peters, S., 262, 263 Phonetic representation, 47, 48 Phrase structure rules, 304 Plural, 52-56, 65-78, 80-91 noun, 52, 53-56,80-84 noun phrase, 52 predication, 73-78 terms, 65-73, 84-91 Possessive forms, 189, 190 Postal, Paul, 124, 169, 264, 297 Predicate modifiers, 222 Predicate nominals, 116-119 Predicate-raising, 206, 207, 210, 211 Prepositional phrases, 187-189 Present-progressive tense, 108 Presuppositions, 272-275 Propositional/automatic dimension of language, see Automatic/propositional dimension of language Propositional language, 197 Pruning, 174, 176

Pseudo-cleft formation, 154-159, 167, 168, 171, 191, 261-272 Psych-movement, 310

Q Quantification, 20, 21, 202, 222, 232, 235 scope of, 232 Quantificational rule, 202 Quantifier-scope ambiguity, 222 Questions, 113-116 Quine, W., 95, 233 Quirk, R., 197 R

"Real verb get," 208 Referentiality, 95, 102, 120-122 Reichenbach, H., 243 Relative clause formation, 167, 168, 171, 267 Relative clauses, in Kihung'an, 127-129 Relative pronoun deletion, 132, 133 RELUCTANTLY, 222,223 Remain to be, and selection restrictions on un-, 315 Robinson, J. 0., 125 Root transformation, 152, 158, 160-162 defined, 152, 162 Rose, 199 Rosenbaum, P. S., 149, 151, 280 Ross, J. R., 25, 134, 149, 151, 156, 162, 176,177,181, 192-194,248,261,289, 297,312 Russell, 68

s Schachter, P., 116, 125, 144, 261 Scope, 218,221,231,235 ambiguity of, 231, 235 Semantic primes, 213 Semantic reading, 213 Semantic representations, 4, 10, 15, 47, 209,213,214 in compositional theories of meaning, 213,214 defined,47 Semantics, see Meaning Sentential relative constructions, 160, 161 Sentential subject constraint, 181

322

Index

Serafetinides, E., 200 Shehadi, F., 297 Should- deletion, 287 Simple tenses and opacity, 108, 109 Smith, A., 199 Some, 65-73 Spanish, 102 Speaker attitude, 207 Staal, J. F., 24 Stalnaker, R., 217, 218, 243 Stark, R. E., 200 Statives, 271, 305, see also Verbs environment for un-passives, 305 Stativity, 316 Structure preserving rules, 165, 166, 171 Structure-preserving transformation, 153, 158,165 Subject-auxiliary inversion, 151 Subject nominals, 119-120 Subject placement, 295 Subject raising, 174 Subject replacement, 150-166, 173-177 Surface structure constraint, 151 Synonymy, 205,206,207, 213-215 Syntactic rules, defined, 3, 4 7

T Tag question formation, 276, 277 Tag questions, 305, 306 Takizala, A., 116 Tarski, A., 241 That- insertion, 251 Thomason, R., 217, 218, 243 Thomason, S. A., 261 Tone-focus, 126 Topicalization, 159, 160, 167, 168, 171, 180, 181 Topicalized constructions, 159, 160 "Topicalized-it-replacement", 160 Tough-movement, 287 Transparency, 234 Truth values, 239, 241, 243

u UCLA phonetics lab, 200-203 Un-, 301-316 as lexical item in deep structure, 314,

:.ms

Un-, (Cont'd) with modals, 305 placement of, 303 selectional restriction on - - , 314, 315 Un-attachment, 316 Un-passive formation, 303 Un-passive putitive derivation, 307-316 Un-passives, 301-316 putative source for, 302 Un-transportation, 304 Un- words, bracketing thereof, 303 Universal quantification, and opacity, 121 Universal semantic calculus, 291-297 V

van Pelt, D., 199 Vennemann, T., 8, 15, 25, 46, 48, 51, 61, 63, 72 Verbs, 12, 13, 19, 20, 97-108, 307-312 coercive-manipulative, 103, 104 cognition, 104-108 factive, 105-108 future implicative, 101-103 implicative, 99-104 intensional, 102 intransitive, 311 middle - - , 311 modal, 19, 20 of modality, 98-101 neg-factive, 105 neg-implicative, 99-101, 103 nonfactive, 12, 13, 105 nonimplicative, 99, 101, 103, 104 perceptual, 107, 108 subcategorized for by-NP, 309, 310 subcategorized for __ NP PP, 312 transitive, 97, 309-311 VSO/SOV/SVO, 27ff, 40ff

w Weinreich, U., 20, 202 Wepman, J. M., 199 Wh- movement, 181 Wh- questions, 140, 141 Without being phrases, 314 Without- phrases, 313, 314 Word, defined, 214